1 |
Author: mpagano |
2 |
Date: 2014-05-05 14:28:48 +0000 (Mon, 05 May 2014) |
3 |
New Revision: 2772 |
4 |
|
5 |
Removed: |
6 |
genpatches-2.6/trunk/3.15/1000_linux-3.14.1.patch |
7 |
genpatches-2.6/trunk/3.15/1001_linux-3.14.2.patch |
8 |
genpatches-2.6/trunk/3.15/4200_fbcondecor-0.9.6.patch |
9 |
genpatches-2.6/trunk/3.15/5000_enable-additional-cpu-optimizations-for-gcc.patch |
10 |
genpatches-2.6/trunk/3.15/5001_BFQ-1-block-cgroups-kconfig-build-bits-for-BFQ-v7r2-3.14.patch |
11 |
genpatches-2.6/trunk/3.15/5002_BFQ-2-block-introduce-the-BFQ-v7r2-I-O-sched-for-3.14.patch1 |
12 |
genpatches-2.6/trunk/3.15/5003_BFQ-3-block-bfq-add-Early-Queue-Merge-EQM-to-BFQ-v7r2-for-3.14.0.patch |
13 |
Modified: |
14 |
genpatches-2.6/trunk/3.15/0000_README |
15 |
Log: |
16 |
Remove patches incompatible with 3.15 until they can be worked on. |
17 |
|
18 |
Modified: genpatches-2.6/trunk/3.15/0000_README |
19 |
=================================================================== |
20 |
--- genpatches-2.6/trunk/3.15/0000_README 2014-05-05 14:22:27 UTC (rev 2771) |
21 |
+++ genpatches-2.6/trunk/3.15/0000_README 2014-05-05 14:28:48 UTC (rev 2772) |
22 |
@@ -42,14 +42,6 @@ |
23 |
|
24 |
Individual Patch Descriptions: |
25 |
-------------------------------------------------------------------------- |
26 |
-Patch: 1000_linux-3.14.1.patch |
27 |
-From: http://www.kernel.org |
28 |
-Desc: Linux 3.14.1 |
29 |
- |
30 |
-Patch: 1001_linux-3.14.2.patch |
31 |
-From: http://www.kernel.org |
32 |
-Desc: Linux 3.14.2 |
33 |
- |
34 |
Patch: 1500_XATTR_USER_PREFIX.patch |
35 |
From: https://bugs.gentoo.org/show_bug.cgi?id=470644 |
36 |
Desc: Support for namespace user.pax.* on tmpfs. |
37 |
@@ -74,10 +66,6 @@ |
38 |
From: Al Viro <viro <at> ZenIV.linux.org.uk> |
39 |
Desc: Do not lock when UMH is waiting on current thread spawned by linuxrc. (bug #481344) |
40 |
|
41 |
-Patch: 4200_fbcondecor-0.9.6.patch |
42 |
-From: http://dev.gentoo.org/~spock |
43 |
-Desc: Bootsplash successor by Michal Januszewski ported by Jeremy (bug #452574) |
44 |
- |
45 |
Patch: 4500_support-for-pogoplug-e02.patch |
46 |
From: Cristoph Junghans <ottxor@g.o> |
47 |
Desc: Support for Pogoplug e02 (bug #460350), adjusted to be opt-in by TomWij. |
48 |
@@ -85,20 +73,3 @@ |
49 |
Patch: 4567_distro-Gentoo-Kconfig.patch |
50 |
From: Tom Wijsman <TomWij@g.o> |
51 |
Desc: Add Gentoo Linux support config settings and defaults. |
52 |
- |
53 |
-Patch: 5000_enable-additional-cpu-optimizations-for-gcc.patch |
54 |
-From: https://github.com/graysky2/kernel_gcc_patch/ |
55 |
-Desc: Kernel patch enables gcc optimizations for additional CPUs. |
56 |
- |
57 |
-Patch: 5001_BFQ-1-block-cgroups-kconfig-build-bits-for-v7r2-3.14.patch |
58 |
-From: http://algo.ing.unimo.it/people/paolo/disk_sched/ |
59 |
-Desc: BFQ v7r2 patch 1 for 3.14: Build, cgroups and kconfig bits |
60 |
- |
61 |
-Patch: 5002_BFQ-2-block-introduce-the-v7r2-I-O-sched-for-3.14.patch1 |
62 |
-From: http://algo.ing.unimo.it/people/paolo/disk_sched/ |
63 |
-Desc: BFQ v7r2 patch 2 for 3.14: BFQ Scheduler |
64 |
- |
65 |
-Patch: 5003_BFQ-3-block-add-Early-Queue-Merge-EQM-v7r2-for-3.14.0.patch |
66 |
-From: http://algo.ing.unimo.it/people/paolo/disk_sched/ |
67 |
-Desc: BFQ v7r2 patch 3 for 3.14: Early Queue Merge (EQM) |
68 |
- |
69 |
|
70 |
Deleted: genpatches-2.6/trunk/3.15/1000_linux-3.14.1.patch |
71 |
=================================================================== |
72 |
--- genpatches-2.6/trunk/3.15/1000_linux-3.14.1.patch 2014-05-05 14:22:27 UTC (rev 2771) |
73 |
+++ genpatches-2.6/trunk/3.15/1000_linux-3.14.1.patch 2014-05-05 14:28:48 UTC (rev 2772) |
74 |
@@ -1,1221 +0,0 @@ |
75 |
-diff --git a/Makefile b/Makefile |
76 |
-index e5ac8a62e6e5..7d0b6992d9ed 100644 |
77 |
---- a/Makefile |
78 |
-+++ b/Makefile |
79 |
-@@ -1,6 +1,6 @@ |
80 |
- VERSION = 3 |
81 |
- PATCHLEVEL = 14 |
82 |
--SUBLEVEL = 0 |
83 |
-+SUBLEVEL = 1 |
84 |
- EXTRAVERSION = |
85 |
- NAME = Shuffling Zombie Juror |
86 |
- |
87 |
-diff --git a/arch/arc/boot/dts/nsimosci.dts b/arch/arc/boot/dts/nsimosci.dts |
88 |
-index ea16d782af58..4f31b2eb5cdf 100644 |
89 |
---- a/arch/arc/boot/dts/nsimosci.dts |
90 |
-+++ b/arch/arc/boot/dts/nsimosci.dts |
91 |
-@@ -11,13 +11,16 @@ |
92 |
- |
93 |
- / { |
94 |
- compatible = "snps,nsimosci"; |
95 |
-- clock-frequency = <80000000>; /* 80 MHZ */ |
96 |
-+ clock-frequency = <20000000>; /* 20 MHZ */ |
97 |
- #address-cells = <1>; |
98 |
- #size-cells = <1>; |
99 |
- interrupt-parent = <&intc>; |
100 |
- |
101 |
- chosen { |
102 |
-- bootargs = "console=tty0 consoleblank=0"; |
103 |
-+ /* this is for console on PGU */ |
104 |
-+ /* bootargs = "console=tty0 consoleblank=0"; */ |
105 |
-+ /* this is for console on serial */ |
106 |
-+ bootargs = "earlycon=uart8250,mmio32,0xc0000000,115200n8 console=ttyS0,115200n8 consoleblank=0 debug"; |
107 |
- }; |
108 |
- |
109 |
- aliases { |
110 |
-@@ -44,15 +47,14 @@ |
111 |
- }; |
112 |
- |
113 |
- uart0: serial@c0000000 { |
114 |
-- compatible = "snps,dw-apb-uart"; |
115 |
-+ compatible = "ns8250"; |
116 |
- reg = <0xc0000000 0x2000>; |
117 |
- interrupts = <11>; |
118 |
-- #clock-frequency = <80000000>; |
119 |
- clock-frequency = <3686400>; |
120 |
- baud = <115200>; |
121 |
- reg-shift = <2>; |
122 |
- reg-io-width = <4>; |
123 |
-- status = "okay"; |
124 |
-+ no-loopback-test = <1>; |
125 |
- }; |
126 |
- |
127 |
- pgu0: pgu@c9000000 { |
128 |
-diff --git a/arch/arc/configs/nsimosci_defconfig b/arch/arc/configs/nsimosci_defconfig |
129 |
-index 451af30914f6..c01ba35a4eff 100644 |
130 |
---- a/arch/arc/configs/nsimosci_defconfig |
131 |
-+++ b/arch/arc/configs/nsimosci_defconfig |
132 |
-@@ -54,6 +54,7 @@ CONFIG_SERIO_ARC_PS2=y |
133 |
- CONFIG_SERIAL_8250=y |
134 |
- CONFIG_SERIAL_8250_CONSOLE=y |
135 |
- CONFIG_SERIAL_8250_DW=y |
136 |
-+CONFIG_SERIAL_OF_PLATFORM=y |
137 |
- CONFIG_SERIAL_ARC=y |
138 |
- CONFIG_SERIAL_ARC_CONSOLE=y |
139 |
- # CONFIG_HW_RANDOM is not set |
140 |
-diff --git a/arch/m68k/Kconfig b/arch/m68k/Kconfig |
141 |
-index dbdd2231c75d..b2e322939256 100644 |
142 |
---- a/arch/m68k/Kconfig |
143 |
-+++ b/arch/m68k/Kconfig |
144 |
-@@ -17,6 +17,7 @@ config M68K |
145 |
- select FPU if MMU |
146 |
- select ARCH_WANT_IPC_PARSE_VERSION |
147 |
- select ARCH_USES_GETTIMEOFFSET if MMU && !COLDFIRE |
148 |
-+ select HAVE_FUTEX_CMPXCHG if MMU && FUTEX |
149 |
- select HAVE_MOD_ARCH_SPECIFIC |
150 |
- select MODULES_USE_ELF_REL |
151 |
- select MODULES_USE_ELF_RELA |
152 |
-diff --git a/arch/s390/Kconfig b/arch/s390/Kconfig |
153 |
-index 65a07750f4f9..bb74b21f007a 100644 |
154 |
---- a/arch/s390/Kconfig |
155 |
-+++ b/arch/s390/Kconfig |
156 |
-@@ -117,6 +117,7 @@ config S390 |
157 |
- select HAVE_FUNCTION_GRAPH_TRACER |
158 |
- select HAVE_FUNCTION_TRACER |
159 |
- select HAVE_FUNCTION_TRACE_MCOUNT_TEST |
160 |
-+ select HAVE_FUTEX_CMPXCHG if FUTEX |
161 |
- select HAVE_KERNEL_BZIP2 |
162 |
- select HAVE_KERNEL_GZIP |
163 |
- select HAVE_KERNEL_LZ4 |
164 |
-diff --git a/arch/x86/crypto/ghash-clmulni-intel_asm.S b/arch/x86/crypto/ghash-clmulni-intel_asm.S |
165 |
-index 586f41aac361..185fad49d86f 100644 |
166 |
---- a/arch/x86/crypto/ghash-clmulni-intel_asm.S |
167 |
-+++ b/arch/x86/crypto/ghash-clmulni-intel_asm.S |
168 |
-@@ -24,10 +24,6 @@ |
169 |
- .align 16 |
170 |
- .Lbswap_mask: |
171 |
- .octa 0x000102030405060708090a0b0c0d0e0f |
172 |
--.Lpoly: |
173 |
-- .octa 0xc2000000000000000000000000000001 |
174 |
--.Ltwo_one: |
175 |
-- .octa 0x00000001000000000000000000000001 |
176 |
- |
177 |
- #define DATA %xmm0 |
178 |
- #define SHASH %xmm1 |
179 |
-@@ -134,28 +130,3 @@ ENTRY(clmul_ghash_update) |
180 |
- .Lupdate_just_ret: |
181 |
- ret |
182 |
- ENDPROC(clmul_ghash_update) |
183 |
-- |
184 |
--/* |
185 |
-- * void clmul_ghash_setkey(be128 *shash, const u8 *key); |
186 |
-- * |
187 |
-- * Calculate hash_key << 1 mod poly |
188 |
-- */ |
189 |
--ENTRY(clmul_ghash_setkey) |
190 |
-- movaps .Lbswap_mask, BSWAP |
191 |
-- movups (%rsi), %xmm0 |
192 |
-- PSHUFB_XMM BSWAP %xmm0 |
193 |
-- movaps %xmm0, %xmm1 |
194 |
-- psllq $1, %xmm0 |
195 |
-- psrlq $63, %xmm1 |
196 |
-- movaps %xmm1, %xmm2 |
197 |
-- pslldq $8, %xmm1 |
198 |
-- psrldq $8, %xmm2 |
199 |
-- por %xmm1, %xmm0 |
200 |
-- # reduction |
201 |
-- pshufd $0b00100100, %xmm2, %xmm1 |
202 |
-- pcmpeqd .Ltwo_one, %xmm1 |
203 |
-- pand .Lpoly, %xmm1 |
204 |
-- pxor %xmm1, %xmm0 |
205 |
-- movups %xmm0, (%rdi) |
206 |
-- ret |
207 |
--ENDPROC(clmul_ghash_setkey) |
208 |
-diff --git a/arch/x86/crypto/ghash-clmulni-intel_glue.c b/arch/x86/crypto/ghash-clmulni-intel_glue.c |
209 |
-index 6759dd1135be..d785cf2c529c 100644 |
210 |
---- a/arch/x86/crypto/ghash-clmulni-intel_glue.c |
211 |
-+++ b/arch/x86/crypto/ghash-clmulni-intel_glue.c |
212 |
-@@ -30,8 +30,6 @@ void clmul_ghash_mul(char *dst, const be128 *shash); |
213 |
- void clmul_ghash_update(char *dst, const char *src, unsigned int srclen, |
214 |
- const be128 *shash); |
215 |
- |
216 |
--void clmul_ghash_setkey(be128 *shash, const u8 *key); |
217 |
-- |
218 |
- struct ghash_async_ctx { |
219 |
- struct cryptd_ahash *cryptd_tfm; |
220 |
- }; |
221 |
-@@ -58,13 +56,23 @@ static int ghash_setkey(struct crypto_shash *tfm, |
222 |
- const u8 *key, unsigned int keylen) |
223 |
- { |
224 |
- struct ghash_ctx *ctx = crypto_shash_ctx(tfm); |
225 |
-+ be128 *x = (be128 *)key; |
226 |
-+ u64 a, b; |
227 |
- |
228 |
- if (keylen != GHASH_BLOCK_SIZE) { |
229 |
- crypto_shash_set_flags(tfm, CRYPTO_TFM_RES_BAD_KEY_LEN); |
230 |
- return -EINVAL; |
231 |
- } |
232 |
- |
233 |
-- clmul_ghash_setkey(&ctx->shash, key); |
234 |
-+ /* perform multiplication by 'x' in GF(2^128) */ |
235 |
-+ a = be64_to_cpu(x->a); |
236 |
-+ b = be64_to_cpu(x->b); |
237 |
-+ |
238 |
-+ ctx->shash.a = (__be64)((b << 1) | (a >> 63)); |
239 |
-+ ctx->shash.b = (__be64)((a << 1) | (b >> 63)); |
240 |
-+ |
241 |
-+ if (a >> 63) |
242 |
-+ ctx->shash.b ^= cpu_to_be64(0xc2); |
243 |
- |
244 |
- return 0; |
245 |
- } |
246 |
-diff --git a/arch/x86/include/asm/efi.h b/arch/x86/include/asm/efi.h |
247 |
-index acd86c850414..f949715e3957 100644 |
248 |
---- a/arch/x86/include/asm/efi.h |
249 |
-+++ b/arch/x86/include/asm/efi.h |
250 |
-@@ -130,7 +130,8 @@ extern void efi_memory_uc(u64 addr, unsigned long size); |
251 |
- extern void __init efi_map_region(efi_memory_desc_t *md); |
252 |
- extern void __init efi_map_region_fixed(efi_memory_desc_t *md); |
253 |
- extern void efi_sync_low_kernel_mappings(void); |
254 |
--extern void efi_setup_page_tables(void); |
255 |
-+extern int efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages); |
256 |
-+extern void efi_cleanup_page_tables(unsigned long pa_memmap, unsigned num_pages); |
257 |
- extern void __init old_map_region(efi_memory_desc_t *md); |
258 |
- extern void __init runtime_code_page_mkexec(void); |
259 |
- extern void __init efi_runtime_mkexec(void); |
260 |
-diff --git a/arch/x86/include/asm/pgtable_types.h b/arch/x86/include/asm/pgtable_types.h |
261 |
-index 1aa9ccd43223..94e40f1efdfd 100644 |
262 |
---- a/arch/x86/include/asm/pgtable_types.h |
263 |
-+++ b/arch/x86/include/asm/pgtable_types.h |
264 |
-@@ -385,6 +385,8 @@ extern pte_t *lookup_address(unsigned long address, unsigned int *level); |
265 |
- extern phys_addr_t slow_virt_to_phys(void *__address); |
266 |
- extern int kernel_map_pages_in_pgd(pgd_t *pgd, u64 pfn, unsigned long address, |
267 |
- unsigned numpages, unsigned long page_flags); |
268 |
-+void kernel_unmap_pages_in_pgd(pgd_t *root, unsigned long address, |
269 |
-+ unsigned numpages); |
270 |
- #endif /* !__ASSEMBLY__ */ |
271 |
- |
272 |
- #endif /* _ASM_X86_PGTABLE_DEFS_H */ |
273 |
-diff --git a/arch/x86/mm/pageattr.c b/arch/x86/mm/pageattr.c |
274 |
-index b3b19f46c016..a3488689e301 100644 |
275 |
---- a/arch/x86/mm/pageattr.c |
276 |
-+++ b/arch/x86/mm/pageattr.c |
277 |
-@@ -692,6 +692,18 @@ static bool try_to_free_pmd_page(pmd_t *pmd) |
278 |
- return true; |
279 |
- } |
280 |
- |
281 |
-+static bool try_to_free_pud_page(pud_t *pud) |
282 |
-+{ |
283 |
-+ int i; |
284 |
-+ |
285 |
-+ for (i = 0; i < PTRS_PER_PUD; i++) |
286 |
-+ if (!pud_none(pud[i])) |
287 |
-+ return false; |
288 |
-+ |
289 |
-+ free_page((unsigned long)pud); |
290 |
-+ return true; |
291 |
-+} |
292 |
-+ |
293 |
- static bool unmap_pte_range(pmd_t *pmd, unsigned long start, unsigned long end) |
294 |
- { |
295 |
- pte_t *pte = pte_offset_kernel(pmd, start); |
296 |
-@@ -805,6 +817,16 @@ static void unmap_pud_range(pgd_t *pgd, unsigned long start, unsigned long end) |
297 |
- */ |
298 |
- } |
299 |
- |
300 |
-+static void unmap_pgd_range(pgd_t *root, unsigned long addr, unsigned long end) |
301 |
-+{ |
302 |
-+ pgd_t *pgd_entry = root + pgd_index(addr); |
303 |
-+ |
304 |
-+ unmap_pud_range(pgd_entry, addr, end); |
305 |
-+ |
306 |
-+ if (try_to_free_pud_page((pud_t *)pgd_page_vaddr(*pgd_entry))) |
307 |
-+ pgd_clear(pgd_entry); |
308 |
-+} |
309 |
-+ |
310 |
- static int alloc_pte_page(pmd_t *pmd) |
311 |
- { |
312 |
- pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL | __GFP_NOTRACK); |
313 |
-@@ -999,9 +1021,8 @@ static int populate_pud(struct cpa_data *cpa, unsigned long start, pgd_t *pgd, |
314 |
- static int populate_pgd(struct cpa_data *cpa, unsigned long addr) |
315 |
- { |
316 |
- pgprot_t pgprot = __pgprot(_KERNPG_TABLE); |
317 |
-- bool allocd_pgd = false; |
318 |
-- pgd_t *pgd_entry; |
319 |
- pud_t *pud = NULL; /* shut up gcc */ |
320 |
-+ pgd_t *pgd_entry; |
321 |
- int ret; |
322 |
- |
323 |
- pgd_entry = cpa->pgd + pgd_index(addr); |
324 |
-@@ -1015,7 +1036,6 @@ static int populate_pgd(struct cpa_data *cpa, unsigned long addr) |
325 |
- return -1; |
326 |
- |
327 |
- set_pgd(pgd_entry, __pgd(__pa(pud) | _KERNPG_TABLE)); |
328 |
-- allocd_pgd = true; |
329 |
- } |
330 |
- |
331 |
- pgprot_val(pgprot) &= ~pgprot_val(cpa->mask_clr); |
332 |
-@@ -1023,19 +1043,11 @@ static int populate_pgd(struct cpa_data *cpa, unsigned long addr) |
333 |
- |
334 |
- ret = populate_pud(cpa, addr, pgd_entry, pgprot); |
335 |
- if (ret < 0) { |
336 |
-- unmap_pud_range(pgd_entry, addr, |
337 |
-+ unmap_pgd_range(cpa->pgd, addr, |
338 |
- addr + (cpa->numpages << PAGE_SHIFT)); |
339 |
-- |
340 |
-- if (allocd_pgd) { |
341 |
-- /* |
342 |
-- * If I allocated this PUD page, I can just as well |
343 |
-- * free it in this error path. |
344 |
-- */ |
345 |
-- pgd_clear(pgd_entry); |
346 |
-- free_page((unsigned long)pud); |
347 |
-- } |
348 |
- return ret; |
349 |
- } |
350 |
-+ |
351 |
- cpa->numpages = ret; |
352 |
- return 0; |
353 |
- } |
354 |
-@@ -1861,6 +1873,12 @@ out: |
355 |
- return retval; |
356 |
- } |
357 |
- |
358 |
-+void kernel_unmap_pages_in_pgd(pgd_t *root, unsigned long address, |
359 |
-+ unsigned numpages) |
360 |
-+{ |
361 |
-+ unmap_pgd_range(root, address, address + (numpages << PAGE_SHIFT)); |
362 |
-+} |
363 |
-+ |
364 |
- /* |
365 |
- * The testcases use internal knowledge of the implementation that shouldn't |
366 |
- * be exposed to the rest of the kernel. Include these directly here. |
367 |
-diff --git a/arch/x86/platform/efi/efi.c b/arch/x86/platform/efi/efi.c |
368 |
-index b97acecf3fd9..abb81b0ad83f 100644 |
369 |
---- a/arch/x86/platform/efi/efi.c |
370 |
-+++ b/arch/x86/platform/efi/efi.c |
371 |
-@@ -939,14 +939,36 @@ static void __init efi_map_regions_fixed(void) |
372 |
- |
373 |
- } |
374 |
- |
375 |
-+static void *realloc_pages(void *old_memmap, int old_shift) |
376 |
-+{ |
377 |
-+ void *ret; |
378 |
-+ |
379 |
-+ ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1); |
380 |
-+ if (!ret) |
381 |
-+ goto out; |
382 |
-+ |
383 |
-+ /* |
384 |
-+ * A first-time allocation doesn't have anything to copy. |
385 |
-+ */ |
386 |
-+ if (!old_memmap) |
387 |
-+ return ret; |
388 |
-+ |
389 |
-+ memcpy(ret, old_memmap, PAGE_SIZE << old_shift); |
390 |
-+ |
391 |
-+out: |
392 |
-+ free_pages((unsigned long)old_memmap, old_shift); |
393 |
-+ return ret; |
394 |
-+} |
395 |
-+ |
396 |
- /* |
397 |
-- * Map efi memory ranges for runtime serivce and update new_memmap with virtual |
398 |
-- * addresses. |
399 |
-+ * Map the efi memory ranges of the runtime services and update new_mmap with |
400 |
-+ * virtual addresses. |
401 |
- */ |
402 |
--static void * __init efi_map_regions(int *count) |
403 |
-+static void * __init efi_map_regions(int *count, int *pg_shift) |
404 |
- { |
405 |
-+ void *p, *new_memmap = NULL; |
406 |
-+ unsigned long left = 0; |
407 |
- efi_memory_desc_t *md; |
408 |
-- void *p, *tmp, *new_memmap = NULL; |
409 |
- |
410 |
- for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) { |
411 |
- md = p; |
412 |
-@@ -961,20 +983,23 @@ static void * __init efi_map_regions(int *count) |
413 |
- efi_map_region(md); |
414 |
- get_systab_virt_addr(md); |
415 |
- |
416 |
-- tmp = krealloc(new_memmap, (*count + 1) * memmap.desc_size, |
417 |
-- GFP_KERNEL); |
418 |
-- if (!tmp) |
419 |
-- goto out; |
420 |
-- new_memmap = tmp; |
421 |
-+ if (left < memmap.desc_size) { |
422 |
-+ new_memmap = realloc_pages(new_memmap, *pg_shift); |
423 |
-+ if (!new_memmap) |
424 |
-+ return NULL; |
425 |
-+ |
426 |
-+ left += PAGE_SIZE << *pg_shift; |
427 |
-+ (*pg_shift)++; |
428 |
-+ } |
429 |
-+ |
430 |
- memcpy(new_memmap + (*count * memmap.desc_size), md, |
431 |
- memmap.desc_size); |
432 |
-+ |
433 |
-+ left -= memmap.desc_size; |
434 |
- (*count)++; |
435 |
- } |
436 |
- |
437 |
- return new_memmap; |
438 |
--out: |
439 |
-- kfree(new_memmap); |
440 |
-- return NULL; |
441 |
- } |
442 |
- |
443 |
- /* |
444 |
-@@ -1000,9 +1025,9 @@ out: |
445 |
- */ |
446 |
- void __init efi_enter_virtual_mode(void) |
447 |
- { |
448 |
-- efi_status_t status; |
449 |
-+ int err, count = 0, pg_shift = 0; |
450 |
- void *new_memmap = NULL; |
451 |
-- int err, count = 0; |
452 |
-+ efi_status_t status; |
453 |
- |
454 |
- efi.systab = NULL; |
455 |
- |
456 |
-@@ -1019,20 +1044,24 @@ void __init efi_enter_virtual_mode(void) |
457 |
- efi_map_regions_fixed(); |
458 |
- } else { |
459 |
- efi_merge_regions(); |
460 |
-- new_memmap = efi_map_regions(&count); |
461 |
-+ new_memmap = efi_map_regions(&count, &pg_shift); |
462 |
- if (!new_memmap) { |
463 |
- pr_err("Error reallocating memory, EFI runtime non-functional!\n"); |
464 |
- return; |
465 |
- } |
466 |
-- } |
467 |
- |
468 |
-- err = save_runtime_map(); |
469 |
-- if (err) |
470 |
-- pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n"); |
471 |
-+ err = save_runtime_map(); |
472 |
-+ if (err) |
473 |
-+ pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n"); |
474 |
-+ } |
475 |
- |
476 |
- BUG_ON(!efi.systab); |
477 |
- |
478 |
-- efi_setup_page_tables(); |
479 |
-+ if (!efi_setup) { |
480 |
-+ if (efi_setup_page_tables(__pa(new_memmap), 1 << pg_shift)) |
481 |
-+ return; |
482 |
-+ } |
483 |
-+ |
484 |
- efi_sync_low_kernel_mappings(); |
485 |
- |
486 |
- if (!efi_setup) { |
487 |
-@@ -1072,7 +1101,35 @@ void __init efi_enter_virtual_mode(void) |
488 |
- |
489 |
- efi_runtime_mkexec(); |
490 |
- |
491 |
-- kfree(new_memmap); |
492 |
-+ |
493 |
-+ /* |
494 |
-+ * We mapped the descriptor array into the EFI pagetable above but we're |
495 |
-+ * not unmapping it here. Here's why: |
496 |
-+ * |
497 |
-+ * We're copying select PGDs from the kernel page table to the EFI page |
498 |
-+ * table and when we do so and make changes to those PGDs like unmapping |
499 |
-+ * stuff from them, those changes appear in the kernel page table and we |
500 |
-+ * go boom. |
501 |
-+ * |
502 |
-+ * From setup_real_mode(): |
503 |
-+ * |
504 |
-+ * ... |
505 |
-+ * trampoline_pgd[0] = init_level4_pgt[pgd_index(__PAGE_OFFSET)].pgd; |
506 |
-+ * |
507 |
-+ * In this particular case, our allocation is in PGD 0 of the EFI page |
508 |
-+ * table but we've copied that PGD from PGD[272] of the EFI page table: |
509 |
-+ * |
510 |
-+ * pgd_index(__PAGE_OFFSET = 0xffff880000000000) = 272 |
511 |
-+ * |
512 |
-+ * where the direct memory mapping in kernel space is. |
513 |
-+ * |
514 |
-+ * new_memmap's VA comes from that direct mapping and thus clearing it, |
515 |
-+ * it would get cleared in the kernel page table too. |
516 |
-+ * |
517 |
-+ * efi_cleanup_page_tables(__pa(new_memmap), 1 << pg_shift); |
518 |
-+ */ |
519 |
-+ if (!efi_setup) |
520 |
-+ free_pages((unsigned long)new_memmap, pg_shift); |
521 |
- |
522 |
- /* clean DUMMY object */ |
523 |
- efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID, |
524 |
-diff --git a/arch/x86/platform/efi/efi_32.c b/arch/x86/platform/efi/efi_32.c |
525 |
-index 0b74cdf7f816..9ee3491e31fb 100644 |
526 |
---- a/arch/x86/platform/efi/efi_32.c |
527 |
-+++ b/arch/x86/platform/efi/efi_32.c |
528 |
-@@ -40,7 +40,12 @@ |
529 |
- static unsigned long efi_rt_eflags; |
530 |
- |
531 |
- void efi_sync_low_kernel_mappings(void) {} |
532 |
--void efi_setup_page_tables(void) {} |
533 |
-+void __init efi_dump_pagetable(void) {} |
534 |
-+int efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages) |
535 |
-+{ |
536 |
-+ return 0; |
537 |
-+} |
538 |
-+void efi_cleanup_page_tables(unsigned long pa_memmap, unsigned num_pages) {} |
539 |
- |
540 |
- void __init efi_map_region(efi_memory_desc_t *md) |
541 |
- { |
542 |
-diff --git a/arch/x86/platform/efi/efi_64.c b/arch/x86/platform/efi/efi_64.c |
543 |
-index 0c2a234fef1e..666b74a09092 100644 |
544 |
---- a/arch/x86/platform/efi/efi_64.c |
545 |
-+++ b/arch/x86/platform/efi/efi_64.c |
546 |
-@@ -137,12 +137,38 @@ void efi_sync_low_kernel_mappings(void) |
547 |
- sizeof(pgd_t) * num_pgds); |
548 |
- } |
549 |
- |
550 |
--void efi_setup_page_tables(void) |
551 |
-+int efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages) |
552 |
- { |
553 |
-+ pgd_t *pgd; |
554 |
-+ |
555 |
-+ if (efi_enabled(EFI_OLD_MEMMAP)) |
556 |
-+ return 0; |
557 |
-+ |
558 |
- efi_scratch.efi_pgt = (pgd_t *)(unsigned long)real_mode_header->trampoline_pgd; |
559 |
-+ pgd = __va(efi_scratch.efi_pgt); |
560 |
- |
561 |
-- if (!efi_enabled(EFI_OLD_MEMMAP)) |
562 |
-- efi_scratch.use_pgd = true; |
563 |
-+ /* |
564 |
-+ * It can happen that the physical address of new_memmap lands in memory |
565 |
-+ * which is not mapped in the EFI page table. Therefore we need to go |
566 |
-+ * and ident-map those pages containing the map before calling |
567 |
-+ * phys_efi_set_virtual_address_map(). |
568 |
-+ */ |
569 |
-+ if (kernel_map_pages_in_pgd(pgd, pa_memmap, pa_memmap, num_pages, _PAGE_NX)) { |
570 |
-+ pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap); |
571 |
-+ return 1; |
572 |
-+ } |
573 |
-+ |
574 |
-+ efi_scratch.use_pgd = true; |
575 |
-+ |
576 |
-+ |
577 |
-+ return 0; |
578 |
-+} |
579 |
-+ |
580 |
-+void efi_cleanup_page_tables(unsigned long pa_memmap, unsigned num_pages) |
581 |
-+{ |
582 |
-+ pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd); |
583 |
-+ |
584 |
-+ kernel_unmap_pages_in_pgd(pgd, pa_memmap, num_pages); |
585 |
- } |
586 |
- |
587 |
- static void __init __map_region(efi_memory_desc_t *md, u64 va) |
588 |
-diff --git a/drivers/isdn/isdnloop/isdnloop.c b/drivers/isdn/isdnloop/isdnloop.c |
589 |
-index 02125e6a9109..5a4da94aefb0 100644 |
590 |
---- a/drivers/isdn/isdnloop/isdnloop.c |
591 |
-+++ b/drivers/isdn/isdnloop/isdnloop.c |
592 |
-@@ -518,9 +518,9 @@ static isdnloop_stat isdnloop_cmd_table[] = |
593 |
- static void |
594 |
- isdnloop_fake_err(isdnloop_card *card) |
595 |
- { |
596 |
-- char buf[60]; |
597 |
-+ char buf[64]; |
598 |
- |
599 |
-- sprintf(buf, "E%s", card->omsg); |
600 |
-+ snprintf(buf, sizeof(buf), "E%s", card->omsg); |
601 |
- isdnloop_fake(card, buf, -1); |
602 |
- isdnloop_fake(card, "NAK", -1); |
603 |
- } |
604 |
-@@ -903,6 +903,8 @@ isdnloop_parse_cmd(isdnloop_card *card) |
605 |
- case 7: |
606 |
- /* 0x;EAZ */ |
607 |
- p += 3; |
608 |
-+ if (strlen(p) >= sizeof(card->eazlist[0])) |
609 |
-+ break; |
610 |
- strcpy(card->eazlist[ch - 1], p); |
611 |
- break; |
612 |
- case 8: |
613 |
-@@ -1070,6 +1072,12 @@ isdnloop_start(isdnloop_card *card, isdnloop_sdef *sdefp) |
614 |
- return -EBUSY; |
615 |
- if (copy_from_user((char *) &sdef, (char *) sdefp, sizeof(sdef))) |
616 |
- return -EFAULT; |
617 |
-+ |
618 |
-+ for (i = 0; i < 3; i++) { |
619 |
-+ if (!memchr(sdef.num[i], 0, sizeof(sdef.num[i]))) |
620 |
-+ return -EINVAL; |
621 |
-+ } |
622 |
-+ |
623 |
- spin_lock_irqsave(&card->isdnloop_lock, flags); |
624 |
- switch (sdef.ptype) { |
625 |
- case ISDN_PTYPE_EURO: |
626 |
-@@ -1127,7 +1135,7 @@ isdnloop_command(isdn_ctrl *c, isdnloop_card *card) |
627 |
- { |
628 |
- ulong a; |
629 |
- int i; |
630 |
-- char cbuf[60]; |
631 |
-+ char cbuf[80]; |
632 |
- isdn_ctrl cmd; |
633 |
- isdnloop_cdef cdef; |
634 |
- |
635 |
-@@ -1192,7 +1200,6 @@ isdnloop_command(isdn_ctrl *c, isdnloop_card *card) |
636 |
- break; |
637 |
- if ((c->arg & 255) < ISDNLOOP_BCH) { |
638 |
- char *p; |
639 |
-- char dial[50]; |
640 |
- char dcode[4]; |
641 |
- |
642 |
- a = c->arg; |
643 |
-@@ -1204,10 +1211,10 @@ isdnloop_command(isdn_ctrl *c, isdnloop_card *card) |
644 |
- } else |
645 |
- /* Normal Dial */ |
646 |
- strcpy(dcode, "CAL"); |
647 |
-- strcpy(dial, p); |
648 |
-- sprintf(cbuf, "%02d;D%s_R%s,%02d,%02d,%s\n", (int) (a + 1), |
649 |
-- dcode, dial, c->parm.setup.si1, |
650 |
-- c->parm.setup.si2, c->parm.setup.eazmsn); |
651 |
-+ snprintf(cbuf, sizeof(cbuf), |
652 |
-+ "%02d;D%s_R%s,%02d,%02d,%s\n", (int) (a + 1), |
653 |
-+ dcode, p, c->parm.setup.si1, |
654 |
-+ c->parm.setup.si2, c->parm.setup.eazmsn); |
655 |
- i = isdnloop_writecmd(cbuf, strlen(cbuf), 0, card); |
656 |
- } |
657 |
- break; |
658 |
-diff --git a/drivers/net/ethernet/cadence/at91_ether.c b/drivers/net/ethernet/cadence/at91_ether.c |
659 |
-index ce75de9bae9e..4a79edaf3885 100644 |
660 |
---- a/drivers/net/ethernet/cadence/at91_ether.c |
661 |
-+++ b/drivers/net/ethernet/cadence/at91_ether.c |
662 |
-@@ -342,6 +342,9 @@ static int __init at91ether_probe(struct platform_device *pdev) |
663 |
- } |
664 |
- clk_enable(lp->pclk); |
665 |
- |
666 |
-+ lp->hclk = ERR_PTR(-ENOENT); |
667 |
-+ lp->tx_clk = ERR_PTR(-ENOENT); |
668 |
-+ |
669 |
- /* Install the interrupt handler */ |
670 |
- dev->irq = platform_get_irq(pdev, 0); |
671 |
- res = devm_request_irq(&pdev->dev, dev->irq, at91ether_interrupt, 0, dev->name, dev); |
672 |
-diff --git a/drivers/net/ethernet/sfc/ef10.c b/drivers/net/ethernet/sfc/ef10.c |
673 |
-index 174a92f5fe51..7645a3ce3854 100644 |
674 |
---- a/drivers/net/ethernet/sfc/ef10.c |
675 |
-+++ b/drivers/net/ethernet/sfc/ef10.c |
676 |
-@@ -565,10 +565,17 @@ static int efx_ef10_dimension_resources(struct efx_nic *efx) |
677 |
- * several of each (in fact that's the only option if host |
678 |
- * page size is >4K). So we may allocate some extra VIs just |
679 |
- * for writing PIO buffers through. |
680 |
-+ * |
681 |
-+ * The UC mapping contains (min_vis - 1) complete VIs and the |
682 |
-+ * first half of the next VI. Then the WC mapping begins with |
683 |
-+ * the second half of this last VI. |
684 |
- */ |
685 |
- uc_mem_map_size = PAGE_ALIGN((min_vis - 1) * EFX_VI_PAGE_SIZE + |
686 |
- ER_DZ_TX_PIOBUF); |
687 |
- if (nic_data->n_piobufs) { |
688 |
-+ /* pio_write_vi_base rounds down to give the number of complete |
689 |
-+ * VIs inside the UC mapping. |
690 |
-+ */ |
691 |
- pio_write_vi_base = uc_mem_map_size / EFX_VI_PAGE_SIZE; |
692 |
- wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base + |
693 |
- nic_data->n_piobufs) * |
694 |
-diff --git a/drivers/net/ethernet/sfc/efx.c b/drivers/net/ethernet/sfc/efx.c |
695 |
-index 83d464347021..f06c790fba5a 100644 |
696 |
---- a/drivers/net/ethernet/sfc/efx.c |
697 |
-+++ b/drivers/net/ethernet/sfc/efx.c |
698 |
-@@ -1603,6 +1603,8 @@ static int efx_probe_nic(struct efx_nic *efx) |
699 |
- if (rc) |
700 |
- goto fail1; |
701 |
- |
702 |
-+ efx_set_channels(efx); |
703 |
-+ |
704 |
- rc = efx->type->dimension_resources(efx); |
705 |
- if (rc) |
706 |
- goto fail2; |
707 |
-@@ -1613,7 +1615,6 @@ static int efx_probe_nic(struct efx_nic *efx) |
708 |
- efx->rx_indir_table[i] = |
709 |
- ethtool_rxfh_indir_default(i, efx->rss_spread); |
710 |
- |
711 |
-- efx_set_channels(efx); |
712 |
- netif_set_real_num_tx_queues(efx->net_dev, efx->n_tx_channels); |
713 |
- netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels); |
714 |
- |
715 |
-diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c |
716 |
-index 1236812c7be6..d091e52b00e1 100644 |
717 |
---- a/drivers/net/vxlan.c |
718 |
-+++ b/drivers/net/vxlan.c |
719 |
-@@ -871,6 +871,9 @@ static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], |
720 |
- if (err) |
721 |
- return err; |
722 |
- |
723 |
-+ if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family) |
724 |
-+ return -EAFNOSUPPORT; |
725 |
-+ |
726 |
- spin_lock_bh(&vxlan->hash_lock); |
727 |
- err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags, |
728 |
- port, vni, ifindex, ndm->ndm_flags); |
729 |
-@@ -2612,9 +2615,10 @@ static int vxlan_newlink(struct net *net, struct net_device *dev, |
730 |
- vni = nla_get_u32(data[IFLA_VXLAN_ID]); |
731 |
- dst->remote_vni = vni; |
732 |
- |
733 |
-+ /* Unless IPv6 is explicitly requested, assume IPv4 */ |
734 |
-+ dst->remote_ip.sa.sa_family = AF_INET; |
735 |
- if (data[IFLA_VXLAN_GROUP]) { |
736 |
- dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]); |
737 |
-- dst->remote_ip.sa.sa_family = AF_INET; |
738 |
- } else if (data[IFLA_VXLAN_GROUP6]) { |
739 |
- if (!IS_ENABLED(CONFIG_IPV6)) |
740 |
- return -EPFNOSUPPORT; |
741 |
-diff --git a/drivers/net/wireless/iwlwifi/mvm/rs.c b/drivers/net/wireless/iwlwifi/mvm/rs.c |
742 |
-index 6abf74e1351f..5bc871513505 100644 |
743 |
---- a/drivers/net/wireless/iwlwifi/mvm/rs.c |
744 |
-+++ b/drivers/net/wireless/iwlwifi/mvm/rs.c |
745 |
-@@ -211,9 +211,9 @@ static const struct rs_tx_column rs_tx_columns[] = { |
746 |
- .next_columns = { |
747 |
- RS_COLUMN_LEGACY_ANT_B, |
748 |
- RS_COLUMN_SISO_ANT_A, |
749 |
-+ RS_COLUMN_SISO_ANT_B, |
750 |
- RS_COLUMN_MIMO2, |
751 |
-- RS_COLUMN_INVALID, |
752 |
-- RS_COLUMN_INVALID, |
753 |
-+ RS_COLUMN_MIMO2_SGI, |
754 |
- }, |
755 |
- }, |
756 |
- [RS_COLUMN_LEGACY_ANT_B] = { |
757 |
-@@ -221,10 +221,10 @@ static const struct rs_tx_column rs_tx_columns[] = { |
758 |
- .ant = ANT_B, |
759 |
- .next_columns = { |
760 |
- RS_COLUMN_LEGACY_ANT_A, |
761 |
-+ RS_COLUMN_SISO_ANT_A, |
762 |
- RS_COLUMN_SISO_ANT_B, |
763 |
- RS_COLUMN_MIMO2, |
764 |
-- RS_COLUMN_INVALID, |
765 |
-- RS_COLUMN_INVALID, |
766 |
-+ RS_COLUMN_MIMO2_SGI, |
767 |
- }, |
768 |
- }, |
769 |
- [RS_COLUMN_SISO_ANT_A] = { |
770 |
-@@ -234,8 +234,8 @@ static const struct rs_tx_column rs_tx_columns[] = { |
771 |
- RS_COLUMN_SISO_ANT_B, |
772 |
- RS_COLUMN_MIMO2, |
773 |
- RS_COLUMN_SISO_ANT_A_SGI, |
774 |
-- RS_COLUMN_INVALID, |
775 |
-- RS_COLUMN_INVALID, |
776 |
-+ RS_COLUMN_SISO_ANT_B_SGI, |
777 |
-+ RS_COLUMN_MIMO2_SGI, |
778 |
- }, |
779 |
- .checks = { |
780 |
- rs_siso_allow, |
781 |
-@@ -248,8 +248,8 @@ static const struct rs_tx_column rs_tx_columns[] = { |
782 |
- RS_COLUMN_SISO_ANT_A, |
783 |
- RS_COLUMN_MIMO2, |
784 |
- RS_COLUMN_SISO_ANT_B_SGI, |
785 |
-- RS_COLUMN_INVALID, |
786 |
-- RS_COLUMN_INVALID, |
787 |
-+ RS_COLUMN_SISO_ANT_A_SGI, |
788 |
-+ RS_COLUMN_MIMO2_SGI, |
789 |
- }, |
790 |
- .checks = { |
791 |
- rs_siso_allow, |
792 |
-@@ -263,8 +263,8 @@ static const struct rs_tx_column rs_tx_columns[] = { |
793 |
- RS_COLUMN_SISO_ANT_B_SGI, |
794 |
- RS_COLUMN_MIMO2_SGI, |
795 |
- RS_COLUMN_SISO_ANT_A, |
796 |
-- RS_COLUMN_INVALID, |
797 |
-- RS_COLUMN_INVALID, |
798 |
-+ RS_COLUMN_SISO_ANT_B, |
799 |
-+ RS_COLUMN_MIMO2, |
800 |
- }, |
801 |
- .checks = { |
802 |
- rs_siso_allow, |
803 |
-@@ -279,8 +279,8 @@ static const struct rs_tx_column rs_tx_columns[] = { |
804 |
- RS_COLUMN_SISO_ANT_A_SGI, |
805 |
- RS_COLUMN_MIMO2_SGI, |
806 |
- RS_COLUMN_SISO_ANT_B, |
807 |
-- RS_COLUMN_INVALID, |
808 |
-- RS_COLUMN_INVALID, |
809 |
-+ RS_COLUMN_SISO_ANT_A, |
810 |
-+ RS_COLUMN_MIMO2, |
811 |
- }, |
812 |
- .checks = { |
813 |
- rs_siso_allow, |
814 |
-@@ -292,10 +292,10 @@ static const struct rs_tx_column rs_tx_columns[] = { |
815 |
- .ant = ANT_AB, |
816 |
- .next_columns = { |
817 |
- RS_COLUMN_SISO_ANT_A, |
818 |
-+ RS_COLUMN_SISO_ANT_B, |
819 |
-+ RS_COLUMN_SISO_ANT_A_SGI, |
820 |
-+ RS_COLUMN_SISO_ANT_B_SGI, |
821 |
- RS_COLUMN_MIMO2_SGI, |
822 |
-- RS_COLUMN_INVALID, |
823 |
-- RS_COLUMN_INVALID, |
824 |
-- RS_COLUMN_INVALID, |
825 |
- }, |
826 |
- .checks = { |
827 |
- rs_mimo_allow, |
828 |
-@@ -307,10 +307,10 @@ static const struct rs_tx_column rs_tx_columns[] = { |
829 |
- .sgi = true, |
830 |
- .next_columns = { |
831 |
- RS_COLUMN_SISO_ANT_A_SGI, |
832 |
-+ RS_COLUMN_SISO_ANT_B_SGI, |
833 |
-+ RS_COLUMN_SISO_ANT_A, |
834 |
-+ RS_COLUMN_SISO_ANT_B, |
835 |
- RS_COLUMN_MIMO2, |
836 |
-- RS_COLUMN_INVALID, |
837 |
-- RS_COLUMN_INVALID, |
838 |
-- RS_COLUMN_INVALID, |
839 |
- }, |
840 |
- .checks = { |
841 |
- rs_mimo_allow, |
842 |
-diff --git a/drivers/net/xen-netback/common.h b/drivers/net/xen-netback/common.h |
843 |
-index ae413a2cbee7..4bf5b334664e 100644 |
844 |
---- a/drivers/net/xen-netback/common.h |
845 |
-+++ b/drivers/net/xen-netback/common.h |
846 |
-@@ -113,6 +113,11 @@ struct xenvif { |
847 |
- domid_t domid; |
848 |
- unsigned int handle; |
849 |
- |
850 |
-+ /* Is this interface disabled? True when backend discovers |
851 |
-+ * frontend is rogue. |
852 |
-+ */ |
853 |
-+ bool disabled; |
854 |
-+ |
855 |
- /* Use NAPI for guest TX */ |
856 |
- struct napi_struct napi; |
857 |
- /* When feature-split-event-channels = 0, tx_irq = rx_irq. */ |
858 |
-diff --git a/drivers/net/xen-netback/interface.c b/drivers/net/xen-netback/interface.c |
859 |
-index 301cc037fda8..2e92d52c0a6d 100644 |
860 |
---- a/drivers/net/xen-netback/interface.c |
861 |
-+++ b/drivers/net/xen-netback/interface.c |
862 |
-@@ -62,6 +62,15 @@ static int xenvif_poll(struct napi_struct *napi, int budget) |
863 |
- struct xenvif *vif = container_of(napi, struct xenvif, napi); |
864 |
- int work_done; |
865 |
- |
866 |
-+ /* This vif is rogue, we pretend we've there is nothing to do |
867 |
-+ * for this vif to deschedule it from NAPI. But this interface |
868 |
-+ * will be turned off in thread context later. |
869 |
-+ */ |
870 |
-+ if (unlikely(vif->disabled)) { |
871 |
-+ napi_complete(napi); |
872 |
-+ return 0; |
873 |
-+ } |
874 |
-+ |
875 |
- work_done = xenvif_tx_action(vif, budget); |
876 |
- |
877 |
- if (work_done < budget) { |
878 |
-@@ -321,6 +330,8 @@ struct xenvif *xenvif_alloc(struct device *parent, domid_t domid, |
879 |
- vif->ip_csum = 1; |
880 |
- vif->dev = dev; |
881 |
- |
882 |
-+ vif->disabled = false; |
883 |
-+ |
884 |
- vif->credit_bytes = vif->remaining_credit = ~0UL; |
885 |
- vif->credit_usec = 0UL; |
886 |
- init_timer(&vif->credit_timeout); |
887 |
-diff --git a/drivers/net/xen-netback/netback.c b/drivers/net/xen-netback/netback.c |
888 |
-index 438d0c09b7e6..97030c193afd 100644 |
889 |
---- a/drivers/net/xen-netback/netback.c |
890 |
-+++ b/drivers/net/xen-netback/netback.c |
891 |
-@@ -192,8 +192,8 @@ static bool start_new_rx_buffer(int offset, unsigned long size, int head) |
892 |
- * into multiple copies tend to give large frags their |
893 |
- * own buffers as before. |
894 |
- */ |
895 |
-- if ((offset + size > MAX_BUFFER_OFFSET) && |
896 |
-- (size <= MAX_BUFFER_OFFSET) && offset && !head) |
897 |
-+ BUG_ON(size > MAX_BUFFER_OFFSET); |
898 |
-+ if ((offset + size > MAX_BUFFER_OFFSET) && offset && !head) |
899 |
- return true; |
900 |
- |
901 |
- return false; |
902 |
-@@ -482,6 +482,8 @@ static void xenvif_rx_action(struct xenvif *vif) |
903 |
- |
904 |
- while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { |
905 |
- RING_IDX max_slots_needed; |
906 |
-+ RING_IDX old_req_cons; |
907 |
-+ RING_IDX ring_slots_used; |
908 |
- int i; |
909 |
- |
910 |
- /* We need a cheap worse case estimate for the number of |
911 |
-@@ -493,9 +495,28 @@ static void xenvif_rx_action(struct xenvif *vif) |
912 |
- PAGE_SIZE); |
913 |
- for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
914 |
- unsigned int size; |
915 |
-+ unsigned int offset; |
916 |
-+ |
917 |
- size = skb_frag_size(&skb_shinfo(skb)->frags[i]); |
918 |
-- max_slots_needed += DIV_ROUND_UP(size, PAGE_SIZE); |
919 |
-+ offset = skb_shinfo(skb)->frags[i].page_offset; |
920 |
-+ |
921 |
-+ /* For a worse-case estimate we need to factor in |
922 |
-+ * the fragment page offset as this will affect the |
923 |
-+ * number of times xenvif_gop_frag_copy() will |
924 |
-+ * call start_new_rx_buffer(). |
925 |
-+ */ |
926 |
-+ max_slots_needed += DIV_ROUND_UP(offset + size, |
927 |
-+ PAGE_SIZE); |
928 |
- } |
929 |
-+ |
930 |
-+ /* To avoid the estimate becoming too pessimal for some |
931 |
-+ * frontends that limit posted rx requests, cap the estimate |
932 |
-+ * at MAX_SKB_FRAGS. |
933 |
-+ */ |
934 |
-+ if (max_slots_needed > MAX_SKB_FRAGS) |
935 |
-+ max_slots_needed = MAX_SKB_FRAGS; |
936 |
-+ |
937 |
-+ /* We may need one more slot for GSO metadata */ |
938 |
- if (skb_is_gso(skb) && |
939 |
- (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || |
940 |
- skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)) |
941 |
-@@ -511,8 +532,12 @@ static void xenvif_rx_action(struct xenvif *vif) |
942 |
- vif->rx_last_skb_slots = 0; |
943 |
- |
944 |
- sco = (struct skb_cb_overlay *)skb->cb; |
945 |
-+ |
946 |
-+ old_req_cons = vif->rx.req_cons; |
947 |
- sco->meta_slots_used = xenvif_gop_skb(skb, &npo); |
948 |
-- BUG_ON(sco->meta_slots_used > max_slots_needed); |
949 |
-+ ring_slots_used = vif->rx.req_cons - old_req_cons; |
950 |
-+ |
951 |
-+ BUG_ON(ring_slots_used > max_slots_needed); |
952 |
- |
953 |
- __skb_queue_tail(&rxq, skb); |
954 |
- } |
955 |
-@@ -655,7 +680,8 @@ static void xenvif_tx_err(struct xenvif *vif, |
956 |
- static void xenvif_fatal_tx_err(struct xenvif *vif) |
957 |
- { |
958 |
- netdev_err(vif->dev, "fatal error; disabling device\n"); |
959 |
-- xenvif_carrier_off(vif); |
960 |
-+ vif->disabled = true; |
961 |
-+ xenvif_kick_thread(vif); |
962 |
- } |
963 |
- |
964 |
- static int xenvif_count_requests(struct xenvif *vif, |
965 |
-@@ -1126,7 +1152,7 @@ static unsigned xenvif_tx_build_gops(struct xenvif *vif, int budget) |
966 |
- vif->tx.sring->req_prod, vif->tx.req_cons, |
967 |
- XEN_NETIF_TX_RING_SIZE); |
968 |
- xenvif_fatal_tx_err(vif); |
969 |
-- continue; |
970 |
-+ break; |
971 |
- } |
972 |
- |
973 |
- work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&vif->tx); |
974 |
-@@ -1548,7 +1574,18 @@ int xenvif_kthread(void *data) |
975 |
- while (!kthread_should_stop()) { |
976 |
- wait_event_interruptible(vif->wq, |
977 |
- rx_work_todo(vif) || |
978 |
-+ vif->disabled || |
979 |
- kthread_should_stop()); |
980 |
-+ |
981 |
-+ /* This frontend is found to be rogue, disable it in |
982 |
-+ * kthread context. Currently this is only set when |
983 |
-+ * netback finds out frontend sends malformed packet, |
984 |
-+ * but we cannot disable the interface in softirq |
985 |
-+ * context so we defer it here. |
986 |
-+ */ |
987 |
-+ if (unlikely(vif->disabled && netif_carrier_ok(vif->dev))) |
988 |
-+ xenvif_carrier_off(vif); |
989 |
-+ |
990 |
- if (kthread_should_stop()) |
991 |
- break; |
992 |
- |
993 |
-diff --git a/include/linux/futex.h b/include/linux/futex.h |
994 |
-index b0d95cac826e..6435f46d6e13 100644 |
995 |
---- a/include/linux/futex.h |
996 |
-+++ b/include/linux/futex.h |
997 |
-@@ -55,7 +55,11 @@ union futex_key { |
998 |
- #ifdef CONFIG_FUTEX |
999 |
- extern void exit_robust_list(struct task_struct *curr); |
1000 |
- extern void exit_pi_state_list(struct task_struct *curr); |
1001 |
-+#ifdef CONFIG_HAVE_FUTEX_CMPXCHG |
1002 |
-+#define futex_cmpxchg_enabled 1 |
1003 |
-+#else |
1004 |
- extern int futex_cmpxchg_enabled; |
1005 |
-+#endif |
1006 |
- #else |
1007 |
- static inline void exit_robust_list(struct task_struct *curr) |
1008 |
- { |
1009 |
-diff --git a/init/Kconfig b/init/Kconfig |
1010 |
-index 009a797dd242..d56cb03c1b49 100644 |
1011 |
---- a/init/Kconfig |
1012 |
-+++ b/init/Kconfig |
1013 |
-@@ -1387,6 +1387,13 @@ config FUTEX |
1014 |
- support for "fast userspace mutexes". The resulting kernel may not |
1015 |
- run glibc-based applications correctly. |
1016 |
- |
1017 |
-+config HAVE_FUTEX_CMPXCHG |
1018 |
-+ bool |
1019 |
-+ help |
1020 |
-+ Architectures should select this if futex_atomic_cmpxchg_inatomic() |
1021 |
-+ is implemented and always working. This removes a couple of runtime |
1022 |
-+ checks. |
1023 |
-+ |
1024 |
- config EPOLL |
1025 |
- bool "Enable eventpoll support" if EXPERT |
1026 |
- default y |
1027 |
-diff --git a/kernel/futex.c b/kernel/futex.c |
1028 |
-index 08ec814ad9d2..6801b3751a95 100644 |
1029 |
---- a/kernel/futex.c |
1030 |
-+++ b/kernel/futex.c |
1031 |
-@@ -157,7 +157,9 @@ |
1032 |
- * enqueue. |
1033 |
- */ |
1034 |
- |
1035 |
-+#ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
1036 |
- int __read_mostly futex_cmpxchg_enabled; |
1037 |
-+#endif |
1038 |
- |
1039 |
- /* |
1040 |
- * Futex flags used to encode options to functions and preserve them across |
1041 |
-@@ -1450,6 +1452,7 @@ retry: |
1042 |
- hb2 = hash_futex(&key2); |
1043 |
- |
1044 |
- retry_private: |
1045 |
-+ hb_waiters_inc(hb2); |
1046 |
- double_lock_hb(hb1, hb2); |
1047 |
- |
1048 |
- if (likely(cmpval != NULL)) { |
1049 |
-@@ -1459,6 +1462,7 @@ retry_private: |
1050 |
- |
1051 |
- if (unlikely(ret)) { |
1052 |
- double_unlock_hb(hb1, hb2); |
1053 |
-+ hb_waiters_dec(hb2); |
1054 |
- |
1055 |
- ret = get_user(curval, uaddr1); |
1056 |
- if (ret) |
1057 |
-@@ -1508,6 +1512,7 @@ retry_private: |
1058 |
- break; |
1059 |
- case -EFAULT: |
1060 |
- double_unlock_hb(hb1, hb2); |
1061 |
-+ hb_waiters_dec(hb2); |
1062 |
- put_futex_key(&key2); |
1063 |
- put_futex_key(&key1); |
1064 |
- ret = fault_in_user_writeable(uaddr2); |
1065 |
-@@ -1517,6 +1522,7 @@ retry_private: |
1066 |
- case -EAGAIN: |
1067 |
- /* The owner was exiting, try again. */ |
1068 |
- double_unlock_hb(hb1, hb2); |
1069 |
-+ hb_waiters_dec(hb2); |
1070 |
- put_futex_key(&key2); |
1071 |
- put_futex_key(&key1); |
1072 |
- cond_resched(); |
1073 |
-@@ -1592,6 +1598,7 @@ retry_private: |
1074 |
- |
1075 |
- out_unlock: |
1076 |
- double_unlock_hb(hb1, hb2); |
1077 |
-+ hb_waiters_dec(hb2); |
1078 |
- |
1079 |
- /* |
1080 |
- * drop_futex_key_refs() must be called outside the spinlocks. During |
1081 |
-@@ -2875,9 +2882,28 @@ SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
1082 |
- return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1083 |
- } |
1084 |
- |
1085 |
--static int __init futex_init(void) |
1086 |
-+static void __init futex_detect_cmpxchg(void) |
1087 |
- { |
1088 |
-+#ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
1089 |
- u32 curval; |
1090 |
-+ |
1091 |
-+ /* |
1092 |
-+ * This will fail and we want it. Some arch implementations do |
1093 |
-+ * runtime detection of the futex_atomic_cmpxchg_inatomic() |
1094 |
-+ * functionality. We want to know that before we call in any |
1095 |
-+ * of the complex code paths. Also we want to prevent |
1096 |
-+ * registration of robust lists in that case. NULL is |
1097 |
-+ * guaranteed to fault and we get -EFAULT on functional |
1098 |
-+ * implementation, the non-functional ones will return |
1099 |
-+ * -ENOSYS. |
1100 |
-+ */ |
1101 |
-+ if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) |
1102 |
-+ futex_cmpxchg_enabled = 1; |
1103 |
-+#endif |
1104 |
-+} |
1105 |
-+ |
1106 |
-+static int __init futex_init(void) |
1107 |
-+{ |
1108 |
- unsigned int futex_shift; |
1109 |
- unsigned long i; |
1110 |
- |
1111 |
-@@ -2893,18 +2919,8 @@ static int __init futex_init(void) |
1112 |
- &futex_shift, NULL, |
1113 |
- futex_hashsize, futex_hashsize); |
1114 |
- futex_hashsize = 1UL << futex_shift; |
1115 |
-- /* |
1116 |
-- * This will fail and we want it. Some arch implementations do |
1117 |
-- * runtime detection of the futex_atomic_cmpxchg_inatomic() |
1118 |
-- * functionality. We want to know that before we call in any |
1119 |
-- * of the complex code paths. Also we want to prevent |
1120 |
-- * registration of robust lists in that case. NULL is |
1121 |
-- * guaranteed to fault and we get -EFAULT on functional |
1122 |
-- * implementation, the non-functional ones will return |
1123 |
-- * -ENOSYS. |
1124 |
-- */ |
1125 |
-- if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) |
1126 |
-- futex_cmpxchg_enabled = 1; |
1127 |
-+ |
1128 |
-+ futex_detect_cmpxchg(); |
1129 |
- |
1130 |
- for (i = 0; i < futex_hashsize; i++) { |
1131 |
- atomic_set(&futex_queues[i].waiters, 0); |
1132 |
-diff --git a/lib/nlattr.c b/lib/nlattr.c |
1133 |
-index 18eca7809b08..fc6754720ced 100644 |
1134 |
---- a/lib/nlattr.c |
1135 |
-+++ b/lib/nlattr.c |
1136 |
-@@ -303,9 +303,15 @@ int nla_memcmp(const struct nlattr *nla, const void *data, |
1137 |
- */ |
1138 |
- int nla_strcmp(const struct nlattr *nla, const char *str) |
1139 |
- { |
1140 |
-- int len = strlen(str) + 1; |
1141 |
-- int d = nla_len(nla) - len; |
1142 |
-+ int len = strlen(str); |
1143 |
-+ char *buf = nla_data(nla); |
1144 |
-+ int attrlen = nla_len(nla); |
1145 |
-+ int d; |
1146 |
- |
1147 |
-+ if (attrlen > 0 && buf[attrlen - 1] == '\0') |
1148 |
-+ attrlen--; |
1149 |
-+ |
1150 |
-+ d = attrlen - len; |
1151 |
- if (d == 0) |
1152 |
- d = memcmp(nla_data(nla), str, len); |
1153 |
- |
1154 |
-diff --git a/net/ipv6/icmp.c b/net/ipv6/icmp.c |
1155 |
-index f2610e157660..7b326529e6a2 100644 |
1156 |
---- a/net/ipv6/icmp.c |
1157 |
-+++ b/net/ipv6/icmp.c |
1158 |
-@@ -520,7 +520,7 @@ static void icmp6_send(struct sk_buff *skb, u8 type, u8 code, __u32 info) |
1159 |
- np->tclass, NULL, &fl6, (struct rt6_info *)dst, |
1160 |
- MSG_DONTWAIT, np->dontfrag); |
1161 |
- if (err) { |
1162 |
-- ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTERRORS); |
1163 |
-+ ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTERRORS); |
1164 |
- ip6_flush_pending_frames(sk); |
1165 |
- } else { |
1166 |
- err = icmpv6_push_pending_frames(sk, &fl6, &tmp_hdr, |
1167 |
-diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c |
1168 |
-index 64d6073731d3..3702d179506d 100644 |
1169 |
---- a/net/ipv6/ip6_output.c |
1170 |
-+++ b/net/ipv6/ip6_output.c |
1171 |
-@@ -1566,8 +1566,8 @@ int ip6_push_pending_frames(struct sock *sk) |
1172 |
- if (proto == IPPROTO_ICMPV6) { |
1173 |
- struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb)); |
1174 |
- |
1175 |
-- ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type); |
1176 |
-- ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS); |
1177 |
-+ ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type); |
1178 |
-+ ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS); |
1179 |
- } |
1180 |
- |
1181 |
- err = ip6_local_out(skb); |
1182 |
-diff --git a/net/ipv6/mcast.c b/net/ipv6/mcast.c |
1183 |
-index e1e47350784b..08b367c6b9cf 100644 |
1184 |
---- a/net/ipv6/mcast.c |
1185 |
-+++ b/net/ipv6/mcast.c |
1186 |
-@@ -1620,11 +1620,12 @@ static void mld_sendpack(struct sk_buff *skb) |
1187 |
- dst_output); |
1188 |
- out: |
1189 |
- if (!err) { |
1190 |
-- ICMP6MSGOUT_INC_STATS_BH(net, idev, ICMPV6_MLD2_REPORT); |
1191 |
-- ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS); |
1192 |
-- IP6_UPD_PO_STATS_BH(net, idev, IPSTATS_MIB_OUTMCAST, payload_len); |
1193 |
-- } else |
1194 |
-- IP6_INC_STATS_BH(net, idev, IPSTATS_MIB_OUTDISCARDS); |
1195 |
-+ ICMP6MSGOUT_INC_STATS(net, idev, ICMPV6_MLD2_REPORT); |
1196 |
-+ ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS); |
1197 |
-+ IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, payload_len); |
1198 |
-+ } else { |
1199 |
-+ IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS); |
1200 |
-+ } |
1201 |
- |
1202 |
- rcu_read_unlock(); |
1203 |
- return; |
1204 |
-diff --git a/net/ipv6/ping.c b/net/ipv6/ping.c |
1205 |
-index 587bbdcb22b4..bda74291c3e0 100644 |
1206 |
---- a/net/ipv6/ping.c |
1207 |
-+++ b/net/ipv6/ping.c |
1208 |
-@@ -182,8 +182,8 @@ int ping_v6_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, |
1209 |
- MSG_DONTWAIT, np->dontfrag); |
1210 |
- |
1211 |
- if (err) { |
1212 |
-- ICMP6_INC_STATS_BH(sock_net(sk), rt->rt6i_idev, |
1213 |
-- ICMP6_MIB_OUTERRORS); |
1214 |
-+ ICMP6_INC_STATS(sock_net(sk), rt->rt6i_idev, |
1215 |
-+ ICMP6_MIB_OUTERRORS); |
1216 |
- ip6_flush_pending_frames(sk); |
1217 |
- } else { |
1218 |
- err = icmpv6_push_pending_frames(sk, &fl6, |
1219 |
-diff --git a/net/rds/iw.c b/net/rds/iw.c |
1220 |
-index 7826d46baa70..589935661d66 100644 |
1221 |
---- a/net/rds/iw.c |
1222 |
-+++ b/net/rds/iw.c |
1223 |
-@@ -239,7 +239,8 @@ static int rds_iw_laddr_check(__be32 addr) |
1224 |
- ret = rdma_bind_addr(cm_id, (struct sockaddr *)&sin); |
1225 |
- /* due to this, we will claim to support IB devices unless we |
1226 |
- check node_type. */ |
1227 |
-- if (ret || cm_id->device->node_type != RDMA_NODE_RNIC) |
1228 |
-+ if (ret || !cm_id->device || |
1229 |
-+ cm_id->device->node_type != RDMA_NODE_RNIC) |
1230 |
- ret = -EADDRNOTAVAIL; |
1231 |
- |
1232 |
- rdsdebug("addr %pI4 ret %d node type %d\n", |
1233 |
-diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c |
1234 |
-index b332e2cc0954..e294b86c8d88 100644 |
1235 |
---- a/security/selinux/hooks.c |
1236 |
-+++ b/security/selinux/hooks.c |
1237 |
-@@ -1418,15 +1418,33 @@ static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dent |
1238 |
- isec->sid = sbsec->sid; |
1239 |
- |
1240 |
- if ((sbsec->flags & SE_SBPROC) && !S_ISLNK(inode->i_mode)) { |
1241 |
-- if (opt_dentry) { |
1242 |
-- isec->sclass = inode_mode_to_security_class(inode->i_mode); |
1243 |
-- rc = selinux_proc_get_sid(opt_dentry, |
1244 |
-- isec->sclass, |
1245 |
-- &sid); |
1246 |
-- if (rc) |
1247 |
-- goto out_unlock; |
1248 |
-- isec->sid = sid; |
1249 |
-- } |
1250 |
-+ /* We must have a dentry to determine the label on |
1251 |
-+ * procfs inodes */ |
1252 |
-+ if (opt_dentry) |
1253 |
-+ /* Called from d_instantiate or |
1254 |
-+ * d_splice_alias. */ |
1255 |
-+ dentry = dget(opt_dentry); |
1256 |
-+ else |
1257 |
-+ /* Called from selinux_complete_init, try to |
1258 |
-+ * find a dentry. */ |
1259 |
-+ dentry = d_find_alias(inode); |
1260 |
-+ /* |
1261 |
-+ * This can be hit on boot when a file is accessed |
1262 |
-+ * before the policy is loaded. When we load policy we |
1263 |
-+ * may find inodes that have no dentry on the |
1264 |
-+ * sbsec->isec_head list. No reason to complain as |
1265 |
-+ * these will get fixed up the next time we go through |
1266 |
-+ * inode_doinit() with a dentry, before these inodes |
1267 |
-+ * could be used again by userspace. |
1268 |
-+ */ |
1269 |
-+ if (!dentry) |
1270 |
-+ goto out_unlock; |
1271 |
-+ isec->sclass = inode_mode_to_security_class(inode->i_mode); |
1272 |
-+ rc = selinux_proc_get_sid(dentry, isec->sclass, &sid); |
1273 |
-+ dput(dentry); |
1274 |
-+ if (rc) |
1275 |
-+ goto out_unlock; |
1276 |
-+ isec->sid = sid; |
1277 |
- } |
1278 |
- break; |
1279 |
- } |
1280 |
-diff --git a/sound/pci/hda/hda_intel.c b/sound/pci/hda/hda_intel.c |
1281 |
-index e354ab1ec20f..a8dec9e9e876 100644 |
1282 |
---- a/sound/pci/hda/hda_intel.c |
1283 |
-+++ b/sound/pci/hda/hda_intel.c |
1284 |
-@@ -297,9 +297,9 @@ enum { SDI0, SDI1, SDI2, SDI3, SDO0, SDO1, SDO2, SDO3 }; |
1285 |
- #define ULI_NUM_CAPTURE 5 |
1286 |
- #define ULI_NUM_PLAYBACK 6 |
1287 |
- |
1288 |
--/* ATI HDMI may have up to 8 playbacks and 0 capture */ |
1289 |
-+/* ATI HDMI has 1 playback and 0 capture */ |
1290 |
- #define ATIHDMI_NUM_CAPTURE 0 |
1291 |
--#define ATIHDMI_NUM_PLAYBACK 8 |
1292 |
-+#define ATIHDMI_NUM_PLAYBACK 1 |
1293 |
- |
1294 |
- /* TERA has 4 playback and 3 capture */ |
1295 |
- #define TERA_NUM_CAPTURE 3 |
1296 |
|
1297 |
Deleted: genpatches-2.6/trunk/3.15/1001_linux-3.14.2.patch |
1298 |
=================================================================== |
1299 |
--- genpatches-2.6/trunk/3.15/1001_linux-3.14.2.patch 2014-05-05 14:22:27 UTC (rev 2771) |
1300 |
+++ genpatches-2.6/trunk/3.15/1001_linux-3.14.2.patch 2014-05-05 14:28:48 UTC (rev 2772) |
1301 |
@@ -1,1201 +0,0 @@ |
1302 |
-diff --git a/Makefile b/Makefile |
1303 |
-index 7d0b6992d9ed..b2f7de81e9a2 100644 |
1304 |
---- a/Makefile |
1305 |
-+++ b/Makefile |
1306 |
-@@ -1,6 +1,6 @@ |
1307 |
- VERSION = 3 |
1308 |
- PATCHLEVEL = 14 |
1309 |
--SUBLEVEL = 1 |
1310 |
-+SUBLEVEL = 2 |
1311 |
- EXTRAVERSION = |
1312 |
- NAME = Shuffling Zombie Juror |
1313 |
- |
1314 |
-diff --git a/arch/x86/kernel/cpu/mshyperv.c b/arch/x86/kernel/cpu/mshyperv.c |
1315 |
-index 9f7ca266864a..832d05a914ba 100644 |
1316 |
---- a/arch/x86/kernel/cpu/mshyperv.c |
1317 |
-+++ b/arch/x86/kernel/cpu/mshyperv.c |
1318 |
-@@ -26,6 +26,7 @@ |
1319 |
- #include <asm/irq_regs.h> |
1320 |
- #include <asm/i8259.h> |
1321 |
- #include <asm/apic.h> |
1322 |
-+#include <asm/timer.h> |
1323 |
- |
1324 |
- struct ms_hyperv_info ms_hyperv; |
1325 |
- EXPORT_SYMBOL_GPL(ms_hyperv); |
1326 |
-@@ -105,6 +106,11 @@ static void __init ms_hyperv_init_platform(void) |
1327 |
- |
1328 |
- if (ms_hyperv.features & HV_X64_MSR_TIME_REF_COUNT_AVAILABLE) |
1329 |
- clocksource_register_hz(&hyperv_cs, NSEC_PER_SEC/100); |
1330 |
-+ |
1331 |
-+#ifdef CONFIG_X86_IO_APIC |
1332 |
-+ no_timer_check = 1; |
1333 |
-+#endif |
1334 |
-+ |
1335 |
- } |
1336 |
- |
1337 |
- const __refconst struct hypervisor_x86 x86_hyper_ms_hyperv = { |
1338 |
-diff --git a/arch/x86/kernel/early-quirks.c b/arch/x86/kernel/early-quirks.c |
1339 |
-index bc4a088f9023..6d7d5a1260a6 100644 |
1340 |
---- a/arch/x86/kernel/early-quirks.c |
1341 |
-+++ b/arch/x86/kernel/early-quirks.c |
1342 |
-@@ -203,18 +203,15 @@ static void __init intel_remapping_check(int num, int slot, int func) |
1343 |
- revision = read_pci_config_byte(num, slot, func, PCI_REVISION_ID); |
1344 |
- |
1345 |
- /* |
1346 |
-- * Revision 13 of all triggering devices id in this quirk have |
1347 |
-- * a problem draining interrupts when irq remapping is enabled, |
1348 |
-- * and should be flagged as broken. Additionally revisions 0x12 |
1349 |
-- * and 0x22 of device id 0x3405 has this problem. |
1350 |
-+ * Revision <= 13 of all triggering devices id in this quirk |
1351 |
-+ * have a problem draining interrupts when irq remapping is |
1352 |
-+ * enabled, and should be flagged as broken. Additionally |
1353 |
-+ * revision 0x22 of device id 0x3405 has this problem. |
1354 |
- */ |
1355 |
-- if (revision == 0x13) |
1356 |
-+ if (revision <= 0x13) |
1357 |
- set_irq_remapping_broken(); |
1358 |
-- else if ((device == 0x3405) && |
1359 |
-- ((revision == 0x12) || |
1360 |
-- (revision == 0x22))) |
1361 |
-+ else if (device == 0x3405 && revision == 0x22) |
1362 |
- set_irq_remapping_broken(); |
1363 |
-- |
1364 |
- } |
1365 |
- |
1366 |
- /* |
1367 |
-diff --git a/drivers/acpi/button.c b/drivers/acpi/button.c |
1368 |
-index 714e957a871a..db35594d4df7 100644 |
1369 |
---- a/drivers/acpi/button.c |
1370 |
-+++ b/drivers/acpi/button.c |
1371 |
-@@ -302,6 +302,10 @@ static void acpi_button_notify(struct acpi_device *device, u32 event) |
1372 |
- input_sync(input); |
1373 |
- |
1374 |
- pm_wakeup_event(&device->dev, 0); |
1375 |
-+ acpi_bus_generate_netlink_event( |
1376 |
-+ device->pnp.device_class, |
1377 |
-+ dev_name(&device->dev), |
1378 |
-+ event, ++button->pushed); |
1379 |
- } |
1380 |
- break; |
1381 |
- default: |
1382 |
-diff --git a/drivers/char/ipmi/ipmi_bt_sm.c b/drivers/char/ipmi/ipmi_bt_sm.c |
1383 |
-index f5e4cd7617f6..61e71616689b 100644 |
1384 |
---- a/drivers/char/ipmi/ipmi_bt_sm.c |
1385 |
-+++ b/drivers/char/ipmi/ipmi_bt_sm.c |
1386 |
-@@ -352,7 +352,7 @@ static inline void write_all_bytes(struct si_sm_data *bt) |
1387 |
- |
1388 |
- static inline int read_all_bytes(struct si_sm_data *bt) |
1389 |
- { |
1390 |
-- unsigned char i; |
1391 |
-+ unsigned int i; |
1392 |
- |
1393 |
- /* |
1394 |
- * length is "framing info", minimum = 4: NetFn, Seq, Cmd, cCode. |
1395 |
-diff --git a/drivers/pci/host/pcie-designware.c b/drivers/pci/host/pcie-designware.c |
1396 |
-index 17ce88f79d2b..f173dd09fce4 100644 |
1397 |
---- a/drivers/pci/host/pcie-designware.c |
1398 |
-+++ b/drivers/pci/host/pcie-designware.c |
1399 |
-@@ -522,13 +522,13 @@ static void dw_pcie_prog_viewport_cfg1(struct pcie_port *pp, u32 busdev) |
1400 |
- dw_pcie_writel_rc(pp, PCIE_ATU_REGION_OUTBOUND | PCIE_ATU_REGION_INDEX1, |
1401 |
- PCIE_ATU_VIEWPORT); |
1402 |
- dw_pcie_writel_rc(pp, PCIE_ATU_TYPE_CFG1, PCIE_ATU_CR1); |
1403 |
-- dw_pcie_writel_rc(pp, PCIE_ATU_ENABLE, PCIE_ATU_CR2); |
1404 |
- dw_pcie_writel_rc(pp, pp->cfg1_base, PCIE_ATU_LOWER_BASE); |
1405 |
- dw_pcie_writel_rc(pp, (pp->cfg1_base >> 32), PCIE_ATU_UPPER_BASE); |
1406 |
- dw_pcie_writel_rc(pp, pp->cfg1_base + pp->config.cfg1_size - 1, |
1407 |
- PCIE_ATU_LIMIT); |
1408 |
- dw_pcie_writel_rc(pp, busdev, PCIE_ATU_LOWER_TARGET); |
1409 |
- dw_pcie_writel_rc(pp, 0, PCIE_ATU_UPPER_TARGET); |
1410 |
-+ dw_pcie_writel_rc(pp, PCIE_ATU_ENABLE, PCIE_ATU_CR2); |
1411 |
- } |
1412 |
- |
1413 |
- static void dw_pcie_prog_viewport_mem_outbound(struct pcie_port *pp) |
1414 |
-@@ -537,7 +537,6 @@ static void dw_pcie_prog_viewport_mem_outbound(struct pcie_port *pp) |
1415 |
- dw_pcie_writel_rc(pp, PCIE_ATU_REGION_OUTBOUND | PCIE_ATU_REGION_INDEX0, |
1416 |
- PCIE_ATU_VIEWPORT); |
1417 |
- dw_pcie_writel_rc(pp, PCIE_ATU_TYPE_MEM, PCIE_ATU_CR1); |
1418 |
-- dw_pcie_writel_rc(pp, PCIE_ATU_ENABLE, PCIE_ATU_CR2); |
1419 |
- dw_pcie_writel_rc(pp, pp->mem_base, PCIE_ATU_LOWER_BASE); |
1420 |
- dw_pcie_writel_rc(pp, (pp->mem_base >> 32), PCIE_ATU_UPPER_BASE); |
1421 |
- dw_pcie_writel_rc(pp, pp->mem_base + pp->config.mem_size - 1, |
1422 |
-@@ -545,6 +544,7 @@ static void dw_pcie_prog_viewport_mem_outbound(struct pcie_port *pp) |
1423 |
- dw_pcie_writel_rc(pp, pp->config.mem_bus_addr, PCIE_ATU_LOWER_TARGET); |
1424 |
- dw_pcie_writel_rc(pp, upper_32_bits(pp->config.mem_bus_addr), |
1425 |
- PCIE_ATU_UPPER_TARGET); |
1426 |
-+ dw_pcie_writel_rc(pp, PCIE_ATU_ENABLE, PCIE_ATU_CR2); |
1427 |
- } |
1428 |
- |
1429 |
- static void dw_pcie_prog_viewport_io_outbound(struct pcie_port *pp) |
1430 |
-@@ -553,7 +553,6 @@ static void dw_pcie_prog_viewport_io_outbound(struct pcie_port *pp) |
1431 |
- dw_pcie_writel_rc(pp, PCIE_ATU_REGION_OUTBOUND | PCIE_ATU_REGION_INDEX1, |
1432 |
- PCIE_ATU_VIEWPORT); |
1433 |
- dw_pcie_writel_rc(pp, PCIE_ATU_TYPE_IO, PCIE_ATU_CR1); |
1434 |
-- dw_pcie_writel_rc(pp, PCIE_ATU_ENABLE, PCIE_ATU_CR2); |
1435 |
- dw_pcie_writel_rc(pp, pp->io_base, PCIE_ATU_LOWER_BASE); |
1436 |
- dw_pcie_writel_rc(pp, (pp->io_base >> 32), PCIE_ATU_UPPER_BASE); |
1437 |
- dw_pcie_writel_rc(pp, pp->io_base + pp->config.io_size - 1, |
1438 |
-@@ -561,6 +560,7 @@ static void dw_pcie_prog_viewport_io_outbound(struct pcie_port *pp) |
1439 |
- dw_pcie_writel_rc(pp, pp->config.io_bus_addr, PCIE_ATU_LOWER_TARGET); |
1440 |
- dw_pcie_writel_rc(pp, upper_32_bits(pp->config.io_bus_addr), |
1441 |
- PCIE_ATU_UPPER_TARGET); |
1442 |
-+ dw_pcie_writel_rc(pp, PCIE_ATU_ENABLE, PCIE_ATU_CR2); |
1443 |
- } |
1444 |
- |
1445 |
- static int dw_pcie_rd_other_conf(struct pcie_port *pp, struct pci_bus *bus, |
1446 |
-@@ -800,7 +800,7 @@ void dw_pcie_setup_rc(struct pcie_port *pp) |
1447 |
- |
1448 |
- /* setup RC BARs */ |
1449 |
- dw_pcie_writel_rc(pp, 0x00000004, PCI_BASE_ADDRESS_0); |
1450 |
-- dw_pcie_writel_rc(pp, 0x00000004, PCI_BASE_ADDRESS_1); |
1451 |
-+ dw_pcie_writel_rc(pp, 0x00000000, PCI_BASE_ADDRESS_1); |
1452 |
- |
1453 |
- /* setup interrupt pins */ |
1454 |
- dw_pcie_readl_rc(pp, PCI_INTERRUPT_LINE, &val); |
1455 |
-diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c |
1456 |
-index 470954aba728..36d1a23f14be 100644 |
1457 |
---- a/drivers/scsi/sd.c |
1458 |
-+++ b/drivers/scsi/sd.c |
1459 |
-@@ -1463,8 +1463,8 @@ static int sd_sync_cache(struct scsi_disk *sdkp) |
1460 |
- sd_print_sense_hdr(sdkp, &sshdr); |
1461 |
- /* we need to evaluate the error return */ |
1462 |
- if (scsi_sense_valid(&sshdr) && |
1463 |
-- /* 0x3a is medium not present */ |
1464 |
-- sshdr.asc == 0x3a) |
1465 |
-+ (sshdr.asc == 0x3a || /* medium not present */ |
1466 |
-+ sshdr.asc == 0x20)) /* invalid command */ |
1467 |
- /* this is no error here */ |
1468 |
- return 0; |
1469 |
- |
1470 |
-diff --git a/drivers/staging/comedi/comedi_buf.c b/drivers/staging/comedi/comedi_buf.c |
1471 |
-index 924fce977985..257595016161 100644 |
1472 |
---- a/drivers/staging/comedi/comedi_buf.c |
1473 |
-+++ b/drivers/staging/comedi/comedi_buf.c |
1474 |
-@@ -61,6 +61,8 @@ static void __comedi_buf_free(struct comedi_device *dev, |
1475 |
- struct comedi_subdevice *s) |
1476 |
- { |
1477 |
- struct comedi_async *async = s->async; |
1478 |
-+ struct comedi_buf_map *bm; |
1479 |
-+ unsigned long flags; |
1480 |
- |
1481 |
- if (async->prealloc_buf) { |
1482 |
- vunmap(async->prealloc_buf); |
1483 |
-@@ -68,8 +70,11 @@ static void __comedi_buf_free(struct comedi_device *dev, |
1484 |
- async->prealloc_bufsz = 0; |
1485 |
- } |
1486 |
- |
1487 |
-- comedi_buf_map_put(async->buf_map); |
1488 |
-+ spin_lock_irqsave(&s->spin_lock, flags); |
1489 |
-+ bm = async->buf_map; |
1490 |
- async->buf_map = NULL; |
1491 |
-+ spin_unlock_irqrestore(&s->spin_lock, flags); |
1492 |
-+ comedi_buf_map_put(bm); |
1493 |
- } |
1494 |
- |
1495 |
- static void __comedi_buf_alloc(struct comedi_device *dev, |
1496 |
-@@ -80,6 +85,7 @@ static void __comedi_buf_alloc(struct comedi_device *dev, |
1497 |
- struct page **pages = NULL; |
1498 |
- struct comedi_buf_map *bm; |
1499 |
- struct comedi_buf_page *buf; |
1500 |
-+ unsigned long flags; |
1501 |
- unsigned i; |
1502 |
- |
1503 |
- if (!IS_ENABLED(CONFIG_HAS_DMA) && s->async_dma_dir != DMA_NONE) { |
1504 |
-@@ -92,8 +98,10 @@ static void __comedi_buf_alloc(struct comedi_device *dev, |
1505 |
- if (!bm) |
1506 |
- return; |
1507 |
- |
1508 |
-- async->buf_map = bm; |
1509 |
- kref_init(&bm->refcount); |
1510 |
-+ spin_lock_irqsave(&s->spin_lock, flags); |
1511 |
-+ async->buf_map = bm; |
1512 |
-+ spin_unlock_irqrestore(&s->spin_lock, flags); |
1513 |
- bm->dma_dir = s->async_dma_dir; |
1514 |
- if (bm->dma_dir != DMA_NONE) |
1515 |
- /* Need ref to hardware device to free buffer later. */ |
1516 |
-@@ -127,7 +135,9 @@ static void __comedi_buf_alloc(struct comedi_device *dev, |
1517 |
- |
1518 |
- pages[i] = virt_to_page(buf->virt_addr); |
1519 |
- } |
1520 |
-+ spin_lock_irqsave(&s->spin_lock, flags); |
1521 |
- bm->n_pages = i; |
1522 |
-+ spin_unlock_irqrestore(&s->spin_lock, flags); |
1523 |
- |
1524 |
- /* vmap the prealloc_buf if all the pages were allocated */ |
1525 |
- if (i == n_pages) |
1526 |
-@@ -150,6 +160,29 @@ int comedi_buf_map_put(struct comedi_buf_map *bm) |
1527 |
- return 1; |
1528 |
- } |
1529 |
- |
1530 |
-+/* returns s->async->buf_map and increments its kref refcount */ |
1531 |
-+struct comedi_buf_map * |
1532 |
-+comedi_buf_map_from_subdev_get(struct comedi_subdevice *s) |
1533 |
-+{ |
1534 |
-+ struct comedi_async *async = s->async; |
1535 |
-+ struct comedi_buf_map *bm = NULL; |
1536 |
-+ unsigned long flags; |
1537 |
-+ |
1538 |
-+ if (!async) |
1539 |
-+ return NULL; |
1540 |
-+ |
1541 |
-+ spin_lock_irqsave(&s->spin_lock, flags); |
1542 |
-+ bm = async->buf_map; |
1543 |
-+ /* only want it if buffer pages allocated */ |
1544 |
-+ if (bm && bm->n_pages) |
1545 |
-+ comedi_buf_map_get(bm); |
1546 |
-+ else |
1547 |
-+ bm = NULL; |
1548 |
-+ spin_unlock_irqrestore(&s->spin_lock, flags); |
1549 |
-+ |
1550 |
-+ return bm; |
1551 |
-+} |
1552 |
-+ |
1553 |
- bool comedi_buf_is_mmapped(struct comedi_async *async) |
1554 |
- { |
1555 |
- struct comedi_buf_map *bm = async->buf_map; |
1556 |
-diff --git a/drivers/staging/comedi/comedi_fops.c b/drivers/staging/comedi/comedi_fops.c |
1557 |
-index c22c617b0da1..eae3ee139330 100644 |
1558 |
---- a/drivers/staging/comedi/comedi_fops.c |
1559 |
-+++ b/drivers/staging/comedi/comedi_fops.c |
1560 |
-@@ -1923,14 +1923,21 @@ static int comedi_mmap(struct file *file, struct vm_area_struct *vma) |
1561 |
- struct comedi_device *dev = file->private_data; |
1562 |
- struct comedi_subdevice *s; |
1563 |
- struct comedi_async *async; |
1564 |
-- struct comedi_buf_map *bm; |
1565 |
-+ struct comedi_buf_map *bm = NULL; |
1566 |
- unsigned long start = vma->vm_start; |
1567 |
- unsigned long size; |
1568 |
- int n_pages; |
1569 |
- int i; |
1570 |
- int retval; |
1571 |
- |
1572 |
-- mutex_lock(&dev->mutex); |
1573 |
-+ /* |
1574 |
-+ * 'trylock' avoids circular dependency with current->mm->mmap_sem |
1575 |
-+ * and down-reading &dev->attach_lock should normally succeed without |
1576 |
-+ * contention unless the device is in the process of being attached |
1577 |
-+ * or detached. |
1578 |
-+ */ |
1579 |
-+ if (!down_read_trylock(&dev->attach_lock)) |
1580 |
-+ return -EAGAIN; |
1581 |
- |
1582 |
- if (!dev->attached) { |
1583 |
- dev_dbg(dev->class_dev, "no driver attached\n"); |
1584 |
-@@ -1970,7 +1977,9 @@ static int comedi_mmap(struct file *file, struct vm_area_struct *vma) |
1585 |
- } |
1586 |
- |
1587 |
- n_pages = size >> PAGE_SHIFT; |
1588 |
-- bm = async->buf_map; |
1589 |
-+ |
1590 |
-+ /* get reference to current buf map (if any) */ |
1591 |
-+ bm = comedi_buf_map_from_subdev_get(s); |
1592 |
- if (!bm || n_pages > bm->n_pages) { |
1593 |
- retval = -EINVAL; |
1594 |
- goto done; |
1595 |
-@@ -1994,7 +2003,8 @@ static int comedi_mmap(struct file *file, struct vm_area_struct *vma) |
1596 |
- |
1597 |
- retval = 0; |
1598 |
- done: |
1599 |
-- mutex_unlock(&dev->mutex); |
1600 |
-+ up_read(&dev->attach_lock); |
1601 |
-+ comedi_buf_map_put(bm); /* put reference to buf map - okay if NULL */ |
1602 |
- return retval; |
1603 |
- } |
1604 |
- |
1605 |
-diff --git a/drivers/staging/comedi/comedi_internal.h b/drivers/staging/comedi/comedi_internal.h |
1606 |
-index 9a746570f161..a492f2d2436e 100644 |
1607 |
---- a/drivers/staging/comedi/comedi_internal.h |
1608 |
-+++ b/drivers/staging/comedi/comedi_internal.h |
1609 |
-@@ -19,6 +19,8 @@ void comedi_buf_reset(struct comedi_async *async); |
1610 |
- bool comedi_buf_is_mmapped(struct comedi_async *async); |
1611 |
- void comedi_buf_map_get(struct comedi_buf_map *bm); |
1612 |
- int comedi_buf_map_put(struct comedi_buf_map *bm); |
1613 |
-+struct comedi_buf_map *comedi_buf_map_from_subdev_get( |
1614 |
-+ struct comedi_subdevice *s); |
1615 |
- unsigned int comedi_buf_write_n_allocated(struct comedi_async *async); |
1616 |
- void comedi_device_cancel_all(struct comedi_device *dev); |
1617 |
- |
1618 |
-diff --git a/drivers/staging/comedi/drivers/8255_pci.c b/drivers/staging/comedi/drivers/8255_pci.c |
1619 |
-index 8a57c3c1ade0..1097dc6a3086 100644 |
1620 |
---- a/drivers/staging/comedi/drivers/8255_pci.c |
1621 |
-+++ b/drivers/staging/comedi/drivers/8255_pci.c |
1622 |
-@@ -56,6 +56,7 @@ Configuration Options: not applicable, uses PCI auto config |
1623 |
- #include "../comedidev.h" |
1624 |
- |
1625 |
- #include "8255.h" |
1626 |
-+#include "mite.h" |
1627 |
- |
1628 |
- enum pci_8255_boardid { |
1629 |
- BOARD_ADLINK_PCI7224, |
1630 |
-@@ -79,6 +80,7 @@ struct pci_8255_boardinfo { |
1631 |
- const char *name; |
1632 |
- int dio_badr; |
1633 |
- int n_8255; |
1634 |
-+ unsigned int has_mite:1; |
1635 |
- }; |
1636 |
- |
1637 |
- static const struct pci_8255_boardinfo pci_8255_boards[] = { |
1638 |
-@@ -126,36 +128,43 @@ static const struct pci_8255_boardinfo pci_8255_boards[] = { |
1639 |
- .name = "ni_pci-dio-96", |
1640 |
- .dio_badr = 1, |
1641 |
- .n_8255 = 4, |
1642 |
-+ .has_mite = 1, |
1643 |
- }, |
1644 |
- [BOARD_NI_PCIDIO96B] = { |
1645 |
- .name = "ni_pci-dio-96b", |
1646 |
- .dio_badr = 1, |
1647 |
- .n_8255 = 4, |
1648 |
-+ .has_mite = 1, |
1649 |
- }, |
1650 |
- [BOARD_NI_PXI6508] = { |
1651 |
- .name = "ni_pxi-6508", |
1652 |
- .dio_badr = 1, |
1653 |
- .n_8255 = 4, |
1654 |
-+ .has_mite = 1, |
1655 |
- }, |
1656 |
- [BOARD_NI_PCI6503] = { |
1657 |
- .name = "ni_pci-6503", |
1658 |
- .dio_badr = 1, |
1659 |
- .n_8255 = 1, |
1660 |
-+ .has_mite = 1, |
1661 |
- }, |
1662 |
- [BOARD_NI_PCI6503B] = { |
1663 |
- .name = "ni_pci-6503b", |
1664 |
- .dio_badr = 1, |
1665 |
- .n_8255 = 1, |
1666 |
-+ .has_mite = 1, |
1667 |
- }, |
1668 |
- [BOARD_NI_PCI6503X] = { |
1669 |
- .name = "ni_pci-6503x", |
1670 |
- .dio_badr = 1, |
1671 |
- .n_8255 = 1, |
1672 |
-+ .has_mite = 1, |
1673 |
- }, |
1674 |
- [BOARD_NI_PXI_6503] = { |
1675 |
- .name = "ni_pxi-6503", |
1676 |
- .dio_badr = 1, |
1677 |
- .n_8255 = 1, |
1678 |
-+ .has_mite = 1, |
1679 |
- }, |
1680 |
- }; |
1681 |
- |
1682 |
-@@ -163,6 +172,25 @@ struct pci_8255_private { |
1683 |
- void __iomem *mmio_base; |
1684 |
- }; |
1685 |
- |
1686 |
-+static int pci_8255_mite_init(struct pci_dev *pcidev) |
1687 |
-+{ |
1688 |
-+ void __iomem *mite_base; |
1689 |
-+ u32 main_phys_addr; |
1690 |
-+ |
1691 |
-+ /* ioremap the MITE registers (BAR 0) temporarily */ |
1692 |
-+ mite_base = pci_ioremap_bar(pcidev, 0); |
1693 |
-+ if (!mite_base) |
1694 |
-+ return -ENOMEM; |
1695 |
-+ |
1696 |
-+ /* set data window to main registers (BAR 1) */ |
1697 |
-+ main_phys_addr = pci_resource_start(pcidev, 1); |
1698 |
-+ writel(main_phys_addr | WENAB, mite_base + MITE_IODWBSR); |
1699 |
-+ |
1700 |
-+ /* finished with MITE registers */ |
1701 |
-+ iounmap(mite_base); |
1702 |
-+ return 0; |
1703 |
-+} |
1704 |
-+ |
1705 |
- static int pci_8255_mmio(int dir, int port, int data, unsigned long iobase) |
1706 |
- { |
1707 |
- void __iomem *mmio_base = (void __iomem *)iobase; |
1708 |
-@@ -201,6 +229,12 @@ static int pci_8255_auto_attach(struct comedi_device *dev, |
1709 |
- if (ret) |
1710 |
- return ret; |
1711 |
- |
1712 |
-+ if (board->has_mite) { |
1713 |
-+ ret = pci_8255_mite_init(pcidev); |
1714 |
-+ if (ret) |
1715 |
-+ return ret; |
1716 |
-+ } |
1717 |
-+ |
1718 |
- is_mmio = (pci_resource_flags(pcidev, board->dio_badr) & |
1719 |
- IORESOURCE_MEM) != 0; |
1720 |
- if (is_mmio) { |
1721 |
-diff --git a/drivers/tty/ipwireless/tty.c b/drivers/tty/ipwireless/tty.c |
1722 |
-index ebd5bff0f5c1..17ee3bf0926b 100644 |
1723 |
---- a/drivers/tty/ipwireless/tty.c |
1724 |
-+++ b/drivers/tty/ipwireless/tty.c |
1725 |
-@@ -176,9 +176,6 @@ void ipwireless_tty_received(struct ipw_tty *tty, unsigned char *data, |
1726 |
- ": %d chars not inserted to flip buffer!\n", |
1727 |
- length - work); |
1728 |
- |
1729 |
-- /* |
1730 |
-- * This may sleep if ->low_latency is set |
1731 |
-- */ |
1732 |
- if (work) |
1733 |
- tty_flip_buffer_push(&tty->port); |
1734 |
- } |
1735 |
-diff --git a/drivers/tty/tty_buffer.c b/drivers/tty/tty_buffer.c |
1736 |
-index 765125dff20e..8ebd9f88a6f6 100644 |
1737 |
---- a/drivers/tty/tty_buffer.c |
1738 |
-+++ b/drivers/tty/tty_buffer.c |
1739 |
-@@ -351,14 +351,11 @@ EXPORT_SYMBOL(tty_insert_flip_string_flags); |
1740 |
- * Takes any pending buffers and transfers their ownership to the |
1741 |
- * ldisc side of the queue. It then schedules those characters for |
1742 |
- * processing by the line discipline. |
1743 |
-- * Note that this function can only be used when the low_latency flag |
1744 |
-- * is unset. Otherwise the workqueue won't be flushed. |
1745 |
- */ |
1746 |
- |
1747 |
- void tty_schedule_flip(struct tty_port *port) |
1748 |
- { |
1749 |
- struct tty_bufhead *buf = &port->buf; |
1750 |
-- WARN_ON(port->low_latency); |
1751 |
- |
1752 |
- buf->tail->commit = buf->tail->used; |
1753 |
- schedule_work(&buf->work); |
1754 |
-@@ -482,17 +479,15 @@ static void flush_to_ldisc(struct work_struct *work) |
1755 |
- */ |
1756 |
- void tty_flush_to_ldisc(struct tty_struct *tty) |
1757 |
- { |
1758 |
-- if (!tty->port->low_latency) |
1759 |
-- flush_work(&tty->port->buf.work); |
1760 |
-+ flush_work(&tty->port->buf.work); |
1761 |
- } |
1762 |
- |
1763 |
- /** |
1764 |
- * tty_flip_buffer_push - terminal |
1765 |
- * @port: tty port to push |
1766 |
- * |
1767 |
-- * Queue a push of the terminal flip buffers to the line discipline. This |
1768 |
-- * function must not be called from IRQ context if port->low_latency is |
1769 |
-- * set. |
1770 |
-+ * Queue a push of the terminal flip buffers to the line discipline. |
1771 |
-+ * Can be called from IRQ/atomic context. |
1772 |
- * |
1773 |
- * In the event of the queue being busy for flipping the work will be |
1774 |
- * held off and retried later. |
1775 |
-@@ -500,14 +495,7 @@ void tty_flush_to_ldisc(struct tty_struct *tty) |
1776 |
- |
1777 |
- void tty_flip_buffer_push(struct tty_port *port) |
1778 |
- { |
1779 |
-- struct tty_bufhead *buf = &port->buf; |
1780 |
-- |
1781 |
-- buf->tail->commit = buf->tail->used; |
1782 |
-- |
1783 |
-- if (port->low_latency) |
1784 |
-- flush_to_ldisc(&buf->work); |
1785 |
-- else |
1786 |
-- schedule_work(&buf->work); |
1787 |
-+ tty_schedule_flip(port); |
1788 |
- } |
1789 |
- EXPORT_SYMBOL(tty_flip_buffer_push); |
1790 |
- |
1791 |
-diff --git a/drivers/tty/tty_io.c b/drivers/tty/tty_io.c |
1792 |
-index c74a00ad7add..d3448a90f0f9 100644 |
1793 |
---- a/drivers/tty/tty_io.c |
1794 |
-+++ b/drivers/tty/tty_io.c |
1795 |
-@@ -1271,12 +1271,13 @@ static void pty_line_name(struct tty_driver *driver, int index, char *p) |
1796 |
- * |
1797 |
- * Locking: None |
1798 |
- */ |
1799 |
--static void tty_line_name(struct tty_driver *driver, int index, char *p) |
1800 |
-+static ssize_t tty_line_name(struct tty_driver *driver, int index, char *p) |
1801 |
- { |
1802 |
- if (driver->flags & TTY_DRIVER_UNNUMBERED_NODE) |
1803 |
-- strcpy(p, driver->name); |
1804 |
-+ return sprintf(p, "%s", driver->name); |
1805 |
- else |
1806 |
-- sprintf(p, "%s%d", driver->name, index + driver->name_base); |
1807 |
-+ return sprintf(p, "%s%d", driver->name, |
1808 |
-+ index + driver->name_base); |
1809 |
- } |
1810 |
- |
1811 |
- /** |
1812 |
-@@ -3545,9 +3546,19 @@ static ssize_t show_cons_active(struct device *dev, |
1813 |
- if (i >= ARRAY_SIZE(cs)) |
1814 |
- break; |
1815 |
- } |
1816 |
-- while (i--) |
1817 |
-- count += sprintf(buf + count, "%s%d%c", |
1818 |
-- cs[i]->name, cs[i]->index, i ? ' ':'\n'); |
1819 |
-+ while (i--) { |
1820 |
-+ int index = cs[i]->index; |
1821 |
-+ struct tty_driver *drv = cs[i]->device(cs[i], &index); |
1822 |
-+ |
1823 |
-+ /* don't resolve tty0 as some programs depend on it */ |
1824 |
-+ if (drv && (cs[i]->index > 0 || drv->major != TTY_MAJOR)) |
1825 |
-+ count += tty_line_name(drv, index, buf + count); |
1826 |
-+ else |
1827 |
-+ count += sprintf(buf + count, "%s%d", |
1828 |
-+ cs[i]->name, cs[i]->index); |
1829 |
-+ |
1830 |
-+ count += sprintf(buf + count, "%c", i ? ' ':'\n'); |
1831 |
-+ } |
1832 |
- console_unlock(); |
1833 |
- |
1834 |
- return count; |
1835 |
-diff --git a/drivers/usb/gadget/u_serial.c b/drivers/usb/gadget/u_serial.c |
1836 |
-index b369292d4b90..ad0aca812002 100644 |
1837 |
---- a/drivers/usb/gadget/u_serial.c |
1838 |
-+++ b/drivers/usb/gadget/u_serial.c |
1839 |
-@@ -549,8 +549,8 @@ static void gs_rx_push(unsigned long _port) |
1840 |
- port->read_started--; |
1841 |
- } |
1842 |
- |
1843 |
-- /* Push from tty to ldisc; without low_latency set this is handled by |
1844 |
-- * a workqueue, so we won't get callbacks and can hold port_lock |
1845 |
-+ /* Push from tty to ldisc; this is handled by a workqueue, |
1846 |
-+ * so we won't get callbacks and can hold port_lock |
1847 |
- */ |
1848 |
- if (do_push) |
1849 |
- tty_flip_buffer_push(&port->port); |
1850 |
-diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c |
1851 |
-index 81ea55314b1f..9a527a1826df 100644 |
1852 |
---- a/fs/btrfs/disk-io.c |
1853 |
-+++ b/fs/btrfs/disk-io.c |
1854 |
-@@ -3244,6 +3244,8 @@ static int barrier_all_devices(struct btrfs_fs_info *info) |
1855 |
- /* send down all the barriers */ |
1856 |
- head = &info->fs_devices->devices; |
1857 |
- list_for_each_entry_rcu(dev, head, dev_list) { |
1858 |
-+ if (dev->missing) |
1859 |
-+ continue; |
1860 |
- if (!dev->bdev) { |
1861 |
- errors_send++; |
1862 |
- continue; |
1863 |
-@@ -3258,6 +3260,8 @@ static int barrier_all_devices(struct btrfs_fs_info *info) |
1864 |
- |
1865 |
- /* wait for all the barriers */ |
1866 |
- list_for_each_entry_rcu(dev, head, dev_list) { |
1867 |
-+ if (dev->missing) |
1868 |
-+ continue; |
1869 |
- if (!dev->bdev) { |
1870 |
- errors_wait++; |
1871 |
- continue; |
1872 |
-diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c |
1873 |
-index 32312e09f0f5..3c8e68da9ef8 100644 |
1874 |
---- a/fs/btrfs/extent-tree.c |
1875 |
-+++ b/fs/btrfs/extent-tree.c |
1876 |
-@@ -2444,7 +2444,8 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
1877 |
- spin_unlock(&locked_ref->lock); |
1878 |
- spin_lock(&delayed_refs->lock); |
1879 |
- spin_lock(&locked_ref->lock); |
1880 |
-- if (rb_first(&locked_ref->ref_root)) { |
1881 |
-+ if (rb_first(&locked_ref->ref_root) || |
1882 |
-+ locked_ref->extent_op) { |
1883 |
- spin_unlock(&locked_ref->lock); |
1884 |
- spin_unlock(&delayed_refs->lock); |
1885 |
- continue; |
1886 |
-diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c |
1887 |
-index 34cd83184c4a..b05bf58b9395 100644 |
1888 |
---- a/fs/btrfs/transaction.c |
1889 |
-+++ b/fs/btrfs/transaction.c |
1890 |
-@@ -683,7 +683,8 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
1891 |
- int lock = (trans->type != TRANS_JOIN_NOLOCK); |
1892 |
- int err = 0; |
1893 |
- |
1894 |
-- if (--trans->use_count) { |
1895 |
-+ if (trans->use_count > 1) { |
1896 |
-+ trans->use_count--; |
1897 |
- trans->block_rsv = trans->orig_rsv; |
1898 |
- return 0; |
1899 |
- } |
1900 |
-@@ -731,17 +732,10 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
1901 |
- } |
1902 |
- |
1903 |
- if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
1904 |
-- if (throttle) { |
1905 |
-- /* |
1906 |
-- * We may race with somebody else here so end up having |
1907 |
-- * to call end_transaction on ourselves again, so inc |
1908 |
-- * our use_count. |
1909 |
-- */ |
1910 |
-- trans->use_count++; |
1911 |
-+ if (throttle) |
1912 |
- return btrfs_commit_transaction(trans, root); |
1913 |
-- } else { |
1914 |
-+ else |
1915 |
- wake_up_process(info->transaction_kthread); |
1916 |
-- } |
1917 |
- } |
1918 |
- |
1919 |
- if (trans->type & __TRANS_FREEZABLE) |
1920 |
-diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c |
1921 |
-index 74bc2d549c58..47188916dd8d 100644 |
1922 |
---- a/fs/ext4/extents.c |
1923 |
-+++ b/fs/ext4/extents.c |
1924 |
-@@ -2585,6 +2585,27 @@ ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, |
1925 |
- ex_ee_block = le32_to_cpu(ex->ee_block); |
1926 |
- ex_ee_len = ext4_ext_get_actual_len(ex); |
1927 |
- |
1928 |
-+ /* |
1929 |
-+ * If we're starting with an extent other than the last one in the |
1930 |
-+ * node, we need to see if it shares a cluster with the extent to |
1931 |
-+ * the right (towards the end of the file). If its leftmost cluster |
1932 |
-+ * is this extent's rightmost cluster and it is not cluster aligned, |
1933 |
-+ * we'll mark it as a partial that is not to be deallocated. |
1934 |
-+ */ |
1935 |
-+ |
1936 |
-+ if (ex != EXT_LAST_EXTENT(eh)) { |
1937 |
-+ ext4_fsblk_t current_pblk, right_pblk; |
1938 |
-+ long long current_cluster, right_cluster; |
1939 |
-+ |
1940 |
-+ current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1; |
1941 |
-+ current_cluster = (long long)EXT4_B2C(sbi, current_pblk); |
1942 |
-+ right_pblk = ext4_ext_pblock(ex + 1); |
1943 |
-+ right_cluster = (long long)EXT4_B2C(sbi, right_pblk); |
1944 |
-+ if (current_cluster == right_cluster && |
1945 |
-+ EXT4_PBLK_COFF(sbi, right_pblk)) |
1946 |
-+ *partial_cluster = -right_cluster; |
1947 |
-+ } |
1948 |
-+ |
1949 |
- trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
1950 |
- |
1951 |
- while (ex >= EXT_FIRST_EXTENT(eh) && |
1952 |
-@@ -2710,10 +2731,15 @@ ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, |
1953 |
- err = ext4_ext_correct_indexes(handle, inode, path); |
1954 |
- |
1955 |
- /* |
1956 |
-- * Free the partial cluster only if the current extent does not |
1957 |
-- * reference it. Otherwise we might free used cluster. |
1958 |
-+ * If there's a partial cluster and at least one extent remains in |
1959 |
-+ * the leaf, free the partial cluster if it isn't shared with the |
1960 |
-+ * current extent. If there's a partial cluster and no extents |
1961 |
-+ * remain in the leaf, it can't be freed here. It can only be |
1962 |
-+ * freed when it's possible to determine if it's not shared with |
1963 |
-+ * any other extent - when the next leaf is processed or when space |
1964 |
-+ * removal is complete. |
1965 |
- */ |
1966 |
-- if (*partial_cluster > 0 && |
1967 |
-+ if (*partial_cluster > 0 && eh->eh_entries && |
1968 |
- (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != |
1969 |
- *partial_cluster)) { |
1970 |
- int flags = get_default_free_blocks_flags(inode); |
1971 |
-@@ -4128,7 +4154,7 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
1972 |
- struct ext4_extent newex, *ex, *ex2; |
1973 |
- struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
1974 |
- ext4_fsblk_t newblock = 0; |
1975 |
-- int free_on_err = 0, err = 0, depth; |
1976 |
-+ int free_on_err = 0, err = 0, depth, ret; |
1977 |
- unsigned int allocated = 0, offset = 0; |
1978 |
- unsigned int allocated_clusters = 0; |
1979 |
- struct ext4_allocation_request ar; |
1980 |
-@@ -4189,9 +4215,13 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
1981 |
- if (!ext4_ext_is_uninitialized(ex)) |
1982 |
- goto out; |
1983 |
- |
1984 |
-- allocated = ext4_ext_handle_uninitialized_extents( |
1985 |
-+ ret = ext4_ext_handle_uninitialized_extents( |
1986 |
- handle, inode, map, path, flags, |
1987 |
- allocated, newblock); |
1988 |
-+ if (ret < 0) |
1989 |
-+ err = ret; |
1990 |
-+ else |
1991 |
-+ allocated = ret; |
1992 |
- goto out3; |
1993 |
- } |
1994 |
- } |
1995 |
-diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c |
1996 |
-index d754e3cf99a8..a16315957ef3 100644 |
1997 |
---- a/fs/fs-writeback.c |
1998 |
-+++ b/fs/fs-writeback.c |
1999 |
-@@ -89,16 +89,29 @@ static inline struct inode *wb_inode(struct list_head *head) |
2000 |
- #define CREATE_TRACE_POINTS |
2001 |
- #include <trace/events/writeback.h> |
2002 |
- |
2003 |
-+static void bdi_wakeup_thread(struct backing_dev_info *bdi) |
2004 |
-+{ |
2005 |
-+ spin_lock_bh(&bdi->wb_lock); |
2006 |
-+ if (test_bit(BDI_registered, &bdi->state)) |
2007 |
-+ mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
2008 |
-+ spin_unlock_bh(&bdi->wb_lock); |
2009 |
-+} |
2010 |
-+ |
2011 |
- static void bdi_queue_work(struct backing_dev_info *bdi, |
2012 |
- struct wb_writeback_work *work) |
2013 |
- { |
2014 |
- trace_writeback_queue(bdi, work); |
2015 |
- |
2016 |
- spin_lock_bh(&bdi->wb_lock); |
2017 |
-+ if (!test_bit(BDI_registered, &bdi->state)) { |
2018 |
-+ if (work->done) |
2019 |
-+ complete(work->done); |
2020 |
-+ goto out_unlock; |
2021 |
-+ } |
2022 |
- list_add_tail(&work->list, &bdi->work_list); |
2023 |
-- spin_unlock_bh(&bdi->wb_lock); |
2024 |
-- |
2025 |
- mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
2026 |
-+out_unlock: |
2027 |
-+ spin_unlock_bh(&bdi->wb_lock); |
2028 |
- } |
2029 |
- |
2030 |
- static void |
2031 |
-@@ -114,7 +127,7 @@ __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
2032 |
- work = kzalloc(sizeof(*work), GFP_ATOMIC); |
2033 |
- if (!work) { |
2034 |
- trace_writeback_nowork(bdi); |
2035 |
-- mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
2036 |
-+ bdi_wakeup_thread(bdi); |
2037 |
- return; |
2038 |
- } |
2039 |
- |
2040 |
-@@ -161,7 +174,7 @@ void bdi_start_background_writeback(struct backing_dev_info *bdi) |
2041 |
- * writeback as soon as there is no other work to do. |
2042 |
- */ |
2043 |
- trace_writeback_wake_background(bdi); |
2044 |
-- mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
2045 |
-+ bdi_wakeup_thread(bdi); |
2046 |
- } |
2047 |
- |
2048 |
- /* |
2049 |
-@@ -1017,7 +1030,7 @@ void bdi_writeback_workfn(struct work_struct *work) |
2050 |
- current->flags |= PF_SWAPWRITE; |
2051 |
- |
2052 |
- if (likely(!current_is_workqueue_rescuer() || |
2053 |
-- list_empty(&bdi->bdi_list))) { |
2054 |
-+ !test_bit(BDI_registered, &bdi->state))) { |
2055 |
- /* |
2056 |
- * The normal path. Keep writing back @bdi until its |
2057 |
- * work_list is empty. Note that this path is also taken |
2058 |
-@@ -1039,10 +1052,10 @@ void bdi_writeback_workfn(struct work_struct *work) |
2059 |
- trace_writeback_pages_written(pages_written); |
2060 |
- } |
2061 |
- |
2062 |
-- if (!list_empty(&bdi->work_list) || |
2063 |
-- (wb_has_dirty_io(wb) && dirty_writeback_interval)) |
2064 |
-- queue_delayed_work(bdi_wq, &wb->dwork, |
2065 |
-- msecs_to_jiffies(dirty_writeback_interval * 10)); |
2066 |
-+ if (!list_empty(&bdi->work_list)) |
2067 |
-+ mod_delayed_work(bdi_wq, &wb->dwork, 0); |
2068 |
-+ else if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
2069 |
-+ bdi_wakeup_thread_delayed(bdi); |
2070 |
- |
2071 |
- current->flags &= ~PF_SWAPWRITE; |
2072 |
- } |
2073 |
-diff --git a/fs/jffs2/compr_rtime.c b/fs/jffs2/compr_rtime.c |
2074 |
-index 16a5047903a6..406d9cc84ba8 100644 |
2075 |
---- a/fs/jffs2/compr_rtime.c |
2076 |
-+++ b/fs/jffs2/compr_rtime.c |
2077 |
-@@ -33,7 +33,7 @@ static int jffs2_rtime_compress(unsigned char *data_in, |
2078 |
- unsigned char *cpage_out, |
2079 |
- uint32_t *sourcelen, uint32_t *dstlen) |
2080 |
- { |
2081 |
-- short positions[256]; |
2082 |
-+ unsigned short positions[256]; |
2083 |
- int outpos = 0; |
2084 |
- int pos=0; |
2085 |
- |
2086 |
-@@ -74,7 +74,7 @@ static int jffs2_rtime_decompress(unsigned char *data_in, |
2087 |
- unsigned char *cpage_out, |
2088 |
- uint32_t srclen, uint32_t destlen) |
2089 |
- { |
2090 |
-- short positions[256]; |
2091 |
-+ unsigned short positions[256]; |
2092 |
- int outpos = 0; |
2093 |
- int pos=0; |
2094 |
- |
2095 |
-diff --git a/fs/jffs2/nodelist.h b/fs/jffs2/nodelist.h |
2096 |
-index e4619b00f7c5..fa35ff79ab35 100644 |
2097 |
---- a/fs/jffs2/nodelist.h |
2098 |
-+++ b/fs/jffs2/nodelist.h |
2099 |
-@@ -231,7 +231,7 @@ struct jffs2_tmp_dnode_info |
2100 |
- uint32_t version; |
2101 |
- uint32_t data_crc; |
2102 |
- uint32_t partial_crc; |
2103 |
-- uint16_t csize; |
2104 |
-+ uint32_t csize; |
2105 |
- uint16_t overlapped; |
2106 |
- }; |
2107 |
- |
2108 |
-diff --git a/fs/jffs2/nodemgmt.c b/fs/jffs2/nodemgmt.c |
2109 |
-index 03310721712f..b6bd4affd9ad 100644 |
2110 |
---- a/fs/jffs2/nodemgmt.c |
2111 |
-+++ b/fs/jffs2/nodemgmt.c |
2112 |
-@@ -179,6 +179,7 @@ int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
2113 |
- spin_unlock(&c->erase_completion_lock); |
2114 |
- |
2115 |
- schedule(); |
2116 |
-+ remove_wait_queue(&c->erase_wait, &wait); |
2117 |
- } else |
2118 |
- spin_unlock(&c->erase_completion_lock); |
2119 |
- } else if (ret) |
2120 |
-@@ -211,20 +212,25 @@ out: |
2121 |
- int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize, |
2122 |
- uint32_t *len, uint32_t sumsize) |
2123 |
- { |
2124 |
-- int ret = -EAGAIN; |
2125 |
-+ int ret; |
2126 |
- minsize = PAD(minsize); |
2127 |
- |
2128 |
- jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize); |
2129 |
- |
2130 |
-- spin_lock(&c->erase_completion_lock); |
2131 |
-- while(ret == -EAGAIN) { |
2132 |
-+ while (true) { |
2133 |
-+ spin_lock(&c->erase_completion_lock); |
2134 |
- ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
2135 |
- if (ret) { |
2136 |
- jffs2_dbg(1, "%s(): looping, ret is %d\n", |
2137 |
- __func__, ret); |
2138 |
- } |
2139 |
-+ spin_unlock(&c->erase_completion_lock); |
2140 |
-+ |
2141 |
-+ if (ret == -EAGAIN) |
2142 |
-+ cond_resched(); |
2143 |
-+ else |
2144 |
-+ break; |
2145 |
- } |
2146 |
-- spin_unlock(&c->erase_completion_lock); |
2147 |
- if (!ret) |
2148 |
- ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1); |
2149 |
- |
2150 |
-diff --git a/fs/kernfs/dir.c b/fs/kernfs/dir.c |
2151 |
-index bd6e18be6e1a..39c0143fb3af 100644 |
2152 |
---- a/fs/kernfs/dir.c |
2153 |
-+++ b/fs/kernfs/dir.c |
2154 |
-@@ -37,7 +37,7 @@ static unsigned int kernfs_name_hash(const char *name, const void *ns) |
2155 |
- hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31)); |
2156 |
- hash &= 0x7fffffffU; |
2157 |
- /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ |
2158 |
-- if (hash < 1) |
2159 |
-+ if (hash < 2) |
2160 |
- hash += 2; |
2161 |
- if (hash >= INT_MAX) |
2162 |
- hash = INT_MAX - 1; |
2163 |
-diff --git a/fs/kernfs/inode.c b/fs/kernfs/inode.c |
2164 |
-index e55126f85bd2..553946c9d952 100644 |
2165 |
---- a/fs/kernfs/inode.c |
2166 |
-+++ b/fs/kernfs/inode.c |
2167 |
-@@ -48,14 +48,18 @@ void __init kernfs_inode_init(void) |
2168 |
- |
2169 |
- static struct kernfs_iattrs *kernfs_iattrs(struct kernfs_node *kn) |
2170 |
- { |
2171 |
-+ static DEFINE_MUTEX(iattr_mutex); |
2172 |
-+ struct kernfs_iattrs *ret; |
2173 |
- struct iattr *iattrs; |
2174 |
- |
2175 |
-+ mutex_lock(&iattr_mutex); |
2176 |
-+ |
2177 |
- if (kn->iattr) |
2178 |
-- return kn->iattr; |
2179 |
-+ goto out_unlock; |
2180 |
- |
2181 |
- kn->iattr = kzalloc(sizeof(struct kernfs_iattrs), GFP_KERNEL); |
2182 |
- if (!kn->iattr) |
2183 |
-- return NULL; |
2184 |
-+ goto out_unlock; |
2185 |
- iattrs = &kn->iattr->ia_iattr; |
2186 |
- |
2187 |
- /* assign default attributes */ |
2188 |
-@@ -65,8 +69,10 @@ static struct kernfs_iattrs *kernfs_iattrs(struct kernfs_node *kn) |
2189 |
- iattrs->ia_atime = iattrs->ia_mtime = iattrs->ia_ctime = CURRENT_TIME; |
2190 |
- |
2191 |
- simple_xattrs_init(&kn->iattr->xattrs); |
2192 |
-- |
2193 |
-- return kn->iattr; |
2194 |
-+out_unlock: |
2195 |
-+ ret = kn->iattr; |
2196 |
-+ mutex_unlock(&iattr_mutex); |
2197 |
-+ return ret; |
2198 |
- } |
2199 |
- |
2200 |
- static int __kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr) |
2201 |
-diff --git a/fs/posix_acl.c b/fs/posix_acl.c |
2202 |
-index 11c54fd51e16..9e363e41dacc 100644 |
2203 |
---- a/fs/posix_acl.c |
2204 |
-+++ b/fs/posix_acl.c |
2205 |
-@@ -723,7 +723,7 @@ posix_acl_to_xattr(struct user_namespace *user_ns, const struct posix_acl *acl, |
2206 |
- void *buffer, size_t size) |
2207 |
- { |
2208 |
- posix_acl_xattr_header *ext_acl = (posix_acl_xattr_header *)buffer; |
2209 |
-- posix_acl_xattr_entry *ext_entry = ext_acl->a_entries; |
2210 |
-+ posix_acl_xattr_entry *ext_entry; |
2211 |
- int real_size, n; |
2212 |
- |
2213 |
- real_size = posix_acl_xattr_size(acl->a_count); |
2214 |
-@@ -731,7 +731,8 @@ posix_acl_to_xattr(struct user_namespace *user_ns, const struct posix_acl *acl, |
2215 |
- return real_size; |
2216 |
- if (real_size > size) |
2217 |
- return -ERANGE; |
2218 |
-- |
2219 |
-+ |
2220 |
-+ ext_entry = ext_acl->a_entries; |
2221 |
- ext_acl->a_version = cpu_to_le32(POSIX_ACL_XATTR_VERSION); |
2222 |
- |
2223 |
- for (n=0; n < acl->a_count; n++, ext_entry++) { |
2224 |
-diff --git a/fs/xfs/xfs_da_btree.c b/fs/xfs/xfs_da_btree.c |
2225 |
-index 796272a2e129..e69d57be866b 100644 |
2226 |
---- a/fs/xfs/xfs_da_btree.c |
2227 |
-+++ b/fs/xfs/xfs_da_btree.c |
2228 |
-@@ -1295,7 +1295,7 @@ xfs_da3_fixhashpath( |
2229 |
- node = blk->bp->b_addr; |
2230 |
- dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
2231 |
- btree = dp->d_ops->node_tree_p(node); |
2232 |
-- if (be32_to_cpu(btree->hashval) == lasthash) |
2233 |
-+ if (be32_to_cpu(btree[blk->index].hashval) == lasthash) |
2234 |
- break; |
2235 |
- blk->hashval = lasthash; |
2236 |
- btree[blk->index].hashval = cpu_to_be32(lasthash); |
2237 |
-diff --git a/include/linux/backing-dev.h b/include/linux/backing-dev.h |
2238 |
-index 24819001f5c8..e488e9459a93 100644 |
2239 |
---- a/include/linux/backing-dev.h |
2240 |
-+++ b/include/linux/backing-dev.h |
2241 |
-@@ -95,7 +95,7 @@ struct backing_dev_info { |
2242 |
- unsigned int max_ratio, max_prop_frac; |
2243 |
- |
2244 |
- struct bdi_writeback wb; /* default writeback info for this bdi */ |
2245 |
-- spinlock_t wb_lock; /* protects work_list */ |
2246 |
-+ spinlock_t wb_lock; /* protects work_list & wb.dwork scheduling */ |
2247 |
- |
2248 |
- struct list_head work_list; |
2249 |
- |
2250 |
-diff --git a/include/linux/tty.h b/include/linux/tty.h |
2251 |
-index 90b4fdc8a61f..b90b5c221ff0 100644 |
2252 |
---- a/include/linux/tty.h |
2253 |
-+++ b/include/linux/tty.h |
2254 |
-@@ -208,7 +208,7 @@ struct tty_port { |
2255 |
- wait_queue_head_t delta_msr_wait; /* Modem status change */ |
2256 |
- unsigned long flags; /* TTY flags ASY_*/ |
2257 |
- unsigned char console:1, /* port is a console */ |
2258 |
-- low_latency:1; /* direct buffer flush */ |
2259 |
-+ low_latency:1; /* optional: tune for latency */ |
2260 |
- struct mutex mutex; /* Locking */ |
2261 |
- struct mutex buf_mutex; /* Buffer alloc lock */ |
2262 |
- unsigned char *xmit_buf; /* Optional buffer */ |
2263 |
-diff --git a/kernel/exit.c b/kernel/exit.c |
2264 |
-index 1e77fc645317..81b3d6789ee8 100644 |
2265 |
---- a/kernel/exit.c |
2266 |
-+++ b/kernel/exit.c |
2267 |
-@@ -560,9 +560,6 @@ static void reparent_leader(struct task_struct *father, struct task_struct *p, |
2268 |
- struct list_head *dead) |
2269 |
- { |
2270 |
- list_move_tail(&p->sibling, &p->real_parent->children); |
2271 |
-- |
2272 |
-- if (p->exit_state == EXIT_DEAD) |
2273 |
-- return; |
2274 |
- /* |
2275 |
- * If this is a threaded reparent there is no need to |
2276 |
- * notify anyone anything has happened. |
2277 |
-@@ -570,9 +567,19 @@ static void reparent_leader(struct task_struct *father, struct task_struct *p, |
2278 |
- if (same_thread_group(p->real_parent, father)) |
2279 |
- return; |
2280 |
- |
2281 |
-- /* We don't want people slaying init. */ |
2282 |
-+ /* |
2283 |
-+ * We don't want people slaying init. |
2284 |
-+ * |
2285 |
-+ * Note: we do this even if it is EXIT_DEAD, wait_task_zombie() |
2286 |
-+ * can change ->exit_state to EXIT_ZOMBIE. If this is the final |
2287 |
-+ * state, do_notify_parent() was already called and ->exit_signal |
2288 |
-+ * doesn't matter. |
2289 |
-+ */ |
2290 |
- p->exit_signal = SIGCHLD; |
2291 |
- |
2292 |
-+ if (p->exit_state == EXIT_DEAD) |
2293 |
-+ return; |
2294 |
-+ |
2295 |
- /* If it has exited notify the new parent about this child's death. */ |
2296 |
- if (!p->ptrace && |
2297 |
- p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) { |
2298 |
-@@ -784,6 +791,8 @@ void do_exit(long code) |
2299 |
- exit_shm(tsk); |
2300 |
- exit_files(tsk); |
2301 |
- exit_fs(tsk); |
2302 |
-+ if (group_dead) |
2303 |
-+ disassociate_ctty(1); |
2304 |
- exit_task_namespaces(tsk); |
2305 |
- exit_task_work(tsk); |
2306 |
- check_stack_usage(); |
2307 |
-@@ -799,13 +808,9 @@ void do_exit(long code) |
2308 |
- |
2309 |
- cgroup_exit(tsk, 1); |
2310 |
- |
2311 |
-- if (group_dead) |
2312 |
-- disassociate_ctty(1); |
2313 |
-- |
2314 |
- module_put(task_thread_info(tsk)->exec_domain->module); |
2315 |
- |
2316 |
- proc_exit_connector(tsk); |
2317 |
-- |
2318 |
- /* |
2319 |
- * FIXME: do that only when needed, using sched_exit tracepoint |
2320 |
- */ |
2321 |
-diff --git a/kernel/pid_namespace.c b/kernel/pid_namespace.c |
2322 |
-index 06c62de9c711..db95d8eb761b 100644 |
2323 |
---- a/kernel/pid_namespace.c |
2324 |
-+++ b/kernel/pid_namespace.c |
2325 |
-@@ -318,7 +318,9 @@ static void *pidns_get(struct task_struct *task) |
2326 |
- struct pid_namespace *ns; |
2327 |
- |
2328 |
- rcu_read_lock(); |
2329 |
-- ns = get_pid_ns(task_active_pid_ns(task)); |
2330 |
-+ ns = task_active_pid_ns(task); |
2331 |
-+ if (ns) |
2332 |
-+ get_pid_ns(ns); |
2333 |
- rcu_read_unlock(); |
2334 |
- |
2335 |
- return ns; |
2336 |
-diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c |
2337 |
-index dd06439b9c84..80a57afd8647 100644 |
2338 |
---- a/kernel/user_namespace.c |
2339 |
-+++ b/kernel/user_namespace.c |
2340 |
-@@ -152,7 +152,7 @@ static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count) |
2341 |
- |
2342 |
- /* Find the matching extent */ |
2343 |
- extents = map->nr_extents; |
2344 |
-- smp_read_barrier_depends(); |
2345 |
-+ smp_rmb(); |
2346 |
- for (idx = 0; idx < extents; idx++) { |
2347 |
- first = map->extent[idx].first; |
2348 |
- last = first + map->extent[idx].count - 1; |
2349 |
-@@ -176,7 +176,7 @@ static u32 map_id_down(struct uid_gid_map *map, u32 id) |
2350 |
- |
2351 |
- /* Find the matching extent */ |
2352 |
- extents = map->nr_extents; |
2353 |
-- smp_read_barrier_depends(); |
2354 |
-+ smp_rmb(); |
2355 |
- for (idx = 0; idx < extents; idx++) { |
2356 |
- first = map->extent[idx].first; |
2357 |
- last = first + map->extent[idx].count - 1; |
2358 |
-@@ -199,7 +199,7 @@ static u32 map_id_up(struct uid_gid_map *map, u32 id) |
2359 |
- |
2360 |
- /* Find the matching extent */ |
2361 |
- extents = map->nr_extents; |
2362 |
-- smp_read_barrier_depends(); |
2363 |
-+ smp_rmb(); |
2364 |
- for (idx = 0; idx < extents; idx++) { |
2365 |
- first = map->extent[idx].lower_first; |
2366 |
- last = first + map->extent[idx].count - 1; |
2367 |
-@@ -615,9 +615,8 @@ static ssize_t map_write(struct file *file, const char __user *buf, |
2368 |
- * were written before the count of the extents. |
2369 |
- * |
2370 |
- * To achieve this smp_wmb() is used on guarantee the write |
2371 |
-- * order and smp_read_barrier_depends() is guaranteed that we |
2372 |
-- * don't have crazy architectures returning stale data. |
2373 |
-- * |
2374 |
-+ * order and smp_rmb() is guaranteed that we don't have crazy |
2375 |
-+ * architectures returning stale data. |
2376 |
- */ |
2377 |
- mutex_lock(&id_map_mutex); |
2378 |
- |
2379 |
-diff --git a/mm/backing-dev.c b/mm/backing-dev.c |
2380 |
-index ce682f7a4f29..09d9591b7708 100644 |
2381 |
---- a/mm/backing-dev.c |
2382 |
-+++ b/mm/backing-dev.c |
2383 |
-@@ -288,13 +288,19 @@ int bdi_has_dirty_io(struct backing_dev_info *bdi) |
2384 |
- * Note, we wouldn't bother setting up the timer, but this function is on the |
2385 |
- * fast-path (used by '__mark_inode_dirty()'), so we save few context switches |
2386 |
- * by delaying the wake-up. |
2387 |
-+ * |
2388 |
-+ * We have to be careful not to postpone flush work if it is scheduled for |
2389 |
-+ * earlier. Thus we use queue_delayed_work(). |
2390 |
- */ |
2391 |
- void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi) |
2392 |
- { |
2393 |
- unsigned long timeout; |
2394 |
- |
2395 |
- timeout = msecs_to_jiffies(dirty_writeback_interval * 10); |
2396 |
-- mod_delayed_work(bdi_wq, &bdi->wb.dwork, timeout); |
2397 |
-+ spin_lock_bh(&bdi->wb_lock); |
2398 |
-+ if (test_bit(BDI_registered, &bdi->state)) |
2399 |
-+ queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout); |
2400 |
-+ spin_unlock_bh(&bdi->wb_lock); |
2401 |
- } |
2402 |
- |
2403 |
- /* |
2404 |
-@@ -307,9 +313,6 @@ static void bdi_remove_from_list(struct backing_dev_info *bdi) |
2405 |
- spin_unlock_bh(&bdi_lock); |
2406 |
- |
2407 |
- synchronize_rcu_expedited(); |
2408 |
-- |
2409 |
-- /* bdi_list is now unused, clear it to mark @bdi dying */ |
2410 |
-- INIT_LIST_HEAD(&bdi->bdi_list); |
2411 |
- } |
2412 |
- |
2413 |
- int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
2414 |
-@@ -360,6 +363,11 @@ static void bdi_wb_shutdown(struct backing_dev_info *bdi) |
2415 |
- */ |
2416 |
- bdi_remove_from_list(bdi); |
2417 |
- |
2418 |
-+ /* Make sure nobody queues further work */ |
2419 |
-+ spin_lock_bh(&bdi->wb_lock); |
2420 |
-+ clear_bit(BDI_registered, &bdi->state); |
2421 |
-+ spin_unlock_bh(&bdi->wb_lock); |
2422 |
-+ |
2423 |
- /* |
2424 |
- * Drain work list and shutdown the delayed_work. At this point, |
2425 |
- * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi |
2426 |
-diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c |
2427 |
-index 5f812455a450..60828cf02eb8 100644 |
2428 |
---- a/net/bluetooth/hci_event.c |
2429 |
-+++ b/net/bluetooth/hci_event.c |
2430 |
-@@ -3593,7 +3593,13 @@ static void hci_le_ltk_request_evt(struct hci_dev *hdev, struct sk_buff *skb) |
2431 |
- |
2432 |
- hci_send_cmd(hdev, HCI_OP_LE_LTK_REPLY, sizeof(cp), &cp); |
2433 |
- |
2434 |
-- if (ltk->type & HCI_SMP_STK) { |
2435 |
-+ /* Ref. Bluetooth Core SPEC pages 1975 and 2004. STK is a |
2436 |
-+ * temporary key used to encrypt a connection following |
2437 |
-+ * pairing. It is used during the Encrypted Session Setup to |
2438 |
-+ * distribute the keys. Later, security can be re-established |
2439 |
-+ * using a distributed LTK. |
2440 |
-+ */ |
2441 |
-+ if (ltk->type == HCI_SMP_STK_SLAVE) { |
2442 |
- list_del(<k->list); |
2443 |
- kfree(ltk); |
2444 |
- } |
2445 |
-diff --git a/security/integrity/ima/ima.h b/security/integrity/ima/ima.h |
2446 |
-index 0356e1d437ca..f79fa8be203c 100644 |
2447 |
---- a/security/integrity/ima/ima.h |
2448 |
-+++ b/security/integrity/ima/ima.h |
2449 |
-@@ -27,7 +27,7 @@ |
2450 |
- #include "../integrity.h" |
2451 |
- |
2452 |
- enum ima_show_type { IMA_SHOW_BINARY, IMA_SHOW_BINARY_NO_FIELD_LEN, |
2453 |
-- IMA_SHOW_ASCII }; |
2454 |
-+ IMA_SHOW_BINARY_OLD_STRING_FMT, IMA_SHOW_ASCII }; |
2455 |
- enum tpm_pcrs { TPM_PCR0 = 0, TPM_PCR8 = 8 }; |
2456 |
- |
2457 |
- /* digest size for IMA, fits SHA1 or MD5 */ |
2458 |
-diff --git a/security/integrity/ima/ima_fs.c b/security/integrity/ima/ima_fs.c |
2459 |
-index db01125926bd..468a3ba3c539 100644 |
2460 |
---- a/security/integrity/ima/ima_fs.c |
2461 |
-+++ b/security/integrity/ima/ima_fs.c |
2462 |
-@@ -160,6 +160,8 @@ static int ima_measurements_show(struct seq_file *m, void *v) |
2463 |
- |
2464 |
- if (is_ima_template && strcmp(field->field_id, "d") == 0) |
2465 |
- show = IMA_SHOW_BINARY_NO_FIELD_LEN; |
2466 |
-+ if (is_ima_template && strcmp(field->field_id, "n") == 0) |
2467 |
-+ show = IMA_SHOW_BINARY_OLD_STRING_FMT; |
2468 |
- field->field_show(m, show, &e->template_data[i]); |
2469 |
- } |
2470 |
- return 0; |
2471 |
-diff --git a/security/integrity/ima/ima_template_lib.c b/security/integrity/ima/ima_template_lib.c |
2472 |
-index 1683bbf289a4..e8592e7bfc21 100644 |
2473 |
---- a/security/integrity/ima/ima_template_lib.c |
2474 |
-+++ b/security/integrity/ima/ima_template_lib.c |
2475 |
-@@ -109,13 +109,16 @@ static void ima_show_template_data_binary(struct seq_file *m, |
2476 |
- enum data_formats datafmt, |
2477 |
- struct ima_field_data *field_data) |
2478 |
- { |
2479 |
-+ u32 len = (show == IMA_SHOW_BINARY_OLD_STRING_FMT) ? |
2480 |
-+ strlen(field_data->data) : field_data->len; |
2481 |
-+ |
2482 |
- if (show != IMA_SHOW_BINARY_NO_FIELD_LEN) |
2483 |
-- ima_putc(m, &field_data->len, sizeof(u32)); |
2484 |
-+ ima_putc(m, &len, sizeof(len)); |
2485 |
- |
2486 |
-- if (!field_data->len) |
2487 |
-+ if (!len) |
2488 |
- return; |
2489 |
- |
2490 |
-- ima_putc(m, field_data->data, field_data->len); |
2491 |
-+ ima_putc(m, field_data->data, len); |
2492 |
- } |
2493 |
- |
2494 |
- static void ima_show_template_field_data(struct seq_file *m, |
2495 |
-@@ -129,6 +132,7 @@ static void ima_show_template_field_data(struct seq_file *m, |
2496 |
- break; |
2497 |
- case IMA_SHOW_BINARY: |
2498 |
- case IMA_SHOW_BINARY_NO_FIELD_LEN: |
2499 |
-+ case IMA_SHOW_BINARY_OLD_STRING_FMT: |
2500 |
- ima_show_template_data_binary(m, show, datafmt, field_data); |
2501 |
- break; |
2502 |
- default: |
2503 |
|
2504 |
Deleted: genpatches-2.6/trunk/3.15/4200_fbcondecor-0.9.6.patch |
2505 |
=================================================================== |
2506 |
--- genpatches-2.6/trunk/3.15/4200_fbcondecor-0.9.6.patch 2014-05-05 14:22:27 UTC (rev 2771) |
2507 |
+++ genpatches-2.6/trunk/3.15/4200_fbcondecor-0.9.6.patch 2014-05-05 14:28:48 UTC (rev 2772) |
2508 |
@@ -1,2177 +0,0 @@ |
2509 |
-diff --git a/Documentation/fb/00-INDEX b/Documentation/fb/00-INDEX |
2510 |
-index 30a7054..9b6a733 100644 |
2511 |
---- a/Documentation/fb/00-INDEX |
2512 |
-+++ b/Documentation/fb/00-INDEX |
2513 |
-@@ -21,6 +21,8 @@ ep93xx-fb.txt |
2514 |
- - info on the driver for EP93xx LCD controller. |
2515 |
- fbcon.txt |
2516 |
- - intro to and usage guide for the framebuffer console (fbcon). |
2517 |
-+fbcondecor.txt |
2518 |
-+ - info on the Framebuffer Console Decoration |
2519 |
- framebuffer.txt |
2520 |
- - introduction to frame buffer devices. |
2521 |
- gxfb.txt |
2522 |
-diff --git a/Documentation/fb/fbcondecor.txt b/Documentation/fb/fbcondecor.txt |
2523 |
-new file mode 100644 |
2524 |
-index 0000000..15889f3 |
2525 |
---- /dev/null |
2526 |
-+++ b/Documentation/fb/fbcondecor.txt |
2527 |
-@@ -0,0 +1,207 @@ |
2528 |
-+What is it? |
2529 |
-+----------- |
2530 |
-+ |
2531 |
-+The framebuffer decorations are a kernel feature which allows displaying a |
2532 |
-+background picture on selected consoles. |
2533 |
-+ |
2534 |
-+What do I need to get it to work? |
2535 |
-+--------------------------------- |
2536 |
-+ |
2537 |
-+To get fbcondecor up-and-running you will have to: |
2538 |
-+ 1) get a copy of splashutils [1] or a similar program |
2539 |
-+ 2) get some fbcondecor themes |
2540 |
-+ 3) build the kernel helper program |
2541 |
-+ 4) build your kernel with the FB_CON_DECOR option enabled. |
2542 |
-+ |
2543 |
-+To get fbcondecor operational right after fbcon initialization is finished, you |
2544 |
-+will have to include a theme and the kernel helper into your initramfs image. |
2545 |
-+Please refer to splashutils documentation for instructions on how to do that. |
2546 |
-+ |
2547 |
-+[1] The splashutils package can be downloaded from: |
2548 |
-+ http://dev.gentoo.org/~spock/projects/splashutils/ |
2549 |
-+ |
2550 |
-+The userspace helper |
2551 |
-+-------------------- |
2552 |
-+ |
2553 |
-+The userspace fbcondecor helper (by default: /sbin/fbcondecor_helper) is called by the |
2554 |
-+kernel whenever an important event occurs and the kernel needs some kind of |
2555 |
-+job to be carried out. Important events include console switches and video |
2556 |
-+mode switches (the kernel requests background images and configuration |
2557 |
-+parameters for the current console). The fbcondecor helper must be accessible at |
2558 |
-+all times. If it's not, fbcondecor will be switched off automatically. |
2559 |
-+ |
2560 |
-+It's possible to set path to the fbcondecor helper by writing it to |
2561 |
-+/proc/sys/kernel/fbcondecor. |
2562 |
-+ |
2563 |
-+***************************************************************************** |
2564 |
-+ |
2565 |
-+The information below is mostly technical stuff. There's probably no need to |
2566 |
-+read it unless you plan to develop a userspace helper. |
2567 |
-+ |
2568 |
-+The fbcondecor protocol |
2569 |
-+----------------------- |
2570 |
-+ |
2571 |
-+The fbcondecor protocol defines a communication interface between the kernel and |
2572 |
-+the userspace fbcondecor helper. |
2573 |
-+ |
2574 |
-+The kernel side is responsible for: |
2575 |
-+ |
2576 |
-+ * rendering console text, using an image as a background (instead of a |
2577 |
-+ standard solid color fbcon uses), |
2578 |
-+ * accepting commands from the user via ioctls on the fbcondecor device, |
2579 |
-+ * calling the userspace helper to set things up as soon as the fb subsystem |
2580 |
-+ is initialized. |
2581 |
-+ |
2582 |
-+The userspace helper is responsible for everything else, including parsing |
2583 |
-+configuration files, decompressing the image files whenever the kernel needs |
2584 |
-+it, and communicating with the kernel if necessary. |
2585 |
-+ |
2586 |
-+The fbcondecor protocol specifies how communication is done in both ways: |
2587 |
-+kernel->userspace and userspace->helper. |
2588 |
-+ |
2589 |
-+Kernel -> Userspace |
2590 |
-+------------------- |
2591 |
-+ |
2592 |
-+The kernel communicates with the userspace helper by calling it and specifying |
2593 |
-+the task to be done in a series of arguments. |
2594 |
-+ |
2595 |
-+The arguments follow the pattern: |
2596 |
-+<fbcondecor protocol version> <command> <parameters> |
2597 |
-+ |
2598 |
-+All commands defined in fbcondecor protocol v2 have the following parameters: |
2599 |
-+ virtual console |
2600 |
-+ framebuffer number |
2601 |
-+ theme |
2602 |
-+ |
2603 |
-+Fbcondecor protocol v1 specified an additional 'fbcondecor mode' after the |
2604 |
-+framebuffer number. Fbcondecor protocol v1 is deprecated and should not be used. |
2605 |
-+ |
2606 |
-+Fbcondecor protocol v2 specifies the following commands: |
2607 |
-+ |
2608 |
-+getpic |
2609 |
-+------ |
2610 |
-+ The kernel issues this command to request image data. It's up to the |
2611 |
-+ userspace helper to find a background image appropriate for the specified |
2612 |
-+ theme and the current resolution. The userspace helper should respond by |
2613 |
-+ issuing the FBIOCONDECOR_SETPIC ioctl. |
2614 |
-+ |
2615 |
-+init |
2616 |
-+---- |
2617 |
-+ The kernel issues this command after the fbcondecor device is created and |
2618 |
-+ the fbcondecor interface is initialized. Upon receiving 'init', the userspace |
2619 |
-+ helper should parse the kernel command line (/proc/cmdline) or otherwise |
2620 |
-+ decide whether fbcondecor is to be activated. |
2621 |
-+ |
2622 |
-+ To activate fbcondecor on the first console the helper should issue the |
2623 |
-+ FBIOCONDECOR_SETCFG, FBIOCONDECOR_SETPIC and FBIOCONDECOR_SETSTATE commands, |
2624 |
-+ in the above-mentioned order. |
2625 |
-+ |
2626 |
-+ When the userspace helper is called in an early phase of the boot process |
2627 |
-+ (right after the initialization of fbcon), no filesystems will be mounted. |
2628 |
-+ The helper program should mount sysfs and then create the appropriate |
2629 |
-+ framebuffer, fbcondecor and tty0 devices (if they don't already exist) to get |
2630 |
-+ current display settings and to be able to communicate with the kernel side. |
2631 |
-+ It should probably also mount the procfs to be able to parse the kernel |
2632 |
-+ command line parameters. |
2633 |
-+ |
2634 |
-+ Note that the console sem is not held when the kernel calls fbcondecor_helper |
2635 |
-+ with the 'init' command. The fbcondecor helper should perform all ioctls with |
2636 |
-+ origin set to FBCON_DECOR_IO_ORIG_USER. |
2637 |
-+ |
2638 |
-+modechange |
2639 |
-+---------- |
2640 |
-+ The kernel issues this command on a mode change. The helper's response should |
2641 |
-+ be similar to the response to the 'init' command. Note that this time the |
2642 |
-+ console sem is held and all ioctls must be performed with origin set to |
2643 |
-+ FBCON_DECOR_IO_ORIG_KERNEL. |
2644 |
-+ |
2645 |
-+ |
2646 |
-+Userspace -> Kernel |
2647 |
-+------------------- |
2648 |
-+ |
2649 |
-+Userspace programs can communicate with fbcondecor via ioctls on the |
2650 |
-+fbcondecor device. These ioctls are to be used by both the userspace helper |
2651 |
-+(called only by the kernel) and userspace configuration tools (run by the users). |
2652 |
-+ |
2653 |
-+The fbcondecor helper should set the origin field to FBCON_DECOR_IO_ORIG_KERNEL |
2654 |
-+when doing the appropriate ioctls. All userspace configuration tools should |
2655 |
-+use FBCON_DECOR_IO_ORIG_USER. Failure to set the appropriate value in the origin |
2656 |
-+field when performing ioctls from the kernel helper will most likely result |
2657 |
-+in a console deadlock. |
2658 |
-+ |
2659 |
-+FBCON_DECOR_IO_ORIG_KERNEL instructs fbcondecor not to try to acquire the console |
2660 |
-+semaphore. Not surprisingly, FBCON_DECOR_IO_ORIG_USER instructs it to acquire |
2661 |
-+the console sem. |
2662 |
-+ |
2663 |
-+The framebuffer console decoration provides the following ioctls (all defined in |
2664 |
-+linux/fb.h): |
2665 |
-+ |
2666 |
-+FBIOCONDECOR_SETPIC |
2667 |
-+description: loads a background picture for a virtual console |
2668 |
-+argument: struct fbcon_decor_iowrapper*; data: struct fb_image* |
2669 |
-+notes: |
2670 |
-+If called for consoles other than the current foreground one, the picture data |
2671 |
-+will be ignored. |
2672 |
-+ |
2673 |
-+If the current virtual console is running in a 8-bpp mode, the cmap substruct |
2674 |
-+of fb_image has to be filled appropriately: start should be set to 16 (first |
2675 |
-+16 colors are reserved for fbcon), len to a value <= 240 and red, green and |
2676 |
-+blue should point to valid cmap data. The transp field is ingored. The fields |
2677 |
-+dx, dy, bg_color, fg_color in fb_image are ignored as well. |
2678 |
-+ |
2679 |
-+FBIOCONDECOR_SETCFG |
2680 |
-+description: sets the fbcondecor config for a virtual console |
2681 |
-+argument: struct fbcon_decor_iowrapper*; data: struct vc_decor* |
2682 |
-+notes: The structure has to be filled with valid data. |
2683 |
-+ |
2684 |
-+FBIOCONDECOR_GETCFG |
2685 |
-+description: gets the fbcondecor config for a virtual console |
2686 |
-+argument: struct fbcon_decor_iowrapper*; data: struct vc_decor* |
2687 |
-+ |
2688 |
-+FBIOCONDECOR_SETSTATE |
2689 |
-+description: sets the fbcondecor state for a virtual console |
2690 |
-+argument: struct fbcon_decor_iowrapper*; data: unsigned int* |
2691 |
-+ values: 0 = disabled, 1 = enabled. |
2692 |
-+ |
2693 |
-+FBIOCONDECOR_GETSTATE |
2694 |
-+description: gets the fbcondecor state for a virtual console |
2695 |
-+argument: struct fbcon_decor_iowrapper*; data: unsigned int* |
2696 |
-+ values: as in FBIOCONDECOR_SETSTATE |
2697 |
-+ |
2698 |
-+Info on used structures: |
2699 |
-+ |
2700 |
-+Definition of struct vc_decor can be found in linux/console_decor.h. It's |
2701 |
-+heavily commented. Note that the 'theme' field should point to a string |
2702 |
-+no longer than FBCON_DECOR_THEME_LEN. When FBIOCONDECOR_GETCFG call is |
2703 |
-+performed, the theme field should point to a char buffer of length |
2704 |
-+FBCON_DECOR_THEME_LEN. |
2705 |
-+ |
2706 |
-+Definition of struct fbcon_decor_iowrapper can be found in linux/fb.h. |
2707 |
-+The fields in this struct have the following meaning: |
2708 |
-+ |
2709 |
-+vc: |
2710 |
-+Virtual console number. |
2711 |
-+ |
2712 |
-+origin: |
2713 |
-+Specifies if the ioctl is performed as a response to a kernel request. The |
2714 |
-+fbcondecor helper should set this field to FBCON_DECOR_IO_ORIG_KERNEL, userspace |
2715 |
-+programs should set it to FBCON_DECOR_IO_ORIG_USER. This field is necessary to |
2716 |
-+avoid console semaphore deadlocks. |
2717 |
-+ |
2718 |
-+data: |
2719 |
-+Pointer to a data structure appropriate for the performed ioctl. Type of |
2720 |
-+the data struct is specified in the ioctls description. |
2721 |
-+ |
2722 |
-+***************************************************************************** |
2723 |
-+ |
2724 |
-+Credit |
2725 |
-+------ |
2726 |
-+ |
2727 |
-+Original 'bootsplash' project & implementation by: |
2728 |
-+ Volker Poplawski <volker@×××××××××.de>, Stefan Reinauer <stepan@××××.de>, |
2729 |
-+ Steffen Winterfeldt <snwint@××××.de>, Michael Schroeder <mls@××××.de>, |
2730 |
-+ Ken Wimer <wimer@××××.de>. |
2731 |
-+ |
2732 |
-+Fbcondecor, fbcondecor protocol design, current implementation & docs by: |
2733 |
-+ Michal Januszewski <spock@g.o> |
2734 |
-+ |
2735 |
-diff --git a/drivers/Makefile b/drivers/Makefile |
2736 |
-index 95952c8..b55db6d 100644 |
2737 |
---- a/drivers/Makefile |
2738 |
-+++ b/drivers/Makefile |
2739 |
-@@ -16,4 +16,8 @@ obj-$(CONFIG_PCI) += pci/ |
2740 |
- obj-$(CONFIG_PARISC) += parisc/ |
2741 |
- obj-$(CONFIG_RAPIDIO) += rapidio/ |
2742 |
-+# tty/ comes before char/ so that the VT console is the boot-time |
2743 |
-+# default. |
2744 |
-+obj-y += tty/ |
2745 |
-+obj-y += char/ |
2746 |
- obj-y += video/ |
2747 |
- obj-y += idle/ |
2748 |
-@@ -37,11 +41,6 @@ obj-$(CONFIG_XEN) += xen/ |
2749 |
- # regulators early, since some subsystems rely on them to initialize |
2750 |
- obj-$(CONFIG_REGULATOR) += regulator/ |
2751 |
- |
2752 |
--# tty/ comes before char/ so that the VT console is the boot-time |
2753 |
--# default. |
2754 |
--obj-y += tty/ |
2755 |
--obj-y += char/ |
2756 |
-- |
2757 |
- # gpu/ comes after char for AGP vs DRM startup |
2758 |
- obj-y += gpu/ |
2759 |
- |
2760 |
-diff --git a/drivers/video/Kconfig b/drivers/video/Kconfig |
2761 |
-index a290be5..3a4ca32 100644 |
2762 |
---- a/drivers/video/Kconfig |
2763 |
-+++ b/drivers/video/Kconfig |
2764 |
-@@ -1229,7 +1229,6 @@ config FB_MATROX |
2765 |
- select FB_CFB_FILLRECT |
2766 |
- select FB_CFB_COPYAREA |
2767 |
- select FB_CFB_IMAGEBLIT |
2768 |
-- select FB_TILEBLITTING |
2769 |
- select FB_MACMODES if PPC_PMAC |
2770 |
- ---help--- |
2771 |
- Say Y here if you have a Matrox Millennium, Matrox Millennium II, |
2772 |
-diff --git a/drivers/video/console/Kconfig b/drivers/video/console/Kconfig |
2773 |
-index c2d11fe..1be9de4 100644 |
2774 |
---- a/drivers/video/console/Kconfig |
2775 |
-+++ b/drivers/video/console/Kconfig |
2776 |
-@@ -120,6 +120,19 @@ config FRAMEBUFFER_CONSOLE_ROTATION |
2777 |
- such that other users of the framebuffer will remain normally |
2778 |
- oriented. |
2779 |
- |
2780 |
-+config FB_CON_DECOR |
2781 |
-+ bool "Support for the Framebuffer Console Decorations" |
2782 |
-+ depends on FRAMEBUFFER_CONSOLE=y && !FB_TILEBLITTING |
2783 |
-+ default n |
2784 |
-+ ---help--- |
2785 |
-+ This option enables support for framebuffer console decorations which |
2786 |
-+ makes it possible to display images in the background of the system |
2787 |
-+ consoles. Note that userspace utilities are necessary in order to take |
2788 |
-+ advantage of these features. Refer to Documentation/fb/fbcondecor.txt |
2789 |
-+ for more information. |
2790 |
-+ |
2791 |
-+ If unsure, say N. |
2792 |
-+ |
2793 |
- config STI_CONSOLE |
2794 |
- bool "STI text console" |
2795 |
- depends on PARISC |
2796 |
---- a/drivers/video/console/Makefile 2013-08-26 14:02:39.905817618 -0400 |
2797 |
-+++ b/drivers/video/console/Makefile 2013-08-26 14:05:06.258848595 -0400 |
2798 |
-@@ -16,4 +16,5 @@ obj-$(CONFIG_FRAMEBUFFER_CONSOLE) += |
2799 |
- fbcon_ccw.o |
2800 |
- endif |
2801 |
- |
2802 |
-+obj-$(CONFIG_FB_CON_DECOR) += fbcondecor.o cfbcondecor.o |
2803 |
- obj-$(CONFIG_FB_STI) += sticore.o |
2804 |
-diff --git a/drivers/video/console/bitblit.c b/drivers/video/console/bitblit.c |
2805 |
-index 28b1a83..33712c0 100644 |
2806 |
---- a/drivers/video/console/bitblit.c |
2807 |
-+++ b/drivers/video/console/bitblit.c |
2808 |
-@@ -18,6 +18,7 @@ |
2809 |
- #include <linux/console.h> |
2810 |
- #include <asm/types.h> |
2811 |
- #include "fbcon.h" |
2812 |
-+#include "fbcondecor.h" |
2813 |
- |
2814 |
- /* |
2815 |
- * Accelerated handlers. |
2816 |
-@@ -55,6 +56,13 @@ static void bit_bmove(struct vc_data *vc, struct fb_info *info, int sy, |
2817 |
- area.height = height * vc->vc_font.height; |
2818 |
- area.width = width * vc->vc_font.width; |
2819 |
- |
2820 |
-+ if (fbcon_decor_active(info, vc)) { |
2821 |
-+ area.sx += vc->vc_decor.tx; |
2822 |
-+ area.sy += vc->vc_decor.ty; |
2823 |
-+ area.dx += vc->vc_decor.tx; |
2824 |
-+ area.dy += vc->vc_decor.ty; |
2825 |
-+ } |
2826 |
-+ |
2827 |
- info->fbops->fb_copyarea(info, &area); |
2828 |
- } |
2829 |
- |
2830 |
-@@ -380,11 +388,15 @@ static void bit_cursor(struct vc_data *vc, struct fb_info *info, int mode, |
2831 |
- cursor.image.depth = 1; |
2832 |
- cursor.rop = ROP_XOR; |
2833 |
- |
2834 |
-- if (info->fbops->fb_cursor) |
2835 |
-- err = info->fbops->fb_cursor(info, &cursor); |
2836 |
-+ if (fbcon_decor_active(info, vc)) { |
2837 |
-+ fbcon_decor_cursor(info, &cursor); |
2838 |
-+ } else { |
2839 |
-+ if (info->fbops->fb_cursor) |
2840 |
-+ err = info->fbops->fb_cursor(info, &cursor); |
2841 |
- |
2842 |
-- if (err) |
2843 |
-- soft_cursor(info, &cursor); |
2844 |
-+ if (err) |
2845 |
-+ soft_cursor(info, &cursor); |
2846 |
-+ } |
2847 |
- |
2848 |
- ops->cursor_reset = 0; |
2849 |
- } |
2850 |
-diff --git a/drivers/video/console/cfbcondecor.c b/drivers/video/console/cfbcondecor.c |
2851 |
-new file mode 100644 |
2852 |
-index 0000000..09381d3 |
2853 |
---- /dev/null |
2854 |
-+++ b/drivers/video/console/cfbcondecor.c |
2855 |
-@@ -0,0 +1,471 @@ |
2856 |
-+/* |
2857 |
-+ * linux/drivers/video/cfbcon_decor.c -- Framebuffer decor render functions |
2858 |
-+ * |
2859 |
-+ * Copyright (C) 2004 Michal Januszewski <spock@g.o> |
2860 |
-+ * |
2861 |
-+ * Code based upon "Bootdecor" (C) 2001-2003 |
2862 |
-+ * Volker Poplawski <volker@×××××××××.de>, |
2863 |
-+ * Stefan Reinauer <stepan@××××.de>, |
2864 |
-+ * Steffen Winterfeldt <snwint@××××.de>, |
2865 |
-+ * Michael Schroeder <mls@××××.de>, |
2866 |
-+ * Ken Wimer <wimer@××××.de>. |
2867 |
-+ * |
2868 |
-+ * This file is subject to the terms and conditions of the GNU General Public |
2869 |
-+ * License. See the file COPYING in the main directory of this archive for |
2870 |
-+ * more details. |
2871 |
-+ */ |
2872 |
-+#include <linux/module.h> |
2873 |
-+#include <linux/types.h> |
2874 |
-+#include <linux/fb.h> |
2875 |
-+#include <linux/selection.h> |
2876 |
-+#include <linux/slab.h> |
2877 |
-+#include <linux/vt_kern.h> |
2878 |
-+#include <asm/irq.h> |
2879 |
-+ |
2880 |
-+#include "fbcon.h" |
2881 |
-+#include "fbcondecor.h" |
2882 |
-+ |
2883 |
-+#define parse_pixel(shift,bpp,type) \ |
2884 |
-+ do { \ |
2885 |
-+ if (d & (0x80 >> (shift))) \ |
2886 |
-+ dd2[(shift)] = fgx; \ |
2887 |
-+ else \ |
2888 |
-+ dd2[(shift)] = transparent ? *(type *)decor_src : bgx; \ |
2889 |
-+ decor_src += (bpp); \ |
2890 |
-+ } while (0) \ |
2891 |
-+ |
2892 |
-+extern int get_color(struct vc_data *vc, struct fb_info *info, |
2893 |
-+ u16 c, int is_fg); |
2894 |
-+ |
2895 |
-+void fbcon_decor_fix_pseudo_pal(struct fb_info *info, struct vc_data *vc) |
2896 |
-+{ |
2897 |
-+ int i, j, k; |
2898 |
-+ int minlen = min(min(info->var.red.length, info->var.green.length), |
2899 |
-+ info->var.blue.length); |
2900 |
-+ u32 col; |
2901 |
-+ |
2902 |
-+ for (j = i = 0; i < 16; i++) { |
2903 |
-+ k = color_table[i]; |
2904 |
-+ |
2905 |
-+ col = ((vc->vc_palette[j++] >> (8-minlen)) |
2906 |
-+ << info->var.red.offset); |
2907 |
-+ col |= ((vc->vc_palette[j++] >> (8-minlen)) |
2908 |
-+ << info->var.green.offset); |
2909 |
-+ col |= ((vc->vc_palette[j++] >> (8-minlen)) |
2910 |
-+ << info->var.blue.offset); |
2911 |
-+ ((u32 *)info->pseudo_palette)[k] = col; |
2912 |
-+ } |
2913 |
-+} |
2914 |
-+ |
2915 |
-+void fbcon_decor_renderc(struct fb_info *info, int ypos, int xpos, int height, |
2916 |
-+ int width, u8* src, u32 fgx, u32 bgx, u8 transparent) |
2917 |
-+{ |
2918 |
-+ unsigned int x, y; |
2919 |
-+ u32 dd; |
2920 |
-+ int bytespp = ((info->var.bits_per_pixel + 7) >> 3); |
2921 |
-+ unsigned int d = ypos * info->fix.line_length + xpos * bytespp; |
2922 |
-+ unsigned int ds = (ypos * info->var.xres + xpos) * bytespp; |
2923 |
-+ u16 dd2[4]; |
2924 |
-+ |
2925 |
-+ u8* decor_src = (u8 *)(info->bgdecor.data + ds); |
2926 |
-+ u8* dst = (u8 *)(info->screen_base + d); |
2927 |
-+ |
2928 |
-+ if ((ypos + height) > info->var.yres || (xpos + width) > info->var.xres) |
2929 |
-+ return; |
2930 |
-+ |
2931 |
-+ for (y = 0; y < height; y++) { |
2932 |
-+ switch (info->var.bits_per_pixel) { |
2933 |
-+ |
2934 |
-+ case 32: |
2935 |
-+ for (x = 0; x < width; x++) { |
2936 |
-+ |
2937 |
-+ if ((x & 7) == 0) |
2938 |
-+ d = *src++; |
2939 |
-+ if (d & 0x80) |
2940 |
-+ dd = fgx; |
2941 |
-+ else |
2942 |
-+ dd = transparent ? |
2943 |
-+ *(u32 *)decor_src : bgx; |
2944 |
-+ |
2945 |
-+ d <<= 1; |
2946 |
-+ decor_src += 4; |
2947 |
-+ fb_writel(dd, dst); |
2948 |
-+ dst += 4; |
2949 |
-+ } |
2950 |
-+ break; |
2951 |
-+ case 24: |
2952 |
-+ for (x = 0; x < width; x++) { |
2953 |
-+ |
2954 |
-+ if ((x & 7) == 0) |
2955 |
-+ d = *src++; |
2956 |
-+ if (d & 0x80) |
2957 |
-+ dd = fgx; |
2958 |
-+ else |
2959 |
-+ dd = transparent ? |
2960 |
-+ (*(u32 *)decor_src & 0xffffff) : bgx; |
2961 |
-+ |
2962 |
-+ d <<= 1; |
2963 |
-+ decor_src += 3; |
2964 |
-+#ifdef __LITTLE_ENDIAN |
2965 |
-+ fb_writew(dd & 0xffff, dst); |
2966 |
-+ dst += 2; |
2967 |
-+ fb_writeb((dd >> 16), dst); |
2968 |
-+#else |
2969 |
-+ fb_writew(dd >> 8, dst); |
2970 |
-+ dst += 2; |
2971 |
-+ fb_writeb(dd & 0xff, dst); |
2972 |
-+#endif |
2973 |
-+ dst++; |
2974 |
-+ } |
2975 |
-+ break; |
2976 |
-+ case 16: |
2977 |
-+ for (x = 0; x < width; x += 2) { |
2978 |
-+ if ((x & 7) == 0) |
2979 |
-+ d = *src++; |
2980 |
-+ |
2981 |
-+ parse_pixel(0, 2, u16); |
2982 |
-+ parse_pixel(1, 2, u16); |
2983 |
-+#ifdef __LITTLE_ENDIAN |
2984 |
-+ dd = dd2[0] | (dd2[1] << 16); |
2985 |
-+#else |
2986 |
-+ dd = dd2[1] | (dd2[0] << 16); |
2987 |
-+#endif |
2988 |
-+ d <<= 2; |
2989 |
-+ fb_writel(dd, dst); |
2990 |
-+ dst += 4; |
2991 |
-+ } |
2992 |
-+ break; |
2993 |
-+ |
2994 |
-+ case 8: |
2995 |
-+ for (x = 0; x < width; x += 4) { |
2996 |
-+ if ((x & 7) == 0) |
2997 |
-+ d = *src++; |
2998 |
-+ |
2999 |
-+ parse_pixel(0, 1, u8); |
3000 |
-+ parse_pixel(1, 1, u8); |
3001 |
-+ parse_pixel(2, 1, u8); |
3002 |
-+ parse_pixel(3, 1, u8); |
3003 |
-+ |
3004 |
-+#ifdef __LITTLE_ENDIAN |
3005 |
-+ dd = dd2[0] | (dd2[1] << 8) | (dd2[2] << 16) | (dd2[3] << 24); |
3006 |
-+#else |
3007 |
-+ dd = dd2[3] | (dd2[2] << 8) | (dd2[1] << 16) | (dd2[0] << 24); |
3008 |
-+#endif |
3009 |
-+ d <<= 4; |
3010 |
-+ fb_writel(dd, dst); |
3011 |
-+ dst += 4; |
3012 |
-+ } |
3013 |
-+ } |
3014 |
-+ |
3015 |
-+ dst += info->fix.line_length - width * bytespp; |
3016 |
-+ decor_src += (info->var.xres - width) * bytespp; |
3017 |
-+ } |
3018 |
-+} |
3019 |
-+ |
3020 |
-+#define cc2cx(a) \ |
3021 |
-+ ((info->fix.visual == FB_VISUAL_TRUECOLOR || \ |
3022 |
-+ info->fix.visual == FB_VISUAL_DIRECTCOLOR) ? \ |
3023 |
-+ ((u32*)info->pseudo_palette)[a] : a) |
3024 |
-+ |
3025 |
-+void fbcon_decor_putcs(struct vc_data *vc, struct fb_info *info, |
3026 |
-+ const unsigned short *s, int count, int yy, int xx) |
3027 |
-+{ |
3028 |
-+ unsigned short charmask = vc->vc_hi_font_mask ? 0x1ff : 0xff; |
3029 |
-+ struct fbcon_ops *ops = info->fbcon_par; |
3030 |
-+ int fg_color, bg_color, transparent; |
3031 |
-+ u8 *src; |
3032 |
-+ u32 bgx, fgx; |
3033 |
-+ u16 c = scr_readw(s); |
3034 |
-+ |
3035 |
-+ fg_color = get_color(vc, info, c, 1); |
3036 |
-+ bg_color = get_color(vc, info, c, 0); |
3037 |
-+ |
3038 |
-+ /* Don't paint the background image if console is blanked */ |
3039 |
-+ transparent = ops->blank_state ? 0 : |
3040 |
-+ (vc->vc_decor.bg_color == bg_color); |
3041 |
-+ |
3042 |
-+ xx = xx * vc->vc_font.width + vc->vc_decor.tx; |
3043 |
-+ yy = yy * vc->vc_font.height + vc->vc_decor.ty; |
3044 |
-+ |
3045 |
-+ fgx = cc2cx(fg_color); |
3046 |
-+ bgx = cc2cx(bg_color); |
3047 |
-+ |
3048 |
-+ while (count--) { |
3049 |
-+ c = scr_readw(s++); |
3050 |
-+ src = vc->vc_font.data + (c & charmask) * vc->vc_font.height * |
3051 |
-+ ((vc->vc_font.width + 7) >> 3); |
3052 |
-+ |
3053 |
-+ fbcon_decor_renderc(info, yy, xx, vc->vc_font.height, |
3054 |
-+ vc->vc_font.width, src, fgx, bgx, transparent); |
3055 |
-+ xx += vc->vc_font.width; |
3056 |
-+ } |
3057 |
-+} |
3058 |
-+ |
3059 |
-+void fbcon_decor_cursor(struct fb_info *info, struct fb_cursor *cursor) |
3060 |
-+{ |
3061 |
-+ int i; |
3062 |
-+ unsigned int dsize, s_pitch; |
3063 |
-+ struct fbcon_ops *ops = info->fbcon_par; |
3064 |
-+ struct vc_data* vc; |
3065 |
-+ u8 *src; |
3066 |
-+ |
3067 |
-+ /* we really don't need any cursors while the console is blanked */ |
3068 |
-+ if (info->state != FBINFO_STATE_RUNNING || ops->blank_state) |
3069 |
-+ return; |
3070 |
-+ |
3071 |
-+ vc = vc_cons[ops->currcon].d; |
3072 |
-+ |
3073 |
-+ src = kmalloc(64 + sizeof(struct fb_image), GFP_ATOMIC); |
3074 |
-+ if (!src) |
3075 |
-+ return; |
3076 |
-+ |
3077 |
-+ s_pitch = (cursor->image.width + 7) >> 3; |
3078 |
-+ dsize = s_pitch * cursor->image.height; |
3079 |
-+ if (cursor->enable) { |
3080 |
-+ switch (cursor->rop) { |
3081 |
-+ case ROP_XOR: |
3082 |
-+ for (i = 0; i < dsize; i++) |
3083 |
-+ src[i] = cursor->image.data[i] ^ cursor->mask[i]; |
3084 |
-+ break; |
3085 |
-+ case ROP_COPY: |
3086 |
-+ default: |
3087 |
-+ for (i = 0; i < dsize; i++) |
3088 |
-+ src[i] = cursor->image.data[i] & cursor->mask[i]; |
3089 |
-+ break; |
3090 |
-+ } |
3091 |
-+ } else |
3092 |
-+ memcpy(src, cursor->image.data, dsize); |
3093 |
-+ |
3094 |
-+ fbcon_decor_renderc(info, |
3095 |
-+ cursor->image.dy + vc->vc_decor.ty, |
3096 |
-+ cursor->image.dx + vc->vc_decor.tx, |
3097 |
-+ cursor->image.height, |
3098 |
-+ cursor->image.width, |
3099 |
-+ (u8*)src, |
3100 |
-+ cc2cx(cursor->image.fg_color), |
3101 |
-+ cc2cx(cursor->image.bg_color), |
3102 |
-+ cursor->image.bg_color == vc->vc_decor.bg_color); |
3103 |
-+ |
3104 |
-+ kfree(src); |
3105 |
-+} |
3106 |
-+ |
3107 |
-+static void decorset(u8 *dst, int height, int width, int dstbytes, |
3108 |
-+ u32 bgx, int bpp) |
3109 |
-+{ |
3110 |
-+ int i; |
3111 |
-+ |
3112 |
-+ if (bpp == 8) |
3113 |
-+ bgx |= bgx << 8; |
3114 |
-+ if (bpp == 16 || bpp == 8) |
3115 |
-+ bgx |= bgx << 16; |
3116 |
-+ |
3117 |
-+ while (height-- > 0) { |
3118 |
-+ u8 *p = dst; |
3119 |
-+ |
3120 |
-+ switch (bpp) { |
3121 |
-+ |
3122 |
-+ case 32: |
3123 |
-+ for (i=0; i < width; i++) { |
3124 |
-+ fb_writel(bgx, p); p += 4; |
3125 |
-+ } |
3126 |
-+ break; |
3127 |
-+ case 24: |
3128 |
-+ for (i=0; i < width; i++) { |
3129 |
-+#ifdef __LITTLE_ENDIAN |
3130 |
-+ fb_writew((bgx & 0xffff),(u16*)p); p += 2; |
3131 |
-+ fb_writeb((bgx >> 16),p++); |
3132 |
-+#else |
3133 |
-+ fb_writew((bgx >> 8),(u16*)p); p += 2; |
3134 |
-+ fb_writeb((bgx & 0xff),p++); |
3135 |
-+#endif |
3136 |
-+ } |
3137 |
-+ case 16: |
3138 |
-+ for (i=0; i < width/4; i++) { |
3139 |
-+ fb_writel(bgx,p); p += 4; |
3140 |
-+ fb_writel(bgx,p); p += 4; |
3141 |
-+ } |
3142 |
-+ if (width & 2) { |
3143 |
-+ fb_writel(bgx,p); p += 4; |
3144 |
-+ } |
3145 |
-+ if (width & 1) |
3146 |
-+ fb_writew(bgx,(u16*)p); |
3147 |
-+ break; |
3148 |
-+ case 8: |
3149 |
-+ for (i=0; i < width/4; i++) { |
3150 |
-+ fb_writel(bgx,p); p += 4; |
3151 |
-+ } |
3152 |
-+ |
3153 |
-+ if (width & 2) { |
3154 |
-+ fb_writew(bgx,p); p += 2; |
3155 |
-+ } |
3156 |
-+ if (width & 1) |
3157 |
-+ fb_writeb(bgx,(u8*)p); |
3158 |
-+ break; |
3159 |
-+ |
3160 |
-+ } |
3161 |
-+ dst += dstbytes; |
3162 |
-+ } |
3163 |
-+} |
3164 |
-+ |
3165 |
-+void fbcon_decor_copy(u8 *dst, u8 *src, int height, int width, int linebytes, |
3166 |
-+ int srclinebytes, int bpp) |
3167 |
-+{ |
3168 |
-+ int i; |
3169 |
-+ |
3170 |
-+ while (height-- > 0) { |
3171 |
-+ u32 *p = (u32 *)dst; |
3172 |
-+ u32 *q = (u32 *)src; |
3173 |
-+ |
3174 |
-+ switch (bpp) { |
3175 |
-+ |
3176 |
-+ case 32: |
3177 |
-+ for (i=0; i < width; i++) |
3178 |
-+ fb_writel(*q++, p++); |
3179 |
-+ break; |
3180 |
-+ case 24: |
3181 |
-+ for (i=0; i < (width*3/4); i++) |
3182 |
-+ fb_writel(*q++, p++); |
3183 |
-+ if ((width*3) % 4) { |
3184 |
-+ if (width & 2) { |
3185 |
-+ fb_writeb(*(u8*)q, (u8*)p); |
3186 |
-+ } else if (width & 1) { |
3187 |
-+ fb_writew(*(u16*)q, (u16*)p); |
3188 |
-+ fb_writeb(*(u8*)((u16*)q+1),(u8*)((u16*)p+2)); |
3189 |
-+ } |
3190 |
-+ } |
3191 |
-+ break; |
3192 |
-+ case 16: |
3193 |
-+ for (i=0; i < width/4; i++) { |
3194 |
-+ fb_writel(*q++, p++); |
3195 |
-+ fb_writel(*q++, p++); |
3196 |
-+ } |
3197 |
-+ if (width & 2) |
3198 |
-+ fb_writel(*q++, p++); |
3199 |
-+ if (width & 1) |
3200 |
-+ fb_writew(*(u16*)q, (u16*)p); |
3201 |
-+ break; |
3202 |
-+ case 8: |
3203 |
-+ for (i=0; i < width/4; i++) |
3204 |
-+ fb_writel(*q++, p++); |
3205 |
-+ |
3206 |
-+ if (width & 2) { |
3207 |
-+ fb_writew(*(u16*)q, (u16*)p); |
3208 |
-+ q = (u32*) ((u16*)q + 1); |
3209 |
-+ p = (u32*) ((u16*)p + 1); |
3210 |
-+ } |
3211 |
-+ if (width & 1) |
3212 |
-+ fb_writeb(*(u8*)q, (u8*)p); |
3213 |
-+ break; |
3214 |
-+ } |
3215 |
-+ |
3216 |
-+ dst += linebytes; |
3217 |
-+ src += srclinebytes; |
3218 |
-+ } |
3219 |
-+} |
3220 |
-+ |
3221 |
-+static void decorfill(struct fb_info *info, int sy, int sx, int height, |
3222 |
-+ int width) |
3223 |
-+{ |
3224 |
-+ int bytespp = ((info->var.bits_per_pixel + 7) >> 3); |
3225 |
-+ int d = sy * info->fix.line_length + sx * bytespp; |
3226 |
-+ int ds = (sy * info->var.xres + sx) * bytespp; |
3227 |
-+ |
3228 |
-+ fbcon_decor_copy((u8 *)(info->screen_base + d), (u8 *)(info->bgdecor.data + ds), |
3229 |
-+ height, width, info->fix.line_length, info->var.xres * bytespp, |
3230 |
-+ info->var.bits_per_pixel); |
3231 |
-+} |
3232 |
-+ |
3233 |
-+void fbcon_decor_clear(struct vc_data *vc, struct fb_info *info, int sy, int sx, |
3234 |
-+ int height, int width) |
3235 |
-+{ |
3236 |
-+ int bgshift = (vc->vc_hi_font_mask) ? 13 : 12; |
3237 |
-+ struct fbcon_ops *ops = info->fbcon_par; |
3238 |
-+ u8 *dst; |
3239 |
-+ int transparent, bg_color = attr_bgcol_ec(bgshift, vc, info); |
3240 |
-+ |
3241 |
-+ transparent = (vc->vc_decor.bg_color == bg_color); |
3242 |
-+ sy = sy * vc->vc_font.height + vc->vc_decor.ty; |
3243 |
-+ sx = sx * vc->vc_font.width + vc->vc_decor.tx; |
3244 |
-+ height *= vc->vc_font.height; |
3245 |
-+ width *= vc->vc_font.width; |
3246 |
-+ |
3247 |
-+ /* Don't paint the background image if console is blanked */ |
3248 |
-+ if (transparent && !ops->blank_state) { |
3249 |
-+ decorfill(info, sy, sx, height, width); |
3250 |
-+ } else { |
3251 |
-+ dst = (u8 *)(info->screen_base + sy * info->fix.line_length + |
3252 |
-+ sx * ((info->var.bits_per_pixel + 7) >> 3)); |
3253 |
-+ decorset(dst, height, width, info->fix.line_length, cc2cx(bg_color), |
3254 |
-+ info->var.bits_per_pixel); |
3255 |
-+ } |
3256 |
-+} |
3257 |
-+ |
3258 |
-+void fbcon_decor_clear_margins(struct vc_data *vc, struct fb_info *info, |
3259 |
-+ int bottom_only) |
3260 |
-+{ |
3261 |
-+ unsigned int tw = vc->vc_cols*vc->vc_font.width; |
3262 |
-+ unsigned int th = vc->vc_rows*vc->vc_font.height; |
3263 |
-+ |
3264 |
-+ if (!bottom_only) { |
3265 |
-+ /* top margin */ |
3266 |
-+ decorfill(info, 0, 0, vc->vc_decor.ty, info->var.xres); |
3267 |
-+ /* left margin */ |
3268 |
-+ decorfill(info, vc->vc_decor.ty, 0, th, vc->vc_decor.tx); |
3269 |
-+ /* right margin */ |
3270 |
-+ decorfill(info, vc->vc_decor.ty, vc->vc_decor.tx + tw, th, |
3271 |
-+ info->var.xres - vc->vc_decor.tx - tw); |
3272 |
-+ } |
3273 |
-+ decorfill(info, vc->vc_decor.ty + th, 0, |
3274 |
-+ info->var.yres - vc->vc_decor.ty - th, info->var.xres); |
3275 |
-+} |
3276 |
-+ |
3277 |
-+void fbcon_decor_bmove_redraw(struct vc_data *vc, struct fb_info *info, int y, |
3278 |
-+ int sx, int dx, int width) |
3279 |
-+{ |
3280 |
-+ u16 *d = (u16 *) (vc->vc_origin + vc->vc_size_row * y + dx * 2); |
3281 |
-+ u16 *s = d + (dx - sx); |
3282 |
-+ u16 *start = d; |
3283 |
-+ u16 *ls = d; |
3284 |
-+ u16 *le = d + width; |
3285 |
-+ u16 c; |
3286 |
-+ int x = dx; |
3287 |
-+ u16 attr = 1; |
3288 |
-+ |
3289 |
-+ do { |
3290 |
-+ c = scr_readw(d); |
3291 |
-+ if (attr != (c & 0xff00)) { |
3292 |
-+ attr = c & 0xff00; |
3293 |
-+ if (d > start) { |
3294 |
-+ fbcon_decor_putcs(vc, info, start, d - start, y, x); |
3295 |
-+ x += d - start; |
3296 |
-+ start = d; |
3297 |
-+ } |
3298 |
-+ } |
3299 |
-+ if (s >= ls && s < le && c == scr_readw(s)) { |
3300 |
-+ if (d > start) { |
3301 |
-+ fbcon_decor_putcs(vc, info, start, d - start, y, x); |
3302 |
-+ x += d - start + 1; |
3303 |
-+ start = d + 1; |
3304 |
-+ } else { |
3305 |
-+ x++; |
3306 |
-+ start++; |
3307 |
-+ } |
3308 |
-+ } |
3309 |
-+ s++; |
3310 |
-+ d++; |
3311 |
-+ } while (d < le); |
3312 |
-+ if (d > start) |
3313 |
-+ fbcon_decor_putcs(vc, info, start, d - start, y, x); |
3314 |
-+} |
3315 |
-+ |
3316 |
-+void fbcon_decor_blank(struct vc_data *vc, struct fb_info *info, int blank) |
3317 |
-+{ |
3318 |
-+ if (blank) { |
3319 |
-+ decorset((u8 *)info->screen_base, info->var.yres, info->var.xres, |
3320 |
-+ info->fix.line_length, 0, info->var.bits_per_pixel); |
3321 |
-+ } else { |
3322 |
-+ update_screen(vc); |
3323 |
-+ fbcon_decor_clear_margins(vc, info, 0); |
3324 |
-+ } |
3325 |
-+} |
3326 |
-+ |
3327 |
-From ea6ca92753106f1e0773acd1f18c71ae79a6f9b0 Mon Sep 17 00:00:00 2001 |
3328 |
-From: Mike Pagano <mpagano@g.o> |
3329 |
-Date: Tue, 27 Aug 2013 07:58:05 -0400 |
3330 |
-Subject: [PATCH] gbcondecor port |
3331 |
- |
3332 |
---- |
3333 |
- drivers/video/console/fbcon.c | 167 ++++++++++++++++++++++++++++++++++++------ |
3334 |
- 1 file changed, 143 insertions(+), 24 deletions(-) |
3335 |
- |
3336 |
-diff --git a/drivers/video/console/fbcon.c b/drivers/video/console/fbcon.c |
3337 |
-index cd8a802..666556c 100644 |
3338 |
---- a/drivers/video/console/fbcon.c |
3339 |
-+++ b/drivers/video/console/fbcon.c |
3340 |
-@@ -79,6 +79,7 @@ |
3341 |
- #include <asm/irq.h> |
3342 |
- |
3343 |
- #include "fbcon.h" |
3344 |
-+#include "fbcondecor.h" |
3345 |
- |
3346 |
- #ifdef FBCONDEBUG |
3347 |
- # define DPRINTK(fmt, args...) printk(KERN_DEBUG "%s: " fmt, __func__ , ## args) |
3348 |
-@@ -94,7 +95,7 @@ enum { |
3349 |
- |
3350 |
- static struct display fb_display[MAX_NR_CONSOLES]; |
3351 |
- |
3352 |
--static signed char con2fb_map[MAX_NR_CONSOLES]; |
3353 |
-+signed char con2fb_map[MAX_NR_CONSOLES]; |
3354 |
- static signed char con2fb_map_boot[MAX_NR_CONSOLES]; |
3355 |
- |
3356 |
- static int logo_lines; |
3357 |
-@@ -286,7 +287,7 @@ static inline int fbcon_is_inactive(struct vc_data *vc, struct fb_info *info) |
3358 |
- !vt_force_oops_output(vc); |
3359 |
- } |
3360 |
- |
3361 |
--static int get_color(struct vc_data *vc, struct fb_info *info, |
3362 |
-+int get_color(struct vc_data *vc, struct fb_info *info, |
3363 |
- u16 c, int is_fg) |
3364 |
- { |
3365 |
- int depth = fb_get_color_depth(&info->var, &info->fix); |
3366 |
-@@ -551,6 +552,9 @@ static int do_fbcon_takeover(int show_logo) |
3367 |
- info_idx = -1; |
3368 |
- } else { |
3369 |
- fbcon_has_console_bind = 1; |
3370 |
-+#ifdef CONFIG_FB_CON_DECOR |
3371 |
-+ fbcon_decor_init(); |
3372 |
-+#endif |
3373 |
- } |
3374 |
- |
3375 |
- return err; |
3376 |
-@@ -1007,6 +1011,12 @@ static const char *fbcon_startup(void) |
3377 |
- rows = FBCON_SWAP(ops->rotate, info->var.yres, info->var.xres); |
3378 |
- cols /= vc->vc_font.width; |
3379 |
- rows /= vc->vc_font.height; |
3380 |
-+ |
3381 |
-+ if (fbcon_decor_active(info, vc)) { |
3382 |
-+ cols = vc->vc_decor.twidth / vc->vc_font.width; |
3383 |
-+ rows = vc->vc_decor.theight / vc->vc_font.height; |
3384 |
-+ } |
3385 |
-+ |
3386 |
- vc_resize(vc, cols, rows); |
3387 |
- |
3388 |
- DPRINTK("mode: %s\n", info->fix.id); |
3389 |
-@@ -1036,7 +1046,7 @@ static void fbcon_init(struct vc_data *vc, int init) |
3390 |
- cap = info->flags; |
3391 |
- |
3392 |
- if (vc != svc || logo_shown == FBCON_LOGO_DONTSHOW || |
3393 |
-- (info->fix.type == FB_TYPE_TEXT)) |
3394 |
-+ (info->fix.type == FB_TYPE_TEXT) || fbcon_decor_active(info, vc)) |
3395 |
- logo = 0; |
3396 |
- |
3397 |
- if (var_to_display(p, &info->var, info)) |
3398 |
-@@ -1260,6 +1270,11 @@ static void fbcon_clear(struct vc_data *vc, int sy, int sx, int height, |
3399 |
- fbcon_clear_margins(vc, 0); |
3400 |
- } |
3401 |
- |
3402 |
-+ if (fbcon_decor_active(info, vc)) { |
3403 |
-+ fbcon_decor_clear(vc, info, sy, sx, height, width); |
3404 |
-+ return; |
3405 |
-+ } |
3406 |
-+ |
3407 |
- /* Split blits that cross physical y_wrap boundary */ |
3408 |
- |
3409 |
- y_break = p->vrows - p->yscroll; |
3410 |
-@@ -1279,10 +1294,15 @@ static void fbcon_putcs(struct vc_data *vc, const unsigned short *s, |
3411 |
- struct display *p = &fb_display[vc->vc_num]; |
3412 |
- struct fbcon_ops *ops = info->fbcon_par; |
3413 |
- |
3414 |
-- if (!fbcon_is_inactive(vc, info)) |
3415 |
-- ops->putcs(vc, info, s, count, real_y(p, ypos), xpos, |
3416 |
-- get_color(vc, info, scr_readw(s), 1), |
3417 |
-- get_color(vc, info, scr_readw(s), 0)); |
3418 |
-+ if (!fbcon_is_inactive(vc, info)) { |
3419 |
-+ |
3420 |
-+ if (fbcon_decor_active(info, vc)) |
3421 |
-+ fbcon_decor_putcs(vc, info, s, count, ypos, xpos); |
3422 |
-+ else |
3423 |
-+ ops->putcs(vc, info, s, count, real_y(p, ypos), xpos, |
3424 |
-+ get_color(vc, info, scr_readw(s), 1), |
3425 |
-+ get_color(vc, info, scr_readw(s), 0)); |
3426 |
-+ } |
3427 |
- } |
3428 |
- |
3429 |
- static void fbcon_putc(struct vc_data *vc, int c, int ypos, int xpos) |
3430 |
-@@ -1297,9 +1317,6 @@ static void fbcon_clear_margins(struct vc_data *vc, int bottom_only) |
3431 |
- { |
3432 |
- struct fb_info *info = registered_fb[con2fb_map[vc->vc_num]]; |
3433 |
- struct fbcon_ops *ops = info->fbcon_par; |
3434 |
-- |
3435 |
-- if (!fbcon_is_inactive(vc, info)) |
3436 |
-- ops->clear_margins(vc, info, bottom_only); |
3437 |
- } |
3438 |
- |
3439 |
- static void fbcon_cursor(struct vc_data *vc, int mode) |
3440 |
-@@ -1819,7 +1836,7 @@ static int fbcon_scroll(struct vc_data *vc, int t, int b, int dir, |
3441 |
- count = vc->vc_rows; |
3442 |
- if (softback_top) |
3443 |
- fbcon_softback_note(vc, t, count); |
3444 |
-- if (logo_shown >= 0) |
3445 |
-+ if (logo_shown >= 0 || fbcon_decor_active(info, vc)) |
3446 |
- goto redraw_up; |
3447 |
- switch (p->scrollmode) { |
3448 |
- case SCROLL_MOVE: |
3449 |
-@@ -1912,6 +1929,8 @@ static int fbcon_scroll(struct vc_data *vc, int t, int b, int dir, |
3450 |
- count = vc->vc_rows; |
3451 |
- if (logo_shown >= 0) |
3452 |
- goto redraw_down; |
3453 |
-+ if (fbcon_decor_active(info, vc)) |
3454 |
-+ goto redraw_down; |
3455 |
- switch (p->scrollmode) { |
3456 |
- case SCROLL_MOVE: |
3457 |
- fbcon_redraw_blit(vc, info, p, b - 1, b - t - count, |
3458 |
-@@ -2060,6 +2079,13 @@ static void fbcon_bmove_rec(struct vc_data *vc, struct display *p, int sy, int s |
3459 |
- } |
3460 |
- return; |
3461 |
- } |
3462 |
-+ |
3463 |
-+ if (fbcon_decor_active(info, vc) && sy == dy && height == 1) { |
3464 |
-+ /* must use slower redraw bmove to keep background pic intact */ |
3465 |
-+ fbcon_decor_bmove_redraw(vc, info, sy, sx, dx, width); |
3466 |
-+ return; |
3467 |
-+ } |
3468 |
-+ |
3469 |
- ops->bmove(vc, info, real_y(p, sy), sx, real_y(p, dy), dx, |
3470 |
- height, width); |
3471 |
- } |
3472 |
-@@ -2130,8 +2156,8 @@ static int fbcon_resize(struct vc_data *vc, unsigned int width, |
3473 |
- var.yres = virt_h * virt_fh; |
3474 |
- x_diff = info->var.xres - var.xres; |
3475 |
- y_diff = info->var.yres - var.yres; |
3476 |
-- if (x_diff < 0 || x_diff > virt_fw || |
3477 |
-- y_diff < 0 || y_diff > virt_fh) { |
3478 |
-+ if ((x_diff < 0 || x_diff > virt_fw || |
3479 |
-+ y_diff < 0 || y_diff > virt_fh) && !vc->vc_decor.state) { |
3480 |
- const struct fb_videomode *mode; |
3481 |
- |
3482 |
- DPRINTK("attempting resize %ix%i\n", var.xres, var.yres); |
3483 |
-@@ -2168,6 +2194,22 @@ static int fbcon_switch(struct vc_data *vc) |
3484 |
- info = registered_fb[con2fb_map[vc->vc_num]]; |
3485 |
- ops = info->fbcon_par; |
3486 |
- |
3487 |
-+ prev_console = ops->currcon; |
3488 |
-+ if (prev_console != -1) |
3489 |
-+ old_info = registered_fb[con2fb_map[prev_console]]; |
3490 |
-+ |
3491 |
-+#ifdef CONFIG_FB_CON_DECOR |
3492 |
-+ if (!fbcon_decor_active_vc(vc) && info->fix.visual == FB_VISUAL_DIRECTCOLOR) { |
3493 |
-+ struct vc_data *vc_curr = vc_cons[prev_console].d; |
3494 |
-+ if (vc_curr && fbcon_decor_active_vc(vc_curr)) { |
3495 |
-+ /* Clear the screen to avoid displaying funky colors during |
3496 |
-+ * * palette updates. */ |
3497 |
-+ memset((u8*)info->screen_base + info->fix.line_length * info->var.yoffset, |
3498 |
-+ 0, info->var.yres * info->fix.line_length); |
3499 |
-+ } |
3500 |
-+ } |
3501 |
-+#endif |
3502 |
-+ |
3503 |
- if (softback_top) { |
3504 |
- if (softback_lines) |
3505 |
- fbcon_set_origin(vc); |
3506 |
-@@ -2185,9 +2227,6 @@ static int fbcon_switch(struct vc_data *vc) |
3507 |
- logo_shown = FBCON_LOGO_CANSHOW; |
3508 |
- } |
3509 |
- |
3510 |
-- prev_console = ops->currcon; |
3511 |
-- if (prev_console != -1) |
3512 |
-- old_info = registered_fb[con2fb_map[prev_console]]; |
3513 |
- /* |
3514 |
- * FIXME: If we have multiple fbdev's loaded, we need to |
3515 |
- * update all info->currcon. Perhaps, we can place this |
3516 |
-@@ -2231,6 +2270,18 @@ static int fbcon_switch(struct vc_data *vc) |
3517 |
- fbcon_del_cursor_timer(old_info); |
3518 |
- } |
3519 |
- |
3520 |
-+ if (fbcon_decor_active_vc(vc)) { |
3521 |
-+ struct vc_data *vc_curr = vc_cons[prev_console].d; |
3522 |
-+ |
3523 |
-+ if (!vc_curr->vc_decor.theme || |
3524 |
-+ strcmp(vc->vc_decor.theme, vc_curr->vc_decor.theme) || |
3525 |
-+ (fbcon_decor_active_nores(info, vc_curr) && |
3526 |
-+ !fbcon_decor_active(info, vc_curr))) { |
3527 |
-+ fbcon_decor_disable(vc, 0); |
3528 |
-+ fbcon_decor_call_helper("modechange", vc->vc_num); |
3529 |
-+ } |
3530 |
-+ } |
3531 |
-+ |
3532 |
- if (fbcon_is_inactive(vc, info) || |
3533 |
- ops->blank_state != FB_BLANK_UNBLANK) |
3534 |
- fbcon_del_cursor_timer(info); |
3535 |
-@@ -2344,10 +2395,14 @@ static int fbcon_blank(struct vc_data *vc, int blank, int mode_switch) |
3536 |
- ops->blank_state = blank; |
3537 |
- fbcon_cursor(vc, blank ? CM_ERASE : CM_DRAW); |
3538 |
- ops->cursor_flash = (!blank); |
3539 |
-- |
3540 |
-- if (!(info->flags & FBINFO_MISC_USEREVENT)) |
3541 |
-- if (fb_blank(info, blank)) |
3542 |
-- fbcon_generic_blank(vc, info, blank); |
3543 |
-+ if (!(info->flags & FBINFO_MISC_USEREVENT)) { |
3544 |
-+ if (fb_blank(info, blank)) { |
3545 |
-+ if (fbcon_decor_active(info, vc)) |
3546 |
-+ fbcon_decor_blank(vc, info, blank); |
3547 |
-+ else |
3548 |
-+ fbcon_generic_blank(vc, info, blank); |
3549 |
-+ } |
3550 |
-+ } |
3551 |
- } |
3552 |
- |
3553 |
- if (!blank) |
3554 |
-@@ -2522,10 +2577,18 @@ static int fbcon_do_set_font(struct vc_data *vc, int w, int h, |
3555 |
- } |
3556 |
- |
3557 |
- if (resize) { |
3558 |
-+ /* reset wrap/pan */ |
3559 |
- int cols, rows; |
3560 |
- |
3561 |
- cols = FBCON_SWAP(ops->rotate, info->var.xres, info->var.yres); |
3562 |
- rows = FBCON_SWAP(ops->rotate, info->var.yres, info->var.xres); |
3563 |
-+ |
3564 |
-+ if (fbcon_decor_active(info, vc)) { |
3565 |
-+ info->var.xoffset = info->var.yoffset = p->yscroll = 0; |
3566 |
-+ cols = vc->vc_decor.twidth; |
3567 |
-+ rows = vc->vc_decor.theight; |
3568 |
-+ } |
3569 |
-+ |
3570 |
- cols /= w; |
3571 |
- rows /= h; |
3572 |
- vc_resize(vc, cols, rows); |
3573 |
-@@ -2657,7 +2720,11 @@ static int fbcon_set_palette(struct vc_data *vc, unsigned char *table) |
3574 |
- int i, j, k, depth; |
3575 |
- u8 val; |
3576 |
- |
3577 |
-- if (fbcon_is_inactive(vc, info)) |
3578 |
-+ if (fbcon_is_inactive(vc, info) |
3579 |
-+#ifdef CONFIG_FB_CON_DECOR |
3580 |
-+ || vc->vc_num != fg_console |
3581 |
-+#endif |
3582 |
-+ ) |
3583 |
- return -EINVAL; |
3584 |
- |
3585 |
- if (!CON_IS_VISIBLE(vc)) |
3586 |
-@@ -2683,7 +2750,49 @@ static int fbcon_set_palette(struct vc_data *vc, unsigned char *table) |
3587 |
- } else |
3588 |
- fb_copy_cmap(fb_default_cmap(1 << depth), &palette_cmap); |
3589 |
- |
3590 |
-- return fb_set_cmap(&palette_cmap, info); |
3591 |
-+ if (fbcon_decor_active(info, vc_cons[fg_console].d) && |
3592 |
-+ info->fix.visual == FB_VISUAL_DIRECTCOLOR) { |
3593 |
-+ |
3594 |
-+ u16 *red, *green, *blue; |
3595 |
-+ int minlen = min(min(info->var.red.length, info->var.green.length), |
3596 |
-+ info->var.blue.length); |
3597 |
-+ int h; |
3598 |
-+ |
3599 |
-+ struct fb_cmap cmap = { |
3600 |
-+ .start = 0, |
3601 |
-+ .len = (1 << minlen), |
3602 |
-+ .red = NULL, |
3603 |
-+ .green = NULL, |
3604 |
-+ .blue = NULL, |
3605 |
-+ .transp = NULL |
3606 |
-+ }; |
3607 |
-+ |
3608 |
-+ red = kmalloc(256 * sizeof(u16) * 3, GFP_KERNEL); |
3609 |
-+ |
3610 |
-+ if (!red) |
3611 |
-+ goto out; |
3612 |
-+ |
3613 |
-+ green = red + 256; |
3614 |
-+ blue = green + 256; |
3615 |
-+ cmap.red = red; |
3616 |
-+ cmap.green = green; |
3617 |
-+ cmap.blue = blue; |
3618 |
-+ |
3619 |
-+ for (i = 0; i < cmap.len; i++) { |
3620 |
-+ red[i] = green[i] = blue[i] = (0xffff * i)/(cmap.len-1); |
3621 |
-+ } |
3622 |
-+ |
3623 |
-+ h = fb_set_cmap(&cmap, info); |
3624 |
-+ fbcon_decor_fix_pseudo_pal(info, vc_cons[fg_console].d); |
3625 |
-+ kfree(red); |
3626 |
-+ |
3627 |
-+ return h; |
3628 |
-+ |
3629 |
-+ } else if (fbcon_decor_active(info, vc_cons[fg_console].d) && |
3630 |
-+ info->var.bits_per_pixel == 8 && info->bgdecor.cmap.red != NULL) |
3631 |
-+ fb_set_cmap(&info->bgdecor.cmap, info); |
3632 |
-+ |
3633 |
-+out: return fb_set_cmap(&palette_cmap, info); |
3634 |
- } |
3635 |
- |
3636 |
- static u16 *fbcon_screen_pos(struct vc_data *vc, int offset) |
3637 |
-@@ -2909,7 +3018,13 @@ static void fbcon_modechanged(struct fb_info *info) |
3638 |
- rows = FBCON_SWAP(ops->rotate, info->var.yres, info->var.xres); |
3639 |
- cols /= vc->vc_font.width; |
3640 |
- rows /= vc->vc_font.height; |
3641 |
-- vc_resize(vc, cols, rows); |
3642 |
-+ if (!fbcon_decor_active_nores(info, vc)) { |
3643 |
-+ vc_resize(vc, cols, rows); |
3644 |
-+ } else { |
3645 |
-+ fbcon_decor_disable(vc, 0); |
3646 |
-+ fbcon_decor_call_helper("modechange", vc->vc_num); |
3647 |
-+ } |
3648 |
-+ |
3649 |
- updatescrollmode(p, info, vc); |
3650 |
- scrollback_max = 0; |
3651 |
- scrollback_current = 0; |
3652 |
-@@ -2954,7 +3069,10 @@ static void fbcon_set_all_vcs(struct fb_info *info) |
3653 |
- rows = FBCON_SWAP(ops->rotate, info->var.yres, info->var.xres); |
3654 |
- cols /= vc->vc_font.width; |
3655 |
- rows /= vc->vc_font.height; |
3656 |
-- vc_resize(vc, cols, rows); |
3657 |
-+ if (!fbcon_decor_active_nores(info, vc)) { |
3658 |
-+ vc_resize(vc, cols, rows); |
3659 |
-+ } |
3660 |
-+ |
3661 |
- } |
3662 |
- |
3663 |
- if (fg != -1) |
3664 |
-@@ -3570,6 +3688,7 @@ static void fbcon_exit(void) |
3665 |
- } |
3666 |
- } |
3667 |
- |
3668 |
-+ fbcon_decor_exit(); |
3669 |
- fbcon_has_exited = 1; |
3670 |
- } |
3671 |
- |
3672 |
--- |
3673 |
-1.8.1.5 |
3674 |
- |
3675 |
-diff --git a/drivers/video/console/fbcondecor.c b/drivers/video/console/fbcondecor.c |
3676 |
-new file mode 100644 |
3677 |
-index 0000000..7189ce6 |
3678 |
---- /dev/null |
3679 |
-+++ b/drivers/video/console/fbcondecor.c |
3680 |
-@@ -0,0 +1,555 @@ |
3681 |
-+/* |
3682 |
-+ * linux/drivers/video/console/fbcondecor.c -- Framebuffer console decorations |
3683 |
-+ * |
3684 |
-+ * Copyright (C) 2004-2009 Michal Januszewski <spock@g.o> |
3685 |
-+ * |
3686 |
-+ * Code based upon "Bootsplash" (C) 2001-2003 |
3687 |
-+ * Volker Poplawski <volker@×××××××××.de>, |
3688 |
-+ * Stefan Reinauer <stepan@××××.de>, |
3689 |
-+ * Steffen Winterfeldt <snwint@××××.de>, |
3690 |
-+ * Michael Schroeder <mls@××××.de>, |
3691 |
-+ * Ken Wimer <wimer@××××.de>. |
3692 |
-+ * |
3693 |
-+ * Compat ioctl support by Thorsten Klein <TK@××××××××××××××.de>. |
3694 |
-+ * |
3695 |
-+ * This file is subject to the terms and conditions of the GNU General Public |
3696 |
-+ * License. See the file COPYING in the main directory of this archive for |
3697 |
-+ * more details. |
3698 |
-+ * |
3699 |
-+ */ |
3700 |
-+#include <linux/module.h> |
3701 |
-+#include <linux/kernel.h> |
3702 |
-+#include <linux/string.h> |
3703 |
-+#include <linux/types.h> |
3704 |
-+#include <linux/fb.h> |
3705 |
-+#include <linux/vt_kern.h> |
3706 |
-+#include <linux/vmalloc.h> |
3707 |
-+#include <linux/unistd.h> |
3708 |
-+#include <linux/syscalls.h> |
3709 |
-+#include <linux/init.h> |
3710 |
-+#include <linux/proc_fs.h> |
3711 |
-+#include <linux/workqueue.h> |
3712 |
-+#include <linux/kmod.h> |
3713 |
-+#include <linux/miscdevice.h> |
3714 |
-+#include <linux/device.h> |
3715 |
-+#include <linux/fs.h> |
3716 |
-+#include <linux/compat.h> |
3717 |
-+#include <linux/console.h> |
3718 |
-+ |
3719 |
-+#include <asm/uaccess.h> |
3720 |
-+#include <asm/irq.h> |
3721 |
-+ |
3722 |
-+#include "fbcon.h" |
3723 |
-+#include "fbcondecor.h" |
3724 |
-+ |
3725 |
-+extern signed char con2fb_map[]; |
3726 |
-+static int fbcon_decor_enable(struct vc_data *vc); |
3727 |
-+char fbcon_decor_path[KMOD_PATH_LEN] = "/sbin/fbcondecor_helper"; |
3728 |
-+static int initialized = 0; |
3729 |
-+ |
3730 |
-+int fbcon_decor_call_helper(char* cmd, unsigned short vc) |
3731 |
-+{ |
3732 |
-+ char *envp[] = { |
3733 |
-+ "HOME=/", |
3734 |
-+ "PATH=/sbin:/bin", |
3735 |
-+ NULL |
3736 |
-+ }; |
3737 |
-+ |
3738 |
-+ char tfb[5]; |
3739 |
-+ char tcons[5]; |
3740 |
-+ unsigned char fb = (int) con2fb_map[vc]; |
3741 |
-+ |
3742 |
-+ char *argv[] = { |
3743 |
-+ fbcon_decor_path, |
3744 |
-+ "2", |
3745 |
-+ cmd, |
3746 |
-+ tcons, |
3747 |
-+ tfb, |
3748 |
-+ vc_cons[vc].d->vc_decor.theme, |
3749 |
-+ NULL |
3750 |
-+ }; |
3751 |
-+ |
3752 |
-+ snprintf(tfb,5,"%d",fb); |
3753 |
-+ snprintf(tcons,5,"%d",vc); |
3754 |
-+ |
3755 |
-+ return call_usermodehelper(fbcon_decor_path, argv, envp, UMH_WAIT_EXEC); |
3756 |
-+} |
3757 |
-+ |
3758 |
-+/* Disables fbcondecor on a virtual console; called with console sem held. */ |
3759 |
-+int fbcon_decor_disable(struct vc_data *vc, unsigned char redraw) |
3760 |
-+{ |
3761 |
-+ struct fb_info* info; |
3762 |
-+ |
3763 |
-+ if (!vc->vc_decor.state) |
3764 |
-+ return -EINVAL; |
3765 |
-+ |
3766 |
-+ info = registered_fb[(int) con2fb_map[vc->vc_num]]; |
3767 |
-+ |
3768 |
-+ if (info == NULL) |
3769 |
-+ return -EINVAL; |
3770 |
-+ |
3771 |
-+ vc->vc_decor.state = 0; |
3772 |
-+ vc_resize(vc, info->var.xres / vc->vc_font.width, |
3773 |
-+ info->var.yres / vc->vc_font.height); |
3774 |
-+ |
3775 |
-+ if (fg_console == vc->vc_num && redraw) { |
3776 |
-+ redraw_screen(vc, 0); |
3777 |
-+ update_region(vc, vc->vc_origin + |
3778 |
-+ vc->vc_size_row * vc->vc_top, |
3779 |
-+ vc->vc_size_row * (vc->vc_bottom - vc->vc_top) / 2); |
3780 |
-+ } |
3781 |
-+ |
3782 |
-+ printk(KERN_INFO "fbcondecor: switched decor state to 'off' on console %d\n", |
3783 |
-+ vc->vc_num); |
3784 |
-+ |
3785 |
-+ return 0; |
3786 |
-+} |
3787 |
-+ |
3788 |
-+/* Enables fbcondecor on a virtual console; called with console sem held. */ |
3789 |
-+static int fbcon_decor_enable(struct vc_data *vc) |
3790 |
-+{ |
3791 |
-+ struct fb_info* info; |
3792 |
-+ |
3793 |
-+ info = registered_fb[(int) con2fb_map[vc->vc_num]]; |
3794 |
-+ |
3795 |
-+ if (vc->vc_decor.twidth == 0 || vc->vc_decor.theight == 0 || |
3796 |
-+ info == NULL || vc->vc_decor.state || (!info->bgdecor.data && |
3797 |
-+ vc->vc_num == fg_console)) |
3798 |
-+ return -EINVAL; |
3799 |
-+ |
3800 |
-+ vc->vc_decor.state = 1; |
3801 |
-+ vc_resize(vc, vc->vc_decor.twidth / vc->vc_font.width, |
3802 |
-+ vc->vc_decor.theight / vc->vc_font.height); |
3803 |
-+ |
3804 |
-+ if (fg_console == vc->vc_num) { |
3805 |
-+ redraw_screen(vc, 0); |
3806 |
-+ update_region(vc, vc->vc_origin + |
3807 |
-+ vc->vc_size_row * vc->vc_top, |
3808 |
-+ vc->vc_size_row * (vc->vc_bottom - vc->vc_top) / 2); |
3809 |
-+ fbcon_decor_clear_margins(vc, info, 0); |
3810 |
-+ } |
3811 |
-+ |
3812 |
-+ printk(KERN_INFO "fbcondecor: switched decor state to 'on' on console %d\n", |
3813 |
-+ vc->vc_num); |
3814 |
-+ |
3815 |
-+ return 0; |
3816 |
-+} |
3817 |
-+ |
3818 |
-+static inline int fbcon_decor_ioctl_dosetstate(struct vc_data *vc, unsigned int state, unsigned char origin) |
3819 |
-+{ |
3820 |
-+ int ret; |
3821 |
-+ |
3822 |
-+// if (origin == FBCON_DECOR_IO_ORIG_USER) |
3823 |
-+ console_lock(); |
3824 |
-+ if (!state) |
3825 |
-+ ret = fbcon_decor_disable(vc, 1); |
3826 |
-+ else |
3827 |
-+ ret = fbcon_decor_enable(vc); |
3828 |
-+// if (origin == FBCON_DECOR_IO_ORIG_USER) |
3829 |
-+ console_unlock(); |
3830 |
-+ |
3831 |
-+ return ret; |
3832 |
-+} |
3833 |
-+ |
3834 |
-+static inline void fbcon_decor_ioctl_dogetstate(struct vc_data *vc, unsigned int *state) |
3835 |
-+{ |
3836 |
-+ *state = vc->vc_decor.state; |
3837 |
-+} |
3838 |
-+ |
3839 |
-+static int fbcon_decor_ioctl_dosetcfg(struct vc_data *vc, struct vc_decor *cfg, unsigned char origin) |
3840 |
-+{ |
3841 |
-+ struct fb_info *info; |
3842 |
-+ int len; |
3843 |
-+ char *tmp; |
3844 |
-+ |
3845 |
-+ info = registered_fb[(int) con2fb_map[vc->vc_num]]; |
3846 |
-+ |
3847 |
-+ if (info == NULL || !cfg->twidth || !cfg->theight || |
3848 |
-+ cfg->tx + cfg->twidth > info->var.xres || |
3849 |
-+ cfg->ty + cfg->theight > info->var.yres) |
3850 |
-+ return -EINVAL; |
3851 |
-+ |
3852 |
-+ len = strlen_user(cfg->theme); |
3853 |
-+ if (!len || len > FBCON_DECOR_THEME_LEN) |
3854 |
-+ return -EINVAL; |
3855 |
-+ tmp = kmalloc(len, GFP_KERNEL); |
3856 |
-+ if (!tmp) |
3857 |
-+ return -ENOMEM; |
3858 |
-+ if (copy_from_user(tmp, (void __user *)cfg->theme, len)) |
3859 |
-+ return -EFAULT; |
3860 |
-+ cfg->theme = tmp; |
3861 |
-+ cfg->state = 0; |
3862 |
-+ |
3863 |
-+ /* If this ioctl is a response to a request from kernel, the console sem |
3864 |
-+ * is already held; we also don't need to disable decor because either the |
3865 |
-+ * new config and background picture will be successfully loaded, and the |
3866 |
-+ * decor will stay on, or in case of a failure it'll be turned off in fbcon. */ |
3867 |
-+// if (origin == FBCON_DECOR_IO_ORIG_USER) { |
3868 |
-+ console_lock(); |
3869 |
-+ if (vc->vc_decor.state) |
3870 |
-+ fbcon_decor_disable(vc, 1); |
3871 |
-+// } |
3872 |
-+ |
3873 |
-+ if (vc->vc_decor.theme) |
3874 |
-+ kfree(vc->vc_decor.theme); |
3875 |
-+ |
3876 |
-+ vc->vc_decor = *cfg; |
3877 |
-+ |
3878 |
-+// if (origin == FBCON_DECOR_IO_ORIG_USER) |
3879 |
-+ console_unlock(); |
3880 |
-+ |
3881 |
-+ printk(KERN_INFO "fbcondecor: console %d using theme '%s'\n", |
3882 |
-+ vc->vc_num, vc->vc_decor.theme); |
3883 |
-+ return 0; |
3884 |
-+} |
3885 |
-+ |
3886 |
-+static int fbcon_decor_ioctl_dogetcfg(struct vc_data *vc, struct vc_decor *decor) |
3887 |
-+{ |
3888 |
-+ char __user *tmp; |
3889 |
-+ |
3890 |
-+ tmp = decor->theme; |
3891 |
-+ *decor = vc->vc_decor; |
3892 |
-+ decor->theme = tmp; |
3893 |
-+ |
3894 |
-+ if (vc->vc_decor.theme) { |
3895 |
-+ if (copy_to_user(tmp, vc->vc_decor.theme, strlen(vc->vc_decor.theme) + 1)) |
3896 |
-+ return -EFAULT; |
3897 |
-+ } else |
3898 |
-+ if (put_user(0, tmp)) |
3899 |
-+ return -EFAULT; |
3900 |
-+ |
3901 |
-+ return 0; |
3902 |
-+} |
3903 |
-+ |
3904 |
-+static int fbcon_decor_ioctl_dosetpic(struct vc_data *vc, struct fb_image *img, unsigned char origin) |
3905 |
-+{ |
3906 |
-+ struct fb_info *info; |
3907 |
-+ int len; |
3908 |
-+ u8 *tmp; |
3909 |
-+ |
3910 |
-+ if (vc->vc_num != fg_console) |
3911 |
-+ return -EINVAL; |
3912 |
-+ |
3913 |
-+ info = registered_fb[(int) con2fb_map[vc->vc_num]]; |
3914 |
-+ |
3915 |
-+ if (info == NULL) |
3916 |
-+ return -EINVAL; |
3917 |
-+ |
3918 |
-+ if (img->width != info->var.xres || img->height != info->var.yres) { |
3919 |
-+ printk(KERN_ERR "fbcondecor: picture dimensions mismatch\n"); |
3920 |
-+ printk(KERN_ERR "%dx%d vs %dx%d\n", img->width, img->height, info->var.xres, info->var.yres); |
3921 |
-+ return -EINVAL; |
3922 |
-+ } |
3923 |
-+ |
3924 |
-+ if (img->depth != info->var.bits_per_pixel) { |
3925 |
-+ printk(KERN_ERR "fbcondecor: picture depth mismatch\n"); |
3926 |
-+ return -EINVAL; |
3927 |
-+ } |
3928 |
-+ |
3929 |
-+ if (img->depth == 8) { |
3930 |
-+ if (!img->cmap.len || !img->cmap.red || !img->cmap.green || |
3931 |
-+ !img->cmap.blue) |
3932 |
-+ return -EINVAL; |
3933 |
-+ |
3934 |
-+ tmp = vmalloc(img->cmap.len * 3 * 2); |
3935 |
-+ if (!tmp) |
3936 |
-+ return -ENOMEM; |
3937 |
-+ |
3938 |
-+ if (copy_from_user(tmp, |
3939 |
-+ (void __user*)img->cmap.red, (img->cmap.len << 1)) || |
3940 |
-+ copy_from_user(tmp + (img->cmap.len << 1), |
3941 |
-+ (void __user*)img->cmap.green, (img->cmap.len << 1)) || |
3942 |
-+ copy_from_user(tmp + (img->cmap.len << 2), |
3943 |
-+ (void __user*)img->cmap.blue, (img->cmap.len << 1))) { |
3944 |
-+ vfree(tmp); |
3945 |
-+ return -EFAULT; |
3946 |
-+ } |
3947 |
-+ |
3948 |
-+ img->cmap.transp = NULL; |
3949 |
-+ img->cmap.red = (u16*)tmp; |
3950 |
-+ img->cmap.green = img->cmap.red + img->cmap.len; |
3951 |
-+ img->cmap.blue = img->cmap.green + img->cmap.len; |
3952 |
-+ } else { |
3953 |
-+ img->cmap.red = NULL; |
3954 |
-+ } |
3955 |
-+ |
3956 |
-+ len = ((img->depth + 7) >> 3) * img->width * img->height; |
3957 |
-+ |
3958 |
-+ /* |
3959 |
-+ * Allocate an additional byte so that we never go outside of the |
3960 |
-+ * buffer boundaries in the rendering functions in a 24 bpp mode. |
3961 |
-+ */ |
3962 |
-+ tmp = vmalloc(len + 1); |
3963 |
-+ |
3964 |
-+ if (!tmp) |
3965 |
-+ goto out; |
3966 |
-+ |
3967 |
-+ if (copy_from_user(tmp, (void __user*)img->data, len)) |
3968 |
-+ goto out; |
3969 |
-+ |
3970 |
-+ img->data = tmp; |
3971 |
-+ |
3972 |
-+ /* If this ioctl is a response to a request from kernel, the console sem |
3973 |
-+ * is already held. */ |
3974 |
-+// if (origin == FBCON_DECOR_IO_ORIG_USER) |
3975 |
-+ console_lock(); |
3976 |
-+ |
3977 |
-+ if (info->bgdecor.data) |
3978 |
-+ vfree((u8*)info->bgdecor.data); |
3979 |
-+ if (info->bgdecor.cmap.red) |
3980 |
-+ vfree(info->bgdecor.cmap.red); |
3981 |
-+ |
3982 |
-+ info->bgdecor = *img; |
3983 |
-+ |
3984 |
-+ if (fbcon_decor_active_vc(vc) && fg_console == vc->vc_num) { |
3985 |
-+ redraw_screen(vc, 0); |
3986 |
-+ update_region(vc, vc->vc_origin + |
3987 |
-+ vc->vc_size_row * vc->vc_top, |
3988 |
-+ vc->vc_size_row * (vc->vc_bottom - vc->vc_top) / 2); |
3989 |
-+ fbcon_decor_clear_margins(vc, info, 0); |
3990 |
-+ } |
3991 |
-+ |
3992 |
-+// if (origin == FBCON_DECOR_IO_ORIG_USER) |
3993 |
-+ console_unlock(); |
3994 |
-+ |
3995 |
-+ return 0; |
3996 |
-+ |
3997 |
-+out: if (img->cmap.red) |
3998 |
-+ vfree(img->cmap.red); |
3999 |
-+ |
4000 |
-+ if (tmp) |
4001 |
-+ vfree(tmp); |
4002 |
-+ return -ENOMEM; |
4003 |
-+} |
4004 |
-+ |
4005 |
-+static long fbcon_decor_ioctl(struct file *filp, u_int cmd, u_long arg) |
4006 |
-+{ |
4007 |
-+ struct fbcon_decor_iowrapper __user *wrapper = (void __user*) arg; |
4008 |
-+ struct vc_data *vc = NULL; |
4009 |
-+ unsigned short vc_num = 0; |
4010 |
-+ unsigned char origin = 0; |
4011 |
-+ void __user *data = NULL; |
4012 |
-+ |
4013 |
-+ if (!access_ok(VERIFY_READ, wrapper, |
4014 |
-+ sizeof(struct fbcon_decor_iowrapper))) |
4015 |
-+ return -EFAULT; |
4016 |
-+ |
4017 |
-+ __get_user(vc_num, &wrapper->vc); |
4018 |
-+ __get_user(origin, &wrapper->origin); |
4019 |
-+ __get_user(data, &wrapper->data); |
4020 |
-+ |
4021 |
-+ if (!vc_cons_allocated(vc_num)) |
4022 |
-+ return -EINVAL; |
4023 |
-+ |
4024 |
-+ vc = vc_cons[vc_num].d; |
4025 |
-+ |
4026 |
-+ switch (cmd) { |
4027 |
-+ case FBIOCONDECOR_SETPIC: |
4028 |
-+ { |
4029 |
-+ struct fb_image img; |
4030 |
-+ if (copy_from_user(&img, (struct fb_image __user *)data, sizeof(struct fb_image))) |
4031 |
-+ return -EFAULT; |
4032 |
-+ |
4033 |
-+ return fbcon_decor_ioctl_dosetpic(vc, &img, origin); |
4034 |
-+ } |
4035 |
-+ case FBIOCONDECOR_SETCFG: |
4036 |
-+ { |
4037 |
-+ struct vc_decor cfg; |
4038 |
-+ if (copy_from_user(&cfg, (struct vc_decor __user *)data, sizeof(struct vc_decor))) |
4039 |
-+ return -EFAULT; |
4040 |
-+ |
4041 |
-+ return fbcon_decor_ioctl_dosetcfg(vc, &cfg, origin); |
4042 |
-+ } |
4043 |
-+ case FBIOCONDECOR_GETCFG: |
4044 |
-+ { |
4045 |
-+ int rval; |
4046 |
-+ struct vc_decor cfg; |
4047 |
-+ |
4048 |
-+ if (copy_from_user(&cfg, (struct vc_decor __user *)data, sizeof(struct vc_decor))) |
4049 |
-+ return -EFAULT; |
4050 |
-+ |
4051 |
-+ rval = fbcon_decor_ioctl_dogetcfg(vc, &cfg); |
4052 |
-+ |
4053 |
-+ if (copy_to_user(data, &cfg, sizeof(struct vc_decor))) |
4054 |
-+ return -EFAULT; |
4055 |
-+ return rval; |
4056 |
-+ } |
4057 |
-+ case FBIOCONDECOR_SETSTATE: |
4058 |
-+ { |
4059 |
-+ unsigned int state = 0; |
4060 |
-+ if (get_user(state, (unsigned int __user *)data)) |
4061 |
-+ return -EFAULT; |
4062 |
-+ return fbcon_decor_ioctl_dosetstate(vc, state, origin); |
4063 |
-+ } |
4064 |
-+ case FBIOCONDECOR_GETSTATE: |
4065 |
-+ { |
4066 |
-+ unsigned int state = 0; |
4067 |
-+ fbcon_decor_ioctl_dogetstate(vc, &state); |
4068 |
-+ return put_user(state, (unsigned int __user *)data); |
4069 |
-+ } |
4070 |
-+ |
4071 |
-+ default: |
4072 |
-+ return -ENOIOCTLCMD; |
4073 |
-+ } |
4074 |
-+} |
4075 |
-+ |
4076 |
-+#ifdef CONFIG_COMPAT |
4077 |
-+ |
4078 |
-+static long fbcon_decor_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) { |
4079 |
-+ |
4080 |
-+ struct fbcon_decor_iowrapper32 __user *wrapper = (void __user *)arg; |
4081 |
-+ struct vc_data *vc = NULL; |
4082 |
-+ unsigned short vc_num = 0; |
4083 |
-+ unsigned char origin = 0; |
4084 |
-+ compat_uptr_t data_compat = 0; |
4085 |
-+ void __user *data = NULL; |
4086 |
-+ |
4087 |
-+ if (!access_ok(VERIFY_READ, wrapper, |
4088 |
-+ sizeof(struct fbcon_decor_iowrapper32))) |
4089 |
-+ return -EFAULT; |
4090 |
-+ |
4091 |
-+ __get_user(vc_num, &wrapper->vc); |
4092 |
-+ __get_user(origin, &wrapper->origin); |
4093 |
-+ __get_user(data_compat, &wrapper->data); |
4094 |
-+ data = compat_ptr(data_compat); |
4095 |
-+ |
4096 |
-+ if (!vc_cons_allocated(vc_num)) |
4097 |
-+ return -EINVAL; |
4098 |
-+ |
4099 |
-+ vc = vc_cons[vc_num].d; |
4100 |
-+ |
4101 |
-+ switch (cmd) { |
4102 |
-+ case FBIOCONDECOR_SETPIC32: |
4103 |
-+ { |
4104 |
-+ struct fb_image32 img_compat; |
4105 |
-+ struct fb_image img; |
4106 |
-+ |
4107 |
-+ if (copy_from_user(&img_compat, (struct fb_image32 __user *)data, sizeof(struct fb_image32))) |
4108 |
-+ return -EFAULT; |
4109 |
-+ |
4110 |
-+ fb_image_from_compat(img, img_compat); |
4111 |
-+ |
4112 |
-+ return fbcon_decor_ioctl_dosetpic(vc, &img, origin); |
4113 |
-+ } |
4114 |
-+ |
4115 |
-+ case FBIOCONDECOR_SETCFG32: |
4116 |
-+ { |
4117 |
-+ struct vc_decor32 cfg_compat; |
4118 |
-+ struct vc_decor cfg; |
4119 |
-+ |
4120 |
-+ if (copy_from_user(&cfg_compat, (struct vc_decor32 __user *)data, sizeof(struct vc_decor32))) |
4121 |
-+ return -EFAULT; |
4122 |
-+ |
4123 |
-+ vc_decor_from_compat(cfg, cfg_compat); |
4124 |
-+ |
4125 |
-+ return fbcon_decor_ioctl_dosetcfg(vc, &cfg, origin); |
4126 |
-+ } |
4127 |
-+ |
4128 |
-+ case FBIOCONDECOR_GETCFG32: |
4129 |
-+ { |
4130 |
-+ int rval; |
4131 |
-+ struct vc_decor32 cfg_compat; |
4132 |
-+ struct vc_decor cfg; |
4133 |
-+ |
4134 |
-+ if (copy_from_user(&cfg_compat, (struct vc_decor32 __user *)data, sizeof(struct vc_decor32))) |
4135 |
-+ return -EFAULT; |
4136 |
-+ cfg.theme = compat_ptr(cfg_compat.theme); |
4137 |
-+ |
4138 |
-+ rval = fbcon_decor_ioctl_dogetcfg(vc, &cfg); |
4139 |
-+ |
4140 |
-+ vc_decor_to_compat(cfg_compat, cfg); |
4141 |
-+ |
4142 |
-+ if (copy_to_user((struct vc_decor32 __user *)data, &cfg_compat, sizeof(struct vc_decor32))) |
4143 |
-+ return -EFAULT; |
4144 |
-+ return rval; |
4145 |
-+ } |
4146 |
-+ |
4147 |
-+ case FBIOCONDECOR_SETSTATE32: |
4148 |
-+ { |
4149 |
-+ compat_uint_t state_compat = 0; |
4150 |
-+ unsigned int state = 0; |
4151 |
-+ |
4152 |
-+ if (get_user(state_compat, (compat_uint_t __user *)data)) |
4153 |
-+ return -EFAULT; |
4154 |
-+ |
4155 |
-+ state = (unsigned int)state_compat; |
4156 |
-+ |
4157 |
-+ return fbcon_decor_ioctl_dosetstate(vc, state, origin); |
4158 |
-+ } |
4159 |
-+ |
4160 |
-+ case FBIOCONDECOR_GETSTATE32: |
4161 |
-+ { |
4162 |
-+ compat_uint_t state_compat = 0; |
4163 |
-+ unsigned int state = 0; |
4164 |
-+ |
4165 |
-+ fbcon_decor_ioctl_dogetstate(vc, &state); |
4166 |
-+ state_compat = (compat_uint_t)state; |
4167 |
-+ |
4168 |
-+ return put_user(state_compat, (compat_uint_t __user *)data); |
4169 |
-+ } |
4170 |
-+ |
4171 |
-+ default: |
4172 |
-+ return -ENOIOCTLCMD; |
4173 |
-+ } |
4174 |
-+} |
4175 |
-+#else |
4176 |
-+ #define fbcon_decor_compat_ioctl NULL |
4177 |
-+#endif |
4178 |
-+ |
4179 |
-+static struct file_operations fbcon_decor_ops = { |
4180 |
-+ .owner = THIS_MODULE, |
4181 |
-+ .unlocked_ioctl = fbcon_decor_ioctl, |
4182 |
-+ .compat_ioctl = fbcon_decor_compat_ioctl |
4183 |
-+}; |
4184 |
-+ |
4185 |
-+static struct miscdevice fbcon_decor_dev = { |
4186 |
-+ .minor = MISC_DYNAMIC_MINOR, |
4187 |
-+ .name = "fbcondecor", |
4188 |
-+ .fops = &fbcon_decor_ops |
4189 |
-+}; |
4190 |
-+ |
4191 |
-+void fbcon_decor_reset() |
4192 |
-+{ |
4193 |
-+ int i; |
4194 |
-+ |
4195 |
-+ for (i = 0; i < num_registered_fb; i++) { |
4196 |
-+ registered_fb[i]->bgdecor.data = NULL; |
4197 |
-+ registered_fb[i]->bgdecor.cmap.red = NULL; |
4198 |
-+ } |
4199 |
-+ |
4200 |
-+ for (i = 0; i < MAX_NR_CONSOLES && vc_cons[i].d; i++) { |
4201 |
-+ vc_cons[i].d->vc_decor.state = vc_cons[i].d->vc_decor.twidth = |
4202 |
-+ vc_cons[i].d->vc_decor.theight = 0; |
4203 |
-+ vc_cons[i].d->vc_decor.theme = NULL; |
4204 |
-+ } |
4205 |
-+ |
4206 |
-+ return; |
4207 |
-+} |
4208 |
-+ |
4209 |
-+int fbcon_decor_init() |
4210 |
-+{ |
4211 |
-+ int i; |
4212 |
-+ |
4213 |
-+ fbcon_decor_reset(); |
4214 |
-+ |
4215 |
-+ if (initialized) |
4216 |
-+ return 0; |
4217 |
-+ |
4218 |
-+ i = misc_register(&fbcon_decor_dev); |
4219 |
-+ if (i) { |
4220 |
-+ printk(KERN_ERR "fbcondecor: failed to register device\n"); |
4221 |
-+ return i; |
4222 |
-+ } |
4223 |
-+ |
4224 |
-+ fbcon_decor_call_helper("init", 0); |
4225 |
-+ initialized = 1; |
4226 |
-+ return 0; |
4227 |
-+} |
4228 |
-+ |
4229 |
-+int fbcon_decor_exit(void) |
4230 |
-+{ |
4231 |
-+ fbcon_decor_reset(); |
4232 |
-+ return 0; |
4233 |
-+} |
4234 |
-+ |
4235 |
-+EXPORT_SYMBOL(fbcon_decor_path); |
4236 |
---- a/drivers/video/console/fbcondecor.h 2014-01-21 09:50:42.229900176 -0500 |
4237 |
-+++ b/drivers/video/console/fbcondecor.h 2014-01-21 09:50:21.489900340 -0500 |
4238 |
-@@ -0,0 +1,79 @@ |
4239 |
-+/* |
4240 |
-+ * linux/drivers/video/console/fbcondecor.h -- Framebuffer Console Decoration headers |
4241 |
-+ * |
4242 |
-+ * Copyright (C) 2004 Michal Januszewski <spock@g.o> |
4243 |
-+ * |
4244 |
-+ */ |
4245 |
-+ |
4246 |
-+#ifndef __FBCON_DECOR_H |
4247 |
-+#define __FBCON_DECOR_H |
4248 |
-+ |
4249 |
-+#ifndef _LINUX_FB_H |
4250 |
-+#include <linux/fb.h> |
4251 |
-+#endif |
4252 |
-+ |
4253 |
-+/* This is needed for vc_cons in fbcmap.c */ |
4254 |
-+#include <linux/vt_kern.h> |
4255 |
-+ |
4256 |
-+struct fb_cursor; |
4257 |
-+struct fb_info; |
4258 |
-+struct vc_data; |
4259 |
-+ |
4260 |
-+#ifdef CONFIG_FB_CON_DECOR |
4261 |
-+/* fbcondecor.c */ |
4262 |
-+int fbcon_decor_init(void); |
4263 |
-+int fbcon_decor_exit(void); |
4264 |
-+int fbcon_decor_call_helper(char* cmd, unsigned short cons); |
4265 |
-+int fbcon_decor_disable(struct vc_data *vc, unsigned char redraw); |
4266 |
-+void fbcon_decor_reset(void); |
4267 |
-+ |
4268 |
-+/* cfbcondecor.c */ |
4269 |
-+void fbcon_decor_putcs(struct vc_data *vc, struct fb_info *info, const unsigned short *s, int count, int yy, int xx); |
4270 |
-+void fbcon_decor_cursor(struct fb_info *info, struct fb_cursor *cursor); |
4271 |
-+void fbcon_decor_clear(struct vc_data *vc, struct fb_info *info, int sy, int sx, int height, int width); |
4272 |
-+void fbcon_decor_clear_margins(struct vc_data *vc, struct fb_info *info, int bottom_only); |
4273 |
-+void fbcon_decor_blank(struct vc_data *vc, struct fb_info *info, int blank); |
4274 |
-+void fbcon_decor_bmove_redraw(struct vc_data *vc, struct fb_info *info, int y, int sx, int dx, int width); |
4275 |
-+void fbcon_decor_copy(u8 *dst, u8 *src, int height, int width, int linebytes, int srclinesbytes, int bpp); |
4276 |
-+void fbcon_decor_fix_pseudo_pal(struct fb_info *info, struct vc_data *vc); |
4277 |
-+ |
4278 |
-+/* vt.c */ |
4279 |
-+void acquire_console_sem(void); |
4280 |
-+void release_console_sem(void); |
4281 |
-+void do_unblank_screen(int entering_gfx); |
4282 |
-+ |
4283 |
-+/* struct vc_data *y */ |
4284 |
-+#define fbcon_decor_active_vc(y) (y->vc_decor.state && y->vc_decor.theme) |
4285 |
-+ |
4286 |
-+/* struct fb_info *x, struct vc_data *y */ |
4287 |
-+#define fbcon_decor_active_nores(x,y) (x->bgdecor.data && fbcon_decor_active_vc(y)) |
4288 |
-+ |
4289 |
-+/* struct fb_info *x, struct vc_data *y */ |
4290 |
-+#define fbcon_decor_active(x,y) (fbcon_decor_active_nores(x,y) && \ |
4291 |
-+ x->bgdecor.width == x->var.xres && \ |
4292 |
-+ x->bgdecor.height == x->var.yres && \ |
4293 |
-+ x->bgdecor.depth == x->var.bits_per_pixel) |
4294 |
-+ |
4295 |
-+ |
4296 |
-+#else /* CONFIG_FB_CON_DECOR */ |
4297 |
-+ |
4298 |
-+static inline void fbcon_decor_putcs(struct vc_data *vc, struct fb_info *info, const unsigned short *s, int count, int yy, int xx) {} |
4299 |
-+static inline void fbcon_decor_putc(struct vc_data *vc, struct fb_info *info, int c, int ypos, int xpos) {} |
4300 |
-+static inline void fbcon_decor_cursor(struct fb_info *info, struct fb_cursor *cursor) {} |
4301 |
-+static inline void fbcon_decor_clear(struct vc_data *vc, struct fb_info *info, int sy, int sx, int height, int width) {} |
4302 |
-+static inline void fbcon_decor_clear_margins(struct vc_data *vc, struct fb_info *info, int bottom_only) {} |
4303 |
-+static inline void fbcon_decor_blank(struct vc_data *vc, struct fb_info *info, int blank) {} |
4304 |
-+static inline void fbcon_decor_bmove_redraw(struct vc_data *vc, struct fb_info *info, int y, int sx, int dx, int width) {} |
4305 |
-+static inline void fbcon_decor_fix_pseudo_pal(struct fb_info *info, struct vc_data *vc) {} |
4306 |
-+static inline int fbcon_decor_call_helper(char* cmd, unsigned short cons) { return 0; } |
4307 |
-+static inline int fbcon_decor_init(void) { return 0; } |
4308 |
-+static inline int fbcon_decor_exit(void) { return 0; } |
4309 |
-+static inline int fbcon_decor_disable(struct vc_data *vc, unsigned char redraw) { return 0; } |
4310 |
-+ |
4311 |
-+#define fbcon_decor_active_vc(y) (0) |
4312 |
-+#define fbcon_decor_active_nores(x,y) (0) |
4313 |
-+#define fbcon_decor_active(x,y) (0) |
4314 |
-+ |
4315 |
-+#endif /* CONFIG_FB_CON_DECOR */ |
4316 |
-+ |
4317 |
-+#endif /* __FBCON_DECOR_H */ |
4318 |
-diff --git a/drivers/video/fbcmap.c b/drivers/video/fbcmap.c |
4319 |
-index 5c3960d..162b5f4 100644 |
4320 |
---- a/drivers/video/fbcmap.c |
4321 |
-+++ b/drivers/video/fbcmap.c |
4322 |
-@@ -17,6 +17,8 @@ |
4323 |
- #include <linux/slab.h> |
4324 |
- #include <linux/uaccess.h> |
4325 |
- |
4326 |
-+#include "console/fbcondecor.h" |
4327 |
-+ |
4328 |
- static u16 red2[] __read_mostly = { |
4329 |
- 0x0000, 0xaaaa |
4330 |
- }; |
4331 |
-@@ -249,14 +251,17 @@ int fb_set_cmap(struct fb_cmap *cmap, struct fb_info *info) |
4332 |
- if (transp) |
4333 |
- htransp = *transp++; |
4334 |
- if (info->fbops->fb_setcolreg(start++, |
4335 |
-- hred, hgreen, hblue, |
4336 |
-+ hred, hgreen, hblue, |
4337 |
- htransp, info)) |
4338 |
- break; |
4339 |
- } |
4340 |
- } |
4341 |
-- if (rc == 0) |
4342 |
-+ if (rc == 0) { |
4343 |
- fb_copy_cmap(cmap, &info->cmap); |
4344 |
-- |
4345 |
-+ if (fbcon_decor_active(info, vc_cons[fg_console].d) && |
4346 |
-+ info->fix.visual == FB_VISUAL_DIRECTCOLOR) |
4347 |
-+ fbcon_decor_fix_pseudo_pal(info, vc_cons[fg_console].d); |
4348 |
-+ } |
4349 |
- return rc; |
4350 |
- } |
4351 |
- |
4352 |
-diff --git a/drivers/video/fbmem.c b/drivers/video/fbmem.c |
4353 |
-index c6ce416..7ce6640 100644 |
4354 |
---- a/drivers/video/fbmem.c |
4355 |
-+++ b/drivers/video/fbmem.c |
4356 |
-@@ -1231,15 +1231,6 @@ struct fb_fix_screeninfo32 { |
4357 |
- u16 reserved[3]; |
4358 |
- }; |
4359 |
- |
4360 |
--struct fb_cmap32 { |
4361 |
-- u32 start; |
4362 |
-- u32 len; |
4363 |
-- compat_caddr_t red; |
4364 |
-- compat_caddr_t green; |
4365 |
-- compat_caddr_t blue; |
4366 |
-- compat_caddr_t transp; |
4367 |
--}; |
4368 |
-- |
4369 |
- static int fb_getput_cmap(struct fb_info *info, unsigned int cmd, |
4370 |
- unsigned long arg) |
4371 |
- { |
4372 |
-diff --git a/include/linux/console_decor.h b/include/linux/console_decor.h |
4373 |
-new file mode 100644 |
4374 |
-index 0000000..04b8d80 |
4375 |
---- /dev/null |
4376 |
-+++ b/include/linux/console_decor.h |
4377 |
-@@ -0,0 +1,46 @@ |
4378 |
-+#ifndef _LINUX_CONSOLE_DECOR_H_ |
4379 |
-+#define _LINUX_CONSOLE_DECOR_H_ 1 |
4380 |
-+ |
4381 |
-+/* A structure used by the framebuffer console decorations (drivers/video/console/fbcondecor.c) */ |
4382 |
-+struct vc_decor { |
4383 |
-+ __u8 bg_color; /* The color that is to be treated as transparent */ |
4384 |
-+ __u8 state; /* Current decor state: 0 = off, 1 = on */ |
4385 |
-+ __u16 tx, ty; /* Top left corner coordinates of the text field */ |
4386 |
-+ __u16 twidth, theight; /* Width and height of the text field */ |
4387 |
-+ char* theme; |
4388 |
-+}; |
4389 |
-+ |
4390 |
-+#ifdef __KERNEL__ |
4391 |
-+#ifdef CONFIG_COMPAT |
4392 |
-+#include <linux/compat.h> |
4393 |
-+ |
4394 |
-+struct vc_decor32 { |
4395 |
-+ __u8 bg_color; /* The color that is to be treated as transparent */ |
4396 |
-+ __u8 state; /* Current decor state: 0 = off, 1 = on */ |
4397 |
-+ __u16 tx, ty; /* Top left corner coordinates of the text field */ |
4398 |
-+ __u16 twidth, theight; /* Width and height of the text field */ |
4399 |
-+ compat_uptr_t theme; |
4400 |
-+}; |
4401 |
-+ |
4402 |
-+#define vc_decor_from_compat(to, from) \ |
4403 |
-+ (to).bg_color = (from).bg_color; \ |
4404 |
-+ (to).state = (from).state; \ |
4405 |
-+ (to).tx = (from).tx; \ |
4406 |
-+ (to).ty = (from).ty; \ |
4407 |
-+ (to).twidth = (from).twidth; \ |
4408 |
-+ (to).theight = (from).theight; \ |
4409 |
-+ (to).theme = compat_ptr((from).theme) |
4410 |
-+ |
4411 |
-+#define vc_decor_to_compat(to, from) \ |
4412 |
-+ (to).bg_color = (from).bg_color; \ |
4413 |
-+ (to).state = (from).state; \ |
4414 |
-+ (to).tx = (from).tx; \ |
4415 |
-+ (to).ty = (from).ty; \ |
4416 |
-+ (to).twidth = (from).twidth; \ |
4417 |
-+ (to).theight = (from).theight; \ |
4418 |
-+ (to).theme = ptr_to_compat((from).theme) |
4419 |
-+ |
4420 |
-+#endif /* CONFIG_COMPAT */ |
4421 |
-+#endif /* __KERNEL__ */ |
4422 |
-+ |
4423 |
-+#endif |
4424 |
-diff --git a/include/linux/console_struct.h b/include/linux/console_struct.h |
4425 |
-index 7f0c329..98f5d60 100644 |
4426 |
---- a/include/linux/console_struct.h |
4427 |
-+++ b/include/linux/console_struct.h |
4428 |
-@@ -19,6 +19,7 @@ |
4429 |
- struct vt_struct; |
4430 |
- |
4431 |
- #define NPAR 16 |
4432 |
-+#include <linux/console_decor.h> |
4433 |
- |
4434 |
- struct vc_data { |
4435 |
- struct tty_port port; /* Upper level data */ |
4436 |
-@@ -107,6 +108,8 @@ struct vc_data { |
4437 |
- unsigned long vc_uni_pagedir; |
4438 |
- unsigned long *vc_uni_pagedir_loc; /* [!] Location of uni_pagedir variable for this console */ |
4439 |
- bool vc_panic_force_write; /* when oops/panic this VC can accept forced output/blanking */ |
4440 |
-+ |
4441 |
-+ struct vc_decor vc_decor; |
4442 |
- /* additional information is in vt_kern.h */ |
4443 |
- }; |
4444 |
- |
4445 |
-diff --git a/include/uapi/linux/fb.h b/include/uapi/linux/fb.h |
4446 |
-index d31cb68..ad161bb 100644 |
4447 |
---- a/include/uapi/linux/fb.h |
4448 |
-+++ b/include/uapi/linux/fb.h |
4449 |
-@@ -8,6 +8,25 @@ |
4450 |
- |
4451 |
- #define FB_MAX 32 /* sufficient for now */ |
4452 |
- |
4453 |
-+struct fbcon_decor_iowrapper |
4454 |
-+{ |
4455 |
-+ unsigned short vc; /* Virtual console */ |
4456 |
-+ unsigned char origin; /* Point of origin of the request */ |
4457 |
-+ void *data; |
4458 |
-+}; |
4459 |
-+ |
4460 |
-+#ifdef __KERNEL__ |
4461 |
-+#ifdef CONFIG_COMPAT |
4462 |
-+#include <linux/compat.h> |
4463 |
-+struct fbcon_decor_iowrapper32 |
4464 |
-+{ |
4465 |
-+ unsigned short vc; /* Virtual console */ |
4466 |
-+ unsigned char origin; /* Point of origin of the request */ |
4467 |
-+ compat_uptr_t data; |
4468 |
-+}; |
4469 |
-+#endif /* CONFIG_COMPAT */ |
4470 |
-+#endif /* __KERNEL__ */ |
4471 |
-+ |
4472 |
- /* ioctls |
4473 |
- 0x46 is 'F' */ |
4474 |
- #define FBIOGET_VSCREENINFO 0x4600 |
4475 |
-@@ -34,6 +53,24 @@ |
4476 |
- #define FBIOPUT_MODEINFO 0x4617 |
4477 |
- #define FBIOGET_DISPINFO 0x4618 |
4478 |
- #define FBIO_WAITFORVSYNC _IOW('F', 0x20, __u32) |
4479 |
-+#define FBIOCONDECOR_SETCFG _IOWR('F', 0x19, struct fbcon_decor_iowrapper) |
4480 |
-+#define FBIOCONDECOR_GETCFG _IOR('F', 0x1A, struct fbcon_decor_iowrapper) |
4481 |
-+#define FBIOCONDECOR_SETSTATE _IOWR('F', 0x1B, struct fbcon_decor_iowrapper) |
4482 |
-+#define FBIOCONDECOR_GETSTATE _IOR('F', 0x1C, struct fbcon_decor_iowrapper) |
4483 |
-+#define FBIOCONDECOR_SETPIC _IOWR('F', 0x1D, struct fbcon_decor_iowrapper) |
4484 |
-+#ifdef __KERNEL__ |
4485 |
-+#ifdef CONFIG_COMPAT |
4486 |
-+#define FBIOCONDECOR_SETCFG32 _IOWR('F', 0x19, struct fbcon_decor_iowrapper32) |
4487 |
-+#define FBIOCONDECOR_GETCFG32 _IOR('F', 0x1A, struct fbcon_decor_iowrapper32) |
4488 |
-+#define FBIOCONDECOR_SETSTATE32 _IOWR('F', 0x1B, struct fbcon_decor_iowrapper32) |
4489 |
-+#define FBIOCONDECOR_GETSTATE32 _IOR('F', 0x1C, struct fbcon_decor_iowrapper32) |
4490 |
-+#define FBIOCONDECOR_SETPIC32 _IOWR('F', 0x1D, struct fbcon_decor_iowrapper32) |
4491 |
-+#endif /* CONFIG_COMPAT */ |
4492 |
-+#endif /* __KERNEL__ */ |
4493 |
-+ |
4494 |
-+#define FBCON_DECOR_THEME_LEN 128 /* Maximum lenght of a theme name */ |
4495 |
-+#define FBCON_DECOR_IO_ORIG_KERNEL 0 /* Kernel ioctl origin */ |
4496 |
-+#define FBCON_DECOR_IO_ORIG_USER 1 /* User ioctl origin */ |
4497 |
- |
4498 |
- #define FB_TYPE_PACKED_PIXELS 0 /* Packed Pixels */ |
4499 |
- #define FB_TYPE_PLANES 1 /* Non interleaved planes */ |
4500 |
-@@ -286,6 +323,28 @@ struct fb_cmap { |
4501 |
- __u16 *transp; /* transparency, can be NULL */ |
4502 |
- }; |
4503 |
- |
4504 |
-+#ifdef __KERNEL__ |
4505 |
-+#ifdef CONFIG_COMPAT |
4506 |
-+struct fb_cmap32 { |
4507 |
-+ __u32 start; |
4508 |
-+ __u32 len; /* Number of entries */ |
4509 |
-+ compat_uptr_t red; /* Red values */ |
4510 |
-+ compat_uptr_t green; |
4511 |
-+ compat_uptr_t blue; |
4512 |
-+ compat_uptr_t transp; /* transparency, can be NULL */ |
4513 |
-+}; |
4514 |
-+ |
4515 |
-+#define fb_cmap_from_compat(to, from) \ |
4516 |
-+ (to).start = (from).start; \ |
4517 |
-+ (to).len = (from).len; \ |
4518 |
-+ (to).red = compat_ptr((from).red); \ |
4519 |
-+ (to).green = compat_ptr((from).green); \ |
4520 |
-+ (to).blue = compat_ptr((from).blue); \ |
4521 |
-+ (to).transp = compat_ptr((from).transp) |
4522 |
-+ |
4523 |
-+#endif /* CONFIG_COMPAT */ |
4524 |
-+#endif /* __KERNEL__ */ |
4525 |
-+ |
4526 |
- struct fb_con2fbmap { |
4527 |
- __u32 console; |
4528 |
- __u32 framebuffer; |
4529 |
-@@ -367,6 +426,34 @@ struct fb_image { |
4530 |
- struct fb_cmap cmap; /* color map info */ |
4531 |
- }; |
4532 |
- |
4533 |
-+#ifdef __KERNEL__ |
4534 |
-+#ifdef CONFIG_COMPAT |
4535 |
-+struct fb_image32 { |
4536 |
-+ __u32 dx; /* Where to place image */ |
4537 |
-+ __u32 dy; |
4538 |
-+ __u32 width; /* Size of image */ |
4539 |
-+ __u32 height; |
4540 |
-+ __u32 fg_color; /* Only used when a mono bitmap */ |
4541 |
-+ __u32 bg_color; |
4542 |
-+ __u8 depth; /* Depth of the image */ |
4543 |
-+ const compat_uptr_t data; /* Pointer to image data */ |
4544 |
-+ struct fb_cmap32 cmap; /* color map info */ |
4545 |
-+}; |
4546 |
-+ |
4547 |
-+#define fb_image_from_compat(to, from) \ |
4548 |
-+ (to).dx = (from).dx; \ |
4549 |
-+ (to).dy = (from).dy; \ |
4550 |
-+ (to).width = (from).width; \ |
4551 |
-+ (to).height = (from).height; \ |
4552 |
-+ (to).fg_color = (from).fg_color; \ |
4553 |
-+ (to).bg_color = (from).bg_color; \ |
4554 |
-+ (to).depth = (from).depth; \ |
4555 |
-+ (to).data = compat_ptr((from).data); \ |
4556 |
-+ fb_cmap_from_compat((to).cmap, (from).cmap) |
4557 |
-+ |
4558 |
-+#endif /* CONFIG_COMPAT */ |
4559 |
-+#endif /* __KERNEL__ */ |
4560 |
-+ |
4561 |
- /* |
4562 |
- * hardware cursor control |
4563 |
- */ |
4564 |
- |
4565 |
-diff --git a/include/linux/fb.h b/include/linux/fb.h |
4566 |
-index d31cb68..ad161bb 100644 |
4567 |
---- a/include/linux/fb.h |
4568 |
-+++ b/include/linux/fb.h |
4569 |
-@@ -488,5 +488,8 @@ #define FBINFO_STATE_SUSPENDED 1 |
4570 |
- u32 state; /* Hardware state i.e suspend */ |
4571 |
- void *fbcon_par; /* fbcon use-only private area */ |
4572 |
-+ |
4573 |
-+ struct fb_image bgdecor; |
4574 |
-+ |
4575 |
- /* From here on everything is device dependent */ |
4576 |
- void *par; |
4577 |
- /* we need the PCI or similar aperture base/size not |
4578 |
- |
4579 |
-diff --git a/kernel/sysctl.c b/kernel/sysctl.c |
4580 |
-index 4ab1187..6561627 100644 |
4581 |
---- a/kernel/sysctl.c |
4582 |
-+++ b/kernel/sysctl.c |
4583 |
-@@ -145,6 +145,10 @@ static int min_percpu_pagelist_fract = 8; |
4584 |
- static int ngroups_max = NGROUPS_MAX; |
4585 |
- static const int cap_last_cap = CAP_LAST_CAP; |
4586 |
- |
4587 |
-+#ifdef CONFIG_FB_CON_DECOR |
4588 |
-+extern char fbcon_decor_path[]; |
4589 |
-+#endif |
4590 |
-+ |
4591 |
- #ifdef CONFIG_INOTIFY_USER |
4592 |
- #include <linux/inotify.h> |
4593 |
- #endif |
4594 |
-@@ -248,6 +252,15 @@ static struct ctl_table sysctl_base_table[] = { |
4595 |
- .mode = 0555, |
4596 |
- .child = dev_table, |
4597 |
- }, |
4598 |
-+#ifdef CONFIG_FB_CON_DECOR |
4599 |
-+ { |
4600 |
-+ .procname = "fbcondecor", |
4601 |
-+ .data = &fbcon_decor_path, |
4602 |
-+ .maxlen = KMOD_PATH_LEN, |
4603 |
-+ .mode = 0644, |
4604 |
-+ .proc_handler = &proc_dostring, |
4605 |
-+ }, |
4606 |
-+#endif |
4607 |
- { } |
4608 |
- }; |
4609 |
- |
4610 |
-@@ -1091,7 +1104,7 @@ static struct ctl_table vm_table[] = { |
4611 |
- .proc_handler = proc_dointvec, |
4612 |
- }, |
4613 |
- { |
4614 |
-- .procname = "page-cluster", |
4615 |
-+ .procname = "page-cluster", |
4616 |
- .data = &page_cluster, |
4617 |
- .maxlen = sizeof(int), |
4618 |
- .mode = 0644, |
4619 |
-@@ -1535,7 +1548,7 @@ static struct ctl_table fs_table[] = { |
4620 |
- .mode = 0555, |
4621 |
- .child = inotify_table, |
4622 |
- }, |
4623 |
--#endif |
4624 |
-+#endif |
4625 |
- #ifdef CONFIG_EPOLL |
4626 |
- { |
4627 |
- .procname = "epoll", |
4628 |
-@@ -1873,12 +1886,12 @@ static int __do_proc_dointvec(void *tbl_data, struct ctl_table *table, |
4629 |
- unsigned long page = 0; |
4630 |
- size_t left; |
4631 |
- char *kbuf; |
4632 |
-- |
4633 |
-+ |
4634 |
- if (!tbl_data || !table->maxlen || !*lenp || (*ppos && !write)) { |
4635 |
- *lenp = 0; |
4636 |
- return 0; |
4637 |
- } |
4638 |
-- |
4639 |
-+ |
4640 |
- i = (int *) tbl_data; |
4641 |
- vleft = table->maxlen / sizeof(*i); |
4642 |
- left = *lenp; |
4643 |
-@@ -1967,7 +1980,7 @@ static int do_proc_dointvec(struct ctl_table *table, int write, |
4644 |
- * @ppos: file position |
4645 |
- * |
4646 |
- * Reads/writes up to table->maxlen/sizeof(unsigned int) integer |
4647 |
-- * values from/to the user buffer, treated as an ASCII string. |
4648 |
-+ * values from/to the user buffer, treated as an ASCII string. |
4649 |
- * |
4650 |
- * Returns 0 on success. |
4651 |
- */ |
4652 |
-@@ -2326,7 +2339,7 @@ static int do_proc_dointvec_ms_jiffies_conv(bool *negp, unsigned long *lvalp, |
4653 |
- * @ppos: file position |
4654 |
- * |
4655 |
- * Reads/writes up to table->maxlen/sizeof(unsigned int) integer |
4656 |
-- * values from/to the user buffer, treated as an ASCII string. |
4657 |
-+ * values from/to the user buffer, treated as an ASCII string. |
4658 |
- * The values read are assumed to be in seconds, and are converted into |
4659 |
- * jiffies. |
4660 |
- * |
4661 |
-@@ -2348,8 +2361,8 @@ int proc_dointvec_jiffies(struct ctl_table *table, int write, |
4662 |
- * @ppos: pointer to the file position |
4663 |
- * |
4664 |
- * Reads/writes up to table->maxlen/sizeof(unsigned int) integer |
4665 |
-- * values from/to the user buffer, treated as an ASCII string. |
4666 |
-- * The values read are assumed to be in 1/USER_HZ seconds, and |
4667 |
-+ * values from/to the user buffer, treated as an ASCII string. |
4668 |
-+ * The values read are assumed to be in 1/USER_HZ seconds, and |
4669 |
- * are converted into jiffies. |
4670 |
- * |
4671 |
- * Returns 0 on success. |
4672 |
-@@ -2371,8 +2384,8 @@ int proc_dointvec_userhz_jiffies(struct ctl_table *table, int write, |
4673 |
- * @ppos: the current position in the file |
4674 |
- * |
4675 |
- * Reads/writes up to table->maxlen/sizeof(unsigned int) integer |
4676 |
-- * values from/to the user buffer, treated as an ASCII string. |
4677 |
-- * The values read are assumed to be in 1/1000 seconds, and |
4678 |
-+ * values from/to the user buffer, treated as an ASCII string. |
4679 |
-+ * The values read are assumed to be in 1/1000 seconds, and |
4680 |
- * are converted into jiffies. |
4681 |
- * |
4682 |
- * Returns 0 on success. |
4683 |
--- |
4684 |
-1.7.10 |
4685 |
- |
4686 |
|
4687 |
Deleted: genpatches-2.6/trunk/3.15/5000_enable-additional-cpu-optimizations-for-gcc.patch |
4688 |
=================================================================== |
4689 |
--- genpatches-2.6/trunk/3.15/5000_enable-additional-cpu-optimizations-for-gcc.patch 2014-05-05 14:22:27 UTC (rev 2771) |
4690 |
+++ genpatches-2.6/trunk/3.15/5000_enable-additional-cpu-optimizations-for-gcc.patch 2014-05-05 14:28:48 UTC (rev 2772) |
4691 |
@@ -1,325 +0,0 @@ |
4692 |
-This patch has been tested on and known to work with kernel versions from 3.2 |
4693 |
-up to the latest git version (pulled on 12/14/2013). |
4694 |
- |
4695 |
-This patch will expand the number of microarchitectures to include new |
4696 |
-processors including: AMD K10-family, AMD Family 10h (Barcelona), AMD Family |
4697 |
-14h (Bobcat), AMD Family 15h (Bulldozer), AMD Family 15h (Piledriver), AMD |
4698 |
-Family 16h (Jaguar), Intel 1st Gen Core i3/i5/i7 (Nehalem), Intel 2nd Gen Core |
4699 |
-i3/i5/i7 (Sandybridge), Intel 3rd Gen Core i3/i5/i7 (Ivybridge), and Intel 4th |
4700 |
-Gen Core i3/i5/i7 (Haswell). It also offers the compiler the 'native' flag. |
4701 |
- |
4702 |
-Small but real speed increases are measurable using a make endpoint comparing |
4703 |
-a generic kernel to one built with one of the respective microarchs. |
4704 |
- |
4705 |
-See the following experimental evidence supporting this statement: |
4706 |
-https://github.com/graysky2/kernel_gcc_patch |
4707 |
- |
4708 |
---- |
4709 |
-diff -uprN a/arch/x86/include/asm/module.h b/arch/x86/include/asm/module.h |
4710 |
---- a/arch/x86/include/asm/module.h 2013-11-03 18:41:51.000000000 -0500 |
4711 |
-+++ b/arch/x86/include/asm/module.h 2013-12-15 06:21:24.351122516 -0500 |
4712 |
-@@ -15,6 +15,16 @@ |
4713 |
- #define MODULE_PROC_FAMILY "586MMX " |
4714 |
- #elif defined CONFIG_MCORE2 |
4715 |
- #define MODULE_PROC_FAMILY "CORE2 " |
4716 |
-+#elif defined CONFIG_MNATIVE |
4717 |
-+#define MODULE_PROC_FAMILY "NATIVE " |
4718 |
-+#elif defined CONFIG_MCOREI7 |
4719 |
-+#define MODULE_PROC_FAMILY "COREI7 " |
4720 |
-+#elif defined CONFIG_MCOREI7AVX |
4721 |
-+#define MODULE_PROC_FAMILY "COREI7AVX " |
4722 |
-+#elif defined CONFIG_MCOREAVXI |
4723 |
-+#define MODULE_PROC_FAMILY "COREAVXI " |
4724 |
-+#elif defined CONFIG_MCOREAVX2 |
4725 |
-+#define MODULE_PROC_FAMILY "COREAVX2 " |
4726 |
- #elif defined CONFIG_MATOM |
4727 |
- #define MODULE_PROC_FAMILY "ATOM " |
4728 |
- #elif defined CONFIG_M686 |
4729 |
-@@ -33,6 +43,18 @@ |
4730 |
- #define MODULE_PROC_FAMILY "K7 " |
4731 |
- #elif defined CONFIG_MK8 |
4732 |
- #define MODULE_PROC_FAMILY "K8 " |
4733 |
-+#elif defined CONFIG_MK10 |
4734 |
-+#define MODULE_PROC_FAMILY "K10 " |
4735 |
-+#elif defined CONFIG_MBARCELONA |
4736 |
-+#define MODULE_PROC_FAMILY "BARCELONA " |
4737 |
-+#elif defined CONFIG_MBOBCAT |
4738 |
-+#define MODULE_PROC_FAMILY "BOBCAT " |
4739 |
-+#elif defined CONFIG_MBULLDOZER |
4740 |
-+#define MODULE_PROC_FAMILY "BULLDOZER " |
4741 |
-+#elif defined CONFIG_MPILEDRIVER |
4742 |
-+#define MODULE_PROC_FAMILY "PILEDRIVER " |
4743 |
-+#elif defined CONFIG_MJAGUAR |
4744 |
-+#define MODULE_PROC_FAMILY "JAGUAR " |
4745 |
- #elif defined CONFIG_MELAN |
4746 |
- #define MODULE_PROC_FAMILY "ELAN " |
4747 |
- #elif defined CONFIG_MCRUSOE |
4748 |
-diff -uprN a/arch/x86/Kconfig.cpu b/arch/x86/Kconfig.cpu |
4749 |
---- a/arch/x86/Kconfig.cpu 2013-11-03 18:41:51.000000000 -0500 |
4750 |
-+++ b/arch/x86/Kconfig.cpu 2013-12-15 06:21:24.351122516 -0500 |
4751 |
-@@ -139,7 +139,7 @@ config MPENTIUM4 |
4752 |
- |
4753 |
- |
4754 |
- config MK6 |
4755 |
-- bool "K6/K6-II/K6-III" |
4756 |
-+ bool "AMD K6/K6-II/K6-III" |
4757 |
- depends on X86_32 |
4758 |
- ---help--- |
4759 |
- Select this for an AMD K6-family processor. Enables use of |
4760 |
-@@ -147,7 +147,7 @@ config MK6 |
4761 |
- flags to GCC. |
4762 |
- |
4763 |
- config MK7 |
4764 |
-- bool "Athlon/Duron/K7" |
4765 |
-+ bool "AMD Athlon/Duron/K7" |
4766 |
- depends on X86_32 |
4767 |
- ---help--- |
4768 |
- Select this for an AMD Athlon K7-family processor. Enables use of |
4769 |
-@@ -155,12 +155,55 @@ config MK7 |
4770 |
- flags to GCC. |
4771 |
- |
4772 |
- config MK8 |
4773 |
-- bool "Opteron/Athlon64/Hammer/K8" |
4774 |
-+ bool "AMD Opteron/Athlon64/Hammer/K8" |
4775 |
- ---help--- |
4776 |
- Select this for an AMD Opteron or Athlon64 Hammer-family processor. |
4777 |
- Enables use of some extended instructions, and passes appropriate |
4778 |
- optimization flags to GCC. |
4779 |
- |
4780 |
-+config MK10 |
4781 |
-+ bool "AMD 61xx/7x50/PhenomX3/X4/II/K10" |
4782 |
-+ ---help--- |
4783 |
-+ Select this for an AMD 61xx Eight-Core Magny-Cours, Athlon X2 7x50, |
4784 |
-+ Phenom X3/X4/II, Athlon II X2/X3/X4, or Turion II-family processor. |
4785 |
-+ Enables use of some extended instructions, and passes appropriate |
4786 |
-+ optimization flags to GCC. |
4787 |
-+ |
4788 |
-+config MBARCELONA |
4789 |
-+ bool "AMD Barcelona" |
4790 |
-+ ---help--- |
4791 |
-+ Select this for AMD Barcelona and newer processors. |
4792 |
-+ |
4793 |
-+ Enables -march=barcelona |
4794 |
-+ |
4795 |
-+config MBOBCAT |
4796 |
-+ bool "AMD Bobcat" |
4797 |
-+ ---help--- |
4798 |
-+ Select this for AMD Bobcat processors. |
4799 |
-+ |
4800 |
-+ Enables -march=btver1 |
4801 |
-+ |
4802 |
-+config MBULLDOZER |
4803 |
-+ bool "AMD Bulldozer" |
4804 |
-+ ---help--- |
4805 |
-+ Select this for AMD Bulldozer processors. |
4806 |
-+ |
4807 |
-+ Enables -march=bdver1 |
4808 |
-+ |
4809 |
-+config MPILEDRIVER |
4810 |
-+ bool "AMD Piledriver" |
4811 |
-+ ---help--- |
4812 |
-+ Select this for AMD Piledriver processors. |
4813 |
-+ |
4814 |
-+ Enables -march=bdver2 |
4815 |
-+ |
4816 |
-+config MJAGUAR |
4817 |
-+ bool "AMD Jaguar" |
4818 |
-+ ---help--- |
4819 |
-+ Select this for AMD Jaguar processors. |
4820 |
-+ |
4821 |
-+ Enables -march=btver2 |
4822 |
-+ |
4823 |
- config MCRUSOE |
4824 |
- bool "Crusoe" |
4825 |
- depends on X86_32 |
4826 |
-@@ -251,8 +294,17 @@ config MPSC |
4827 |
- using the cpu family field |
4828 |
- in /proc/cpuinfo. Family 15 is an older Xeon, Family 6 a newer one. |
4829 |
- |
4830 |
-+config MATOM |
4831 |
-+ bool "Intel Atom" |
4832 |
-+ ---help--- |
4833 |
-+ |
4834 |
-+ Select this for the Intel Atom platform. Intel Atom CPUs have an |
4835 |
-+ in-order pipelining architecture and thus can benefit from |
4836 |
-+ accordingly optimized code. Use a recent GCC with specific Atom |
4837 |
-+ support in order to fully benefit from selecting this option. |
4838 |
-+ |
4839 |
- config MCORE2 |
4840 |
-- bool "Core 2/newer Xeon" |
4841 |
-+ bool "Intel Core 2" |
4842 |
- ---help--- |
4843 |
- |
4844 |
- Select this for Intel Core 2 and newer Core 2 Xeons (Xeon 51xx and |
4845 |
-@@ -260,14 +312,40 @@ config MCORE2 |
4846 |
- family in /proc/cpuinfo. Newer ones have 6 and older ones 15 |
4847 |
- (not a typo) |
4848 |
- |
4849 |
--config MATOM |
4850 |
-- bool "Intel Atom" |
4851 |
-+ Enables -march=core2 |
4852 |
-+ |
4853 |
-+config MCOREI7 |
4854 |
-+ bool "Intel Core i7" |
4855 |
- ---help--- |
4856 |
- |
4857 |
-- Select this for the Intel Atom platform. Intel Atom CPUs have an |
4858 |
-- in-order pipelining architecture and thus can benefit from |
4859 |
-- accordingly optimized code. Use a recent GCC with specific Atom |
4860 |
-- support in order to fully benefit from selecting this option. |
4861 |
-+ Select this for the Intel Nehalem platform. Intel Nehalem proecessors |
4862 |
-+ include Core i3, i5, i7, Xeon: 34xx, 35xx, 55xx, 56xx, 75xx processors. |
4863 |
-+ |
4864 |
-+ Enables -march=corei7 |
4865 |
-+ |
4866 |
-+config MCOREI7AVX |
4867 |
-+ bool "Intel Core 2nd Gen AVX" |
4868 |
-+ ---help--- |
4869 |
-+ |
4870 |
-+ Select this for 2nd Gen Core processors including Sandy Bridge. |
4871 |
-+ |
4872 |
-+ Enables -march=corei7-avx |
4873 |
-+ |
4874 |
-+config MCOREAVXI |
4875 |
-+ bool "Intel Core 3rd Gen AVX" |
4876 |
-+ ---help--- |
4877 |
-+ |
4878 |
-+ Select this for 3rd Gen Core processors including Ivy Bridge. |
4879 |
-+ |
4880 |
-+ Enables -march=core-avx-i |
4881 |
-+ |
4882 |
-+config MCOREAVX2 |
4883 |
-+ bool "Intel Core AVX2" |
4884 |
-+ ---help--- |
4885 |
-+ |
4886 |
-+ Select this for AVX2 enabled processors including Haswell. |
4887 |
-+ |
4888 |
-+ Enables -march=core-avx2 |
4889 |
- |
4890 |
- config GENERIC_CPU |
4891 |
- bool "Generic-x86-64" |
4892 |
-@@ -276,6 +354,19 @@ config GENERIC_CPU |
4893 |
- Generic x86-64 CPU. |
4894 |
- Run equally well on all x86-64 CPUs. |
4895 |
- |
4896 |
-+config MNATIVE |
4897 |
-+ bool "Native optimizations autodetected by GCC" |
4898 |
-+ ---help--- |
4899 |
-+ |
4900 |
-+ GCC 4.2 and above support -march=native, which automatically detects |
4901 |
-+ the optimum settings to use based on your processor. -march=native |
4902 |
-+ also detects and applies additional settings beyond -march specific |
4903 |
-+ to your CPU, (eg. -msse4). Unless you have a specific reason not to |
4904 |
-+ (e.g. distcc cross-compiling), you should probably be using |
4905 |
-+ -march=native rather than anything listed below. |
4906 |
-+ |
4907 |
-+ Enables -march=native |
4908 |
-+ |
4909 |
- endchoice |
4910 |
- |
4911 |
- config X86_GENERIC |
4912 |
-@@ -300,7 +391,7 @@ config X86_INTERNODE_CACHE_SHIFT |
4913 |
- config X86_L1_CACHE_SHIFT |
4914 |
- int |
4915 |
- default "7" if MPENTIUM4 || MPSC |
4916 |
-- default "6" if MK7 || MK8 || MPENTIUMM || MCORE2 || MATOM || MVIAC7 || X86_GENERIC || GENERIC_CPU |
4917 |
-+ default "6" if MK7 || MK8 || MK10 || MBARCELONA || MBOBCAT || MBULLDOZER || MPILEDRIVER || MJAGUAR || MPENTIUMM || MCORE2 || MCOREI7 || MCOREI7AVX || MCOREAVXI || MCOREAVX2 || MATOM || MVIAC7 || X86_GENERIC || MNATIVE || GENERIC_CPU |
4918 |
- default "4" if MELAN || M486 || MGEODEGX1 |
4919 |
- default "5" if MWINCHIP3D || MWINCHIPC6 || MCRUSOE || MEFFICEON || MCYRIXIII || MK6 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || M586 || MVIAC3_2 || MGEODE_LX |
4920 |
- |
4921 |
-@@ -331,11 +422,11 @@ config X86_ALIGNMENT_16 |
4922 |
- |
4923 |
- config X86_INTEL_USERCOPY |
4924 |
- def_bool y |
4925 |
-- depends on MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M586MMX || X86_GENERIC || MK8 || MK7 || MEFFICEON || MCORE2 |
4926 |
-+ depends on MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M586MMX || MNATIVE || X86_GENERIC || MK8 || MK7 || MK10 || MBARCELONA || MEFFICEON || MCORE2 || MCOREI7 || MCOREI7AVX || MCOREAVXI || MCOREAVX2 |
4927 |
- |
4928 |
- config X86_USE_PPRO_CHECKSUM |
4929 |
- def_bool y |
4930 |
-- depends on MWINCHIP3D || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2 || MVIAC7 || MEFFICEON || MGEODE_LX || MCORE2 || MATOM |
4931 |
-+ depends on MWINCHIP3D || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MK10 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2 || MVIAC7 || MEFFICEON || MGEODE_LX || MCORE2 || MCOREI7 || MCOREI7AVX || MCOREAVXI || MCOREAVX2 || MATOM || MNATIVE |
4932 |
- |
4933 |
- config X86_USE_3DNOW |
4934 |
- def_bool y |
4935 |
-@@ -363,17 +454,17 @@ config X86_P6_NOP |
4936 |
- |
4937 |
- config X86_TSC |
4938 |
- def_bool y |
4939 |
-- depends on ((MWINCHIP3D || MCRUSOE || MEFFICEON || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2 || MVIAC7 || MGEODEGX1 || MGEODE_LX || MCORE2 || MATOM) && !X86_NUMAQ) || X86_64 |
4940 |
-+ depends on ((MWINCHIP3D || MCRUSOE || MEFFICEON || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MK10 || MBARCELONA || MBOBCAT || MBULLDOZER || MPILEDRIVER || MJAGUAR || MVIAC3_2 || MVIAC7 || MGEODEGX1 || MGEODE_LX || MCORE2 || MCOREI7 || MCOREI7-AVX || MATOM) && !X86_NUMAQ) || X86_64 || MNATIVE |
4941 |
- |
4942 |
- config X86_CMPXCHG64 |
4943 |
- def_bool y |
4944 |
-- depends on X86_PAE || X86_64 || MCORE2 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MATOM |
4945 |
-+ depends on X86_PAE || X86_64 || MCORE2 || MCOREI7 || MCOREI7AVX || MCOREAVXI || MCOREAVX2 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MATOM || MNATIVE |
4946 |
- |
4947 |
- # this should be set for all -march=.. options where the compiler |
4948 |
- # generates cmov. |
4949 |
- config X86_CMOV |
4950 |
- def_bool y |
4951 |
-- depends on (MK8 || MK7 || MCORE2 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MVIAC3_2 || MVIAC7 || MCRUSOE || MEFFICEON || X86_64 || MATOM || MGEODE_LX) |
4952 |
-+ depends on (MK8 || MK10 || MBARCELONA || MBOBCAT || MBULLDOZER || MPILEDRIVER || MJAGUAR || MK7 || MCORE2 || MCOREI7 || MCOREI7AVX || MCOREAVXI || MCOREAVX2 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MVIAC3_2 || MVIAC7 || MCRUSOE || MEFFICEON || X86_64 || MNATIVE || MATOM || MGEODE_LX) |
4953 |
- |
4954 |
- config X86_MINIMUM_CPU_FAMILY |
4955 |
- int |
4956 |
-diff -uprN a/arch/x86/Makefile b/arch/x86/Makefile |
4957 |
---- a/arch/x86/Makefile 2013-11-03 18:41:51.000000000 -0500 |
4958 |
-+++ b/arch/x86/Makefile 2013-12-15 06:21:24.354455723 -0500 |
4959 |
-@@ -61,11 +61,26 @@ else |
4960 |
- KBUILD_CFLAGS += $(call cc-option,-mno-sse -mpreferred-stack-boundary=3) |
4961 |
- |
4962 |
- # FIXME - should be integrated in Makefile.cpu (Makefile_32.cpu) |
4963 |
-+ cflags-$(CONFIG_MNATIVE) += $(call cc-option,-march=native) |
4964 |
- cflags-$(CONFIG_MK8) += $(call cc-option,-march=k8) |
4965 |
-+ cflags-$(CONFIG_MK10) += $(call cc-option,-march=amdfam10) |
4966 |
-+ cflags-$(CONFIG_MBARCELONA) += $(call cc-option,-march=barcelona) |
4967 |
-+ cflags-$(CONFIG_MBOBCAT) += $(call cc-option,-march=btver1) |
4968 |
-+ cflags-$(CONFIG_MBULLDOZER) += $(call cc-option,-march=bdver1) |
4969 |
-+ cflags-$(CONFIG_MPILEDRIVER) += $(call cc-option,-march=bdver2) |
4970 |
-+ cflags-$(CONFIG_MJAGUAR) += $(call cc-option,-march=btver2) |
4971 |
- cflags-$(CONFIG_MPSC) += $(call cc-option,-march=nocona) |
4972 |
- |
4973 |
- cflags-$(CONFIG_MCORE2) += \ |
4974 |
-- $(call cc-option,-march=core2,$(call cc-option,-mtune=generic)) |
4975 |
-+ $(call cc-option,-march=core2,$(call cc-option,-mtune=core2)) |
4976 |
-+ cflags-$(CONFIG_MCOREI7) += \ |
4977 |
-+ $(call cc-option,-march=corei7,$(call cc-option,-mtune=corei7)) |
4978 |
-+ cflags-$(CONFIG_MCOREI7AVX) += \ |
4979 |
-+ $(call cc-option,-march=corei7-avx,$(call cc-option,-mtune=corei7-avx)) |
4980 |
-+ cflags-$(CONFIG_MCOREAVXI) += \ |
4981 |
-+ $(call cc-option,-march=core-avx-i,$(call cc-option,-mtune=core-avx-i)) |
4982 |
-+ cflags-$(CONFIG_MCOREAVX2) += \ |
4983 |
-+ $(call cc-option,-march=core-avx2,$(call cc-option,-mtune=core-avx2)) |
4984 |
- cflags-$(CONFIG_MATOM) += $(call cc-option,-march=atom) \ |
4985 |
- $(call cc-option,-mtune=atom,$(call cc-option,-mtune=generic)) |
4986 |
- cflags-$(CONFIG_GENERIC_CPU) += $(call cc-option,-mtune=generic) |
4987 |
-diff -uprN a/arch/x86/Makefile_32.cpu b/arch/x86/Makefile_32.cpu |
4988 |
---- a/arch/x86/Makefile_32.cpu 2013-11-03 18:41:51.000000000 -0500 |
4989 |
-+++ b/arch/x86/Makefile_32.cpu 2013-12-15 06:21:24.354455723 -0500 |
4990 |
-@@ -23,7 +23,14 @@ cflags-$(CONFIG_MK6) += -march=k6 |
4991 |
- # Please note, that patches that add -march=athlon-xp and friends are pointless. |
4992 |
- # They make zero difference whatsosever to performance at this time. |
4993 |
- cflags-$(CONFIG_MK7) += -march=athlon |
4994 |
-+cflags-$(CONFIG_MNATIVE) += $(call cc-option,-march=native) |
4995 |
- cflags-$(CONFIG_MK8) += $(call cc-option,-march=k8,-march=athlon) |
4996 |
-+cflags-$(CONFIG_MK10) += $(call cc-option,-march=amdfam10,-march=athlon) |
4997 |
-+cflags-$(CONFIG_MBARCELONA) += $(call cc-option,-march=barcelona,-march=athlon) |
4998 |
-+cflags-$(CONFIG_MBOBCAT) += $(call cc-option,-march=btver1,-march=athlon) |
4999 |
-+cflags-$(CONFIG_MBULLDOZER) += $(call cc-option,-march=bdver1,-march=athlon) |
5000 |
-+cflags-$(CONFIG_MPILEDRIVER) += $(call cc-option,-march=bdver2,-march=athlon) |
5001 |
-+cflags-$(CONFIG_MJAGUAR) += $(call cc-option,-march=btver2,-march=athlon) |
5002 |
- cflags-$(CONFIG_MCRUSOE) += -march=i686 $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0 |
5003 |
- cflags-$(CONFIG_MEFFICEON) += -march=i686 $(call tune,pentium3) $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0 |
5004 |
- cflags-$(CONFIG_MWINCHIPC6) += $(call cc-option,-march=winchip-c6,-march=i586) |
5005 |
-@@ -32,6 +39,10 @@ cflags-$(CONFIG_MCYRIXIII) += $(call cc- |
5006 |
- cflags-$(CONFIG_MVIAC3_2) += $(call cc-option,-march=c3-2,-march=i686) |
5007 |
- cflags-$(CONFIG_MVIAC7) += -march=i686 |
5008 |
- cflags-$(CONFIG_MCORE2) += -march=i686 $(call tune,core2) |
5009 |
-+cflags-$(CONFIG_MCOREI7) += -march=i686 $(call tune,corei7) |
5010 |
-+cflags-$(CONFIG_MCOREI7AVX) += -march=i686 $(call tune,corei7-avx) |
5011 |
-+cflags-$(CONFIG_MCOREAVXI) += -march=i686 $(call tune,core-avx-i) |
5012 |
-+cflags-$(CONFIG_MCOREAVX2) += -march=i686 $(call tune,core-avx2) |
5013 |
- cflags-$(CONFIG_MATOM) += $(call cc-option,-march=atom,$(call cc-option,-march=core2,-march=i686)) \ |
5014 |
- $(call cc-option,-mtune=atom,$(call cc-option,-mtune=generic)) |
5015 |
- |
5016 |
- |
5017 |
|
5018 |
Deleted: genpatches-2.6/trunk/3.15/5001_BFQ-1-block-cgroups-kconfig-build-bits-for-BFQ-v7r2-3.14.patch |
5019 |
=================================================================== |
5020 |
--- genpatches-2.6/trunk/3.15/5001_BFQ-1-block-cgroups-kconfig-build-bits-for-BFQ-v7r2-3.14.patch 2014-05-05 14:22:27 UTC (rev 2771) |
5021 |
+++ genpatches-2.6/trunk/3.15/5001_BFQ-1-block-cgroups-kconfig-build-bits-for-BFQ-v7r2-3.14.patch 2014-05-05 14:28:48 UTC (rev 2772) |
5022 |
@@ -1,104 +0,0 @@ |
5023 |
-From c3280db98437c9520f04ecacfdf1a868d7a4b7b3 Mon Sep 17 00:00:00 2001 |
5024 |
-From: Paolo Valente <paolo.valente@×××××××.it> |
5025 |
-Date: Tue, 3 Sep 2013 16:50:42 +0200 |
5026 |
-Subject: [PATCH 1/3] block: cgroups, kconfig, build bits for BFQ-v7r2-3.14 |
5027 |
- |
5028 |
-Update Kconfig.iosched and do the related Makefile changes to include |
5029 |
-kernel configuration options for BFQ. Also add the bfqio controller |
5030 |
-to the cgroups subsystem. |
5031 |
- |
5032 |
-Signed-off-by: Paolo Valente <paolo.valente@×××××××.it> |
5033 |
-Signed-off-by: Arianna Avanzini <avanzini.arianna@×××××.com> |
5034 |
---- |
5035 |
- block/Kconfig.iosched | 32 ++++++++++++++++++++++++++++++++ |
5036 |
- block/Makefile | 1 + |
5037 |
- include/linux/cgroup_subsys.h | 4 ++++ |
5038 |
- 3 files changed, 37 insertions(+) |
5039 |
- |
5040 |
-diff --git a/block/Kconfig.iosched b/block/Kconfig.iosched |
5041 |
-index 421bef9..0ee5f0f 100644 |
5042 |
---- a/block/Kconfig.iosched |
5043 |
-+++ b/block/Kconfig.iosched |
5044 |
-@@ -39,6 +39,27 @@ config CFQ_GROUP_IOSCHED |
5045 |
- ---help--- |
5046 |
- Enable group IO scheduling in CFQ. |
5047 |
- |
5048 |
-+config IOSCHED_BFQ |
5049 |
-+ tristate "BFQ I/O scheduler" |
5050 |
-+ default n |
5051 |
-+ ---help--- |
5052 |
-+ The BFQ I/O scheduler tries to distribute bandwidth among |
5053 |
-+ all processes according to their weights. |
5054 |
-+ It aims at distributing the bandwidth as desired, independently of |
5055 |
-+ the disk parameters and with any workload. It also tries to |
5056 |
-+ guarantee low latency to interactive and soft real-time |
5057 |
-+ applications. If compiled built-in (saying Y here), BFQ can |
5058 |
-+ be configured to support hierarchical scheduling. |
5059 |
-+ |
5060 |
-+config CGROUP_BFQIO |
5061 |
-+ bool "BFQ hierarchical scheduling support" |
5062 |
-+ depends on CGROUPS && IOSCHED_BFQ=y |
5063 |
-+ default n |
5064 |
-+ ---help--- |
5065 |
-+ Enable hierarchical scheduling in BFQ, using the cgroups |
5066 |
-+ filesystem interface. The name of the subsystem will be |
5067 |
-+ bfqio. |
5068 |
-+ |
5069 |
- choice |
5070 |
- prompt "Default I/O scheduler" |
5071 |
- default DEFAULT_CFQ |
5072 |
-@@ -52,6 +73,16 @@ choice |
5073 |
- config DEFAULT_CFQ |
5074 |
- bool "CFQ" if IOSCHED_CFQ=y |
5075 |
- |
5076 |
-+ config DEFAULT_BFQ |
5077 |
-+ bool "BFQ" if IOSCHED_BFQ=y |
5078 |
-+ help |
5079 |
-+ Selects BFQ as the default I/O scheduler which will be |
5080 |
-+ used by default for all block devices. |
5081 |
-+ The BFQ I/O scheduler aims at distributing the bandwidth |
5082 |
-+ as desired, independently of the disk parameters and with |
5083 |
-+ any workload. It also tries to guarantee low latency to |
5084 |
-+ interactive and soft real-time applications. |
5085 |
-+ |
5086 |
- config DEFAULT_NOOP |
5087 |
- bool "No-op" |
5088 |
- |
5089 |
-@@ -61,6 +92,7 @@ config DEFAULT_IOSCHED |
5090 |
- string |
5091 |
- default "deadline" if DEFAULT_DEADLINE |
5092 |
- default "cfq" if DEFAULT_CFQ |
5093 |
-+ default "bfq" if DEFAULT_BFQ |
5094 |
- default "noop" if DEFAULT_NOOP |
5095 |
- |
5096 |
- endmenu |
5097 |
-diff --git a/block/Makefile b/block/Makefile |
5098 |
-index 20645e8..cbd83fb 100644 |
5099 |
---- a/block/Makefile |
5100 |
-+++ b/block/Makefile |
5101 |
-@@ -16,6 +16,7 @@ obj-$(CONFIG_BLK_DEV_THROTTLING) += blk-throttle.o |
5102 |
- obj-$(CONFIG_IOSCHED_NOOP) += noop-iosched.o |
5103 |
- obj-$(CONFIG_IOSCHED_DEADLINE) += deadline-iosched.o |
5104 |
- obj-$(CONFIG_IOSCHED_CFQ) += cfq-iosched.o |
5105 |
-+obj-$(CONFIG_IOSCHED_BFQ) += bfq-iosched.o |
5106 |
- |
5107 |
- obj-$(CONFIG_BLOCK_COMPAT) += compat_ioctl.o |
5108 |
- obj-$(CONFIG_BLK_DEV_INTEGRITY) += blk-integrity.o |
5109 |
-diff --git a/include/linux/cgroup_subsys.h b/include/linux/cgroup_subsys.h |
5110 |
-index 7b99d71..4e8c0ff 100644 |
5111 |
---- a/include/linux/cgroup_subsys.h |
5112 |
-+++ b/include/linux/cgroup_subsys.h |
5113 |
-@@ -39,6 +39,10 @@ SUBSYS(net_cls) |
5114 |
- SUBSYS(blkio) |
5115 |
- #endif |
5116 |
- |
5117 |
-+#if IS_SUBSYS_ENABLED(CONFIG_CGROUP_BFQIO) |
5118 |
-+SUBSYS(bfqio) |
5119 |
-+#endif |
5120 |
-+ |
5121 |
- #if IS_SUBSYS_ENABLED(CONFIG_CGROUP_PERF) |
5122 |
- SUBSYS(perf) |
5123 |
- #endif |
5124 |
--- |
5125 |
-1.9.0 |
5126 |
- |
5127 |
|
5128 |
Deleted: genpatches-2.6/trunk/3.15/5002_BFQ-2-block-introduce-the-BFQ-v7r2-I-O-sched-for-3.14.patch1 |
5129 |
=================================================================== |
5130 |
--- genpatches-2.6/trunk/3.15/5002_BFQ-2-block-introduce-the-BFQ-v7r2-I-O-sched-for-3.14.patch1 2014-05-05 14:22:27 UTC (rev 2771) |
5131 |
+++ genpatches-2.6/trunk/3.15/5002_BFQ-2-block-introduce-the-BFQ-v7r2-I-O-sched-for-3.14.patch1 2014-05-05 14:28:48 UTC (rev 2772) |
5132 |
@@ -1,6065 +0,0 @@ |
5133 |
-From 5055277df59d9280da6b60cf90bed8e5e57dc44d Mon Sep 17 00:00:00 2001 |
5134 |
-From: Paolo Valente <paolo.valente@×××××××.it> |
5135 |
-Date: Thu, 9 May 2013 19:10:02 +0200 |
5136 |
-Subject: [PATCH 2/3] block: introduce the BFQ-v7r2 I/O sched for 3.14 |
5137 |
- |
5138 |
-Add the BFQ-v7r2 I/O scheduler to 3.14. |
5139 |
-The general structure is borrowed from CFQ, as much of the code for |
5140 |
-handling I/O contexts. Over time, several useful features have been |
5141 |
-ported from CFQ as well (details in the changelog in README.BFQ). A |
5142 |
-(bfq_)queue is associated to each task doing I/O on a device, and each |
5143 |
-time a scheduling decision has to be made a queue is selected and served |
5144 |
-until it expires. |
5145 |
- |
5146 |
- - Slices are given in the service domain: tasks are assigned |
5147 |
- budgets, measured in number of sectors. Once got the disk, a task |
5148 |
- must however consume its assigned budget within a configurable |
5149 |
- maximum time (by default, the maximum possible value of the |
5150 |
- budgets is automatically computed to comply with this timeout). |
5151 |
- This allows the desired latency vs "throughput boosting" tradeoff |
5152 |
- to be set. |
5153 |
- |
5154 |
- - Budgets are scheduled according to a variant of WF2Q+, implemented |
5155 |
- using an augmented rb-tree to take eligibility into account while |
5156 |
- preserving an O(log N) overall complexity. |
5157 |
- |
5158 |
- - A low-latency tunable is provided; if enabled, both interactive |
5159 |
- and soft real-time applications are guaranteed a very low latency. |
5160 |
- |
5161 |
- - Latency guarantees are preserved also in the presence of NCQ. |
5162 |
- |
5163 |
- - Also with flash-based devices, a high throughput is achieved |
5164 |
- while still preserving latency guarantees. |
5165 |
- |
5166 |
- - BFQ features Early Queue Merge (EQM), a sort of fusion of the |
5167 |
- cooperating-queue-merging and the preemption mechanisms present |
5168 |
- in CFQ. EQM is in fact a unified mechanism that tries to get a |
5169 |
- sequential read pattern, and hence a high throughput, with any |
5170 |
- set of processes performing interleaved I/O over a contiguous |
5171 |
- sequence of sectors. |
5172 |
- |
5173 |
- - BFQ supports full hierarchical scheduling, exporting a cgroups |
5174 |
- interface. Since each node has a full scheduler, each group can |
5175 |
- be assigned its own weight. |
5176 |
- |
5177 |
- - If the cgroups interface is not used, only I/O priorities can be |
5178 |
- assigned to processes, with ioprio values mapped to weights |
5179 |
- with the relation weight = IOPRIO_BE_NR - ioprio. |
5180 |
- |
5181 |
- - ioprio classes are served in strict priority order, i.e., lower |
5182 |
- priority queues are not served as long as there are higher |
5183 |
- priority queues. Among queues in the same class the bandwidth is |
5184 |
- distributed in proportion to the weight of each queue. A very |
5185 |
- thin extra bandwidth is however guaranteed to the Idle class, to |
5186 |
- prevent it from starving. |
5187 |
- |
5188 |
-Signed-off-by: Paolo Valente <paolo.valente@×××××××.it> |
5189 |
-Signed-off-by: Arianna Avanzini <avanzini.arianna@×××××.com> |
5190 |
---- |
5191 |
- block/bfq-cgroup.c | 926 +++++++++++++++ |
5192 |
- block/bfq-ioc.c | 36 + |
5193 |
- block/bfq-iosched.c | 3300 +++++++++++++++++++++++++++++++++++++++++++++++++++ |
5194 |
- block/bfq-sched.c | 1078 +++++++++++++++++ |
5195 |
- block/bfq.h | 622 ++++++++++ |
5196 |
- 5 files changed, 5962 insertions(+) |
5197 |
- create mode 100644 block/bfq-cgroup.c |
5198 |
- create mode 100644 block/bfq-ioc.c |
5199 |
- create mode 100644 block/bfq-iosched.c |
5200 |
- create mode 100644 block/bfq-sched.c |
5201 |
- create mode 100644 block/bfq.h |
5202 |
- |
5203 |
-diff --git a/block/bfq-cgroup.c b/block/bfq-cgroup.c |
5204 |
-new file mode 100644 |
5205 |
-index 0000000..bcecdb4 |
5206 |
---- /dev/null |
5207 |
-+++ b/block/bfq-cgroup.c |
5208 |
-@@ -0,0 +1,926 @@ |
5209 |
-+/* |
5210 |
-+ * BFQ: CGROUPS support. |
5211 |
-+ * |
5212 |
-+ * Based on ideas and code from CFQ: |
5213 |
-+ * Copyright (C) 2003 Jens Axboe <axboe@××××××.dk> |
5214 |
-+ * |
5215 |
-+ * Copyright (C) 2008 Fabio Checconi <fabio@×××××××××××××.it> |
5216 |
-+ * Paolo Valente <paolo.valente@×××××××.it> |
5217 |
-+ * |
5218 |
-+ * Copyright (C) 2010 Paolo Valente <paolo.valente@×××××××.it> |
5219 |
-+ * |
5220 |
-+ * Licensed under the GPL-2 as detailed in the accompanying COPYING.BFQ file. |
5221 |
-+ */ |
5222 |
-+ |
5223 |
-+#ifdef CONFIG_CGROUP_BFQIO |
5224 |
-+ |
5225 |
-+static DEFINE_MUTEX(bfqio_mutex); |
5226 |
-+ |
5227 |
-+static bool bfqio_is_removed(struct bfqio_cgroup *bgrp) |
5228 |
-+{ |
5229 |
-+ return bgrp ? !bgrp->online : false; |
5230 |
-+} |
5231 |
-+ |
5232 |
-+static struct bfqio_cgroup bfqio_root_cgroup = { |
5233 |
-+ .weight = BFQ_DEFAULT_GRP_WEIGHT, |
5234 |
-+ .ioprio = BFQ_DEFAULT_GRP_IOPRIO, |
5235 |
-+ .ioprio_class = BFQ_DEFAULT_GRP_CLASS, |
5236 |
-+}; |
5237 |
-+ |
5238 |
-+static inline void bfq_init_entity(struct bfq_entity *entity, |
5239 |
-+ struct bfq_group *bfqg) |
5240 |
-+{ |
5241 |
-+ entity->weight = entity->new_weight; |
5242 |
-+ entity->orig_weight = entity->new_weight; |
5243 |
-+ entity->ioprio = entity->new_ioprio; |
5244 |
-+ entity->ioprio_class = entity->new_ioprio_class; |
5245 |
-+ entity->parent = bfqg->my_entity; |
5246 |
-+ entity->sched_data = &bfqg->sched_data; |
5247 |
-+} |
5248 |
-+ |
5249 |
-+static struct bfqio_cgroup *css_to_bfqio(struct cgroup_subsys_state *css) |
5250 |
-+{ |
5251 |
-+ return css ? container_of(css, struct bfqio_cgroup, css) : NULL; |
5252 |
-+} |
5253 |
-+ |
5254 |
-+/* |
5255 |
-+ * Search the bfq_group for bfqd into the hash table (by now only a list) |
5256 |
-+ * of bgrp. Must be called under rcu_read_lock(). |
5257 |
-+ */ |
5258 |
-+static struct bfq_group *bfqio_lookup_group(struct bfqio_cgroup *bgrp, |
5259 |
-+ struct bfq_data *bfqd) |
5260 |
-+{ |
5261 |
-+ struct bfq_group *bfqg; |
5262 |
-+ void *key; |
5263 |
-+ |
5264 |
-+ hlist_for_each_entry_rcu(bfqg, &bgrp->group_data, group_node) { |
5265 |
-+ key = rcu_dereference(bfqg->bfqd); |
5266 |
-+ if (key == bfqd) |
5267 |
-+ return bfqg; |
5268 |
-+ } |
5269 |
-+ |
5270 |
-+ return NULL; |
5271 |
-+} |
5272 |
-+ |
5273 |
-+static inline void bfq_group_init_entity(struct bfqio_cgroup *bgrp, |
5274 |
-+ struct bfq_group *bfqg) |
5275 |
-+{ |
5276 |
-+ struct bfq_entity *entity = &bfqg->entity; |
5277 |
-+ |
5278 |
-+ /* |
5279 |
-+ * If the weight of the entity has never been set via the sysfs |
5280 |
-+ * interface, then bgrp->weight == 0. In this case we initialize |
5281 |
-+ * the weight from the current ioprio value. Otherwise, the group |
5282 |
-+ * weight, if set, has priority over the ioprio value. |
5283 |
-+ */ |
5284 |
-+ if (bgrp->weight == 0) { |
5285 |
-+ entity->new_weight = bfq_ioprio_to_weight(bgrp->ioprio); |
5286 |
-+ entity->new_ioprio = bgrp->ioprio; |
5287 |
-+ } else { |
5288 |
-+ entity->new_weight = bgrp->weight; |
5289 |
-+ entity->new_ioprio = bfq_weight_to_ioprio(bgrp->weight); |
5290 |
-+ } |
5291 |
-+ entity->orig_weight = entity->weight = entity->new_weight; |
5292 |
-+ entity->ioprio = entity->new_ioprio; |
5293 |
-+ entity->ioprio_class = entity->new_ioprio_class = bgrp->ioprio_class; |
5294 |
-+ entity->my_sched_data = &bfqg->sched_data; |
5295 |
-+} |
5296 |
-+ |
5297 |
-+static inline void bfq_group_set_parent(struct bfq_group *bfqg, |
5298 |
-+ struct bfq_group *parent) |
5299 |
-+{ |
5300 |
-+ struct bfq_entity *entity; |
5301 |
-+ |
5302 |
-+ BUG_ON(parent == NULL); |
5303 |
-+ BUG_ON(bfqg == NULL); |
5304 |
-+ |
5305 |
-+ entity = &bfqg->entity; |
5306 |
-+ entity->parent = parent->my_entity; |
5307 |
-+ entity->sched_data = &parent->sched_data; |
5308 |
-+} |
5309 |
-+ |
5310 |
-+/** |
5311 |
-+ * bfq_group_chain_alloc - allocate a chain of groups. |
5312 |
-+ * @bfqd: queue descriptor. |
5313 |
-+ * @css: the leaf cgroup_subsys_state this chain starts from. |
5314 |
-+ * |
5315 |
-+ * Allocate a chain of groups starting from the one belonging to |
5316 |
-+ * @cgroup up to the root cgroup. Stop if a cgroup on the chain |
5317 |
-+ * to the root has already an allocated group on @bfqd. |
5318 |
-+ */ |
5319 |
-+static struct bfq_group *bfq_group_chain_alloc(struct bfq_data *bfqd, |
5320 |
-+ struct cgroup_subsys_state *css) |
5321 |
-+{ |
5322 |
-+ struct bfqio_cgroup *bgrp; |
5323 |
-+ struct bfq_group *bfqg, *prev = NULL, *leaf = NULL; |
5324 |
-+ |
5325 |
-+ for (; css != NULL; css = css->parent) { |
5326 |
-+ bgrp = css_to_bfqio(css); |
5327 |
-+ |
5328 |
-+ bfqg = bfqio_lookup_group(bgrp, bfqd); |
5329 |
-+ if (bfqg != NULL) { |
5330 |
-+ /* |
5331 |
-+ * All the cgroups in the path from there to the |
5332 |
-+ * root must have a bfq_group for bfqd, so we don't |
5333 |
-+ * need any more allocations. |
5334 |
-+ */ |
5335 |
-+ break; |
5336 |
-+ } |
5337 |
-+ |
5338 |
-+ bfqg = kzalloc(sizeof(*bfqg), GFP_ATOMIC); |
5339 |
-+ if (bfqg == NULL) |
5340 |
-+ goto cleanup; |
5341 |
-+ |
5342 |
-+ bfq_group_init_entity(bgrp, bfqg); |
5343 |
-+ bfqg->my_entity = &bfqg->entity; |
5344 |
-+ |
5345 |
-+ if (leaf == NULL) { |
5346 |
-+ leaf = bfqg; |
5347 |
-+ prev = leaf; |
5348 |
-+ } else { |
5349 |
-+ bfq_group_set_parent(prev, bfqg); |
5350 |
-+ /* |
5351 |
-+ * Build a list of allocated nodes using the bfqd |
5352 |
-+ * filed, that is still unused and will be initialized |
5353 |
-+ * only after the node will be connected. |
5354 |
-+ */ |
5355 |
-+ prev->bfqd = bfqg; |
5356 |
-+ prev = bfqg; |
5357 |
-+ } |
5358 |
-+ } |
5359 |
-+ |
5360 |
-+ return leaf; |
5361 |
-+ |
5362 |
-+cleanup: |
5363 |
-+ while (leaf != NULL) { |
5364 |
-+ prev = leaf; |
5365 |
-+ leaf = leaf->bfqd; |
5366 |
-+ kfree(prev); |
5367 |
-+ } |
5368 |
-+ |
5369 |
-+ return NULL; |
5370 |
-+} |
5371 |
-+ |
5372 |
-+/** |
5373 |
-+ * bfq_group_chain_link - link an allocated group chain to a cgroup hierarchy. |
5374 |
-+ * @bfqd: the queue descriptor. |
5375 |
-+ * @css: the leaf cgroup_subsys_state to start from. |
5376 |
-+ * @leaf: the leaf group (to be associated to @cgroup). |
5377 |
-+ * |
5378 |
-+ * Try to link a chain of groups to a cgroup hierarchy, connecting the |
5379 |
-+ * nodes bottom-up, so we can be sure that when we find a cgroup in the |
5380 |
-+ * hierarchy that already as a group associated to @bfqd all the nodes |
5381 |
-+ * in the path to the root cgroup have one too. |
5382 |
-+ * |
5383 |
-+ * On locking: the queue lock protects the hierarchy (there is a hierarchy |
5384 |
-+ * per device) while the bfqio_cgroup lock protects the list of groups |
5385 |
-+ * belonging to the same cgroup. |
5386 |
-+ */ |
5387 |
-+static void bfq_group_chain_link(struct bfq_data *bfqd, |
5388 |
-+ struct cgroup_subsys_state *css, |
5389 |
-+ struct bfq_group *leaf) |
5390 |
-+{ |
5391 |
-+ struct bfqio_cgroup *bgrp; |
5392 |
-+ struct bfq_group *bfqg, *next, *prev = NULL; |
5393 |
-+ unsigned long flags; |
5394 |
-+ |
5395 |
-+ assert_spin_locked(bfqd->queue->queue_lock); |
5396 |
-+ |
5397 |
-+ for (; css != NULL && leaf != NULL; css = css->parent) { |
5398 |
-+ bgrp = css_to_bfqio(css); |
5399 |
-+ next = leaf->bfqd; |
5400 |
-+ |
5401 |
-+ bfqg = bfqio_lookup_group(bgrp, bfqd); |
5402 |
-+ BUG_ON(bfqg != NULL); |
5403 |
-+ |
5404 |
-+ spin_lock_irqsave(&bgrp->lock, flags); |
5405 |
-+ |
5406 |
-+ rcu_assign_pointer(leaf->bfqd, bfqd); |
5407 |
-+ hlist_add_head_rcu(&leaf->group_node, &bgrp->group_data); |
5408 |
-+ hlist_add_head(&leaf->bfqd_node, &bfqd->group_list); |
5409 |
-+ |
5410 |
-+ spin_unlock_irqrestore(&bgrp->lock, flags); |
5411 |
-+ |
5412 |
-+ prev = leaf; |
5413 |
-+ leaf = next; |
5414 |
-+ } |
5415 |
-+ |
5416 |
-+ BUG_ON(css == NULL && leaf != NULL); |
5417 |
-+ if (css != NULL && prev != NULL) { |
5418 |
-+ bgrp = css_to_bfqio(css); |
5419 |
-+ bfqg = bfqio_lookup_group(bgrp, bfqd); |
5420 |
-+ bfq_group_set_parent(prev, bfqg); |
5421 |
-+ } |
5422 |
-+} |
5423 |
-+ |
5424 |
-+/** |
5425 |
-+ * bfq_find_alloc_group - return the group associated to @bfqd in @cgroup. |
5426 |
-+ * @bfqd: queue descriptor. |
5427 |
-+ * @cgroup: cgroup being searched for. |
5428 |
-+ * |
5429 |
-+ * Return a group associated to @bfqd in @cgroup, allocating one if |
5430 |
-+ * necessary. When a group is returned all the cgroups in the path |
5431 |
-+ * to the root have a group associated to @bfqd. |
5432 |
-+ * |
5433 |
-+ * If the allocation fails, return the root group: this breaks guarantees |
5434 |
-+ * but is a safe fallback. If this loss becomes a problem it can be |
5435 |
-+ * mitigated using the equivalent weight (given by the product of the |
5436 |
-+ * weights of the groups in the path from @group to the root) in the |
5437 |
-+ * root scheduler. |
5438 |
-+ * |
5439 |
-+ * We allocate all the missing nodes in the path from the leaf cgroup |
5440 |
-+ * to the root and we connect the nodes only after all the allocations |
5441 |
-+ * have been successful. |
5442 |
-+ */ |
5443 |
-+static struct bfq_group *bfq_find_alloc_group(struct bfq_data *bfqd, |
5444 |
-+ struct cgroup_subsys_state *css) |
5445 |
-+{ |
5446 |
-+ struct bfqio_cgroup *bgrp = css_to_bfqio(css); |
5447 |
-+ struct bfq_group *bfqg; |
5448 |
-+ |
5449 |
-+ bfqg = bfqio_lookup_group(bgrp, bfqd); |
5450 |
-+ if (bfqg != NULL) |
5451 |
-+ return bfqg; |
5452 |
-+ |
5453 |
-+ bfqg = bfq_group_chain_alloc(bfqd, css); |
5454 |
-+ if (bfqg != NULL) |
5455 |
-+ bfq_group_chain_link(bfqd, css, bfqg); |
5456 |
-+ else |
5457 |
-+ bfqg = bfqd->root_group; |
5458 |
-+ |
5459 |
-+ return bfqg; |
5460 |
-+} |
5461 |
-+ |
5462 |
-+/** |
5463 |
-+ * bfq_bfqq_move - migrate @bfqq to @bfqg. |
5464 |
-+ * @bfqd: queue descriptor. |
5465 |
-+ * @bfqq: the queue to move. |
5466 |
-+ * @entity: @bfqq's entity. |
5467 |
-+ * @bfqg: the group to move to. |
5468 |
-+ * |
5469 |
-+ * Move @bfqq to @bfqg, deactivating it from its old group and reactivating |
5470 |
-+ * it on the new one. Avoid putting the entity on the old group idle tree. |
5471 |
-+ * |
5472 |
-+ * Must be called under the queue lock; the cgroup owning @bfqg must |
5473 |
-+ * not disappear (by now this just means that we are called under |
5474 |
-+ * rcu_read_lock()). |
5475 |
-+ */ |
5476 |
-+static void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
5477 |
-+ struct bfq_entity *entity, struct bfq_group *bfqg) |
5478 |
-+{ |
5479 |
-+ int busy, resume; |
5480 |
-+ |
5481 |
-+ busy = bfq_bfqq_busy(bfqq); |
5482 |
-+ resume = !RB_EMPTY_ROOT(&bfqq->sort_list); |
5483 |
-+ |
5484 |
-+ BUG_ON(resume && !entity->on_st); |
5485 |
-+ BUG_ON(busy && !resume && entity->on_st && |
5486 |
-+ bfqq != bfqd->in_service_queue); |
5487 |
-+ |
5488 |
-+ if (busy) { |
5489 |
-+ BUG_ON(atomic_read(&bfqq->ref) < 2); |
5490 |
-+ |
5491 |
-+ if (!resume) |
5492 |
-+ bfq_del_bfqq_busy(bfqd, bfqq, 0); |
5493 |
-+ else |
5494 |
-+ bfq_deactivate_bfqq(bfqd, bfqq, 0); |
5495 |
-+ } else if (entity->on_st) |
5496 |
-+ bfq_put_idle_entity(bfq_entity_service_tree(entity), entity); |
5497 |
-+ |
5498 |
-+ /* |
5499 |
-+ * Here we use a reference to bfqg. We don't need a refcounter |
5500 |
-+ * as the cgroup reference will not be dropped, so that its |
5501 |
-+ * destroy() callback will not be invoked. |
5502 |
-+ */ |
5503 |
-+ entity->parent = bfqg->my_entity; |
5504 |
-+ entity->sched_data = &bfqg->sched_data; |
5505 |
-+ |
5506 |
-+ if (busy && resume) |
5507 |
-+ bfq_activate_bfqq(bfqd, bfqq); |
5508 |
-+ |
5509 |
-+ if (bfqd->in_service_queue == NULL && !bfqd->rq_in_driver) |
5510 |
-+ bfq_schedule_dispatch(bfqd); |
5511 |
-+} |
5512 |
-+ |
5513 |
-+/** |
5514 |
-+ * __bfq_bic_change_cgroup - move @bic to @cgroup. |
5515 |
-+ * @bfqd: the queue descriptor. |
5516 |
-+ * @bic: the bic to move. |
5517 |
-+ * @cgroup: the cgroup to move to. |
5518 |
-+ * |
5519 |
-+ * Move bic to cgroup, assuming that bfqd->queue is locked; the caller |
5520 |
-+ * has to make sure that the reference to cgroup is valid across the call. |
5521 |
-+ * |
5522 |
-+ * NOTE: an alternative approach might have been to store the current |
5523 |
-+ * cgroup in bfqq and getting a reference to it, reducing the lookup |
5524 |
-+ * time here, at the price of slightly more complex code. |
5525 |
-+ */ |
5526 |
-+static struct bfq_group *__bfq_bic_change_cgroup(struct bfq_data *bfqd, |
5527 |
-+ struct bfq_io_cq *bic, |
5528 |
-+ struct cgroup_subsys_state *css) |
5529 |
-+{ |
5530 |
-+ struct bfq_queue *async_bfqq = bic_to_bfqq(bic, 0); |
5531 |
-+ struct bfq_queue *sync_bfqq = bic_to_bfqq(bic, 1); |
5532 |
-+ struct bfq_entity *entity; |
5533 |
-+ struct bfq_group *bfqg; |
5534 |
-+ struct bfqio_cgroup *bgrp; |
5535 |
-+ |
5536 |
-+ bgrp = css_to_bfqio(css); |
5537 |
-+ |
5538 |
-+ bfqg = bfq_find_alloc_group(bfqd, css); |
5539 |
-+ if (async_bfqq != NULL) { |
5540 |
-+ entity = &async_bfqq->entity; |
5541 |
-+ |
5542 |
-+ if (entity->sched_data != &bfqg->sched_data) { |
5543 |
-+ bic_set_bfqq(bic, NULL, 0); |
5544 |
-+ bfq_log_bfqq(bfqd, async_bfqq, |
5545 |
-+ "bic_change_group: %p %d", |
5546 |
-+ async_bfqq, atomic_read(&async_bfqq->ref)); |
5547 |
-+ bfq_put_queue(async_bfqq); |
5548 |
-+ } |
5549 |
-+ } |
5550 |
-+ |
5551 |
-+ if (sync_bfqq != NULL) { |
5552 |
-+ entity = &sync_bfqq->entity; |
5553 |
-+ if (entity->sched_data != &bfqg->sched_data) |
5554 |
-+ bfq_bfqq_move(bfqd, sync_bfqq, entity, bfqg); |
5555 |
-+ } |
5556 |
-+ |
5557 |
-+ return bfqg; |
5558 |
-+} |
5559 |
-+ |
5560 |
-+/** |
5561 |
-+ * bfq_bic_change_cgroup - move @bic to @cgroup. |
5562 |
-+ * @bic: the bic being migrated. |
5563 |
-+ * @cgroup: the destination cgroup. |
5564 |
-+ * |
5565 |
-+ * When the task owning @bic is moved to @cgroup, @bic is immediately |
5566 |
-+ * moved into its new parent group. |
5567 |
-+ */ |
5568 |
-+static void bfq_bic_change_cgroup(struct bfq_io_cq *bic, |
5569 |
-+ struct cgroup_subsys_state *css) |
5570 |
-+{ |
5571 |
-+ struct bfq_data *bfqd; |
5572 |
-+ unsigned long uninitialized_var(flags); |
5573 |
-+ |
5574 |
-+ bfqd = bfq_get_bfqd_locked(&(bic->icq.q->elevator->elevator_data), |
5575 |
-+ &flags); |
5576 |
-+ if (bfqd != NULL) { |
5577 |
-+ __bfq_bic_change_cgroup(bfqd, bic, css); |
5578 |
-+ bfq_put_bfqd_unlock(bfqd, &flags); |
5579 |
-+ } |
5580 |
-+} |
5581 |
-+ |
5582 |
-+/** |
5583 |
-+ * bfq_bic_update_cgroup - update the cgroup of @bic. |
5584 |
-+ * @bic: the @bic to update. |
5585 |
-+ * |
5586 |
-+ * Make sure that @bic is enqueued in the cgroup of the current task. |
5587 |
-+ * We need this in addition to moving bics during the cgroup attach |
5588 |
-+ * phase because the task owning @bic could be at its first disk |
5589 |
-+ * access or we may end up in the root cgroup as the result of a |
5590 |
-+ * memory allocation failure and here we try to move to the right |
5591 |
-+ * group. |
5592 |
-+ * |
5593 |
-+ * Must be called under the queue lock. It is safe to use the returned |
5594 |
-+ * value even after the rcu_read_unlock() as the migration/destruction |
5595 |
-+ * paths act under the queue lock too. IOW it is impossible to race with |
5596 |
-+ * group migration/destruction and end up with an invalid group as: |
5597 |
-+ * a) here cgroup has not yet been destroyed, nor its destroy callback |
5598 |
-+ * has started execution, as current holds a reference to it, |
5599 |
-+ * b) if it is destroyed after rcu_read_unlock() [after current is |
5600 |
-+ * migrated to a different cgroup] its attach() callback will have |
5601 |
-+ * taken care of remove all the references to the old cgroup data. |
5602 |
-+ */ |
5603 |
-+static struct bfq_group *bfq_bic_update_cgroup(struct bfq_io_cq *bic) |
5604 |
-+{ |
5605 |
-+ struct bfq_data *bfqd = bic_to_bfqd(bic); |
5606 |
-+ struct bfq_group *bfqg; |
5607 |
-+ struct cgroup_subsys_state *css; |
5608 |
-+ |
5609 |
-+ BUG_ON(bfqd == NULL); |
5610 |
-+ |
5611 |
-+ rcu_read_lock(); |
5612 |
-+ css = task_css(current, bfqio_subsys_id); |
5613 |
-+ bfqg = __bfq_bic_change_cgroup(bfqd, bic, css); |
5614 |
-+ rcu_read_unlock(); |
5615 |
-+ |
5616 |
-+ return bfqg; |
5617 |
-+} |
5618 |
-+ |
5619 |
-+/** |
5620 |
-+ * bfq_flush_idle_tree - deactivate any entity on the idle tree of @st. |
5621 |
-+ * @st: the service tree being flushed. |
5622 |
-+ */ |
5623 |
-+static inline void bfq_flush_idle_tree(struct bfq_service_tree *st) |
5624 |
-+{ |
5625 |
-+ struct bfq_entity *entity = st->first_idle; |
5626 |
-+ |
5627 |
-+ for (; entity != NULL; entity = st->first_idle) |
5628 |
-+ __bfq_deactivate_entity(entity, 0); |
5629 |
-+} |
5630 |
-+ |
5631 |
-+/** |
5632 |
-+ * bfq_reparent_leaf_entity - move leaf entity to the root_group. |
5633 |
-+ * @bfqd: the device data structure with the root group. |
5634 |
-+ * @entity: the entity to move. |
5635 |
-+ */ |
5636 |
-+static inline void bfq_reparent_leaf_entity(struct bfq_data *bfqd, |
5637 |
-+ struct bfq_entity *entity) |
5638 |
-+{ |
5639 |
-+ struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); |
5640 |
-+ |
5641 |
-+ BUG_ON(bfqq == NULL); |
5642 |
-+ bfq_bfqq_move(bfqd, bfqq, entity, bfqd->root_group); |
5643 |
-+ return; |
5644 |
-+} |
5645 |
-+ |
5646 |
-+/** |
5647 |
-+ * bfq_reparent_active_entities - move to the root group all active entities. |
5648 |
-+ * @bfqd: the device data structure with the root group. |
5649 |
-+ * @bfqg: the group to move from. |
5650 |
-+ * @st: the service tree with the entities. |
5651 |
-+ * |
5652 |
-+ * Needs queue_lock to be taken and reference to be valid over the call. |
5653 |
-+ */ |
5654 |
-+static inline void bfq_reparent_active_entities(struct bfq_data *bfqd, |
5655 |
-+ struct bfq_group *bfqg, |
5656 |
-+ struct bfq_service_tree *st) |
5657 |
-+{ |
5658 |
-+ struct rb_root *active = &st->active; |
5659 |
-+ struct bfq_entity *entity = NULL; |
5660 |
-+ |
5661 |
-+ if (!RB_EMPTY_ROOT(&st->active)) |
5662 |
-+ entity = bfq_entity_of(rb_first(active)); |
5663 |
-+ |
5664 |
-+ for (; entity != NULL; entity = bfq_entity_of(rb_first(active))) |
5665 |
-+ bfq_reparent_leaf_entity(bfqd, entity); |
5666 |
-+ |
5667 |
-+ if (bfqg->sched_data.in_service_entity != NULL) |
5668 |
-+ bfq_reparent_leaf_entity(bfqd, |
5669 |
-+ bfqg->sched_data.in_service_entity); |
5670 |
-+ |
5671 |
-+ return; |
5672 |
-+} |
5673 |
-+ |
5674 |
-+/** |
5675 |
-+ * bfq_destroy_group - destroy @bfqg. |
5676 |
-+ * @bgrp: the bfqio_cgroup containing @bfqg. |
5677 |
-+ * @bfqg: the group being destroyed. |
5678 |
-+ * |
5679 |
-+ * Destroy @bfqg, making sure that it is not referenced from its parent. |
5680 |
-+ */ |
5681 |
-+static void bfq_destroy_group(struct bfqio_cgroup *bgrp, struct bfq_group *bfqg) |
5682 |
-+{ |
5683 |
-+ struct bfq_data *bfqd; |
5684 |
-+ struct bfq_service_tree *st; |
5685 |
-+ struct bfq_entity *entity = bfqg->my_entity; |
5686 |
-+ unsigned long uninitialized_var(flags); |
5687 |
-+ int i; |
5688 |
-+ |
5689 |
-+ hlist_del(&bfqg->group_node); |
5690 |
-+ |
5691 |
-+ /* |
5692 |
-+ * Empty all service_trees belonging to this group before deactivating |
5693 |
-+ * the group itself. |
5694 |
-+ */ |
5695 |
-+ for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) { |
5696 |
-+ st = bfqg->sched_data.service_tree + i; |
5697 |
-+ |
5698 |
-+ /* |
5699 |
-+ * The idle tree may still contain bfq_queues belonging |
5700 |
-+ * to exited task because they never migrated to a different |
5701 |
-+ * cgroup from the one being destroyed now. No one else |
5702 |
-+ * can access them so it's safe to act without any lock. |
5703 |
-+ */ |
5704 |
-+ bfq_flush_idle_tree(st); |
5705 |
-+ |
5706 |
-+ /* |
5707 |
-+ * It may happen that some queues are still active |
5708 |
-+ * (busy) upon group destruction (if the corresponding |
5709 |
-+ * processes have been forced to terminate). We move |
5710 |
-+ * all the leaf entities corresponding to these queues |
5711 |
-+ * to the root_group. |
5712 |
-+ * Also, it may happen that the group has an entity |
5713 |
-+ * under service, which is disconnected from the active |
5714 |
-+ * tree: it must be moved, too. |
5715 |
-+ * There is no need to put the sync queues, as the |
5716 |
-+ * scheduler has taken no reference. |
5717 |
-+ */ |
5718 |
-+ bfqd = bfq_get_bfqd_locked(&bfqg->bfqd, &flags); |
5719 |
-+ if (bfqd != NULL) { |
5720 |
-+ bfq_reparent_active_entities(bfqd, bfqg, st); |
5721 |
-+ bfq_put_bfqd_unlock(bfqd, &flags); |
5722 |
-+ } |
5723 |
-+ BUG_ON(!RB_EMPTY_ROOT(&st->active)); |
5724 |
-+ BUG_ON(!RB_EMPTY_ROOT(&st->idle)); |
5725 |
-+ } |
5726 |
-+ BUG_ON(bfqg->sched_data.next_in_service != NULL); |
5727 |
-+ BUG_ON(bfqg->sched_data.in_service_entity != NULL); |
5728 |
-+ |
5729 |
-+ /* |
5730 |
-+ * We may race with device destruction, take extra care when |
5731 |
-+ * dereferencing bfqg->bfqd. |
5732 |
-+ */ |
5733 |
-+ bfqd = bfq_get_bfqd_locked(&bfqg->bfqd, &flags); |
5734 |
-+ if (bfqd != NULL) { |
5735 |
-+ hlist_del(&bfqg->bfqd_node); |
5736 |
-+ __bfq_deactivate_entity(entity, 0); |
5737 |
-+ bfq_put_async_queues(bfqd, bfqg); |
5738 |
-+ bfq_put_bfqd_unlock(bfqd, &flags); |
5739 |
-+ } |
5740 |
-+ BUG_ON(entity->tree != NULL); |
5741 |
-+ |
5742 |
-+ /* |
5743 |
-+ * No need to defer the kfree() to the end of the RCU grace |
5744 |
-+ * period: we are called from the destroy() callback of our |
5745 |
-+ * cgroup, so we can be sure that no one is a) still using |
5746 |
-+ * this cgroup or b) doing lookups in it. |
5747 |
-+ */ |
5748 |
-+ kfree(bfqg); |
5749 |
-+} |
5750 |
-+ |
5751 |
-+static void bfq_end_raising_async(struct bfq_data *bfqd) |
5752 |
-+{ |
5753 |
-+ struct hlist_node *tmp; |
5754 |
-+ struct bfq_group *bfqg; |
5755 |
-+ |
5756 |
-+ hlist_for_each_entry_safe(bfqg, tmp, &bfqd->group_list, bfqd_node) |
5757 |
-+ bfq_end_raising_async_queues(bfqd, bfqg); |
5758 |
-+ bfq_end_raising_async_queues(bfqd, bfqd->root_group); |
5759 |
-+} |
5760 |
-+ |
5761 |
-+/** |
5762 |
-+ * bfq_disconnect_groups - disconnect @bfqd from all its groups. |
5763 |
-+ * @bfqd: the device descriptor being exited. |
5764 |
-+ * |
5765 |
-+ * When the device exits we just make sure that no lookup can return |
5766 |
-+ * the now unused group structures. They will be deallocated on cgroup |
5767 |
-+ * destruction. |
5768 |
-+ */ |
5769 |
-+static void bfq_disconnect_groups(struct bfq_data *bfqd) |
5770 |
-+{ |
5771 |
-+ struct hlist_node *tmp; |
5772 |
-+ struct bfq_group *bfqg; |
5773 |
-+ |
5774 |
-+ bfq_log(bfqd, "disconnect_groups beginning"); |
5775 |
-+ hlist_for_each_entry_safe(bfqg, tmp, &bfqd->group_list, bfqd_node) { |
5776 |
-+ hlist_del(&bfqg->bfqd_node); |
5777 |
-+ |
5778 |
-+ __bfq_deactivate_entity(bfqg->my_entity, 0); |
5779 |
-+ |
5780 |
-+ /* |
5781 |
-+ * Don't remove from the group hash, just set an |
5782 |
-+ * invalid key. No lookups can race with the |
5783 |
-+ * assignment as bfqd is being destroyed; this |
5784 |
-+ * implies also that new elements cannot be added |
5785 |
-+ * to the list. |
5786 |
-+ */ |
5787 |
-+ rcu_assign_pointer(bfqg->bfqd, NULL); |
5788 |
-+ |
5789 |
-+ bfq_log(bfqd, "disconnect_groups: put async for group %p", |
5790 |
-+ bfqg); |
5791 |
-+ bfq_put_async_queues(bfqd, bfqg); |
5792 |
-+ } |
5793 |
-+} |
5794 |
-+ |
5795 |
-+static inline void bfq_free_root_group(struct bfq_data *bfqd) |
5796 |
-+{ |
5797 |
-+ struct bfqio_cgroup *bgrp = &bfqio_root_cgroup; |
5798 |
-+ struct bfq_group *bfqg = bfqd->root_group; |
5799 |
-+ |
5800 |
-+ bfq_put_async_queues(bfqd, bfqg); |
5801 |
-+ |
5802 |
-+ spin_lock_irq(&bgrp->lock); |
5803 |
-+ hlist_del_rcu(&bfqg->group_node); |
5804 |
-+ spin_unlock_irq(&bgrp->lock); |
5805 |
-+ |
5806 |
-+ /* |
5807 |
-+ * No need to synchronize_rcu() here: since the device is gone |
5808 |
-+ * there cannot be any read-side access to its root_group. |
5809 |
-+ */ |
5810 |
-+ kfree(bfqg); |
5811 |
-+} |
5812 |
-+ |
5813 |
-+static struct bfq_group *bfq_alloc_root_group(struct bfq_data *bfqd, int node) |
5814 |
-+{ |
5815 |
-+ struct bfq_group *bfqg; |
5816 |
-+ struct bfqio_cgroup *bgrp; |
5817 |
-+ int i; |
5818 |
-+ |
5819 |
-+ bfqg = kzalloc_node(sizeof(*bfqg), GFP_KERNEL, node); |
5820 |
-+ if (bfqg == NULL) |
5821 |
-+ return NULL; |
5822 |
-+ |
5823 |
-+ bfqg->entity.parent = NULL; |
5824 |
-+ for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) |
5825 |
-+ bfqg->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT; |
5826 |
-+ |
5827 |
-+ bgrp = &bfqio_root_cgroup; |
5828 |
-+ spin_lock_irq(&bgrp->lock); |
5829 |
-+ rcu_assign_pointer(bfqg->bfqd, bfqd); |
5830 |
-+ hlist_add_head_rcu(&bfqg->group_node, &bgrp->group_data); |
5831 |
-+ spin_unlock_irq(&bgrp->lock); |
5832 |
-+ |
5833 |
-+ return bfqg; |
5834 |
-+} |
5835 |
-+ |
5836 |
-+#define SHOW_FUNCTION(__VAR) \ |
5837 |
-+static u64 bfqio_cgroup_##__VAR##_read(struct cgroup_subsys_state *css, \ |
5838 |
-+ struct cftype *cftype) \ |
5839 |
-+{ \ |
5840 |
-+ struct bfqio_cgroup *bgrp = css_to_bfqio(css); \ |
5841 |
-+ u64 ret = -ENODEV; \ |
5842 |
-+ \ |
5843 |
-+ mutex_lock(&bfqio_mutex); \ |
5844 |
-+ if (bfqio_is_removed(bgrp)) \ |
5845 |
-+ goto out_unlock; \ |
5846 |
-+ \ |
5847 |
-+ spin_lock_irq(&bgrp->lock); \ |
5848 |
-+ ret = bgrp->__VAR; \ |
5849 |
-+ spin_unlock_irq(&bgrp->lock); \ |
5850 |
-+ \ |
5851 |
-+out_unlock: \ |
5852 |
-+ mutex_unlock(&bfqio_mutex); \ |
5853 |
-+ return ret; \ |
5854 |
-+} |
5855 |
-+ |
5856 |
-+SHOW_FUNCTION(weight); |
5857 |
-+SHOW_FUNCTION(ioprio); |
5858 |
-+SHOW_FUNCTION(ioprio_class); |
5859 |
-+#undef SHOW_FUNCTION |
5860 |
-+ |
5861 |
-+#define STORE_FUNCTION(__VAR, __MIN, __MAX) \ |
5862 |
-+static int bfqio_cgroup_##__VAR##_write(struct cgroup_subsys_state *css,\ |
5863 |
-+ struct cftype *cftype, \ |
5864 |
-+ u64 val) \ |
5865 |
-+{ \ |
5866 |
-+ struct bfqio_cgroup *bgrp = css_to_bfqio(css); \ |
5867 |
-+ struct bfq_group *bfqg; \ |
5868 |
-+ int ret = -EINVAL; \ |
5869 |
-+ \ |
5870 |
-+ if (val < (__MIN) || val > (__MAX)) \ |
5871 |
-+ return ret; \ |
5872 |
-+ \ |
5873 |
-+ ret = -ENODEV; \ |
5874 |
-+ mutex_lock(&bfqio_mutex); \ |
5875 |
-+ if (bfqio_is_removed(bgrp)) \ |
5876 |
-+ goto out_unlock; \ |
5877 |
-+ ret = 0; \ |
5878 |
-+ \ |
5879 |
-+ spin_lock_irq(&bgrp->lock); \ |
5880 |
-+ bgrp->__VAR = (unsigned short)val; \ |
5881 |
-+ hlist_for_each_entry(bfqg, &bgrp->group_data, group_node) { \ |
5882 |
-+ /* \ |
5883 |
-+ * Setting the ioprio_changed flag of the entity \ |
5884 |
-+ * to 1 with new_##__VAR == ##__VAR would re-set \ |
5885 |
-+ * the value of the weight to its ioprio mapping. \ |
5886 |
-+ * Set the flag only if necessary. \ |
5887 |
-+ */ \ |
5888 |
-+ if ((unsigned short)val != bfqg->entity.new_##__VAR) { \ |
5889 |
-+ bfqg->entity.new_##__VAR = (unsigned short)val; \ |
5890 |
-+ /* \ |
5891 |
-+ * Make sure that the above new value has been \ |
5892 |
-+ * stored in bfqg->entity.new_##__VAR before \ |
5893 |
-+ * setting the ioprio_changed flag. In fact, \ |
5894 |
-+ * this flag may be read asynchronously (in \ |
5895 |
-+ * critical sections protected by a different \ |
5896 |
-+ * lock than that held here), and finding this \ |
5897 |
-+ * flag set may cause the execution of the code \ |
5898 |
-+ * for updating parameters whose value may \ |
5899 |
-+ * depend also on bfqg->entity.new_##__VAR (in \ |
5900 |
-+ * __bfq_entity_update_weight_prio). \ |
5901 |
-+ * This barrier makes sure that the new value \ |
5902 |
-+ * of bfqg->entity.new_##__VAR is correctly \ |
5903 |
-+ * seen in that code. \ |
5904 |
-+ */ \ |
5905 |
-+ smp_wmb(); \ |
5906 |
-+ bfqg->entity.ioprio_changed = 1; \ |
5907 |
-+ } \ |
5908 |
-+ } \ |
5909 |
-+ spin_unlock_irq(&bgrp->lock); \ |
5910 |
-+ \ |
5911 |
-+out_unlock: \ |
5912 |
-+ mutex_unlock(&bfqio_mutex); \ |
5913 |
-+ return ret; \ |
5914 |
-+} |
5915 |
-+ |
5916 |
-+STORE_FUNCTION(weight, BFQ_MIN_WEIGHT, BFQ_MAX_WEIGHT); |
5917 |
-+STORE_FUNCTION(ioprio, 0, IOPRIO_BE_NR - 1); |
5918 |
-+STORE_FUNCTION(ioprio_class, IOPRIO_CLASS_RT, IOPRIO_CLASS_IDLE); |
5919 |
-+#undef STORE_FUNCTION |
5920 |
-+ |
5921 |
-+static struct cftype bfqio_files[] = { |
5922 |
-+ { |
5923 |
-+ .name = "weight", |
5924 |
-+ .read_u64 = bfqio_cgroup_weight_read, |
5925 |
-+ .write_u64 = bfqio_cgroup_weight_write, |
5926 |
-+ }, |
5927 |
-+ { |
5928 |
-+ .name = "ioprio", |
5929 |
-+ .read_u64 = bfqio_cgroup_ioprio_read, |
5930 |
-+ .write_u64 = bfqio_cgroup_ioprio_write, |
5931 |
-+ }, |
5932 |
-+ { |
5933 |
-+ .name = "ioprio_class", |
5934 |
-+ .read_u64 = bfqio_cgroup_ioprio_class_read, |
5935 |
-+ .write_u64 = bfqio_cgroup_ioprio_class_write, |
5936 |
-+ }, |
5937 |
-+ { }, /* terminate */ |
5938 |
-+}; |
5939 |
-+ |
5940 |
-+static struct cgroup_subsys_state *bfqio_create(struct cgroup_subsys_state |
5941 |
-+ *parent_css) |
5942 |
-+{ |
5943 |
-+ struct bfqio_cgroup *bgrp; |
5944 |
-+ |
5945 |
-+ if (parent_css != NULL) { |
5946 |
-+ bgrp = kzalloc(sizeof(*bgrp), GFP_KERNEL); |
5947 |
-+ if (bgrp == NULL) |
5948 |
-+ return ERR_PTR(-ENOMEM); |
5949 |
-+ } else |
5950 |
-+ bgrp = &bfqio_root_cgroup; |
5951 |
-+ |
5952 |
-+ spin_lock_init(&bgrp->lock); |
5953 |
-+ INIT_HLIST_HEAD(&bgrp->group_data); |
5954 |
-+ bgrp->ioprio = BFQ_DEFAULT_GRP_IOPRIO; |
5955 |
-+ bgrp->ioprio_class = BFQ_DEFAULT_GRP_CLASS; |
5956 |
-+ |
5957 |
-+ return &bgrp->css; |
5958 |
-+} |
5959 |
-+ |
5960 |
-+/* |
5961 |
-+ * We cannot support shared io contexts, as we have no means to support |
5962 |
-+ * two tasks with the same ioc in two different groups without major rework |
5963 |
-+ * of the main bic/bfqq data structures. By now we allow a task to change |
5964 |
-+ * its cgroup only if it's the only owner of its ioc; the drawback of this |
5965 |
-+ * behavior is that a group containing a task that forked using CLONE_IO |
5966 |
-+ * will not be destroyed until the tasks sharing the ioc die. |
5967 |
-+ */ |
5968 |
-+static int bfqio_can_attach(struct cgroup_subsys_state *css, |
5969 |
-+ struct cgroup_taskset *tset) |
5970 |
-+{ |
5971 |
-+ struct task_struct *task; |
5972 |
-+ struct io_context *ioc; |
5973 |
-+ int ret = 0; |
5974 |
-+ |
5975 |
-+ cgroup_taskset_for_each(task, css, tset) { |
5976 |
-+ /* |
5977 |
-+ * task_lock() is needed to avoid races with |
5978 |
-+ * exit_io_context() |
5979 |
-+ */ |
5980 |
-+ task_lock(task); |
5981 |
-+ ioc = task->io_context; |
5982 |
-+ if (ioc != NULL && atomic_read(&ioc->nr_tasks) > 1) |
5983 |
-+ /* |
5984 |
-+ * ioc == NULL means that the task is either too young |
5985 |
-+ * or exiting: if it has still no ioc the ioc can't be |
5986 |
-+ * shared, if the task is exiting the attach will fail |
5987 |
-+ * anyway, no matter what we return here. |
5988 |
-+ */ |
5989 |
-+ ret = -EINVAL; |
5990 |
-+ task_unlock(task); |
5991 |
-+ if (ret) |
5992 |
-+ break; |
5993 |
-+ } |
5994 |
-+ |
5995 |
-+ return ret; |
5996 |
-+} |
5997 |
-+ |
5998 |
-+static void bfqio_attach(struct cgroup_subsys_state *css, |
5999 |
-+ struct cgroup_taskset *tset) |
6000 |
-+{ |
6001 |
-+ struct task_struct *task; |
6002 |
-+ struct io_context *ioc; |
6003 |
-+ struct io_cq *icq; |
6004 |
-+ |
6005 |
-+ /* |
6006 |
-+ * IMPORTANT NOTE: The move of more than one process at a time to a |
6007 |
-+ * new group has not yet been tested. |
6008 |
-+ */ |
6009 |
-+ cgroup_taskset_for_each(task, css, tset) { |
6010 |
-+ ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE); |
6011 |
-+ if (ioc) { |
6012 |
-+ /* |
6013 |
-+ * Handle cgroup change here. |
6014 |
-+ */ |
6015 |
-+ rcu_read_lock(); |
6016 |
-+ hlist_for_each_entry_rcu(icq, &ioc->icq_list, ioc_node) |
6017 |
-+ if (!strncmp( |
6018 |
-+ icq->q->elevator->type->elevator_name, |
6019 |
-+ "bfq", ELV_NAME_MAX)) |
6020 |
-+ bfq_bic_change_cgroup(icq_to_bic(icq), |
6021 |
-+ css); |
6022 |
-+ rcu_read_unlock(); |
6023 |
-+ put_io_context(ioc); |
6024 |
-+ } |
6025 |
-+ } |
6026 |
-+} |
6027 |
-+ |
6028 |
-+static void bfqio_destroy(struct cgroup_subsys_state *css) |
6029 |
-+{ |
6030 |
-+ struct bfqio_cgroup *bgrp = css_to_bfqio(css); |
6031 |
-+ struct hlist_node *tmp; |
6032 |
-+ struct bfq_group *bfqg; |
6033 |
-+ |
6034 |
-+ /* |
6035 |
-+ * Since we are destroying the cgroup, there are no more tasks |
6036 |
-+ * referencing it, and all the RCU grace periods that may have |
6037 |
-+ * referenced it are ended (as the destruction of the parent |
6038 |
-+ * cgroup is RCU-safe); bgrp->group_data will not be accessed by |
6039 |
-+ * anything else and we don't need any synchronization. |
6040 |
-+ */ |
6041 |
-+ hlist_for_each_entry_safe(bfqg, tmp, &bgrp->group_data, group_node) |
6042 |
-+ bfq_destroy_group(bgrp, bfqg); |
6043 |
-+ |
6044 |
-+ BUG_ON(!hlist_empty(&bgrp->group_data)); |
6045 |
-+ |
6046 |
-+ kfree(bgrp); |
6047 |
-+} |
6048 |
-+ |
6049 |
-+static int bfqio_css_online(struct cgroup_subsys_state *css) |
6050 |
-+{ |
6051 |
-+ struct bfqio_cgroup *bgrp = css_to_bfqio(css); |
6052 |
-+ |
6053 |
-+ mutex_lock(&bfqio_mutex); |
6054 |
-+ bgrp->online = true; |
6055 |
-+ mutex_unlock(&bfqio_mutex); |
6056 |
-+ |
6057 |
-+ return 0; |
6058 |
-+} |
6059 |
-+ |
6060 |
-+static void bfqio_css_offline(struct cgroup_subsys_state *css) |
6061 |
-+{ |
6062 |
-+ struct bfqio_cgroup *bgrp = css_to_bfqio(css); |
6063 |
-+ |
6064 |
-+ mutex_lock(&bfqio_mutex); |
6065 |
-+ bgrp->online = false; |
6066 |
-+ mutex_unlock(&bfqio_mutex); |
6067 |
-+} |
6068 |
-+ |
6069 |
-+struct cgroup_subsys bfqio_subsys = { |
6070 |
-+ .name = "bfqio", |
6071 |
-+ .css_alloc = bfqio_create, |
6072 |
-+ .css_online = bfqio_css_online, |
6073 |
-+ .css_offline = bfqio_css_offline, |
6074 |
-+ .can_attach = bfqio_can_attach, |
6075 |
-+ .attach = bfqio_attach, |
6076 |
-+ .css_free = bfqio_destroy, |
6077 |
-+ .subsys_id = bfqio_subsys_id, |
6078 |
-+ .base_cftypes = bfqio_files, |
6079 |
-+}; |
6080 |
-+#else |
6081 |
-+static inline void bfq_init_entity(struct bfq_entity *entity, |
6082 |
-+ struct bfq_group *bfqg) |
6083 |
-+{ |
6084 |
-+ entity->weight = entity->new_weight; |
6085 |
-+ entity->orig_weight = entity->new_weight; |
6086 |
-+ entity->ioprio = entity->new_ioprio; |
6087 |
-+ entity->ioprio_class = entity->new_ioprio_class; |
6088 |
-+ entity->sched_data = &bfqg->sched_data; |
6089 |
-+} |
6090 |
-+ |
6091 |
-+static inline struct bfq_group * |
6092 |
-+bfq_bic_update_cgroup(struct bfq_io_cq *bic) |
6093 |
-+{ |
6094 |
-+ struct bfq_data *bfqd = bic_to_bfqd(bic); |
6095 |
-+ return bfqd->root_group; |
6096 |
-+} |
6097 |
-+ |
6098 |
-+static inline void bfq_bfqq_move(struct bfq_data *bfqd, |
6099 |
-+ struct bfq_queue *bfqq, |
6100 |
-+ struct bfq_entity *entity, |
6101 |
-+ struct bfq_group *bfqg) |
6102 |
-+{ |
6103 |
-+} |
6104 |
-+ |
6105 |
-+static void bfq_end_raising_async(struct bfq_data *bfqd) |
6106 |
-+{ |
6107 |
-+ bfq_end_raising_async_queues(bfqd, bfqd->root_group); |
6108 |
-+} |
6109 |
-+ |
6110 |
-+static inline void bfq_disconnect_groups(struct bfq_data *bfqd) |
6111 |
-+{ |
6112 |
-+ bfq_put_async_queues(bfqd, bfqd->root_group); |
6113 |
-+} |
6114 |
-+ |
6115 |
-+static inline void bfq_free_root_group(struct bfq_data *bfqd) |
6116 |
-+{ |
6117 |
-+ kfree(bfqd->root_group); |
6118 |
-+} |
6119 |
-+ |
6120 |
-+static struct bfq_group *bfq_alloc_root_group(struct bfq_data *bfqd, int node) |
6121 |
-+{ |
6122 |
-+ struct bfq_group *bfqg; |
6123 |
-+ int i; |
6124 |
-+ |
6125 |
-+ bfqg = kmalloc_node(sizeof(*bfqg), GFP_KERNEL | __GFP_ZERO, node); |
6126 |
-+ if (bfqg == NULL) |
6127 |
-+ return NULL; |
6128 |
-+ |
6129 |
-+ for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) |
6130 |
-+ bfqg->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT; |
6131 |
-+ |
6132 |
-+ return bfqg; |
6133 |
-+} |
6134 |
-+#endif |
6135 |
-diff --git a/block/bfq-ioc.c b/block/bfq-ioc.c |
6136 |
-new file mode 100644 |
6137 |
-index 0000000..7f6b000 |
6138 |
---- /dev/null |
6139 |
-+++ b/block/bfq-ioc.c |
6140 |
-@@ -0,0 +1,36 @@ |
6141 |
-+/* |
6142 |
-+ * BFQ: I/O context handling. |
6143 |
-+ * |
6144 |
-+ * Based on ideas and code from CFQ: |
6145 |
-+ * Copyright (C) 2003 Jens Axboe <axboe@××××××.dk> |
6146 |
-+ * |
6147 |
-+ * Copyright (C) 2008 Fabio Checconi <fabio@×××××××××××××.it> |
6148 |
-+ * Paolo Valente <paolo.valente@×××××××.it> |
6149 |
-+ * |
6150 |
-+ * Copyright (C) 2010 Paolo Valente <paolo.valente@×××××××.it> |
6151 |
-+ */ |
6152 |
-+ |
6153 |
-+/** |
6154 |
-+ * icq_to_bic - convert iocontext queue structure to bfq_io_cq. |
6155 |
-+ * @icq: the iocontext queue. |
6156 |
-+ */ |
6157 |
-+static inline struct bfq_io_cq *icq_to_bic(struct io_cq *icq) |
6158 |
-+{ |
6159 |
-+ /* bic->icq is the first member, %NULL will convert to %NULL */ |
6160 |
-+ return container_of(icq, struct bfq_io_cq, icq); |
6161 |
-+} |
6162 |
-+ |
6163 |
-+/** |
6164 |
-+ * bfq_bic_lookup - search into @ioc a bic associated to @bfqd. |
6165 |
-+ * @bfqd: the lookup key. |
6166 |
-+ * @ioc: the io_context of the process doing I/O. |
6167 |
-+ * |
6168 |
-+ * Queue lock must be held. |
6169 |
-+ */ |
6170 |
-+static inline struct bfq_io_cq *bfq_bic_lookup(struct bfq_data *bfqd, |
6171 |
-+ struct io_context *ioc) |
6172 |
-+{ |
6173 |
-+ if (ioc) |
6174 |
-+ return icq_to_bic(ioc_lookup_icq(ioc, bfqd->queue)); |
6175 |
-+ return NULL; |
6176 |
-+} |
6177 |
-diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c |
6178 |
-new file mode 100644 |
6179 |
-index 0000000..f5f71e4 |
6180 |
---- /dev/null |
6181 |
-+++ b/block/bfq-iosched.c |
6182 |
-@@ -0,0 +1,3300 @@ |
6183 |
-+/* |
6184 |
-+ * Budget Fair Queueing (BFQ) disk scheduler. |
6185 |
-+ * |
6186 |
-+ * Based on ideas and code from CFQ: |
6187 |
-+ * Copyright (C) 2003 Jens Axboe <axboe@××××××.dk> |
6188 |
-+ * |
6189 |
-+ * Copyright (C) 2008 Fabio Checconi <fabio@×××××××××××××.it> |
6190 |
-+ * Paolo Valente <paolo.valente@×××××××.it> |
6191 |
-+ * |
6192 |
-+ * Copyright (C) 2010 Paolo Valente <paolo.valente@×××××××.it> |
6193 |
-+ * |
6194 |
-+ * Licensed under the GPL-2 as detailed in the accompanying COPYING.BFQ file. |
6195 |
-+ * |
6196 |
-+ * BFQ is a proportional share disk scheduling algorithm based on the |
6197 |
-+ * slice-by-slice service scheme of CFQ. But BFQ assigns budgets, measured in |
6198 |
-+ * number of sectors, to tasks instead of time slices. The disk is not granted |
6199 |
-+ * to the in-service task for a given time slice, but until it has exhausted |
6200 |
-+ * its assigned budget. This change from the time to the service domain allows |
6201 |
-+ * BFQ to distribute the disk bandwidth among tasks as desired, without any |
6202 |
-+ * distortion due to ZBR, workload fluctuations or other factors. BFQ uses an |
6203 |
-+ * ad hoc internal scheduler, called B-WF2Q+, to schedule tasks according to |
6204 |
-+ * their budgets (more precisely BFQ schedules queues associated to tasks). |
6205 |
-+ * Thanks to this accurate scheduler, BFQ can afford to assign high budgets to |
6206 |
-+ * disk-bound non-seeky tasks (to boost the throughput), and yet guarantee low |
6207 |
-+ * latencies to interactive and soft real-time applications. |
6208 |
-+ * |
6209 |
-+ * BFQ is described in [1], where also a reference to the initial, more |
6210 |
-+ * theoretical paper on BFQ can be found. The interested reader can find in |
6211 |
-+ * the latter paper full details on the main algorithm as well as formulas of |
6212 |
-+ * the guarantees, plus formal proofs of all the properties. With respect to |
6213 |
-+ * the version of BFQ presented in these papers, this implementation adds a |
6214 |
-+ * few more heuristics, such as the one that guarantees a low latency to soft |
6215 |
-+ * real-time applications, and a hierarchical extension based on H-WF2Q+. |
6216 |
-+ * |
6217 |
-+ * B-WF2Q+ is based on WF2Q+, that is described in [2], together with |
6218 |
-+ * H-WF2Q+, while the augmented tree used to implement B-WF2Q+ with O(log N) |
6219 |
-+ * complexity derives from the one introduced with EEVDF in [3]. |
6220 |
-+ * |
6221 |
-+ * [1] P. Valente and M. Andreolini, ``Improving Application Responsiveness |
6222 |
-+ * with the BFQ Disk I/O Scheduler'', |
6223 |
-+ * Proceedings of the 5th Annual International Systems and Storage |
6224 |
-+ * Conference (SYSTOR '12), June 2012. |
6225 |
-+ * |
6226 |
-+ * http://algogroup.unimo.it/people/paolo/disk_sched/bf1-v1-suite-results.pdf |
6227 |
-+ * |
6228 |
-+ * [2] Jon C.R. Bennett and H. Zhang, ``Hierarchical Packet Fair Queueing |
6229 |
-+ * Algorithms,'' IEEE/ACM Transactions on Networking, 5(5):675-689, |
6230 |
-+ * Oct 1997. |
6231 |
-+ * |
6232 |
-+ * http://www.cs.cmu.edu/~hzhang/papers/TON-97-Oct.ps.gz |
6233 |
-+ * |
6234 |
-+ * [3] I. Stoica and H. Abdel-Wahab, ``Earliest Eligible Virtual Deadline |
6235 |
-+ * First: A Flexible and Accurate Mechanism for Proportional Share |
6236 |
-+ * Resource Allocation,'' technical report. |
6237 |
-+ * |
6238 |
-+ * http://www.cs.berkeley.edu/~istoica/papers/eevdf-tr-95.pdf |
6239 |
-+ */ |
6240 |
-+#include <linux/module.h> |
6241 |
-+#include <linux/slab.h> |
6242 |
-+#include <linux/blkdev.h> |
6243 |
-+#include <linux/cgroup.h> |
6244 |
-+#include <linux/elevator.h> |
6245 |
-+#include <linux/jiffies.h> |
6246 |
-+#include <linux/rbtree.h> |
6247 |
-+#include <linux/ioprio.h> |
6248 |
-+#include "bfq.h" |
6249 |
-+#include "blk.h" |
6250 |
-+ |
6251 |
-+/* Max number of dispatches in one round of service. */ |
6252 |
-+static const int bfq_quantum = 4; |
6253 |
-+ |
6254 |
-+/* Expiration time of sync (0) and async (1) requests, in jiffies. */ |
6255 |
-+static const int bfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
6256 |
-+ |
6257 |
-+/* Maximum backwards seek, in KiB. */ |
6258 |
-+static const int bfq_back_max = 16 * 1024; |
6259 |
-+ |
6260 |
-+/* Penalty of a backwards seek, in number of sectors. */ |
6261 |
-+static const int bfq_back_penalty = 2; |
6262 |
-+ |
6263 |
-+/* Idling period duration, in jiffies. */ |
6264 |
-+static int bfq_slice_idle = HZ / 125; |
6265 |
-+ |
6266 |
-+/* Default maximum budget values, in sectors and number of requests. */ |
6267 |
-+static const int bfq_default_max_budget = 16 * 1024; |
6268 |
-+static const int bfq_max_budget_async_rq = 4; |
6269 |
-+ |
6270 |
-+/* |
6271 |
-+ * Async to sync throughput distribution is controlled as follows: |
6272 |
-+ * when an async request is served, the entity is charged the number |
6273 |
-+ * of sectors of the request, multiplied by the factor below |
6274 |
-+ */ |
6275 |
-+static const int bfq_async_charge_factor = 10; |
6276 |
-+ |
6277 |
-+/* Default timeout values, in jiffies, approximating CFQ defaults. */ |
6278 |
-+static const int bfq_timeout_sync = HZ / 8; |
6279 |
-+static int bfq_timeout_async = HZ / 25; |
6280 |
-+ |
6281 |
-+struct kmem_cache *bfq_pool; |
6282 |
-+ |
6283 |
-+/* Below this threshold (in ms), we consider thinktime immediate. */ |
6284 |
-+#define BFQ_MIN_TT 2 |
6285 |
-+ |
6286 |
-+/* hw_tag detection: parallel requests threshold and min samples needed. */ |
6287 |
-+#define BFQ_HW_QUEUE_THRESHOLD 4 |
6288 |
-+#define BFQ_HW_QUEUE_SAMPLES 32 |
6289 |
-+ |
6290 |
-+#define BFQQ_SEEK_THR (sector_t)(8 * 1024) |
6291 |
-+#define BFQQ_SEEKY(bfqq) ((bfqq)->seek_mean > BFQQ_SEEK_THR) |
6292 |
-+ |
6293 |
-+/* Min samples used for peak rate estimation (for autotuning). */ |
6294 |
-+#define BFQ_PEAK_RATE_SAMPLES 32 |
6295 |
-+ |
6296 |
-+/* Shift used for peak rate fixed precision calculations. */ |
6297 |
-+#define BFQ_RATE_SHIFT 16 |
6298 |
-+ |
6299 |
-+/* |
6300 |
-+ * The duration of the weight raising for interactive applications is |
6301 |
-+ * computed automatically (as default behaviour), using the following |
6302 |
-+ * formula: duration = (R / r) * T, where r is the peak rate of the |
6303 |
-+ * disk, and R and T are two reference parameters. In particular, R is |
6304 |
-+ * the peak rate of a reference disk, and T is about the maximum time |
6305 |
-+ * for starting popular large applications on that disk, under BFQ and |
6306 |
-+ * while reading two files in parallel. Finally, BFQ uses two |
6307 |
-+ * different pairs (R, T) depending on whether the disk is rotational |
6308 |
-+ * or non-rotational. |
6309 |
-+ */ |
6310 |
-+#define T_rot (msecs_to_jiffies(5500)) |
6311 |
-+#define T_nonrot (msecs_to_jiffies(2000)) |
6312 |
-+/* Next two quantities are in sectors/usec, left-shifted by BFQ_RATE_SHIFT */ |
6313 |
-+#define R_rot 17415 |
6314 |
-+#define R_nonrot 34791 |
6315 |
-+ |
6316 |
-+#define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \ |
6317 |
-+ { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 }) |
6318 |
-+ |
6319 |
-+#define RQ_BIC(rq) ((struct bfq_io_cq *) (rq)->elv.priv[0]) |
6320 |
-+#define RQ_BFQQ(rq) ((rq)->elv.priv[1]) |
6321 |
-+ |
6322 |
-+static inline void bfq_schedule_dispatch(struct bfq_data *bfqd); |
6323 |
-+ |
6324 |
-+#include "bfq-ioc.c" |
6325 |
-+#include "bfq-sched.c" |
6326 |
-+#include "bfq-cgroup.c" |
6327 |
-+ |
6328 |
-+#define bfq_class_idle(bfqq) ((bfqq)->entity.ioprio_class ==\ |
6329 |
-+ IOPRIO_CLASS_IDLE) |
6330 |
-+#define bfq_class_rt(bfqq) ((bfqq)->entity.ioprio_class ==\ |
6331 |
-+ IOPRIO_CLASS_RT) |
6332 |
-+ |
6333 |
-+#define bfq_sample_valid(samples) ((samples) > 80) |
6334 |
-+ |
6335 |
-+/* |
6336 |
-+ * We regard a request as SYNC, if either it's a read or has the SYNC bit |
6337 |
-+ * set (in which case it could also be a direct WRITE). |
6338 |
-+ */ |
6339 |
-+static inline int bfq_bio_sync(struct bio *bio) |
6340 |
-+{ |
6341 |
-+ if (bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC)) |
6342 |
-+ return 1; |
6343 |
-+ |
6344 |
-+ return 0; |
6345 |
-+} |
6346 |
-+ |
6347 |
-+/* |
6348 |
-+ * Scheduler run of queue, if there are requests pending and no one in the |
6349 |
-+ * driver that will restart queueing. |
6350 |
-+ */ |
6351 |
-+static inline void bfq_schedule_dispatch(struct bfq_data *bfqd) |
6352 |
-+{ |
6353 |
-+ if (bfqd->queued != 0) { |
6354 |
-+ bfq_log(bfqd, "schedule dispatch"); |
6355 |
-+ kblockd_schedule_work(bfqd->queue, &bfqd->unplug_work); |
6356 |
-+ } |
6357 |
-+} |
6358 |
-+ |
6359 |
-+/* |
6360 |
-+ * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
6361 |
-+ * We choose the request that is closesr to the head right now. Distance |
6362 |
-+ * behind the head is penalized and only allowed to a certain extent. |
6363 |
-+ */ |
6364 |
-+static struct request *bfq_choose_req(struct bfq_data *bfqd, |
6365 |
-+ struct request *rq1, |
6366 |
-+ struct request *rq2, |
6367 |
-+ sector_t last) |
6368 |
-+{ |
6369 |
-+ sector_t s1, s2, d1 = 0, d2 = 0; |
6370 |
-+ unsigned long back_max; |
6371 |
-+#define BFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
6372 |
-+#define BFQ_RQ2_WRAP 0x02 /* request 2 wraps */ |
6373 |
-+ unsigned wrap = 0; /* bit mask: requests behind the disk head? */ |
6374 |
-+ |
6375 |
-+ if (rq1 == NULL || rq1 == rq2) |
6376 |
-+ return rq2; |
6377 |
-+ if (rq2 == NULL) |
6378 |
-+ return rq1; |
6379 |
-+ |
6380 |
-+ if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
6381 |
-+ return rq1; |
6382 |
-+ else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) |
6383 |
-+ return rq2; |
6384 |
-+ if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META)) |
6385 |
-+ return rq1; |
6386 |
-+ else if ((rq2->cmd_flags & REQ_META) && !(rq1->cmd_flags & REQ_META)) |
6387 |
-+ return rq2; |
6388 |
-+ |
6389 |
-+ s1 = blk_rq_pos(rq1); |
6390 |
-+ s2 = blk_rq_pos(rq2); |
6391 |
-+ |
6392 |
-+ /* |
6393 |
-+ * By definition, 1KiB is 2 sectors. |
6394 |
-+ */ |
6395 |
-+ back_max = bfqd->bfq_back_max * 2; |
6396 |
-+ |
6397 |
-+ /* |
6398 |
-+ * Strict one way elevator _except_ in the case where we allow |
6399 |
-+ * short backward seeks which are biased as twice the cost of a |
6400 |
-+ * similar forward seek. |
6401 |
-+ */ |
6402 |
-+ if (s1 >= last) |
6403 |
-+ d1 = s1 - last; |
6404 |
-+ else if (s1 + back_max >= last) |
6405 |
-+ d1 = (last - s1) * bfqd->bfq_back_penalty; |
6406 |
-+ else |
6407 |
-+ wrap |= BFQ_RQ1_WRAP; |
6408 |
-+ |
6409 |
-+ if (s2 >= last) |
6410 |
-+ d2 = s2 - last; |
6411 |
-+ else if (s2 + back_max >= last) |
6412 |
-+ d2 = (last - s2) * bfqd->bfq_back_penalty; |
6413 |
-+ else |
6414 |
-+ wrap |= BFQ_RQ2_WRAP; |
6415 |
-+ |
6416 |
-+ /* Found required data */ |
6417 |
-+ |
6418 |
-+ /* |
6419 |
-+ * By doing switch() on the bit mask "wrap" we avoid having to |
6420 |
-+ * check two variables for all permutations: --> faster! |
6421 |
-+ */ |
6422 |
-+ switch (wrap) { |
6423 |
-+ case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
6424 |
-+ if (d1 < d2) |
6425 |
-+ return rq1; |
6426 |
-+ else if (d2 < d1) |
6427 |
-+ return rq2; |
6428 |
-+ else { |
6429 |
-+ if (s1 >= s2) |
6430 |
-+ return rq1; |
6431 |
-+ else |
6432 |
-+ return rq2; |
6433 |
-+ } |
6434 |
-+ |
6435 |
-+ case BFQ_RQ2_WRAP: |
6436 |
-+ return rq1; |
6437 |
-+ case BFQ_RQ1_WRAP: |
6438 |
-+ return rq2; |
6439 |
-+ case (BFQ_RQ1_WRAP|BFQ_RQ2_WRAP): /* both rqs wrapped */ |
6440 |
-+ default: |
6441 |
-+ /* |
6442 |
-+ * Since both rqs are wrapped, |
6443 |
-+ * start with the one that's further behind head |
6444 |
-+ * (--> only *one* back seek required), |
6445 |
-+ * since back seek takes more time than forward. |
6446 |
-+ */ |
6447 |
-+ if (s1 <= s2) |
6448 |
-+ return rq1; |
6449 |
-+ else |
6450 |
-+ return rq2; |
6451 |
-+ } |
6452 |
-+} |
6453 |
-+ |
6454 |
-+static struct bfq_queue * |
6455 |
-+bfq_rq_pos_tree_lookup(struct bfq_data *bfqd, struct rb_root *root, |
6456 |
-+ sector_t sector, struct rb_node **ret_parent, |
6457 |
-+ struct rb_node ***rb_link) |
6458 |
-+{ |
6459 |
-+ struct rb_node **p, *parent; |
6460 |
-+ struct bfq_queue *bfqq = NULL; |
6461 |
-+ |
6462 |
-+ parent = NULL; |
6463 |
-+ p = &root->rb_node; |
6464 |
-+ while (*p) { |
6465 |
-+ struct rb_node **n; |
6466 |
-+ |
6467 |
-+ parent = *p; |
6468 |
-+ bfqq = rb_entry(parent, struct bfq_queue, pos_node); |
6469 |
-+ |
6470 |
-+ /* |
6471 |
-+ * Sort strictly based on sector. Smallest to the left, |
6472 |
-+ * largest to the right. |
6473 |
-+ */ |
6474 |
-+ if (sector > blk_rq_pos(bfqq->next_rq)) |
6475 |
-+ n = &(*p)->rb_right; |
6476 |
-+ else if (sector < blk_rq_pos(bfqq->next_rq)) |
6477 |
-+ n = &(*p)->rb_left; |
6478 |
-+ else |
6479 |
-+ break; |
6480 |
-+ p = n; |
6481 |
-+ bfqq = NULL; |
6482 |
-+ } |
6483 |
-+ |
6484 |
-+ *ret_parent = parent; |
6485 |
-+ if (rb_link) |
6486 |
-+ *rb_link = p; |
6487 |
-+ |
6488 |
-+ bfq_log(bfqd, "rq_pos_tree_lookup %llu: returning %d", |
6489 |
-+ (long long unsigned)sector, |
6490 |
-+ bfqq != NULL ? bfqq->pid : 0); |
6491 |
-+ |
6492 |
-+ return bfqq; |
6493 |
-+} |
6494 |
-+ |
6495 |
-+static void bfq_rq_pos_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq) |
6496 |
-+{ |
6497 |
-+ struct rb_node **p, *parent; |
6498 |
-+ struct bfq_queue *__bfqq; |
6499 |
-+ |
6500 |
-+ if (bfqq->pos_root != NULL) { |
6501 |
-+ rb_erase(&bfqq->pos_node, bfqq->pos_root); |
6502 |
-+ bfqq->pos_root = NULL; |
6503 |
-+ } |
6504 |
-+ |
6505 |
-+ if (bfq_class_idle(bfqq)) |
6506 |
-+ return; |
6507 |
-+ if (!bfqq->next_rq) |
6508 |
-+ return; |
6509 |
-+ |
6510 |
-+ bfqq->pos_root = &bfqd->rq_pos_tree; |
6511 |
-+ __bfqq = bfq_rq_pos_tree_lookup(bfqd, bfqq->pos_root, |
6512 |
-+ blk_rq_pos(bfqq->next_rq), &parent, &p); |
6513 |
-+ if (__bfqq == NULL) { |
6514 |
-+ rb_link_node(&bfqq->pos_node, parent, p); |
6515 |
-+ rb_insert_color(&bfqq->pos_node, bfqq->pos_root); |
6516 |
-+ } else |
6517 |
-+ bfqq->pos_root = NULL; |
6518 |
-+} |
6519 |
-+ |
6520 |
-+static struct request *bfq_find_next_rq(struct bfq_data *bfqd, |
6521 |
-+ struct bfq_queue *bfqq, |
6522 |
-+ struct request *last) |
6523 |
-+{ |
6524 |
-+ struct rb_node *rbnext = rb_next(&last->rb_node); |
6525 |
-+ struct rb_node *rbprev = rb_prev(&last->rb_node); |
6526 |
-+ struct request *next = NULL, *prev = NULL; |
6527 |
-+ |
6528 |
-+ BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
6529 |
-+ |
6530 |
-+ if (rbprev != NULL) |
6531 |
-+ prev = rb_entry_rq(rbprev); |
6532 |
-+ |
6533 |
-+ if (rbnext != NULL) |
6534 |
-+ next = rb_entry_rq(rbnext); |
6535 |
-+ else { |
6536 |
-+ rbnext = rb_first(&bfqq->sort_list); |
6537 |
-+ if (rbnext && rbnext != &last->rb_node) |
6538 |
-+ next = rb_entry_rq(rbnext); |
6539 |
-+ } |
6540 |
-+ |
6541 |
-+ return bfq_choose_req(bfqd, next, prev, blk_rq_pos(last)); |
6542 |
-+} |
6543 |
-+ |
6544 |
-+static void bfq_del_rq_rb(struct request *rq) |
6545 |
-+{ |
6546 |
-+ struct bfq_queue *bfqq = RQ_BFQQ(rq); |
6547 |
-+ struct bfq_data *bfqd = bfqq->bfqd; |
6548 |
-+ const int sync = rq_is_sync(rq); |
6549 |
-+ |
6550 |
-+ BUG_ON(bfqq->queued[sync] == 0); |
6551 |
-+ bfqq->queued[sync]--; |
6552 |
-+ bfqd->queued--; |
6553 |
-+ |
6554 |
-+ elv_rb_del(&bfqq->sort_list, rq); |
6555 |
-+ |
6556 |
-+ if (RB_EMPTY_ROOT(&bfqq->sort_list)) { |
6557 |
-+ if (bfq_bfqq_busy(bfqq) && bfqq != bfqd->in_service_queue) |
6558 |
-+ bfq_del_bfqq_busy(bfqd, bfqq, 1); |
6559 |
-+ /* |
6560 |
-+ * Remove queue from request-position tree as it is empty. |
6561 |
-+ */ |
6562 |
-+ if (bfqq->pos_root != NULL) { |
6563 |
-+ rb_erase(&bfqq->pos_node, bfqq->pos_root); |
6564 |
-+ bfqq->pos_root = NULL; |
6565 |
-+ } |
6566 |
-+ } |
6567 |
-+} |
6568 |
-+ |
6569 |
-+/* see the definition of bfq_async_charge_factor for details */ |
6570 |
-+static inline unsigned long bfq_serv_to_charge(struct request *rq, |
6571 |
-+ struct bfq_queue *bfqq) |
6572 |
-+{ |
6573 |
-+ return blk_rq_sectors(rq) * |
6574 |
-+ (1 + ((!bfq_bfqq_sync(bfqq)) * (bfqq->raising_coeff == 1) * |
6575 |
-+ bfq_async_charge_factor)); |
6576 |
-+} |
6577 |
-+ |
6578 |
-+/** |
6579 |
-+ * bfq_updated_next_req - update the queue after a new next_rq selection. |
6580 |
-+ * @bfqd: the device data the queue belongs to. |
6581 |
-+ * @bfqq: the queue to update. |
6582 |
-+ * |
6583 |
-+ * If the first request of a queue changes we make sure that the queue |
6584 |
-+ * has enough budget to serve at least its first request (if the |
6585 |
-+ * request has grown). We do this because if the queue has not enough |
6586 |
-+ * budget for its first request, it has to go through two dispatch |
6587 |
-+ * rounds to actually get it dispatched. |
6588 |
-+ */ |
6589 |
-+static void bfq_updated_next_req(struct bfq_data *bfqd, |
6590 |
-+ struct bfq_queue *bfqq) |
6591 |
-+{ |
6592 |
-+ struct bfq_entity *entity = &bfqq->entity; |
6593 |
-+ struct bfq_service_tree *st = bfq_entity_service_tree(entity); |
6594 |
-+ struct request *next_rq = bfqq->next_rq; |
6595 |
-+ unsigned long new_budget; |
6596 |
-+ |
6597 |
-+ if (next_rq == NULL) |
6598 |
-+ return; |
6599 |
-+ |
6600 |
-+ if (bfqq == bfqd->in_service_queue) |
6601 |
-+ /* |
6602 |
-+ * In order not to break guarantees, budgets cannot be |
6603 |
-+ * changed after an entity has been selected. |
6604 |
-+ */ |
6605 |
-+ return; |
6606 |
-+ |
6607 |
-+ BUG_ON(entity->tree != &st->active); |
6608 |
-+ BUG_ON(entity == entity->sched_data->in_service_entity); |
6609 |
-+ |
6610 |
-+ new_budget = max_t(unsigned long, bfqq->max_budget, |
6611 |
-+ bfq_serv_to_charge(next_rq, bfqq)); |
6612 |
-+ entity->budget = new_budget; |
6613 |
-+ bfq_log_bfqq(bfqd, bfqq, "updated next rq: new budget %lu", new_budget); |
6614 |
-+ bfq_activate_bfqq(bfqd, bfqq); |
6615 |
-+} |
6616 |
-+ |
6617 |
-+static inline unsigned int bfq_wrais_duration(struct bfq_data *bfqd) |
6618 |
-+{ |
6619 |
-+ u64 dur; |
6620 |
-+ |
6621 |
-+ if (bfqd->bfq_raising_max_time > 0) |
6622 |
-+ return bfqd->bfq_raising_max_time; |
6623 |
-+ |
6624 |
-+ dur = bfqd->RT_prod; |
6625 |
-+ do_div(dur, bfqd->peak_rate); |
6626 |
-+ |
6627 |
-+ return dur; |
6628 |
-+} |
6629 |
-+ |
6630 |
-+static void bfq_add_rq_rb(struct request *rq) |
6631 |
-+{ |
6632 |
-+ struct bfq_queue *bfqq = RQ_BFQQ(rq); |
6633 |
-+ struct bfq_entity *entity = &bfqq->entity; |
6634 |
-+ struct bfq_data *bfqd = bfqq->bfqd; |
6635 |
-+ struct request *next_rq, *prev; |
6636 |
-+ unsigned long old_raising_coeff = bfqq->raising_coeff; |
6637 |
-+ int idle_for_long_time = 0; |
6638 |
-+ |
6639 |
-+ bfq_log_bfqq(bfqd, bfqq, "add_rq_rb %d", rq_is_sync(rq)); |
6640 |
-+ bfqq->queued[rq_is_sync(rq)]++; |
6641 |
-+ bfqd->queued++; |
6642 |
-+ |
6643 |
-+ elv_rb_add(&bfqq->sort_list, rq); |
6644 |
-+ |
6645 |
-+ /* |
6646 |
-+ * Check if this request is a better next-serve candidate. |
6647 |
-+ */ |
6648 |
-+ prev = bfqq->next_rq; |
6649 |
-+ next_rq = bfq_choose_req(bfqd, bfqq->next_rq, rq, bfqd->last_position); |
6650 |
-+ BUG_ON(next_rq == NULL); |
6651 |
-+ bfqq->next_rq = next_rq; |
6652 |
-+ |
6653 |
-+ /* |
6654 |
-+ * Adjust priority tree position, if next_rq changes. |
6655 |
-+ */ |
6656 |
-+ if (prev != bfqq->next_rq) |
6657 |
-+ bfq_rq_pos_tree_add(bfqd, bfqq); |
6658 |
-+ |
6659 |
-+ if (!bfq_bfqq_busy(bfqq)) { |
6660 |
-+ int soft_rt = bfqd->bfq_raising_max_softrt_rate > 0 && |
6661 |
-+ time_is_before_jiffies(bfqq->soft_rt_next_start); |
6662 |
-+ idle_for_long_time = time_is_before_jiffies( |
6663 |
-+ bfqq->budget_timeout + |
6664 |
-+ bfqd->bfq_raising_min_idle_time); |
6665 |
-+ entity->budget = max_t(unsigned long, bfqq->max_budget, |
6666 |
-+ bfq_serv_to_charge(next_rq, bfqq)); |
6667 |
-+ |
6668 |
-+ if (!bfqd->low_latency) |
6669 |
-+ goto add_bfqq_busy; |
6670 |
-+ |
6671 |
-+ /* |
6672 |
-+ * If the queue is not being boosted and has been idle |
6673 |
-+ * for enough time, start a weight-raising period |
6674 |
-+ */ |
6675 |
-+ if (old_raising_coeff == 1 && |
6676 |
-+ (idle_for_long_time || soft_rt)) { |
6677 |
-+ bfqq->raising_coeff = bfqd->bfq_raising_coeff; |
6678 |
-+ if (idle_for_long_time) |
6679 |
-+ bfqq->raising_cur_max_time = |
6680 |
-+ bfq_wrais_duration(bfqd); |
6681 |
-+ else |
6682 |
-+ bfqq->raising_cur_max_time = |
6683 |
-+ bfqd->bfq_raising_rt_max_time; |
6684 |
-+ bfq_log_bfqq(bfqd, bfqq, |
6685 |
-+ "wrais starting at %lu, " |
6686 |
-+ "rais_max_time %u", |
6687 |
-+ jiffies, |
6688 |
-+ jiffies_to_msecs(bfqq-> |
6689 |
-+ raising_cur_max_time)); |
6690 |
-+ } else if (old_raising_coeff > 1) { |
6691 |
-+ if (idle_for_long_time) |
6692 |
-+ bfqq->raising_cur_max_time = |
6693 |
-+ bfq_wrais_duration(bfqd); |
6694 |
-+ else if (bfqq->raising_cur_max_time == |
6695 |
-+ bfqd->bfq_raising_rt_max_time && |
6696 |
-+ !soft_rt) { |
6697 |
-+ bfqq->raising_coeff = 1; |
6698 |
-+ bfq_log_bfqq(bfqd, bfqq, |
6699 |
-+ "wrais ending at %lu, " |
6700 |
-+ "rais_max_time %u", |
6701 |
-+ jiffies, |
6702 |
-+ jiffies_to_msecs(bfqq-> |
6703 |
-+ raising_cur_max_time)); |
6704 |
-+ } else if (time_before( |
6705 |
-+ bfqq->last_rais_start_finish + |
6706 |
-+ bfqq->raising_cur_max_time, |
6707 |
-+ jiffies + |
6708 |
-+ bfqd->bfq_raising_rt_max_time) && |
6709 |
-+ soft_rt) { |
6710 |
-+ /* |
6711 |
-+ * |
6712 |
-+ * The remaining weight-raising time is lower |
6713 |
-+ * than bfqd->bfq_raising_rt_max_time, which |
6714 |
-+ * means that the application is enjoying |
6715 |
-+ * weight raising either because deemed soft- |
6716 |
-+ * rt in the near past, or because deemed |
6717 |
-+ * interactive a long ago. In both cases, |
6718 |
-+ * resetting now the current remaining weight- |
6719 |
-+ * raising time for the application to the |
6720 |
-+ * weight-raising duration for soft rt |
6721 |
-+ * applications would not cause any latency |
6722 |
-+ * increase for the application (as the new |
6723 |
-+ * duration would be higher than the remaining |
6724 |
-+ * time). |
6725 |
-+ * |
6726 |
-+ * In addition, the application is now meeting |
6727 |
-+ * the requirements for being deemed soft rt. |
6728 |
-+ * In the end we can correctly and safely |
6729 |
-+ * (re)charge the weight-raising duration for |
6730 |
-+ * the application with the weight-raising |
6731 |
-+ * duration for soft rt applications. |
6732 |
-+ * |
6733 |
-+ * In particular, doing this recharge now, i.e., |
6734 |
-+ * before the weight-raising period for the |
6735 |
-+ * application finishes, reduces the probability |
6736 |
-+ * of the following negative scenario: |
6737 |
-+ * 1) the weight of a soft rt application is |
6738 |
-+ * raised at startup (as for any newly |
6739 |
-+ * created application), |
6740 |
-+ * 2) since the application is not interactive, |
6741 |
-+ * at a certain time weight-raising is |
6742 |
-+ * stopped for the application, |
6743 |
-+ * 3) at that time the application happens to |
6744 |
-+ * still have pending requests, and hence |
6745 |
-+ * is destined to not have a chance to be |
6746 |
-+ * deemed soft rt before these requests are |
6747 |
-+ * completed (see the comments to the |
6748 |
-+ * function bfq_bfqq_softrt_next_start() |
6749 |
-+ * for details on soft rt detection), |
6750 |
-+ * 4) these pending requests experience a high |
6751 |
-+ * latency because the application is not |
6752 |
-+ * weight-raised while they are pending. |
6753 |
-+ */ |
6754 |
-+ bfqq->last_rais_start_finish = jiffies; |
6755 |
-+ bfqq->raising_cur_max_time = |
6756 |
-+ bfqd->bfq_raising_rt_max_time; |
6757 |
-+ } |
6758 |
-+ } |
6759 |
-+ if (old_raising_coeff != bfqq->raising_coeff) |
6760 |
-+ entity->ioprio_changed = 1; |
6761 |
-+add_bfqq_busy: |
6762 |
-+ bfqq->last_idle_bklogged = jiffies; |
6763 |
-+ bfqq->service_from_backlogged = 0; |
6764 |
-+ bfq_clear_bfqq_softrt_update(bfqq); |
6765 |
-+ bfq_add_bfqq_busy(bfqd, bfqq); |
6766 |
-+ } else { |
6767 |
-+ if (bfqd->low_latency && old_raising_coeff == 1 && |
6768 |
-+ !rq_is_sync(rq) && |
6769 |
-+ time_is_before_jiffies( |
6770 |
-+ bfqq->last_rais_start_finish + |
6771 |
-+ bfqd->bfq_raising_min_inter_arr_async)) { |
6772 |
-+ bfqq->raising_coeff = bfqd->bfq_raising_coeff; |
6773 |
-+ bfqq->raising_cur_max_time = bfq_wrais_duration(bfqd); |
6774 |
-+ |
6775 |
-+ bfqd->raised_busy_queues++; |
6776 |
-+ entity->ioprio_changed = 1; |
6777 |
-+ bfq_log_bfqq(bfqd, bfqq, |
6778 |
-+ "non-idle wrais starting at %lu, " |
6779 |
-+ "rais_max_time %u", |
6780 |
-+ jiffies, |
6781 |
-+ jiffies_to_msecs(bfqq-> |
6782 |
-+ raising_cur_max_time)); |
6783 |
-+ } |
6784 |
-+ bfq_updated_next_req(bfqd, bfqq); |
6785 |
-+ } |
6786 |
-+ |
6787 |
-+ if (bfqd->low_latency && |
6788 |
-+ (old_raising_coeff == 1 || bfqq->raising_coeff == 1 || |
6789 |
-+ idle_for_long_time)) |
6790 |
-+ bfqq->last_rais_start_finish = jiffies; |
6791 |
-+} |
6792 |
-+ |
6793 |
-+static void bfq_reposition_rq_rb(struct bfq_queue *bfqq, struct request *rq) |
6794 |
-+{ |
6795 |
-+ elv_rb_del(&bfqq->sort_list, rq); |
6796 |
-+ bfqq->queued[rq_is_sync(rq)]--; |
6797 |
-+ bfqq->bfqd->queued--; |
6798 |
-+ bfq_add_rq_rb(rq); |
6799 |
-+} |
6800 |
-+ |
6801 |
-+static struct request *bfq_find_rq_fmerge(struct bfq_data *bfqd, |
6802 |
-+ struct bio *bio) |
6803 |
-+{ |
6804 |
-+ struct task_struct *tsk = current; |
6805 |
-+ struct bfq_io_cq *bic; |
6806 |
-+ struct bfq_queue *bfqq; |
6807 |
-+ |
6808 |
-+ bic = bfq_bic_lookup(bfqd, tsk->io_context); |
6809 |
-+ if (bic == NULL) |
6810 |
-+ return NULL; |
6811 |
-+ |
6812 |
-+ bfqq = bic_to_bfqq(bic, bfq_bio_sync(bio)); |
6813 |
-+ if (bfqq != NULL) |
6814 |
-+ return elv_rb_find(&bfqq->sort_list, bio_end_sector(bio)); |
6815 |
-+ |
6816 |
-+ return NULL; |
6817 |
-+} |
6818 |
-+ |
6819 |
-+static void bfq_activate_request(struct request_queue *q, struct request *rq) |
6820 |
-+{ |
6821 |
-+ struct bfq_data *bfqd = q->elevator->elevator_data; |
6822 |
-+ |
6823 |
-+ bfqd->rq_in_driver++; |
6824 |
-+ bfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
6825 |
-+ bfq_log(bfqd, "activate_request: new bfqd->last_position %llu", |
6826 |
-+ (long long unsigned)bfqd->last_position); |
6827 |
-+} |
6828 |
-+ |
6829 |
-+static void bfq_deactivate_request(struct request_queue *q, struct request *rq) |
6830 |
-+{ |
6831 |
-+ struct bfq_data *bfqd = q->elevator->elevator_data; |
6832 |
-+ |
6833 |
-+ WARN_ON(bfqd->rq_in_driver == 0); |
6834 |
-+ bfqd->rq_in_driver--; |
6835 |
-+} |
6836 |
-+ |
6837 |
-+static void bfq_remove_request(struct request *rq) |
6838 |
-+{ |
6839 |
-+ struct bfq_queue *bfqq = RQ_BFQQ(rq); |
6840 |
-+ struct bfq_data *bfqd = bfqq->bfqd; |
6841 |
-+ |
6842 |
-+ if (bfqq->next_rq == rq) { |
6843 |
-+ bfqq->next_rq = bfq_find_next_rq(bfqd, bfqq, rq); |
6844 |
-+ bfq_updated_next_req(bfqd, bfqq); |
6845 |
-+ } |
6846 |
-+ |
6847 |
-+ list_del_init(&rq->queuelist); |
6848 |
-+ bfq_del_rq_rb(rq); |
6849 |
-+ |
6850 |
-+ if (rq->cmd_flags & REQ_META) { |
6851 |
-+ WARN_ON(bfqq->meta_pending == 0); |
6852 |
-+ bfqq->meta_pending--; |
6853 |
-+ } |
6854 |
-+} |
6855 |
-+ |
6856 |
-+static int bfq_merge(struct request_queue *q, struct request **req, |
6857 |
-+ struct bio *bio) |
6858 |
-+{ |
6859 |
-+ struct bfq_data *bfqd = q->elevator->elevator_data; |
6860 |
-+ struct request *__rq; |
6861 |
-+ |
6862 |
-+ __rq = bfq_find_rq_fmerge(bfqd, bio); |
6863 |
-+ if (__rq != NULL && elv_rq_merge_ok(__rq, bio)) { |
6864 |
-+ *req = __rq; |
6865 |
-+ return ELEVATOR_FRONT_MERGE; |
6866 |
-+ } |
6867 |
-+ |
6868 |
-+ return ELEVATOR_NO_MERGE; |
6869 |
-+} |
6870 |
-+ |
6871 |
-+static void bfq_merged_request(struct request_queue *q, struct request *req, |
6872 |
-+ int type) |
6873 |
-+{ |
6874 |
-+ if (type == ELEVATOR_FRONT_MERGE) { |
6875 |
-+ struct bfq_queue *bfqq = RQ_BFQQ(req); |
6876 |
-+ |
6877 |
-+ bfq_reposition_rq_rb(bfqq, req); |
6878 |
-+ } |
6879 |
-+} |
6880 |
-+ |
6881 |
-+static void bfq_merged_requests(struct request_queue *q, struct request *rq, |
6882 |
-+ struct request *next) |
6883 |
-+{ |
6884 |
-+ struct bfq_queue *bfqq = RQ_BFQQ(rq); |
6885 |
-+ |
6886 |
-+ /* |
6887 |
-+ * Reposition in fifo if next is older than rq. |
6888 |
-+ */ |
6889 |
-+ if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && |
6890 |
-+ time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
6891 |
-+ list_move(&rq->queuelist, &next->queuelist); |
6892 |
-+ rq_set_fifo_time(rq, rq_fifo_time(next)); |
6893 |
-+ } |
6894 |
-+ |
6895 |
-+ if (bfqq->next_rq == next) |
6896 |
-+ bfqq->next_rq = rq; |
6897 |
-+ |
6898 |
-+ bfq_remove_request(next); |
6899 |
-+} |
6900 |
-+ |
6901 |
-+/* Must be called with bfqq != NULL */ |
6902 |
-+static inline void bfq_bfqq_end_raising(struct bfq_queue *bfqq) |
6903 |
-+{ |
6904 |
-+ BUG_ON(bfqq == NULL); |
6905 |
-+ if (bfq_bfqq_busy(bfqq)) |
6906 |
-+ bfqq->bfqd->raised_busy_queues--; |
6907 |
-+ bfqq->raising_coeff = 1; |
6908 |
-+ bfqq->raising_cur_max_time = 0; |
6909 |
-+ /* Trigger a weight change on the next activation of the queue */ |
6910 |
-+ bfqq->entity.ioprio_changed = 1; |
6911 |
-+} |
6912 |
-+ |
6913 |
-+static void bfq_end_raising_async_queues(struct bfq_data *bfqd, |
6914 |
-+ struct bfq_group *bfqg) |
6915 |
-+{ |
6916 |
-+ int i, j; |
6917 |
-+ |
6918 |
-+ for (i = 0; i < 2; i++) |
6919 |
-+ for (j = 0; j < IOPRIO_BE_NR; j++) |
6920 |
-+ if (bfqg->async_bfqq[i][j] != NULL) |
6921 |
-+ bfq_bfqq_end_raising(bfqg->async_bfqq[i][j]); |
6922 |
-+ if (bfqg->async_idle_bfqq != NULL) |
6923 |
-+ bfq_bfqq_end_raising(bfqg->async_idle_bfqq); |
6924 |
-+} |
6925 |
-+ |
6926 |
-+static void bfq_end_raising(struct bfq_data *bfqd) |
6927 |
-+{ |
6928 |
-+ struct bfq_queue *bfqq; |
6929 |
-+ |
6930 |
-+ spin_lock_irq(bfqd->queue->queue_lock); |
6931 |
-+ |
6932 |
-+ list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list) |
6933 |
-+ bfq_bfqq_end_raising(bfqq); |
6934 |
-+ list_for_each_entry(bfqq, &bfqd->idle_list, bfqq_list) |
6935 |
-+ bfq_bfqq_end_raising(bfqq); |
6936 |
-+ bfq_end_raising_async(bfqd); |
6937 |
-+ |
6938 |
-+ spin_unlock_irq(bfqd->queue->queue_lock); |
6939 |
-+} |
6940 |
-+ |
6941 |
-+static int bfq_allow_merge(struct request_queue *q, struct request *rq, |
6942 |
-+ struct bio *bio) |
6943 |
-+{ |
6944 |
-+ struct bfq_data *bfqd = q->elevator->elevator_data; |
6945 |
-+ struct bfq_io_cq *bic; |
6946 |
-+ struct bfq_queue *bfqq; |
6947 |
-+ |
6948 |
-+ /* |
6949 |
-+ * Disallow merge of a sync bio into an async request. |
6950 |
-+ */ |
6951 |
-+ if (bfq_bio_sync(bio) && !rq_is_sync(rq)) |
6952 |
-+ return 0; |
6953 |
-+ |
6954 |
-+ /* |
6955 |
-+ * Lookup the bfqq that this bio will be queued with. Allow |
6956 |
-+ * merge only if rq is queued there. |
6957 |
-+ * Queue lock is held here. |
6958 |
-+ */ |
6959 |
-+ bic = bfq_bic_lookup(bfqd, current->io_context); |
6960 |
-+ if (bic == NULL) |
6961 |
-+ return 0; |
6962 |
-+ |
6963 |
-+ bfqq = bic_to_bfqq(bic, bfq_bio_sync(bio)); |
6964 |
-+ return bfqq == RQ_BFQQ(rq); |
6965 |
-+} |
6966 |
-+ |
6967 |
-+static void __bfq_set_in_service_queue(struct bfq_data *bfqd, |
6968 |
-+ struct bfq_queue *bfqq) |
6969 |
-+{ |
6970 |
-+ if (bfqq != NULL) { |
6971 |
-+ bfq_mark_bfqq_must_alloc(bfqq); |
6972 |
-+ bfq_mark_bfqq_budget_new(bfqq); |
6973 |
-+ bfq_clear_bfqq_fifo_expire(bfqq); |
6974 |
-+ |
6975 |
-+ bfqd->budgets_assigned = (bfqd->budgets_assigned*7 + 256) / 8; |
6976 |
-+ |
6977 |
-+ bfq_log_bfqq(bfqd, bfqq, |
6978 |
-+ "set_in_service_queue, cur-budget = %lu", |
6979 |
-+ bfqq->entity.budget); |
6980 |
-+ } |
6981 |
-+ |
6982 |
-+ bfqd->in_service_queue = bfqq; |
6983 |
-+} |
6984 |
-+ |
6985 |
-+/* |
6986 |
-+ * Get and set a new queue for service. |
6987 |
-+ */ |
6988 |
-+static struct bfq_queue *bfq_set_in_service_queue(struct bfq_data *bfqd, |
6989 |
-+ struct bfq_queue *bfqq) |
6990 |
-+{ |
6991 |
-+ if (!bfqq) |
6992 |
-+ bfqq = bfq_get_next_queue(bfqd); |
6993 |
-+ else |
6994 |
-+ bfq_get_next_queue_forced(bfqd, bfqq); |
6995 |
-+ |
6996 |
-+ __bfq_set_in_service_queue(bfqd, bfqq); |
6997 |
-+ return bfqq; |
6998 |
-+} |
6999 |
-+ |
7000 |
-+static inline sector_t bfq_dist_from_last(struct bfq_data *bfqd, |
7001 |
-+ struct request *rq) |
7002 |
-+{ |
7003 |
-+ if (blk_rq_pos(rq) >= bfqd->last_position) |
7004 |
-+ return blk_rq_pos(rq) - bfqd->last_position; |
7005 |
-+ else |
7006 |
-+ return bfqd->last_position - blk_rq_pos(rq); |
7007 |
-+} |
7008 |
-+ |
7009 |
-+/* |
7010 |
-+ * Return true if bfqq has no request pending and rq is close enough to |
7011 |
-+ * bfqd->last_position, or if rq is closer to bfqd->last_position than |
7012 |
-+ * bfqq->next_rq |
7013 |
-+ */ |
7014 |
-+static inline int bfq_rq_close(struct bfq_data *bfqd, struct request *rq) |
7015 |
-+{ |
7016 |
-+ return bfq_dist_from_last(bfqd, rq) <= BFQQ_SEEK_THR; |
7017 |
-+} |
7018 |
-+ |
7019 |
-+static struct bfq_queue *bfqq_close(struct bfq_data *bfqd) |
7020 |
-+{ |
7021 |
-+ struct rb_root *root = &bfqd->rq_pos_tree; |
7022 |
-+ struct rb_node *parent, *node; |
7023 |
-+ struct bfq_queue *__bfqq; |
7024 |
-+ sector_t sector = bfqd->last_position; |
7025 |
-+ |
7026 |
-+ if (RB_EMPTY_ROOT(root)) |
7027 |
-+ return NULL; |
7028 |
-+ |
7029 |
-+ /* |
7030 |
-+ * First, if we find a request starting at the end of the last |
7031 |
-+ * request, choose it. |
7032 |
-+ */ |
7033 |
-+ __bfqq = bfq_rq_pos_tree_lookup(bfqd, root, sector, &parent, NULL); |
7034 |
-+ if (__bfqq != NULL) |
7035 |
-+ return __bfqq; |
7036 |
-+ |
7037 |
-+ /* |
7038 |
-+ * If the exact sector wasn't found, the parent of the NULL leaf |
7039 |
-+ * will contain the closest sector (rq_pos_tree sorted by next_request |
7040 |
-+ * position). |
7041 |
-+ */ |
7042 |
-+ __bfqq = rb_entry(parent, struct bfq_queue, pos_node); |
7043 |
-+ if (bfq_rq_close(bfqd, __bfqq->next_rq)) |
7044 |
-+ return __bfqq; |
7045 |
-+ |
7046 |
-+ if (blk_rq_pos(__bfqq->next_rq) < sector) |
7047 |
-+ node = rb_next(&__bfqq->pos_node); |
7048 |
-+ else |
7049 |
-+ node = rb_prev(&__bfqq->pos_node); |
7050 |
-+ if (node == NULL) |
7051 |
-+ return NULL; |
7052 |
-+ |
7053 |
-+ __bfqq = rb_entry(node, struct bfq_queue, pos_node); |
7054 |
-+ if (bfq_rq_close(bfqd, __bfqq->next_rq)) |
7055 |
-+ return __bfqq; |
7056 |
-+ |
7057 |
-+ return NULL; |
7058 |
-+} |
7059 |
-+ |
7060 |
-+/* |
7061 |
-+ * bfqd - obvious |
7062 |
-+ * cur_bfqq - passed in so that we don't decide that the current queue |
7063 |
-+ * is closely cooperating with itself. |
7064 |
-+ * |
7065 |
-+ * We are assuming that cur_bfqq has dispatched at least one request, |
7066 |
-+ * and that bfqd->last_position reflects a position on the disk associated |
7067 |
-+ * with the I/O issued by cur_bfqq. |
7068 |
-+ */ |
7069 |
-+static struct bfq_queue *bfq_close_cooperator(struct bfq_data *bfqd, |
7070 |
-+ struct bfq_queue *cur_bfqq) |
7071 |
-+{ |
7072 |
-+ struct bfq_queue *bfqq; |
7073 |
-+ |
7074 |
-+ if (bfq_class_idle(cur_bfqq)) |
7075 |
-+ return NULL; |
7076 |
-+ if (!bfq_bfqq_sync(cur_bfqq)) |
7077 |
-+ return NULL; |
7078 |
-+ if (BFQQ_SEEKY(cur_bfqq)) |
7079 |
-+ return NULL; |
7080 |
-+ |
7081 |
-+ /* If device has only one backlogged bfq_queue, don't search. */ |
7082 |
-+ if (bfqd->busy_queues == 1) |
7083 |
-+ return NULL; |
7084 |
-+ |
7085 |
-+ /* |
7086 |
-+ * We should notice if some of the queues are cooperating, e.g. |
7087 |
-+ * working closely on the same area of the disk. In that case, |
7088 |
-+ * we can group them together and don't waste time idling. |
7089 |
-+ */ |
7090 |
-+ bfqq = bfqq_close(bfqd); |
7091 |
-+ if (bfqq == NULL || bfqq == cur_bfqq) |
7092 |
-+ return NULL; |
7093 |
-+ |
7094 |
-+ /* |
7095 |
-+ * Do not merge queues from different bfq_groups. |
7096 |
-+ */ |
7097 |
-+ if (bfqq->entity.parent != cur_bfqq->entity.parent) |
7098 |
-+ return NULL; |
7099 |
-+ |
7100 |
-+ /* |
7101 |
-+ * It only makes sense to merge sync queues. |
7102 |
-+ */ |
7103 |
-+ if (!bfq_bfqq_sync(bfqq)) |
7104 |
-+ return NULL; |
7105 |
-+ if (BFQQ_SEEKY(bfqq)) |
7106 |
-+ return NULL; |
7107 |
-+ |
7108 |
-+ /* |
7109 |
-+ * Do not merge queues of different priority classes. |
7110 |
-+ */ |
7111 |
-+ if (bfq_class_rt(bfqq) != bfq_class_rt(cur_bfqq)) |
7112 |
-+ return NULL; |
7113 |
-+ |
7114 |
-+ return bfqq; |
7115 |
-+} |
7116 |
-+ |
7117 |
-+/* |
7118 |
-+ * If enough samples have been computed, return the current max budget |
7119 |
-+ * stored in bfqd, which is dynamically updated according to the |
7120 |
-+ * estimated disk peak rate; otherwise return the default max budget |
7121 |
-+ */ |
7122 |
-+static inline unsigned long bfq_max_budget(struct bfq_data *bfqd) |
7123 |
-+{ |
7124 |
-+ if (bfqd->budgets_assigned < 194) |
7125 |
-+ return bfq_default_max_budget; |
7126 |
-+ else |
7127 |
-+ return bfqd->bfq_max_budget; |
7128 |
-+} |
7129 |
-+ |
7130 |
-+/* |
7131 |
-+ * Return min budget, which is a fraction of the current or default |
7132 |
-+ * max budget (trying with 1/32) |
7133 |
-+ */ |
7134 |
-+static inline unsigned long bfq_min_budget(struct bfq_data *bfqd) |
7135 |
-+{ |
7136 |
-+ if (bfqd->budgets_assigned < 194) |
7137 |
-+ return bfq_default_max_budget / 32; |
7138 |
-+ else |
7139 |
-+ return bfqd->bfq_max_budget / 32; |
7140 |
-+} |
7141 |
-+ |
7142 |
-+/* |
7143 |
-+ * Decides whether idling should be done for given device and |
7144 |
-+ * given in-service queue. |
7145 |
-+ */ |
7146 |
-+static inline bool bfq_queue_nonrot_noidle(struct bfq_data *bfqd, |
7147 |
-+ struct bfq_queue *in_service_bfqq) |
7148 |
-+{ |
7149 |
-+ if (in_service_bfqq == NULL) |
7150 |
-+ return false; |
7151 |
-+ /* |
7152 |
-+ * If the device is non-rotational, and hence has no seek penalty, |
7153 |
-+ * disable idling; but do so only if: |
7154 |
-+ * - device does not support queuing, otherwise we still have |
7155 |
-+ * a problem with sync vs async workloads; |
7156 |
-+ * - the queue is not weight-raised, to preserve guarantees. |
7157 |
-+ */ |
7158 |
-+ return blk_queue_nonrot(bfqd->queue) && bfqd->hw_tag && |
7159 |
-+ (in_service_bfqq->raising_coeff == 1); |
7160 |
-+} |
7161 |
-+ |
7162 |
-+static void bfq_arm_slice_timer(struct bfq_data *bfqd) |
7163 |
-+{ |
7164 |
-+ struct bfq_queue *bfqq = bfqd->in_service_queue; |
7165 |
-+ struct bfq_io_cq *bic; |
7166 |
-+ unsigned long sl; |
7167 |
-+ |
7168 |
-+ WARN_ON(!RB_EMPTY_ROOT(&bfqq->sort_list)); |
7169 |
-+ |
7170 |
-+ /* Tasks have exited, don't wait. */ |
7171 |
-+ bic = bfqd->in_service_bic; |
7172 |
-+ if (bic == NULL || atomic_read(&bic->icq.ioc->active_ref) == 0) |
7173 |
-+ return; |
7174 |
-+ |
7175 |
-+ bfq_mark_bfqq_wait_request(bfqq); |
7176 |
-+ |
7177 |
-+ /* |
7178 |
-+ * We don't want to idle for seeks, but we do want to allow |
7179 |
-+ * fair distribution of slice time for a process doing back-to-back |
7180 |
-+ * seeks. So allow a little bit of time for him to submit a new rq. |
7181 |
-+ * |
7182 |
-+ * To prevent processes with (partly) seeky workloads from |
7183 |
-+ * being too ill-treated, grant them a small fraction of the |
7184 |
-+ * assigned budget before reducing the waiting time to |
7185 |
-+ * BFQ_MIN_TT. This happened to help reduce latency. |
7186 |
-+ */ |
7187 |
-+ sl = bfqd->bfq_slice_idle; |
7188 |
-+ if (bfq_sample_valid(bfqq->seek_samples) && BFQQ_SEEKY(bfqq) && |
7189 |
-+ bfqq->entity.service > bfq_max_budget(bfqd) / 8 && |
7190 |
-+ bfqq->raising_coeff == 1) |
7191 |
-+ sl = min(sl, msecs_to_jiffies(BFQ_MIN_TT)); |
7192 |
-+ else if (bfqq->raising_coeff > 1) |
7193 |
-+ sl = sl * 3; |
7194 |
-+ bfqd->last_idling_start = ktime_get(); |
7195 |
-+ mod_timer(&bfqd->idle_slice_timer, jiffies + sl); |
7196 |
-+ bfq_log(bfqd, "arm idle: %u/%u ms", |
7197 |
-+ jiffies_to_msecs(sl), jiffies_to_msecs(bfqd->bfq_slice_idle)); |
7198 |
-+} |
7199 |
-+ |
7200 |
-+/* |
7201 |
-+ * Set the maximum time for the in-service queue to consume its |
7202 |
-+ * budget. This prevents seeky processes from lowering the disk |
7203 |
-+ * throughput (always guaranteed with a time slice scheme as in CFQ). |
7204 |
-+ */ |
7205 |
-+static void bfq_set_budget_timeout(struct bfq_data *bfqd) |
7206 |
-+{ |
7207 |
-+ struct bfq_queue *bfqq = bfqd->in_service_queue; |
7208 |
-+ unsigned int timeout_coeff; |
7209 |
-+ if (bfqq->raising_cur_max_time == bfqd->bfq_raising_rt_max_time) |
7210 |
-+ timeout_coeff = 1; |
7211 |
-+ else |
7212 |
-+ timeout_coeff = bfqq->entity.weight / bfqq->entity.orig_weight; |
7213 |
-+ |
7214 |
-+ bfqd->last_budget_start = ktime_get(); |
7215 |
-+ |
7216 |
-+ bfq_clear_bfqq_budget_new(bfqq); |
7217 |
-+ bfqq->budget_timeout = jiffies + |
7218 |
-+ bfqd->bfq_timeout[bfq_bfqq_sync(bfqq)] * timeout_coeff; |
7219 |
-+ |
7220 |
-+ bfq_log_bfqq(bfqd, bfqq, "set budget_timeout %u", |
7221 |
-+ jiffies_to_msecs(bfqd->bfq_timeout[bfq_bfqq_sync(bfqq)] * |
7222 |
-+ timeout_coeff)); |
7223 |
-+} |
7224 |
-+ |
7225 |
-+/* |
7226 |
-+ * Move request from internal lists to the request queue dispatch list. |
7227 |
-+ */ |
7228 |
-+static void bfq_dispatch_insert(struct request_queue *q, struct request *rq) |
7229 |
-+{ |
7230 |
-+ struct bfq_data *bfqd = q->elevator->elevator_data; |
7231 |
-+ struct bfq_queue *bfqq = RQ_BFQQ(rq); |
7232 |
-+ |
7233 |
-+ bfq_remove_request(rq); |
7234 |
-+ bfqq->dispatched++; |
7235 |
-+ elv_dispatch_sort(q, rq); |
7236 |
-+ |
7237 |
-+ if (bfq_bfqq_sync(bfqq)) |
7238 |
-+ bfqd->sync_flight++; |
7239 |
-+} |
7240 |
-+ |
7241 |
-+/* |
7242 |
-+ * Return expired entry, or NULL to just start from scratch in rbtree. |
7243 |
-+ */ |
7244 |
-+static struct request *bfq_check_fifo(struct bfq_queue *bfqq) |
7245 |
-+{ |
7246 |
-+ struct request *rq = NULL; |
7247 |
-+ |
7248 |
-+ if (bfq_bfqq_fifo_expire(bfqq)) |
7249 |
-+ return NULL; |
7250 |
-+ |
7251 |
-+ bfq_mark_bfqq_fifo_expire(bfqq); |
7252 |
-+ |
7253 |
-+ if (list_empty(&bfqq->fifo)) |
7254 |
-+ return NULL; |
7255 |
-+ |
7256 |
-+ rq = rq_entry_fifo(bfqq->fifo.next); |
7257 |
-+ |
7258 |
-+ if (time_before(jiffies, rq_fifo_time(rq))) |
7259 |
-+ return NULL; |
7260 |
-+ |
7261 |
-+ return rq; |
7262 |
-+} |
7263 |
-+ |
7264 |
-+/* |
7265 |
-+ * Must be called with the queue_lock held. |
7266 |
-+ */ |
7267 |
-+static int bfqq_process_refs(struct bfq_queue *bfqq) |
7268 |
-+{ |
7269 |
-+ int process_refs, io_refs; |
7270 |
-+ |
7271 |
-+ io_refs = bfqq->allocated[READ] + bfqq->allocated[WRITE]; |
7272 |
-+ process_refs = atomic_read(&bfqq->ref) - io_refs - bfqq->entity.on_st; |
7273 |
-+ BUG_ON(process_refs < 0); |
7274 |
-+ return process_refs; |
7275 |
-+} |
7276 |
-+ |
7277 |
-+static void bfq_setup_merge(struct bfq_queue *bfqq, struct bfq_queue *new_bfqq) |
7278 |
-+{ |
7279 |
-+ int process_refs, new_process_refs; |
7280 |
-+ struct bfq_queue *__bfqq; |
7281 |
-+ |
7282 |
-+ /* |
7283 |
-+ * If there are no process references on the new_bfqq, then it is |
7284 |
-+ * unsafe to follow the ->new_bfqq chain as other bfqq's in the chain |
7285 |
-+ * may have dropped their last reference (not just their last process |
7286 |
-+ * reference). |
7287 |
-+ */ |
7288 |
-+ if (!bfqq_process_refs(new_bfqq)) |
7289 |
-+ return; |
7290 |
-+ |
7291 |
-+ /* Avoid a circular list and skip interim queue merges. */ |
7292 |
-+ while ((__bfqq = new_bfqq->new_bfqq)) { |
7293 |
-+ if (__bfqq == bfqq) |
7294 |
-+ return; |
7295 |
-+ new_bfqq = __bfqq; |
7296 |
-+ } |
7297 |
-+ |
7298 |
-+ process_refs = bfqq_process_refs(bfqq); |
7299 |
-+ new_process_refs = bfqq_process_refs(new_bfqq); |
7300 |
-+ /* |
7301 |
-+ * If the process for the bfqq has gone away, there is no |
7302 |
-+ * sense in merging the queues. |
7303 |
-+ */ |
7304 |
-+ if (process_refs == 0 || new_process_refs == 0) |
7305 |
-+ return; |
7306 |
-+ |
7307 |
-+ /* |
7308 |
-+ * Merge in the direction of the lesser amount of work. |
7309 |
-+ */ |
7310 |
-+ if (new_process_refs >= process_refs) { |
7311 |
-+ bfqq->new_bfqq = new_bfqq; |
7312 |
-+ atomic_add(process_refs, &new_bfqq->ref); |
7313 |
-+ } else { |
7314 |
-+ new_bfqq->new_bfqq = bfqq; |
7315 |
-+ atomic_add(new_process_refs, &bfqq->ref); |
7316 |
-+ } |
7317 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, "scheduling merge with queue %d", |
7318 |
-+ new_bfqq->pid); |
7319 |
-+} |
7320 |
-+ |
7321 |
-+static inline unsigned long bfq_bfqq_budget_left(struct bfq_queue *bfqq) |
7322 |
-+{ |
7323 |
-+ struct bfq_entity *entity = &bfqq->entity; |
7324 |
-+ return entity->budget - entity->service; |
7325 |
-+} |
7326 |
-+ |
7327 |
-+static void __bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq) |
7328 |
-+{ |
7329 |
-+ BUG_ON(bfqq != bfqd->in_service_queue); |
7330 |
-+ |
7331 |
-+ __bfq_bfqd_reset_in_service(bfqd); |
7332 |
-+ |
7333 |
-+ /* |
7334 |
-+ * If this bfqq is shared between multiple processes, check |
7335 |
-+ * to make sure that those processes are still issuing I/Os |
7336 |
-+ * within the mean seek distance. If not, it may be time to |
7337 |
-+ * break the queues apart again. |
7338 |
-+ */ |
7339 |
-+ if (bfq_bfqq_coop(bfqq) && BFQQ_SEEKY(bfqq)) |
7340 |
-+ bfq_mark_bfqq_split_coop(bfqq); |
7341 |
-+ |
7342 |
-+ if (RB_EMPTY_ROOT(&bfqq->sort_list)) { |
7343 |
-+ /* |
7344 |
-+ * overloading budget_timeout field to store when |
7345 |
-+ * the queue remains with no backlog, used by |
7346 |
-+ * the weight-raising mechanism |
7347 |
-+ */ |
7348 |
-+ bfqq->budget_timeout = jiffies; |
7349 |
-+ bfq_del_bfqq_busy(bfqd, bfqq, 1); |
7350 |
-+ } else { |
7351 |
-+ bfq_activate_bfqq(bfqd, bfqq); |
7352 |
-+ /* |
7353 |
-+ * Resort priority tree of potential close cooperators. |
7354 |
-+ */ |
7355 |
-+ bfq_rq_pos_tree_add(bfqd, bfqq); |
7356 |
-+ } |
7357 |
-+} |
7358 |
-+ |
7359 |
-+/** |
7360 |
-+ * __bfq_bfqq_recalc_budget - try to adapt the budget to the @bfqq behavior. |
7361 |
-+ * @bfqd: device data. |
7362 |
-+ * @bfqq: queue to update. |
7363 |
-+ * @reason: reason for expiration. |
7364 |
-+ * |
7365 |
-+ * Handle the feedback on @bfqq budget. See the body for detailed |
7366 |
-+ * comments. |
7367 |
-+ */ |
7368 |
-+static void __bfq_bfqq_recalc_budget(struct bfq_data *bfqd, |
7369 |
-+ struct bfq_queue *bfqq, |
7370 |
-+ enum bfqq_expiration reason) |
7371 |
-+{ |
7372 |
-+ struct request *next_rq; |
7373 |
-+ unsigned long budget, min_budget; |
7374 |
-+ |
7375 |
-+ budget = bfqq->max_budget; |
7376 |
-+ min_budget = bfq_min_budget(bfqd); |
7377 |
-+ |
7378 |
-+ BUG_ON(bfqq != bfqd->in_service_queue); |
7379 |
-+ |
7380 |
-+ bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last budg %lu, budg left %lu", |
7381 |
-+ bfqq->entity.budget, bfq_bfqq_budget_left(bfqq)); |
7382 |
-+ bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last max_budg %lu, min budg %lu", |
7383 |
-+ budget, bfq_min_budget(bfqd)); |
7384 |
-+ bfq_log_bfqq(bfqd, bfqq, "recalc_budg: sync %d, seeky %d", |
7385 |
-+ bfq_bfqq_sync(bfqq), BFQQ_SEEKY(bfqd->in_service_queue)); |
7386 |
-+ |
7387 |
-+ if (bfq_bfqq_sync(bfqq)) { |
7388 |
-+ switch (reason) { |
7389 |
-+ /* |
7390 |
-+ * Caveat: in all the following cases we trade latency |
7391 |
-+ * for throughput. |
7392 |
-+ */ |
7393 |
-+ case BFQ_BFQQ_TOO_IDLE: |
7394 |
-+ /* |
7395 |
-+ * This is the only case where we may reduce |
7396 |
-+ * the budget: if there is no request of the |
7397 |
-+ * process still waiting for completion, then |
7398 |
-+ * we assume (tentatively) that the timer has |
7399 |
-+ * expired because the batch of requests of |
7400 |
-+ * the process could have been served with a |
7401 |
-+ * smaller budget. Hence, betting that |
7402 |
-+ * process will behave in the same way when it |
7403 |
-+ * becomes backlogged again, we reduce its |
7404 |
-+ * next budget. As long as we guess right, |
7405 |
-+ * this budget cut reduces the latency |
7406 |
-+ * experienced by the process. |
7407 |
-+ * |
7408 |
-+ * However, if there are still outstanding |
7409 |
-+ * requests, then the process may have not yet |
7410 |
-+ * issued its next request just because it is |
7411 |
-+ * still waiting for the completion of some of |
7412 |
-+ * the still outstanding ones. So in this |
7413 |
-+ * subcase we do not reduce its budget, on the |
7414 |
-+ * contrary we increase it to possibly boost |
7415 |
-+ * the throughput, as discussed in the |
7416 |
-+ * comments to the BUDGET_TIMEOUT case. |
7417 |
-+ */ |
7418 |
-+ if (bfqq->dispatched > 0) /* still outstanding reqs */ |
7419 |
-+ budget = min(budget * 2, bfqd->bfq_max_budget); |
7420 |
-+ else { |
7421 |
-+ if (budget > 5 * min_budget) |
7422 |
-+ budget -= 4 * min_budget; |
7423 |
-+ else |
7424 |
-+ budget = min_budget; |
7425 |
-+ } |
7426 |
-+ break; |
7427 |
-+ case BFQ_BFQQ_BUDGET_TIMEOUT: |
7428 |
-+ /* |
7429 |
-+ * We double the budget here because: 1) it |
7430 |
-+ * gives the chance to boost the throughput if |
7431 |
-+ * this is not a seeky process (which may have |
7432 |
-+ * bumped into this timeout because of, e.g., |
7433 |
-+ * ZBR), 2) together with charge_full_budget |
7434 |
-+ * it helps give seeky processes higher |
7435 |
-+ * timestamps, and hence be served less |
7436 |
-+ * frequently. |
7437 |
-+ */ |
7438 |
-+ budget = min(budget * 2, bfqd->bfq_max_budget); |
7439 |
-+ break; |
7440 |
-+ case BFQ_BFQQ_BUDGET_EXHAUSTED: |
7441 |
-+ /* |
7442 |
-+ * The process still has backlog, and did not |
7443 |
-+ * let either the budget timeout or the disk |
7444 |
-+ * idling timeout expire. Hence it is not |
7445 |
-+ * seeky, has a short thinktime and may be |
7446 |
-+ * happy with a higher budget too. So |
7447 |
-+ * definitely increase the budget of this good |
7448 |
-+ * candidate to boost the disk throughput. |
7449 |
-+ */ |
7450 |
-+ budget = min(budget * 4, bfqd->bfq_max_budget); |
7451 |
-+ break; |
7452 |
-+ case BFQ_BFQQ_NO_MORE_REQUESTS: |
7453 |
-+ /* |
7454 |
-+ * Leave the budget unchanged. |
7455 |
-+ */ |
7456 |
-+ default: |
7457 |
-+ return; |
7458 |
-+ } |
7459 |
-+ } else /* async queue */ |
7460 |
-+ /* async queues get always the maximum possible budget |
7461 |
-+ * (their ability to dispatch is limited by |
7462 |
-+ * @bfqd->bfq_max_budget_async_rq). |
7463 |
-+ */ |
7464 |
-+ budget = bfqd->bfq_max_budget; |
7465 |
-+ |
7466 |
-+ bfqq->max_budget = budget; |
7467 |
-+ |
7468 |
-+ if (bfqd->budgets_assigned >= 194 && bfqd->bfq_user_max_budget == 0 && |
7469 |
-+ bfqq->max_budget > bfqd->bfq_max_budget) |
7470 |
-+ bfqq->max_budget = bfqd->bfq_max_budget; |
7471 |
-+ |
7472 |
-+ /* |
7473 |
-+ * Make sure that we have enough budget for the next request. |
7474 |
-+ * Since the finish time of the bfqq must be kept in sync with |
7475 |
-+ * the budget, be sure to call __bfq_bfqq_expire() after the |
7476 |
-+ * update. |
7477 |
-+ */ |
7478 |
-+ next_rq = bfqq->next_rq; |
7479 |
-+ if (next_rq != NULL) |
7480 |
-+ bfqq->entity.budget = max_t(unsigned long, bfqq->max_budget, |
7481 |
-+ bfq_serv_to_charge(next_rq, bfqq)); |
7482 |
-+ else |
7483 |
-+ bfqq->entity.budget = bfqq->max_budget; |
7484 |
-+ |
7485 |
-+ bfq_log_bfqq(bfqd, bfqq, "head sect: %u, new budget %lu", |
7486 |
-+ next_rq != NULL ? blk_rq_sectors(next_rq) : 0, |
7487 |
-+ bfqq->entity.budget); |
7488 |
-+} |
7489 |
-+ |
7490 |
-+static unsigned long bfq_calc_max_budget(u64 peak_rate, u64 timeout) |
7491 |
-+{ |
7492 |
-+ unsigned long max_budget; |
7493 |
-+ |
7494 |
-+ /* |
7495 |
-+ * The max_budget calculated when autotuning is equal to the |
7496 |
-+ * amount of sectors transfered in timeout_sync at the |
7497 |
-+ * estimated peak rate. |
7498 |
-+ */ |
7499 |
-+ max_budget = (unsigned long)(peak_rate * 1000 * |
7500 |
-+ timeout >> BFQ_RATE_SHIFT); |
7501 |
-+ |
7502 |
-+ return max_budget; |
7503 |
-+} |
7504 |
-+ |
7505 |
-+/* |
7506 |
-+ * In addition to updating the peak rate, checks whether the process |
7507 |
-+ * is "slow", and returns 1 if so. This slow flag is used, in addition |
7508 |
-+ * to the budget timeout, to reduce the amount of service provided to |
7509 |
-+ * seeky processes, and hence reduce their chances to lower the |
7510 |
-+ * throughput. See the code for more details. |
7511 |
-+ */ |
7512 |
-+static int bfq_update_peak_rate(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
7513 |
-+ int compensate, enum bfqq_expiration reason) |
7514 |
-+{ |
7515 |
-+ u64 bw, usecs, expected, timeout; |
7516 |
-+ ktime_t delta; |
7517 |
-+ int update = 0; |
7518 |
-+ |
7519 |
-+ if (!bfq_bfqq_sync(bfqq) || bfq_bfqq_budget_new(bfqq)) |
7520 |
-+ return 0; |
7521 |
-+ |
7522 |
-+ if (compensate) |
7523 |
-+ delta = bfqd->last_idling_start; |
7524 |
-+ else |
7525 |
-+ delta = ktime_get(); |
7526 |
-+ delta = ktime_sub(delta, bfqd->last_budget_start); |
7527 |
-+ usecs = ktime_to_us(delta); |
7528 |
-+ |
7529 |
-+ /* Don't trust short/unrealistic values. */ |
7530 |
-+ if (usecs < 100 || usecs >= LONG_MAX) |
7531 |
-+ return 0; |
7532 |
-+ |
7533 |
-+ /* |
7534 |
-+ * Calculate the bandwidth for the last slice. We use a 64 bit |
7535 |
-+ * value to store the peak rate, in sectors per usec in fixed |
7536 |
-+ * point math. We do so to have enough precision in the estimate |
7537 |
-+ * and to avoid overflows. |
7538 |
-+ */ |
7539 |
-+ bw = (u64)bfqq->entity.service << BFQ_RATE_SHIFT; |
7540 |
-+ do_div(bw, (unsigned long)usecs); |
7541 |
-+ |
7542 |
-+ timeout = jiffies_to_msecs(bfqd->bfq_timeout[BLK_RW_SYNC]); |
7543 |
-+ |
7544 |
-+ /* |
7545 |
-+ * Use only long (> 20ms) intervals to filter out spikes for |
7546 |
-+ * the peak rate estimation. |
7547 |
-+ */ |
7548 |
-+ if (usecs > 20000) { |
7549 |
-+ if (bw > bfqd->peak_rate || |
7550 |
-+ (!BFQQ_SEEKY(bfqq) && |
7551 |
-+ reason == BFQ_BFQQ_BUDGET_TIMEOUT)) { |
7552 |
-+ bfq_log(bfqd, "measured bw =%llu", bw); |
7553 |
-+ /* |
7554 |
-+ * To smooth oscillations use a low-pass filter with |
7555 |
-+ * alpha=7/8, i.e., |
7556 |
-+ * new_rate = (7/8) * old_rate + (1/8) * bw |
7557 |
-+ */ |
7558 |
-+ do_div(bw, 8); |
7559 |
-+ if (bw == 0) |
7560 |
-+ return 0; |
7561 |
-+ bfqd->peak_rate *= 7; |
7562 |
-+ do_div(bfqd->peak_rate, 8); |
7563 |
-+ bfqd->peak_rate += bw; |
7564 |
-+ update = 1; |
7565 |
-+ bfq_log(bfqd, "new peak_rate=%llu", bfqd->peak_rate); |
7566 |
-+ } |
7567 |
-+ |
7568 |
-+ update |= bfqd->peak_rate_samples == BFQ_PEAK_RATE_SAMPLES - 1; |
7569 |
-+ |
7570 |
-+ if (bfqd->peak_rate_samples < BFQ_PEAK_RATE_SAMPLES) |
7571 |
-+ bfqd->peak_rate_samples++; |
7572 |
-+ |
7573 |
-+ if (bfqd->peak_rate_samples == BFQ_PEAK_RATE_SAMPLES && |
7574 |
-+ update && bfqd->bfq_user_max_budget == 0) { |
7575 |
-+ bfqd->bfq_max_budget = |
7576 |
-+ bfq_calc_max_budget(bfqd->peak_rate, timeout); |
7577 |
-+ bfq_log(bfqd, "new max_budget=%lu", |
7578 |
-+ bfqd->bfq_max_budget); |
7579 |
-+ } |
7580 |
-+ } |
7581 |
-+ |
7582 |
-+ /* |
7583 |
-+ * If the process has been served for a too short time |
7584 |
-+ * interval to let its possible sequential accesses prevail on |
7585 |
-+ * the initial seek time needed to move the disk head on the |
7586 |
-+ * first sector it requested, then give the process a chance |
7587 |
-+ * and for the moment return false. |
7588 |
-+ */ |
7589 |
-+ if (bfqq->entity.budget <= bfq_max_budget(bfqd) / 8) |
7590 |
-+ return 0; |
7591 |
-+ |
7592 |
-+ /* |
7593 |
-+ * A process is considered ``slow'' (i.e., seeky, so that we |
7594 |
-+ * cannot treat it fairly in the service domain, as it would |
7595 |
-+ * slow down too much the other processes) if, when a slice |
7596 |
-+ * ends for whatever reason, it has received service at a |
7597 |
-+ * rate that would not be high enough to complete the budget |
7598 |
-+ * before the budget timeout expiration. |
7599 |
-+ */ |
7600 |
-+ expected = bw * 1000 * timeout >> BFQ_RATE_SHIFT; |
7601 |
-+ |
7602 |
-+ /* |
7603 |
-+ * Caveat: processes doing IO in the slower disk zones will |
7604 |
-+ * tend to be slow(er) even if not seeky. And the estimated |
7605 |
-+ * peak rate will actually be an average over the disk |
7606 |
-+ * surface. Hence, to not be too harsh with unlucky processes, |
7607 |
-+ * we keep a budget/3 margin of safety before declaring a |
7608 |
-+ * process slow. |
7609 |
-+ */ |
7610 |
-+ return expected > (4 * bfqq->entity.budget) / 3; |
7611 |
-+} |
7612 |
-+ |
7613 |
-+/* |
7614 |
-+ * To be deemed as soft real-time, an application must meet two requirements. |
7615 |
-+ * First, the application must not require an average bandwidth higher than |
7616 |
-+ * the approximate bandwidth required to playback or record a compressed high- |
7617 |
-+ * definition video. |
7618 |
-+ * The next function is invoked on the completion of the last request of a |
7619 |
-+ * batch, to compute the next-start time instant, soft_rt_next_start, such |
7620 |
-+ * that, if the next request of the application does not arrive before |
7621 |
-+ * soft_rt_next_start, then the above requirement on the bandwidth is met. |
7622 |
-+ * |
7623 |
-+ * The second requirement is that the request pattern of the application is |
7624 |
-+ * isochronous, i.e., that, after issuing a request or a batch of requests, |
7625 |
-+ * the application stops issuing new requests until all its pending requests |
7626 |
-+ * have been completed. After that, the application may issue a new batch, |
7627 |
-+ * and so on. |
7628 |
-+ * For this reason the next function is invoked to compute soft_rt_next_start |
7629 |
-+ * only for applications that meet this requirement, whereas soft_rt_next_start |
7630 |
-+ * is set to infinity for applications that do not. |
7631 |
-+ * |
7632 |
-+ * Unfortunately, even a greedy application may happen to behave in an |
7633 |
-+ * isochronous way if the CPU load is high. In fact, the application may stop |
7634 |
-+ * issuing requests while the CPUs are busy serving other processes, then |
7635 |
-+ * restart, then stop again for a while, and so on. In addition, if the disk |
7636 |
-+ * achieves a low enough throughput with the request pattern issued by the |
7637 |
-+ * application (e.g., because the request pattern is random and/or the device |
7638 |
-+ * is slow), then the application may meet the above bandwidth requirement too. |
7639 |
-+ * To prevent such a greedy application to be deemed as soft real-time, a |
7640 |
-+ * further rule is used in the computation of soft_rt_next_start: |
7641 |
-+ * soft_rt_next_start must be higher than the current time plus the maximum |
7642 |
-+ * time for which the arrival of a request is waited for when a sync queue |
7643 |
-+ * becomes idle, namely bfqd->bfq_slice_idle. |
7644 |
-+ * This filters out greedy applications, as the latter issue instead their next |
7645 |
-+ * request as soon as possible after the last one has been completed (in |
7646 |
-+ * contrast, when a batch of requests is completed, a soft real-time application |
7647 |
-+ * spends some time processing data). |
7648 |
-+ * |
7649 |
-+ * Unfortunately, the last filter may easily generate false positives if only |
7650 |
-+ * bfqd->bfq_slice_idle is used as a reference time interval and one or both |
7651 |
-+ * the following cases occur: |
7652 |
-+ * 1) HZ is so low that the duration of a jiffy is comparable to or higher |
7653 |
-+ * than bfqd->bfq_slice_idle. This happens, e.g., on slow devices with |
7654 |
-+ * HZ=100. |
7655 |
-+ * 2) jiffies, instead of increasing at a constant rate, may stop increasing |
7656 |
-+ * for a while, then suddenly 'jump' by several units to recover the lost |
7657 |
-+ * increments. This seems to happen, e.g., inside virtual machines. |
7658 |
-+ * To address this issue, we do not use as a reference time interval just |
7659 |
-+ * bfqd->bfq_slice_idle, but bfqd->bfq_slice_idle plus a few jiffies. In |
7660 |
-+ * particular we add the minimum number of jiffies for which the filter seems |
7661 |
-+ * to be quite precise also in embedded systems and KVM/QEMU virtual machines. |
7662 |
-+ */ |
7663 |
-+static inline unsigned long bfq_bfqq_softrt_next_start(struct bfq_data *bfqd, |
7664 |
-+ struct bfq_queue *bfqq) |
7665 |
-+{ |
7666 |
-+ return max(bfqq->last_idle_bklogged + |
7667 |
-+ HZ * bfqq->service_from_backlogged / |
7668 |
-+ bfqd->bfq_raising_max_softrt_rate, |
7669 |
-+ jiffies + bfqq->bfqd->bfq_slice_idle + 4); |
7670 |
-+} |
7671 |
-+ |
7672 |
-+/* |
7673 |
-+ * Return the largest-possible time instant such that, for as long as possible, |
7674 |
-+ * the current time will be lower than this time instant according to the macro |
7675 |
-+ * time_is_before_jiffies(). |
7676 |
-+ */ |
7677 |
-+static inline unsigned long bfq_infinity_from_now(unsigned long now) |
7678 |
-+{ |
7679 |
-+ return now + ULONG_MAX / 2; |
7680 |
-+} |
7681 |
-+ |
7682 |
-+/** |
7683 |
-+ * bfq_bfqq_expire - expire a queue. |
7684 |
-+ * @bfqd: device owning the queue. |
7685 |
-+ * @bfqq: the queue to expire. |
7686 |
-+ * @compensate: if true, compensate for the time spent idling. |
7687 |
-+ * @reason: the reason causing the expiration. |
7688 |
-+ * |
7689 |
-+ * |
7690 |
-+ * If the process associated to the queue is slow (i.e., seeky), or in |
7691 |
-+ * case of budget timeout, or, finally, if it is async, we |
7692 |
-+ * artificially charge it an entire budget (independently of the |
7693 |
-+ * actual service it received). As a consequence, the queue will get |
7694 |
-+ * higher timestamps than the correct ones upon reactivation, and |
7695 |
-+ * hence it will be rescheduled as if it had received more service |
7696 |
-+ * than what it actually received. In the end, this class of processes |
7697 |
-+ * will receive less service in proportion to how slowly they consume |
7698 |
-+ * their budgets (and hence how seriously they tend to lower the |
7699 |
-+ * throughput). |
7700 |
-+ * |
7701 |
-+ * In contrast, when a queue expires because it has been idling for |
7702 |
-+ * too much or because it exhausted its budget, we do not touch the |
7703 |
-+ * amount of service it has received. Hence when the queue will be |
7704 |
-+ * reactivated and its timestamps updated, the latter will be in sync |
7705 |
-+ * with the actual service received by the queue until expiration. |
7706 |
-+ * |
7707 |
-+ * Charging a full budget to the first type of queues and the exact |
7708 |
-+ * service to the others has the effect of using the WF2Q+ policy to |
7709 |
-+ * schedule the former on a timeslice basis, without violating the |
7710 |
-+ * service domain guarantees of the latter. |
7711 |
-+ */ |
7712 |
-+static void bfq_bfqq_expire(struct bfq_data *bfqd, |
7713 |
-+ struct bfq_queue *bfqq, |
7714 |
-+ int compensate, |
7715 |
-+ enum bfqq_expiration reason) |
7716 |
-+{ |
7717 |
-+ int slow; |
7718 |
-+ BUG_ON(bfqq != bfqd->in_service_queue); |
7719 |
-+ |
7720 |
-+ /* Update disk peak rate for autotuning and check whether the |
7721 |
-+ * process is slow (see bfq_update_peak_rate). |
7722 |
-+ */ |
7723 |
-+ slow = bfq_update_peak_rate(bfqd, bfqq, compensate, reason); |
7724 |
-+ |
7725 |
-+ /* |
7726 |
-+ * As above explained, 'punish' slow (i.e., seeky), timed-out |
7727 |
-+ * and async queues, to favor sequential sync workloads. |
7728 |
-+ * |
7729 |
-+ * Processes doing IO in the slower disk zones will tend to be |
7730 |
-+ * slow(er) even if not seeky. Hence, since the estimated peak |
7731 |
-+ * rate is actually an average over the disk surface, these |
7732 |
-+ * processes may timeout just for bad luck. To avoid punishing |
7733 |
-+ * them we do not charge a full budget to a process that |
7734 |
-+ * succeeded in consuming at least 2/3 of its budget. |
7735 |
-+ */ |
7736 |
-+ if (slow || (reason == BFQ_BFQQ_BUDGET_TIMEOUT && |
7737 |
-+ bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3)) |
7738 |
-+ bfq_bfqq_charge_full_budget(bfqq); |
7739 |
-+ |
7740 |
-+ bfqq->service_from_backlogged += bfqq->entity.service; |
7741 |
-+ |
7742 |
-+ if (bfqd->low_latency && bfqq->raising_coeff == 1) |
7743 |
-+ bfqq->last_rais_start_finish = jiffies; |
7744 |
-+ |
7745 |
-+ if (bfqd->low_latency && bfqd->bfq_raising_max_softrt_rate > 0 && |
7746 |
-+ RB_EMPTY_ROOT(&bfqq->sort_list)) { |
7747 |
-+ /* |
7748 |
-+ * If we get here, and there are no outstanding requests, |
7749 |
-+ * then the request pattern is isochronous (see the comments |
7750 |
-+ * to the function bfq_bfqq_softrt_next_start()). Hence we can |
7751 |
-+ * compute soft_rt_next_start. If, instead, the queue still |
7752 |
-+ * has outstanding requests, then we have to wait for the |
7753 |
-+ * completion of all the outstanding requests to discover |
7754 |
-+ * whether the request pattern is actually isochronous. |
7755 |
-+ */ |
7756 |
-+ if (bfqq->dispatched == 0) |
7757 |
-+ bfqq->soft_rt_next_start = |
7758 |
-+ bfq_bfqq_softrt_next_start(bfqd, bfqq); |
7759 |
-+ else { |
7760 |
-+ /* |
7761 |
-+ * The application is still waiting for the |
7762 |
-+ * completion of one or more requests: |
7763 |
-+ * prevent it from possibly being incorrectly |
7764 |
-+ * deemed as soft real-time by setting its |
7765 |
-+ * soft_rt_next_start to infinity. In fact, |
7766 |
-+ * without this assignment, the application |
7767 |
-+ * would be incorrectly deemed as soft |
7768 |
-+ * real-time if: |
7769 |
-+ * 1) it issued a new request before the |
7770 |
-+ * completion of all its in-flight |
7771 |
-+ * requests, and |
7772 |
-+ * 2) at that time, its soft_rt_next_start |
7773 |
-+ * happened to be in the past. |
7774 |
-+ */ |
7775 |
-+ bfqq->soft_rt_next_start = |
7776 |
-+ bfq_infinity_from_now(jiffies); |
7777 |
-+ /* |
7778 |
-+ * Schedule an update of soft_rt_next_start to when |
7779 |
-+ * the task may be discovered to be isochronous. |
7780 |
-+ */ |
7781 |
-+ bfq_mark_bfqq_softrt_update(bfqq); |
7782 |
-+ } |
7783 |
-+ } |
7784 |
-+ |
7785 |
-+ bfq_log_bfqq(bfqd, bfqq, |
7786 |
-+ "expire (%d, slow %d, num_disp %d, idle_win %d)", reason, slow, |
7787 |
-+ bfqq->dispatched, bfq_bfqq_idle_window(bfqq)); |
7788 |
-+ |
7789 |
-+ /* Increase, decrease or leave budget unchanged according to reason */ |
7790 |
-+ __bfq_bfqq_recalc_budget(bfqd, bfqq, reason); |
7791 |
-+ __bfq_bfqq_expire(bfqd, bfqq); |
7792 |
-+} |
7793 |
-+ |
7794 |
-+/* |
7795 |
-+ * Budget timeout is not implemented through a dedicated timer, but |
7796 |
-+ * just checked on request arrivals and completions, as well as on |
7797 |
-+ * idle timer expirations. |
7798 |
-+ */ |
7799 |
-+static int bfq_bfqq_budget_timeout(struct bfq_queue *bfqq) |
7800 |
-+{ |
7801 |
-+ if (bfq_bfqq_budget_new(bfqq)) |
7802 |
-+ return 0; |
7803 |
-+ |
7804 |
-+ if (time_before(jiffies, bfqq->budget_timeout)) |
7805 |
-+ return 0; |
7806 |
-+ |
7807 |
-+ return 1; |
7808 |
-+} |
7809 |
-+ |
7810 |
-+/* |
7811 |
-+ * If we expire a queue that is waiting for the arrival of a new |
7812 |
-+ * request, we may prevent the fictitious timestamp back-shifting that |
7813 |
-+ * allows the guarantees of the queue to be preserved (see [1] for |
7814 |
-+ * this tricky aspect). Hence we return true only if this condition |
7815 |
-+ * does not hold, or if the queue is slow enough to deserve only to be |
7816 |
-+ * kicked off for preserving a high throughput. |
7817 |
-+*/ |
7818 |
-+static inline int bfq_may_expire_for_budg_timeout(struct bfq_queue *bfqq) |
7819 |
-+{ |
7820 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, |
7821 |
-+ "may_budget_timeout: wr %d left %d timeout %d", |
7822 |
-+ bfq_bfqq_wait_request(bfqq), |
7823 |
-+ bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3, |
7824 |
-+ bfq_bfqq_budget_timeout(bfqq)); |
7825 |
-+ |
7826 |
-+ return (!bfq_bfqq_wait_request(bfqq) || |
7827 |
-+ bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3) |
7828 |
-+ && |
7829 |
-+ bfq_bfqq_budget_timeout(bfqq); |
7830 |
-+} |
7831 |
-+ |
7832 |
-+/* |
7833 |
-+ * For weight-raised queues issuing sync requests, idling is always performed, |
7834 |
-+ * as this is instrumental in guaranteeing a high fraction of the throughput |
7835 |
-+ * to these queues, and hence in guaranteeing a lower latency for their |
7836 |
-+ * requests. See [1] for details. |
7837 |
-+ * |
7838 |
-+ * For non-weight-raised queues, idling is instead disabled if the device is |
7839 |
-+ * NCQ-enabled and non-rotational, as this boosts the throughput on such |
7840 |
-+ * devices. |
7841 |
-+ */ |
7842 |
-+static inline bool bfq_bfqq_must_not_expire(struct bfq_queue *bfqq) |
7843 |
-+{ |
7844 |
-+ struct bfq_data *bfqd = bfqq->bfqd; |
7845 |
-+ |
7846 |
-+ return bfq_bfqq_sync(bfqq) && ( |
7847 |
-+ bfqq->raising_coeff > 1 || |
7848 |
-+ (bfq_bfqq_idle_window(bfqq) && |
7849 |
-+ !(bfqd->hw_tag && |
7850 |
-+ (blk_queue_nonrot(bfqd->queue) || |
7851 |
-+ /* |
7852 |
-+ * If there are weight-raised busy queues, then do not idle |
7853 |
-+ * the disk for a sync non-weight-raised queue, and hence |
7854 |
-+ * expire the queue immediately if empty. Combined with the |
7855 |
-+ * timestamping rules of BFQ (see [1] for details), this |
7856 |
-+ * causes sync non-weight-raised queues to get a lower |
7857 |
-+ * fraction of the disk throughput, and hence reduces the rate |
7858 |
-+ * at which the processes associated to these queues ask for |
7859 |
-+ * requests from the request pool. |
7860 |
-+ * |
7861 |
-+ * This is beneficial for weight-raised processes, when the |
7862 |
-+ * system operates in request-pool saturation conditions |
7863 |
-+ * (e.g., in the presence of write hogs). In fact, if |
7864 |
-+ * non-weight-raised processes ask for requests at a lower |
7865 |
-+ * rate, then weight-raised processes have a higher |
7866 |
-+ * probability to get a request from the pool immediately |
7867 |
-+ * (or at least soon) when they need one. Hence they have a |
7868 |
-+ * higher probability to actually get a fraction of the disk |
7869 |
-+ * throughput proportional to their high weight. This is |
7870 |
-+ * especially true with NCQ-enabled drives, which enqueue |
7871 |
-+ * several requests in advance and further reorder |
7872 |
-+ * internally-queued requests. |
7873 |
-+ * |
7874 |
-+ * Mistreating non-weight-raised queues in the above-described |
7875 |
-+ * way, when there are busy weight-raised queues, seems to |
7876 |
-+ * mitigate starvation problems in the presence of heavy write |
7877 |
-+ * workloads and NCQ, and hence to guarantee a higher |
7878 |
-+ * application and system responsiveness in these hostile |
7879 |
-+ * scenarios. |
7880 |
-+ */ |
7881 |
-+ bfqd->raised_busy_queues > 0) |
7882 |
-+ ) |
7883 |
-+ ) |
7884 |
-+ ); |
7885 |
-+} |
7886 |
-+ |
7887 |
-+/* |
7888 |
-+ * If the in-service queue is empty, but it is sync and either of the following |
7889 |
-+ * conditions holds, then: 1) the queue must remain in service and cannot be |
7890 |
-+ * expired, and 2) the disk must be idled to wait for the possible arrival |
7891 |
-+ * of a new request for the queue. The conditions are: |
7892 |
-+ * - the device is rotational and not performing NCQ, and the queue has its |
7893 |
-+ * idle window set (in this case, waiting for a new request for the queue |
7894 |
-+ * is likely to boost the disk throughput); |
7895 |
-+ * - the queue is weight-raised (waiting for the request is necessary to |
7896 |
-+ * provide the queue with fairness and latency guarantees, see [1] for |
7897 |
-+ * details). |
7898 |
-+ */ |
7899 |
-+static inline bool bfq_bfqq_must_idle(struct bfq_queue *bfqq) |
7900 |
-+{ |
7901 |
-+ struct bfq_data *bfqd = bfqq->bfqd; |
7902 |
-+ |
7903 |
-+ return RB_EMPTY_ROOT(&bfqq->sort_list) && bfqd->bfq_slice_idle != 0 && |
7904 |
-+ bfq_bfqq_must_not_expire(bfqq) && |
7905 |
-+ !bfq_queue_nonrot_noidle(bfqd, bfqq); |
7906 |
-+} |
7907 |
-+ |
7908 |
-+/* |
7909 |
-+ * Select a queue for service. If we have a current queue in service, |
7910 |
-+ * check whether to continue servicing it, or retrieve and set a new one. |
7911 |
-+ */ |
7912 |
-+static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd) |
7913 |
-+{ |
7914 |
-+ struct bfq_queue *bfqq, *new_bfqq = NULL; |
7915 |
-+ struct request *next_rq; |
7916 |
-+ enum bfqq_expiration reason = BFQ_BFQQ_BUDGET_TIMEOUT; |
7917 |
-+ |
7918 |
-+ bfqq = bfqd->in_service_queue; |
7919 |
-+ if (bfqq == NULL) |
7920 |
-+ goto new_queue; |
7921 |
-+ |
7922 |
-+ bfq_log_bfqq(bfqd, bfqq, "select_queue: already in-service queue"); |
7923 |
-+ |
7924 |
-+ /* |
7925 |
-+ * If another queue has a request waiting within our mean seek |
7926 |
-+ * distance, let it run. The expire code will check for close |
7927 |
-+ * cooperators and put the close queue at the front of the |
7928 |
-+ * service tree. If possible, merge the expiring queue with the |
7929 |
-+ * new bfqq. |
7930 |
-+ */ |
7931 |
-+ new_bfqq = bfq_close_cooperator(bfqd, bfqq); |
7932 |
-+ if (new_bfqq != NULL && bfqq->new_bfqq == NULL) |
7933 |
-+ bfq_setup_merge(bfqq, new_bfqq); |
7934 |
-+ |
7935 |
-+ if (bfq_may_expire_for_budg_timeout(bfqq) && |
7936 |
-+ !timer_pending(&bfqd->idle_slice_timer) && |
7937 |
-+ !bfq_bfqq_must_idle(bfqq)) |
7938 |
-+ goto expire; |
7939 |
-+ |
7940 |
-+ next_rq = bfqq->next_rq; |
7941 |
-+ /* |
7942 |
-+ * If bfqq has requests queued and it has enough budget left to |
7943 |
-+ * serve them, keep the queue, otherwise expire it. |
7944 |
-+ */ |
7945 |
-+ if (next_rq != NULL) { |
7946 |
-+ if (bfq_serv_to_charge(next_rq, bfqq) > |
7947 |
-+ bfq_bfqq_budget_left(bfqq)) { |
7948 |
-+ reason = BFQ_BFQQ_BUDGET_EXHAUSTED; |
7949 |
-+ goto expire; |
7950 |
-+ } else { |
7951 |
-+ /* |
7952 |
-+ * The idle timer may be pending because we may not |
7953 |
-+ * disable disk idling even when a new request arrives |
7954 |
-+ */ |
7955 |
-+ if (timer_pending(&bfqd->idle_slice_timer)) { |
7956 |
-+ /* |
7957 |
-+ * If we get here: 1) at least a new request |
7958 |
-+ * has arrived but we have not disabled the |
7959 |
-+ * timer because the request was too small, |
7960 |
-+ * 2) then the block layer has unplugged the |
7961 |
-+ * device, causing the dispatch to be invoked. |
7962 |
-+ * |
7963 |
-+ * Since the device is unplugged, now the |
7964 |
-+ * requests are probably large enough to |
7965 |
-+ * provide a reasonable throughput. |
7966 |
-+ * So we disable idling. |
7967 |
-+ */ |
7968 |
-+ bfq_clear_bfqq_wait_request(bfqq); |
7969 |
-+ del_timer(&bfqd->idle_slice_timer); |
7970 |
-+ } |
7971 |
-+ if (new_bfqq == NULL) |
7972 |
-+ goto keep_queue; |
7973 |
-+ else |
7974 |
-+ goto expire; |
7975 |
-+ } |
7976 |
-+ } |
7977 |
-+ |
7978 |
-+ /* |
7979 |
-+ * No requests pending. If the in-service queue has no cooperator and |
7980 |
-+ * still has requests in flight (possibly waiting for a completion) |
7981 |
-+ * or is idling for a new request, then keep it. |
7982 |
-+ */ |
7983 |
-+ if (new_bfqq == NULL && (timer_pending(&bfqd->idle_slice_timer) || |
7984 |
-+ (bfqq->dispatched != 0 && bfq_bfqq_must_not_expire(bfqq)))) { |
7985 |
-+ bfqq = NULL; |
7986 |
-+ goto keep_queue; |
7987 |
-+ } else if (new_bfqq != NULL && timer_pending(&bfqd->idle_slice_timer)) { |
7988 |
-+ /* |
7989 |
-+ * Expiring the queue because there is a close cooperator, |
7990 |
-+ * cancel timer. |
7991 |
-+ */ |
7992 |
-+ bfq_clear_bfqq_wait_request(bfqq); |
7993 |
-+ del_timer(&bfqd->idle_slice_timer); |
7994 |
-+ } |
7995 |
-+ |
7996 |
-+ reason = BFQ_BFQQ_NO_MORE_REQUESTS; |
7997 |
-+expire: |
7998 |
-+ bfq_bfqq_expire(bfqd, bfqq, 0, reason); |
7999 |
-+new_queue: |
8000 |
-+ bfqq = bfq_set_in_service_queue(bfqd, new_bfqq); |
8001 |
-+ bfq_log(bfqd, "select_queue: new queue %d returned", |
8002 |
-+ bfqq != NULL ? bfqq->pid : 0); |
8003 |
-+keep_queue: |
8004 |
-+ return bfqq; |
8005 |
-+} |
8006 |
-+ |
8007 |
-+static void bfq_update_raising_data(struct bfq_data *bfqd, |
8008 |
-+ struct bfq_queue *bfqq) |
8009 |
-+{ |
8010 |
-+ if (bfqq->raising_coeff > 1) { /* queue is being boosted */ |
8011 |
-+ struct bfq_entity *entity = &bfqq->entity; |
8012 |
-+ |
8013 |
-+ bfq_log_bfqq(bfqd, bfqq, |
8014 |
-+ "raising period dur %u/%u msec, " |
8015 |
-+ "old raising coeff %u, w %d(%d)", |
8016 |
-+ jiffies_to_msecs(jiffies - |
8017 |
-+ bfqq->last_rais_start_finish), |
8018 |
-+ jiffies_to_msecs(bfqq->raising_cur_max_time), |
8019 |
-+ bfqq->raising_coeff, |
8020 |
-+ bfqq->entity.weight, bfqq->entity.orig_weight); |
8021 |
-+ |
8022 |
-+ BUG_ON(bfqq != bfqd->in_service_queue && entity->weight != |
8023 |
-+ entity->orig_weight * bfqq->raising_coeff); |
8024 |
-+ if (entity->ioprio_changed) |
8025 |
-+ bfq_log_bfqq(bfqd, bfqq, |
8026 |
-+ "WARN: pending prio change"); |
8027 |
-+ /* |
8028 |
-+ * If too much time has elapsed from the beginning |
8029 |
-+ * of this weight-raising, stop it. |
8030 |
-+ */ |
8031 |
-+ if (time_is_before_jiffies(bfqq->last_rais_start_finish + |
8032 |
-+ bfqq->raising_cur_max_time)) { |
8033 |
-+ bfqq->last_rais_start_finish = jiffies; |
8034 |
-+ bfq_log_bfqq(bfqd, bfqq, |
8035 |
-+ "wrais ending at %lu, " |
8036 |
-+ "rais_max_time %u", |
8037 |
-+ bfqq->last_rais_start_finish, |
8038 |
-+ jiffies_to_msecs(bfqq-> |
8039 |
-+ raising_cur_max_time)); |
8040 |
-+ bfq_bfqq_end_raising(bfqq); |
8041 |
-+ __bfq_entity_update_weight_prio( |
8042 |
-+ bfq_entity_service_tree(entity), |
8043 |
-+ entity); |
8044 |
-+ } |
8045 |
-+ } |
8046 |
-+} |
8047 |
-+ |
8048 |
-+/* |
8049 |
-+ * Dispatch one request from bfqq, moving it to the request queue |
8050 |
-+ * dispatch list. |
8051 |
-+ */ |
8052 |
-+static int bfq_dispatch_request(struct bfq_data *bfqd, |
8053 |
-+ struct bfq_queue *bfqq) |
8054 |
-+{ |
8055 |
-+ int dispatched = 0; |
8056 |
-+ struct request *rq; |
8057 |
-+ unsigned long service_to_charge; |
8058 |
-+ |
8059 |
-+ BUG_ON(RB_EMPTY_ROOT(&bfqq->sort_list)); |
8060 |
-+ |
8061 |
-+ /* Follow expired path, else get first next available. */ |
8062 |
-+ rq = bfq_check_fifo(bfqq); |
8063 |
-+ if (rq == NULL) |
8064 |
-+ rq = bfqq->next_rq; |
8065 |
-+ service_to_charge = bfq_serv_to_charge(rq, bfqq); |
8066 |
-+ |
8067 |
-+ if (service_to_charge > bfq_bfqq_budget_left(bfqq)) { |
8068 |
-+ /* |
8069 |
-+ * This may happen if the next rq is chosen |
8070 |
-+ * in fifo order instead of sector order. |
8071 |
-+ * The budget is properly dimensioned |
8072 |
-+ * to be always sufficient to serve the next request |
8073 |
-+ * only if it is chosen in sector order. The reason is |
8074 |
-+ * that it would be quite inefficient and little useful |
8075 |
-+ * to always make sure that the budget is large enough |
8076 |
-+ * to serve even the possible next rq in fifo order. |
8077 |
-+ * In fact, requests are seldom served in fifo order. |
8078 |
-+ * |
8079 |
-+ * Expire the queue for budget exhaustion, and |
8080 |
-+ * make sure that the next act_budget is enough |
8081 |
-+ * to serve the next request, even if it comes |
8082 |
-+ * from the fifo expired path. |
8083 |
-+ */ |
8084 |
-+ bfqq->next_rq = rq; |
8085 |
-+ /* |
8086 |
-+ * Since this dispatch is failed, make sure that |
8087 |
-+ * a new one will be performed |
8088 |
-+ */ |
8089 |
-+ if (!bfqd->rq_in_driver) |
8090 |
-+ bfq_schedule_dispatch(bfqd); |
8091 |
-+ goto expire; |
8092 |
-+ } |
8093 |
-+ |
8094 |
-+ /* Finally, insert request into driver dispatch list. */ |
8095 |
-+ bfq_bfqq_served(bfqq, service_to_charge); |
8096 |
-+ bfq_dispatch_insert(bfqd->queue, rq); |
8097 |
-+ |
8098 |
-+ bfq_update_raising_data(bfqd, bfqq); |
8099 |
-+ |
8100 |
-+ bfq_log_bfqq(bfqd, bfqq, |
8101 |
-+ "dispatched %u sec req (%llu), budg left %lu", |
8102 |
-+ blk_rq_sectors(rq), |
8103 |
-+ (long long unsigned)blk_rq_pos(rq), |
8104 |
-+ bfq_bfqq_budget_left(bfqq)); |
8105 |
-+ |
8106 |
-+ dispatched++; |
8107 |
-+ |
8108 |
-+ if (bfqd->in_service_bic == NULL) { |
8109 |
-+ atomic_long_inc(&RQ_BIC(rq)->icq.ioc->refcount); |
8110 |
-+ bfqd->in_service_bic = RQ_BIC(rq); |
8111 |
-+ } |
8112 |
-+ |
8113 |
-+ if (bfqd->busy_queues > 1 && ((!bfq_bfqq_sync(bfqq) && |
8114 |
-+ dispatched >= bfqd->bfq_max_budget_async_rq) || |
8115 |
-+ bfq_class_idle(bfqq))) |
8116 |
-+ goto expire; |
8117 |
-+ |
8118 |
-+ return dispatched; |
8119 |
-+ |
8120 |
-+expire: |
8121 |
-+ bfq_bfqq_expire(bfqd, bfqq, 0, BFQ_BFQQ_BUDGET_EXHAUSTED); |
8122 |
-+ return dispatched; |
8123 |
-+} |
8124 |
-+ |
8125 |
-+static int __bfq_forced_dispatch_bfqq(struct bfq_queue *bfqq) |
8126 |
-+{ |
8127 |
-+ int dispatched = 0; |
8128 |
-+ |
8129 |
-+ while (bfqq->next_rq != NULL) { |
8130 |
-+ bfq_dispatch_insert(bfqq->bfqd->queue, bfqq->next_rq); |
8131 |
-+ dispatched++; |
8132 |
-+ } |
8133 |
-+ |
8134 |
-+ BUG_ON(!list_empty(&bfqq->fifo)); |
8135 |
-+ return dispatched; |
8136 |
-+} |
8137 |
-+ |
8138 |
-+/* |
8139 |
-+ * Drain our current requests. Used for barriers and when switching |
8140 |
-+ * io schedulers on-the-fly. |
8141 |
-+ */ |
8142 |
-+static int bfq_forced_dispatch(struct bfq_data *bfqd) |
8143 |
-+{ |
8144 |
-+ struct bfq_queue *bfqq, *n; |
8145 |
-+ struct bfq_service_tree *st; |
8146 |
-+ int dispatched = 0; |
8147 |
-+ |
8148 |
-+ bfqq = bfqd->in_service_queue; |
8149 |
-+ if (bfqq != NULL) |
8150 |
-+ __bfq_bfqq_expire(bfqd, bfqq); |
8151 |
-+ |
8152 |
-+ /* |
8153 |
-+ * Loop through classes, and be careful to leave the scheduler |
8154 |
-+ * in a consistent state, as feedback mechanisms and vtime |
8155 |
-+ * updates cannot be disabled during the process. |
8156 |
-+ */ |
8157 |
-+ list_for_each_entry_safe(bfqq, n, &bfqd->active_list, bfqq_list) { |
8158 |
-+ st = bfq_entity_service_tree(&bfqq->entity); |
8159 |
-+ |
8160 |
-+ dispatched += __bfq_forced_dispatch_bfqq(bfqq); |
8161 |
-+ bfqq->max_budget = bfq_max_budget(bfqd); |
8162 |
-+ |
8163 |
-+ bfq_forget_idle(st); |
8164 |
-+ } |
8165 |
-+ |
8166 |
-+ BUG_ON(bfqd->busy_queues != 0); |
8167 |
-+ |
8168 |
-+ return dispatched; |
8169 |
-+} |
8170 |
-+ |
8171 |
-+static int bfq_dispatch_requests(struct request_queue *q, int force) |
8172 |
-+{ |
8173 |
-+ struct bfq_data *bfqd = q->elevator->elevator_data; |
8174 |
-+ struct bfq_queue *bfqq; |
8175 |
-+ int max_dispatch; |
8176 |
-+ |
8177 |
-+ bfq_log(bfqd, "dispatch requests: %d busy queues", bfqd->busy_queues); |
8178 |
-+ if (bfqd->busy_queues == 0) |
8179 |
-+ return 0; |
8180 |
-+ |
8181 |
-+ if (unlikely(force)) |
8182 |
-+ return bfq_forced_dispatch(bfqd); |
8183 |
-+ |
8184 |
-+ bfqq = bfq_select_queue(bfqd); |
8185 |
-+ if (bfqq == NULL) |
8186 |
-+ return 0; |
8187 |
-+ |
8188 |
-+ max_dispatch = bfqd->bfq_quantum; |
8189 |
-+ if (bfq_class_idle(bfqq)) |
8190 |
-+ max_dispatch = 1; |
8191 |
-+ |
8192 |
-+ if (!bfq_bfqq_sync(bfqq)) |
8193 |
-+ max_dispatch = bfqd->bfq_max_budget_async_rq; |
8194 |
-+ |
8195 |
-+ if (bfqq->dispatched >= max_dispatch) { |
8196 |
-+ if (bfqd->busy_queues > 1) |
8197 |
-+ return 0; |
8198 |
-+ if (bfqq->dispatched >= 4 * max_dispatch) |
8199 |
-+ return 0; |
8200 |
-+ } |
8201 |
-+ |
8202 |
-+ if (bfqd->sync_flight != 0 && !bfq_bfqq_sync(bfqq)) |
8203 |
-+ return 0; |
8204 |
-+ |
8205 |
-+ bfq_clear_bfqq_wait_request(bfqq); |
8206 |
-+ BUG_ON(timer_pending(&bfqd->idle_slice_timer)); |
8207 |
-+ |
8208 |
-+ if (!bfq_dispatch_request(bfqd, bfqq)) |
8209 |
-+ return 0; |
8210 |
-+ |
8211 |
-+ bfq_log_bfqq(bfqd, bfqq, "dispatched one request of %d (max_disp %d)", |
8212 |
-+ bfqq->pid, max_dispatch); |
8213 |
-+ |
8214 |
-+ return 1; |
8215 |
-+} |
8216 |
-+ |
8217 |
-+/* |
8218 |
-+ * Task holds one reference to the queue, dropped when task exits. Each rq |
8219 |
-+ * in-flight on this queue also holds a reference, dropped when rq is freed. |
8220 |
-+ * |
8221 |
-+ * Queue lock must be held here. |
8222 |
-+ */ |
8223 |
-+static void bfq_put_queue(struct bfq_queue *bfqq) |
8224 |
-+{ |
8225 |
-+ struct bfq_data *bfqd = bfqq->bfqd; |
8226 |
-+ |
8227 |
-+ BUG_ON(atomic_read(&bfqq->ref) <= 0); |
8228 |
-+ |
8229 |
-+ bfq_log_bfqq(bfqd, bfqq, "put_queue: %p %d", bfqq, |
8230 |
-+ atomic_read(&bfqq->ref)); |
8231 |
-+ if (!atomic_dec_and_test(&bfqq->ref)) |
8232 |
-+ return; |
8233 |
-+ |
8234 |
-+ BUG_ON(rb_first(&bfqq->sort_list) != NULL); |
8235 |
-+ BUG_ON(bfqq->allocated[READ] + bfqq->allocated[WRITE] != 0); |
8236 |
-+ BUG_ON(bfqq->entity.tree != NULL); |
8237 |
-+ BUG_ON(bfq_bfqq_busy(bfqq)); |
8238 |
-+ BUG_ON(bfqd->in_service_queue == bfqq); |
8239 |
-+ |
8240 |
-+ bfq_log_bfqq(bfqd, bfqq, "put_queue: %p freed", bfqq); |
8241 |
-+ |
8242 |
-+ kmem_cache_free(bfq_pool, bfqq); |
8243 |
-+} |
8244 |
-+ |
8245 |
-+static void bfq_put_cooperator(struct bfq_queue *bfqq) |
8246 |
-+{ |
8247 |
-+ struct bfq_queue *__bfqq, *next; |
8248 |
-+ |
8249 |
-+ /* |
8250 |
-+ * If this queue was scheduled to merge with another queue, be |
8251 |
-+ * sure to drop the reference taken on that queue (and others in |
8252 |
-+ * the merge chain). See bfq_setup_merge and bfq_merge_bfqqs. |
8253 |
-+ */ |
8254 |
-+ __bfqq = bfqq->new_bfqq; |
8255 |
-+ while (__bfqq) { |
8256 |
-+ if (__bfqq == bfqq) { |
8257 |
-+ WARN(1, "bfqq->new_bfqq loop detected.\n"); |
8258 |
-+ break; |
8259 |
-+ } |
8260 |
-+ next = __bfqq->new_bfqq; |
8261 |
-+ bfq_put_queue(__bfqq); |
8262 |
-+ __bfqq = next; |
8263 |
-+ } |
8264 |
-+} |
8265 |
-+ |
8266 |
-+static void bfq_exit_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq) |
8267 |
-+{ |
8268 |
-+ if (bfqq == bfqd->in_service_queue) { |
8269 |
-+ __bfq_bfqq_expire(bfqd, bfqq); |
8270 |
-+ bfq_schedule_dispatch(bfqd); |
8271 |
-+ } |
8272 |
-+ |
8273 |
-+ bfq_log_bfqq(bfqd, bfqq, "exit_bfqq: %p, %d", bfqq, |
8274 |
-+ atomic_read(&bfqq->ref)); |
8275 |
-+ |
8276 |
-+ bfq_put_cooperator(bfqq); |
8277 |
-+ |
8278 |
-+ bfq_put_queue(bfqq); |
8279 |
-+} |
8280 |
-+ |
8281 |
-+static void bfq_init_icq(struct io_cq *icq) |
8282 |
-+{ |
8283 |
-+ struct bfq_io_cq *bic = icq_to_bic(icq); |
8284 |
-+ |
8285 |
-+ bic->ttime.last_end_request = jiffies; |
8286 |
-+} |
8287 |
-+ |
8288 |
-+static void bfq_exit_icq(struct io_cq *icq) |
8289 |
-+{ |
8290 |
-+ struct bfq_io_cq *bic = icq_to_bic(icq); |
8291 |
-+ struct bfq_data *bfqd = bic_to_bfqd(bic); |
8292 |
-+ |
8293 |
-+ if (bic->bfqq[BLK_RW_ASYNC]) { |
8294 |
-+ bfq_exit_bfqq(bfqd, bic->bfqq[BLK_RW_ASYNC]); |
8295 |
-+ bic->bfqq[BLK_RW_ASYNC] = NULL; |
8296 |
-+ } |
8297 |
-+ |
8298 |
-+ if (bic->bfqq[BLK_RW_SYNC]) { |
8299 |
-+ bfq_exit_bfqq(bfqd, bic->bfqq[BLK_RW_SYNC]); |
8300 |
-+ bic->bfqq[BLK_RW_SYNC] = NULL; |
8301 |
-+ } |
8302 |
-+} |
8303 |
-+ |
8304 |
-+/* |
8305 |
-+ * Update the entity prio values; note that the new values will not |
8306 |
-+ * be used until the next (re)activation. |
8307 |
-+ */ |
8308 |
-+static void bfq_init_prio_data(struct bfq_queue *bfqq, struct bfq_io_cq *bic) |
8309 |
-+{ |
8310 |
-+ struct task_struct *tsk = current; |
8311 |
-+ int ioprio_class; |
8312 |
-+ |
8313 |
-+ if (!bfq_bfqq_prio_changed(bfqq)) |
8314 |
-+ return; |
8315 |
-+ |
8316 |
-+ ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio); |
8317 |
-+ switch (ioprio_class) { |
8318 |
-+ default: |
8319 |
-+ dev_err(bfqq->bfqd->queue->backing_dev_info.dev, |
8320 |
-+ "bfq: bad prio %x\n", ioprio_class); |
8321 |
-+ case IOPRIO_CLASS_NONE: |
8322 |
-+ /* |
8323 |
-+ * No prio set, inherit CPU scheduling settings. |
8324 |
-+ */ |
8325 |
-+ bfqq->entity.new_ioprio = task_nice_ioprio(tsk); |
8326 |
-+ bfqq->entity.new_ioprio_class = task_nice_ioclass(tsk); |
8327 |
-+ break; |
8328 |
-+ case IOPRIO_CLASS_RT: |
8329 |
-+ bfqq->entity.new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio); |
8330 |
-+ bfqq->entity.new_ioprio_class = IOPRIO_CLASS_RT; |
8331 |
-+ break; |
8332 |
-+ case IOPRIO_CLASS_BE: |
8333 |
-+ bfqq->entity.new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio); |
8334 |
-+ bfqq->entity.new_ioprio_class = IOPRIO_CLASS_BE; |
8335 |
-+ break; |
8336 |
-+ case IOPRIO_CLASS_IDLE: |
8337 |
-+ bfqq->entity.new_ioprio_class = IOPRIO_CLASS_IDLE; |
8338 |
-+ bfqq->entity.new_ioprio = 7; |
8339 |
-+ bfq_clear_bfqq_idle_window(bfqq); |
8340 |
-+ break; |
8341 |
-+ } |
8342 |
-+ |
8343 |
-+ bfqq->entity.ioprio_changed = 1; |
8344 |
-+ |
8345 |
-+ /* |
8346 |
-+ * Keep track of original prio settings in case we have to temporarily |
8347 |
-+ * elevate the priority of this queue. |
8348 |
-+ */ |
8349 |
-+ bfqq->org_ioprio = bfqq->entity.new_ioprio; |
8350 |
-+ bfq_clear_bfqq_prio_changed(bfqq); |
8351 |
-+} |
8352 |
-+ |
8353 |
-+static void bfq_changed_ioprio(struct bfq_io_cq *bic) |
8354 |
-+{ |
8355 |
-+ struct bfq_data *bfqd; |
8356 |
-+ struct bfq_queue *bfqq, *new_bfqq; |
8357 |
-+ struct bfq_group *bfqg; |
8358 |
-+ unsigned long uninitialized_var(flags); |
8359 |
-+ int ioprio = bic->icq.ioc->ioprio; |
8360 |
-+ |
8361 |
-+ bfqd = bfq_get_bfqd_locked(&(bic->icq.q->elevator->elevator_data), |
8362 |
-+ &flags); |
8363 |
-+ /* |
8364 |
-+ * This condition may trigger on a newly created bic, be sure to drop |
8365 |
-+ * the lock before returning. |
8366 |
-+ */ |
8367 |
-+ if (unlikely(bfqd == NULL) || likely(bic->ioprio == ioprio)) |
8368 |
-+ goto out; |
8369 |
-+ |
8370 |
-+ bfqq = bic->bfqq[BLK_RW_ASYNC]; |
8371 |
-+ if (bfqq != NULL) { |
8372 |
-+ bfqg = container_of(bfqq->entity.sched_data, struct bfq_group, |
8373 |
-+ sched_data); |
8374 |
-+ new_bfqq = bfq_get_queue(bfqd, bfqg, BLK_RW_ASYNC, bic, |
8375 |
-+ GFP_ATOMIC); |
8376 |
-+ if (new_bfqq != NULL) { |
8377 |
-+ bic->bfqq[BLK_RW_ASYNC] = new_bfqq; |
8378 |
-+ bfq_log_bfqq(bfqd, bfqq, |
8379 |
-+ "changed_ioprio: bfqq %p %d", |
8380 |
-+ bfqq, atomic_read(&bfqq->ref)); |
8381 |
-+ bfq_put_queue(bfqq); |
8382 |
-+ } |
8383 |
-+ } |
8384 |
-+ |
8385 |
-+ bfqq = bic->bfqq[BLK_RW_SYNC]; |
8386 |
-+ if (bfqq != NULL) |
8387 |
-+ bfq_mark_bfqq_prio_changed(bfqq); |
8388 |
-+ |
8389 |
-+ bic->ioprio = ioprio; |
8390 |
-+ |
8391 |
-+out: |
8392 |
-+ bfq_put_bfqd_unlock(bfqd, &flags); |
8393 |
-+} |
8394 |
-+ |
8395 |
-+static void bfq_init_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
8396 |
-+ pid_t pid, int is_sync) |
8397 |
-+{ |
8398 |
-+ RB_CLEAR_NODE(&bfqq->entity.rb_node); |
8399 |
-+ INIT_LIST_HEAD(&bfqq->fifo); |
8400 |
-+ |
8401 |
-+ atomic_set(&bfqq->ref, 0); |
8402 |
-+ bfqq->bfqd = bfqd; |
8403 |
-+ |
8404 |
-+ bfq_mark_bfqq_prio_changed(bfqq); |
8405 |
-+ |
8406 |
-+ if (is_sync) { |
8407 |
-+ if (!bfq_class_idle(bfqq)) |
8408 |
-+ bfq_mark_bfqq_idle_window(bfqq); |
8409 |
-+ bfq_mark_bfqq_sync(bfqq); |
8410 |
-+ } |
8411 |
-+ |
8412 |
-+ /* Tentative initial value to trade off between thr and lat */ |
8413 |
-+ bfqq->max_budget = (2 * bfq_max_budget(bfqd)) / 3; |
8414 |
-+ bfqq->pid = pid; |
8415 |
-+ |
8416 |
-+ bfqq->raising_coeff = 1; |
8417 |
-+ bfqq->last_rais_start_finish = 0; |
8418 |
-+ /* |
8419 |
-+ * Set to the value for which bfqq will not be deemed as |
8420 |
-+ * soft rt when it becomes backlogged. |
8421 |
-+ */ |
8422 |
-+ bfqq->soft_rt_next_start = bfq_infinity_from_now(jiffies); |
8423 |
-+} |
8424 |
-+ |
8425 |
-+static struct bfq_queue *bfq_find_alloc_queue(struct bfq_data *bfqd, |
8426 |
-+ struct bfq_group *bfqg, |
8427 |
-+ int is_sync, |
8428 |
-+ struct bfq_io_cq *bic, |
8429 |
-+ gfp_t gfp_mask) |
8430 |
-+{ |
8431 |
-+ struct bfq_queue *bfqq, *new_bfqq = NULL; |
8432 |
-+ |
8433 |
-+retry: |
8434 |
-+ /* bic always exists here */ |
8435 |
-+ bfqq = bic_to_bfqq(bic, is_sync); |
8436 |
-+ |
8437 |
-+ /* |
8438 |
-+ * Always try a new alloc if we fall back to the OOM bfqq |
8439 |
-+ * originally, since it should just be a temporary situation. |
8440 |
-+ */ |
8441 |
-+ if (bfqq == NULL || bfqq == &bfqd->oom_bfqq) { |
8442 |
-+ bfqq = NULL; |
8443 |
-+ if (new_bfqq != NULL) { |
8444 |
-+ bfqq = new_bfqq; |
8445 |
-+ new_bfqq = NULL; |
8446 |
-+ } else if (gfp_mask & __GFP_WAIT) { |
8447 |
-+ spin_unlock_irq(bfqd->queue->queue_lock); |
8448 |
-+ new_bfqq = kmem_cache_alloc_node(bfq_pool, |
8449 |
-+ gfp_mask | __GFP_ZERO, |
8450 |
-+ bfqd->queue->node); |
8451 |
-+ spin_lock_irq(bfqd->queue->queue_lock); |
8452 |
-+ if (new_bfqq != NULL) |
8453 |
-+ goto retry; |
8454 |
-+ } else { |
8455 |
-+ bfqq = kmem_cache_alloc_node(bfq_pool, |
8456 |
-+ gfp_mask | __GFP_ZERO, |
8457 |
-+ bfqd->queue->node); |
8458 |
-+ } |
8459 |
-+ |
8460 |
-+ if (bfqq != NULL) { |
8461 |
-+ bfq_init_bfqq(bfqd, bfqq, current->pid, is_sync); |
8462 |
-+ bfq_log_bfqq(bfqd, bfqq, "allocated"); |
8463 |
-+ } else { |
8464 |
-+ bfqq = &bfqd->oom_bfqq; |
8465 |
-+ bfq_log_bfqq(bfqd, bfqq, "using oom bfqq"); |
8466 |
-+ } |
8467 |
-+ |
8468 |
-+ bfq_init_prio_data(bfqq, bic); |
8469 |
-+ bfq_init_entity(&bfqq->entity, bfqg); |
8470 |
-+ } |
8471 |
-+ |
8472 |
-+ if (new_bfqq != NULL) |
8473 |
-+ kmem_cache_free(bfq_pool, new_bfqq); |
8474 |
-+ |
8475 |
-+ return bfqq; |
8476 |
-+} |
8477 |
-+ |
8478 |
-+static struct bfq_queue **bfq_async_queue_prio(struct bfq_data *bfqd, |
8479 |
-+ struct bfq_group *bfqg, |
8480 |
-+ int ioprio_class, int ioprio) |
8481 |
-+{ |
8482 |
-+ switch (ioprio_class) { |
8483 |
-+ case IOPRIO_CLASS_RT: |
8484 |
-+ return &bfqg->async_bfqq[0][ioprio]; |
8485 |
-+ case IOPRIO_CLASS_NONE: |
8486 |
-+ ioprio = IOPRIO_NORM; |
8487 |
-+ /* fall through */ |
8488 |
-+ case IOPRIO_CLASS_BE: |
8489 |
-+ return &bfqg->async_bfqq[1][ioprio]; |
8490 |
-+ case IOPRIO_CLASS_IDLE: |
8491 |
-+ return &bfqg->async_idle_bfqq; |
8492 |
-+ default: |
8493 |
-+ BUG(); |
8494 |
-+ } |
8495 |
-+} |
8496 |
-+ |
8497 |
-+static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd, |
8498 |
-+ struct bfq_group *bfqg, int is_sync, |
8499 |
-+ struct bfq_io_cq *bic, gfp_t gfp_mask) |
8500 |
-+{ |
8501 |
-+ const int ioprio = IOPRIO_PRIO_DATA(bic->ioprio); |
8502 |
-+ const int ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio); |
8503 |
-+ struct bfq_queue **async_bfqq = NULL; |
8504 |
-+ struct bfq_queue *bfqq = NULL; |
8505 |
-+ |
8506 |
-+ if (!is_sync) { |
8507 |
-+ async_bfqq = bfq_async_queue_prio(bfqd, bfqg, ioprio_class, |
8508 |
-+ ioprio); |
8509 |
-+ bfqq = *async_bfqq; |
8510 |
-+ } |
8511 |
-+ |
8512 |
-+ if (bfqq == NULL) |
8513 |
-+ bfqq = bfq_find_alloc_queue(bfqd, bfqg, is_sync, bic, gfp_mask); |
8514 |
-+ |
8515 |
-+ /* |
8516 |
-+ * Pin the queue now that it's allocated, scheduler exit will prune it. |
8517 |
-+ */ |
8518 |
-+ if (!is_sync && *async_bfqq == NULL) { |
8519 |
-+ atomic_inc(&bfqq->ref); |
8520 |
-+ bfq_log_bfqq(bfqd, bfqq, "get_queue, bfqq not in async: %p, %d", |
8521 |
-+ bfqq, atomic_read(&bfqq->ref)); |
8522 |
-+ *async_bfqq = bfqq; |
8523 |
-+ } |
8524 |
-+ |
8525 |
-+ atomic_inc(&bfqq->ref); |
8526 |
-+ bfq_log_bfqq(bfqd, bfqq, "get_queue, at end: %p, %d", bfqq, |
8527 |
-+ atomic_read(&bfqq->ref)); |
8528 |
-+ return bfqq; |
8529 |
-+} |
8530 |
-+ |
8531 |
-+static void bfq_update_io_thinktime(struct bfq_data *bfqd, |
8532 |
-+ struct bfq_io_cq *bic) |
8533 |
-+{ |
8534 |
-+ unsigned long elapsed = jiffies - bic->ttime.last_end_request; |
8535 |
-+ unsigned long ttime = min(elapsed, 2UL * bfqd->bfq_slice_idle); |
8536 |
-+ |
8537 |
-+ bic->ttime.ttime_samples = (7*bic->ttime.ttime_samples + 256) / 8; |
8538 |
-+ bic->ttime.ttime_total = (7*bic->ttime.ttime_total + 256*ttime) / 8; |
8539 |
-+ bic->ttime.ttime_mean = (bic->ttime.ttime_total + 128) / |
8540 |
-+ bic->ttime.ttime_samples; |
8541 |
-+} |
8542 |
-+ |
8543 |
-+static void bfq_update_io_seektime(struct bfq_data *bfqd, |
8544 |
-+ struct bfq_queue *bfqq, |
8545 |
-+ struct request *rq) |
8546 |
-+{ |
8547 |
-+ sector_t sdist; |
8548 |
-+ u64 total; |
8549 |
-+ |
8550 |
-+ if (bfqq->last_request_pos < blk_rq_pos(rq)) |
8551 |
-+ sdist = blk_rq_pos(rq) - bfqq->last_request_pos; |
8552 |
-+ else |
8553 |
-+ sdist = bfqq->last_request_pos - blk_rq_pos(rq); |
8554 |
-+ |
8555 |
-+ /* |
8556 |
-+ * Don't allow the seek distance to get too large from the |
8557 |
-+ * odd fragment, pagein, etc. |
8558 |
-+ */ |
8559 |
-+ if (bfqq->seek_samples == 0) /* first request, not really a seek */ |
8560 |
-+ sdist = 0; |
8561 |
-+ else if (bfqq->seek_samples <= 60) /* second & third seek */ |
8562 |
-+ sdist = min(sdist, (bfqq->seek_mean * 4) + 2*1024*1024); |
8563 |
-+ else |
8564 |
-+ sdist = min(sdist, (bfqq->seek_mean * 4) + 2*1024*64); |
8565 |
-+ |
8566 |
-+ bfqq->seek_samples = (7*bfqq->seek_samples + 256) / 8; |
8567 |
-+ bfqq->seek_total = (7*bfqq->seek_total + (u64)256*sdist) / 8; |
8568 |
-+ total = bfqq->seek_total + (bfqq->seek_samples/2); |
8569 |
-+ do_div(total, bfqq->seek_samples); |
8570 |
-+ bfqq->seek_mean = (sector_t)total; |
8571 |
-+ |
8572 |
-+ bfq_log_bfqq(bfqd, bfqq, "dist=%llu mean=%llu", (u64)sdist, |
8573 |
-+ (u64)bfqq->seek_mean); |
8574 |
-+} |
8575 |
-+ |
8576 |
-+/* |
8577 |
-+ * Disable idle window if the process thinks too long or seeks so much that |
8578 |
-+ * it doesn't matter. |
8579 |
-+ */ |
8580 |
-+static void bfq_update_idle_window(struct bfq_data *bfqd, |
8581 |
-+ struct bfq_queue *bfqq, |
8582 |
-+ struct bfq_io_cq *bic) |
8583 |
-+{ |
8584 |
-+ int enable_idle; |
8585 |
-+ |
8586 |
-+ /* Don't idle for async or idle io prio class. */ |
8587 |
-+ if (!bfq_bfqq_sync(bfqq) || bfq_class_idle(bfqq)) |
8588 |
-+ return; |
8589 |
-+ |
8590 |
-+ enable_idle = bfq_bfqq_idle_window(bfqq); |
8591 |
-+ |
8592 |
-+ if (atomic_read(&bic->icq.ioc->active_ref) == 0 || |
8593 |
-+ bfqd->bfq_slice_idle == 0 || |
8594 |
-+ (bfqd->hw_tag && BFQQ_SEEKY(bfqq) && |
8595 |
-+ bfqq->raising_coeff == 1)) |
8596 |
-+ enable_idle = 0; |
8597 |
-+ else if (bfq_sample_valid(bic->ttime.ttime_samples)) { |
8598 |
-+ if (bic->ttime.ttime_mean > bfqd->bfq_slice_idle && |
8599 |
-+ bfqq->raising_coeff == 1) |
8600 |
-+ enable_idle = 0; |
8601 |
-+ else |
8602 |
-+ enable_idle = 1; |
8603 |
-+ } |
8604 |
-+ bfq_log_bfqq(bfqd, bfqq, "update_idle_window: enable_idle %d", |
8605 |
-+ enable_idle); |
8606 |
-+ |
8607 |
-+ if (enable_idle) |
8608 |
-+ bfq_mark_bfqq_idle_window(bfqq); |
8609 |
-+ else |
8610 |
-+ bfq_clear_bfqq_idle_window(bfqq); |
8611 |
-+} |
8612 |
-+ |
8613 |
-+/* |
8614 |
-+ * Called when a new fs request (rq) is added to bfqq. Check if there's |
8615 |
-+ * something we should do about it. |
8616 |
-+ */ |
8617 |
-+static void bfq_rq_enqueued(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
8618 |
-+ struct request *rq) |
8619 |
-+{ |
8620 |
-+ struct bfq_io_cq *bic = RQ_BIC(rq); |
8621 |
-+ |
8622 |
-+ if (rq->cmd_flags & REQ_META) |
8623 |
-+ bfqq->meta_pending++; |
8624 |
-+ |
8625 |
-+ bfq_update_io_thinktime(bfqd, bic); |
8626 |
-+ bfq_update_io_seektime(bfqd, bfqq, rq); |
8627 |
-+ if (bfqq->entity.service > bfq_max_budget(bfqd) / 8 || |
8628 |
-+ !BFQQ_SEEKY(bfqq)) |
8629 |
-+ bfq_update_idle_window(bfqd, bfqq, bic); |
8630 |
-+ |
8631 |
-+ bfq_log_bfqq(bfqd, bfqq, |
8632 |
-+ "rq_enqueued: idle_window=%d (seeky %d, mean %llu)", |
8633 |
-+ bfq_bfqq_idle_window(bfqq), BFQQ_SEEKY(bfqq), |
8634 |
-+ (long long unsigned)bfqq->seek_mean); |
8635 |
-+ |
8636 |
-+ bfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
8637 |
-+ |
8638 |
-+ if (bfqq == bfqd->in_service_queue && bfq_bfqq_wait_request(bfqq)) { |
8639 |
-+ int small_req = bfqq->queued[rq_is_sync(rq)] == 1 && |
8640 |
-+ blk_rq_sectors(rq) < 32; |
8641 |
-+ int budget_timeout = bfq_bfqq_budget_timeout(bfqq); |
8642 |
-+ |
8643 |
-+ /* |
8644 |
-+ * There is just this request queued: if the request |
8645 |
-+ * is small and the queue is not to be expired, then |
8646 |
-+ * just exit. |
8647 |
-+ * |
8648 |
-+ * In this way, if the disk is being idled to wait for |
8649 |
-+ * a new request from the in-service queue, we avoid |
8650 |
-+ * unplugging the device and committing the disk to serve |
8651 |
-+ * just a small request. On the contrary, we wait for |
8652 |
-+ * the block layer to decide when to unplug the device: |
8653 |
-+ * hopefully, new requests will be merged to this one |
8654 |
-+ * quickly, then the device will be unplugged and |
8655 |
-+ * larger requests will be dispatched. |
8656 |
-+ */ |
8657 |
-+ if (small_req && !budget_timeout) |
8658 |
-+ return; |
8659 |
-+ |
8660 |
-+ /* |
8661 |
-+ * A large enough request arrived, or the queue is to |
8662 |
-+ * be expired: in both cases disk idling is to be |
8663 |
-+ * stopped, so clear wait_request flag and reset |
8664 |
-+ * timer. |
8665 |
-+ */ |
8666 |
-+ bfq_clear_bfqq_wait_request(bfqq); |
8667 |
-+ del_timer(&bfqd->idle_slice_timer); |
8668 |
-+ |
8669 |
-+ /* |
8670 |
-+ * The queue is not empty, because a new request just |
8671 |
-+ * arrived. Hence we can safely expire the queue, in |
8672 |
-+ * case of budget timeout, without risking that the |
8673 |
-+ * timestamps of the queue are not updated correctly. |
8674 |
-+ * See [1] for more details. |
8675 |
-+ */ |
8676 |
-+ if (budget_timeout) |
8677 |
-+ bfq_bfqq_expire(bfqd, bfqq, 0, BFQ_BFQQ_BUDGET_TIMEOUT); |
8678 |
-+ |
8679 |
-+ /* |
8680 |
-+ * Let the request rip immediately, or let a new queue be |
8681 |
-+ * selected if bfqq has just been expired. |
8682 |
-+ */ |
8683 |
-+ __blk_run_queue(bfqd->queue); |
8684 |
-+ } |
8685 |
-+} |
8686 |
-+ |
8687 |
-+static void bfq_insert_request(struct request_queue *q, struct request *rq) |
8688 |
-+{ |
8689 |
-+ struct bfq_data *bfqd = q->elevator->elevator_data; |
8690 |
-+ struct bfq_queue *bfqq = RQ_BFQQ(rq); |
8691 |
-+ |
8692 |
-+ assert_spin_locked(bfqd->queue->queue_lock); |
8693 |
-+ bfq_init_prio_data(bfqq, RQ_BIC(rq)); |
8694 |
-+ |
8695 |
-+ bfq_add_rq_rb(rq); |
8696 |
-+ |
8697 |
-+ rq_set_fifo_time(rq, jiffies + bfqd->bfq_fifo_expire[rq_is_sync(rq)]); |
8698 |
-+ list_add_tail(&rq->queuelist, &bfqq->fifo); |
8699 |
-+ |
8700 |
-+ bfq_rq_enqueued(bfqd, bfqq, rq); |
8701 |
-+} |
8702 |
-+ |
8703 |
-+static void bfq_update_hw_tag(struct bfq_data *bfqd) |
8704 |
-+{ |
8705 |
-+ bfqd->max_rq_in_driver = max(bfqd->max_rq_in_driver, |
8706 |
-+ bfqd->rq_in_driver); |
8707 |
-+ |
8708 |
-+ if (bfqd->hw_tag == 1) |
8709 |
-+ return; |
8710 |
-+ |
8711 |
-+ /* |
8712 |
-+ * This sample is valid if the number of outstanding requests |
8713 |
-+ * is large enough to allow a queueing behavior. Note that the |
8714 |
-+ * sum is not exact, as it's not taking into account deactivated |
8715 |
-+ * requests. |
8716 |
-+ */ |
8717 |
-+ if (bfqd->rq_in_driver + bfqd->queued < BFQ_HW_QUEUE_THRESHOLD) |
8718 |
-+ return; |
8719 |
-+ |
8720 |
-+ if (bfqd->hw_tag_samples++ < BFQ_HW_QUEUE_SAMPLES) |
8721 |
-+ return; |
8722 |
-+ |
8723 |
-+ bfqd->hw_tag = bfqd->max_rq_in_driver > BFQ_HW_QUEUE_THRESHOLD; |
8724 |
-+ bfqd->max_rq_in_driver = 0; |
8725 |
-+ bfqd->hw_tag_samples = 0; |
8726 |
-+} |
8727 |
-+ |
8728 |
-+static void bfq_completed_request(struct request_queue *q, struct request *rq) |
8729 |
-+{ |
8730 |
-+ struct bfq_queue *bfqq = RQ_BFQQ(rq); |
8731 |
-+ struct bfq_data *bfqd = bfqq->bfqd; |
8732 |
-+ const int sync = rq_is_sync(rq); |
8733 |
-+ |
8734 |
-+ bfq_log_bfqq(bfqd, bfqq, "completed %u sects req (%d)", |
8735 |
-+ blk_rq_sectors(rq), sync); |
8736 |
-+ |
8737 |
-+ bfq_update_hw_tag(bfqd); |
8738 |
-+ |
8739 |
-+ WARN_ON(!bfqd->rq_in_driver); |
8740 |
-+ WARN_ON(!bfqq->dispatched); |
8741 |
-+ bfqd->rq_in_driver--; |
8742 |
-+ bfqq->dispatched--; |
8743 |
-+ |
8744 |
-+ if (bfq_bfqq_sync(bfqq)) |
8745 |
-+ bfqd->sync_flight--; |
8746 |
-+ |
8747 |
-+ if (sync) |
8748 |
-+ RQ_BIC(rq)->ttime.last_end_request = jiffies; |
8749 |
-+ |
8750 |
-+ /* |
8751 |
-+ * If we are waiting to discover whether the request pattern of the |
8752 |
-+ * task associated with the queue is actually isochronous, and |
8753 |
-+ * both requisites for this condition to hold are satisfied, then |
8754 |
-+ * compute soft_rt_next_start (see the comments to the function |
8755 |
-+ * bfq_bfqq_softrt_next_start()). |
8756 |
-+ */ |
8757 |
-+ if (bfq_bfqq_softrt_update(bfqq) && bfqq->dispatched == 0 && |
8758 |
-+ RB_EMPTY_ROOT(&bfqq->sort_list)) |
8759 |
-+ bfqq->soft_rt_next_start = |
8760 |
-+ bfq_bfqq_softrt_next_start(bfqd, bfqq); |
8761 |
-+ |
8762 |
-+ /* |
8763 |
-+ * If this is the in-service queue, check if it needs to be expired, |
8764 |
-+ * or if we want to idle in case it has no pending requests. |
8765 |
-+ */ |
8766 |
-+ if (bfqd->in_service_queue == bfqq) { |
8767 |
-+ if (bfq_bfqq_budget_new(bfqq)) |
8768 |
-+ bfq_set_budget_timeout(bfqd); |
8769 |
-+ |
8770 |
-+ if (bfq_bfqq_must_idle(bfqq)) { |
8771 |
-+ bfq_arm_slice_timer(bfqd); |
8772 |
-+ goto out; |
8773 |
-+ } else if (bfq_may_expire_for_budg_timeout(bfqq)) |
8774 |
-+ bfq_bfqq_expire(bfqd, bfqq, 0, BFQ_BFQQ_BUDGET_TIMEOUT); |
8775 |
-+ else if (RB_EMPTY_ROOT(&bfqq->sort_list) && |
8776 |
-+ (bfqq->dispatched == 0 || |
8777 |
-+ !bfq_bfqq_must_not_expire(bfqq))) |
8778 |
-+ bfq_bfqq_expire(bfqd, bfqq, 0, |
8779 |
-+ BFQ_BFQQ_NO_MORE_REQUESTS); |
8780 |
-+ } |
8781 |
-+ |
8782 |
-+ if (!bfqd->rq_in_driver) |
8783 |
-+ bfq_schedule_dispatch(bfqd); |
8784 |
-+ |
8785 |
-+out: |
8786 |
-+ return; |
8787 |
-+} |
8788 |
-+ |
8789 |
-+static inline int __bfq_may_queue(struct bfq_queue *bfqq) |
8790 |
-+{ |
8791 |
-+ if (bfq_bfqq_wait_request(bfqq) && bfq_bfqq_must_alloc(bfqq)) { |
8792 |
-+ bfq_clear_bfqq_must_alloc(bfqq); |
8793 |
-+ return ELV_MQUEUE_MUST; |
8794 |
-+ } |
8795 |
-+ |
8796 |
-+ return ELV_MQUEUE_MAY; |
8797 |
-+} |
8798 |
-+ |
8799 |
-+static int bfq_may_queue(struct request_queue *q, int rw) |
8800 |
-+{ |
8801 |
-+ struct bfq_data *bfqd = q->elevator->elevator_data; |
8802 |
-+ struct task_struct *tsk = current; |
8803 |
-+ struct bfq_io_cq *bic; |
8804 |
-+ struct bfq_queue *bfqq; |
8805 |
-+ |
8806 |
-+ /* |
8807 |
-+ * Don't force setup of a queue from here, as a call to may_queue |
8808 |
-+ * does not necessarily imply that a request actually will be queued. |
8809 |
-+ * So just lookup a possibly existing queue, or return 'may queue' |
8810 |
-+ * if that fails. |
8811 |
-+ */ |
8812 |
-+ bic = bfq_bic_lookup(bfqd, tsk->io_context); |
8813 |
-+ if (bic == NULL) |
8814 |
-+ return ELV_MQUEUE_MAY; |
8815 |
-+ |
8816 |
-+ bfqq = bic_to_bfqq(bic, rw_is_sync(rw)); |
8817 |
-+ if (bfqq != NULL) { |
8818 |
-+ bfq_init_prio_data(bfqq, bic); |
8819 |
-+ |
8820 |
-+ return __bfq_may_queue(bfqq); |
8821 |
-+ } |
8822 |
-+ |
8823 |
-+ return ELV_MQUEUE_MAY; |
8824 |
-+} |
8825 |
-+ |
8826 |
-+/* |
8827 |
-+ * Queue lock held here. |
8828 |
-+ */ |
8829 |
-+static void bfq_put_request(struct request *rq) |
8830 |
-+{ |
8831 |
-+ struct bfq_queue *bfqq = RQ_BFQQ(rq); |
8832 |
-+ |
8833 |
-+ if (bfqq != NULL) { |
8834 |
-+ const int rw = rq_data_dir(rq); |
8835 |
-+ |
8836 |
-+ BUG_ON(!bfqq->allocated[rw]); |
8837 |
-+ bfqq->allocated[rw]--; |
8838 |
-+ |
8839 |
-+ rq->elv.priv[0] = NULL; |
8840 |
-+ rq->elv.priv[1] = NULL; |
8841 |
-+ |
8842 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, "put_request %p, %d", |
8843 |
-+ bfqq, atomic_read(&bfqq->ref)); |
8844 |
-+ bfq_put_queue(bfqq); |
8845 |
-+ } |
8846 |
-+} |
8847 |
-+ |
8848 |
-+static struct bfq_queue * |
8849 |
-+bfq_merge_bfqqs(struct bfq_data *bfqd, struct bfq_io_cq *bic, |
8850 |
-+ struct bfq_queue *bfqq) |
8851 |
-+{ |
8852 |
-+ bfq_log_bfqq(bfqd, bfqq, "merging with queue %lu", |
8853 |
-+ (long unsigned)bfqq->new_bfqq->pid); |
8854 |
-+ bic_set_bfqq(bic, bfqq->new_bfqq, 1); |
8855 |
-+ bfq_mark_bfqq_coop(bfqq->new_bfqq); |
8856 |
-+ bfq_put_queue(bfqq); |
8857 |
-+ return bic_to_bfqq(bic, 1); |
8858 |
-+} |
8859 |
-+ |
8860 |
-+/* |
8861 |
-+ * Returns NULL if a new bfqq should be allocated, or the old bfqq if this |
8862 |
-+ * was the last process referring to said bfqq. |
8863 |
-+ */ |
8864 |
-+static struct bfq_queue * |
8865 |
-+bfq_split_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq) |
8866 |
-+{ |
8867 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, "splitting queue"); |
8868 |
-+ if (bfqq_process_refs(bfqq) == 1) { |
8869 |
-+ bfqq->pid = current->pid; |
8870 |
-+ bfq_clear_bfqq_coop(bfqq); |
8871 |
-+ bfq_clear_bfqq_split_coop(bfqq); |
8872 |
-+ return bfqq; |
8873 |
-+ } |
8874 |
-+ |
8875 |
-+ bic_set_bfqq(bic, NULL, 1); |
8876 |
-+ |
8877 |
-+ bfq_put_cooperator(bfqq); |
8878 |
-+ |
8879 |
-+ bfq_put_queue(bfqq); |
8880 |
-+ return NULL; |
8881 |
-+} |
8882 |
-+ |
8883 |
-+/* |
8884 |
-+ * Allocate bfq data structures associated with this request. |
8885 |
-+ */ |
8886 |
-+static int bfq_set_request(struct request_queue *q, struct request *rq, |
8887 |
-+ struct bio *bio, gfp_t gfp_mask) |
8888 |
-+{ |
8889 |
-+ struct bfq_data *bfqd = q->elevator->elevator_data; |
8890 |
-+ struct bfq_io_cq *bic = icq_to_bic(rq->elv.icq); |
8891 |
-+ const int rw = rq_data_dir(rq); |
8892 |
-+ const int is_sync = rq_is_sync(rq); |
8893 |
-+ struct bfq_queue *bfqq; |
8894 |
-+ struct bfq_group *bfqg; |
8895 |
-+ unsigned long flags; |
8896 |
-+ |
8897 |
-+ might_sleep_if(gfp_mask & __GFP_WAIT); |
8898 |
-+ |
8899 |
-+ bfq_changed_ioprio(bic); |
8900 |
-+ |
8901 |
-+ spin_lock_irqsave(q->queue_lock, flags); |
8902 |
-+ |
8903 |
-+ if (bic == NULL) |
8904 |
-+ goto queue_fail; |
8905 |
-+ |
8906 |
-+ bfqg = bfq_bic_update_cgroup(bic); |
8907 |
-+ |
8908 |
-+new_queue: |
8909 |
-+ bfqq = bic_to_bfqq(bic, is_sync); |
8910 |
-+ if (bfqq == NULL || bfqq == &bfqd->oom_bfqq) { |
8911 |
-+ bfqq = bfq_get_queue(bfqd, bfqg, is_sync, bic, gfp_mask); |
8912 |
-+ bic_set_bfqq(bic, bfqq, is_sync); |
8913 |
-+ } else { |
8914 |
-+ /* |
8915 |
-+ * If the queue was seeky for too long, break it apart. |
8916 |
-+ */ |
8917 |
-+ if (bfq_bfqq_coop(bfqq) && bfq_bfqq_split_coop(bfqq)) { |
8918 |
-+ bfq_log_bfqq(bfqd, bfqq, "breaking apart bfqq"); |
8919 |
-+ bfqq = bfq_split_bfqq(bic, bfqq); |
8920 |
-+ if (!bfqq) |
8921 |
-+ goto new_queue; |
8922 |
-+ } |
8923 |
-+ |
8924 |
-+ /* |
8925 |
-+ * Check to see if this queue is scheduled to merge with |
8926 |
-+ * another closely cooperating queue. The merging of queues |
8927 |
-+ * happens here as it must be done in process context. |
8928 |
-+ * The reference on new_bfqq was taken in merge_bfqqs. |
8929 |
-+ */ |
8930 |
-+ if (bfqq->new_bfqq != NULL) |
8931 |
-+ bfqq = bfq_merge_bfqqs(bfqd, bic, bfqq); |
8932 |
-+ } |
8933 |
-+ |
8934 |
-+ bfqq->allocated[rw]++; |
8935 |
-+ atomic_inc(&bfqq->ref); |
8936 |
-+ bfq_log_bfqq(bfqd, bfqq, "set_request: bfqq %p, %d", bfqq, |
8937 |
-+ atomic_read(&bfqq->ref)); |
8938 |
-+ |
8939 |
-+ rq->elv.priv[0] = bic; |
8940 |
-+ rq->elv.priv[1] = bfqq; |
8941 |
-+ |
8942 |
-+ spin_unlock_irqrestore(q->queue_lock, flags); |
8943 |
-+ |
8944 |
-+ return 0; |
8945 |
-+ |
8946 |
-+queue_fail: |
8947 |
-+ bfq_schedule_dispatch(bfqd); |
8948 |
-+ spin_unlock_irqrestore(q->queue_lock, flags); |
8949 |
-+ |
8950 |
-+ return 1; |
8951 |
-+} |
8952 |
-+ |
8953 |
-+static void bfq_kick_queue(struct work_struct *work) |
8954 |
-+{ |
8955 |
-+ struct bfq_data *bfqd = |
8956 |
-+ container_of(work, struct bfq_data, unplug_work); |
8957 |
-+ struct request_queue *q = bfqd->queue; |
8958 |
-+ |
8959 |
-+ spin_lock_irq(q->queue_lock); |
8960 |
-+ __blk_run_queue(q); |
8961 |
-+ spin_unlock_irq(q->queue_lock); |
8962 |
-+} |
8963 |
-+ |
8964 |
-+/* |
8965 |
-+ * Handler of the expiration of the timer running if the in-service queue |
8966 |
-+ * is idling inside its time slice. |
8967 |
-+ */ |
8968 |
-+static void bfq_idle_slice_timer(unsigned long data) |
8969 |
-+{ |
8970 |
-+ struct bfq_data *bfqd = (struct bfq_data *)data; |
8971 |
-+ struct bfq_queue *bfqq; |
8972 |
-+ unsigned long flags; |
8973 |
-+ enum bfqq_expiration reason; |
8974 |
-+ |
8975 |
-+ spin_lock_irqsave(bfqd->queue->queue_lock, flags); |
8976 |
-+ |
8977 |
-+ bfqq = bfqd->in_service_queue; |
8978 |
-+ /* |
8979 |
-+ * Theoretical race here: the in-service queue can be NULL or different |
8980 |
-+ * from the queue that was idling if the timer handler spins on |
8981 |
-+ * the queue_lock and a new request arrives for the current |
8982 |
-+ * queue and there is a full dispatch cycle that changes the |
8983 |
-+ * in-service queue. This can hardly happen, but in the worst case |
8984 |
-+ * we just expire a queue too early. |
8985 |
-+ */ |
8986 |
-+ if (bfqq != NULL) { |
8987 |
-+ bfq_log_bfqq(bfqd, bfqq, "slice_timer expired"); |
8988 |
-+ if (bfq_bfqq_budget_timeout(bfqq)) |
8989 |
-+ /* |
8990 |
-+ * Also here the queue can be safely expired |
8991 |
-+ * for budget timeout without wasting |
8992 |
-+ * guarantees |
8993 |
-+ */ |
8994 |
-+ reason = BFQ_BFQQ_BUDGET_TIMEOUT; |
8995 |
-+ else if (bfqq->queued[0] == 0 && bfqq->queued[1] == 0) |
8996 |
-+ /* |
8997 |
-+ * The queue may not be empty upon timer expiration, |
8998 |
-+ * because we may not disable the timer when the first |
8999 |
-+ * request of the in-service queue arrives during |
9000 |
-+ * disk idling |
9001 |
-+ */ |
9002 |
-+ reason = BFQ_BFQQ_TOO_IDLE; |
9003 |
-+ else |
9004 |
-+ goto schedule_dispatch; |
9005 |
-+ |
9006 |
-+ bfq_bfqq_expire(bfqd, bfqq, 1, reason); |
9007 |
-+ } |
9008 |
-+ |
9009 |
-+schedule_dispatch: |
9010 |
-+ bfq_schedule_dispatch(bfqd); |
9011 |
-+ |
9012 |
-+ spin_unlock_irqrestore(bfqd->queue->queue_lock, flags); |
9013 |
-+} |
9014 |
-+ |
9015 |
-+static void bfq_shutdown_timer_wq(struct bfq_data *bfqd) |
9016 |
-+{ |
9017 |
-+ del_timer_sync(&bfqd->idle_slice_timer); |
9018 |
-+ cancel_work_sync(&bfqd->unplug_work); |
9019 |
-+} |
9020 |
-+ |
9021 |
-+static inline void __bfq_put_async_bfqq(struct bfq_data *bfqd, |
9022 |
-+ struct bfq_queue **bfqq_ptr) |
9023 |
-+{ |
9024 |
-+ struct bfq_group *root_group = bfqd->root_group; |
9025 |
-+ struct bfq_queue *bfqq = *bfqq_ptr; |
9026 |
-+ |
9027 |
-+ bfq_log(bfqd, "put_async_bfqq: %p", bfqq); |
9028 |
-+ if (bfqq != NULL) { |
9029 |
-+ bfq_bfqq_move(bfqd, bfqq, &bfqq->entity, root_group); |
9030 |
-+ bfq_log_bfqq(bfqd, bfqq, "put_async_bfqq: putting %p, %d", |
9031 |
-+ bfqq, atomic_read(&bfqq->ref)); |
9032 |
-+ bfq_put_queue(bfqq); |
9033 |
-+ *bfqq_ptr = NULL; |
9034 |
-+ } |
9035 |
-+} |
9036 |
-+ |
9037 |
-+/* |
9038 |
-+ * Release all the bfqg references to its async queues. If we are |
9039 |
-+ * deallocating the group these queues may still contain requests, so |
9040 |
-+ * we reparent them to the root cgroup (i.e., the only one that will |
9041 |
-+ * exist for sure until all the requests on a device are gone). |
9042 |
-+ */ |
9043 |
-+static void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg) |
9044 |
-+{ |
9045 |
-+ int i, j; |
9046 |
-+ |
9047 |
-+ for (i = 0; i < 2; i++) |
9048 |
-+ for (j = 0; j < IOPRIO_BE_NR; j++) |
9049 |
-+ __bfq_put_async_bfqq(bfqd, &bfqg->async_bfqq[i][j]); |
9050 |
-+ |
9051 |
-+ __bfq_put_async_bfqq(bfqd, &bfqg->async_idle_bfqq); |
9052 |
-+} |
9053 |
-+ |
9054 |
-+static void bfq_exit_queue(struct elevator_queue *e) |
9055 |
-+{ |
9056 |
-+ struct bfq_data *bfqd = e->elevator_data; |
9057 |
-+ struct request_queue *q = bfqd->queue; |
9058 |
-+ struct bfq_queue *bfqq, *n; |
9059 |
-+ |
9060 |
-+ bfq_shutdown_timer_wq(bfqd); |
9061 |
-+ |
9062 |
-+ spin_lock_irq(q->queue_lock); |
9063 |
-+ |
9064 |
-+ BUG_ON(bfqd->in_service_queue != NULL); |
9065 |
-+ list_for_each_entry_safe(bfqq, n, &bfqd->idle_list, bfqq_list) |
9066 |
-+ bfq_deactivate_bfqq(bfqd, bfqq, 0); |
9067 |
-+ |
9068 |
-+ bfq_disconnect_groups(bfqd); |
9069 |
-+ spin_unlock_irq(q->queue_lock); |
9070 |
-+ |
9071 |
-+ bfq_shutdown_timer_wq(bfqd); |
9072 |
-+ |
9073 |
-+ synchronize_rcu(); |
9074 |
-+ |
9075 |
-+ BUG_ON(timer_pending(&bfqd->idle_slice_timer)); |
9076 |
-+ |
9077 |
-+ bfq_free_root_group(bfqd); |
9078 |
-+ kfree(bfqd); |
9079 |
-+} |
9080 |
-+ |
9081 |
-+static int bfq_init_queue(struct request_queue *q, struct elevator_type *e) |
9082 |
-+{ |
9083 |
-+ struct bfq_group *bfqg; |
9084 |
-+ struct bfq_data *bfqd; |
9085 |
-+ struct elevator_queue *eq; |
9086 |
-+ |
9087 |
-+ eq = elevator_alloc(q, e); |
9088 |
-+ if (eq == NULL) |
9089 |
-+ return -ENOMEM; |
9090 |
-+ |
9091 |
-+ bfqd = kzalloc_node(sizeof(*bfqd), GFP_KERNEL, q->node); |
9092 |
-+ if (bfqd == NULL) { |
9093 |
-+ kobject_put(&eq->kobj); |
9094 |
-+ return -ENOMEM; |
9095 |
-+ } |
9096 |
-+ eq->elevator_data = bfqd; |
9097 |
-+ |
9098 |
-+ /* |
9099 |
-+ * Our fallback bfqq if bfq_find_alloc_queue() runs into OOM issues. |
9100 |
-+ * Grab a permanent reference to it, so that the normal code flow |
9101 |
-+ * will not attempt to free it. |
9102 |
-+ */ |
9103 |
-+ bfq_init_bfqq(bfqd, &bfqd->oom_bfqq, 1, 0); |
9104 |
-+ atomic_inc(&bfqd->oom_bfqq.ref); |
9105 |
-+ |
9106 |
-+ bfqd->queue = q; |
9107 |
-+ |
9108 |
-+ spin_lock_irq(q->queue_lock); |
9109 |
-+ q->elevator = eq; |
9110 |
-+ spin_unlock_irq(q->queue_lock); |
9111 |
-+ |
9112 |
-+ bfqg = bfq_alloc_root_group(bfqd, q->node); |
9113 |
-+ if (bfqg == NULL) { |
9114 |
-+ kfree(bfqd); |
9115 |
-+ kobject_put(&eq->kobj); |
9116 |
-+ return -ENOMEM; |
9117 |
-+ } |
9118 |
-+ |
9119 |
-+ bfqd->root_group = bfqg; |
9120 |
-+ |
9121 |
-+ init_timer(&bfqd->idle_slice_timer); |
9122 |
-+ bfqd->idle_slice_timer.function = bfq_idle_slice_timer; |
9123 |
-+ bfqd->idle_slice_timer.data = (unsigned long)bfqd; |
9124 |
-+ |
9125 |
-+ bfqd->rq_pos_tree = RB_ROOT; |
9126 |
-+ |
9127 |
-+ INIT_WORK(&bfqd->unplug_work, bfq_kick_queue); |
9128 |
-+ |
9129 |
-+ INIT_LIST_HEAD(&bfqd->active_list); |
9130 |
-+ INIT_LIST_HEAD(&bfqd->idle_list); |
9131 |
-+ |
9132 |
-+ bfqd->hw_tag = -1; |
9133 |
-+ |
9134 |
-+ bfqd->bfq_max_budget = bfq_default_max_budget; |
9135 |
-+ |
9136 |
-+ bfqd->bfq_quantum = bfq_quantum; |
9137 |
-+ bfqd->bfq_fifo_expire[0] = bfq_fifo_expire[0]; |
9138 |
-+ bfqd->bfq_fifo_expire[1] = bfq_fifo_expire[1]; |
9139 |
-+ bfqd->bfq_back_max = bfq_back_max; |
9140 |
-+ bfqd->bfq_back_penalty = bfq_back_penalty; |
9141 |
-+ bfqd->bfq_slice_idle = bfq_slice_idle; |
9142 |
-+ bfqd->bfq_class_idle_last_service = 0; |
9143 |
-+ bfqd->bfq_max_budget_async_rq = bfq_max_budget_async_rq; |
9144 |
-+ bfqd->bfq_timeout[BLK_RW_ASYNC] = bfq_timeout_async; |
9145 |
-+ bfqd->bfq_timeout[BLK_RW_SYNC] = bfq_timeout_sync; |
9146 |
-+ |
9147 |
-+ bfqd->low_latency = true; |
9148 |
-+ |
9149 |
-+ bfqd->bfq_raising_coeff = 20; |
9150 |
-+ bfqd->bfq_raising_rt_max_time = msecs_to_jiffies(300); |
9151 |
-+ bfqd->bfq_raising_max_time = 0; |
9152 |
-+ bfqd->bfq_raising_min_idle_time = msecs_to_jiffies(2000); |
9153 |
-+ bfqd->bfq_raising_min_inter_arr_async = msecs_to_jiffies(500); |
9154 |
-+ bfqd->bfq_raising_max_softrt_rate = 7000; /* |
9155 |
-+ * Approximate rate required |
9156 |
-+ * to playback or record a |
9157 |
-+ * high-definition compressed |
9158 |
-+ * video. |
9159 |
-+ */ |
9160 |
-+ bfqd->raised_busy_queues = 0; |
9161 |
-+ |
9162 |
-+ /* Initially estimate the device's peak rate as the reference rate */ |
9163 |
-+ if (blk_queue_nonrot(bfqd->queue)) { |
9164 |
-+ bfqd->RT_prod = R_nonrot * T_nonrot; |
9165 |
-+ bfqd->peak_rate = R_nonrot; |
9166 |
-+ } else { |
9167 |
-+ bfqd->RT_prod = R_rot * T_rot; |
9168 |
-+ bfqd->peak_rate = R_rot; |
9169 |
-+ } |
9170 |
-+ |
9171 |
-+ return 0; |
9172 |
-+} |
9173 |
-+ |
9174 |
-+static void bfq_slab_kill(void) |
9175 |
-+{ |
9176 |
-+ if (bfq_pool != NULL) |
9177 |
-+ kmem_cache_destroy(bfq_pool); |
9178 |
-+} |
9179 |
-+ |
9180 |
-+static int __init bfq_slab_setup(void) |
9181 |
-+{ |
9182 |
-+ bfq_pool = KMEM_CACHE(bfq_queue, 0); |
9183 |
-+ if (bfq_pool == NULL) |
9184 |
-+ return -ENOMEM; |
9185 |
-+ return 0; |
9186 |
-+} |
9187 |
-+ |
9188 |
-+static ssize_t bfq_var_show(unsigned int var, char *page) |
9189 |
-+{ |
9190 |
-+ return sprintf(page, "%d\n", var); |
9191 |
-+} |
9192 |
-+ |
9193 |
-+static ssize_t bfq_var_store(unsigned long *var, const char *page, size_t count) |
9194 |
-+{ |
9195 |
-+ unsigned long new_val; |
9196 |
-+ int ret = kstrtoul(page, 10, &new_val); |
9197 |
-+ |
9198 |
-+ if (ret == 0) |
9199 |
-+ *var = new_val; |
9200 |
-+ |
9201 |
-+ return count; |
9202 |
-+} |
9203 |
-+ |
9204 |
-+static ssize_t bfq_raising_max_time_show(struct elevator_queue *e, char *page) |
9205 |
-+{ |
9206 |
-+ struct bfq_data *bfqd = e->elevator_data; |
9207 |
-+ return sprintf(page, "%d\n", bfqd->bfq_raising_max_time > 0 ? |
9208 |
-+ jiffies_to_msecs(bfqd->bfq_raising_max_time) : |
9209 |
-+ jiffies_to_msecs(bfq_wrais_duration(bfqd))); |
9210 |
-+} |
9211 |
-+ |
9212 |
-+static ssize_t bfq_weights_show(struct elevator_queue *e, char *page) |
9213 |
-+{ |
9214 |
-+ struct bfq_queue *bfqq; |
9215 |
-+ struct bfq_data *bfqd = e->elevator_data; |
9216 |
-+ ssize_t num_char = 0; |
9217 |
-+ |
9218 |
-+ num_char += sprintf(page + num_char, "Tot reqs queued %d\n\n", |
9219 |
-+ bfqd->queued); |
9220 |
-+ |
9221 |
-+ spin_lock_irq(bfqd->queue->queue_lock); |
9222 |
-+ |
9223 |
-+ num_char += sprintf(page + num_char, "Active:\n"); |
9224 |
-+ list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list) { |
9225 |
-+ num_char += sprintf(page + num_char, |
9226 |
-+ "pid%d: weight %hu, nr_queued %d %d," |
9227 |
-+ " dur %d/%u\n", |
9228 |
-+ bfqq->pid, |
9229 |
-+ bfqq->entity.weight, |
9230 |
-+ bfqq->queued[0], |
9231 |
-+ bfqq->queued[1], |
9232 |
-+ jiffies_to_msecs(jiffies - |
9233 |
-+ bfqq->last_rais_start_finish), |
9234 |
-+ jiffies_to_msecs(bfqq->raising_cur_max_time)); |
9235 |
-+ } |
9236 |
-+ |
9237 |
-+ num_char += sprintf(page + num_char, "Idle:\n"); |
9238 |
-+ list_for_each_entry(bfqq, &bfqd->idle_list, bfqq_list) { |
9239 |
-+ num_char += sprintf(page + num_char, |
9240 |
-+ "pid%d: weight %hu, dur %d/%u\n", |
9241 |
-+ bfqq->pid, |
9242 |
-+ bfqq->entity.weight, |
9243 |
-+ jiffies_to_msecs(jiffies - |
9244 |
-+ bfqq->last_rais_start_finish), |
9245 |
-+ jiffies_to_msecs(bfqq->raising_cur_max_time)); |
9246 |
-+ } |
9247 |
-+ |
9248 |
-+ spin_unlock_irq(bfqd->queue->queue_lock); |
9249 |
-+ |
9250 |
-+ return num_char; |
9251 |
-+} |
9252 |
-+ |
9253 |
-+#define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
9254 |
-+static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
9255 |
-+{ \ |
9256 |
-+ struct bfq_data *bfqd = e->elevator_data; \ |
9257 |
-+ unsigned int __data = __VAR; \ |
9258 |
-+ if (__CONV) \ |
9259 |
-+ __data = jiffies_to_msecs(__data); \ |
9260 |
-+ return bfq_var_show(__data, (page)); \ |
9261 |
-+} |
9262 |
-+SHOW_FUNCTION(bfq_quantum_show, bfqd->bfq_quantum, 0); |
9263 |
-+SHOW_FUNCTION(bfq_fifo_expire_sync_show, bfqd->bfq_fifo_expire[1], 1); |
9264 |
-+SHOW_FUNCTION(bfq_fifo_expire_async_show, bfqd->bfq_fifo_expire[0], 1); |
9265 |
-+SHOW_FUNCTION(bfq_back_seek_max_show, bfqd->bfq_back_max, 0); |
9266 |
-+SHOW_FUNCTION(bfq_back_seek_penalty_show, bfqd->bfq_back_penalty, 0); |
9267 |
-+SHOW_FUNCTION(bfq_slice_idle_show, bfqd->bfq_slice_idle, 1); |
9268 |
-+SHOW_FUNCTION(bfq_max_budget_show, bfqd->bfq_user_max_budget, 0); |
9269 |
-+SHOW_FUNCTION(bfq_max_budget_async_rq_show, bfqd->bfq_max_budget_async_rq, 0); |
9270 |
-+SHOW_FUNCTION(bfq_timeout_sync_show, bfqd->bfq_timeout[BLK_RW_SYNC], 1); |
9271 |
-+SHOW_FUNCTION(bfq_timeout_async_show, bfqd->bfq_timeout[BLK_RW_ASYNC], 1); |
9272 |
-+SHOW_FUNCTION(bfq_low_latency_show, bfqd->low_latency, 0); |
9273 |
-+SHOW_FUNCTION(bfq_raising_coeff_show, bfqd->bfq_raising_coeff, 0); |
9274 |
-+SHOW_FUNCTION(bfq_raising_rt_max_time_show, bfqd->bfq_raising_rt_max_time, 1); |
9275 |
-+SHOW_FUNCTION(bfq_raising_min_idle_time_show, bfqd->bfq_raising_min_idle_time, |
9276 |
-+ 1); |
9277 |
-+SHOW_FUNCTION(bfq_raising_min_inter_arr_async_show, |
9278 |
-+ bfqd->bfq_raising_min_inter_arr_async, |
9279 |
-+ 1); |
9280 |
-+SHOW_FUNCTION(bfq_raising_max_softrt_rate_show, |
9281 |
-+ bfqd->bfq_raising_max_softrt_rate, 0); |
9282 |
-+#undef SHOW_FUNCTION |
9283 |
-+ |
9284 |
-+#define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ |
9285 |
-+static ssize_t \ |
9286 |
-+__FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
9287 |
-+{ \ |
9288 |
-+ struct bfq_data *bfqd = e->elevator_data; \ |
9289 |
-+ unsigned long uninitialized_var(__data); \ |
9290 |
-+ int ret = bfq_var_store(&__data, (page), count); \ |
9291 |
-+ if (__data < (MIN)) \ |
9292 |
-+ __data = (MIN); \ |
9293 |
-+ else if (__data > (MAX)) \ |
9294 |
-+ __data = (MAX); \ |
9295 |
-+ if (__CONV) \ |
9296 |
-+ *(__PTR) = msecs_to_jiffies(__data); \ |
9297 |
-+ else \ |
9298 |
-+ *(__PTR) = __data; \ |
9299 |
-+ return ret; \ |
9300 |
-+} |
9301 |
-+STORE_FUNCTION(bfq_quantum_store, &bfqd->bfq_quantum, 1, INT_MAX, 0); |
9302 |
-+STORE_FUNCTION(bfq_fifo_expire_sync_store, &bfqd->bfq_fifo_expire[1], 1, |
9303 |
-+ INT_MAX, 1); |
9304 |
-+STORE_FUNCTION(bfq_fifo_expire_async_store, &bfqd->bfq_fifo_expire[0], 1, |
9305 |
-+ INT_MAX, 1); |
9306 |
-+STORE_FUNCTION(bfq_back_seek_max_store, &bfqd->bfq_back_max, 0, INT_MAX, 0); |
9307 |
-+STORE_FUNCTION(bfq_back_seek_penalty_store, &bfqd->bfq_back_penalty, 1, |
9308 |
-+ INT_MAX, 0); |
9309 |
-+STORE_FUNCTION(bfq_slice_idle_store, &bfqd->bfq_slice_idle, 0, INT_MAX, 1); |
9310 |
-+STORE_FUNCTION(bfq_max_budget_async_rq_store, &bfqd->bfq_max_budget_async_rq, |
9311 |
-+ 1, INT_MAX, 0); |
9312 |
-+STORE_FUNCTION(bfq_timeout_async_store, &bfqd->bfq_timeout[BLK_RW_ASYNC], 0, |
9313 |
-+ INT_MAX, 1); |
9314 |
-+STORE_FUNCTION(bfq_raising_coeff_store, &bfqd->bfq_raising_coeff, 1, |
9315 |
-+ INT_MAX, 0); |
9316 |
-+STORE_FUNCTION(bfq_raising_max_time_store, &bfqd->bfq_raising_max_time, 0, |
9317 |
-+ INT_MAX, 1); |
9318 |
-+STORE_FUNCTION(bfq_raising_rt_max_time_store, &bfqd->bfq_raising_rt_max_time, 0, |
9319 |
-+ INT_MAX, 1); |
9320 |
-+STORE_FUNCTION(bfq_raising_min_idle_time_store, |
9321 |
-+ &bfqd->bfq_raising_min_idle_time, 0, INT_MAX, 1); |
9322 |
-+STORE_FUNCTION(bfq_raising_min_inter_arr_async_store, |
9323 |
-+ &bfqd->bfq_raising_min_inter_arr_async, 0, INT_MAX, 1); |
9324 |
-+STORE_FUNCTION(bfq_raising_max_softrt_rate_store, |
9325 |
-+ &bfqd->bfq_raising_max_softrt_rate, 0, INT_MAX, 0); |
9326 |
-+#undef STORE_FUNCTION |
9327 |
-+ |
9328 |
-+/* do nothing for the moment */ |
9329 |
-+static ssize_t bfq_weights_store(struct elevator_queue *e, |
9330 |
-+ const char *page, size_t count) |
9331 |
-+{ |
9332 |
-+ return count; |
9333 |
-+} |
9334 |
-+ |
9335 |
-+static inline unsigned long bfq_estimated_max_budget(struct bfq_data *bfqd) |
9336 |
-+{ |
9337 |
-+ u64 timeout = jiffies_to_msecs(bfqd->bfq_timeout[BLK_RW_SYNC]); |
9338 |
-+ |
9339 |
-+ if (bfqd->peak_rate_samples >= BFQ_PEAK_RATE_SAMPLES) |
9340 |
-+ return bfq_calc_max_budget(bfqd->peak_rate, timeout); |
9341 |
-+ else |
9342 |
-+ return bfq_default_max_budget; |
9343 |
-+} |
9344 |
-+ |
9345 |
-+static ssize_t bfq_max_budget_store(struct elevator_queue *e, |
9346 |
-+ const char *page, size_t count) |
9347 |
-+{ |
9348 |
-+ struct bfq_data *bfqd = e->elevator_data; |
9349 |
-+ unsigned long uninitialized_var(__data); |
9350 |
-+ int ret = bfq_var_store(&__data, (page), count); |
9351 |
-+ |
9352 |
-+ if (__data == 0) |
9353 |
-+ bfqd->bfq_max_budget = bfq_estimated_max_budget(bfqd); |
9354 |
-+ else { |
9355 |
-+ if (__data > INT_MAX) |
9356 |
-+ __data = INT_MAX; |
9357 |
-+ bfqd->bfq_max_budget = __data; |
9358 |
-+ } |
9359 |
-+ |
9360 |
-+ bfqd->bfq_user_max_budget = __data; |
9361 |
-+ |
9362 |
-+ return ret; |
9363 |
-+} |
9364 |
-+ |
9365 |
-+static ssize_t bfq_timeout_sync_store(struct elevator_queue *e, |
9366 |
-+ const char *page, size_t count) |
9367 |
-+{ |
9368 |
-+ struct bfq_data *bfqd = e->elevator_data; |
9369 |
-+ unsigned long uninitialized_var(__data); |
9370 |
-+ int ret = bfq_var_store(&__data, (page), count); |
9371 |
-+ |
9372 |
-+ if (__data < 1) |
9373 |
-+ __data = 1; |
9374 |
-+ else if (__data > INT_MAX) |
9375 |
-+ __data = INT_MAX; |
9376 |
-+ |
9377 |
-+ bfqd->bfq_timeout[BLK_RW_SYNC] = msecs_to_jiffies(__data); |
9378 |
-+ if (bfqd->bfq_user_max_budget == 0) |
9379 |
-+ bfqd->bfq_max_budget = bfq_estimated_max_budget(bfqd); |
9380 |
-+ |
9381 |
-+ return ret; |
9382 |
-+} |
9383 |
-+ |
9384 |
-+static ssize_t bfq_low_latency_store(struct elevator_queue *e, |
9385 |
-+ const char *page, size_t count) |
9386 |
-+{ |
9387 |
-+ struct bfq_data *bfqd = e->elevator_data; |
9388 |
-+ unsigned long uninitialized_var(__data); |
9389 |
-+ int ret = bfq_var_store(&__data, (page), count); |
9390 |
-+ |
9391 |
-+ if (__data > 1) |
9392 |
-+ __data = 1; |
9393 |
-+ if (__data == 0 && bfqd->low_latency != 0) |
9394 |
-+ bfq_end_raising(bfqd); |
9395 |
-+ bfqd->low_latency = __data; |
9396 |
-+ |
9397 |
-+ return ret; |
9398 |
-+} |
9399 |
-+ |
9400 |
-+#define BFQ_ATTR(name) \ |
9401 |
-+ __ATTR(name, S_IRUGO|S_IWUSR, bfq_##name##_show, bfq_##name##_store) |
9402 |
-+ |
9403 |
-+static struct elv_fs_entry bfq_attrs[] = { |
9404 |
-+ BFQ_ATTR(quantum), |
9405 |
-+ BFQ_ATTR(fifo_expire_sync), |
9406 |
-+ BFQ_ATTR(fifo_expire_async), |
9407 |
-+ BFQ_ATTR(back_seek_max), |
9408 |
-+ BFQ_ATTR(back_seek_penalty), |
9409 |
-+ BFQ_ATTR(slice_idle), |
9410 |
-+ BFQ_ATTR(max_budget), |
9411 |
-+ BFQ_ATTR(max_budget_async_rq), |
9412 |
-+ BFQ_ATTR(timeout_sync), |
9413 |
-+ BFQ_ATTR(timeout_async), |
9414 |
-+ BFQ_ATTR(low_latency), |
9415 |
-+ BFQ_ATTR(raising_coeff), |
9416 |
-+ BFQ_ATTR(raising_max_time), |
9417 |
-+ BFQ_ATTR(raising_rt_max_time), |
9418 |
-+ BFQ_ATTR(raising_min_idle_time), |
9419 |
-+ BFQ_ATTR(raising_min_inter_arr_async), |
9420 |
-+ BFQ_ATTR(raising_max_softrt_rate), |
9421 |
-+ BFQ_ATTR(weights), |
9422 |
-+ __ATTR_NULL |
9423 |
-+}; |
9424 |
-+ |
9425 |
-+static struct elevator_type iosched_bfq = { |
9426 |
-+ .ops = { |
9427 |
-+ .elevator_merge_fn = bfq_merge, |
9428 |
-+ .elevator_merged_fn = bfq_merged_request, |
9429 |
-+ .elevator_merge_req_fn = bfq_merged_requests, |
9430 |
-+ .elevator_allow_merge_fn = bfq_allow_merge, |
9431 |
-+ .elevator_dispatch_fn = bfq_dispatch_requests, |
9432 |
-+ .elevator_add_req_fn = bfq_insert_request, |
9433 |
-+ .elevator_activate_req_fn = bfq_activate_request, |
9434 |
-+ .elevator_deactivate_req_fn = bfq_deactivate_request, |
9435 |
-+ .elevator_completed_req_fn = bfq_completed_request, |
9436 |
-+ .elevator_former_req_fn = elv_rb_former_request, |
9437 |
-+ .elevator_latter_req_fn = elv_rb_latter_request, |
9438 |
-+ .elevator_init_icq_fn = bfq_init_icq, |
9439 |
-+ .elevator_exit_icq_fn = bfq_exit_icq, |
9440 |
-+ .elevator_set_req_fn = bfq_set_request, |
9441 |
-+ .elevator_put_req_fn = bfq_put_request, |
9442 |
-+ .elevator_may_queue_fn = bfq_may_queue, |
9443 |
-+ .elevator_init_fn = bfq_init_queue, |
9444 |
-+ .elevator_exit_fn = bfq_exit_queue, |
9445 |
-+ }, |
9446 |
-+ .icq_size = sizeof(struct bfq_io_cq), |
9447 |
-+ .icq_align = __alignof__(struct bfq_io_cq), |
9448 |
-+ .elevator_attrs = bfq_attrs, |
9449 |
-+ .elevator_name = "bfq", |
9450 |
-+ .elevator_owner = THIS_MODULE, |
9451 |
-+}; |
9452 |
-+ |
9453 |
-+static int __init bfq_init(void) |
9454 |
-+{ |
9455 |
-+ /* |
9456 |
-+ * Can be 0 on HZ < 1000 setups. |
9457 |
-+ */ |
9458 |
-+ if (bfq_slice_idle == 0) |
9459 |
-+ bfq_slice_idle = 1; |
9460 |
-+ |
9461 |
-+ if (bfq_timeout_async == 0) |
9462 |
-+ bfq_timeout_async = 1; |
9463 |
-+ |
9464 |
-+ if (bfq_slab_setup()) |
9465 |
-+ return -ENOMEM; |
9466 |
-+ |
9467 |
-+ elv_register(&iosched_bfq); |
9468 |
-+ pr_info("BFQ I/O-scheduler version: v7r2"); |
9469 |
-+ |
9470 |
-+ return 0; |
9471 |
-+} |
9472 |
-+ |
9473 |
-+static void __exit bfq_exit(void) |
9474 |
-+{ |
9475 |
-+ elv_unregister(&iosched_bfq); |
9476 |
-+ bfq_slab_kill(); |
9477 |
-+} |
9478 |
-+ |
9479 |
-+module_init(bfq_init); |
9480 |
-+module_exit(bfq_exit); |
9481 |
-+ |
9482 |
-+MODULE_AUTHOR("Fabio Checconi, Paolo Valente"); |
9483 |
-diff --git a/block/bfq-sched.c b/block/bfq-sched.c |
9484 |
-new file mode 100644 |
9485 |
-index 0000000..999b475 |
9486 |
---- /dev/null |
9487 |
-+++ b/block/bfq-sched.c |
9488 |
-@@ -0,0 +1,1078 @@ |
9489 |
-+/* |
9490 |
-+ * BFQ: Hierarchical B-WF2Q+ scheduler. |
9491 |
-+ * |
9492 |
-+ * Based on ideas and code from CFQ: |
9493 |
-+ * Copyright (C) 2003 Jens Axboe <axboe@××××××.dk> |
9494 |
-+ * |
9495 |
-+ * Copyright (C) 2008 Fabio Checconi <fabio@×××××××××××××.it> |
9496 |
-+ * Paolo Valente <paolo.valente@×××××××.it> |
9497 |
-+ * |
9498 |
-+ * Copyright (C) 2010 Paolo Valente <paolo.valente@×××××××.it> |
9499 |
-+ */ |
9500 |
-+ |
9501 |
-+#ifdef CONFIG_CGROUP_BFQIO |
9502 |
-+#define for_each_entity(entity) \ |
9503 |
-+ for (; entity != NULL; entity = entity->parent) |
9504 |
-+ |
9505 |
-+#define for_each_entity_safe(entity, parent) \ |
9506 |
-+ for (; entity && ({ parent = entity->parent; 1; }); entity = parent) |
9507 |
-+ |
9508 |
-+static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd, |
9509 |
-+ int extract, |
9510 |
-+ struct bfq_data *bfqd); |
9511 |
-+ |
9512 |
-+static inline void bfq_update_budget(struct bfq_entity *next_in_service) |
9513 |
-+{ |
9514 |
-+ struct bfq_entity *bfqg_entity; |
9515 |
-+ struct bfq_group *bfqg; |
9516 |
-+ struct bfq_sched_data *group_sd; |
9517 |
-+ |
9518 |
-+ BUG_ON(next_in_service == NULL); |
9519 |
-+ |
9520 |
-+ group_sd = next_in_service->sched_data; |
9521 |
-+ |
9522 |
-+ bfqg = container_of(group_sd, struct bfq_group, sched_data); |
9523 |
-+ /* |
9524 |
-+ * bfq_group's my_entity field is not NULL only if the group |
9525 |
-+ * is not the root group. We must not touch the root entity |
9526 |
-+ * as it must never become an in-service entity. |
9527 |
-+ */ |
9528 |
-+ bfqg_entity = bfqg->my_entity; |
9529 |
-+ if (bfqg_entity != NULL) |
9530 |
-+ bfqg_entity->budget = next_in_service->budget; |
9531 |
-+} |
9532 |
-+ |
9533 |
-+static int bfq_update_next_in_service(struct bfq_sched_data *sd) |
9534 |
-+{ |
9535 |
-+ struct bfq_entity *next_in_service; |
9536 |
-+ |
9537 |
-+ if (sd->in_service_entity != NULL) |
9538 |
-+ /* will update/requeue at the end of service */ |
9539 |
-+ return 0; |
9540 |
-+ |
9541 |
-+ /* |
9542 |
-+ * NOTE: this can be improved in many ways, such as returning |
9543 |
-+ * 1 (and thus propagating upwards the update) only when the |
9544 |
-+ * budget changes, or caching the bfqq that will be scheduled |
9545 |
-+ * next from this subtree. By now we worry more about |
9546 |
-+ * correctness than about performance... |
9547 |
-+ */ |
9548 |
-+ next_in_service = bfq_lookup_next_entity(sd, 0, NULL); |
9549 |
-+ sd->next_in_service = next_in_service; |
9550 |
-+ |
9551 |
-+ if (next_in_service != NULL) |
9552 |
-+ bfq_update_budget(next_in_service); |
9553 |
-+ |
9554 |
-+ return 1; |
9555 |
-+} |
9556 |
-+ |
9557 |
-+static inline void bfq_check_next_in_service(struct bfq_sched_data *sd, |
9558 |
-+ struct bfq_entity *entity) |
9559 |
-+{ |
9560 |
-+ BUG_ON(sd->next_in_service != entity); |
9561 |
-+} |
9562 |
-+#else |
9563 |
-+#define for_each_entity(entity) \ |
9564 |
-+ for (; entity != NULL; entity = NULL) |
9565 |
-+ |
9566 |
-+#define for_each_entity_safe(entity, parent) \ |
9567 |
-+ for (parent = NULL; entity != NULL; entity = parent) |
9568 |
-+ |
9569 |
-+static inline int bfq_update_next_in_service(struct bfq_sched_data *sd) |
9570 |
-+{ |
9571 |
-+ return 0; |
9572 |
-+} |
9573 |
-+ |
9574 |
-+static inline void bfq_check_next_in_service(struct bfq_sched_data *sd, |
9575 |
-+ struct bfq_entity *entity) |
9576 |
-+{ |
9577 |
-+} |
9578 |
-+ |
9579 |
-+static inline void bfq_update_budget(struct bfq_entity *next_in_service) |
9580 |
-+{ |
9581 |
-+} |
9582 |
-+#endif |
9583 |
-+ |
9584 |
-+/* |
9585 |
-+ * Shift for timestamp calculations. This actually limits the maximum |
9586 |
-+ * service allowed in one timestamp delta (small shift values increase it), |
9587 |
-+ * the maximum total weight that can be used for the queues in the system |
9588 |
-+ * (big shift values increase it), and the period of virtual time wraparounds. |
9589 |
-+ */ |
9590 |
-+#define WFQ_SERVICE_SHIFT 22 |
9591 |
-+ |
9592 |
-+/** |
9593 |
-+ * bfq_gt - compare two timestamps. |
9594 |
-+ * @a: first ts. |
9595 |
-+ * @b: second ts. |
9596 |
-+ * |
9597 |
-+ * Return @a > @b, dealing with wrapping correctly. |
9598 |
-+ */ |
9599 |
-+static inline int bfq_gt(u64 a, u64 b) |
9600 |
-+{ |
9601 |
-+ return (s64)(a - b) > 0; |
9602 |
-+} |
9603 |
-+ |
9604 |
-+static inline struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity) |
9605 |
-+{ |
9606 |
-+ struct bfq_queue *bfqq = NULL; |
9607 |
-+ |
9608 |
-+ BUG_ON(entity == NULL); |
9609 |
-+ |
9610 |
-+ if (entity->my_sched_data == NULL) |
9611 |
-+ bfqq = container_of(entity, struct bfq_queue, entity); |
9612 |
-+ |
9613 |
-+ return bfqq; |
9614 |
-+} |
9615 |
-+ |
9616 |
-+ |
9617 |
-+/** |
9618 |
-+ * bfq_delta - map service into the virtual time domain. |
9619 |
-+ * @service: amount of service. |
9620 |
-+ * @weight: scale factor (weight of an entity or weight sum). |
9621 |
-+ */ |
9622 |
-+static inline u64 bfq_delta(unsigned long service, |
9623 |
-+ unsigned long weight) |
9624 |
-+{ |
9625 |
-+ u64 d = (u64)service << WFQ_SERVICE_SHIFT; |
9626 |
-+ |
9627 |
-+ do_div(d, weight); |
9628 |
-+ return d; |
9629 |
-+} |
9630 |
-+ |
9631 |
-+/** |
9632 |
-+ * bfq_calc_finish - assign the finish time to an entity. |
9633 |
-+ * @entity: the entity to act upon. |
9634 |
-+ * @service: the service to be charged to the entity. |
9635 |
-+ */ |
9636 |
-+static inline void bfq_calc_finish(struct bfq_entity *entity, |
9637 |
-+ unsigned long service) |
9638 |
-+{ |
9639 |
-+ struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); |
9640 |
-+ |
9641 |
-+ BUG_ON(entity->weight == 0); |
9642 |
-+ |
9643 |
-+ entity->finish = entity->start + |
9644 |
-+ bfq_delta(service, entity->weight); |
9645 |
-+ |
9646 |
-+ if (bfqq != NULL) { |
9647 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, |
9648 |
-+ "calc_finish: serv %lu, w %d", |
9649 |
-+ service, entity->weight); |
9650 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, |
9651 |
-+ "calc_finish: start %llu, finish %llu, delta %llu", |
9652 |
-+ entity->start, entity->finish, |
9653 |
-+ bfq_delta(service, entity->weight)); |
9654 |
-+ } |
9655 |
-+} |
9656 |
-+ |
9657 |
-+/** |
9658 |
-+ * bfq_entity_of - get an entity from a node. |
9659 |
-+ * @node: the node field of the entity. |
9660 |
-+ * |
9661 |
-+ * Convert a node pointer to the relative entity. This is used only |
9662 |
-+ * to simplify the logic of some functions and not as the generic |
9663 |
-+ * conversion mechanism because, e.g., in the tree walking functions, |
9664 |
-+ * the check for a %NULL value would be redundant. |
9665 |
-+ */ |
9666 |
-+static inline struct bfq_entity *bfq_entity_of(struct rb_node *node) |
9667 |
-+{ |
9668 |
-+ struct bfq_entity *entity = NULL; |
9669 |
-+ |
9670 |
-+ if (node != NULL) |
9671 |
-+ entity = rb_entry(node, struct bfq_entity, rb_node); |
9672 |
-+ |
9673 |
-+ return entity; |
9674 |
-+} |
9675 |
-+ |
9676 |
-+/** |
9677 |
-+ * bfq_extract - remove an entity from a tree. |
9678 |
-+ * @root: the tree root. |
9679 |
-+ * @entity: the entity to remove. |
9680 |
-+ */ |
9681 |
-+static inline void bfq_extract(struct rb_root *root, |
9682 |
-+ struct bfq_entity *entity) |
9683 |
-+{ |
9684 |
-+ BUG_ON(entity->tree != root); |
9685 |
-+ |
9686 |
-+ entity->tree = NULL; |
9687 |
-+ rb_erase(&entity->rb_node, root); |
9688 |
-+} |
9689 |
-+ |
9690 |
-+/** |
9691 |
-+ * bfq_idle_extract - extract an entity from the idle tree. |
9692 |
-+ * @st: the service tree of the owning @entity. |
9693 |
-+ * @entity: the entity being removed. |
9694 |
-+ */ |
9695 |
-+static void bfq_idle_extract(struct bfq_service_tree *st, |
9696 |
-+ struct bfq_entity *entity) |
9697 |
-+{ |
9698 |
-+ struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); |
9699 |
-+ struct rb_node *next; |
9700 |
-+ |
9701 |
-+ BUG_ON(entity->tree != &st->idle); |
9702 |
-+ |
9703 |
-+ if (entity == st->first_idle) { |
9704 |
-+ next = rb_next(&entity->rb_node); |
9705 |
-+ st->first_idle = bfq_entity_of(next); |
9706 |
-+ } |
9707 |
-+ |
9708 |
-+ if (entity == st->last_idle) { |
9709 |
-+ next = rb_prev(&entity->rb_node); |
9710 |
-+ st->last_idle = bfq_entity_of(next); |
9711 |
-+ } |
9712 |
-+ |
9713 |
-+ bfq_extract(&st->idle, entity); |
9714 |
-+ |
9715 |
-+ if (bfqq != NULL) |
9716 |
-+ list_del(&bfqq->bfqq_list); |
9717 |
-+} |
9718 |
-+ |
9719 |
-+/** |
9720 |
-+ * bfq_insert - generic tree insertion. |
9721 |
-+ * @root: tree root. |
9722 |
-+ * @entity: entity to insert. |
9723 |
-+ * |
9724 |
-+ * This is used for the idle and the active tree, since they are both |
9725 |
-+ * ordered by finish time. |
9726 |
-+ */ |
9727 |
-+static void bfq_insert(struct rb_root *root, struct bfq_entity *entity) |
9728 |
-+{ |
9729 |
-+ struct bfq_entity *entry; |
9730 |
-+ struct rb_node **node = &root->rb_node; |
9731 |
-+ struct rb_node *parent = NULL; |
9732 |
-+ |
9733 |
-+ BUG_ON(entity->tree != NULL); |
9734 |
-+ |
9735 |
-+ while (*node != NULL) { |
9736 |
-+ parent = *node; |
9737 |
-+ entry = rb_entry(parent, struct bfq_entity, rb_node); |
9738 |
-+ |
9739 |
-+ if (bfq_gt(entry->finish, entity->finish)) |
9740 |
-+ node = &parent->rb_left; |
9741 |
-+ else |
9742 |
-+ node = &parent->rb_right; |
9743 |
-+ } |
9744 |
-+ |
9745 |
-+ rb_link_node(&entity->rb_node, parent, node); |
9746 |
-+ rb_insert_color(&entity->rb_node, root); |
9747 |
-+ |
9748 |
-+ entity->tree = root; |
9749 |
-+} |
9750 |
-+ |
9751 |
-+/** |
9752 |
-+ * bfq_update_min - update the min_start field of a entity. |
9753 |
-+ * @entity: the entity to update. |
9754 |
-+ * @node: one of its children. |
9755 |
-+ * |
9756 |
-+ * This function is called when @entity may store an invalid value for |
9757 |
-+ * min_start due to updates to the active tree. The function assumes |
9758 |
-+ * that the subtree rooted at @node (which may be its left or its right |
9759 |
-+ * child) has a valid min_start value. |
9760 |
-+ */ |
9761 |
-+static inline void bfq_update_min(struct bfq_entity *entity, |
9762 |
-+ struct rb_node *node) |
9763 |
-+{ |
9764 |
-+ struct bfq_entity *child; |
9765 |
-+ |
9766 |
-+ if (node != NULL) { |
9767 |
-+ child = rb_entry(node, struct bfq_entity, rb_node); |
9768 |
-+ if (bfq_gt(entity->min_start, child->min_start)) |
9769 |
-+ entity->min_start = child->min_start; |
9770 |
-+ } |
9771 |
-+} |
9772 |
-+ |
9773 |
-+/** |
9774 |
-+ * bfq_update_active_node - recalculate min_start. |
9775 |
-+ * @node: the node to update. |
9776 |
-+ * |
9777 |
-+ * @node may have changed position or one of its children may have moved, |
9778 |
-+ * this function updates its min_start value. The left and right subtrees |
9779 |
-+ * are assumed to hold a correct min_start value. |
9780 |
-+ */ |
9781 |
-+static inline void bfq_update_active_node(struct rb_node *node) |
9782 |
-+{ |
9783 |
-+ struct bfq_entity *entity = rb_entry(node, struct bfq_entity, rb_node); |
9784 |
-+ |
9785 |
-+ entity->min_start = entity->start; |
9786 |
-+ bfq_update_min(entity, node->rb_right); |
9787 |
-+ bfq_update_min(entity, node->rb_left); |
9788 |
-+} |
9789 |
-+ |
9790 |
-+/** |
9791 |
-+ * bfq_update_active_tree - update min_start for the whole active tree. |
9792 |
-+ * @node: the starting node. |
9793 |
-+ * |
9794 |
-+ * @node must be the deepest modified node after an update. This function |
9795 |
-+ * updates its min_start using the values held by its children, assuming |
9796 |
-+ * that they did not change, and then updates all the nodes that may have |
9797 |
-+ * changed in the path to the root. The only nodes that may have changed |
9798 |
-+ * are the ones in the path or their siblings. |
9799 |
-+ */ |
9800 |
-+static void bfq_update_active_tree(struct rb_node *node) |
9801 |
-+{ |
9802 |
-+ struct rb_node *parent; |
9803 |
-+ |
9804 |
-+up: |
9805 |
-+ bfq_update_active_node(node); |
9806 |
-+ |
9807 |
-+ parent = rb_parent(node); |
9808 |
-+ if (parent == NULL) |
9809 |
-+ return; |
9810 |
-+ |
9811 |
-+ if (node == parent->rb_left && parent->rb_right != NULL) |
9812 |
-+ bfq_update_active_node(parent->rb_right); |
9813 |
-+ else if (parent->rb_left != NULL) |
9814 |
-+ bfq_update_active_node(parent->rb_left); |
9815 |
-+ |
9816 |
-+ node = parent; |
9817 |
-+ goto up; |
9818 |
-+} |
9819 |
-+ |
9820 |
-+/** |
9821 |
-+ * bfq_active_insert - insert an entity in the active tree of its group/device. |
9822 |
-+ * @st: the service tree of the entity. |
9823 |
-+ * @entity: the entity being inserted. |
9824 |
-+ * |
9825 |
-+ * The active tree is ordered by finish time, but an extra key is kept |
9826 |
-+ * per each node, containing the minimum value for the start times of |
9827 |
-+ * its children (and the node itself), so it's possible to search for |
9828 |
-+ * the eligible node with the lowest finish time in logarithmic time. |
9829 |
-+ */ |
9830 |
-+static void bfq_active_insert(struct bfq_service_tree *st, |
9831 |
-+ struct bfq_entity *entity) |
9832 |
-+{ |
9833 |
-+ struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); |
9834 |
-+ struct rb_node *node = &entity->rb_node; |
9835 |
-+ |
9836 |
-+ bfq_insert(&st->active, entity); |
9837 |
-+ |
9838 |
-+ if (node->rb_left != NULL) |
9839 |
-+ node = node->rb_left; |
9840 |
-+ else if (node->rb_right != NULL) |
9841 |
-+ node = node->rb_right; |
9842 |
-+ |
9843 |
-+ bfq_update_active_tree(node); |
9844 |
-+ |
9845 |
-+ if (bfqq != NULL) |
9846 |
-+ list_add(&bfqq->bfqq_list, &bfqq->bfqd->active_list); |
9847 |
-+} |
9848 |
-+ |
9849 |
-+/** |
9850 |
-+ * bfq_ioprio_to_weight - calc a weight from an ioprio. |
9851 |
-+ * @ioprio: the ioprio value to convert. |
9852 |
-+ */ |
9853 |
-+static unsigned short bfq_ioprio_to_weight(int ioprio) |
9854 |
-+{ |
9855 |
-+ WARN_ON(ioprio < 0 || ioprio >= IOPRIO_BE_NR); |
9856 |
-+ return IOPRIO_BE_NR - ioprio; |
9857 |
-+} |
9858 |
-+ |
9859 |
-+/** |
9860 |
-+ * bfq_weight_to_ioprio - calc an ioprio from a weight. |
9861 |
-+ * @weight: the weight value to convert. |
9862 |
-+ * |
9863 |
-+ * To preserve as mush as possible the old only-ioprio user interface, |
9864 |
-+ * 0 is used as an escape ioprio value for weights (numerically) equal or |
9865 |
-+ * larger than IOPRIO_BE_NR |
9866 |
-+ */ |
9867 |
-+static unsigned short bfq_weight_to_ioprio(int weight) |
9868 |
-+{ |
9869 |
-+ WARN_ON(weight < BFQ_MIN_WEIGHT || weight > BFQ_MAX_WEIGHT); |
9870 |
-+ return IOPRIO_BE_NR - weight < 0 ? 0 : IOPRIO_BE_NR - weight; |
9871 |
-+} |
9872 |
-+ |
9873 |
-+static inline void bfq_get_entity(struct bfq_entity *entity) |
9874 |
-+{ |
9875 |
-+ struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); |
9876 |
-+ struct bfq_sched_data *sd; |
9877 |
-+ |
9878 |
-+ if (bfqq != NULL) { |
9879 |
-+ sd = entity->sched_data; |
9880 |
-+ atomic_inc(&bfqq->ref); |
9881 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, "get_entity: %p %d", |
9882 |
-+ bfqq, atomic_read(&bfqq->ref)); |
9883 |
-+ } |
9884 |
-+} |
9885 |
-+ |
9886 |
-+/** |
9887 |
-+ * bfq_find_deepest - find the deepest node that an extraction can modify. |
9888 |
-+ * @node: the node being removed. |
9889 |
-+ * |
9890 |
-+ * Do the first step of an extraction in an rb tree, looking for the |
9891 |
-+ * node that will replace @node, and returning the deepest node that |
9892 |
-+ * the following modifications to the tree can touch. If @node is the |
9893 |
-+ * last node in the tree return %NULL. |
9894 |
-+ */ |
9895 |
-+static struct rb_node *bfq_find_deepest(struct rb_node *node) |
9896 |
-+{ |
9897 |
-+ struct rb_node *deepest; |
9898 |
-+ |
9899 |
-+ if (node->rb_right == NULL && node->rb_left == NULL) |
9900 |
-+ deepest = rb_parent(node); |
9901 |
-+ else if (node->rb_right == NULL) |
9902 |
-+ deepest = node->rb_left; |
9903 |
-+ else if (node->rb_left == NULL) |
9904 |
-+ deepest = node->rb_right; |
9905 |
-+ else { |
9906 |
-+ deepest = rb_next(node); |
9907 |
-+ if (deepest->rb_right != NULL) |
9908 |
-+ deepest = deepest->rb_right; |
9909 |
-+ else if (rb_parent(deepest) != node) |
9910 |
-+ deepest = rb_parent(deepest); |
9911 |
-+ } |
9912 |
-+ |
9913 |
-+ return deepest; |
9914 |
-+} |
9915 |
-+ |
9916 |
-+/** |
9917 |
-+ * bfq_active_extract - remove an entity from the active tree. |
9918 |
-+ * @st: the service_tree containing the tree. |
9919 |
-+ * @entity: the entity being removed. |
9920 |
-+ */ |
9921 |
-+static void bfq_active_extract(struct bfq_service_tree *st, |
9922 |
-+ struct bfq_entity *entity) |
9923 |
-+{ |
9924 |
-+ struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); |
9925 |
-+ struct rb_node *node; |
9926 |
-+ |
9927 |
-+ node = bfq_find_deepest(&entity->rb_node); |
9928 |
-+ bfq_extract(&st->active, entity); |
9929 |
-+ |
9930 |
-+ if (node != NULL) |
9931 |
-+ bfq_update_active_tree(node); |
9932 |
-+ |
9933 |
-+ if (bfqq != NULL) |
9934 |
-+ list_del(&bfqq->bfqq_list); |
9935 |
-+} |
9936 |
-+ |
9937 |
-+/** |
9938 |
-+ * bfq_idle_insert - insert an entity into the idle tree. |
9939 |
-+ * @st: the service tree containing the tree. |
9940 |
-+ * @entity: the entity to insert. |
9941 |
-+ */ |
9942 |
-+static void bfq_idle_insert(struct bfq_service_tree *st, |
9943 |
-+ struct bfq_entity *entity) |
9944 |
-+{ |
9945 |
-+ struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); |
9946 |
-+ struct bfq_entity *first_idle = st->first_idle; |
9947 |
-+ struct bfq_entity *last_idle = st->last_idle; |
9948 |
-+ |
9949 |
-+ if (first_idle == NULL || bfq_gt(first_idle->finish, entity->finish)) |
9950 |
-+ st->first_idle = entity; |
9951 |
-+ if (last_idle == NULL || bfq_gt(entity->finish, last_idle->finish)) |
9952 |
-+ st->last_idle = entity; |
9953 |
-+ |
9954 |
-+ bfq_insert(&st->idle, entity); |
9955 |
-+ |
9956 |
-+ if (bfqq != NULL) |
9957 |
-+ list_add(&bfqq->bfqq_list, &bfqq->bfqd->idle_list); |
9958 |
-+} |
9959 |
-+ |
9960 |
-+/** |
9961 |
-+ * bfq_forget_entity - remove an entity from the wfq trees. |
9962 |
-+ * @st: the service tree. |
9963 |
-+ * @entity: the entity being removed. |
9964 |
-+ * |
9965 |
-+ * Update the device status and forget everything about @entity, putting |
9966 |
-+ * the device reference to it, if it is a queue. Entities belonging to |
9967 |
-+ * groups are not refcounted. |
9968 |
-+ */ |
9969 |
-+static void bfq_forget_entity(struct bfq_service_tree *st, |
9970 |
-+ struct bfq_entity *entity) |
9971 |
-+{ |
9972 |
-+ struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); |
9973 |
-+ struct bfq_sched_data *sd; |
9974 |
-+ |
9975 |
-+ BUG_ON(!entity->on_st); |
9976 |
-+ |
9977 |
-+ entity->on_st = 0; |
9978 |
-+ st->wsum -= entity->weight; |
9979 |
-+ if (bfqq != NULL) { |
9980 |
-+ sd = entity->sched_data; |
9981 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, "forget_entity: %p %d", |
9982 |
-+ bfqq, atomic_read(&bfqq->ref)); |
9983 |
-+ bfq_put_queue(bfqq); |
9984 |
-+ } |
9985 |
-+} |
9986 |
-+ |
9987 |
-+/** |
9988 |
-+ * bfq_put_idle_entity - release the idle tree ref of an entity. |
9989 |
-+ * @st: service tree for the entity. |
9990 |
-+ * @entity: the entity being released. |
9991 |
-+ */ |
9992 |
-+static void bfq_put_idle_entity(struct bfq_service_tree *st, |
9993 |
-+ struct bfq_entity *entity) |
9994 |
-+{ |
9995 |
-+ bfq_idle_extract(st, entity); |
9996 |
-+ bfq_forget_entity(st, entity); |
9997 |
-+} |
9998 |
-+ |
9999 |
-+/** |
10000 |
-+ * bfq_forget_idle - update the idle tree if necessary. |
10001 |
-+ * @st: the service tree to act upon. |
10002 |
-+ * |
10003 |
-+ * To preserve the global O(log N) complexity we only remove one entry here; |
10004 |
-+ * as the idle tree will not grow indefinitely this can be done safely. |
10005 |
-+ */ |
10006 |
-+static void bfq_forget_idle(struct bfq_service_tree *st) |
10007 |
-+{ |
10008 |
-+ struct bfq_entity *first_idle = st->first_idle; |
10009 |
-+ struct bfq_entity *last_idle = st->last_idle; |
10010 |
-+ |
10011 |
-+ if (RB_EMPTY_ROOT(&st->active) && last_idle != NULL && |
10012 |
-+ !bfq_gt(last_idle->finish, st->vtime)) { |
10013 |
-+ /* |
10014 |
-+ * Forget the whole idle tree, increasing the vtime past |
10015 |
-+ * the last finish time of idle entities. |
10016 |
-+ */ |
10017 |
-+ st->vtime = last_idle->finish; |
10018 |
-+ } |
10019 |
-+ |
10020 |
-+ if (first_idle != NULL && !bfq_gt(first_idle->finish, st->vtime)) |
10021 |
-+ bfq_put_idle_entity(st, first_idle); |
10022 |
-+} |
10023 |
-+ |
10024 |
-+static struct bfq_service_tree * |
10025 |
-+__bfq_entity_update_weight_prio(struct bfq_service_tree *old_st, |
10026 |
-+ struct bfq_entity *entity) |
10027 |
-+{ |
10028 |
-+ struct bfq_service_tree *new_st = old_st; |
10029 |
-+ |
10030 |
-+ if (entity->ioprio_changed) { |
10031 |
-+ struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); |
10032 |
-+ |
10033 |
-+ BUG_ON(old_st->wsum < entity->weight); |
10034 |
-+ old_st->wsum -= entity->weight; |
10035 |
-+ |
10036 |
-+ if (entity->new_weight != entity->orig_weight) { |
10037 |
-+ entity->orig_weight = entity->new_weight; |
10038 |
-+ entity->ioprio = |
10039 |
-+ bfq_weight_to_ioprio(entity->orig_weight); |
10040 |
-+ } else if (entity->new_ioprio != entity->ioprio) { |
10041 |
-+ entity->ioprio = entity->new_ioprio; |
10042 |
-+ entity->orig_weight = |
10043 |
-+ bfq_ioprio_to_weight(entity->ioprio); |
10044 |
-+ } else |
10045 |
-+ entity->new_weight = entity->orig_weight = |
10046 |
-+ bfq_ioprio_to_weight(entity->ioprio); |
10047 |
-+ |
10048 |
-+ entity->ioprio_class = entity->new_ioprio_class; |
10049 |
-+ entity->ioprio_changed = 0; |
10050 |
-+ |
10051 |
-+ /* |
10052 |
-+ * NOTE: here we may be changing the weight too early, |
10053 |
-+ * this will cause unfairness. The correct approach |
10054 |
-+ * would have required additional complexity to defer |
10055 |
-+ * weight changes to the proper time instants (i.e., |
10056 |
-+ * when entity->finish <= old_st->vtime). |
10057 |
-+ */ |
10058 |
-+ new_st = bfq_entity_service_tree(entity); |
10059 |
-+ entity->weight = entity->orig_weight * |
10060 |
-+ (bfqq != NULL ? bfqq->raising_coeff : 1); |
10061 |
-+ new_st->wsum += entity->weight; |
10062 |
-+ |
10063 |
-+ if (new_st != old_st) |
10064 |
-+ entity->start = new_st->vtime; |
10065 |
-+ } |
10066 |
-+ |
10067 |
-+ return new_st; |
10068 |
-+} |
10069 |
-+ |
10070 |
-+/** |
10071 |
-+ * bfq_bfqq_served - update the scheduler status after selection for service. |
10072 |
-+ * @bfqq: the queue being served. |
10073 |
-+ * @served: bytes to transfer. |
10074 |
-+ * |
10075 |
-+ * NOTE: this can be optimized, as the timestamps of upper level entities |
10076 |
-+ * are synchronized every time a new bfqq is selected for service. By now, |
10077 |
-+ * we keep it to better check consistency. |
10078 |
-+ */ |
10079 |
-+static void bfq_bfqq_served(struct bfq_queue *bfqq, unsigned long served) |
10080 |
-+{ |
10081 |
-+ struct bfq_entity *entity = &bfqq->entity; |
10082 |
-+ struct bfq_service_tree *st; |
10083 |
-+ |
10084 |
-+ for_each_entity(entity) { |
10085 |
-+ st = bfq_entity_service_tree(entity); |
10086 |
-+ |
10087 |
-+ entity->service += served; |
10088 |
-+ BUG_ON(entity->service > entity->budget); |
10089 |
-+ BUG_ON(st->wsum == 0); |
10090 |
-+ |
10091 |
-+ st->vtime += bfq_delta(served, st->wsum); |
10092 |
-+ bfq_forget_idle(st); |
10093 |
-+ } |
10094 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, "bfqq_served %lu secs", served); |
10095 |
-+} |
10096 |
-+ |
10097 |
-+/** |
10098 |
-+ * bfq_bfqq_charge_full_budget - set the service to the entity budget. |
10099 |
-+ * @bfqq: the queue that needs a service update. |
10100 |
-+ * |
10101 |
-+ * When it's not possible to be fair in the service domain, because |
10102 |
-+ * a queue is not consuming its budget fast enough (the meaning of |
10103 |
-+ * fast depends on the timeout parameter), we charge it a full |
10104 |
-+ * budget. In this way we should obtain a sort of time-domain |
10105 |
-+ * fairness among all the seeky/slow queues. |
10106 |
-+ */ |
10107 |
-+static inline void bfq_bfqq_charge_full_budget(struct bfq_queue *bfqq) |
10108 |
-+{ |
10109 |
-+ struct bfq_entity *entity = &bfqq->entity; |
10110 |
-+ |
10111 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, "charge_full_budget"); |
10112 |
-+ |
10113 |
-+ bfq_bfqq_served(bfqq, entity->budget - entity->service); |
10114 |
-+} |
10115 |
-+ |
10116 |
-+/** |
10117 |
-+ * __bfq_activate_entity - activate an entity. |
10118 |
-+ * @entity: the entity being activated. |
10119 |
-+ * |
10120 |
-+ * Called whenever an entity is activated, i.e., it is not active and one |
10121 |
-+ * of its children receives a new request, or has to be reactivated due to |
10122 |
-+ * budget exhaustion. It uses the current budget of the entity (and the |
10123 |
-+ * service received if @entity is active) of the queue to calculate its |
10124 |
-+ * timestamps. |
10125 |
-+ */ |
10126 |
-+static void __bfq_activate_entity(struct bfq_entity *entity) |
10127 |
-+{ |
10128 |
-+ struct bfq_sched_data *sd = entity->sched_data; |
10129 |
-+ struct bfq_service_tree *st = bfq_entity_service_tree(entity); |
10130 |
-+ |
10131 |
-+ if (entity == sd->in_service_entity) { |
10132 |
-+ BUG_ON(entity->tree != NULL); |
10133 |
-+ /* |
10134 |
-+ * If we are requeueing the current entity we have |
10135 |
-+ * to take care of not charging to it service it has |
10136 |
-+ * not received. |
10137 |
-+ */ |
10138 |
-+ bfq_calc_finish(entity, entity->service); |
10139 |
-+ entity->start = entity->finish; |
10140 |
-+ sd->in_service_entity = NULL; |
10141 |
-+ } else if (entity->tree == &st->active) { |
10142 |
-+ /* |
10143 |
-+ * Requeueing an entity due to a change of some |
10144 |
-+ * next_in_service entity below it. We reuse the |
10145 |
-+ * old start time. |
10146 |
-+ */ |
10147 |
-+ bfq_active_extract(st, entity); |
10148 |
-+ } else if (entity->tree == &st->idle) { |
10149 |
-+ /* |
10150 |
-+ * Must be on the idle tree, bfq_idle_extract() will |
10151 |
-+ * check for that. |
10152 |
-+ */ |
10153 |
-+ bfq_idle_extract(st, entity); |
10154 |
-+ entity->start = bfq_gt(st->vtime, entity->finish) ? |
10155 |
-+ st->vtime : entity->finish; |
10156 |
-+ } else { |
10157 |
-+ /* |
10158 |
-+ * The finish time of the entity may be invalid, and |
10159 |
-+ * it is in the past for sure, otherwise the queue |
10160 |
-+ * would have been on the idle tree. |
10161 |
-+ */ |
10162 |
-+ entity->start = st->vtime; |
10163 |
-+ st->wsum += entity->weight; |
10164 |
-+ bfq_get_entity(entity); |
10165 |
-+ |
10166 |
-+ BUG_ON(entity->on_st); |
10167 |
-+ entity->on_st = 1; |
10168 |
-+ } |
10169 |
-+ |
10170 |
-+ st = __bfq_entity_update_weight_prio(st, entity); |
10171 |
-+ bfq_calc_finish(entity, entity->budget); |
10172 |
-+ bfq_active_insert(st, entity); |
10173 |
-+} |
10174 |
-+ |
10175 |
-+/** |
10176 |
-+ * bfq_activate_entity - activate an entity and its ancestors if necessary. |
10177 |
-+ * @entity: the entity to activate. |
10178 |
-+ * |
10179 |
-+ * Activate @entity and all the entities on the path from it to the root. |
10180 |
-+ */ |
10181 |
-+static void bfq_activate_entity(struct bfq_entity *entity) |
10182 |
-+{ |
10183 |
-+ struct bfq_sched_data *sd; |
10184 |
-+ |
10185 |
-+ for_each_entity(entity) { |
10186 |
-+ __bfq_activate_entity(entity); |
10187 |
-+ |
10188 |
-+ sd = entity->sched_data; |
10189 |
-+ if (!bfq_update_next_in_service(sd)) |
10190 |
-+ /* |
10191 |
-+ * No need to propagate the activation to the |
10192 |
-+ * upper entities, as they will be updated when |
10193 |
-+ * the in-service entity is rescheduled. |
10194 |
-+ */ |
10195 |
-+ break; |
10196 |
-+ } |
10197 |
-+} |
10198 |
-+ |
10199 |
-+/** |
10200 |
-+ * __bfq_deactivate_entity - deactivate an entity from its service tree. |
10201 |
-+ * @entity: the entity to deactivate. |
10202 |
-+ * @requeue: if false, the entity will not be put into the idle tree. |
10203 |
-+ * |
10204 |
-+ * Deactivate an entity, independently from its previous state. If the |
10205 |
-+ * entity was not on a service tree just return, otherwise if it is on |
10206 |
-+ * any scheduler tree, extract it from that tree, and if necessary |
10207 |
-+ * and if the caller did not specify @requeue, put it on the idle tree. |
10208 |
-+ * |
10209 |
-+ * Return %1 if the caller should update the entity hierarchy, i.e., |
10210 |
-+ * if the entity was under service or if it was the next_in_service for |
10211 |
-+ * its sched_data; return %0 otherwise. |
10212 |
-+ */ |
10213 |
-+static int __bfq_deactivate_entity(struct bfq_entity *entity, int requeue) |
10214 |
-+{ |
10215 |
-+ struct bfq_sched_data *sd = entity->sched_data; |
10216 |
-+ struct bfq_service_tree *st = bfq_entity_service_tree(entity); |
10217 |
-+ int was_in_service = entity == sd->in_service_entity; |
10218 |
-+ int ret = 0; |
10219 |
-+ |
10220 |
-+ if (!entity->on_st) |
10221 |
-+ return 0; |
10222 |
-+ |
10223 |
-+ BUG_ON(was_in_service && entity->tree != NULL); |
10224 |
-+ |
10225 |
-+ if (was_in_service) { |
10226 |
-+ bfq_calc_finish(entity, entity->service); |
10227 |
-+ sd->in_service_entity = NULL; |
10228 |
-+ } else if (entity->tree == &st->active) |
10229 |
-+ bfq_active_extract(st, entity); |
10230 |
-+ else if (entity->tree == &st->idle) |
10231 |
-+ bfq_idle_extract(st, entity); |
10232 |
-+ else if (entity->tree != NULL) |
10233 |
-+ BUG(); |
10234 |
-+ |
10235 |
-+ if (was_in_service || sd->next_in_service == entity) |
10236 |
-+ ret = bfq_update_next_in_service(sd); |
10237 |
-+ |
10238 |
-+ if (!requeue || !bfq_gt(entity->finish, st->vtime)) |
10239 |
-+ bfq_forget_entity(st, entity); |
10240 |
-+ else |
10241 |
-+ bfq_idle_insert(st, entity); |
10242 |
-+ |
10243 |
-+ BUG_ON(sd->in_service_entity == entity); |
10244 |
-+ BUG_ON(sd->next_in_service == entity); |
10245 |
-+ |
10246 |
-+ return ret; |
10247 |
-+} |
10248 |
-+ |
10249 |
-+/** |
10250 |
-+ * bfq_deactivate_entity - deactivate an entity. |
10251 |
-+ * @entity: the entity to deactivate. |
10252 |
-+ * @requeue: true if the entity can be put on the idle tree |
10253 |
-+ */ |
10254 |
-+static void bfq_deactivate_entity(struct bfq_entity *entity, int requeue) |
10255 |
-+{ |
10256 |
-+ struct bfq_sched_data *sd; |
10257 |
-+ struct bfq_entity *parent; |
10258 |
-+ |
10259 |
-+ for_each_entity_safe(entity, parent) { |
10260 |
-+ sd = entity->sched_data; |
10261 |
-+ |
10262 |
-+ if (!__bfq_deactivate_entity(entity, requeue)) |
10263 |
-+ /* |
10264 |
-+ * The parent entity is still backlogged, and |
10265 |
-+ * we don't need to update it as it is still |
10266 |
-+ * under service. |
10267 |
-+ */ |
10268 |
-+ break; |
10269 |
-+ |
10270 |
-+ if (sd->next_in_service != NULL) |
10271 |
-+ /* |
10272 |
-+ * The parent entity is still backlogged and |
10273 |
-+ * the budgets on the path towards the root |
10274 |
-+ * need to be updated. |
10275 |
-+ */ |
10276 |
-+ goto update; |
10277 |
-+ |
10278 |
-+ /* |
10279 |
-+ * If we reach there the parent is no more backlogged and |
10280 |
-+ * we want to propagate the dequeue upwards. |
10281 |
-+ */ |
10282 |
-+ requeue = 1; |
10283 |
-+ } |
10284 |
-+ |
10285 |
-+ return; |
10286 |
-+ |
10287 |
-+update: |
10288 |
-+ entity = parent; |
10289 |
-+ for_each_entity(entity) { |
10290 |
-+ __bfq_activate_entity(entity); |
10291 |
-+ |
10292 |
-+ sd = entity->sched_data; |
10293 |
-+ if (!bfq_update_next_in_service(sd)) |
10294 |
-+ break; |
10295 |
-+ } |
10296 |
-+} |
10297 |
-+ |
10298 |
-+/** |
10299 |
-+ * bfq_update_vtime - update vtime if necessary. |
10300 |
-+ * @st: the service tree to act upon. |
10301 |
-+ * |
10302 |
-+ * If necessary update the service tree vtime to have at least one |
10303 |
-+ * eligible entity, skipping to its start time. Assumes that the |
10304 |
-+ * active tree of the device is not empty. |
10305 |
-+ * |
10306 |
-+ * NOTE: this hierarchical implementation updates vtimes quite often, |
10307 |
-+ * we may end up with reactivated tasks getting timestamps after a |
10308 |
-+ * vtime skip done because we needed a ->first_active entity on some |
10309 |
-+ * intermediate node. |
10310 |
-+ */ |
10311 |
-+static void bfq_update_vtime(struct bfq_service_tree *st) |
10312 |
-+{ |
10313 |
-+ struct bfq_entity *entry; |
10314 |
-+ struct rb_node *node = st->active.rb_node; |
10315 |
-+ |
10316 |
-+ entry = rb_entry(node, struct bfq_entity, rb_node); |
10317 |
-+ if (bfq_gt(entry->min_start, st->vtime)) { |
10318 |
-+ st->vtime = entry->min_start; |
10319 |
-+ bfq_forget_idle(st); |
10320 |
-+ } |
10321 |
-+} |
10322 |
-+ |
10323 |
-+/** |
10324 |
-+ * bfq_first_active_entity - find the eligible entity with |
10325 |
-+ * the smallest finish time |
10326 |
-+ * @st: the service tree to select from. |
10327 |
-+ * |
10328 |
-+ * This function searches the first schedulable entity, starting from the |
10329 |
-+ * root of the tree and going on the left every time on this side there is |
10330 |
-+ * a subtree with at least one eligible (start >= vtime) entity. The path |
10331 |
-+ * on the right is followed only if a) the left subtree contains no eligible |
10332 |
-+ * entities and b) no eligible entity has been found yet. |
10333 |
-+ */ |
10334 |
-+static struct bfq_entity *bfq_first_active_entity(struct bfq_service_tree *st) |
10335 |
-+{ |
10336 |
-+ struct bfq_entity *entry, *first = NULL; |
10337 |
-+ struct rb_node *node = st->active.rb_node; |
10338 |
-+ |
10339 |
-+ while (node != NULL) { |
10340 |
-+ entry = rb_entry(node, struct bfq_entity, rb_node); |
10341 |
-+left: |
10342 |
-+ if (!bfq_gt(entry->start, st->vtime)) |
10343 |
-+ first = entry; |
10344 |
-+ |
10345 |
-+ BUG_ON(bfq_gt(entry->min_start, st->vtime)); |
10346 |
-+ |
10347 |
-+ if (node->rb_left != NULL) { |
10348 |
-+ entry = rb_entry(node->rb_left, |
10349 |
-+ struct bfq_entity, rb_node); |
10350 |
-+ if (!bfq_gt(entry->min_start, st->vtime)) { |
10351 |
-+ node = node->rb_left; |
10352 |
-+ goto left; |
10353 |
-+ } |
10354 |
-+ } |
10355 |
-+ if (first != NULL) |
10356 |
-+ break; |
10357 |
-+ node = node->rb_right; |
10358 |
-+ } |
10359 |
-+ |
10360 |
-+ BUG_ON(first == NULL && !RB_EMPTY_ROOT(&st->active)); |
10361 |
-+ return first; |
10362 |
-+} |
10363 |
-+ |
10364 |
-+/** |
10365 |
-+ * __bfq_lookup_next_entity - return the first eligible entity in @st. |
10366 |
-+ * @st: the service tree. |
10367 |
-+ * |
10368 |
-+ * Update the virtual time in @st and return the first eligible entity |
10369 |
-+ * it contains. |
10370 |
-+ */ |
10371 |
-+static struct bfq_entity *__bfq_lookup_next_entity(struct bfq_service_tree *st, |
10372 |
-+ bool force) |
10373 |
-+{ |
10374 |
-+ struct bfq_entity *entity, *new_next_in_service = NULL; |
10375 |
-+ |
10376 |
-+ if (RB_EMPTY_ROOT(&st->active)) |
10377 |
-+ return NULL; |
10378 |
-+ |
10379 |
-+ bfq_update_vtime(st); |
10380 |
-+ entity = bfq_first_active_entity(st); |
10381 |
-+ BUG_ON(bfq_gt(entity->start, st->vtime)); |
10382 |
-+ |
10383 |
-+ /* |
10384 |
-+ * If the chosen entity does not match with the sched_data's |
10385 |
-+ * next_in_service and we are forcedly serving the IDLE priority |
10386 |
-+ * class tree, bubble up budget update. |
10387 |
-+ */ |
10388 |
-+ if (unlikely(force && entity != entity->sched_data->next_in_service)) { |
10389 |
-+ new_next_in_service = entity; |
10390 |
-+ for_each_entity(new_next_in_service) |
10391 |
-+ bfq_update_budget(new_next_in_service); |
10392 |
-+ } |
10393 |
-+ |
10394 |
-+ return entity; |
10395 |
-+} |
10396 |
-+ |
10397 |
-+/** |
10398 |
-+ * bfq_lookup_next_entity - return the first eligible entity in @sd. |
10399 |
-+ * @sd: the sched_data. |
10400 |
-+ * @extract: if true the returned entity will be also extracted from @sd. |
10401 |
-+ * |
10402 |
-+ * NOTE: since we cache the next_in_service entity at each level of the |
10403 |
-+ * hierarchy, the complexity of the lookup can be decreased with |
10404 |
-+ * absolutely no effort just returning the cached next_in_service value; |
10405 |
-+ * we prefer to do full lookups to test the consistency of * the data |
10406 |
-+ * structures. |
10407 |
-+ */ |
10408 |
-+static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd, |
10409 |
-+ int extract, |
10410 |
-+ struct bfq_data *bfqd) |
10411 |
-+{ |
10412 |
-+ struct bfq_service_tree *st = sd->service_tree; |
10413 |
-+ struct bfq_entity *entity; |
10414 |
-+ int i = 0; |
10415 |
-+ |
10416 |
-+ BUG_ON(sd->in_service_entity != NULL); |
10417 |
-+ |
10418 |
-+ if (bfqd != NULL && |
10419 |
-+ jiffies - bfqd->bfq_class_idle_last_service > BFQ_CL_IDLE_TIMEOUT) { |
10420 |
-+ entity = __bfq_lookup_next_entity(st + BFQ_IOPRIO_CLASSES - 1, |
10421 |
-+ true); |
10422 |
-+ if (entity != NULL) { |
10423 |
-+ i = BFQ_IOPRIO_CLASSES - 1; |
10424 |
-+ bfqd->bfq_class_idle_last_service = jiffies; |
10425 |
-+ sd->next_in_service = entity; |
10426 |
-+ } |
10427 |
-+ } |
10428 |
-+ for (; i < BFQ_IOPRIO_CLASSES; i++) { |
10429 |
-+ entity = __bfq_lookup_next_entity(st + i, false); |
10430 |
-+ if (entity != NULL) { |
10431 |
-+ if (extract) { |
10432 |
-+ bfq_check_next_in_service(sd, entity); |
10433 |
-+ bfq_active_extract(st + i, entity); |
10434 |
-+ sd->in_service_entity = entity; |
10435 |
-+ sd->next_in_service = NULL; |
10436 |
-+ } |
10437 |
-+ break; |
10438 |
-+ } |
10439 |
-+ } |
10440 |
-+ |
10441 |
-+ return entity; |
10442 |
-+} |
10443 |
-+ |
10444 |
-+/* |
10445 |
-+ * Get next queue for service. |
10446 |
-+ */ |
10447 |
-+static struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd) |
10448 |
-+{ |
10449 |
-+ struct bfq_entity *entity = NULL; |
10450 |
-+ struct bfq_sched_data *sd; |
10451 |
-+ struct bfq_queue *bfqq; |
10452 |
-+ |
10453 |
-+ BUG_ON(bfqd->in_service_queue != NULL); |
10454 |
-+ |
10455 |
-+ if (bfqd->busy_queues == 0) |
10456 |
-+ return NULL; |
10457 |
-+ |
10458 |
-+ sd = &bfqd->root_group->sched_data; |
10459 |
-+ for (; sd != NULL; sd = entity->my_sched_data) { |
10460 |
-+ entity = bfq_lookup_next_entity(sd, 1, bfqd); |
10461 |
-+ BUG_ON(entity == NULL); |
10462 |
-+ entity->service = 0; |
10463 |
-+ } |
10464 |
-+ |
10465 |
-+ bfqq = bfq_entity_to_bfqq(entity); |
10466 |
-+ BUG_ON(bfqq == NULL); |
10467 |
-+ |
10468 |
-+ return bfqq; |
10469 |
-+} |
10470 |
-+ |
10471 |
-+/* |
10472 |
-+ * Forced extraction of the given queue. |
10473 |
-+ */ |
10474 |
-+static void bfq_get_next_queue_forced(struct bfq_data *bfqd, |
10475 |
-+ struct bfq_queue *bfqq) |
10476 |
-+{ |
10477 |
-+ struct bfq_entity *entity; |
10478 |
-+ struct bfq_sched_data *sd; |
10479 |
-+ |
10480 |
-+ BUG_ON(bfqd->in_service_queue != NULL); |
10481 |
-+ |
10482 |
-+ entity = &bfqq->entity; |
10483 |
-+ /* |
10484 |
-+ * Bubble up extraction/update from the leaf to the root. |
10485 |
-+ */ |
10486 |
-+ for_each_entity(entity) { |
10487 |
-+ sd = entity->sched_data; |
10488 |
-+ bfq_update_budget(entity); |
10489 |
-+ bfq_update_vtime(bfq_entity_service_tree(entity)); |
10490 |
-+ bfq_active_extract(bfq_entity_service_tree(entity), entity); |
10491 |
-+ sd->active_entity = entity; |
10492 |
-+ sd->next_active = NULL; |
10493 |
-+ entity->service = 0; |
10494 |
-+ } |
10495 |
-+ |
10496 |
-+ return; |
10497 |
-+} |
10498 |
-+ |
10499 |
-+static void __bfq_bfqd_reset_in_service(struct bfq_data *bfqd) |
10500 |
-+{ |
10501 |
-+ if (bfqd->in_service_bic != NULL) { |
10502 |
-+ put_io_context(bfqd->in_service_bic->icq.ioc); |
10503 |
-+ bfqd->in_service_bic = NULL; |
10504 |
-+ } |
10505 |
-+ |
10506 |
-+ bfqd->in_service_queue = NULL; |
10507 |
-+ del_timer(&bfqd->idle_slice_timer); |
10508 |
-+} |
10509 |
-+ |
10510 |
-+static void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
10511 |
-+ int requeue) |
10512 |
-+{ |
10513 |
-+ struct bfq_entity *entity = &bfqq->entity; |
10514 |
-+ |
10515 |
-+ if (bfqq == bfqd->in_service_queue) |
10516 |
-+ __bfq_bfqd_reset_in_service(bfqd); |
10517 |
-+ |
10518 |
-+ bfq_deactivate_entity(entity, requeue); |
10519 |
-+} |
10520 |
-+ |
10521 |
-+static void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq) |
10522 |
-+{ |
10523 |
-+ struct bfq_entity *entity = &bfqq->entity; |
10524 |
-+ |
10525 |
-+ bfq_activate_entity(entity); |
10526 |
-+} |
10527 |
-+ |
10528 |
-+/* |
10529 |
-+ * Called when the bfqq no longer has requests pending, remove it from |
10530 |
-+ * the service tree. |
10531 |
-+ */ |
10532 |
-+static void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
10533 |
-+ int requeue) |
10534 |
-+{ |
10535 |
-+ BUG_ON(!bfq_bfqq_busy(bfqq)); |
10536 |
-+ BUG_ON(!RB_EMPTY_ROOT(&bfqq->sort_list)); |
10537 |
-+ |
10538 |
-+ bfq_log_bfqq(bfqd, bfqq, "del from busy"); |
10539 |
-+ |
10540 |
-+ bfq_clear_bfqq_busy(bfqq); |
10541 |
-+ |
10542 |
-+ BUG_ON(bfqd->busy_queues == 0); |
10543 |
-+ bfqd->busy_queues--; |
10544 |
-+ if (bfqq->raising_coeff > 1) |
10545 |
-+ bfqd->raised_busy_queues--; |
10546 |
-+ |
10547 |
-+ bfq_deactivate_bfqq(bfqd, bfqq, requeue); |
10548 |
-+} |
10549 |
-+ |
10550 |
-+/* |
10551 |
-+ * Called when an inactive queue receives a new request. |
10552 |
-+ */ |
10553 |
-+static void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq) |
10554 |
-+{ |
10555 |
-+ BUG_ON(bfq_bfqq_busy(bfqq)); |
10556 |
-+ BUG_ON(bfqq == bfqd->in_service_queue); |
10557 |
-+ |
10558 |
-+ bfq_log_bfqq(bfqd, bfqq, "add to busy"); |
10559 |
-+ |
10560 |
-+ bfq_activate_bfqq(bfqd, bfqq); |
10561 |
-+ |
10562 |
-+ bfq_mark_bfqq_busy(bfqq); |
10563 |
-+ bfqd->busy_queues++; |
10564 |
-+ if (bfqq->raising_coeff > 1) |
10565 |
-+ bfqd->raised_busy_queues++; |
10566 |
-+} |
10567 |
-diff --git a/block/bfq.h b/block/bfq.h |
10568 |
-new file mode 100644 |
10569 |
-index 0000000..3ca8482 |
10570 |
---- /dev/null |
10571 |
-+++ b/block/bfq.h |
10572 |
-@@ -0,0 +1,622 @@ |
10573 |
-+/* |
10574 |
-+ * BFQ-v7r2 for 3.14.0: data structures and common functions prototypes. |
10575 |
-+ * |
10576 |
-+ * Based on ideas and code from CFQ: |
10577 |
-+ * Copyright (C) 2003 Jens Axboe <axboe@××××××.dk> |
10578 |
-+ * |
10579 |
-+ * Copyright (C) 2008 Fabio Checconi <fabio@×××××××××××××.it> |
10580 |
-+ * Paolo Valente <paolo.valente@×××××××.it> |
10581 |
-+ * |
10582 |
-+ * Copyright (C) 2010 Paolo Valente <paolo.valente@×××××××.it> |
10583 |
-+ */ |
10584 |
-+ |
10585 |
-+#ifndef _BFQ_H |
10586 |
-+#define _BFQ_H |
10587 |
-+ |
10588 |
-+#include <linux/blktrace_api.h> |
10589 |
-+#include <linux/hrtimer.h> |
10590 |
-+#include <linux/ioprio.h> |
10591 |
-+#include <linux/rbtree.h> |
10592 |
-+ |
10593 |
-+#define BFQ_IOPRIO_CLASSES 3 |
10594 |
-+#define BFQ_CL_IDLE_TIMEOUT (HZ/5) |
10595 |
-+ |
10596 |
-+#define BFQ_MIN_WEIGHT 1 |
10597 |
-+#define BFQ_MAX_WEIGHT 1000 |
10598 |
-+ |
10599 |
-+#define BFQ_DEFAULT_GRP_WEIGHT 10 |
10600 |
-+#define BFQ_DEFAULT_GRP_IOPRIO 0 |
10601 |
-+#define BFQ_DEFAULT_GRP_CLASS IOPRIO_CLASS_BE |
10602 |
-+ |
10603 |
-+struct bfq_entity; |
10604 |
-+ |
10605 |
-+/** |
10606 |
-+ * struct bfq_service_tree - per ioprio_class service tree. |
10607 |
-+ * @active: tree for active entities (i.e., those backlogged). |
10608 |
-+ * @idle: tree for idle entities (i.e., those not backlogged, with V <= F_i). |
10609 |
-+ * @first_idle: idle entity with minimum F_i. |
10610 |
-+ * @last_idle: idle entity with maximum F_i. |
10611 |
-+ * @vtime: scheduler virtual time. |
10612 |
-+ * @wsum: scheduler weight sum; active and idle entities contribute to it. |
10613 |
-+ * |
10614 |
-+ * Each service tree represents a B-WF2Q+ scheduler on its own. Each |
10615 |
-+ * ioprio_class has its own independent scheduler, and so its own |
10616 |
-+ * bfq_service_tree. All the fields are protected by the queue lock |
10617 |
-+ * of the containing bfqd. |
10618 |
-+ */ |
10619 |
-+struct bfq_service_tree { |
10620 |
-+ struct rb_root active; |
10621 |
-+ struct rb_root idle; |
10622 |
-+ |
10623 |
-+ struct bfq_entity *first_idle; |
10624 |
-+ struct bfq_entity *last_idle; |
10625 |
-+ |
10626 |
-+ u64 vtime; |
10627 |
-+ unsigned long wsum; |
10628 |
-+}; |
10629 |
-+ |
10630 |
-+/** |
10631 |
-+ * struct bfq_sched_data - multi-class scheduler. |
10632 |
-+ * @in_service_entity: entity under service. |
10633 |
-+ * @next_in_service: head-of-the-line entity in the scheduler. |
10634 |
-+ * @service_tree: array of service trees, one per ioprio_class. |
10635 |
-+ * |
10636 |
-+ * bfq_sched_data is the basic scheduler queue. It supports three |
10637 |
-+ * ioprio_classes, and can be used either as a toplevel queue or as |
10638 |
-+ * an intermediate queue on a hierarchical setup. |
10639 |
-+ * @next_in_service points to the active entity of the sched_data |
10640 |
-+ * service trees that will be scheduled next. |
10641 |
-+ * |
10642 |
-+ * The supported ioprio_classes are the same as in CFQ, in descending |
10643 |
-+ * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE. |
10644 |
-+ * Requests from higher priority queues are served before all the |
10645 |
-+ * requests from lower priority queues; among requests of the same |
10646 |
-+ * queue requests are served according to B-WF2Q+. |
10647 |
-+ * All the fields are protected by the queue lock of the containing bfqd. |
10648 |
-+ */ |
10649 |
-+struct bfq_sched_data { |
10650 |
-+ struct bfq_entity *in_service_entity; |
10651 |
-+ struct bfq_entity *next_in_service; |
10652 |
-+ struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES]; |
10653 |
-+}; |
10654 |
-+ |
10655 |
-+/** |
10656 |
-+ * struct bfq_entity - schedulable entity. |
10657 |
-+ * @rb_node: service_tree member. |
10658 |
-+ * @on_st: flag, true if the entity is on a tree (either the active or |
10659 |
-+ * the idle one of its service_tree). |
10660 |
-+ * @finish: B-WF2Q+ finish timestamp (aka F_i). |
10661 |
-+ * @start: B-WF2Q+ start timestamp (aka S_i). |
10662 |
-+ * @tree: tree the entity is enqueued into; %NULL if not on a tree. |
10663 |
-+ * @min_start: minimum start time of the (active) subtree rooted at |
10664 |
-+ * this entity; used for O(log N) lookups into active trees. |
10665 |
-+ * @service: service received during the last round of service. |
10666 |
-+ * @budget: budget used to calculate F_i; F_i = S_i + @budget / @weight. |
10667 |
-+ * @weight: weight of the queue |
10668 |
-+ * @parent: parent entity, for hierarchical scheduling. |
10669 |
-+ * @my_sched_data: for non-leaf nodes in the cgroup hierarchy, the |
10670 |
-+ * associated scheduler queue, %NULL on leaf nodes. |
10671 |
-+ * @sched_data: the scheduler queue this entity belongs to. |
10672 |
-+ * @ioprio: the ioprio in use. |
10673 |
-+ * @new_weight: when a weight change is requested, the new weight value. |
10674 |
-+ * @orig_weight: original weight, used to implement weight boosting |
10675 |
-+ * @new_ioprio: when an ioprio change is requested, the new ioprio value. |
10676 |
-+ * @ioprio_class: the ioprio_class in use. |
10677 |
-+ * @new_ioprio_class: when an ioprio_class change is requested, the new |
10678 |
-+ * ioprio_class value. |
10679 |
-+ * @ioprio_changed: flag, true when the user requested a weight, ioprio or |
10680 |
-+ * ioprio_class change. |
10681 |
-+ * |
10682 |
-+ * A bfq_entity is used to represent either a bfq_queue (leaf node in the |
10683 |
-+ * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each |
10684 |
-+ * entity belongs to the sched_data of the parent group in the cgroup |
10685 |
-+ * hierarchy. Non-leaf entities have also their own sched_data, stored |
10686 |
-+ * in @my_sched_data. |
10687 |
-+ * |
10688 |
-+ * Each entity stores independently its priority values; this would |
10689 |
-+ * allow different weights on different devices, but this |
10690 |
-+ * functionality is not exported to userspace by now. Priorities and |
10691 |
-+ * weights are updated lazily, first storing the new values into the |
10692 |
-+ * new_* fields, then setting the @ioprio_changed flag. As soon as |
10693 |
-+ * there is a transition in the entity state that allows the priority |
10694 |
-+ * update to take place the effective and the requested priority |
10695 |
-+ * values are synchronized. |
10696 |
-+ * |
10697 |
-+ * Unless cgroups are used, the weight value is calculated from the |
10698 |
-+ * ioprio to export the same interface as CFQ. When dealing with |
10699 |
-+ * ``well-behaved'' queues (i.e., queues that do not spend too much |
10700 |
-+ * time to consume their budget and have true sequential behavior, and |
10701 |
-+ * when there are no external factors breaking anticipation) the |
10702 |
-+ * relative weights at each level of the cgroups hierarchy should be |
10703 |
-+ * guaranteed. All the fields are protected by the queue lock of the |
10704 |
-+ * containing bfqd. |
10705 |
-+ */ |
10706 |
-+struct bfq_entity { |
10707 |
-+ struct rb_node rb_node; |
10708 |
-+ |
10709 |
-+ int on_st; |
10710 |
-+ |
10711 |
-+ u64 finish; |
10712 |
-+ u64 start; |
10713 |
-+ |
10714 |
-+ struct rb_root *tree; |
10715 |
-+ |
10716 |
-+ u64 min_start; |
10717 |
-+ |
10718 |
-+ unsigned long service, budget; |
10719 |
-+ unsigned short weight, new_weight; |
10720 |
-+ unsigned short orig_weight; |
10721 |
-+ |
10722 |
-+ struct bfq_entity *parent; |
10723 |
-+ |
10724 |
-+ struct bfq_sched_data *my_sched_data; |
10725 |
-+ struct bfq_sched_data *sched_data; |
10726 |
-+ |
10727 |
-+ unsigned short ioprio, new_ioprio; |
10728 |
-+ unsigned short ioprio_class, new_ioprio_class; |
10729 |
-+ |
10730 |
-+ int ioprio_changed; |
10731 |
-+}; |
10732 |
-+ |
10733 |
-+struct bfq_group; |
10734 |
-+ |
10735 |
-+/** |
10736 |
-+ * struct bfq_queue - leaf schedulable entity. |
10737 |
-+ * @ref: reference counter. |
10738 |
-+ * @bfqd: parent bfq_data. |
10739 |
-+ * @new_bfqq: shared bfq_queue if queue is cooperating with |
10740 |
-+ * one or more other queues. |
10741 |
-+ * @pos_node: request-position tree member (see bfq_data's @rq_pos_tree). |
10742 |
-+ * @pos_root: request-position tree root (see bfq_data's @rq_pos_tree). |
10743 |
-+ * @sort_list: sorted list of pending requests. |
10744 |
-+ * @next_rq: if fifo isn't expired, next request to serve. |
10745 |
-+ * @queued: nr of requests queued in @sort_list. |
10746 |
-+ * @allocated: currently allocated requests. |
10747 |
-+ * @meta_pending: pending metadata requests. |
10748 |
-+ * @fifo: fifo list of requests in sort_list. |
10749 |
-+ * @entity: entity representing this queue in the scheduler. |
10750 |
-+ * @max_budget: maximum budget allowed from the feedback mechanism. |
10751 |
-+ * @budget_timeout: budget expiration (in jiffies). |
10752 |
-+ * @dispatched: number of requests on the dispatch list or inside driver. |
10753 |
-+ * @org_ioprio: saved ioprio during boosted periods. |
10754 |
-+ * @flags: status flags. |
10755 |
-+ * @bfqq_list: node for active/idle bfqq list inside our bfqd. |
10756 |
-+ * @seek_samples: number of seeks sampled |
10757 |
-+ * @seek_total: sum of the distances of the seeks sampled |
10758 |
-+ * @seek_mean: mean seek distance |
10759 |
-+ * @last_request_pos: position of the last request enqueued |
10760 |
-+ * @pid: pid of the process owning the queue, used for logging purposes. |
10761 |
-+ * @last_rais_start_finish: start time of the current weight-raising period if |
10762 |
-+ * the @bfq-queue is being weight-raised, otherwise |
10763 |
-+ * finish time of the last weight-raising period |
10764 |
-+ * @raising_cur_max_time: current max raising time for this queue |
10765 |
-+ * @soft_rt_next_start: minimum time instant such that, only if a new request |
10766 |
-+ * is enqueued after this time instant in an idle |
10767 |
-+ * @bfq_queue with no outstanding requests, then the |
10768 |
-+ * task associated with the queue it is deemed as soft |
10769 |
-+ * real-time (see the comments to the function |
10770 |
-+ * bfq_bfqq_softrt_next_start()) |
10771 |
-+ * @last_idle_bklogged: time of the last transition of the @bfq_queue from |
10772 |
-+ * idle to backlogged |
10773 |
-+ * @service_from_backlogged: cumulative service received from the @bfq_queue |
10774 |
-+ * since the last transition from idle to backlogged |
10775 |
-+ * |
10776 |
-+ * A bfq_queue is a leaf request queue; it can be associated with an io_context |
10777 |
-+ * or more, if it is async or shared between cooperating processes. @cgroup |
10778 |
-+ * holds a reference to the cgroup, to be sure that it does not disappear while |
10779 |
-+ * a bfqq still references it (mostly to avoid races between request issuing and |
10780 |
-+ * task migration followed by cgroup destruction). |
10781 |
-+ * All the fields are protected by the queue lock of the containing bfqd. |
10782 |
-+ */ |
10783 |
-+struct bfq_queue { |
10784 |
-+ atomic_t ref; |
10785 |
-+ struct bfq_data *bfqd; |
10786 |
-+ |
10787 |
-+ /* fields for cooperating queues handling */ |
10788 |
-+ struct bfq_queue *new_bfqq; |
10789 |
-+ struct rb_node pos_node; |
10790 |
-+ struct rb_root *pos_root; |
10791 |
-+ |
10792 |
-+ struct rb_root sort_list; |
10793 |
-+ struct request *next_rq; |
10794 |
-+ int queued[2]; |
10795 |
-+ int allocated[2]; |
10796 |
-+ int meta_pending; |
10797 |
-+ struct list_head fifo; |
10798 |
-+ |
10799 |
-+ struct bfq_entity entity; |
10800 |
-+ |
10801 |
-+ unsigned long max_budget; |
10802 |
-+ unsigned long budget_timeout; |
10803 |
-+ |
10804 |
-+ int dispatched; |
10805 |
-+ |
10806 |
-+ unsigned short org_ioprio; |
10807 |
-+ |
10808 |
-+ unsigned int flags; |
10809 |
-+ |
10810 |
-+ struct list_head bfqq_list; |
10811 |
-+ |
10812 |
-+ unsigned int seek_samples; |
10813 |
-+ u64 seek_total; |
10814 |
-+ sector_t seek_mean; |
10815 |
-+ sector_t last_request_pos; |
10816 |
-+ |
10817 |
-+ pid_t pid; |
10818 |
-+ |
10819 |
-+ /* weight-raising fields */ |
10820 |
-+ unsigned long raising_cur_max_time; |
10821 |
-+ unsigned long soft_rt_next_start; |
10822 |
-+ unsigned long last_rais_start_finish; |
10823 |
-+ unsigned int raising_coeff; |
10824 |
-+ unsigned long last_idle_bklogged; |
10825 |
-+ unsigned long service_from_backlogged; |
10826 |
-+}; |
10827 |
-+ |
10828 |
-+/** |
10829 |
-+ * struct bfq_ttime - per process thinktime stats. |
10830 |
-+ * @ttime_total: total process thinktime |
10831 |
-+ * @ttime_samples: number of thinktime samples |
10832 |
-+ * @ttime_mean: average process thinktime |
10833 |
-+ */ |
10834 |
-+struct bfq_ttime { |
10835 |
-+ unsigned long last_end_request; |
10836 |
-+ |
10837 |
-+ unsigned long ttime_total; |
10838 |
-+ unsigned long ttime_samples; |
10839 |
-+ unsigned long ttime_mean; |
10840 |
-+}; |
10841 |
-+ |
10842 |
-+/** |
10843 |
-+ * struct bfq_io_cq - per (request_queue, io_context) structure. |
10844 |
-+ * @icq: associated io_cq structure |
10845 |
-+ * @bfqq: array of two process queues, the sync and the async |
10846 |
-+ * @ttime: associated @bfq_ttime struct |
10847 |
-+ */ |
10848 |
-+struct bfq_io_cq { |
10849 |
-+ struct io_cq icq; /* must be the first member */ |
10850 |
-+ struct bfq_queue *bfqq[2]; |
10851 |
-+ struct bfq_ttime ttime; |
10852 |
-+ int ioprio; |
10853 |
-+}; |
10854 |
-+ |
10855 |
-+/** |
10856 |
-+ * struct bfq_data - per device data structure. |
10857 |
-+ * @queue: request queue for the managed device. |
10858 |
-+ * @root_group: root bfq_group for the device. |
10859 |
-+ * @rq_pos_tree: rbtree sorted by next_request position, |
10860 |
-+ * used when determining if two or more queues |
10861 |
-+ * have interleaving requests (see bfq_close_cooperator). |
10862 |
-+ * @busy_queues: number of bfq_queues containing requests (including the |
10863 |
-+ * queue under service, even if it is idling). |
10864 |
-+ * @raised_busy_queues: number of weight-raised busy bfq_queues. |
10865 |
-+ * @queued: number of queued requests. |
10866 |
-+ * @rq_in_driver: number of requests dispatched and waiting for completion. |
10867 |
-+ * @sync_flight: number of sync requests in the driver. |
10868 |
-+ * @max_rq_in_driver: max number of reqs in driver in the last @hw_tag_samples |
10869 |
-+ * completed requests . |
10870 |
-+ * @hw_tag_samples: nr of samples used to calculate hw_tag. |
10871 |
-+ * @hw_tag: flag set to one if the driver is showing a queueing behavior. |
10872 |
-+ * @budgets_assigned: number of budgets assigned. |
10873 |
-+ * @idle_slice_timer: timer set when idling for the next sequential request |
10874 |
-+ * from the queue under service. |
10875 |
-+ * @unplug_work: delayed work to restart dispatching on the request queue. |
10876 |
-+ * @in_service_queue: bfq_queue under service. |
10877 |
-+ * @in_service_bic: bfq_io_cq (bic) associated with the @in_service_queue. |
10878 |
-+ * @last_position: on-disk position of the last served request. |
10879 |
-+ * @last_budget_start: beginning of the last budget. |
10880 |
-+ * @last_idling_start: beginning of the last idle slice. |
10881 |
-+ * @peak_rate: peak transfer rate observed for a budget. |
10882 |
-+ * @peak_rate_samples: number of samples used to calculate @peak_rate. |
10883 |
-+ * @bfq_max_budget: maximum budget allotted to a bfq_queue before rescheduling. |
10884 |
-+ * @group_list: list of all the bfq_groups active on the device. |
10885 |
-+ * @active_list: list of all the bfq_queues active on the device. |
10886 |
-+ * @idle_list: list of all the bfq_queues idle on the device. |
10887 |
-+ * @bfq_quantum: max number of requests dispatched per dispatch round. |
10888 |
-+ * @bfq_fifo_expire: timeout for async/sync requests; when it expires |
10889 |
-+ * requests are served in fifo order. |
10890 |
-+ * @bfq_back_penalty: weight of backward seeks wrt forward ones. |
10891 |
-+ * @bfq_back_max: maximum allowed backward seek. |
10892 |
-+ * @bfq_slice_idle: maximum idling time. |
10893 |
-+ * @bfq_user_max_budget: user-configured max budget value (0 for auto-tuning). |
10894 |
-+ * @bfq_max_budget_async_rq: maximum budget (in nr of requests) allotted to |
10895 |
-+ * async queues. |
10896 |
-+ * @bfq_timeout: timeout for bfq_queues to consume their budget; used to |
10897 |
-+ * to prevent seeky queues to impose long latencies to well |
10898 |
-+ * behaved ones (this also implies that seeky queues cannot |
10899 |
-+ * receive guarantees in the service domain; after a timeout |
10900 |
-+ * they are charged for the whole allocated budget, to try |
10901 |
-+ * to preserve a behavior reasonably fair among them, but |
10902 |
-+ * without service-domain guarantees). |
10903 |
-+ * @bfq_raising_coeff: Maximum factor by which the weight of a boosted |
10904 |
-+ * queue is multiplied |
10905 |
-+ * @bfq_raising_max_time: maximum duration of a weight-raising period (jiffies) |
10906 |
-+ * @bfq_raising_rt_max_time: maximum duration for soft real-time processes |
10907 |
-+ * @bfq_raising_min_idle_time: minimum idle period after which weight-raising |
10908 |
-+ * may be reactivated for a queue (in jiffies) |
10909 |
-+ * @bfq_raising_min_inter_arr_async: minimum period between request arrivals |
10910 |
-+ * after which weight-raising may be |
10911 |
-+ * reactivated for an already busy queue |
10912 |
-+ * (in jiffies) |
10913 |
-+ * @bfq_raising_max_softrt_rate: max service-rate for a soft real-time queue, |
10914 |
-+ * sectors per seconds |
10915 |
-+ * @RT_prod: cached value of the product R*T used for computing the maximum |
10916 |
-+ * duration of the weight raising automatically |
10917 |
-+ * @oom_bfqq: fallback dummy bfqq for extreme OOM conditions |
10918 |
-+ * |
10919 |
-+ * All the fields are protected by the @queue lock. |
10920 |
-+ */ |
10921 |
-+struct bfq_data { |
10922 |
-+ struct request_queue *queue; |
10923 |
-+ |
10924 |
-+ struct bfq_group *root_group; |
10925 |
-+ |
10926 |
-+ struct rb_root rq_pos_tree; |
10927 |
-+ |
10928 |
-+ int busy_queues; |
10929 |
-+ int raised_busy_queues; |
10930 |
-+ int queued; |
10931 |
-+ int rq_in_driver; |
10932 |
-+ int sync_flight; |
10933 |
-+ |
10934 |
-+ int max_rq_in_driver; |
10935 |
-+ int hw_tag_samples; |
10936 |
-+ int hw_tag; |
10937 |
-+ |
10938 |
-+ int budgets_assigned; |
10939 |
-+ |
10940 |
-+ struct timer_list idle_slice_timer; |
10941 |
-+ struct work_struct unplug_work; |
10942 |
-+ |
10943 |
-+ struct bfq_queue *in_service_queue; |
10944 |
-+ struct bfq_io_cq *in_service_bic; |
10945 |
-+ |
10946 |
-+ sector_t last_position; |
10947 |
-+ |
10948 |
-+ ktime_t last_budget_start; |
10949 |
-+ ktime_t last_idling_start; |
10950 |
-+ int peak_rate_samples; |
10951 |
-+ u64 peak_rate; |
10952 |
-+ unsigned long bfq_max_budget; |
10953 |
-+ |
10954 |
-+ struct hlist_head group_list; |
10955 |
-+ struct list_head active_list; |
10956 |
-+ struct list_head idle_list; |
10957 |
-+ |
10958 |
-+ unsigned int bfq_quantum; |
10959 |
-+ unsigned int bfq_fifo_expire[2]; |
10960 |
-+ unsigned int bfq_back_penalty; |
10961 |
-+ unsigned int bfq_back_max; |
10962 |
-+ unsigned int bfq_slice_idle; |
10963 |
-+ u64 bfq_class_idle_last_service; |
10964 |
-+ |
10965 |
-+ unsigned int bfq_user_max_budget; |
10966 |
-+ unsigned int bfq_max_budget_async_rq; |
10967 |
-+ unsigned int bfq_timeout[2]; |
10968 |
-+ |
10969 |
-+ bool low_latency; |
10970 |
-+ |
10971 |
-+ /* parameters of the low_latency heuristics */ |
10972 |
-+ unsigned int bfq_raising_coeff; |
10973 |
-+ unsigned int bfq_raising_max_time; |
10974 |
-+ unsigned int bfq_raising_rt_max_time; |
10975 |
-+ unsigned int bfq_raising_min_idle_time; |
10976 |
-+ unsigned long bfq_raising_min_inter_arr_async; |
10977 |
-+ unsigned int bfq_raising_max_softrt_rate; |
10978 |
-+ u64 RT_prod; |
10979 |
-+ |
10980 |
-+ struct bfq_queue oom_bfqq; |
10981 |
-+}; |
10982 |
-+ |
10983 |
-+enum bfqq_state_flags { |
10984 |
-+ BFQ_BFQQ_FLAG_busy = 0, /* has requests or is under service */ |
10985 |
-+ BFQ_BFQQ_FLAG_wait_request, /* waiting for a request */ |
10986 |
-+ BFQ_BFQQ_FLAG_must_alloc, /* must be allowed rq alloc */ |
10987 |
-+ BFQ_BFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
10988 |
-+ BFQ_BFQQ_FLAG_idle_window, /* slice idling enabled */ |
10989 |
-+ BFQ_BFQQ_FLAG_prio_changed, /* task priority has changed */ |
10990 |
-+ BFQ_BFQQ_FLAG_sync, /* synchronous queue */ |
10991 |
-+ BFQ_BFQQ_FLAG_budget_new, /* no completion with this budget */ |
10992 |
-+ BFQ_BFQQ_FLAG_coop, /* bfqq is shared */ |
10993 |
-+ BFQ_BFQQ_FLAG_split_coop, /* shared bfqq will be splitted */ |
10994 |
-+ BFQ_BFQQ_FLAG_softrt_update, /* needs softrt-next-start update */ |
10995 |
-+}; |
10996 |
-+ |
10997 |
-+#define BFQ_BFQQ_FNS(name) \ |
10998 |
-+static inline void bfq_mark_bfqq_##name(struct bfq_queue *bfqq) \ |
10999 |
-+{ \ |
11000 |
-+ (bfqq)->flags |= (1 << BFQ_BFQQ_FLAG_##name); \ |
11001 |
-+} \ |
11002 |
-+static inline void bfq_clear_bfqq_##name(struct bfq_queue *bfqq) \ |
11003 |
-+{ \ |
11004 |
-+ (bfqq)->flags &= ~(1 << BFQ_BFQQ_FLAG_##name); \ |
11005 |
-+} \ |
11006 |
-+static inline int bfq_bfqq_##name(const struct bfq_queue *bfqq) \ |
11007 |
-+{ \ |
11008 |
-+ return ((bfqq)->flags & (1 << BFQ_BFQQ_FLAG_##name)) != 0; \ |
11009 |
-+} |
11010 |
-+ |
11011 |
-+BFQ_BFQQ_FNS(busy); |
11012 |
-+BFQ_BFQQ_FNS(wait_request); |
11013 |
-+BFQ_BFQQ_FNS(must_alloc); |
11014 |
-+BFQ_BFQQ_FNS(fifo_expire); |
11015 |
-+BFQ_BFQQ_FNS(idle_window); |
11016 |
-+BFQ_BFQQ_FNS(prio_changed); |
11017 |
-+BFQ_BFQQ_FNS(sync); |
11018 |
-+BFQ_BFQQ_FNS(budget_new); |
11019 |
-+BFQ_BFQQ_FNS(coop); |
11020 |
-+BFQ_BFQQ_FNS(split_coop); |
11021 |
-+BFQ_BFQQ_FNS(softrt_update); |
11022 |
-+#undef BFQ_BFQQ_FNS |
11023 |
-+ |
11024 |
-+/* Logging facilities. */ |
11025 |
-+#define bfq_log_bfqq(bfqd, bfqq, fmt, args...) \ |
11026 |
-+ blk_add_trace_msg((bfqd)->queue, "bfq%d " fmt, (bfqq)->pid, ##args) |
11027 |
-+ |
11028 |
-+#define bfq_log(bfqd, fmt, args...) \ |
11029 |
-+ blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args) |
11030 |
-+ |
11031 |
-+/* Expiration reasons. */ |
11032 |
-+enum bfqq_expiration { |
11033 |
-+ BFQ_BFQQ_TOO_IDLE = 0, /* queue has been idling for too long */ |
11034 |
-+ BFQ_BFQQ_BUDGET_TIMEOUT, /* budget took too long to be used */ |
11035 |
-+ BFQ_BFQQ_BUDGET_EXHAUSTED, /* budget consumed */ |
11036 |
-+ BFQ_BFQQ_NO_MORE_REQUESTS, /* the queue has no more requests */ |
11037 |
-+}; |
11038 |
-+ |
11039 |
-+#ifdef CONFIG_CGROUP_BFQIO |
11040 |
-+/** |
11041 |
-+ * struct bfq_group - per (device, cgroup) data structure. |
11042 |
-+ * @entity: schedulable entity to insert into the parent group sched_data. |
11043 |
-+ * @sched_data: own sched_data, to contain child entities (they may be |
11044 |
-+ * both bfq_queues and bfq_groups). |
11045 |
-+ * @group_node: node to be inserted into the bfqio_cgroup->group_data |
11046 |
-+ * list of the containing cgroup's bfqio_cgroup. |
11047 |
-+ * @bfqd_node: node to be inserted into the @bfqd->group_list list |
11048 |
-+ * of the groups active on the same device; used for cleanup. |
11049 |
-+ * @bfqd: the bfq_data for the device this group acts upon. |
11050 |
-+ * @async_bfqq: array of async queues for all the tasks belonging to |
11051 |
-+ * the group, one queue per ioprio value per ioprio_class, |
11052 |
-+ * except for the idle class that has only one queue. |
11053 |
-+ * @async_idle_bfqq: async queue for the idle class (ioprio is ignored). |
11054 |
-+ * @my_entity: pointer to @entity, %NULL for the toplevel group; used |
11055 |
-+ * to avoid too many special cases during group creation/migration. |
11056 |
-+ * |
11057 |
-+ * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup |
11058 |
-+ * there is a set of bfq_groups, each one collecting the lower-level |
11059 |
-+ * entities belonging to the group that are acting on the same device. |
11060 |
-+ * |
11061 |
-+ * Locking works as follows: |
11062 |
-+ * o @group_node is protected by the bfqio_cgroup lock, and is accessed |
11063 |
-+ * via RCU from its readers. |
11064 |
-+ * o @bfqd is protected by the queue lock, RCU is used to access it |
11065 |
-+ * from the readers. |
11066 |
-+ * o All the other fields are protected by the @bfqd queue lock. |
11067 |
-+ */ |
11068 |
-+struct bfq_group { |
11069 |
-+ struct bfq_entity entity; |
11070 |
-+ struct bfq_sched_data sched_data; |
11071 |
-+ |
11072 |
-+ struct hlist_node group_node; |
11073 |
-+ struct hlist_node bfqd_node; |
11074 |
-+ |
11075 |
-+ void *bfqd; |
11076 |
-+ |
11077 |
-+ struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR]; |
11078 |
-+ struct bfq_queue *async_idle_bfqq; |
11079 |
-+ |
11080 |
-+ struct bfq_entity *my_entity; |
11081 |
-+}; |
11082 |
-+ |
11083 |
-+/** |
11084 |
-+ * struct bfqio_cgroup - bfq cgroup data structure. |
11085 |
-+ * @css: subsystem state for bfq in the containing cgroup. |
11086 |
-+ * @online: flag marked when the subsystem is inserted. |
11087 |
-+ * @weight: cgroup weight. |
11088 |
-+ * @ioprio: cgroup ioprio. |
11089 |
-+ * @ioprio_class: cgroup ioprio_class. |
11090 |
-+ * @lock: spinlock that protects @ioprio, @ioprio_class and @group_data. |
11091 |
-+ * @group_data: list containing the bfq_group belonging to this cgroup. |
11092 |
-+ * |
11093 |
-+ * @group_data is accessed using RCU, with @lock protecting the updates, |
11094 |
-+ * @ioprio and @ioprio_class are protected by @lock. |
11095 |
-+ */ |
11096 |
-+struct bfqio_cgroup { |
11097 |
-+ struct cgroup_subsys_state css; |
11098 |
-+ bool online; |
11099 |
-+ |
11100 |
-+ unsigned short weight, ioprio, ioprio_class; |
11101 |
-+ |
11102 |
-+ spinlock_t lock; |
11103 |
-+ struct hlist_head group_data; |
11104 |
-+}; |
11105 |
-+#else |
11106 |
-+struct bfq_group { |
11107 |
-+ struct bfq_sched_data sched_data; |
11108 |
-+ |
11109 |
-+ struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR]; |
11110 |
-+ struct bfq_queue *async_idle_bfqq; |
11111 |
-+}; |
11112 |
-+#endif |
11113 |
-+ |
11114 |
-+static inline struct bfq_service_tree * |
11115 |
-+bfq_entity_service_tree(struct bfq_entity *entity) |
11116 |
-+{ |
11117 |
-+ struct bfq_sched_data *sched_data = entity->sched_data; |
11118 |
-+ unsigned int idx = entity->ioprio_class - 1; |
11119 |
-+ |
11120 |
-+ BUG_ON(idx >= BFQ_IOPRIO_CLASSES); |
11121 |
-+ BUG_ON(sched_data == NULL); |
11122 |
-+ |
11123 |
-+ return sched_data->service_tree + idx; |
11124 |
-+} |
11125 |
-+ |
11126 |
-+static inline struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, |
11127 |
-+ int is_sync) |
11128 |
-+{ |
11129 |
-+ return bic->bfqq[!!is_sync]; |
11130 |
-+} |
11131 |
-+ |
11132 |
-+static inline void bic_set_bfqq(struct bfq_io_cq *bic, |
11133 |
-+ struct bfq_queue *bfqq, int is_sync) |
11134 |
-+{ |
11135 |
-+ bic->bfqq[!!is_sync] = bfqq; |
11136 |
-+} |
11137 |
-+ |
11138 |
-+static inline struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic) |
11139 |
-+{ |
11140 |
-+ return bic->icq.q->elevator->elevator_data; |
11141 |
-+} |
11142 |
-+ |
11143 |
-+/** |
11144 |
-+ * bfq_get_bfqd_locked - get a lock to a bfqd using a RCU protected pointer. |
11145 |
-+ * @ptr: a pointer to a bfqd. |
11146 |
-+ * @flags: storage for the flags to be saved. |
11147 |
-+ * |
11148 |
-+ * This function allows bfqg->bfqd to be protected by the |
11149 |
-+ * queue lock of the bfqd they reference; the pointer is dereferenced |
11150 |
-+ * under RCU, so the storage for bfqd is assured to be safe as long |
11151 |
-+ * as the RCU read side critical section does not end. After the |
11152 |
-+ * bfqd->queue->queue_lock is taken the pointer is rechecked, to be |
11153 |
-+ * sure that no other writer accessed it. If we raced with a writer, |
11154 |
-+ * the function returns NULL, with the queue unlocked, otherwise it |
11155 |
-+ * returns the dereferenced pointer, with the queue locked. |
11156 |
-+ */ |
11157 |
-+static inline struct bfq_data *bfq_get_bfqd_locked(void **ptr, |
11158 |
-+ unsigned long *flags) |
11159 |
-+{ |
11160 |
-+ struct bfq_data *bfqd; |
11161 |
-+ |
11162 |
-+ rcu_read_lock(); |
11163 |
-+ bfqd = rcu_dereference(*(struct bfq_data **)ptr); |
11164 |
-+ |
11165 |
-+ if (bfqd != NULL) { |
11166 |
-+ spin_lock_irqsave(bfqd->queue->queue_lock, *flags); |
11167 |
-+ if (*ptr == bfqd) |
11168 |
-+ goto out; |
11169 |
-+ spin_unlock_irqrestore(bfqd->queue->queue_lock, *flags); |
11170 |
-+ } |
11171 |
-+ |
11172 |
-+ bfqd = NULL; |
11173 |
-+out: |
11174 |
-+ rcu_read_unlock(); |
11175 |
-+ return bfqd; |
11176 |
-+} |
11177 |
-+ |
11178 |
-+static inline void bfq_put_bfqd_unlock(struct bfq_data *bfqd, |
11179 |
-+ unsigned long *flags) |
11180 |
-+{ |
11181 |
-+ spin_unlock_irqrestore(bfqd->queue->queue_lock, *flags); |
11182 |
-+} |
11183 |
-+ |
11184 |
-+static void bfq_changed_ioprio(struct bfq_io_cq *bic); |
11185 |
-+static void bfq_put_queue(struct bfq_queue *bfqq); |
11186 |
-+static void bfq_dispatch_insert(struct request_queue *q, struct request *rq); |
11187 |
-+static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd, |
11188 |
-+ struct bfq_group *bfqg, int is_sync, |
11189 |
-+ struct bfq_io_cq *bic, gfp_t gfp_mask); |
11190 |
-+static void bfq_end_raising_async_queues(struct bfq_data *bfqd, |
11191 |
-+ struct bfq_group *bfqg); |
11192 |
-+static void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg); |
11193 |
-+static void bfq_exit_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq); |
11194 |
-+#endif |
11195 |
--- |
11196 |
-1.9.0 |
11197 |
- |
11198 |
|
11199 |
Deleted: genpatches-2.6/trunk/3.15/5003_BFQ-3-block-bfq-add-Early-Queue-Merge-EQM-to-BFQ-v7r2-for-3.14.0.patch |
11200 |
=================================================================== |
11201 |
--- genpatches-2.6/trunk/3.15/5003_BFQ-3-block-bfq-add-Early-Queue-Merge-EQM-to-BFQ-v7r2-for-3.14.0.patch 2014-05-05 14:22:27 UTC (rev 2771) |
11202 |
+++ genpatches-2.6/trunk/3.15/5003_BFQ-3-block-bfq-add-Early-Queue-Merge-EQM-to-BFQ-v7r2-for-3.14.0.patch 2014-05-05 14:28:48 UTC (rev 2772) |
11203 |
@@ -1,1038 +0,0 @@ |
11204 |
-From 4fbeda28a90d7fccd05d28a89d9fc409b2344e0a Mon Sep 17 00:00:00 2001 |
11205 |
-From: Mauro Andreolini <mauro.andreolini@×××××××.it> |
11206 |
-Date: Fri, 14 Feb 2014 12:52:49 +0100 |
11207 |
-Subject: [PATCH 3/3] block, bfq: add Early Queue Merge (EQM) to BFQ-v7r2 for |
11208 |
- 3.14.0 |
11209 |
- |
11210 |
-A set of processes may happen to perform interleaved reads, i.e., requests |
11211 |
-whose union would give rise to a sequential read pattern. There are two |
11212 |
-typical cases: in the first case, processes read fixed-size chunks of |
11213 |
-data at a fixed distance from each other, while in the second case processes |
11214 |
-may read variable-size chunks at variable distances. The latter case occurs |
11215 |
-for example with KVM, which splits the I/O generated by the guest into |
11216 |
-multiple chunks, and lets these chunks be served by a pool of cooperating |
11217 |
-processes, iteratively assigning the next chunk of I/O to the first |
11218 |
-available process. CFQ uses actual queue merging for the first type of |
11219 |
-rocesses, whereas it uses preemption to get a sequential read pattern out |
11220 |
-of the read requests performed by the second type of processes. In the end |
11221 |
-it uses two different mechanisms to achieve the same goal: boosting the |
11222 |
-throughput with interleaved I/O. |
11223 |
- |
11224 |
-This patch introduces Early Queue Merge (EQM), a unified mechanism to get a |
11225 |
-sequential read pattern with both types of processes. The main idea is |
11226 |
-checking newly arrived requests against the next request of the active queue |
11227 |
-both in case of actual request insert and in case of request merge. By doing |
11228 |
-so, both the types of processes can be handled by just merging their queues. |
11229 |
-EQM is then simpler and more compact than the pair of mechanisms used in |
11230 |
-CFQ. |
11231 |
- |
11232 |
-Finally, EQM also preserves the typical low-latency properties of BFQ, by |
11233 |
-properly restoring the weight-raising state of a queue when it gets back to |
11234 |
-a non-merged state. |
11235 |
- |
11236 |
-Signed-off-by: Mauro Andreolini <mauro.andreolini@×××××××.it> |
11237 |
-Signed-off-by: Arianna Avanzini <avanzini.arianna@×××××.com> |
11238 |
-Reviewed-by: Paolo Valente <paolo.valente@×××××××.it> |
11239 |
---- |
11240 |
- block/bfq-iosched.c | 657 ++++++++++++++++++++++++++++++++++++---------------- |
11241 |
- block/bfq-sched.c | 28 --- |
11242 |
- block/bfq.h | 20 +- |
11243 |
- 3 files changed, 476 insertions(+), 229 deletions(-) |
11244 |
- |
11245 |
-diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c |
11246 |
-index f5f71e4..0d3503d 100644 |
11247 |
---- a/block/bfq-iosched.c |
11248 |
-+++ b/block/bfq-iosched.c |
11249 |
-@@ -445,6 +445,46 @@ static inline unsigned int bfq_wrais_duration(struct bfq_data *bfqd) |
11250 |
- return dur; |
11251 |
- } |
11252 |
- |
11253 |
-+static inline void |
11254 |
-+bfq_bfqq_resume_state(struct bfq_queue *bfqq, struct bfq_io_cq *bic) |
11255 |
-+{ |
11256 |
-+ if (bic->saved_idle_window) |
11257 |
-+ bfq_mark_bfqq_idle_window(bfqq); |
11258 |
-+ else |
11259 |
-+ bfq_clear_bfqq_idle_window(bfqq); |
11260 |
-+ if (bic->raising_time_left && bfqq->bfqd->low_latency) { |
11261 |
-+ /* |
11262 |
-+ * Start a weight raising period with the duration given by |
11263 |
-+ * the raising_time_left snapshot. |
11264 |
-+ */ |
11265 |
-+ if (bfq_bfqq_busy(bfqq)) |
11266 |
-+ bfqq->bfqd->raised_busy_queues++; |
11267 |
-+ bfqq->raising_coeff = bfqq->bfqd->bfq_raising_coeff; |
11268 |
-+ bfqq->raising_cur_max_time = bic->raising_time_left; |
11269 |
-+ bfqq->last_rais_start_finish = jiffies; |
11270 |
-+ bfqq->entity.ioprio_changed = 1; |
11271 |
-+ } |
11272 |
-+ /* |
11273 |
-+ * Clear raising_time_left to prevent bfq_bfqq_save_state() from |
11274 |
-+ * getting confused about the queue's need of a weight-raising |
11275 |
-+ * period. |
11276 |
-+ */ |
11277 |
-+ bic->raising_time_left = 0; |
11278 |
-+} |
11279 |
-+ |
11280 |
-+/* |
11281 |
-+ * Must be called with the queue_lock held. |
11282 |
-+ */ |
11283 |
-+static int bfqq_process_refs(struct bfq_queue *bfqq) |
11284 |
-+{ |
11285 |
-+ int process_refs, io_refs; |
11286 |
-+ |
11287 |
-+ io_refs = bfqq->allocated[READ] + bfqq->allocated[WRITE]; |
11288 |
-+ process_refs = atomic_read(&bfqq->ref) - io_refs - bfqq->entity.on_st; |
11289 |
-+ BUG_ON(process_refs < 0); |
11290 |
-+ return process_refs; |
11291 |
-+} |
11292 |
-+ |
11293 |
- static void bfq_add_rq_rb(struct request *rq) |
11294 |
- { |
11295 |
- struct bfq_queue *bfqq = RQ_BFQQ(rq); |
11296 |
-@@ -486,12 +526,20 @@ static void bfq_add_rq_rb(struct request *rq) |
11297 |
- if (!bfqd->low_latency) |
11298 |
- goto add_bfqq_busy; |
11299 |
- |
11300 |
-+ if (bfq_bfqq_just_split(bfqq)) |
11301 |
-+ goto set_ioprio_changed; |
11302 |
-+ |
11303 |
- /* |
11304 |
-- * If the queue is not being boosted and has been idle |
11305 |
-- * for enough time, start a weight-raising period |
11306 |
-+ * If the queue: |
11307 |
-+ * - is not being boosted, |
11308 |
-+ * - has been idle for enough time, |
11309 |
-+ * - is not a sync queue or is linked to a bfq_io_cq (it is |
11310 |
-+ * shared "for its nature" or it is not shared and its |
11311 |
-+ * requests have not been redirected to a shared queue) |
11312 |
-+ * start a weight-raising period. |
11313 |
- */ |
11314 |
-- if (old_raising_coeff == 1 && |
11315 |
-- (idle_for_long_time || soft_rt)) { |
11316 |
-+ if (old_raising_coeff == 1 && (idle_for_long_time || soft_rt) && |
11317 |
-+ (!bfq_bfqq_sync(bfqq) || bfqq->bic != NULL)) { |
11318 |
- bfqq->raising_coeff = bfqd->bfq_raising_coeff; |
11319 |
- if (idle_for_long_time) |
11320 |
- bfqq->raising_cur_max_time = |
11321 |
-@@ -574,6 +622,7 @@ static void bfq_add_rq_rb(struct request *rq) |
11322 |
- bfqd->bfq_raising_rt_max_time; |
11323 |
- } |
11324 |
- } |
11325 |
-+set_ioprio_changed: |
11326 |
- if (old_raising_coeff != bfqq->raising_coeff) |
11327 |
- entity->ioprio_changed = 1; |
11328 |
- add_bfqq_busy: |
11329 |
-@@ -756,90 +805,35 @@ static void bfq_end_raising(struct bfq_data *bfqd) |
11330 |
- spin_unlock_irq(bfqd->queue->queue_lock); |
11331 |
- } |
11332 |
- |
11333 |
--static int bfq_allow_merge(struct request_queue *q, struct request *rq, |
11334 |
-- struct bio *bio) |
11335 |
--{ |
11336 |
-- struct bfq_data *bfqd = q->elevator->elevator_data; |
11337 |
-- struct bfq_io_cq *bic; |
11338 |
-- struct bfq_queue *bfqq; |
11339 |
-- |
11340 |
-- /* |
11341 |
-- * Disallow merge of a sync bio into an async request. |
11342 |
-- */ |
11343 |
-- if (bfq_bio_sync(bio) && !rq_is_sync(rq)) |
11344 |
-- return 0; |
11345 |
-- |
11346 |
-- /* |
11347 |
-- * Lookup the bfqq that this bio will be queued with. Allow |
11348 |
-- * merge only if rq is queued there. |
11349 |
-- * Queue lock is held here. |
11350 |
-- */ |
11351 |
-- bic = bfq_bic_lookup(bfqd, current->io_context); |
11352 |
-- if (bic == NULL) |
11353 |
-- return 0; |
11354 |
-- |
11355 |
-- bfqq = bic_to_bfqq(bic, bfq_bio_sync(bio)); |
11356 |
-- return bfqq == RQ_BFQQ(rq); |
11357 |
--} |
11358 |
-- |
11359 |
--static void __bfq_set_in_service_queue(struct bfq_data *bfqd, |
11360 |
-- struct bfq_queue *bfqq) |
11361 |
--{ |
11362 |
-- if (bfqq != NULL) { |
11363 |
-- bfq_mark_bfqq_must_alloc(bfqq); |
11364 |
-- bfq_mark_bfqq_budget_new(bfqq); |
11365 |
-- bfq_clear_bfqq_fifo_expire(bfqq); |
11366 |
-- |
11367 |
-- bfqd->budgets_assigned = (bfqd->budgets_assigned*7 + 256) / 8; |
11368 |
-- |
11369 |
-- bfq_log_bfqq(bfqd, bfqq, |
11370 |
-- "set_in_service_queue, cur-budget = %lu", |
11371 |
-- bfqq->entity.budget); |
11372 |
-- } |
11373 |
-- |
11374 |
-- bfqd->in_service_queue = bfqq; |
11375 |
--} |
11376 |
-- |
11377 |
--/* |
11378 |
-- * Get and set a new queue for service. |
11379 |
-- */ |
11380 |
--static struct bfq_queue *bfq_set_in_service_queue(struct bfq_data *bfqd, |
11381 |
-- struct bfq_queue *bfqq) |
11382 |
-+static inline sector_t bfq_io_struct_pos(void *io_struct, bool request) |
11383 |
- { |
11384 |
-- if (!bfqq) |
11385 |
-- bfqq = bfq_get_next_queue(bfqd); |
11386 |
-+ if (request) |
11387 |
-+ return blk_rq_pos(io_struct); |
11388 |
- else |
11389 |
-- bfq_get_next_queue_forced(bfqd, bfqq); |
11390 |
-- |
11391 |
-- __bfq_set_in_service_queue(bfqd, bfqq); |
11392 |
-- return bfqq; |
11393 |
-+ return ((struct bio *)io_struct)->bi_iter.bi_sector; |
11394 |
- } |
11395 |
- |
11396 |
--static inline sector_t bfq_dist_from_last(struct bfq_data *bfqd, |
11397 |
-- struct request *rq) |
11398 |
-+static inline sector_t bfq_dist_from(sector_t pos1, |
11399 |
-+ sector_t pos2) |
11400 |
- { |
11401 |
-- if (blk_rq_pos(rq) >= bfqd->last_position) |
11402 |
-- return blk_rq_pos(rq) - bfqd->last_position; |
11403 |
-+ if (pos1 >= pos2) |
11404 |
-+ return pos1 - pos2; |
11405 |
- else |
11406 |
-- return bfqd->last_position - blk_rq_pos(rq); |
11407 |
-+ return pos2 - pos1; |
11408 |
- } |
11409 |
- |
11410 |
--/* |
11411 |
-- * Return true if bfqq has no request pending and rq is close enough to |
11412 |
-- * bfqd->last_position, or if rq is closer to bfqd->last_position than |
11413 |
-- * bfqq->next_rq |
11414 |
-- */ |
11415 |
--static inline int bfq_rq_close(struct bfq_data *bfqd, struct request *rq) |
11416 |
-+static inline int bfq_rq_close_to_sector(void *io_struct, bool request, |
11417 |
-+ sector_t sector) |
11418 |
- { |
11419 |
-- return bfq_dist_from_last(bfqd, rq) <= BFQQ_SEEK_THR; |
11420 |
-+ return bfq_dist_from(bfq_io_struct_pos(io_struct, request), sector) <= |
11421 |
-+ BFQQ_SEEK_THR; |
11422 |
- } |
11423 |
- |
11424 |
--static struct bfq_queue *bfqq_close(struct bfq_data *bfqd) |
11425 |
-+static struct bfq_queue *bfqq_close(struct bfq_data *bfqd, sector_t sector) |
11426 |
- { |
11427 |
- struct rb_root *root = &bfqd->rq_pos_tree; |
11428 |
- struct rb_node *parent, *node; |
11429 |
- struct bfq_queue *__bfqq; |
11430 |
-- sector_t sector = bfqd->last_position; |
11431 |
- |
11432 |
- if (RB_EMPTY_ROOT(root)) |
11433 |
- return NULL; |
11434 |
-@@ -858,7 +852,7 @@ static struct bfq_queue *bfqq_close(struct bfq_data *bfqd) |
11435 |
- * position). |
11436 |
- */ |
11437 |
- __bfqq = rb_entry(parent, struct bfq_queue, pos_node); |
11438 |
-- if (bfq_rq_close(bfqd, __bfqq->next_rq)) |
11439 |
-+ if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector)) |
11440 |
- return __bfqq; |
11441 |
- |
11442 |
- if (blk_rq_pos(__bfqq->next_rq) < sector) |
11443 |
-@@ -869,7 +863,7 @@ static struct bfq_queue *bfqq_close(struct bfq_data *bfqd) |
11444 |
- return NULL; |
11445 |
- |
11446 |
- __bfqq = rb_entry(node, struct bfq_queue, pos_node); |
11447 |
-- if (bfq_rq_close(bfqd, __bfqq->next_rq)) |
11448 |
-+ if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector)) |
11449 |
- return __bfqq; |
11450 |
- |
11451 |
- return NULL; |
11452 |
-@@ -878,14 +872,12 @@ static struct bfq_queue *bfqq_close(struct bfq_data *bfqd) |
11453 |
- /* |
11454 |
- * bfqd - obvious |
11455 |
- * cur_bfqq - passed in so that we don't decide that the current queue |
11456 |
-- * is closely cooperating with itself. |
11457 |
-- * |
11458 |
-- * We are assuming that cur_bfqq has dispatched at least one request, |
11459 |
-- * and that bfqd->last_position reflects a position on the disk associated |
11460 |
-- * with the I/O issued by cur_bfqq. |
11461 |
-+ * is closely cooperating with itself |
11462 |
-+ * sector - used as a reference point to search for a close queue |
11463 |
- */ |
11464 |
- static struct bfq_queue *bfq_close_cooperator(struct bfq_data *bfqd, |
11465 |
-- struct bfq_queue *cur_bfqq) |
11466 |
-+ struct bfq_queue *cur_bfqq, |
11467 |
-+ sector_t sector) |
11468 |
- { |
11469 |
- struct bfq_queue *bfqq; |
11470 |
- |
11471 |
-@@ -905,7 +897,7 @@ static struct bfq_queue *bfq_close_cooperator(struct bfq_data *bfqd, |
11472 |
- * working closely on the same area of the disk. In that case, |
11473 |
- * we can group them together and don't waste time idling. |
11474 |
- */ |
11475 |
-- bfqq = bfqq_close(bfqd); |
11476 |
-+ bfqq = bfqq_close(bfqd, sector); |
11477 |
- if (bfqq == NULL || bfqq == cur_bfqq) |
11478 |
- return NULL; |
11479 |
- |
11480 |
-@@ -932,6 +924,282 @@ static struct bfq_queue *bfq_close_cooperator(struct bfq_data *bfqd, |
11481 |
- return bfqq; |
11482 |
- } |
11483 |
- |
11484 |
-+static struct bfq_queue * |
11485 |
-+bfq_setup_merge(struct bfq_queue *bfqq, struct bfq_queue *new_bfqq) |
11486 |
-+{ |
11487 |
-+ int process_refs, new_process_refs; |
11488 |
-+ struct bfq_queue *__bfqq; |
11489 |
-+ |
11490 |
-+ /* |
11491 |
-+ * If there are no process references on the new_bfqq, then it is |
11492 |
-+ * unsafe to follow the ->new_bfqq chain as other bfqq's in the chain |
11493 |
-+ * may have dropped their last reference (not just their last process |
11494 |
-+ * reference). |
11495 |
-+ */ |
11496 |
-+ if (!bfqq_process_refs(new_bfqq)) |
11497 |
-+ return NULL; |
11498 |
-+ |
11499 |
-+ /* Avoid a circular list and skip interim queue merges. */ |
11500 |
-+ while ((__bfqq = new_bfqq->new_bfqq)) { |
11501 |
-+ if (__bfqq == bfqq) |
11502 |
-+ return NULL; |
11503 |
-+ new_bfqq = __bfqq; |
11504 |
-+ } |
11505 |
-+ |
11506 |
-+ process_refs = bfqq_process_refs(bfqq); |
11507 |
-+ new_process_refs = bfqq_process_refs(new_bfqq); |
11508 |
-+ /* |
11509 |
-+ * If the process for the bfqq has gone away, there is no |
11510 |
-+ * sense in merging the queues. |
11511 |
-+ */ |
11512 |
-+ if (process_refs == 0 || new_process_refs == 0) |
11513 |
-+ return NULL; |
11514 |
-+ |
11515 |
-+ bfq_log_bfqq(bfqq->bfqd, bfqq, "scheduling merge with queue %d", |
11516 |
-+ new_bfqq->pid); |
11517 |
-+ |
11518 |
-+ /* |
11519 |
-+ * Merging is just a redirection: the requests of the process owning |
11520 |
-+ * one of the two queues are redirected to the other queue. The latter |
11521 |
-+ * queue, in its turn, is set as shared if this is the first time that |
11522 |
-+ * the requests of some process are redirected to it. |
11523 |
-+ * |
11524 |
-+ * We redirect bfqq to new_bfqq and not the opposite, because we |
11525 |
-+ * are in the context of the process owning bfqq, hence we have the |
11526 |
-+ * io_cq of this process. So we can immediately configure this io_cq |
11527 |
-+ * to redirect the requests of the process to new_bfqq. |
11528 |
-+ * |
11529 |
-+ * NOTE, even if new_bfqq coincides with the in-service queue, the |
11530 |
-+ * io_cq of new_bfqq is not available, because, if the in-service queue |
11531 |
-+ * is shared, bfqd->in_service_bic may not point to the io_cq of the |
11532 |
-+ * in-service queue. |
11533 |
-+ * Redirecting the requests of the process owning bfqq to the currently |
11534 |
-+ * in-service queue is in any case the best option, as we feed the |
11535 |
-+ * in-service queue with new requests close to the last request served |
11536 |
-+ * and, by doing so, hopefully increase the throughput. |
11537 |
-+ */ |
11538 |
-+ bfqq->new_bfqq = new_bfqq; |
11539 |
-+ atomic_add(process_refs, &new_bfqq->ref); |
11540 |
-+ return new_bfqq; |
11541 |
-+} |
11542 |
-+ |
11543 |
-+/* |
11544 |
-+ * Attempt to schedule a merge of bfqq with the currently in-service queue or |
11545 |
-+ * with a close queue among the scheduled queues. |
11546 |
-+ * Return NULL if no merge was scheduled, a pointer to the shared bfq_queue |
11547 |
-+ * structure otherwise. |
11548 |
-+ */ |
11549 |
-+static struct bfq_queue * |
11550 |
-+bfq_setup_cooperator(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
11551 |
-+ void *io_struct, bool request) |
11552 |
-+{ |
11553 |
-+ struct bfq_queue *in_service_bfqq, *new_bfqq; |
11554 |
-+ |
11555 |
-+ if (bfqq->new_bfqq) |
11556 |
-+ return bfqq->new_bfqq; |
11557 |
-+ |
11558 |
-+ if (!io_struct) |
11559 |
-+ return NULL; |
11560 |
-+ |
11561 |
-+ in_service_bfqq = bfqd->in_service_queue; |
11562 |
-+ |
11563 |
-+ if (in_service_bfqq == NULL || in_service_bfqq == bfqq || |
11564 |
-+ !bfqd->in_service_bic) |
11565 |
-+ goto check_scheduled; |
11566 |
-+ |
11567 |
-+ if (bfq_class_idle(in_service_bfqq) || bfq_class_idle(bfqq)) |
11568 |
-+ goto check_scheduled; |
11569 |
-+ |
11570 |
-+ if (bfq_class_rt(in_service_bfqq) != bfq_class_rt(bfqq)) |
11571 |
-+ goto check_scheduled; |
11572 |
-+ |
11573 |
-+ if (in_service_bfqq->entity.parent != bfqq->entity.parent) |
11574 |
-+ goto check_scheduled; |
11575 |
-+ |
11576 |
-+ if (bfq_rq_close_to_sector(io_struct, request, bfqd->last_position) && |
11577 |
-+ bfq_bfqq_sync(in_service_bfqq) && bfq_bfqq_sync(bfqq)) { |
11578 |
-+ new_bfqq = bfq_setup_merge(bfqq, in_service_bfqq); |
11579 |
-+ if (new_bfqq != NULL) |
11580 |
-+ return new_bfqq; /* Merge with the in-service queue */ |
11581 |
-+ } |
11582 |
-+ |
11583 |
-+ /* |
11584 |
-+ * Check whether there is a cooperator among currently scheduled |
11585 |
-+ * queues. The only thing we need is that the bio/request is not |
11586 |
-+ * NULL, as we need it to establish whether a cooperator exists. |
11587 |
-+ */ |
11588 |
-+check_scheduled: |
11589 |
-+ new_bfqq = bfq_close_cooperator(bfqd, bfqq, |
11590 |
-+ bfq_io_struct_pos(io_struct, request)); |
11591 |
-+ if (new_bfqq) |
11592 |
-+ return bfq_setup_merge(bfqq, new_bfqq); |
11593 |
-+ |
11594 |
-+ return NULL; |
11595 |
-+} |
11596 |
-+ |
11597 |
-+static inline void |
11598 |
-+bfq_bfqq_save_state(struct bfq_queue *bfqq) |
11599 |
-+{ |
11600 |
-+ /* |
11601 |
-+ * If bfqq->bic == NULL, the queue is already shared or its requests |
11602 |
-+ * have already been redirected to a shared queue; both idle window |
11603 |
-+ * and weight raising state have already been saved. Do nothing. |
11604 |
-+ */ |
11605 |
-+ if (bfqq->bic == NULL) |
11606 |
-+ return; |
11607 |
-+ if (bfqq->bic->raising_time_left) |
11608 |
-+ /* |
11609 |
-+ * This is the queue of a just-started process, and would |
11610 |
-+ * deserve weight raising: we set raising_time_left to the full |
11611 |
-+ * weight-raising duration to trigger weight-raising when and |
11612 |
-+ * if the queue is split and the first request of the queue |
11613 |
-+ * is enqueued. |
11614 |
-+ */ |
11615 |
-+ bfqq->bic->raising_time_left = bfq_wrais_duration(bfqq->bfqd); |
11616 |
-+ else if (bfqq->raising_coeff > 1) { |
11617 |
-+ unsigned long wrais_duration = |
11618 |
-+ jiffies - bfqq->last_rais_start_finish; |
11619 |
-+ /* |
11620 |
-+ * It may happen that a queue's weight raising period lasts |
11621 |
-+ * longer than its raising_cur_max_time, as weight raising is |
11622 |
-+ * handled only when a request is enqueued or dispatched (it |
11623 |
-+ * does not use any timer). If the weight raising period is |
11624 |
-+ * about to end, don't save it. |
11625 |
-+ */ |
11626 |
-+ if (bfqq->raising_cur_max_time <= wrais_duration) |
11627 |
-+ bfqq->bic->raising_time_left = 0; |
11628 |
-+ else |
11629 |
-+ bfqq->bic->raising_time_left = |
11630 |
-+ bfqq->raising_cur_max_time - wrais_duration; |
11631 |
-+ /* |
11632 |
-+ * The bfq_queue is becoming shared or the requests of the |
11633 |
-+ * process owning the queue are being redirected to a shared |
11634 |
-+ * queue. Stop the weight raising period of the queue, as in |
11635 |
-+ * both cases it should not be owned by an interactive or soft |
11636 |
-+ * real-time application. |
11637 |
-+ */ |
11638 |
-+ bfq_bfqq_end_raising(bfqq); |
11639 |
-+ } else |
11640 |
-+ bfqq->bic->raising_time_left = 0; |
11641 |
-+ bfqq->bic->saved_idle_window = bfq_bfqq_idle_window(bfqq); |
11642 |
-+} |
11643 |
-+ |
11644 |
-+static inline void |
11645 |
-+bfq_get_bic_reference(struct bfq_queue *bfqq) |
11646 |
-+{ |
11647 |
-+ /* |
11648 |
-+ * If bfqq->bic has a non-NULL value, the bic to which it belongs |
11649 |
-+ * is about to begin using a shared bfq_queue. |
11650 |
-+ */ |
11651 |
-+ if (bfqq->bic) |
11652 |
-+ atomic_long_inc(&bfqq->bic->icq.ioc->refcount); |
11653 |
-+} |
11654 |
-+ |
11655 |
-+static void |
11656 |
-+bfq_merge_bfqqs(struct bfq_data *bfqd, struct bfq_io_cq *bic, |
11657 |
-+ struct bfq_queue *bfqq, struct bfq_queue *new_bfqq) |
11658 |
-+{ |
11659 |
-+ bfq_log_bfqq(bfqd, bfqq, "merging with queue %lu", |
11660 |
-+ (long unsigned)new_bfqq->pid); |
11661 |
-+ /* Save weight raising and idle window of the merged queues */ |
11662 |
-+ bfq_bfqq_save_state(bfqq); |
11663 |
-+ bfq_bfqq_save_state(new_bfqq); |
11664 |
-+ /* |
11665 |
-+ * Grab a reference to the bic, to prevent it from being destroyed |
11666 |
-+ * before being possibly touched by a bfq_split_bfqq(). |
11667 |
-+ */ |
11668 |
-+ bfq_get_bic_reference(bfqq); |
11669 |
-+ bfq_get_bic_reference(new_bfqq); |
11670 |
-+ /* Merge queues (that is, let bic redirect its requests to new_bfqq) */ |
11671 |
-+ bic_set_bfqq(bic, new_bfqq, 1); |
11672 |
-+ bfq_mark_bfqq_coop(new_bfqq); |
11673 |
-+ /* |
11674 |
-+ * new_bfqq now belongs to at least two bics (it is a shared queue): set |
11675 |
-+ * new_bfqq->bic to NULL. bfqq either: |
11676 |
-+ * - does not belong to any bic any more, and hence bfqq->bic must |
11677 |
-+ * be set to NULL, or |
11678 |
-+ * - is a queue whose owning bics have already been redirected to a |
11679 |
-+ * different queue, hence the queue is destined to not belong to any |
11680 |
-+ * bic soon and bfqq->bic is already NULL (therefore the next |
11681 |
-+ * assignment causes no harm). |
11682 |
-+ */ |
11683 |
-+ new_bfqq->bic = NULL; |
11684 |
-+ bfqq->bic = NULL; |
11685 |
-+ bfq_put_queue(bfqq); |
11686 |
-+} |
11687 |
-+ |
11688 |
-+static int bfq_allow_merge(struct request_queue *q, struct request *rq, |
11689 |
-+ struct bio *bio) |
11690 |
-+{ |
11691 |
-+ struct bfq_data *bfqd = q->elevator->elevator_data; |
11692 |
-+ struct bfq_io_cq *bic; |
11693 |
-+ struct bfq_queue *bfqq, *new_bfqq; |
11694 |
-+ |
11695 |
-+ /* |
11696 |
-+ * Disallow merge of a sync bio into an async request. |
11697 |
-+ */ |
11698 |
-+ if (bfq_bio_sync(bio) && !rq_is_sync(rq)) |
11699 |
-+ return 0; |
11700 |
-+ |
11701 |
-+ /* |
11702 |
-+ * Lookup the bfqq that this bio will be queued with. Allow |
11703 |
-+ * merge only if rq is queued there. |
11704 |
-+ * Queue lock is held here. |
11705 |
-+ */ |
11706 |
-+ bic = bfq_bic_lookup(bfqd, current->io_context); |
11707 |
-+ if (bic == NULL) |
11708 |
-+ return 0; |
11709 |
-+ |
11710 |
-+ bfqq = bic_to_bfqq(bic, bfq_bio_sync(bio)); |
11711 |
-+ /* |
11712 |
-+ * We take advantage of this function to perform an early merge |
11713 |
-+ * of the queues of possible cooperating processes. |
11714 |
-+ */ |
11715 |
-+ if (bfqq != NULL) { |
11716 |
-+ new_bfqq = bfq_setup_cooperator(bfqd, bfqq, bio, false); |
11717 |
-+ if (new_bfqq != NULL) { |
11718 |
-+ bfq_merge_bfqqs(bfqd, bic, bfqq, new_bfqq); |
11719 |
-+ /* |
11720 |
-+ * If we get here, the bio will be queued in the shared |
11721 |
-+ * queue, i.e., new_bfqq, so use new_bfqq to decide |
11722 |
-+ * whether bio and rq can be merged. |
11723 |
-+ */ |
11724 |
-+ bfqq = new_bfqq; |
11725 |
-+ } |
11726 |
-+ } |
11727 |
-+ |
11728 |
-+ return bfqq == RQ_BFQQ(rq); |
11729 |
-+} |
11730 |
-+ |
11731 |
-+static void __bfq_set_in_service_queue(struct bfq_data *bfqd, |
11732 |
-+ struct bfq_queue *bfqq) |
11733 |
-+{ |
11734 |
-+ if (bfqq != NULL) { |
11735 |
-+ bfq_mark_bfqq_must_alloc(bfqq); |
11736 |
-+ bfq_mark_bfqq_budget_new(bfqq); |
11737 |
-+ bfq_clear_bfqq_fifo_expire(bfqq); |
11738 |
-+ |
11739 |
-+ bfqd->budgets_assigned = (bfqd->budgets_assigned*7 + 256) / 8; |
11740 |
-+ |
11741 |
-+ bfq_log_bfqq(bfqd, bfqq, |
11742 |
-+ "set_in_service_queue, cur-budget = %lu", |
11743 |
-+ bfqq->entity.budget); |
11744 |
-+ } |
11745 |
-+ |
11746 |
-+ bfqd->in_service_queue = bfqq; |
11747 |
-+} |
11748 |
-+ |
11749 |
-+/* |
11750 |
-+ * Get and set a new queue for service. |
11751 |
-+ */ |
11752 |
-+static struct bfq_queue *bfq_set_in_service_queue(struct bfq_data *bfqd) |
11753 |
-+{ |
11754 |
-+ struct bfq_queue *bfqq = bfq_get_next_queue(bfqd); |
11755 |
-+ |
11756 |
-+ __bfq_set_in_service_queue(bfqd, bfqq); |
11757 |
-+ return bfqq; |
11758 |
-+} |
11759 |
-+ |
11760 |
- /* |
11761 |
- * If enough samples have been computed, return the current max budget |
11762 |
- * stored in bfqd, which is dynamically updated according to the |
11763 |
-@@ -1079,63 +1347,6 @@ static struct request *bfq_check_fifo(struct bfq_queue *bfqq) |
11764 |
- return rq; |
11765 |
- } |
11766 |
- |
11767 |
--/* |
11768 |
-- * Must be called with the queue_lock held. |
11769 |
-- */ |
11770 |
--static int bfqq_process_refs(struct bfq_queue *bfqq) |
11771 |
--{ |
11772 |
-- int process_refs, io_refs; |
11773 |
-- |
11774 |
-- io_refs = bfqq->allocated[READ] + bfqq->allocated[WRITE]; |
11775 |
-- process_refs = atomic_read(&bfqq->ref) - io_refs - bfqq->entity.on_st; |
11776 |
-- BUG_ON(process_refs < 0); |
11777 |
-- return process_refs; |
11778 |
--} |
11779 |
-- |
11780 |
--static void bfq_setup_merge(struct bfq_queue *bfqq, struct bfq_queue *new_bfqq) |
11781 |
--{ |
11782 |
-- int process_refs, new_process_refs; |
11783 |
-- struct bfq_queue *__bfqq; |
11784 |
-- |
11785 |
-- /* |
11786 |
-- * If there are no process references on the new_bfqq, then it is |
11787 |
-- * unsafe to follow the ->new_bfqq chain as other bfqq's in the chain |
11788 |
-- * may have dropped their last reference (not just their last process |
11789 |
-- * reference). |
11790 |
-- */ |
11791 |
-- if (!bfqq_process_refs(new_bfqq)) |
11792 |
-- return; |
11793 |
-- |
11794 |
-- /* Avoid a circular list and skip interim queue merges. */ |
11795 |
-- while ((__bfqq = new_bfqq->new_bfqq)) { |
11796 |
-- if (__bfqq == bfqq) |
11797 |
-- return; |
11798 |
-- new_bfqq = __bfqq; |
11799 |
-- } |
11800 |
-- |
11801 |
-- process_refs = bfqq_process_refs(bfqq); |
11802 |
-- new_process_refs = bfqq_process_refs(new_bfqq); |
11803 |
-- /* |
11804 |
-- * If the process for the bfqq has gone away, there is no |
11805 |
-- * sense in merging the queues. |
11806 |
-- */ |
11807 |
-- if (process_refs == 0 || new_process_refs == 0) |
11808 |
-- return; |
11809 |
-- |
11810 |
-- /* |
11811 |
-- * Merge in the direction of the lesser amount of work. |
11812 |
-- */ |
11813 |
-- if (new_process_refs >= process_refs) { |
11814 |
-- bfqq->new_bfqq = new_bfqq; |
11815 |
-- atomic_add(process_refs, &new_bfqq->ref); |
11816 |
-- } else { |
11817 |
-- new_bfqq->new_bfqq = bfqq; |
11818 |
-- atomic_add(new_process_refs, &bfqq->ref); |
11819 |
-- } |
11820 |
-- bfq_log_bfqq(bfqq->bfqd, bfqq, "scheduling merge with queue %d", |
11821 |
-- new_bfqq->pid); |
11822 |
--} |
11823 |
-- |
11824 |
- static inline unsigned long bfq_bfqq_budget_left(struct bfq_queue *bfqq) |
11825 |
- { |
11826 |
- struct bfq_entity *entity = &bfqq->entity; |
11827 |
-@@ -1729,7 +1940,7 @@ static inline bool bfq_bfqq_must_idle(struct bfq_queue *bfqq) |
11828 |
- */ |
11829 |
- static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd) |
11830 |
- { |
11831 |
-- struct bfq_queue *bfqq, *new_bfqq = NULL; |
11832 |
-+ struct bfq_queue *bfqq; |
11833 |
- struct request *next_rq; |
11834 |
- enum bfqq_expiration reason = BFQ_BFQQ_BUDGET_TIMEOUT; |
11835 |
- |
11836 |
-@@ -1739,17 +1950,6 @@ static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd) |
11837 |
- |
11838 |
- bfq_log_bfqq(bfqd, bfqq, "select_queue: already in-service queue"); |
11839 |
- |
11840 |
-- /* |
11841 |
-- * If another queue has a request waiting within our mean seek |
11842 |
-- * distance, let it run. The expire code will check for close |
11843 |
-- * cooperators and put the close queue at the front of the |
11844 |
-- * service tree. If possible, merge the expiring queue with the |
11845 |
-- * new bfqq. |
11846 |
-- */ |
11847 |
-- new_bfqq = bfq_close_cooperator(bfqd, bfqq); |
11848 |
-- if (new_bfqq != NULL && bfqq->new_bfqq == NULL) |
11849 |
-- bfq_setup_merge(bfqq, new_bfqq); |
11850 |
-- |
11851 |
- if (bfq_may_expire_for_budg_timeout(bfqq) && |
11852 |
- !timer_pending(&bfqd->idle_slice_timer) && |
11853 |
- !bfq_bfqq_must_idle(bfqq)) |
11854 |
-@@ -1786,36 +1986,26 @@ static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd) |
11855 |
- bfq_clear_bfqq_wait_request(bfqq); |
11856 |
- del_timer(&bfqd->idle_slice_timer); |
11857 |
- } |
11858 |
-- if (new_bfqq == NULL) |
11859 |
-- goto keep_queue; |
11860 |
-- else |
11861 |
-- goto expire; |
11862 |
-+ goto keep_queue; |
11863 |
- } |
11864 |
- } |
11865 |
- |
11866 |
- /* |
11867 |
-- * No requests pending. If the in-service queue has no cooperator and |
11868 |
-- * still has requests in flight (possibly waiting for a completion) |
11869 |
-- * or is idling for a new request, then keep it. |
11870 |
-+ * No requests pending. If the in-service queue still has requests in |
11871 |
-+ * flight (possibly waiting for a completion) or is idling for a new |
11872 |
-+ * request, then keep it. |
11873 |
- */ |
11874 |
-- if (new_bfqq == NULL && (timer_pending(&bfqd->idle_slice_timer) || |
11875 |
-- (bfqq->dispatched != 0 && bfq_bfqq_must_not_expire(bfqq)))) { |
11876 |
-+ if (timer_pending(&bfqd->idle_slice_timer) || |
11877 |
-+ (bfqq->dispatched != 0 && bfq_bfqq_must_not_expire(bfqq))) { |
11878 |
- bfqq = NULL; |
11879 |
- goto keep_queue; |
11880 |
-- } else if (new_bfqq != NULL && timer_pending(&bfqd->idle_slice_timer)) { |
11881 |
-- /* |
11882 |
-- * Expiring the queue because there is a close cooperator, |
11883 |
-- * cancel timer. |
11884 |
-- */ |
11885 |
-- bfq_clear_bfqq_wait_request(bfqq); |
11886 |
-- del_timer(&bfqd->idle_slice_timer); |
11887 |
- } |
11888 |
- |
11889 |
- reason = BFQ_BFQQ_NO_MORE_REQUESTS; |
11890 |
- expire: |
11891 |
- bfq_bfqq_expire(bfqd, bfqq, 0, reason); |
11892 |
- new_queue: |
11893 |
-- bfqq = bfq_set_in_service_queue(bfqd, new_bfqq); |
11894 |
-+ bfqq = bfq_set_in_service_queue(bfqd); |
11895 |
- bfq_log(bfqd, "select_queue: new queue %d returned", |
11896 |
- bfqq != NULL ? bfqq->pid : 0); |
11897 |
- keep_queue: |
11898 |
-@@ -1825,9 +2015,8 @@ keep_queue: |
11899 |
- static void bfq_update_raising_data(struct bfq_data *bfqd, |
11900 |
- struct bfq_queue *bfqq) |
11901 |
- { |
11902 |
-+ struct bfq_entity *entity = &bfqq->entity; |
11903 |
- if (bfqq->raising_coeff > 1) { /* queue is being boosted */ |
11904 |
-- struct bfq_entity *entity = &bfqq->entity; |
11905 |
-- |
11906 |
- bfq_log_bfqq(bfqd, bfqq, |
11907 |
- "raising period dur %u/%u msec, " |
11908 |
- "old raising coeff %u, w %d(%d)", |
11909 |
-@@ -1844,7 +2033,7 @@ static void bfq_update_raising_data(struct bfq_data *bfqd, |
11910 |
- "WARN: pending prio change"); |
11911 |
- /* |
11912 |
- * If too much time has elapsed from the beginning |
11913 |
-- * of this weight-raising, stop it. |
11914 |
-+ * of this weight-raising period, stop it. |
11915 |
- */ |
11916 |
- if (time_is_before_jiffies(bfqq->last_rais_start_finish + |
11917 |
- bfqq->raising_cur_max_time)) { |
11918 |
-@@ -1856,11 +2045,13 @@ static void bfq_update_raising_data(struct bfq_data *bfqd, |
11919 |
- jiffies_to_msecs(bfqq-> |
11920 |
- raising_cur_max_time)); |
11921 |
- bfq_bfqq_end_raising(bfqq); |
11922 |
-- __bfq_entity_update_weight_prio( |
11923 |
-- bfq_entity_service_tree(entity), |
11924 |
-- entity); |
11925 |
- } |
11926 |
- } |
11927 |
-+ /* Update weight both if it must be raised and if it must be lowered */ |
11928 |
-+ if ((entity->weight > entity->orig_weight) != (bfqq->raising_coeff > 1)) |
11929 |
-+ __bfq_entity_update_weight_prio( |
11930 |
-+ bfq_entity_service_tree(entity), |
11931 |
-+ entity); |
11932 |
- } |
11933 |
- |
11934 |
- /* |
11935 |
-@@ -2101,6 +2292,25 @@ static void bfq_init_icq(struct io_cq *icq) |
11936 |
- struct bfq_io_cq *bic = icq_to_bic(icq); |
11937 |
- |
11938 |
- bic->ttime.last_end_request = jiffies; |
11939 |
-+ /* |
11940 |
-+ * A newly created bic indicates that the process has just |
11941 |
-+ * started doing I/O, and is probably mapping into memory its |
11942 |
-+ * executable and libraries: it definitely needs weight raising. |
11943 |
-+ * There is however the possibility that the process performs, |
11944 |
-+ * for a while, I/O close to some other process. EQM intercepts |
11945 |
-+ * this behavior and may merge the queue corresponding to the |
11946 |
-+ * process with some other queue, BEFORE the weight of the queue |
11947 |
-+ * is raised. Merged queues are not weight-raised (they are assumed |
11948 |
-+ * to belong to processes that benefit only from high throughput). |
11949 |
-+ * If the merge is basically the consequence of an accident, then |
11950 |
-+ * the queue will be split soon and will get back its old weight. |
11951 |
-+ * It is then important to write down somewhere that this queue |
11952 |
-+ * does need weight raising, even if it did not make it to get its |
11953 |
-+ * weight raised before being merged. To this purpose, we overload |
11954 |
-+ * the field raising_time_left and assign 1 to it, to mark the queue |
11955 |
-+ * as needing weight raising. |
11956 |
-+ */ |
11957 |
-+ bic->raising_time_left = 1; |
11958 |
- } |
11959 |
- |
11960 |
- static void bfq_exit_icq(struct io_cq *icq) |
11961 |
-@@ -2114,6 +2324,13 @@ static void bfq_exit_icq(struct io_cq *icq) |
11962 |
- } |
11963 |
- |
11964 |
- if (bic->bfqq[BLK_RW_SYNC]) { |
11965 |
-+ /* |
11966 |
-+ * If the bic is using a shared queue, put the reference |
11967 |
-+ * taken on the io_context when the bic started using a |
11968 |
-+ * shared bfq_queue. |
11969 |
-+ */ |
11970 |
-+ if (bfq_bfqq_coop(bic->bfqq[BLK_RW_SYNC])) |
11971 |
-+ put_io_context(icq->ioc); |
11972 |
- bfq_exit_bfqq(bfqd, bic->bfqq[BLK_RW_SYNC]); |
11973 |
- bic->bfqq[BLK_RW_SYNC] = NULL; |
11974 |
- } |
11975 |
-@@ -2405,6 +2622,10 @@ static void bfq_update_idle_window(struct bfq_data *bfqd, |
11976 |
- if (!bfq_bfqq_sync(bfqq) || bfq_class_idle(bfqq)) |
11977 |
- return; |
11978 |
- |
11979 |
-+ /* Idle window just restored, statistics are meaningless. */ |
11980 |
-+ if (bfq_bfqq_just_split(bfqq)) |
11981 |
-+ return; |
11982 |
-+ |
11983 |
- enable_idle = bfq_bfqq_idle_window(bfqq); |
11984 |
- |
11985 |
- if (atomic_read(&bic->icq.ioc->active_ref) == 0 || |
11986 |
-@@ -2445,6 +2666,7 @@ static void bfq_rq_enqueued(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
11987 |
- if (bfqq->entity.service > bfq_max_budget(bfqd) / 8 || |
11988 |
- !BFQQ_SEEKY(bfqq)) |
11989 |
- bfq_update_idle_window(bfqd, bfqq, bic); |
11990 |
-+ bfq_clear_bfqq_just_split(bfqq); |
11991 |
- |
11992 |
- bfq_log_bfqq(bfqd, bfqq, |
11993 |
- "rq_enqueued: idle_window=%d (seeky %d, mean %llu)", |
11994 |
-@@ -2505,13 +2727,48 @@ static void bfq_rq_enqueued(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
11995 |
- static void bfq_insert_request(struct request_queue *q, struct request *rq) |
11996 |
- { |
11997 |
- struct bfq_data *bfqd = q->elevator->elevator_data; |
11998 |
-- struct bfq_queue *bfqq = RQ_BFQQ(rq); |
11999 |
-+ struct bfq_queue *bfqq = RQ_BFQQ(rq), *new_bfqq; |
12000 |
- |
12001 |
- assert_spin_locked(bfqd->queue->queue_lock); |
12002 |
-+ |
12003 |
-+ /* |
12004 |
-+ * An unplug may trigger a requeue of a request from the device |
12005 |
-+ * driver: make sure we are in process context while trying to |
12006 |
-+ * merge two bfq_queues. |
12007 |
-+ */ |
12008 |
-+ if (!in_interrupt()) { |
12009 |
-+ new_bfqq = bfq_setup_cooperator(bfqd, bfqq, rq, true); |
12010 |
-+ if (new_bfqq != NULL) { |
12011 |
-+ if (bic_to_bfqq(RQ_BIC(rq), 1) != bfqq) |
12012 |
-+ new_bfqq = bic_to_bfqq(RQ_BIC(rq), 1); |
12013 |
-+ /* |
12014 |
-+ * Release the request's reference to the old bfqq |
12015 |
-+ * and make sure one is taken to the shared queue. |
12016 |
-+ */ |
12017 |
-+ new_bfqq->allocated[rq_data_dir(rq)]++; |
12018 |
-+ bfqq->allocated[rq_data_dir(rq)]--; |
12019 |
-+ atomic_inc(&new_bfqq->ref); |
12020 |
-+ bfq_put_queue(bfqq); |
12021 |
-+ if (bic_to_bfqq(RQ_BIC(rq), 1) == bfqq) |
12022 |
-+ bfq_merge_bfqqs(bfqd, RQ_BIC(rq), |
12023 |
-+ bfqq, new_bfqq); |
12024 |
-+ rq->elv.priv[1] = new_bfqq; |
12025 |
-+ bfqq = new_bfqq; |
12026 |
-+ } |
12027 |
-+ } |
12028 |
-+ |
12029 |
- bfq_init_prio_data(bfqq, RQ_BIC(rq)); |
12030 |
- |
12031 |
- bfq_add_rq_rb(rq); |
12032 |
- |
12033 |
-+ /* |
12034 |
-+ * Here a newly-created bfq_queue has already started a weight-raising |
12035 |
-+ * period: clear raising_time_left to prevent bfq_bfqq_save_state() |
12036 |
-+ * from assigning it a full weight-raising period. See the detailed |
12037 |
-+ * comments about this field in bfq_init_icq(). |
12038 |
-+ */ |
12039 |
-+ if (bfqq->bic != NULL) |
12040 |
-+ bfqq->bic->raising_time_left = 0; |
12041 |
- rq_set_fifo_time(rq, jiffies + bfqd->bfq_fifo_expire[rq_is_sync(rq)]); |
12042 |
- list_add_tail(&rq->queuelist, &bfqq->fifo); |
12043 |
- |
12044 |
-@@ -2663,18 +2920,6 @@ static void bfq_put_request(struct request *rq) |
12045 |
- } |
12046 |
- } |
12047 |
- |
12048 |
--static struct bfq_queue * |
12049 |
--bfq_merge_bfqqs(struct bfq_data *bfqd, struct bfq_io_cq *bic, |
12050 |
-- struct bfq_queue *bfqq) |
12051 |
--{ |
12052 |
-- bfq_log_bfqq(bfqd, bfqq, "merging with queue %lu", |
12053 |
-- (long unsigned)bfqq->new_bfqq->pid); |
12054 |
-- bic_set_bfqq(bic, bfqq->new_bfqq, 1); |
12055 |
-- bfq_mark_bfqq_coop(bfqq->new_bfqq); |
12056 |
-- bfq_put_queue(bfqq); |
12057 |
-- return bic_to_bfqq(bic, 1); |
12058 |
--} |
12059 |
-- |
12060 |
- /* |
12061 |
- * Returns NULL if a new bfqq should be allocated, or the old bfqq if this |
12062 |
- * was the last process referring to said bfqq. |
12063 |
-@@ -2683,6 +2928,9 @@ static struct bfq_queue * |
12064 |
- bfq_split_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq) |
12065 |
- { |
12066 |
- bfq_log_bfqq(bfqq->bfqd, bfqq, "splitting queue"); |
12067 |
-+ |
12068 |
-+ put_io_context(bic->icq.ioc); |
12069 |
-+ |
12070 |
- if (bfqq_process_refs(bfqq) == 1) { |
12071 |
- bfqq->pid = current->pid; |
12072 |
- bfq_clear_bfqq_coop(bfqq); |
12073 |
-@@ -2711,6 +2959,7 @@ static int bfq_set_request(struct request_queue *q, struct request *rq, |
12074 |
- struct bfq_queue *bfqq; |
12075 |
- struct bfq_group *bfqg; |
12076 |
- unsigned long flags; |
12077 |
-+ bool split = false; |
12078 |
- |
12079 |
- might_sleep_if(gfp_mask & __GFP_WAIT); |
12080 |
- |
12081 |
-@@ -2729,24 +2978,14 @@ new_queue: |
12082 |
- bfqq = bfq_get_queue(bfqd, bfqg, is_sync, bic, gfp_mask); |
12083 |
- bic_set_bfqq(bic, bfqq, is_sync); |
12084 |
- } else { |
12085 |
-- /* |
12086 |
-- * If the queue was seeky for too long, break it apart. |
12087 |
-- */ |
12088 |
-+ /* If the queue was seeky for too long, break it apart. */ |
12089 |
- if (bfq_bfqq_coop(bfqq) && bfq_bfqq_split_coop(bfqq)) { |
12090 |
- bfq_log_bfqq(bfqd, bfqq, "breaking apart bfqq"); |
12091 |
- bfqq = bfq_split_bfqq(bic, bfqq); |
12092 |
-+ split = true; |
12093 |
- if (!bfqq) |
12094 |
- goto new_queue; |
12095 |
- } |
12096 |
-- |
12097 |
-- /* |
12098 |
-- * Check to see if this queue is scheduled to merge with |
12099 |
-- * another closely cooperating queue. The merging of queues |
12100 |
-- * happens here as it must be done in process context. |
12101 |
-- * The reference on new_bfqq was taken in merge_bfqqs. |
12102 |
-- */ |
12103 |
-- if (bfqq->new_bfqq != NULL) |
12104 |
-- bfqq = bfq_merge_bfqqs(bfqd, bic, bfqq); |
12105 |
- } |
12106 |
- |
12107 |
- bfqq->allocated[rw]++; |
12108 |
-@@ -2757,6 +2996,26 @@ new_queue: |
12109 |
- rq->elv.priv[0] = bic; |
12110 |
- rq->elv.priv[1] = bfqq; |
12111 |
- |
12112 |
-+ /* |
12113 |
-+ * If a bfq_queue has only one process reference, it is owned |
12114 |
-+ * by only one bfq_io_cq: we can set the bic field of the |
12115 |
-+ * bfq_queue to the address of that structure. Also, if the |
12116 |
-+ * queue has just been split, mark a flag so that the |
12117 |
-+ * information is available to the other scheduler hooks. |
12118 |
-+ */ |
12119 |
-+ if (bfqq_process_refs(bfqq) == 1) { |
12120 |
-+ bfqq->bic = bic; |
12121 |
-+ if (split) { |
12122 |
-+ bfq_mark_bfqq_just_split(bfqq); |
12123 |
-+ /* |
12124 |
-+ * If the queue has just been split from a shared queue, |
12125 |
-+ * restore the idle window and the possible weight |
12126 |
-+ * raising period. |
12127 |
-+ */ |
12128 |
-+ bfq_bfqq_resume_state(bfqq, bic); |
12129 |
-+ } |
12130 |
-+ } |
12131 |
-+ |
12132 |
- spin_unlock_irqrestore(q->queue_lock, flags); |
12133 |
- |
12134 |
- return 0; |
12135 |
-diff --git a/block/bfq-sched.c b/block/bfq-sched.c |
12136 |
-index 999b475..e54ea33 100644 |
12137 |
---- a/block/bfq-sched.c |
12138 |
-+++ b/block/bfq-sched.c |
12139 |
-@@ -980,34 +980,6 @@ static struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd) |
12140 |
- return bfqq; |
12141 |
- } |
12142 |
- |
12143 |
--/* |
12144 |
-- * Forced extraction of the given queue. |
12145 |
-- */ |
12146 |
--static void bfq_get_next_queue_forced(struct bfq_data *bfqd, |
12147 |
-- struct bfq_queue *bfqq) |
12148 |
--{ |
12149 |
-- struct bfq_entity *entity; |
12150 |
-- struct bfq_sched_data *sd; |
12151 |
-- |
12152 |
-- BUG_ON(bfqd->in_service_queue != NULL); |
12153 |
-- |
12154 |
-- entity = &bfqq->entity; |
12155 |
-- /* |
12156 |
-- * Bubble up extraction/update from the leaf to the root. |
12157 |
-- */ |
12158 |
-- for_each_entity(entity) { |
12159 |
-- sd = entity->sched_data; |
12160 |
-- bfq_update_budget(entity); |
12161 |
-- bfq_update_vtime(bfq_entity_service_tree(entity)); |
12162 |
-- bfq_active_extract(bfq_entity_service_tree(entity), entity); |
12163 |
-- sd->active_entity = entity; |
12164 |
-- sd->next_active = NULL; |
12165 |
-- entity->service = 0; |
12166 |
-- } |
12167 |
-- |
12168 |
-- return; |
12169 |
--} |
12170 |
-- |
12171 |
- static void __bfq_bfqd_reset_in_service(struct bfq_data *bfqd) |
12172 |
- { |
12173 |
- if (bfqd->in_service_bic != NULL) { |
12174 |
-diff --git a/block/bfq.h b/block/bfq.h |
12175 |
-index 3ca8482..c278796 100644 |
12176 |
---- a/block/bfq.h |
12177 |
-+++ b/block/bfq.h |
12178 |
-@@ -200,6 +200,8 @@ struct bfq_group; |
12179 |
- * idle to backlogged |
12180 |
- * @service_from_backlogged: cumulative service received from the @bfq_queue |
12181 |
- * since the last transition from idle to backlogged |
12182 |
-+ * @bic: pointer to the bfq_io_cq owning the bfq_queue, set to %NULL if the |
12183 |
-+ * queue is shared |
12184 |
- * |
12185 |
- * A bfq_queue is a leaf request queue; it can be associated with an io_context |
12186 |
- * or more, if it is async or shared between cooperating processes. @cgroup |
12187 |
-@@ -243,6 +245,7 @@ struct bfq_queue { |
12188 |
- sector_t last_request_pos; |
12189 |
- |
12190 |
- pid_t pid; |
12191 |
-+ struct bfq_io_cq *bic; |
12192 |
- |
12193 |
- /* weight-raising fields */ |
12194 |
- unsigned long raising_cur_max_time; |
12195 |
-@@ -272,12 +275,23 @@ struct bfq_ttime { |
12196 |
- * @icq: associated io_cq structure |
12197 |
- * @bfqq: array of two process queues, the sync and the async |
12198 |
- * @ttime: associated @bfq_ttime struct |
12199 |
-+ * @raising_time_left: snapshot of the time left before weight raising ends |
12200 |
-+ * for the sync queue associated to this process; this |
12201 |
-+ * snapshot is taken to remember this value while the weight |
12202 |
-+ * raising is suspended because the queue is merged with a |
12203 |
-+ * shared queue, and is used to set @raising_cur_max_time |
12204 |
-+ * when the queue is split from the shared queue and its |
12205 |
-+ * weight is raised again |
12206 |
-+ * @saved_idle_window: same purpose as the previous field for the idle window |
12207 |
- */ |
12208 |
- struct bfq_io_cq { |
12209 |
- struct io_cq icq; /* must be the first member */ |
12210 |
- struct bfq_queue *bfqq[2]; |
12211 |
- struct bfq_ttime ttime; |
12212 |
- int ioprio; |
12213 |
-+ |
12214 |
-+ unsigned int raising_time_left; |
12215 |
-+ unsigned int saved_idle_window; |
12216 |
- }; |
12217 |
- |
12218 |
- /** |
12219 |
-@@ -418,8 +432,9 @@ enum bfqq_state_flags { |
12220 |
- BFQ_BFQQ_FLAG_sync, /* synchronous queue */ |
12221 |
- BFQ_BFQQ_FLAG_budget_new, /* no completion with this budget */ |
12222 |
- BFQ_BFQQ_FLAG_coop, /* bfqq is shared */ |
12223 |
-- BFQ_BFQQ_FLAG_split_coop, /* shared bfqq will be splitted */ |
12224 |
-- BFQ_BFQQ_FLAG_softrt_update, /* needs softrt-next-start update */ |
12225 |
-+ BFQ_BFQQ_FLAG_split_coop, /* shared bfqq will be split */ |
12226 |
-+ BFQ_BFQQ_FLAG_just_split, /* queue has just been split */ |
12227 |
-+ BFQ_BFQQ_FLAG_softrt_update, /* may need softrt-next-start update */ |
12228 |
- }; |
12229 |
- |
12230 |
- #define BFQ_BFQQ_FNS(name) \ |
12231 |
-@@ -446,6 +461,7 @@ BFQ_BFQQ_FNS(sync); |
12232 |
- BFQ_BFQQ_FNS(budget_new); |
12233 |
- BFQ_BFQQ_FNS(coop); |
12234 |
- BFQ_BFQQ_FNS(split_coop); |
12235 |
-+BFQ_BFQQ_FNS(just_split); |
12236 |
- BFQ_BFQQ_FNS(softrt_update); |
12237 |
- #undef BFQ_BFQQ_FNS |
12238 |
- |
12239 |
--- |
12240 |
-1.9.0 |
12241 |
- |