1 |
commit: 5d6d793f8ed82a72cedf644601ba72883f9ab36f |
2 |
Author: Mike Pagano <mpagano <AT> gentoo <DOT> org> |
3 |
AuthorDate: Sun Jul 11 14:45:27 2021 +0000 |
4 |
Commit: Mike Pagano <mpagano <AT> gentoo <DOT> org> |
5 |
CommitDate: Sun Jul 11 14:45:27 2021 +0000 |
6 |
URL: https://gitweb.gentoo.org/proj/linux-patches.git/commit/?id=5d6d793f |
7 |
|
8 |
Linux patch 4.19.197 |
9 |
|
10 |
Signed-off-by: Mike Pagano <mpagano <AT> gentoo.org> |
11 |
|
12 |
0000_README | 4 + |
13 |
1196_linux-4.19.197.patch | 1701 +++++++++++++++++++++++++++++++++++++++++++++ |
14 |
2 files changed, 1705 insertions(+) |
15 |
|
16 |
diff --git a/0000_README b/0000_README |
17 |
index 2009bef..dd8ede9 100644 |
18 |
--- a/0000_README |
19 |
+++ b/0000_README |
20 |
@@ -823,6 +823,10 @@ Patch: 1195_linux-4.19.196.patch |
21 |
From: https://www.kernel.org |
22 |
Desc: Linux 4.19.196 |
23 |
|
24 |
+Patch: 1196_linux-4.19.197.patch |
25 |
+From: https://www.kernel.org |
26 |
+Desc: Linux 4.19.197 |
27 |
+ |
28 |
Patch: 1500_XATTR_USER_PREFIX.patch |
29 |
From: https://bugs.gentoo.org/show_bug.cgi?id=470644 |
30 |
Desc: Support for namespace user.pax.* on tmpfs. |
31 |
|
32 |
diff --git a/1196_linux-4.19.197.patch b/1196_linux-4.19.197.patch |
33 |
new file mode 100644 |
34 |
index 0000000..e8a9416 |
35 |
--- /dev/null |
36 |
+++ b/1196_linux-4.19.197.patch |
37 |
@@ -0,0 +1,1701 @@ |
38 |
+diff --git a/Makefile b/Makefile |
39 |
+index 63b0bc92a0fa6..42073a4c6e2e3 100644 |
40 |
+--- a/Makefile |
41 |
++++ b/Makefile |
42 |
+@@ -1,7 +1,7 @@ |
43 |
+ # SPDX-License-Identifier: GPL-2.0 |
44 |
+ VERSION = 4 |
45 |
+ PATCHLEVEL = 19 |
46 |
+-SUBLEVEL = 196 |
47 |
++SUBLEVEL = 197 |
48 |
+ EXTRAVERSION = |
49 |
+ NAME = "People's Front" |
50 |
+ |
51 |
+diff --git a/arch/arm/boot/dts/dra7.dtsi b/arch/arm/boot/dts/dra7.dtsi |
52 |
+index 0c0781a37c5a7..7f1fe4a724472 100644 |
53 |
+--- a/arch/arm/boot/dts/dra7.dtsi |
54 |
++++ b/arch/arm/boot/dts/dra7.dtsi |
55 |
+@@ -48,6 +48,7 @@ |
56 |
+ |
57 |
+ timer { |
58 |
+ compatible = "arm,armv7-timer"; |
59 |
++ status = "disabled"; /* See ARM architected timer wrap erratum i940 */ |
60 |
+ interrupts = <GIC_PPI 13 (GIC_CPU_MASK_SIMPLE(2) | IRQ_TYPE_LEVEL_LOW)>, |
61 |
+ <GIC_PPI 14 (GIC_CPU_MASK_SIMPLE(2) | IRQ_TYPE_LEVEL_LOW)>, |
62 |
+ <GIC_PPI 11 (GIC_CPU_MASK_SIMPLE(2) | IRQ_TYPE_LEVEL_LOW)>, |
63 |
+@@ -910,6 +911,8 @@ |
64 |
+ reg = <0x48032000 0x80>; |
65 |
+ interrupts = <GIC_SPI 33 IRQ_TYPE_LEVEL_HIGH>; |
66 |
+ ti,hwmods = "timer2"; |
67 |
++ clock-names = "fck"; |
68 |
++ clocks = <&l4per_clkctrl DRA7_TIMER2_CLKCTRL 24>; |
69 |
+ }; |
70 |
+ |
71 |
+ timer3: timer@48034000 { |
72 |
+@@ -917,6 +920,10 @@ |
73 |
+ reg = <0x48034000 0x80>; |
74 |
+ interrupts = <GIC_SPI 34 IRQ_TYPE_LEVEL_HIGH>; |
75 |
+ ti,hwmods = "timer3"; |
76 |
++ clock-names = "fck"; |
77 |
++ clocks = <&l4per_clkctrl DRA7_TIMER3_CLKCTRL 24>; |
78 |
++ assigned-clocks = <&l4per_clkctrl DRA7_TIMER3_CLKCTRL 24>; |
79 |
++ assigned-clock-parents = <&timer_sys_clk_div>; |
80 |
+ }; |
81 |
+ |
82 |
+ timer4: timer@48036000 { |
83 |
+@@ -924,6 +931,10 @@ |
84 |
+ reg = <0x48036000 0x80>; |
85 |
+ interrupts = <GIC_SPI 35 IRQ_TYPE_LEVEL_HIGH>; |
86 |
+ ti,hwmods = "timer4"; |
87 |
++ clock-names = "fck"; |
88 |
++ clocks = <&l4per_clkctrl DRA7_TIMER4_CLKCTRL 24>; |
89 |
++ assigned-clocks = <&l4per_clkctrl DRA7_TIMER4_CLKCTRL 24>; |
90 |
++ assigned-clock-parents = <&timer_sys_clk_div>; |
91 |
+ }; |
92 |
+ |
93 |
+ timer5: timer@48820000 { |
94 |
+diff --git a/arch/arm/boot/dts/imx6qdl-sabresd.dtsi b/arch/arm/boot/dts/imx6qdl-sabresd.dtsi |
95 |
+index 41384bbd2f60c..03357d39870ee 100644 |
96 |
+--- a/arch/arm/boot/dts/imx6qdl-sabresd.dtsi |
97 |
++++ b/arch/arm/boot/dts/imx6qdl-sabresd.dtsi |
98 |
+@@ -675,10 +675,6 @@ |
99 |
+ vin-supply = <&vgen5_reg>; |
100 |
+ }; |
101 |
+ |
102 |
+-®_vdd3p0 { |
103 |
+- vin-supply = <&sw2_reg>; |
104 |
+-}; |
105 |
+- |
106 |
+ ®_vdd2p5 { |
107 |
+ vin-supply = <&vgen5_reg>; |
108 |
+ }; |
109 |
+diff --git a/arch/arm/mach-omap1/pm.c b/arch/arm/mach-omap1/pm.c |
110 |
+index 3e1de14805e48..65b91a8379c90 100644 |
111 |
+--- a/arch/arm/mach-omap1/pm.c |
112 |
++++ b/arch/arm/mach-omap1/pm.c |
113 |
+@@ -610,11 +610,6 @@ static irqreturn_t omap_wakeup_interrupt(int irq, void *dev) |
114 |
+ return IRQ_HANDLED; |
115 |
+ } |
116 |
+ |
117 |
+-static struct irqaction omap_wakeup_irq = { |
118 |
+- .name = "peripheral wakeup", |
119 |
+- .handler = omap_wakeup_interrupt |
120 |
+-}; |
121 |
+- |
122 |
+ |
123 |
+ |
124 |
+ static const struct platform_suspend_ops omap_pm_ops = { |
125 |
+@@ -627,6 +622,7 @@ static const struct platform_suspend_ops omap_pm_ops = { |
126 |
+ static int __init omap_pm_init(void) |
127 |
+ { |
128 |
+ int error = 0; |
129 |
++ int irq; |
130 |
+ |
131 |
+ if (!cpu_class_is_omap1()) |
132 |
+ return -ENODEV; |
133 |
+@@ -670,9 +666,12 @@ static int __init omap_pm_init(void) |
134 |
+ arm_pm_idle = omap1_pm_idle; |
135 |
+ |
136 |
+ if (cpu_is_omap7xx()) |
137 |
+- setup_irq(INT_7XX_WAKE_UP_REQ, &omap_wakeup_irq); |
138 |
++ irq = INT_7XX_WAKE_UP_REQ; |
139 |
+ else if (cpu_is_omap16xx()) |
140 |
+- setup_irq(INT_1610_WAKE_UP_REQ, &omap_wakeup_irq); |
141 |
++ irq = INT_1610_WAKE_UP_REQ; |
142 |
++ if (request_irq(irq, omap_wakeup_interrupt, 0, "peripheral wakeup", |
143 |
++ NULL)) |
144 |
++ pr_err("Failed to request irq %d (peripheral wakeup)\n", irq); |
145 |
+ |
146 |
+ /* Program new power ramp-up time |
147 |
+ * (0 for most boards since we don't lower voltage when in deep sleep) |
148 |
+diff --git a/arch/arm/mach-omap1/time.c b/arch/arm/mach-omap1/time.c |
149 |
+index 524977a31a49c..de590a85a42b3 100644 |
150 |
+--- a/arch/arm/mach-omap1/time.c |
151 |
++++ b/arch/arm/mach-omap1/time.c |
152 |
+@@ -155,15 +155,11 @@ static irqreturn_t omap_mpu_timer1_interrupt(int irq, void *dev_id) |
153 |
+ return IRQ_HANDLED; |
154 |
+ } |
155 |
+ |
156 |
+-static struct irqaction omap_mpu_timer1_irq = { |
157 |
+- .name = "mpu_timer1", |
158 |
+- .flags = IRQF_TIMER | IRQF_IRQPOLL, |
159 |
+- .handler = omap_mpu_timer1_interrupt, |
160 |
+-}; |
161 |
+- |
162 |
+ static __init void omap_init_mpu_timer(unsigned long rate) |
163 |
+ { |
164 |
+- setup_irq(INT_TIMER1, &omap_mpu_timer1_irq); |
165 |
++ if (request_irq(INT_TIMER1, omap_mpu_timer1_interrupt, |
166 |
++ IRQF_TIMER | IRQF_IRQPOLL, "mpu_timer1", NULL)) |
167 |
++ pr_err("Failed to request irq %d (mpu_timer1)\n", INT_TIMER1); |
168 |
+ omap_mpu_timer_start(0, (rate / HZ) - 1, 1); |
169 |
+ |
170 |
+ clockevent_mpu_timer1.cpumask = cpumask_of(0); |
171 |
+diff --git a/arch/arm/mach-omap1/timer32k.c b/arch/arm/mach-omap1/timer32k.c |
172 |
+index 0ae6c52a7d70b..780fdf03c3cee 100644 |
173 |
+--- a/arch/arm/mach-omap1/timer32k.c |
174 |
++++ b/arch/arm/mach-omap1/timer32k.c |
175 |
+@@ -148,15 +148,11 @@ static irqreturn_t omap_32k_timer_interrupt(int irq, void *dev_id) |
176 |
+ return IRQ_HANDLED; |
177 |
+ } |
178 |
+ |
179 |
+-static struct irqaction omap_32k_timer_irq = { |
180 |
+- .name = "32KHz timer", |
181 |
+- .flags = IRQF_TIMER | IRQF_IRQPOLL, |
182 |
+- .handler = omap_32k_timer_interrupt, |
183 |
+-}; |
184 |
+- |
185 |
+ static __init void omap_init_32k_timer(void) |
186 |
+ { |
187 |
+- setup_irq(INT_OS_TIMER, &omap_32k_timer_irq); |
188 |
++ if (request_irq(INT_OS_TIMER, omap_32k_timer_interrupt, |
189 |
++ IRQF_TIMER | IRQF_IRQPOLL, "32KHz timer", NULL)) |
190 |
++ pr_err("Failed to request irq %d(32KHz timer)\n", INT_OS_TIMER); |
191 |
+ |
192 |
+ clockevent_32k_timer.cpumask = cpumask_of(0); |
193 |
+ clockevents_config_and_register(&clockevent_32k_timer, |
194 |
+diff --git a/arch/arm/mach-omap2/board-generic.c b/arch/arm/mach-omap2/board-generic.c |
195 |
+index 6b4f4975cf7a6..6e59c11131c48 100644 |
196 |
+--- a/arch/arm/mach-omap2/board-generic.c |
197 |
++++ b/arch/arm/mach-omap2/board-generic.c |
198 |
+@@ -330,7 +330,7 @@ DT_MACHINE_START(DRA74X_DT, "Generic DRA74X (Flattened Device Tree)") |
199 |
+ .init_late = dra7xx_init_late, |
200 |
+ .init_irq = omap_gic_of_init, |
201 |
+ .init_machine = omap_generic_init, |
202 |
+- .init_time = omap5_realtime_timer_init, |
203 |
++ .init_time = omap3_gptimer_timer_init, |
204 |
+ .dt_compat = dra74x_boards_compat, |
205 |
+ .restart = omap44xx_restart, |
206 |
+ MACHINE_END |
207 |
+@@ -353,7 +353,7 @@ DT_MACHINE_START(DRA72X_DT, "Generic DRA72X (Flattened Device Tree)") |
208 |
+ .init_late = dra7xx_init_late, |
209 |
+ .init_irq = omap_gic_of_init, |
210 |
+ .init_machine = omap_generic_init, |
211 |
+- .init_time = omap5_realtime_timer_init, |
212 |
++ .init_time = omap3_gptimer_timer_init, |
213 |
+ .dt_compat = dra72x_boards_compat, |
214 |
+ .restart = omap44xx_restart, |
215 |
+ MACHINE_END |
216 |
+diff --git a/arch/arm/mach-omap2/timer.c b/arch/arm/mach-omap2/timer.c |
217 |
+index 98ed5ac073bc1..c4ba848e8af62 100644 |
218 |
+--- a/arch/arm/mach-omap2/timer.c |
219 |
++++ b/arch/arm/mach-omap2/timer.c |
220 |
+@@ -42,6 +42,7 @@ |
221 |
+ #include <linux/platform_device.h> |
222 |
+ #include <linux/platform_data/dmtimer-omap.h> |
223 |
+ #include <linux/sched_clock.h> |
224 |
++#include <linux/cpu.h> |
225 |
+ |
226 |
+ #include <asm/mach/time.h> |
227 |
+ #include <asm/smp_twd.h> |
228 |
+@@ -64,15 +65,28 @@ |
229 |
+ |
230 |
+ /* Clockevent code */ |
231 |
+ |
232 |
+-static struct omap_dm_timer clkev; |
233 |
+-static struct clock_event_device clockevent_gpt; |
234 |
+- |
235 |
+ /* Clockevent hwmod for am335x and am437x suspend */ |
236 |
+ static struct omap_hwmod *clockevent_gpt_hwmod; |
237 |
+ |
238 |
+ /* Clockesource hwmod for am437x suspend */ |
239 |
+ static struct omap_hwmod *clocksource_gpt_hwmod; |
240 |
+ |
241 |
++struct dmtimer_clockevent { |
242 |
++ struct clock_event_device dev; |
243 |
++ struct omap_dm_timer timer; |
244 |
++}; |
245 |
++ |
246 |
++static struct dmtimer_clockevent clockevent; |
247 |
++ |
248 |
++static struct omap_dm_timer *to_dmtimer(struct clock_event_device *clockevent) |
249 |
++{ |
250 |
++ struct dmtimer_clockevent *clkevt = |
251 |
++ container_of(clockevent, struct dmtimer_clockevent, dev); |
252 |
++ struct omap_dm_timer *timer = &clkevt->timer; |
253 |
++ |
254 |
++ return timer; |
255 |
++} |
256 |
++ |
257 |
+ #ifdef CONFIG_SOC_HAS_REALTIME_COUNTER |
258 |
+ static unsigned long arch_timer_freq; |
259 |
+ |
260 |
+@@ -84,24 +98,21 @@ void set_cntfreq(void) |
261 |
+ |
262 |
+ static irqreturn_t omap2_gp_timer_interrupt(int irq, void *dev_id) |
263 |
+ { |
264 |
+- struct clock_event_device *evt = &clockevent_gpt; |
265 |
+- |
266 |
+- __omap_dm_timer_write_status(&clkev, OMAP_TIMER_INT_OVERFLOW); |
267 |
++ struct dmtimer_clockevent *clkevt = dev_id; |
268 |
++ struct clock_event_device *evt = &clkevt->dev; |
269 |
++ struct omap_dm_timer *timer = &clkevt->timer; |
270 |
+ |
271 |
++ __omap_dm_timer_write_status(timer, OMAP_TIMER_INT_OVERFLOW); |
272 |
+ evt->event_handler(evt); |
273 |
+ return IRQ_HANDLED; |
274 |
+ } |
275 |
+ |
276 |
+-static struct irqaction omap2_gp_timer_irq = { |
277 |
+- .name = "gp_timer", |
278 |
+- .flags = IRQF_TIMER | IRQF_IRQPOLL, |
279 |
+- .handler = omap2_gp_timer_interrupt, |
280 |
+-}; |
281 |
+- |
282 |
+ static int omap2_gp_timer_set_next_event(unsigned long cycles, |
283 |
+ struct clock_event_device *evt) |
284 |
+ { |
285 |
+- __omap_dm_timer_load_start(&clkev, OMAP_TIMER_CTRL_ST, |
286 |
++ struct omap_dm_timer *timer = to_dmtimer(evt); |
287 |
++ |
288 |
++ __omap_dm_timer_load_start(timer, OMAP_TIMER_CTRL_ST, |
289 |
+ 0xffffffff - cycles, OMAP_TIMER_POSTED); |
290 |
+ |
291 |
+ return 0; |
292 |
+@@ -109,22 +120,26 @@ static int omap2_gp_timer_set_next_event(unsigned long cycles, |
293 |
+ |
294 |
+ static int omap2_gp_timer_shutdown(struct clock_event_device *evt) |
295 |
+ { |
296 |
+- __omap_dm_timer_stop(&clkev, OMAP_TIMER_POSTED, clkev.rate); |
297 |
++ struct omap_dm_timer *timer = to_dmtimer(evt); |
298 |
++ |
299 |
++ __omap_dm_timer_stop(timer, OMAP_TIMER_POSTED, timer->rate); |
300 |
++ |
301 |
+ return 0; |
302 |
+ } |
303 |
+ |
304 |
+ static int omap2_gp_timer_set_periodic(struct clock_event_device *evt) |
305 |
+ { |
306 |
++ struct omap_dm_timer *timer = to_dmtimer(evt); |
307 |
+ u32 period; |
308 |
+ |
309 |
+- __omap_dm_timer_stop(&clkev, OMAP_TIMER_POSTED, clkev.rate); |
310 |
++ __omap_dm_timer_stop(timer, OMAP_TIMER_POSTED, timer->rate); |
311 |
+ |
312 |
+- period = clkev.rate / HZ; |
313 |
++ period = timer->rate / HZ; |
314 |
+ period -= 1; |
315 |
+ /* Looks like we need to first set the load value separately */ |
316 |
+- __omap_dm_timer_write(&clkev, OMAP_TIMER_LOAD_REG, 0xffffffff - period, |
317 |
++ __omap_dm_timer_write(timer, OMAP_TIMER_LOAD_REG, 0xffffffff - period, |
318 |
+ OMAP_TIMER_POSTED); |
319 |
+- __omap_dm_timer_load_start(&clkev, |
320 |
++ __omap_dm_timer_load_start(timer, |
321 |
+ OMAP_TIMER_CTRL_AR | OMAP_TIMER_CTRL_ST, |
322 |
+ 0xffffffff - period, OMAP_TIMER_POSTED); |
323 |
+ return 0; |
324 |
+@@ -138,25 +153,16 @@ static void omap_clkevt_idle(struct clock_event_device *unused) |
325 |
+ omap_hwmod_idle(clockevent_gpt_hwmod); |
326 |
+ } |
327 |
+ |
328 |
+-static void omap_clkevt_unidle(struct clock_event_device *unused) |
329 |
++static void omap_clkevt_unidle(struct clock_event_device *evt) |
330 |
+ { |
331 |
++ struct omap_dm_timer *timer = to_dmtimer(evt); |
332 |
++ |
333 |
+ if (!clockevent_gpt_hwmod) |
334 |
+ return; |
335 |
+ |
336 |
+ omap_hwmod_enable(clockevent_gpt_hwmod); |
337 |
+- __omap_dm_timer_int_enable(&clkev, OMAP_TIMER_INT_OVERFLOW); |
338 |
+-} |
339 |
+- |
340 |
+-static struct clock_event_device clockevent_gpt = { |
341 |
+- .features = CLOCK_EVT_FEAT_PERIODIC | |
342 |
+- CLOCK_EVT_FEAT_ONESHOT, |
343 |
+- .rating = 300, |
344 |
+- .set_next_event = omap2_gp_timer_set_next_event, |
345 |
+- .set_state_shutdown = omap2_gp_timer_shutdown, |
346 |
+- .set_state_periodic = omap2_gp_timer_set_periodic, |
347 |
+- .set_state_oneshot = omap2_gp_timer_shutdown, |
348 |
+- .tick_resume = omap2_gp_timer_shutdown, |
349 |
+-}; |
350 |
++ __omap_dm_timer_int_enable(timer, OMAP_TIMER_INT_OVERFLOW); |
351 |
++} |
352 |
+ |
353 |
+ static const struct of_device_id omap_timer_match[] __initconst = { |
354 |
+ { .compatible = "ti,omap2420-timer", }, |
355 |
+@@ -363,47 +369,104 @@ void tick_broadcast(const struct cpumask *mask) |
356 |
+ } |
357 |
+ #endif |
358 |
+ |
359 |
+-static void __init omap2_gp_clockevent_init(int gptimer_id, |
360 |
+- const char *fck_source, |
361 |
+- const char *property) |
362 |
++static void __init dmtimer_clkevt_init_common(struct dmtimer_clockevent *clkevt, |
363 |
++ int gptimer_id, |
364 |
++ const char *fck_source, |
365 |
++ unsigned int features, |
366 |
++ const struct cpumask *cpumask, |
367 |
++ const char *property, |
368 |
++ int rating, const char *name) |
369 |
+ { |
370 |
++ struct omap_dm_timer *timer = &clkevt->timer; |
371 |
+ int res; |
372 |
+ |
373 |
+- clkev.id = gptimer_id; |
374 |
+- clkev.errata = omap_dm_timer_get_errata(); |
375 |
++ timer->id = gptimer_id; |
376 |
++ timer->errata = omap_dm_timer_get_errata(); |
377 |
++ clkevt->dev.features = features; |
378 |
++ clkevt->dev.rating = rating; |
379 |
++ clkevt->dev.set_next_event = omap2_gp_timer_set_next_event; |
380 |
++ clkevt->dev.set_state_shutdown = omap2_gp_timer_shutdown; |
381 |
++ clkevt->dev.set_state_periodic = omap2_gp_timer_set_periodic; |
382 |
++ clkevt->dev.set_state_oneshot = omap2_gp_timer_shutdown; |
383 |
++ clkevt->dev.tick_resume = omap2_gp_timer_shutdown; |
384 |
+ |
385 |
+ /* |
386 |
+ * For clock-event timers we never read the timer counter and |
387 |
+ * so we are not impacted by errata i103 and i767. Therefore, |
388 |
+ * we can safely ignore this errata for clock-event timers. |
389 |
+ */ |
390 |
+- __omap_dm_timer_override_errata(&clkev, OMAP_TIMER_ERRATA_I103_I767); |
391 |
++ __omap_dm_timer_override_errata(timer, OMAP_TIMER_ERRATA_I103_I767); |
392 |
+ |
393 |
+- res = omap_dm_timer_init_one(&clkev, fck_source, property, |
394 |
+- &clockevent_gpt.name, OMAP_TIMER_POSTED); |
395 |
++ res = omap_dm_timer_init_one(timer, fck_source, property, |
396 |
++ &clkevt->dev.name, OMAP_TIMER_POSTED); |
397 |
+ BUG_ON(res); |
398 |
+ |
399 |
+- omap2_gp_timer_irq.dev_id = &clkev; |
400 |
+- setup_irq(clkev.irq, &omap2_gp_timer_irq); |
401 |
++ clkevt->dev.cpumask = cpumask; |
402 |
++ clkevt->dev.irq = omap_dm_timer_get_irq(timer); |
403 |
+ |
404 |
+- __omap_dm_timer_int_enable(&clkev, OMAP_TIMER_INT_OVERFLOW); |
405 |
++ if (request_irq(clkevt->dev.irq, omap2_gp_timer_interrupt, |
406 |
++ IRQF_TIMER | IRQF_IRQPOLL, name, clkevt)) |
407 |
++ pr_err("Failed to request irq %d (gp_timer)\n", clkevt->dev.irq); |
408 |
+ |
409 |
+- clockevent_gpt.cpumask = cpu_possible_mask; |
410 |
+- clockevent_gpt.irq = omap_dm_timer_get_irq(&clkev); |
411 |
+- clockevents_config_and_register(&clockevent_gpt, clkev.rate, |
412 |
+- 3, /* Timer internal resynch latency */ |
413 |
+- 0xffffffff); |
414 |
++ __omap_dm_timer_int_enable(timer, OMAP_TIMER_INT_OVERFLOW); |
415 |
+ |
416 |
+ if (soc_is_am33xx() || soc_is_am43xx()) { |
417 |
+- clockevent_gpt.suspend = omap_clkevt_idle; |
418 |
+- clockevent_gpt.resume = omap_clkevt_unidle; |
419 |
++ clkevt->dev.suspend = omap_clkevt_idle; |
420 |
++ clkevt->dev.resume = omap_clkevt_unidle; |
421 |
+ |
422 |
+ clockevent_gpt_hwmod = |
423 |
+- omap_hwmod_lookup(clockevent_gpt.name); |
424 |
++ omap_hwmod_lookup(clkevt->dev.name); |
425 |
++ } |
426 |
++ |
427 |
++ pr_info("OMAP clockevent source: %s at %lu Hz\n", clkevt->dev.name, |
428 |
++ timer->rate); |
429 |
++} |
430 |
++ |
431 |
++static DEFINE_PER_CPU(struct dmtimer_clockevent, dmtimer_percpu_timer); |
432 |
++ |
433 |
++static int omap_gptimer_starting_cpu(unsigned int cpu) |
434 |
++{ |
435 |
++ struct dmtimer_clockevent *clkevt = per_cpu_ptr(&dmtimer_percpu_timer, cpu); |
436 |
++ struct clock_event_device *dev = &clkevt->dev; |
437 |
++ struct omap_dm_timer *timer = &clkevt->timer; |
438 |
++ |
439 |
++ clockevents_config_and_register(dev, timer->rate, 3, ULONG_MAX); |
440 |
++ irq_force_affinity(dev->irq, cpumask_of(cpu)); |
441 |
++ |
442 |
++ return 0; |
443 |
++} |
444 |
++ |
445 |
++static int __init dmtimer_percpu_quirk_init(void) |
446 |
++{ |
447 |
++ struct dmtimer_clockevent *clkevt; |
448 |
++ struct clock_event_device *dev; |
449 |
++ struct device_node *arm_timer; |
450 |
++ struct omap_dm_timer *timer; |
451 |
++ int cpu = 0; |
452 |
++ |
453 |
++ arm_timer = of_find_compatible_node(NULL, NULL, "arm,armv7-timer"); |
454 |
++ if (of_device_is_available(arm_timer)) { |
455 |
++ pr_warn_once("ARM architected timer wrap issue i940 detected\n"); |
456 |
++ return 0; |
457 |
++ } |
458 |
++ |
459 |
++ for_each_possible_cpu(cpu) { |
460 |
++ clkevt = per_cpu_ptr(&dmtimer_percpu_timer, cpu); |
461 |
++ dev = &clkevt->dev; |
462 |
++ timer = &clkevt->timer; |
463 |
++ |
464 |
++ dmtimer_clkevt_init_common(clkevt, 0, "timer_sys_ck", |
465 |
++ CLOCK_EVT_FEAT_ONESHOT, |
466 |
++ cpumask_of(cpu), |
467 |
++ "assigned-clock-parents", |
468 |
++ 500, "percpu timer"); |
469 |
+ } |
470 |
+ |
471 |
+- pr_info("OMAP clockevent source: %s at %lu Hz\n", clockevent_gpt.name, |
472 |
+- clkev.rate); |
473 |
++ cpuhp_setup_state(CPUHP_AP_OMAP_DM_TIMER_STARTING, |
474 |
++ "clockevents/omap/gptimer:starting", |
475 |
++ omap_gptimer_starting_cpu, NULL); |
476 |
++ |
477 |
++ return 0; |
478 |
+ } |
479 |
+ |
480 |
+ /* Clocksource code */ |
481 |
+@@ -543,7 +606,15 @@ static void __init __omap_sync32k_timer_init(int clkev_nr, const char *clkev_src |
482 |
+ { |
483 |
+ omap_clk_init(); |
484 |
+ omap_dmtimer_init(); |
485 |
+- omap2_gp_clockevent_init(clkev_nr, clkev_src, clkev_prop); |
486 |
++ dmtimer_clkevt_init_common(&clockevent, clkev_nr, clkev_src, |
487 |
++ CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT, |
488 |
++ cpu_possible_mask, clkev_prop, 300, "clockevent"); |
489 |
++ clockevents_config_and_register(&clockevent.dev, clockevent.timer.rate, |
490 |
++ 3, /* Timer internal resynch latency */ |
491 |
++ 0xffffffff); |
492 |
++ |
493 |
++ if (soc_is_dra7xx()) |
494 |
++ dmtimer_percpu_quirk_init(); |
495 |
+ |
496 |
+ /* Enable the use of clocksource="gp_timer" kernel parameter */ |
497 |
+ if (use_gptimer_clksrc || gptimer) |
498 |
+@@ -572,7 +643,7 @@ void __init omap3_secure_sync32k_timer_init(void) |
499 |
+ #endif /* CONFIG_ARCH_OMAP3 */ |
500 |
+ |
501 |
+ #if defined(CONFIG_ARCH_OMAP3) || defined(CONFIG_SOC_AM33XX) || \ |
502 |
+- defined(CONFIG_SOC_AM43XX) |
503 |
++ defined(CONFIG_SOC_AM43XX) || defined(CONFIG_SOC_DRA7XX) |
504 |
+ void __init omap3_gptimer_timer_init(void) |
505 |
+ { |
506 |
+ __omap_sync32k_timer_init(2, "timer_sys_ck", NULL, |
507 |
+diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c |
508 |
+index ad24e67772778..bd463d6842370 100644 |
509 |
+--- a/arch/x86/kvm/svm.c |
510 |
++++ b/arch/x86/kvm/svm.c |
511 |
+@@ -1791,9 +1791,25 @@ static void sev_asid_free(struct kvm *kvm) |
512 |
+ __sev_asid_free(sev->asid); |
513 |
+ } |
514 |
+ |
515 |
+-static void sev_unbind_asid(struct kvm *kvm, unsigned int handle) |
516 |
++static void sev_decommission(unsigned int handle) |
517 |
+ { |
518 |
+ struct sev_data_decommission *decommission; |
519 |
++ |
520 |
++ if (!handle) |
521 |
++ return; |
522 |
++ |
523 |
++ decommission = kzalloc(sizeof(*decommission), GFP_KERNEL); |
524 |
++ if (!decommission) |
525 |
++ return; |
526 |
++ |
527 |
++ decommission->handle = handle; |
528 |
++ sev_guest_decommission(decommission, NULL); |
529 |
++ |
530 |
++ kfree(decommission); |
531 |
++} |
532 |
++ |
533 |
++static void sev_unbind_asid(struct kvm *kvm, unsigned int handle) |
534 |
++{ |
535 |
+ struct sev_data_deactivate *data; |
536 |
+ |
537 |
+ if (!handle) |
538 |
+@@ -1811,15 +1827,7 @@ static void sev_unbind_asid(struct kvm *kvm, unsigned int handle) |
539 |
+ sev_guest_df_flush(NULL); |
540 |
+ kfree(data); |
541 |
+ |
542 |
+- decommission = kzalloc(sizeof(*decommission), GFP_KERNEL); |
543 |
+- if (!decommission) |
544 |
+- return; |
545 |
+- |
546 |
+- /* decommission handle */ |
547 |
+- decommission->handle = handle; |
548 |
+- sev_guest_decommission(decommission, NULL); |
549 |
+- |
550 |
+- kfree(decommission); |
551 |
++ sev_decommission(handle); |
552 |
+ } |
553 |
+ |
554 |
+ static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr, |
555 |
+@@ -1954,6 +1962,7 @@ static void sev_vm_destroy(struct kvm *kvm) |
556 |
+ list_for_each_safe(pos, q, head) { |
557 |
+ __unregister_enc_region_locked(kvm, |
558 |
+ list_entry(pos, struct enc_region, list)); |
559 |
++ cond_resched(); |
560 |
+ } |
561 |
+ } |
562 |
+ |
563 |
+@@ -6468,8 +6477,10 @@ static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp) |
564 |
+ |
565 |
+ /* Bind ASID to this guest */ |
566 |
+ ret = sev_bind_asid(kvm, start->handle, error); |
567 |
+- if (ret) |
568 |
++ if (ret) { |
569 |
++ sev_decommission(start->handle); |
570 |
+ goto e_free_session; |
571 |
++ } |
572 |
+ |
573 |
+ /* return handle to userspace */ |
574 |
+ params.handle = start->handle; |
575 |
+diff --git a/drivers/clk/ti/clk-7xx.c b/drivers/clk/ti/clk-7xx.c |
576 |
+index 71a122b2dc67e..b6d1ec49fa01a 100644 |
577 |
+--- a/drivers/clk/ti/clk-7xx.c |
578 |
++++ b/drivers/clk/ti/clk-7xx.c |
579 |
+@@ -733,6 +733,7 @@ const struct omap_clkctrl_data dra7_clkctrl_data[] __initconst = { |
580 |
+ static struct ti_dt_clk dra7xx_clks[] = { |
581 |
+ DT_CLK(NULL, "timer_32k_ck", "sys_32k_ck"), |
582 |
+ DT_CLK(NULL, "sys_clkin_ck", "timer_sys_clk_div"), |
583 |
++ DT_CLK(NULL, "timer_sys_ck", "timer_sys_clk_div"), |
584 |
+ DT_CLK(NULL, "sys_clkin", "sys_clkin1"), |
585 |
+ DT_CLK(NULL, "atl_dpll_clk_mux", "atl_cm:0000:24"), |
586 |
+ DT_CLK(NULL, "atl_gfclk_mux", "atl_cm:0000:26"), |
587 |
+diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c |
588 |
+index 7214022dfb911..d230536e7086d 100644 |
589 |
+--- a/drivers/gpu/drm/nouveau/nouveau_bo.c |
590 |
++++ b/drivers/gpu/drm/nouveau/nouveau_bo.c |
591 |
+@@ -512,7 +512,7 @@ nouveau_bo_sync_for_device(struct nouveau_bo *nvbo) |
592 |
+ struct ttm_dma_tt *ttm_dma = (struct ttm_dma_tt *)nvbo->bo.ttm; |
593 |
+ int i; |
594 |
+ |
595 |
+- if (!ttm_dma) |
596 |
++ if (!ttm_dma || !ttm_dma->dma_address) |
597 |
+ return; |
598 |
+ |
599 |
+ /* Don't waste time looping if the object is coherent */ |
600 |
+@@ -532,7 +532,7 @@ nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo) |
601 |
+ struct ttm_dma_tt *ttm_dma = (struct ttm_dma_tt *)nvbo->bo.ttm; |
602 |
+ int i; |
603 |
+ |
604 |
+- if (!ttm_dma) |
605 |
++ if (!ttm_dma || !ttm_dma->dma_address) |
606 |
+ return; |
607 |
+ |
608 |
+ /* Don't waste time looping if the object is coherent */ |
609 |
+diff --git a/drivers/scsi/sr.c b/drivers/scsi/sr.c |
610 |
+index 45c8bf39ad238..acf0c244141f4 100644 |
611 |
+--- a/drivers/scsi/sr.c |
612 |
++++ b/drivers/scsi/sr.c |
613 |
+@@ -216,6 +216,8 @@ static unsigned int sr_get_events(struct scsi_device *sdev) |
614 |
+ return DISK_EVENT_EJECT_REQUEST; |
615 |
+ else if (med->media_event_code == 2) |
616 |
+ return DISK_EVENT_MEDIA_CHANGE; |
617 |
++ else if (med->media_event_code == 3) |
618 |
++ return DISK_EVENT_EJECT_REQUEST; |
619 |
+ return 0; |
620 |
+ } |
621 |
+ |
622 |
+diff --git a/drivers/xen/events/events_base.c b/drivers/xen/events/events_base.c |
623 |
+index b370144682ed5..a2f8130e18fec 100644 |
624 |
+--- a/drivers/xen/events/events_base.c |
625 |
++++ b/drivers/xen/events/events_base.c |
626 |
+@@ -524,6 +524,9 @@ static void xen_irq_lateeoi_locked(struct irq_info *info, bool spurious) |
627 |
+ } |
628 |
+ |
629 |
+ info->eoi_time = 0; |
630 |
++ |
631 |
++ /* is_active hasn't been reset yet, do it now. */ |
632 |
++ smp_store_release(&info->is_active, 0); |
633 |
+ do_unmask(info, EVT_MASK_REASON_EOI_PENDING); |
634 |
+ } |
635 |
+ |
636 |
+@@ -1780,10 +1783,22 @@ static void lateeoi_ack_dynirq(struct irq_data *data) |
637 |
+ struct irq_info *info = info_for_irq(data->irq); |
638 |
+ evtchn_port_t evtchn = info ? info->evtchn : 0; |
639 |
+ |
640 |
+- if (VALID_EVTCHN(evtchn)) { |
641 |
+- do_mask(info, EVT_MASK_REASON_EOI_PENDING); |
642 |
+- ack_dynirq(data); |
643 |
+- } |
644 |
++ if (!VALID_EVTCHN(evtchn)) |
645 |
++ return; |
646 |
++ |
647 |
++ do_mask(info, EVT_MASK_REASON_EOI_PENDING); |
648 |
++ |
649 |
++ if (unlikely(irqd_is_setaffinity_pending(data)) && |
650 |
++ likely(!irqd_irq_disabled(data))) { |
651 |
++ do_mask(info, EVT_MASK_REASON_TEMPORARY); |
652 |
++ |
653 |
++ clear_evtchn(evtchn); |
654 |
++ |
655 |
++ irq_move_masked_irq(data); |
656 |
++ |
657 |
++ do_unmask(info, EVT_MASK_REASON_TEMPORARY); |
658 |
++ } else |
659 |
++ clear_evtchn(evtchn); |
660 |
+ } |
661 |
+ |
662 |
+ static void lateeoi_mask_ack_dynirq(struct irq_data *data) |
663 |
+diff --git a/fs/ext4/block_validity.c b/fs/ext4/block_validity.c |
664 |
+index 1ea8fc9ff048f..1bc65ecd4bd6d 100644 |
665 |
+--- a/fs/ext4/block_validity.c |
666 |
++++ b/fs/ext4/block_validity.c |
667 |
+@@ -171,8 +171,10 @@ static int ext4_data_block_valid_rcu(struct ext4_sb_info *sbi, |
668 |
+ else if (start_blk >= (entry->start_blk + entry->count)) |
669 |
+ n = n->rb_right; |
670 |
+ else { |
671 |
++ if (entry->ino == ino) |
672 |
++ return 1; |
673 |
+ sbi->s_es->s_last_error_block = cpu_to_le64(start_blk); |
674 |
+- return entry->ino == ino; |
675 |
++ return 0; |
676 |
+ } |
677 |
+ } |
678 |
+ return 1; |
679 |
+diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h |
680 |
+index d67c0035165c2..3d323c6c85260 100644 |
681 |
+--- a/include/linux/cpuhotplug.h |
682 |
++++ b/include/linux/cpuhotplug.h |
683 |
+@@ -118,6 +118,7 @@ enum cpuhp_state { |
684 |
+ CPUHP_AP_ARM_L2X0_STARTING, |
685 |
+ CPUHP_AP_EXYNOS4_MCT_TIMER_STARTING, |
686 |
+ CPUHP_AP_ARM_ARCH_TIMER_STARTING, |
687 |
++ CPUHP_AP_OMAP_DM_TIMER_STARTING, |
688 |
+ CPUHP_AP_ARM_GLOBAL_TIMER_STARTING, |
689 |
+ CPUHP_AP_JCORE_TIMER_STARTING, |
690 |
+ CPUHP_AP_ARM_TWD_STARTING, |
691 |
+diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h |
692 |
+index e375f2249f520..becf9b1eae5a1 100644 |
693 |
+--- a/include/linux/huge_mm.h |
694 |
++++ b/include/linux/huge_mm.h |
695 |
+@@ -224,6 +224,7 @@ struct page *follow_devmap_pud(struct vm_area_struct *vma, unsigned long addr, |
696 |
+ extern vm_fault_t do_huge_pmd_numa_page(struct vm_fault *vmf, pmd_t orig_pmd); |
697 |
+ |
698 |
+ extern struct page *huge_zero_page; |
699 |
++extern unsigned long huge_zero_pfn; |
700 |
+ |
701 |
+ static inline bool is_huge_zero_page(struct page *page) |
702 |
+ { |
703 |
+@@ -232,7 +233,7 @@ static inline bool is_huge_zero_page(struct page *page) |
704 |
+ |
705 |
+ static inline bool is_huge_zero_pmd(pmd_t pmd) |
706 |
+ { |
707 |
+- return is_huge_zero_page(pmd_page(pmd)); |
708 |
++ return READ_ONCE(huge_zero_pfn) == pmd_pfn(pmd) && pmd_present(pmd); |
709 |
+ } |
710 |
+ |
711 |
+ static inline bool is_huge_zero_pud(pud_t pud) |
712 |
+@@ -342,6 +343,11 @@ static inline bool is_huge_zero_page(struct page *page) |
713 |
+ return false; |
714 |
+ } |
715 |
+ |
716 |
++static inline bool is_huge_zero_pmd(pmd_t pmd) |
717 |
++{ |
718 |
++ return false; |
719 |
++} |
720 |
++ |
721 |
+ static inline bool is_huge_zero_pud(pud_t pud) |
722 |
+ { |
723 |
+ return false; |
724 |
+diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h |
725 |
+index c129c1c14c5f2..2df83a6598182 100644 |
726 |
+--- a/include/linux/hugetlb.h |
727 |
++++ b/include/linux/hugetlb.h |
728 |
+@@ -477,17 +477,6 @@ static inline int hstate_index(struct hstate *h) |
729 |
+ return h - hstates; |
730 |
+ } |
731 |
+ |
732 |
+-pgoff_t __basepage_index(struct page *page); |
733 |
+- |
734 |
+-/* Return page->index in PAGE_SIZE units */ |
735 |
+-static inline pgoff_t basepage_index(struct page *page) |
736 |
+-{ |
737 |
+- if (!PageCompound(page)) |
738 |
+- return page->index; |
739 |
+- |
740 |
+- return __basepage_index(page); |
741 |
+-} |
742 |
+- |
743 |
+ extern int dissolve_free_huge_page(struct page *page); |
744 |
+ extern int dissolve_free_huge_pages(unsigned long start_pfn, |
745 |
+ unsigned long end_pfn); |
746 |
+@@ -582,11 +571,6 @@ static inline int hstate_index(struct hstate *h) |
747 |
+ return 0; |
748 |
+ } |
749 |
+ |
750 |
+-static inline pgoff_t basepage_index(struct page *page) |
751 |
+-{ |
752 |
+- return page->index; |
753 |
+-} |
754 |
+- |
755 |
+ static inline int dissolve_free_huge_page(struct page *page) |
756 |
+ { |
757 |
+ return 0; |
758 |
+diff --git a/include/linux/mm.h b/include/linux/mm.h |
759 |
+index f6ecf41aea83d..c736c677b876d 100644 |
760 |
+--- a/include/linux/mm.h |
761 |
++++ b/include/linux/mm.h |
762 |
+@@ -1338,6 +1338,7 @@ struct zap_details { |
763 |
+ struct address_space *check_mapping; /* Check page->mapping if set */ |
764 |
+ pgoff_t first_index; /* Lowest page->index to unmap */ |
765 |
+ pgoff_t last_index; /* Highest page->index to unmap */ |
766 |
++ struct page *single_page; /* Locked page to be unmapped */ |
767 |
+ }; |
768 |
+ |
769 |
+ struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
770 |
+@@ -1428,6 +1429,7 @@ extern vm_fault_t handle_mm_fault(struct vm_area_struct *vma, |
771 |
+ extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm, |
772 |
+ unsigned long address, unsigned int fault_flags, |
773 |
+ bool *unlocked); |
774 |
++void unmap_mapping_page(struct page *page); |
775 |
+ void unmap_mapping_pages(struct address_space *mapping, |
776 |
+ pgoff_t start, pgoff_t nr, bool even_cows); |
777 |
+ void unmap_mapping_range(struct address_space *mapping, |
778 |
+@@ -1448,6 +1450,7 @@ static inline int fixup_user_fault(struct task_struct *tsk, |
779 |
+ BUG(); |
780 |
+ return -EFAULT; |
781 |
+ } |
782 |
++static inline void unmap_mapping_page(struct page *page) { } |
783 |
+ static inline void unmap_mapping_pages(struct address_space *mapping, |
784 |
+ pgoff_t start, pgoff_t nr, bool even_cows) { } |
785 |
+ static inline void unmap_mapping_range(struct address_space *mapping, |
786 |
+diff --git a/include/linux/mmdebug.h b/include/linux/mmdebug.h |
787 |
+index 2ad72d2c8cc52..5d0767cb424aa 100644 |
788 |
+--- a/include/linux/mmdebug.h |
789 |
++++ b/include/linux/mmdebug.h |
790 |
+@@ -37,6 +37,18 @@ void dump_mm(const struct mm_struct *mm); |
791 |
+ BUG(); \ |
792 |
+ } \ |
793 |
+ } while (0) |
794 |
++#define VM_WARN_ON_ONCE_PAGE(cond, page) ({ \ |
795 |
++ static bool __section(".data.once") __warned; \ |
796 |
++ int __ret_warn_once = !!(cond); \ |
797 |
++ \ |
798 |
++ if (unlikely(__ret_warn_once && !__warned)) { \ |
799 |
++ dump_page(page, "VM_WARN_ON_ONCE_PAGE(" __stringify(cond)")");\ |
800 |
++ __warned = true; \ |
801 |
++ WARN_ON(1); \ |
802 |
++ } \ |
803 |
++ unlikely(__ret_warn_once); \ |
804 |
++}) |
805 |
++ |
806 |
+ #define VM_WARN_ON(cond) (void)WARN_ON(cond) |
807 |
+ #define VM_WARN_ON_ONCE(cond) (void)WARN_ON_ONCE(cond) |
808 |
+ #define VM_WARN_ONCE(cond, format...) (void)WARN_ONCE(cond, format) |
809 |
+@@ -48,6 +60,7 @@ void dump_mm(const struct mm_struct *mm); |
810 |
+ #define VM_BUG_ON_MM(cond, mm) VM_BUG_ON(cond) |
811 |
+ #define VM_WARN_ON(cond) BUILD_BUG_ON_INVALID(cond) |
812 |
+ #define VM_WARN_ON_ONCE(cond) BUILD_BUG_ON_INVALID(cond) |
813 |
++#define VM_WARN_ON_ONCE_PAGE(cond, page) BUILD_BUG_ON_INVALID(cond) |
814 |
+ #define VM_WARN_ONCE(cond, format...) BUILD_BUG_ON_INVALID(cond) |
815 |
+ #define VM_WARN(cond, format...) BUILD_BUG_ON_INVALID(cond) |
816 |
+ #endif |
817 |
+diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h |
818 |
+index b1bd2186e6d2b..33b63b2a163f9 100644 |
819 |
+--- a/include/linux/pagemap.h |
820 |
++++ b/include/linux/pagemap.h |
821 |
+@@ -403,7 +403,7 @@ static inline struct page *read_mapping_page(struct address_space *mapping, |
822 |
+ } |
823 |
+ |
824 |
+ /* |
825 |
+- * Get index of the page with in radix-tree |
826 |
++ * Get index of the page within radix-tree (but not for hugetlb pages). |
827 |
+ * (TODO: remove once hugetlb pages will have ->index in PAGE_SIZE) |
828 |
+ */ |
829 |
+ static inline pgoff_t page_to_index(struct page *page) |
830 |
+@@ -422,15 +422,16 @@ static inline pgoff_t page_to_index(struct page *page) |
831 |
+ return pgoff; |
832 |
+ } |
833 |
+ |
834 |
++extern pgoff_t hugetlb_basepage_index(struct page *page); |
835 |
++ |
836 |
+ /* |
837 |
+- * Get the offset in PAGE_SIZE. |
838 |
+- * (TODO: hugepage should have ->index in PAGE_SIZE) |
839 |
++ * Get the offset in PAGE_SIZE (even for hugetlb pages). |
840 |
++ * (TODO: hugetlb pages should have ->index in PAGE_SIZE) |
841 |
+ */ |
842 |
+ static inline pgoff_t page_to_pgoff(struct page *page) |
843 |
+ { |
844 |
+- if (unlikely(PageHeadHuge(page))) |
845 |
+- return page->index << compound_order(page); |
846 |
+- |
847 |
++ if (unlikely(PageHuge(page))) |
848 |
++ return hugetlb_basepage_index(page); |
849 |
+ return page_to_index(page); |
850 |
+ } |
851 |
+ |
852 |
+diff --git a/include/linux/rmap.h b/include/linux/rmap.h |
853 |
+index d7d6d4eb17949..91ccae9467164 100644 |
854 |
+--- a/include/linux/rmap.h |
855 |
++++ b/include/linux/rmap.h |
856 |
+@@ -98,7 +98,8 @@ enum ttu_flags { |
857 |
+ * do a final flush if necessary */ |
858 |
+ TTU_RMAP_LOCKED = 0x80, /* do not grab rmap lock: |
859 |
+ * caller holds it */ |
860 |
+- TTU_SPLIT_FREEZE = 0x100, /* freeze pte under splitting thp */ |
861 |
++ TTU_SPLIT_FREEZE = 0x100, /* freeze pte under splitting thp */ |
862 |
++ TTU_SYNC = 0x200, /* avoid racy checks with PVMW_SYNC */ |
863 |
+ }; |
864 |
+ |
865 |
+ #ifdef CONFIG_MMU |
866 |
+diff --git a/kernel/futex.c b/kernel/futex.c |
867 |
+index 526ebcff5a0a9..3c67da9b84086 100644 |
868 |
+--- a/kernel/futex.c |
869 |
++++ b/kernel/futex.c |
870 |
+@@ -719,7 +719,7 @@ again: |
871 |
+ |
872 |
+ key->both.offset |= FUT_OFF_INODE; /* inode-based key */ |
873 |
+ key->shared.i_seq = get_inode_sequence_number(inode); |
874 |
+- key->shared.pgoff = basepage_index(tail); |
875 |
++ key->shared.pgoff = page_to_pgoff(tail); |
876 |
+ rcu_read_unlock(); |
877 |
+ } |
878 |
+ |
879 |
+diff --git a/kernel/kthread.c b/kernel/kthread.c |
880 |
+index 81abfac351272..9750f4f7f9010 100644 |
881 |
+--- a/kernel/kthread.c |
882 |
++++ b/kernel/kthread.c |
883 |
+@@ -1010,8 +1010,38 @@ void kthread_flush_work(struct kthread_work *work) |
884 |
+ EXPORT_SYMBOL_GPL(kthread_flush_work); |
885 |
+ |
886 |
+ /* |
887 |
+- * This function removes the work from the worker queue. Also it makes sure |
888 |
+- * that it won't get queued later via the delayed work's timer. |
889 |
++ * Make sure that the timer is neither set nor running and could |
890 |
++ * not manipulate the work list_head any longer. |
891 |
++ * |
892 |
++ * The function is called under worker->lock. The lock is temporary |
893 |
++ * released but the timer can't be set again in the meantime. |
894 |
++ */ |
895 |
++static void kthread_cancel_delayed_work_timer(struct kthread_work *work, |
896 |
++ unsigned long *flags) |
897 |
++{ |
898 |
++ struct kthread_delayed_work *dwork = |
899 |
++ container_of(work, struct kthread_delayed_work, work); |
900 |
++ struct kthread_worker *worker = work->worker; |
901 |
++ |
902 |
++ /* |
903 |
++ * del_timer_sync() must be called to make sure that the timer |
904 |
++ * callback is not running. The lock must be temporary released |
905 |
++ * to avoid a deadlock with the callback. In the meantime, |
906 |
++ * any queuing is blocked by setting the canceling counter. |
907 |
++ */ |
908 |
++ work->canceling++; |
909 |
++ spin_unlock_irqrestore(&worker->lock, *flags); |
910 |
++ del_timer_sync(&dwork->timer); |
911 |
++ spin_lock_irqsave(&worker->lock, *flags); |
912 |
++ work->canceling--; |
913 |
++} |
914 |
++ |
915 |
++/* |
916 |
++ * This function removes the work from the worker queue. |
917 |
++ * |
918 |
++ * It is called under worker->lock. The caller must make sure that |
919 |
++ * the timer used by delayed work is not running, e.g. by calling |
920 |
++ * kthread_cancel_delayed_work_timer(). |
921 |
+ * |
922 |
+ * The work might still be in use when this function finishes. See the |
923 |
+ * current_work proceed by the worker. |
924 |
+@@ -1019,28 +1049,8 @@ EXPORT_SYMBOL_GPL(kthread_flush_work); |
925 |
+ * Return: %true if @work was pending and successfully canceled, |
926 |
+ * %false if @work was not pending |
927 |
+ */ |
928 |
+-static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork, |
929 |
+- unsigned long *flags) |
930 |
++static bool __kthread_cancel_work(struct kthread_work *work) |
931 |
+ { |
932 |
+- /* Try to cancel the timer if exists. */ |
933 |
+- if (is_dwork) { |
934 |
+- struct kthread_delayed_work *dwork = |
935 |
+- container_of(work, struct kthread_delayed_work, work); |
936 |
+- struct kthread_worker *worker = work->worker; |
937 |
+- |
938 |
+- /* |
939 |
+- * del_timer_sync() must be called to make sure that the timer |
940 |
+- * callback is not running. The lock must be temporary released |
941 |
+- * to avoid a deadlock with the callback. In the meantime, |
942 |
+- * any queuing is blocked by setting the canceling counter. |
943 |
+- */ |
944 |
+- work->canceling++; |
945 |
+- spin_unlock_irqrestore(&worker->lock, *flags); |
946 |
+- del_timer_sync(&dwork->timer); |
947 |
+- spin_lock_irqsave(&worker->lock, *flags); |
948 |
+- work->canceling--; |
949 |
+- } |
950 |
+- |
951 |
+ /* |
952 |
+ * Try to remove the work from a worker list. It might either |
953 |
+ * be from worker->work_list or from worker->delayed_work_list. |
954 |
+@@ -1093,11 +1103,23 @@ bool kthread_mod_delayed_work(struct kthread_worker *worker, |
955 |
+ /* Work must not be used with >1 worker, see kthread_queue_work() */ |
956 |
+ WARN_ON_ONCE(work->worker != worker); |
957 |
+ |
958 |
+- /* Do not fight with another command that is canceling this work. */ |
959 |
++ /* |
960 |
++ * Temporary cancel the work but do not fight with another command |
961 |
++ * that is canceling the work as well. |
962 |
++ * |
963 |
++ * It is a bit tricky because of possible races with another |
964 |
++ * mod_delayed_work() and cancel_delayed_work() callers. |
965 |
++ * |
966 |
++ * The timer must be canceled first because worker->lock is released |
967 |
++ * when doing so. But the work can be removed from the queue (list) |
968 |
++ * only when it can be queued again so that the return value can |
969 |
++ * be used for reference counting. |
970 |
++ */ |
971 |
++ kthread_cancel_delayed_work_timer(work, &flags); |
972 |
+ if (work->canceling) |
973 |
+ goto out; |
974 |
++ ret = __kthread_cancel_work(work); |
975 |
+ |
976 |
+- ret = __kthread_cancel_work(work, true, &flags); |
977 |
+ fast_queue: |
978 |
+ __kthread_queue_delayed_work(worker, dwork, delay); |
979 |
+ out: |
980 |
+@@ -1119,7 +1141,10 @@ static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork) |
981 |
+ /* Work must not be used with >1 worker, see kthread_queue_work(). */ |
982 |
+ WARN_ON_ONCE(work->worker != worker); |
983 |
+ |
984 |
+- ret = __kthread_cancel_work(work, is_dwork, &flags); |
985 |
++ if (is_dwork) |
986 |
++ kthread_cancel_delayed_work_timer(work, &flags); |
987 |
++ |
988 |
++ ret = __kthread_cancel_work(work); |
989 |
+ |
990 |
+ if (worker->current_work != work) |
991 |
+ goto out_fast; |
992 |
+diff --git a/mm/huge_memory.c b/mm/huge_memory.c |
993 |
+index 7c374c0fcf0c7..4400957d8e4e2 100644 |
994 |
+--- a/mm/huge_memory.c |
995 |
++++ b/mm/huge_memory.c |
996 |
+@@ -62,6 +62,7 @@ static struct shrinker deferred_split_shrinker; |
997 |
+ |
998 |
+ static atomic_t huge_zero_refcount; |
999 |
+ struct page *huge_zero_page __read_mostly; |
1000 |
++unsigned long huge_zero_pfn __read_mostly = ~0UL; |
1001 |
+ |
1002 |
+ bool transparent_hugepage_enabled(struct vm_area_struct *vma) |
1003 |
+ { |
1004 |
+@@ -93,6 +94,7 @@ retry: |
1005 |
+ __free_pages(zero_page, compound_order(zero_page)); |
1006 |
+ goto retry; |
1007 |
+ } |
1008 |
++ WRITE_ONCE(huge_zero_pfn, page_to_pfn(zero_page)); |
1009 |
+ |
1010 |
+ /* We take additional reference here. It will be put back by shrinker */ |
1011 |
+ atomic_set(&huge_zero_refcount, 2); |
1012 |
+@@ -142,6 +144,7 @@ static unsigned long shrink_huge_zero_page_scan(struct shrinker *shrink, |
1013 |
+ if (atomic_cmpxchg(&huge_zero_refcount, 1, 0) == 1) { |
1014 |
+ struct page *zero_page = xchg(&huge_zero_page, NULL); |
1015 |
+ BUG_ON(zero_page == NULL); |
1016 |
++ WRITE_ONCE(huge_zero_pfn, ~0UL); |
1017 |
+ __free_pages(zero_page, compound_order(zero_page)); |
1018 |
+ return HPAGE_PMD_NR; |
1019 |
+ } |
1020 |
+@@ -2125,7 +2128,7 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd, |
1021 |
+ count_vm_event(THP_SPLIT_PMD); |
1022 |
+ |
1023 |
+ if (!vma_is_anonymous(vma)) { |
1024 |
+- _pmd = pmdp_huge_clear_flush_notify(vma, haddr, pmd); |
1025 |
++ old_pmd = pmdp_huge_clear_flush_notify(vma, haddr, pmd); |
1026 |
+ /* |
1027 |
+ * We are going to unmap this huge page. So |
1028 |
+ * just go ahead and zap it |
1029 |
+@@ -2134,16 +2137,25 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd, |
1030 |
+ zap_deposited_table(mm, pmd); |
1031 |
+ if (vma_is_dax(vma)) |
1032 |
+ return; |
1033 |
+- page = pmd_page(_pmd); |
1034 |
+- if (!PageDirty(page) && pmd_dirty(_pmd)) |
1035 |
+- set_page_dirty(page); |
1036 |
+- if (!PageReferenced(page) && pmd_young(_pmd)) |
1037 |
+- SetPageReferenced(page); |
1038 |
+- page_remove_rmap(page, true); |
1039 |
+- put_page(page); |
1040 |
++ if (unlikely(is_pmd_migration_entry(old_pmd))) { |
1041 |
++ swp_entry_t entry; |
1042 |
++ |
1043 |
++ entry = pmd_to_swp_entry(old_pmd); |
1044 |
++ page = migration_entry_to_page(entry); |
1045 |
++ } else { |
1046 |
++ page = pmd_page(old_pmd); |
1047 |
++ if (!PageDirty(page) && pmd_dirty(old_pmd)) |
1048 |
++ set_page_dirty(page); |
1049 |
++ if (!PageReferenced(page) && pmd_young(old_pmd)) |
1050 |
++ SetPageReferenced(page); |
1051 |
++ page_remove_rmap(page, true); |
1052 |
++ put_page(page); |
1053 |
++ } |
1054 |
+ add_mm_counter(mm, mm_counter_file(page), -HPAGE_PMD_NR); |
1055 |
+ return; |
1056 |
+- } else if (pmd_trans_huge(*pmd) && is_huge_zero_pmd(*pmd)) { |
1057 |
++ } |
1058 |
++ |
1059 |
++ if (is_huge_zero_pmd(*pmd)) { |
1060 |
+ /* |
1061 |
+ * FIXME: Do we want to invalidate secondary mmu by calling |
1062 |
+ * mmu_notifier_invalidate_range() see comments below inside |
1063 |
+@@ -2418,16 +2430,16 @@ void vma_adjust_trans_huge(struct vm_area_struct *vma, |
1064 |
+ static void unmap_page(struct page *page) |
1065 |
+ { |
1066 |
+ enum ttu_flags ttu_flags = TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS | |
1067 |
+- TTU_RMAP_LOCKED | TTU_SPLIT_HUGE_PMD; |
1068 |
+- bool unmap_success; |
1069 |
++ TTU_RMAP_LOCKED | TTU_SPLIT_HUGE_PMD | TTU_SYNC; |
1070 |
+ |
1071 |
+ VM_BUG_ON_PAGE(!PageHead(page), page); |
1072 |
+ |
1073 |
+ if (PageAnon(page)) |
1074 |
+ ttu_flags |= TTU_SPLIT_FREEZE; |
1075 |
+ |
1076 |
+- unmap_success = try_to_unmap(page, ttu_flags); |
1077 |
+- VM_BUG_ON_PAGE(!unmap_success, page); |
1078 |
++ try_to_unmap(page, ttu_flags); |
1079 |
++ |
1080 |
++ VM_WARN_ON_ONCE_PAGE(page_mapped(page), page); |
1081 |
+ } |
1082 |
+ |
1083 |
+ static void remap_page(struct page *page) |
1084 |
+@@ -2686,7 +2698,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) |
1085 |
+ struct pglist_data *pgdata = NODE_DATA(page_to_nid(head)); |
1086 |
+ struct anon_vma *anon_vma = NULL; |
1087 |
+ struct address_space *mapping = NULL; |
1088 |
+- int count, mapcount, extra_pins, ret; |
1089 |
++ int extra_pins, ret; |
1090 |
+ bool mlocked; |
1091 |
+ unsigned long flags; |
1092 |
+ pgoff_t end; |
1093 |
+@@ -2748,7 +2760,6 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) |
1094 |
+ |
1095 |
+ mlocked = PageMlocked(page); |
1096 |
+ unmap_page(head); |
1097 |
+- VM_BUG_ON_PAGE(compound_mapcount(head), head); |
1098 |
+ |
1099 |
+ /* Make sure the page is not on per-CPU pagevec as it takes pin */ |
1100 |
+ if (mlocked) |
1101 |
+@@ -2774,9 +2785,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) |
1102 |
+ |
1103 |
+ /* Prevent deferred_split_scan() touching ->_refcount */ |
1104 |
+ spin_lock(&pgdata->split_queue_lock); |
1105 |
+- count = page_count(head); |
1106 |
+- mapcount = total_mapcount(head); |
1107 |
+- if (!mapcount && page_ref_freeze(head, 1 + extra_pins)) { |
1108 |
++ if (page_ref_freeze(head, 1 + extra_pins)) { |
1109 |
+ if (!list_empty(page_deferred_list(head))) { |
1110 |
+ pgdata->split_queue_len--; |
1111 |
+ list_del(page_deferred_list(head)); |
1112 |
+@@ -2792,16 +2801,9 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) |
1113 |
+ } else |
1114 |
+ ret = 0; |
1115 |
+ } else { |
1116 |
+- if (IS_ENABLED(CONFIG_DEBUG_VM) && mapcount) { |
1117 |
+- pr_alert("total_mapcount: %u, page_count(): %u\n", |
1118 |
+- mapcount, count); |
1119 |
+- if (PageTail(page)) |
1120 |
+- dump_page(head, NULL); |
1121 |
+- dump_page(page, "total_mapcount(head) > 0"); |
1122 |
+- BUG(); |
1123 |
+- } |
1124 |
+ spin_unlock(&pgdata->split_queue_lock); |
1125 |
+-fail: if (mapping) |
1126 |
++fail: |
1127 |
++ if (mapping) |
1128 |
+ xa_unlock(&mapping->i_pages); |
1129 |
+ spin_unlock_irqrestore(zone_lru_lock(page_zone(head)), flags); |
1130 |
+ remap_page(head); |
1131 |
+diff --git a/mm/hugetlb.c b/mm/hugetlb.c |
1132 |
+index c69f12e4c1499..ebcf26bc4cd4b 100644 |
1133 |
+--- a/mm/hugetlb.c |
1134 |
++++ b/mm/hugetlb.c |
1135 |
+@@ -1391,15 +1391,12 @@ int PageHeadHuge(struct page *page_head) |
1136 |
+ return get_compound_page_dtor(page_head) == free_huge_page; |
1137 |
+ } |
1138 |
+ |
1139 |
+-pgoff_t __basepage_index(struct page *page) |
1140 |
++pgoff_t hugetlb_basepage_index(struct page *page) |
1141 |
+ { |
1142 |
+ struct page *page_head = compound_head(page); |
1143 |
+ pgoff_t index = page_index(page_head); |
1144 |
+ unsigned long compound_idx; |
1145 |
+ |
1146 |
+- if (!PageHuge(page_head)) |
1147 |
+- return page_index(page); |
1148 |
+- |
1149 |
+ if (compound_order(page_head) >= MAX_ORDER) |
1150 |
+ compound_idx = page_to_pfn(page) - page_to_pfn(page_head); |
1151 |
+ else |
1152 |
+diff --git a/mm/internal.h b/mm/internal.h |
1153 |
+index 397183c8fe47b..3a2e973138d36 100644 |
1154 |
+--- a/mm/internal.h |
1155 |
++++ b/mm/internal.h |
1156 |
+@@ -331,27 +331,52 @@ static inline void mlock_migrate_page(struct page *newpage, struct page *page) |
1157 |
+ extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma); |
1158 |
+ |
1159 |
+ /* |
1160 |
+- * At what user virtual address is page expected in @vma? |
1161 |
++ * At what user virtual address is page expected in vma? |
1162 |
++ * Returns -EFAULT if all of the page is outside the range of vma. |
1163 |
++ * If page is a compound head, the entire compound page is considered. |
1164 |
+ */ |
1165 |
+ static inline unsigned long |
1166 |
+-__vma_address(struct page *page, struct vm_area_struct *vma) |
1167 |
++vma_address(struct page *page, struct vm_area_struct *vma) |
1168 |
+ { |
1169 |
+- pgoff_t pgoff = page_to_pgoff(page); |
1170 |
+- return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
1171 |
++ pgoff_t pgoff; |
1172 |
++ unsigned long address; |
1173 |
++ |
1174 |
++ VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */ |
1175 |
++ pgoff = page_to_pgoff(page); |
1176 |
++ if (pgoff >= vma->vm_pgoff) { |
1177 |
++ address = vma->vm_start + |
1178 |
++ ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
1179 |
++ /* Check for address beyond vma (or wrapped through 0?) */ |
1180 |
++ if (address < vma->vm_start || address >= vma->vm_end) |
1181 |
++ address = -EFAULT; |
1182 |
++ } else if (PageHead(page) && |
1183 |
++ pgoff + (1UL << compound_order(page)) - 1 >= vma->vm_pgoff) { |
1184 |
++ /* Test above avoids possibility of wrap to 0 on 32-bit */ |
1185 |
++ address = vma->vm_start; |
1186 |
++ } else { |
1187 |
++ address = -EFAULT; |
1188 |
++ } |
1189 |
++ return address; |
1190 |
+ } |
1191 |
+ |
1192 |
++/* |
1193 |
++ * Then at what user virtual address will none of the page be found in vma? |
1194 |
++ * Assumes that vma_address() already returned a good starting address. |
1195 |
++ * If page is a compound head, the entire compound page is considered. |
1196 |
++ */ |
1197 |
+ static inline unsigned long |
1198 |
+-vma_address(struct page *page, struct vm_area_struct *vma) |
1199 |
++vma_address_end(struct page *page, struct vm_area_struct *vma) |
1200 |
+ { |
1201 |
+- unsigned long start, end; |
1202 |
+- |
1203 |
+- start = __vma_address(page, vma); |
1204 |
+- end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1); |
1205 |
+- |
1206 |
+- /* page should be within @vma mapping range */ |
1207 |
+- VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma); |
1208 |
+- |
1209 |
+- return max(start, vma->vm_start); |
1210 |
++ pgoff_t pgoff; |
1211 |
++ unsigned long address; |
1212 |
++ |
1213 |
++ VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */ |
1214 |
++ pgoff = page_to_pgoff(page) + (1UL << compound_order(page)); |
1215 |
++ address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
1216 |
++ /* Check for address beyond vma (or wrapped through 0?) */ |
1217 |
++ if (address < vma->vm_start || address > vma->vm_end) |
1218 |
++ address = vma->vm_end; |
1219 |
++ return address; |
1220 |
+ } |
1221 |
+ |
1222 |
+ #else /* !CONFIG_MMU */ |
1223 |
+diff --git a/mm/memory.c b/mm/memory.c |
1224 |
+index c2011c51f15de..49b546cdce0d2 100644 |
1225 |
+--- a/mm/memory.c |
1226 |
++++ b/mm/memory.c |
1227 |
+@@ -1439,7 +1439,18 @@ static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
1228 |
+ else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1229 |
+ goto next; |
1230 |
+ /* fall through */ |
1231 |
++ } else if (details && details->single_page && |
1232 |
++ PageTransCompound(details->single_page) && |
1233 |
++ next - addr == HPAGE_PMD_SIZE && pmd_none(*pmd)) { |
1234 |
++ spinlock_t *ptl = pmd_lock(tlb->mm, pmd); |
1235 |
++ /* |
1236 |
++ * Take and drop THP pmd lock so that we cannot return |
1237 |
++ * prematurely, while zap_huge_pmd() has cleared *pmd, |
1238 |
++ * but not yet decremented compound_mapcount(). |
1239 |
++ */ |
1240 |
++ spin_unlock(ptl); |
1241 |
+ } |
1242 |
++ |
1243 |
+ /* |
1244 |
+ * Here there can be other concurrent MADV_DONTNEED or |
1245 |
+ * trans huge page faults running, and if the pmd is |
1246 |
+@@ -2924,6 +2935,36 @@ static inline void unmap_mapping_range_tree(struct rb_root_cached *root, |
1247 |
+ } |
1248 |
+ } |
1249 |
+ |
1250 |
++/** |
1251 |
++ * unmap_mapping_page() - Unmap single page from processes. |
1252 |
++ * @page: The locked page to be unmapped. |
1253 |
++ * |
1254 |
++ * Unmap this page from any userspace process which still has it mmaped. |
1255 |
++ * Typically, for efficiency, the range of nearby pages has already been |
1256 |
++ * unmapped by unmap_mapping_pages() or unmap_mapping_range(). But once |
1257 |
++ * truncation or invalidation holds the lock on a page, it may find that |
1258 |
++ * the page has been remapped again: and then uses unmap_mapping_page() |
1259 |
++ * to unmap it finally. |
1260 |
++ */ |
1261 |
++void unmap_mapping_page(struct page *page) |
1262 |
++{ |
1263 |
++ struct address_space *mapping = page->mapping; |
1264 |
++ struct zap_details details = { }; |
1265 |
++ |
1266 |
++ VM_BUG_ON(!PageLocked(page)); |
1267 |
++ VM_BUG_ON(PageTail(page)); |
1268 |
++ |
1269 |
++ details.check_mapping = mapping; |
1270 |
++ details.first_index = page->index; |
1271 |
++ details.last_index = page->index + hpage_nr_pages(page) - 1; |
1272 |
++ details.single_page = page; |
1273 |
++ |
1274 |
++ i_mmap_lock_write(mapping); |
1275 |
++ if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) |
1276 |
++ unmap_mapping_range_tree(&mapping->i_mmap, &details); |
1277 |
++ i_mmap_unlock_write(mapping); |
1278 |
++} |
1279 |
++ |
1280 |
+ /** |
1281 |
+ * unmap_mapping_pages() - Unmap pages from processes. |
1282 |
+ * @mapping: The address space containing pages to be unmapped. |
1283 |
+diff --git a/mm/page_vma_mapped.c b/mm/page_vma_mapped.c |
1284 |
+index 11df03e71288c..edca786093187 100644 |
1285 |
+--- a/mm/page_vma_mapped.c |
1286 |
++++ b/mm/page_vma_mapped.c |
1287 |
+@@ -111,6 +111,13 @@ static bool check_pte(struct page_vma_mapped_walk *pvmw) |
1288 |
+ return pfn_in_hpage(pvmw->page, pfn); |
1289 |
+ } |
1290 |
+ |
1291 |
++static void step_forward(struct page_vma_mapped_walk *pvmw, unsigned long size) |
1292 |
++{ |
1293 |
++ pvmw->address = (pvmw->address + size) & ~(size - 1); |
1294 |
++ if (!pvmw->address) |
1295 |
++ pvmw->address = ULONG_MAX; |
1296 |
++} |
1297 |
++ |
1298 |
+ /** |
1299 |
+ * page_vma_mapped_walk - check if @pvmw->page is mapped in @pvmw->vma at |
1300 |
+ * @pvmw->address |
1301 |
+@@ -139,6 +146,7 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw) |
1302 |
+ { |
1303 |
+ struct mm_struct *mm = pvmw->vma->vm_mm; |
1304 |
+ struct page *page = pvmw->page; |
1305 |
++ unsigned long end; |
1306 |
+ pgd_t *pgd; |
1307 |
+ p4d_t *p4d; |
1308 |
+ pud_t *pud; |
1309 |
+@@ -148,10 +156,11 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw) |
1310 |
+ if (pvmw->pmd && !pvmw->pte) |
1311 |
+ return not_found(pvmw); |
1312 |
+ |
1313 |
+- if (pvmw->pte) |
1314 |
+- goto next_pte; |
1315 |
++ if (unlikely(PageHuge(page))) { |
1316 |
++ /* The only possible mapping was handled on last iteration */ |
1317 |
++ if (pvmw->pte) |
1318 |
++ return not_found(pvmw); |
1319 |
+ |
1320 |
+- if (unlikely(PageHuge(pvmw->page))) { |
1321 |
+ /* when pud is not present, pte will be NULL */ |
1322 |
+ pvmw->pte = huge_pte_offset(mm, pvmw->address, |
1323 |
+ PAGE_SIZE << compound_order(page)); |
1324 |
+@@ -164,78 +173,108 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw) |
1325 |
+ return not_found(pvmw); |
1326 |
+ return true; |
1327 |
+ } |
1328 |
+-restart: |
1329 |
+- pgd = pgd_offset(mm, pvmw->address); |
1330 |
+- if (!pgd_present(*pgd)) |
1331 |
+- return false; |
1332 |
+- p4d = p4d_offset(pgd, pvmw->address); |
1333 |
+- if (!p4d_present(*p4d)) |
1334 |
+- return false; |
1335 |
+- pud = pud_offset(p4d, pvmw->address); |
1336 |
+- if (!pud_present(*pud)) |
1337 |
+- return false; |
1338 |
+- pvmw->pmd = pmd_offset(pud, pvmw->address); |
1339 |
++ |
1340 |
+ /* |
1341 |
+- * Make sure the pmd value isn't cached in a register by the |
1342 |
+- * compiler and used as a stale value after we've observed a |
1343 |
+- * subsequent update. |
1344 |
++ * Seek to next pte only makes sense for THP. |
1345 |
++ * But more important than that optimization, is to filter out |
1346 |
++ * any PageKsm page: whose page->index misleads vma_address() |
1347 |
++ * and vma_address_end() to disaster. |
1348 |
+ */ |
1349 |
+- pmde = READ_ONCE(*pvmw->pmd); |
1350 |
+- if (pmd_trans_huge(pmde) || is_pmd_migration_entry(pmde)) { |
1351 |
+- pvmw->ptl = pmd_lock(mm, pvmw->pmd); |
1352 |
+- if (likely(pmd_trans_huge(*pvmw->pmd))) { |
1353 |
+- if (pvmw->flags & PVMW_MIGRATION) |
1354 |
+- return not_found(pvmw); |
1355 |
+- if (pmd_page(*pvmw->pmd) != page) |
1356 |
+- return not_found(pvmw); |
1357 |
+- return true; |
1358 |
+- } else if (!pmd_present(*pvmw->pmd)) { |
1359 |
+- if (thp_migration_supported()) { |
1360 |
+- if (!(pvmw->flags & PVMW_MIGRATION)) |
1361 |
++ end = PageTransCompound(page) ? |
1362 |
++ vma_address_end(page, pvmw->vma) : |
1363 |
++ pvmw->address + PAGE_SIZE; |
1364 |
++ if (pvmw->pte) |
1365 |
++ goto next_pte; |
1366 |
++restart: |
1367 |
++ do { |
1368 |
++ pgd = pgd_offset(mm, pvmw->address); |
1369 |
++ if (!pgd_present(*pgd)) { |
1370 |
++ step_forward(pvmw, PGDIR_SIZE); |
1371 |
++ continue; |
1372 |
++ } |
1373 |
++ p4d = p4d_offset(pgd, pvmw->address); |
1374 |
++ if (!p4d_present(*p4d)) { |
1375 |
++ step_forward(pvmw, P4D_SIZE); |
1376 |
++ continue; |
1377 |
++ } |
1378 |
++ pud = pud_offset(p4d, pvmw->address); |
1379 |
++ if (!pud_present(*pud)) { |
1380 |
++ step_forward(pvmw, PUD_SIZE); |
1381 |
++ continue; |
1382 |
++ } |
1383 |
++ |
1384 |
++ pvmw->pmd = pmd_offset(pud, pvmw->address); |
1385 |
++ /* |
1386 |
++ * Make sure the pmd value isn't cached in a register by the |
1387 |
++ * compiler and used as a stale value after we've observed a |
1388 |
++ * subsequent update. |
1389 |
++ */ |
1390 |
++ pmde = READ_ONCE(*pvmw->pmd); |
1391 |
++ |
1392 |
++ if (pmd_trans_huge(pmde) || is_pmd_migration_entry(pmde)) { |
1393 |
++ pvmw->ptl = pmd_lock(mm, pvmw->pmd); |
1394 |
++ pmde = *pvmw->pmd; |
1395 |
++ if (likely(pmd_trans_huge(pmde))) { |
1396 |
++ if (pvmw->flags & PVMW_MIGRATION) |
1397 |
+ return not_found(pvmw); |
1398 |
+- if (is_migration_entry(pmd_to_swp_entry(*pvmw->pmd))) { |
1399 |
+- swp_entry_t entry = pmd_to_swp_entry(*pvmw->pmd); |
1400 |
++ if (pmd_page(pmde) != page) |
1401 |
++ return not_found(pvmw); |
1402 |
++ return true; |
1403 |
++ } |
1404 |
++ if (!pmd_present(pmde)) { |
1405 |
++ swp_entry_t entry; |
1406 |
+ |
1407 |
+- if (migration_entry_to_page(entry) != page) |
1408 |
+- return not_found(pvmw); |
1409 |
+- return true; |
1410 |
+- } |
1411 |
++ if (!thp_migration_supported() || |
1412 |
++ !(pvmw->flags & PVMW_MIGRATION)) |
1413 |
++ return not_found(pvmw); |
1414 |
++ entry = pmd_to_swp_entry(pmde); |
1415 |
++ if (!is_migration_entry(entry) || |
1416 |
++ migration_entry_to_page(entry) != page) |
1417 |
++ return not_found(pvmw); |
1418 |
++ return true; |
1419 |
+ } |
1420 |
+- return not_found(pvmw); |
1421 |
+- } else { |
1422 |
+ /* THP pmd was split under us: handle on pte level */ |
1423 |
+ spin_unlock(pvmw->ptl); |
1424 |
+ pvmw->ptl = NULL; |
1425 |
++ } else if (!pmd_present(pmde)) { |
1426 |
++ /* |
1427 |
++ * If PVMW_SYNC, take and drop THP pmd lock so that we |
1428 |
++ * cannot return prematurely, while zap_huge_pmd() has |
1429 |
++ * cleared *pmd but not decremented compound_mapcount(). |
1430 |
++ */ |
1431 |
++ if ((pvmw->flags & PVMW_SYNC) && |
1432 |
++ PageTransCompound(page)) { |
1433 |
++ spinlock_t *ptl = pmd_lock(mm, pvmw->pmd); |
1434 |
++ |
1435 |
++ spin_unlock(ptl); |
1436 |
++ } |
1437 |
++ step_forward(pvmw, PMD_SIZE); |
1438 |
++ continue; |
1439 |
+ } |
1440 |
+- } else if (!pmd_present(pmde)) { |
1441 |
+- return false; |
1442 |
+- } |
1443 |
+- if (!map_pte(pvmw)) |
1444 |
+- goto next_pte; |
1445 |
+- while (1) { |
1446 |
++ if (!map_pte(pvmw)) |
1447 |
++ goto next_pte; |
1448 |
++this_pte: |
1449 |
+ if (check_pte(pvmw)) |
1450 |
+ return true; |
1451 |
+ next_pte: |
1452 |
+- /* Seek to next pte only makes sense for THP */ |
1453 |
+- if (!PageTransHuge(pvmw->page) || PageHuge(pvmw->page)) |
1454 |
+- return not_found(pvmw); |
1455 |
+ do { |
1456 |
+ pvmw->address += PAGE_SIZE; |
1457 |
+- if (pvmw->address >= pvmw->vma->vm_end || |
1458 |
+- pvmw->address >= |
1459 |
+- __vma_address(pvmw->page, pvmw->vma) + |
1460 |
+- hpage_nr_pages(pvmw->page) * PAGE_SIZE) |
1461 |
++ if (pvmw->address >= end) |
1462 |
+ return not_found(pvmw); |
1463 |
+ /* Did we cross page table boundary? */ |
1464 |
+- if (pvmw->address % PMD_SIZE == 0) { |
1465 |
+- pte_unmap(pvmw->pte); |
1466 |
++ if ((pvmw->address & (PMD_SIZE - PAGE_SIZE)) == 0) { |
1467 |
+ if (pvmw->ptl) { |
1468 |
+ spin_unlock(pvmw->ptl); |
1469 |
+ pvmw->ptl = NULL; |
1470 |
+ } |
1471 |
++ pte_unmap(pvmw->pte); |
1472 |
++ pvmw->pte = NULL; |
1473 |
+ goto restart; |
1474 |
+- } else { |
1475 |
+- pvmw->pte++; |
1476 |
++ } |
1477 |
++ pvmw->pte++; |
1478 |
++ if ((pvmw->flags & PVMW_SYNC) && !pvmw->ptl) { |
1479 |
++ pvmw->ptl = pte_lockptr(mm, pvmw->pmd); |
1480 |
++ spin_lock(pvmw->ptl); |
1481 |
+ } |
1482 |
+ } while (pte_none(*pvmw->pte)); |
1483 |
+ |
1484 |
+@@ -243,7 +282,10 @@ next_pte: |
1485 |
+ pvmw->ptl = pte_lockptr(mm, pvmw->pmd); |
1486 |
+ spin_lock(pvmw->ptl); |
1487 |
+ } |
1488 |
+- } |
1489 |
++ goto this_pte; |
1490 |
++ } while (pvmw->address < end); |
1491 |
++ |
1492 |
++ return false; |
1493 |
+ } |
1494 |
+ |
1495 |
+ /** |
1496 |
+@@ -262,14 +304,10 @@ int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
1497 |
+ .vma = vma, |
1498 |
+ .flags = PVMW_SYNC, |
1499 |
+ }; |
1500 |
+- unsigned long start, end; |
1501 |
+- |
1502 |
+- start = __vma_address(page, vma); |
1503 |
+- end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1); |
1504 |
+ |
1505 |
+- if (unlikely(end < vma->vm_start || start >= vma->vm_end)) |
1506 |
++ pvmw.address = vma_address(page, vma); |
1507 |
++ if (pvmw.address == -EFAULT) |
1508 |
+ return 0; |
1509 |
+- pvmw.address = max(start, vma->vm_start); |
1510 |
+ if (!page_vma_mapped_walk(&pvmw)) |
1511 |
+ return 0; |
1512 |
+ page_vma_mapped_walk_done(&pvmw); |
1513 |
+diff --git a/mm/pgtable-generic.c b/mm/pgtable-generic.c |
1514 |
+index cf2af04b34b97..36770fcdc3582 100644 |
1515 |
+--- a/mm/pgtable-generic.c |
1516 |
++++ b/mm/pgtable-generic.c |
1517 |
+@@ -125,8 +125,8 @@ pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma, unsigned long address, |
1518 |
+ { |
1519 |
+ pmd_t pmd; |
1520 |
+ VM_BUG_ON(address & ~HPAGE_PMD_MASK); |
1521 |
+- VM_BUG_ON((pmd_present(*pmdp) && !pmd_trans_huge(*pmdp) && |
1522 |
+- !pmd_devmap(*pmdp)) || !pmd_present(*pmdp)); |
1523 |
++ VM_BUG_ON(pmd_present(*pmdp) && !pmd_trans_huge(*pmdp) && |
1524 |
++ !pmd_devmap(*pmdp)); |
1525 |
+ pmd = pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp); |
1526 |
+ flush_pmd_tlb_range(vma, address, address + HPAGE_PMD_SIZE); |
1527 |
+ return pmd; |
1528 |
+diff --git a/mm/rmap.c b/mm/rmap.c |
1529 |
+index 1bd94ea62f7f1..699f445e3e78c 100644 |
1530 |
+--- a/mm/rmap.c |
1531 |
++++ b/mm/rmap.c |
1532 |
+@@ -686,7 +686,6 @@ static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) |
1533 |
+ */ |
1534 |
+ unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) |
1535 |
+ { |
1536 |
+- unsigned long address; |
1537 |
+ if (PageAnon(page)) { |
1538 |
+ struct anon_vma *page__anon_vma = page_anon_vma(page); |
1539 |
+ /* |
1540 |
+@@ -696,15 +695,13 @@ unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) |
1541 |
+ if (!vma->anon_vma || !page__anon_vma || |
1542 |
+ vma->anon_vma->root != page__anon_vma->root) |
1543 |
+ return -EFAULT; |
1544 |
+- } else if (page->mapping) { |
1545 |
+- if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping) |
1546 |
+- return -EFAULT; |
1547 |
+- } else |
1548 |
++ } else if (!vma->vm_file) { |
1549 |
+ return -EFAULT; |
1550 |
+- address = __vma_address(page, vma); |
1551 |
+- if (unlikely(address < vma->vm_start || address >= vma->vm_end)) |
1552 |
++ } else if (vma->vm_file->f_mapping != compound_head(page)->mapping) { |
1553 |
+ return -EFAULT; |
1554 |
+- return address; |
1555 |
++ } |
1556 |
++ |
1557 |
++ return vma_address(page, vma); |
1558 |
+ } |
1559 |
+ |
1560 |
+ pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address) |
1561 |
+@@ -896,7 +893,7 @@ static bool page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
1562 |
+ * We have to assume the worse case ie pmd for invalidation. Note that |
1563 |
+ * the page can not be free from this function. |
1564 |
+ */ |
1565 |
+- end = min(vma->vm_end, start + (PAGE_SIZE << compound_order(page))); |
1566 |
++ end = vma_address_end(page, vma); |
1567 |
+ mmu_notifier_invalidate_range_start(vma->vm_mm, start, end); |
1568 |
+ |
1569 |
+ while (page_vma_mapped_walk(&pvmw)) { |
1570 |
+@@ -1348,6 +1345,15 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
1571 |
+ unsigned long start = address, end; |
1572 |
+ enum ttu_flags flags = (enum ttu_flags)arg; |
1573 |
+ |
1574 |
++ /* |
1575 |
++ * When racing against e.g. zap_pte_range() on another cpu, |
1576 |
++ * in between its ptep_get_and_clear_full() and page_remove_rmap(), |
1577 |
++ * try_to_unmap() may return false when it is about to become true, |
1578 |
++ * if page table locking is skipped: use TTU_SYNC to wait for that. |
1579 |
++ */ |
1580 |
++ if (flags & TTU_SYNC) |
1581 |
++ pvmw.flags = PVMW_SYNC; |
1582 |
++ |
1583 |
+ /* munlock has nothing to gain from examining un-locked vmas */ |
1584 |
+ if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED)) |
1585 |
+ return true; |
1586 |
+@@ -1369,7 +1375,8 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
1587 |
+ * Note that the page can not be free in this function as call of |
1588 |
+ * try_to_unmap() must hold a reference on the page. |
1589 |
+ */ |
1590 |
+- end = min(vma->vm_end, start + (PAGE_SIZE << compound_order(page))); |
1591 |
++ end = PageKsm(page) ? |
1592 |
++ address + PAGE_SIZE : vma_address_end(page, vma); |
1593 |
+ if (PageHuge(page)) { |
1594 |
+ /* |
1595 |
+ * If sharing is possible, start and end will be adjusted |
1596 |
+@@ -1682,9 +1689,9 @@ static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg) |
1597 |
+ return is_vma_temporary_stack(vma); |
1598 |
+ } |
1599 |
+ |
1600 |
+-static int page_mapcount_is_zero(struct page *page) |
1601 |
++static int page_not_mapped(struct page *page) |
1602 |
+ { |
1603 |
+- return !total_mapcount(page); |
1604 |
++ return !page_mapped(page); |
1605 |
+ } |
1606 |
+ |
1607 |
+ /** |
1608 |
+@@ -1702,7 +1709,7 @@ bool try_to_unmap(struct page *page, enum ttu_flags flags) |
1609 |
+ struct rmap_walk_control rwc = { |
1610 |
+ .rmap_one = try_to_unmap_one, |
1611 |
+ .arg = (void *)flags, |
1612 |
+- .done = page_mapcount_is_zero, |
1613 |
++ .done = page_not_mapped, |
1614 |
+ .anon_lock = page_lock_anon_vma_read, |
1615 |
+ }; |
1616 |
+ |
1617 |
+@@ -1723,14 +1730,15 @@ bool try_to_unmap(struct page *page, enum ttu_flags flags) |
1618 |
+ else |
1619 |
+ rmap_walk(page, &rwc); |
1620 |
+ |
1621 |
+- return !page_mapcount(page) ? true : false; |
1622 |
++ /* |
1623 |
++ * When racing against e.g. zap_pte_range() on another cpu, |
1624 |
++ * in between its ptep_get_and_clear_full() and page_remove_rmap(), |
1625 |
++ * try_to_unmap() may return false when it is about to become true, |
1626 |
++ * if page table locking is skipped: use TTU_SYNC to wait for that. |
1627 |
++ */ |
1628 |
++ return !page_mapcount(page); |
1629 |
+ } |
1630 |
+ |
1631 |
+-static int page_not_mapped(struct page *page) |
1632 |
+-{ |
1633 |
+- return !page_mapped(page); |
1634 |
+-}; |
1635 |
+- |
1636 |
+ /** |
1637 |
+ * try_to_munlock - try to munlock a page |
1638 |
+ * @page: the page to be munlocked |
1639 |
+@@ -1825,6 +1833,7 @@ static void rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc, |
1640 |
+ struct vm_area_struct *vma = avc->vma; |
1641 |
+ unsigned long address = vma_address(page, vma); |
1642 |
+ |
1643 |
++ VM_BUG_ON_VMA(address == -EFAULT, vma); |
1644 |
+ cond_resched(); |
1645 |
+ |
1646 |
+ if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1647 |
+@@ -1879,6 +1888,7 @@ static void rmap_walk_file(struct page *page, struct rmap_walk_control *rwc, |
1648 |
+ pgoff_start, pgoff_end) { |
1649 |
+ unsigned long address = vma_address(page, vma); |
1650 |
+ |
1651 |
++ VM_BUG_ON_VMA(address == -EFAULT, vma); |
1652 |
+ cond_resched(); |
1653 |
+ |
1654 |
+ if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1655 |
+diff --git a/mm/truncate.c b/mm/truncate.c |
1656 |
+index 71b65aab80775..43c73db17a0a6 100644 |
1657 |
+--- a/mm/truncate.c |
1658 |
++++ b/mm/truncate.c |
1659 |
+@@ -175,13 +175,10 @@ void do_invalidatepage(struct page *page, unsigned int offset, |
1660 |
+ * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1661 |
+ * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
1662 |
+ */ |
1663 |
+-static void |
1664 |
+-truncate_cleanup_page(struct address_space *mapping, struct page *page) |
1665 |
++static void truncate_cleanup_page(struct page *page) |
1666 |
+ { |
1667 |
+- if (page_mapped(page)) { |
1668 |
+- pgoff_t nr = PageTransHuge(page) ? HPAGE_PMD_NR : 1; |
1669 |
+- unmap_mapping_pages(mapping, page->index, nr, false); |
1670 |
+- } |
1671 |
++ if (page_mapped(page)) |
1672 |
++ unmap_mapping_page(page); |
1673 |
+ |
1674 |
+ if (page_has_private(page)) |
1675 |
+ do_invalidatepage(page, 0, PAGE_SIZE); |
1676 |
+@@ -226,7 +223,7 @@ int truncate_inode_page(struct address_space *mapping, struct page *page) |
1677 |
+ if (page->mapping != mapping) |
1678 |
+ return -EIO; |
1679 |
+ |
1680 |
+- truncate_cleanup_page(mapping, page); |
1681 |
++ truncate_cleanup_page(page); |
1682 |
+ delete_from_page_cache(page); |
1683 |
+ return 0; |
1684 |
+ } |
1685 |
+@@ -364,7 +361,7 @@ void truncate_inode_pages_range(struct address_space *mapping, |
1686 |
+ pagevec_add(&locked_pvec, page); |
1687 |
+ } |
1688 |
+ for (i = 0; i < pagevec_count(&locked_pvec); i++) |
1689 |
+- truncate_cleanup_page(mapping, locked_pvec.pages[i]); |
1690 |
++ truncate_cleanup_page(locked_pvec.pages[i]); |
1691 |
+ delete_from_page_cache_batch(mapping, &locked_pvec); |
1692 |
+ for (i = 0; i < pagevec_count(&locked_pvec); i++) |
1693 |
+ unlock_page(locked_pvec.pages[i]); |
1694 |
+@@ -703,6 +700,16 @@ int invalidate_inode_pages2_range(struct address_space *mapping, |
1695 |
+ continue; |
1696 |
+ } |
1697 |
+ |
1698 |
++ if (!did_range_unmap && page_mapped(page)) { |
1699 |
++ /* |
1700 |
++ * If page is mapped, before taking its lock, |
1701 |
++ * zap the rest of the file in one hit. |
1702 |
++ */ |
1703 |
++ unmap_mapping_pages(mapping, index, |
1704 |
++ (1 + end - index), false); |
1705 |
++ did_range_unmap = 1; |
1706 |
++ } |
1707 |
++ |
1708 |
+ lock_page(page); |
1709 |
+ WARN_ON(page_to_index(page) != index); |
1710 |
+ if (page->mapping != mapping) { |
1711 |
+@@ -710,23 +717,11 @@ int invalidate_inode_pages2_range(struct address_space *mapping, |
1712 |
+ continue; |
1713 |
+ } |
1714 |
+ wait_on_page_writeback(page); |
1715 |
+- if (page_mapped(page)) { |
1716 |
+- if (!did_range_unmap) { |
1717 |
+- /* |
1718 |
+- * Zap the rest of the file in one hit. |
1719 |
+- */ |
1720 |
+- unmap_mapping_pages(mapping, index, |
1721 |
+- (1 + end - index), false); |
1722 |
+- did_range_unmap = 1; |
1723 |
+- } else { |
1724 |
+- /* |
1725 |
+- * Just zap this page |
1726 |
+- */ |
1727 |
+- unmap_mapping_pages(mapping, index, |
1728 |
+- 1, false); |
1729 |
+- } |
1730 |
+- } |
1731 |
++ |
1732 |
++ if (page_mapped(page)) |
1733 |
++ unmap_mapping_page(page); |
1734 |
+ BUG_ON(page_mapped(page)); |
1735 |
++ |
1736 |
+ ret2 = do_launder_page(mapping, page); |
1737 |
+ if (ret2 == 0) { |
1738 |
+ if (!invalidate_complete_page2(mapping, page)) |