Gentoo Archives: gentoo-commits

From: Mike Pagano <mpagano@g.o>
To: gentoo-commits@l.g.o
Subject: [gentoo-commits] proj/linux-patches:5.4 commit in: /
Date: Sat, 09 May 2020 22:12:20
Message-Id: 1589062296.69d12816ea9f46070999271a2932ee9ff8954950.mpagano@gentoo
1 commit: 69d12816ea9f46070999271a2932ee9ff8954950
2 Author: Mike Pagano <mpagano <AT> gentoo <DOT> org>
3 AuthorDate: Sat May 9 22:11:36 2020 +0000
4 Commit: Mike Pagano <mpagano <AT> gentoo <DOT> org>
5 CommitDate: Sat May 9 22:11:36 2020 +0000
6 URL: https://gitweb.gentoo.org/proj/linux-patches.git/commit/?id=69d12816
7
8 x86: Fix early boot crash on gcc-10
9
10 Bug: https://bugs.gentoo.org/721734
11
12 Signed-off-by: Mike Pagano <mpagano <AT> gentoo.org>
13
14 0000_README | 4 +
15 1700_x86-gcc-10-early-boot-crash-fix.patch | 131 +++++++++++++++++++++++++++++
16 2 files changed, 135 insertions(+)
17
18 diff --git a/0000_README b/0000_README
19 index c1ac1b1..ea8edd3 100644
20 --- a/0000_README
21 +++ b/0000_README
22 @@ -207,6 +207,10 @@ Patch: 1510_fs-enable-link-security-restrictions-by-default.patch
23 From: http://sources.debian.net/src/linux/3.16.7-ckt4-3/debian/patches/debian/fs-enable-link-security-restrictions-by-default.patch/
24 Desc: Enable link security restrictions by default.
25
26 +Patch: 1700_x86-gcc-10-early-boot-crash-fix.patch
27 +From: https://https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/patch/?id=f670269a42bfdd2c83a1118cc3d1b475547eac22
28 +Desc: x86: Fix early boot crash on gcc-10
29 +
30 Patch: 2000_BT-Check-key-sizes-only-if-Secure-Simple-Pairing-enabled.patch
31 From: https://lore.kernel.org/linux-bluetooth/20190522070540.48895-1-marcel@××××××××.org/raw
32 Desc: Bluetooth: Check key sizes only when Secure Simple Pairing is enabled. See bug #686758
33
34 diff --git a/1700_x86-gcc-10-early-boot-crash-fix.patch b/1700_x86-gcc-10-early-boot-crash-fix.patch
35 new file mode 100644
36 index 0000000..8cdf651
37 --- /dev/null
38 +++ b/1700_x86-gcc-10-early-boot-crash-fix.patch
39 @@ -0,0 +1,131 @@
40 +From f670269a42bfdd2c83a1118cc3d1b475547eac22 Mon Sep 17 00:00:00 2001
41 +From: Borislav Petkov <bp@××××.de>
42 +Date: Wed, 22 Apr 2020 18:11:30 +0200
43 +Subject: x86: Fix early boot crash on gcc-10, next try
44 +MIME-Version: 1.0
45 +Content-Type: text/plain; charset=UTF-8
46 +Content-Transfer-Encoding: 8bit
47 +
48 +... or the odyssey of trying to disable the stack protector for the
49 +function which generates the stack canary value.
50 +
51 +The whole story started with Sergei reporting a boot crash with a kernel
52 +built with gcc-10:
53 +
54 + Kernel panic — not syncing: stack-protector: Kernel stack is corrupted in: start_secondary
55 + CPU: 1 PID: 0 Comm: swapper/1 Not tainted 5.6.0-rc5—00235—gfffb08b37df9 #139
56 + Hardware name: Gigabyte Technology Co., Ltd. To be filled by O.E.M./H77M—D3H, BIOS F12 11/14/2013
57 + Call Trace:
58 + dump_stack
59 + panic
60 + ? start_secondary
61 + __stack_chk_fail
62 + start_secondary
63 + secondary_startup_64
64 + -—-[ end Kernel panic — not syncing: stack—protector: Kernel stack is corrupted in: start_secondary
65 +
66 +This happens because gcc-10 tail-call optimizes the last function call
67 +in start_secondary() - cpu_startup_entry() - and thus emits a stack
68 +canary check which fails because the canary value changes after the
69 +boot_init_stack_canary() call.
70 +
71 +To fix that, the initial attempt was to mark the one function which
72 +generates the stack canary with:
73 +
74 + __attribute__((optimize("-fno-stack-protector"))) ... start_secondary(void *unused)
75 +
76 +however, using the optimize attribute doesn't work cumulatively
77 +as the attribute does not add to but rather replaces previously
78 +supplied optimization options - roughly all -fxxx options.
79 +
80 +The key one among them being -fno-omit-frame-pointer and thus leading to
81 +not present frame pointer - frame pointer which the kernel needs.
82 +
83 +The next attempt to prevent compilers from tail-call optimizing
84 +the last function call cpu_startup_entry(), shy of carving out
85 +start_secondary() into a separate compilation unit and building it with
86 +-fno-stack-protector, is this one.
87 +
88 +The current solution is short and sweet, and reportedly, is supported by
89 +both compilers so let's see how far we'll get this time.
90 +
91 +Reported-by: Sergei Trofimovich <slyfox@g.o>
92 +Signed-off-by: Borislav Petkov <bp@××××.de>
93 +Reviewed-by: Nick Desaulniers <ndesaulniers@××××××.com>
94 +Reviewed-by: Kees Cook <keescook@××××××××.org>
95 +Link: https://lkml.kernel.org/r/20200314164451.346497-1-slyfox@g.o
96 +---
97 + arch/x86/include/asm/stackprotector.h | 7 ++++++-
98 + arch/x86/kernel/smpboot.c | 8 ++++++++
99 + arch/x86/xen/smp_pv.c | 1 +
100 + include/linux/compiler.h | 6 ++++++
101 + 4 files changed, 21 insertions(+), 1 deletion(-)
102 +
103 +diff --git a/arch/x86/include/asm/stackprotector.h b/arch/x86/include/asm/stackprotector.h
104 +index 91e29b6a86a5..9804a7957f4e 100644
105 +--- a/arch/x86/include/asm/stackprotector.h
106 ++++ b/arch/x86/include/asm/stackprotector.h
107 +@@ -55,8 +55,13 @@
108 + /*
109 + * Initialize the stackprotector canary value.
110 + *
111 +- * NOTE: this must only be called from functions that never return,
112 ++ * NOTE: this must only be called from functions that never return
113 + * and it must always be inlined.
114 ++ *
115 ++ * In addition, it should be called from a compilation unit for which
116 ++ * stack protector is disabled. Alternatively, the caller should not end
117 ++ * with a function call which gets tail-call optimized as that would
118 ++ * lead to checking a modified canary value.
119 + */
120 + static __always_inline void boot_init_stack_canary(void)
121 + {
122 +diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
123 +index fe3ab9632f3b..4f275ac7830b 100644
124 +--- a/arch/x86/kernel/smpboot.c
125 ++++ b/arch/x86/kernel/smpboot.c
126 +@@ -266,6 +266,14 @@ static void notrace start_secondary(void *unused)
127 +
128 + wmb();
129 + cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
130 ++
131 ++ /*
132 ++ * Prevent tail call to cpu_startup_entry() because the stack protector
133 ++ * guard has been changed a couple of function calls up, in
134 ++ * boot_init_stack_canary() and must not be checked before tail calling
135 ++ * another function.
136 ++ */
137 ++ prevent_tail_call_optimization();
138 + }
139 +
140 + /**
141 +diff --git a/arch/x86/xen/smp_pv.c b/arch/x86/xen/smp_pv.c
142 +index 8fb8a50a28b4..f2adb63b2d7c 100644
143 +--- a/arch/x86/xen/smp_pv.c
144 ++++ b/arch/x86/xen/smp_pv.c
145 +@@ -93,6 +93,7 @@ asmlinkage __visible void cpu_bringup_and_idle(void)
146 + cpu_bringup();
147 + boot_init_stack_canary();
148 + cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
149 ++ prevent_tail_call_optimization();
150 + }
151 +
152 + void xen_smp_intr_free_pv(unsigned int cpu)
153 +diff --git a/include/linux/compiler.h b/include/linux/compiler.h
154 +index 034b0a644efc..732754d96039 100644
155 +--- a/include/linux/compiler.h
156 ++++ b/include/linux/compiler.h
157 +@@ -356,4 +356,10 @@ static inline void *offset_to_ptr(const int *off)
158 + /* &a[0] degrades to a pointer: a different type from an array */
159 + #define __must_be_array(a) BUILD_BUG_ON_ZERO(__same_type((a), &(a)[0]))
160 +
161 ++/*
162 ++ * This is needed in functions which generate the stack canary, see
163 ++ * arch/x86/kernel/smpboot.c::start_secondary() for an example.
164 ++ */
165 ++#define prevent_tail_call_optimization() asm("")
166 ++
167 + #endif /* __LINUX_COMPILER_H */
168 +--
169 +cgit 1.2-0.3.lf.el7
170 +