1 From 455bd4c430b0c0a361f38e8658a0d6cb469942b5 Mon Sep 17 00:00:00 2001
2 From: Ivan Djelic <ivan.djelic@parrot.com>
3 Date: Wed, 6 Mar 2013 20:09:27 +0100
4 Subject: [PATCH] ARM: 7668/1: fix memset-related crashes caused by recent GCC
7 Recent GCC versions (e.g. GCC-4.7.2) perform optimizations based on
8 assumptions about the implementation of memset and similar functions.
9 The current ARM optimized memset code does not return the value of
10 its first argument, as is usually expected from standard implementations.
12 For instance in the following function:
14 void debug_mutex_lock_common(struct mutex *lock, struct mutex_waiter *waiter)
16 memset(waiter, MUTEX_DEBUG_INIT, sizeof(*waiter));
17 waiter->magic = waiter;
18 INIT_LIST_HEAD(&waiter->list);
23 800554d0 <debug_mutex_lock_common>:
24 800554d0: e92d4008 push {r3, lr}
25 800554d4: e1a00001 mov r0, r1
26 800554d8: e3a02010 mov r2, #16 ; 0x10
27 800554dc: e3a01011 mov r1, #17 ; 0x11
28 800554e0: eb04426e bl 80165ea0 <memset>
29 800554e4: e1a03000 mov r3, r0
30 800554e8: e583000c str r0, [r3, #12]
31 800554ec: e5830000 str r0, [r3]
32 800554f0: e5830004 str r0, [r3, #4]
33 800554f4: e8bd8008 pop {r3, pc}
35 GCC assumes memset returns the value of pointer 'waiter' in register r0; causing
36 register/memory corruptions.
38 This patch fixes the return value of the assembly version of memset.
39 It adds a 'mov' instruction and merges an additional load+store into
40 existing load/store instructions.
41 For ease of review, here is a breakdown of the patch into 4 simple steps:
45 Perform the following substitutions:
48 and insert 'mov ip, r0' as the first statement of the function.
49 At this point, we have a memset() implementation returning the proper result,
50 but corrupting r8 on some paths (the ones that were using ip).
54 Make sure r8 is saved and restored when (! CALGN(1)+0) == 1:
60 and restore r8 on both exit paths:
61 - ldmeqfd sp!, {pc} @ Now <64 bytes to go.
62 + ldmeqfd sp!, {r8, pc} @ Now <64 bytes to go.
65 stmneia ip!, {r1, r3, r8, lr}
71 Make sure r8 is saved and restored when (! CALGN(1)+0) == 0:
74 - stmfd sp!, {r4-r7, lr}
75 + stmfd sp!, {r4-r8, lr}
77 and restore r8 on both exit paths:
79 - ldmeqfd sp!, {r4-r7, pc}
80 + ldmeqfd sp!, {r4-r8, pc}
84 - ldmfd sp!, {r4-r7, lr}
85 + ldmfd sp!, {r4-r8, lr}
89 Rewrite register list "r4-r7, r8" as "r4-r8".
91 Signed-off-by: Ivan Djelic <ivan.djelic@parrot.com>
92 Reviewed-by: Nicolas Pitre <nico@linaro.org>
93 Signed-off-by: Dirk Behme <dirk.behme@gmail.com>
94 Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
96 arch/arm/lib/memset.S | 85 ++++++++++++++++++++++++++-------------------------
97 1 file changed, 44 insertions(+), 41 deletions(-)
99 --- a/arch/arm/lib/memset.S
100 +++ b/arch/arm/lib/memset.S
102 1: subs r2, r2, #4 @ 1 do we have enough
103 blt 5f @ 1 bytes to align with?
105 - strltb r1, [r0], #1 @ 1
106 - strleb r1, [r0], #1 @ 1
107 - strb r1, [r0], #1 @ 1
108 + strltb r1, [ip], #1 @ 1
109 + strleb r1, [ip], #1 @ 1
110 + strb r1, [ip], #1 @ 1
111 add r2, r2, r3 @ 1 (r2 = r2 - (4 - r3))
113 * The pointer is now aligned and the length is adjusted. Try doing the
118 - ands r3, r0, #3 @ 1 unaligned?
120 + * Preserve the contents of r0 for the return value.
123 + ands r3, ip, #3 @ 1 unaligned?
126 - * we know that the pointer in r0 is aligned to a word boundary.
127 + * we know that the pointer in ip is aligned to a word boundary.
129 orr r1, r1, r1, lsl #8
130 orr r1, r1, r1, lsl #16
131 @@ -43,29 +47,28 @@ ENTRY(memset)
135 - * We need an extra register for this loop - save the return address and
137 + * We need 2 extra registers for this loop - use r8 and the LR
141 + stmfd sp!, {r8, lr}
146 - stmgeia r0!, {r1, r3, ip, lr} @ 64 bytes at a time.
147 - stmgeia r0!, {r1, r3, ip, lr}
148 - stmgeia r0!, {r1, r3, ip, lr}
149 - stmgeia r0!, {r1, r3, ip, lr}
150 + stmgeia ip!, {r1, r3, r8, lr} @ 64 bytes at a time.
151 + stmgeia ip!, {r1, r3, r8, lr}
152 + stmgeia ip!, {r1, r3, r8, lr}
153 + stmgeia ip!, {r1, r3, r8, lr}
155 - ldmeqfd sp!, {pc} @ Now <64 bytes to go.
156 + ldmeqfd sp!, {r8, pc} @ Now <64 bytes to go.
158 * No need to correct the count; we're only testing bits from now on
161 - stmneia r0!, {r1, r3, ip, lr}
162 - stmneia r0!, {r1, r3, ip, lr}
163 + stmneia ip!, {r1, r3, r8, lr}
164 + stmneia ip!, {r1, r3, r8, lr}
166 - stmneia r0!, {r1, r3, ip, lr}
168 + stmneia ip!, {r1, r3, r8, lr}
169 + ldmfd sp!, {r8, lr}
173 @@ -74,54 +77,54 @@ ENTRY(memset)
174 * whole cache lines at once.
177 - stmfd sp!, {r4-r7, lr}
178 + stmfd sp!, {r4-r8, lr}
195 - movs ip, ip, lsl #(32 - 4)
196 - stmcsia r0!, {r4, r5, r6, r7}
197 - stmmiia r0!, {r4, r5}
204 + movs r8, r8, lsl #(32 - 4)
205 + stmcsia ip!, {r4, r5, r6, r7}
206 + stmmiia ip!, {r4, r5}
212 - stmgeia r0!, {r1, r3-r7, ip, lr}
213 - stmgeia r0!, {r1, r3-r7, ip, lr}
214 + stmgeia ip!, {r1, r3-r8, lr}
215 + stmgeia ip!, {r1, r3-r8, lr}
217 - ldmeqfd sp!, {r4-r7, pc}
218 + ldmeqfd sp!, {r4-r8, pc}
221 - stmneia r0!, {r1, r3-r7, ip, lr}
222 + stmneia ip!, {r1, r3-r8, lr}
224 - stmneia r0!, {r4-r7}
225 - ldmfd sp!, {r4-r7, lr}
226 + stmneia ip!, {r4-r7}
227 + ldmfd sp!, {r4-r8, lr}
232 - stmneia r0!, {r1, r3}
233 + stmneia ip!, {r1, r3}
238 * When we get here, we've got less than 4 bytes to zero. We
239 * may have an unaligned pointer as well.
242 - strneb r1, [r0], #1
243 - strneb r1, [r0], #1
244 + strneb r1, [ip], #1
245 + strneb r1, [ip], #1
247 - strneb r1, [r0], #1
248 + strneb r1, [ip], #1