Fix i686 memchr for large input sizes
Commit Message
Similar to BZ#19387 and BZ#20971, both i686 memchr optimized assembly
implementations (memchr-sse2-bsf and memchr-sse2) do not handle the
size overflow correctly.
It is shown by the new tests added by commit 3daef2c8ee4df29, where
both implementation fails with size as SIZE_MAX.
This patch uses a similar strategy used on 3daef2c8ee4df2, where
saturared math is used for overflow case.
Checked on i686-linux-gnu.
[BZ #21014]
* sysdeps/i386/i686/multiarch/memchr-sse2-bsf.S (MEMCHR): Avoid overflow
in pointer addition.
* sysdeps/i386/i686/multiarch/memchr-sse2.S (MEMCHR): Likewise.
---
ChangeLog | 7 +++++++
sysdeps/i386/i686/multiarch/memchr-sse2-bsf.S | 10 ++++++++--
sysdeps/i386/i686/multiarch/memchr-sse2.S | 8 +++++++-
3 files changed, 22 insertions(+), 3 deletions(-)
Comments
On Monday 02 January 2017 07:56 PM, Adhemerval Zanella wrote:
> Similar to BZ#19387 and BZ#20971, both i686 memchr optimized assembly
> implementations (memchr-sse2-bsf and memchr-sse2) do not handle the
> size overflow correctly.
>
> It is shown by the new tests added by commit 3daef2c8ee4df29, where
> both implementation fails with size as SIZE_MAX.
>
> This patch uses a similar strategy used on 3daef2c8ee4df2, where
> saturared math is used for overflow case.
>
> Checked on i686-linux-gnu.
>
> [BZ #21014]
> * sysdeps/i386/i686/multiarch/memchr-sse2-bsf.S (MEMCHR): Avoid overflow
> in pointer addition.
> * sysdeps/i386/i686/multiarch/memchr-sse2.S (MEMCHR): Likewise.
Looks good to me.
Siddhesh
@@ -149,9 +149,15 @@ L(crosscache):
.p2align 4
L(unaligned_no_match):
# ifndef USE_AS_RAWMEMCHR
- sub $16, %edx
+ /* Calculate the last acceptable address and check for possible
+ addition overflow by using satured math:
+ edx = ecx + edx
+ edx |= -(edx < ecx) */
add %ecx, %edx
- jle L(return_null)
+ sbb %eax, %eax
+ or %eax, %edx
+ sub $16, %edx
+ jbe L(return_null)
add $16, %edi
# else
add $16, %edx
@@ -118,8 +118,14 @@ L(crosscache):
# ifndef USE_AS_RAWMEMCHR
jnz L(match_case2_prolog1)
lea -16(%edx), %edx
+ /* Calculate the last acceptable address and check for possible
+ addition overflow by using satured math:
+ edx = ecx + edx
+ edx |= -(edx < ecx) */
add %ecx, %edx
- jle L(return_null)
+ sbb %eax, %eax
+ or %eax, %edx
+ jbe L(return_null)
lea 16(%edi), %edi
# else
jnz L(match_case1_prolog1)