On 3/20/25 15:03, Alex Coplan wrote:
External email: Use caution opening links or attachments
On 20/03/2025 14:31, Alfie Richards wrote:
Hello,
This is a cherry pick of 20385cb92cbd4a1934661ab97a162c1e25935836 which didn't
apply cleanly
so needed a very minor edit to for it to apply for GCC 12 and 13.
Regtested and bootstrapped for gcc-13 on aarch64-linux-gnu.
This applies cleanly to gcc-12 but there is a regression for
`c-c++-common/hwasan/large-aligned-1.c`.
The output of execution was
```
==2242916==ERROR: HWAddressSanitizer: tag-mismatch on address 0xfffffffff680 at
pc 0xfffff780e08c
READ of size 4 at 0xfffffffff680 tags: 02/01(00) (ptr/mem) in thread T0
#0 0xfffff780e08c in SigTrap<2>
../../../../src/libsanitizer/hwasan/hwasan_checks.h:28
```
and has changed to
```
==2242927==ERROR: HWAddressSanitizer: tag-mismatch on address 0xfffffffff690 at
pc 0xfffff780e08c
READ of size 4 at 0xfffffffff690 tags: 02/00 (ptr/mem) in thread T0
#0 0xfffff780e08c in SigTrap<2>
../../../../src/libsanitizer/hwasan/hwasan_checks.h:28
```
Note the difference in the tags.
Does anyone with experience of HWAddressSanitizer know what caused this and if
it's an issue?
Yeah, I ran into this too in the past when testing GCC 12 backports.
I think those hwasan tests are flaky due to the dg-output checks being
overly strict. Martin Liska pushed a fix to GCC 13 to relax them
(r13-100-g3771486daa1e904ceae6f3e135b28e58af33849). I think that patch
needs backporting to the GCC 12 branch, which I requested here:
https://gcc.gnu.org/pipermail/gcc-patches/2024-February/645278.html but
that backport request was never ACKed. I still think we should backport
it.
Just for context: From having a look at the output on my local machine
it seems that for some reason the stack layout has been changed after
this patch so that the order of the naturally aligned and artificially
aligned objects are different. That has meant that the access past the
end of the naturally aligned object hits something different.
I'm honestly a bit baffled about why this patch would change the stack
layout.
So agree this is a flaky test, and not demonstrating a problem with the
relevant patch. I'm still really surprised that it happened.
Alex
Kind regards,
Alfie Richards
Andrew Carlotti (1):
aarch64: Use PAUTH instead of V8_3A in some places
gcc/config/aarch64/aarch64.cc | 6 +++---
gcc/config/aarch64/aarch64.md | 8 ++++----
2 files changed, 7 insertions(+), 7 deletions(-)
--
2.34.1
