[Bug sanitizer/61422] False Asan positive in libopus
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61422 Marat Zakirov m.zakirov at samsung dot com changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED Resolution|--- |FIXED --- Comment #8 from Marat Zakirov m.zakirov at samsung dot com --- Already fixed by Gribov commit.
[Bug sanitizer/61422] False Asan positive in libopus
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61422 --- Comment #1 from Marat Zakirov m.zakirov at samsung dot com --- Created attachment 32896 -- https://gcc.gnu.org/bugzilla/attachment.cgi?id=32896action=edit Proposed patch Only tested with asan testsuite on x64.
[Bug sanitizer/61422] False Asan positive in libopus
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61422 --- Comment #2 from Jakub Jelinek jakub at gcc dot gnu.org --- Comment on attachment 32896 -- https://gcc.gnu.org/bugzilla/attachment.cgi?id=32896 Proposed patch Can't reproduce it, are you sure you don't have extra patches in the tree (stringops currently don't use the the real_size_in_bytes != size_in_bytes accesses). The fix is good, except for the bogus whitespace after [ and incorrect testcase (you should be really using +m instead of =m, otherwise the compiler could optimize away the stores of 1 or 3. Also, please don't add * sizeof (char) there, that is * 1 by definition. And please use __builtin_memcpy instead of unprototyped memcpy. And copying around uninitialized bytes probably isn't a good idea either, so you should first memset it or fill in or something.
[Bug sanitizer/61422] False Asan positive in libopus
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61422 --- Comment #3 from Marat Zakirov m.zakirov at samsung dot com --- I fix it.
[Bug sanitizer/61422] False Asan positive in libopus
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61422
--- Comment #4 from Jakub Jelinek jakub at gcc dot gnu.org ---
If you want a testcase that crashes even for me with current trunk, then e.g.:
struct S { char c[16]; } __attribute__((packed));
__attribute__((noinline, noclone)) struct S
foo (struct S *p)
{
asm volatile ( : : r (p) : memory);
return *p;
}
int
main ()
{
struct S a __attribute__((aligned (16)));
struct S b __attribute__((aligned (16)));
__builtin_memset (b, 0, sizeof b);
a = foo (b);
asm volatile ( : : m (a), m (b));
return 0;
}
[Bug sanitizer/61422] False Asan positive in libopus
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61422
--- Comment #5 from Marat Zakirov m.zakirov at samsung dot com ---
Thank you for your quick response Jacub. Actually I take this issue from
existing ffmpeg source so the test is just a truncated version.
Following fail in my 4.10 without discovered fix:
cat test.c
#define N 4
#define M 16
int main ()
{
int i = 1, j = 3;
char ind[N][M];
__builtin_memset( ind, 0, M * N);
__asm (\n
: +m(i), +m(j));
__builtin_memcpy (ind[j], ind[i], M * 1);
return 0;
}
gcc test.c -fsanitize=address -static-libasan
./a.out
[Bug sanitizer/61422] False Asan positive in libopus
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61422 --- Comment #6 from Marat Zakirov m.zakirov at samsung dot com --- Created attachment 32898 -- https://gcc.gnu.org/bugzilla/attachment.cgi?id=32898action=edit Proposed patch Try this. It is mostly the same. No additional patches is needed. I hope it's reproducible.
[Bug sanitizer/61422] False Asan positive in libopus
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61422 --- Comment #7 from Jakub Jelinek jakub at gcc dot gnu.org --- You still have: tree shadow_ptr_type = shadow_ptr_types[ real_size_in_bytes == 16 ? 1 : 0]; instead of: tree shadow_ptr_type = shadow_ptr_types[real_size_in_bytes == 16 ? 1 : 0]; And, I can't really reproduce using your testcase, so either you have local patches other than this (I think Yuri Gribov has been working on something in this area), or it really would be better to use a testcase with explicit 16 byte unaligned access (or both). Still no point to use * 1, or \n in the asm pattern ( is good enough). That said, patches should be posted to gcc-patches mailing list.
