[Bug target/50818] va_list is filled incorrectly in functions with ms_abi attribute on amd64
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=50818
Steven Shi changed:
What|Removed |Added
CC||steven.shi at intel dot com
--- Comment #5 from Steven Shi ---
Hi Bizjak,
Your Executable testcase cannot really work with -mabi=ms on the command line.
If add a printf() in the while loop to output the sum value as below example,
the testcase will fail again with Segmentation fault even building with with
-mabi=ms.
$ gcc -O2 -mabi=ms pr50818.c
$ ./a.out
Segmentation fault (core dumped)
#include
int
__attribute__((ms_abi))
foo (int n, ...)
{
__builtin_va_list ap;
int sum = 0;
__builtin_va_start (ap, n);
while (n--) {
sum += __builtin_va_arg (ap, int);
printf("sum = %d\n", sum);
}
__builtin_va_end (ap);
return sum;
}
int main ()
{
int res = foo (10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10);
if (res != 55)
__builtin_abort ();
return 0;
}
[Bug target/50818] va_list is filled incorrectly in functions with ms_abi attribute on amd64
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=50818 --- Comment #6 from Steven Shi --- BTW, latest LLVM/Clang 3.8 release has fixed this va_list ms_abi issue, please see below email thread. Does GCC really cannot fix it as LLVM/Clang? This issue really impact Uefi firmware building with GCC. http://lists.llvm.org/pipermail/llvm-dev/2016-January/093778.html
[Bug target/50818] va_list is filled incorrectly in functions with ms_abi attribute on amd64
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=50818
--- Comment #9 from Steven Shi ---
GCC6.10 lto fails to support below MS ABI buildin of va list, but GCC5 works.
This is a regression bug for GCC6. I suspect current GCC trunk also has this
regression bug.
__builtin_ms_va_list ap;
__builtin_ms_va_start (ap, n);
__builtin_ms_va_end (ap);
Below is my testcase:
$cat test2.c
#include
int
__attribute__((ms_abi))
foo (int n, ...)
{
__builtin_ms_va_list ap;
int sum = 0;
__builtin_ms_va_start (ap, n);
while (n--) {
sum += __builtin_va_arg (ap, int);
printf("sum = %d\n", sum);
}
__builtin_ms_va_end (ap);
return sum;
}
int main ()
{
int res = foo (10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10);
if (res != 55)
__builtin_abort ();
return 0;
}
$ gcc --version
gcc (GCC) 6.1.0
$ gcc -flto -Os test2.c
$ ./a.out
Segmentation fault (core dumped)
$ gcc-5 -flto -Os test2.c
$ ./a.out
sum = 1
sum = 3
sum = 6
sum = 10
sum = 15
sum = 21
sum = 28
sum = 36
sum = 45
sum = 55
[Bug c/77341] New: GCC6 va_list MS ABI buildin regression bug
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77341
Bug ID: 77341
Summary: GCC6 va_list MS ABI buildin regression bug
Product: gcc
Version: 6.1.0
Status: UNCONFIRMED
Severity: critical
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: steven.shi at intel dot com
Target Milestone: ---
GCC6.10 lto fails to support below MS ABI buildin of va list, but GCC5 works.
This is a regression bug for GCC6. I suspect current GCC trunk also has this
regression bug.
__builtin_ms_va_list ap;
__builtin_ms_va_start (ap, n);
__builtin_ms_va_end (ap);
Below is my testcase:
$cat test2.c
#include
int
__attribute__((ms_abi))
foo (int n, ...)
{
__builtin_ms_va_list ap;
int sum = 0;
__builtin_ms_va_start (ap, n);
while (n--) {
sum += __builtin_va_arg (ap, int);
printf("sum = %d\n", sum);
}
__builtin_ms_va_end (ap);
return sum;
}
int main ()
{
int res = foo (10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10);
if (res != 55)
__builtin_abort ();
return 0;
}
$ gcc --version
gcc (GCC) 6.1.0
$ gcc -flto -Os test2.c
$ ./a.out
Segmentation fault (core dumped)
$ gcc-5 -flto -Os test2.c
$ ./a.out
sum = 1
sum = 3
sum = 6
sum = 10
sum = 15
sum = 21
sum = 28
sum = 36
sum = 45
sum = 55
BTW, I've raised this issue in below bug comments:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=50818
[Bug lto/70955] [6/7 Regression] Wrong code generation for __builtin_ms_va_list with -flto
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70955 --- Comment #16 from Steven Shi --- How to checkout the 6.3 code from GCC trunk? I hope to verify this fix on Uefi firmware GCC LTO build.
[Bug lto/70955] [6/7 Regression] Wrong code generation for __builtin_ms_va_list with -flto
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70955
--- Comment #19 from Steven Shi ---
(In reply to vries from comment #18)
> Created attachment 39484 [details]
> Tentative patch using attributes ms_abi/sysv_abi
>
> not build or regression tested yet, but at least makes test-case pass.
Vries,
I tried apply your patch on latest gcc trunk, build and install the new gcc7,
it still cannot work. Below is my test case.
$ gcc --version
gcc (GCC) 7.0.0 20160823 (experimental)
$ cat test2.c
#include
int
__attribute__((ms_abi))
foo (int n, ...)
{
__builtin_ms_va_list ap;
int sum = 0;
__builtin_ms_va_start (ap, n);
while (n--) {
sum += __builtin_va_arg (ap, int);
printf("sum = %d\n", sum);
}
__builtin_ms_va_end (ap);
return sum;
}
int main ()
{
int res = foo (10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10);
if (res != 55)
__builtin_abort ();
return 0;
}
$ gcc -flto -Os test2.c
$ ./a.out
Segmentation fault (core dumped)
But GCC5 works.
$ gcc-5 -flto -Os test2.c
$ ./a.out
sum = 1
sum = 3
sum = 6
sum = 10
sum = 15
sum = 21
sum = 28
sum = 36
sum = 45
sum = 55
[Bug lto/70955] [6/7 Regression] Wrong code generation for __builtin_ms_va_list with -flto
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70955 --- Comment #21 from Steven Shi --- Vries, Sorry, my bad. I didn't apply your patch completely. Yes, your patch works. And with your patch, my UEFU firmware gcc lto build pass and boot successfully. Hope we could check in a formal fix ASAP, because the lto build is Uefi firmware GCC default build now. This regression bug block us to adopt new version GCC on Uefi firmware. Thanks!
