Gary V. Vaughan wrote:
> hppa2.0w-hp-hpux11.11-hpc1111 m4 tests pass, gnulib fails: test-c-stack.sh
> ./test-c-stack.sh[7]: 1752 Memory fault(coredump)
> FAIL: test-c-stack.sh
I reproduce this on HP-UX 11.11, without libsigsegv.
$ ./test-c-stack
./test-c-stack: program error
Segmentation fault (core dumped)
The situation on HP-UX 11.11 is that HAVE_XSI_STACK_OVERFLOW_HEURISTIC is not
defined. This is correct: there are two problems with the system here:
1) info->si_code is zero, not a positive number as expected by Paul's code.
2) the mcontext's uc_stack.ss_size field is unusable.
Breakpoint 1, segv_handler (signo=11, info=0x400017a8, context=0x400012d0)
at c-stack.c:244
244 if (0 < info->si_code)
(gdb) print *info
$1 = {si_signo = 11, si_code = 0, si_errno = 0, si_value = {__svi_int_s = {
__svi_int = 0}, __sival_ptr = 0x0}, __data = {__proc = {
__pid = 2047868944, __pdata = {__kill = {__uid = 0}, __SIGCLD = {
__status = 0}}}, __fault = {__addr = 0x7a100010}, __file = {
__fd = 2047868944, __band = 0}, __gfault = {__addr = 0x7a100010,
__lockpage = 0x0, __dev_id = 0}}, __pad = {0, 0, 0, 0, 0, 0, 0, 0, 0}}
Stack grows, previous stack values are ...0x7a0ffe6c...0x7a0fff6c
Stack apparently ends at 0x7a100000
Failure address 0x7a100010 is plausible.
si_code is 0 - unusual but acceptable.
But uc_stack = {ss_sp = 0x7a001100, ss_flags = 0, ss_size = 4352}
which is unusable.
I'm committing the change below. With it, the test passes:
PASS: test-c-stack.sh
cannot tell stack overflow from crash; consider installing libsigsegv
SKIP: test-c-stack2.sh
2009-01-24 Bruno Haible <br...@clisp.org>
* lib/c-stack.c (segv_handler): If !HAVE_XSI_STACK_OVERFLOW_HEURISTIC,
set signo = 0 also if info->si_code <= 0. Needed on HP-UX 11.11.
Reported by Gary V. Vaughan <g...@gnu.org>.
*** lib/c-stack.c.orig 2009-01-24 14:02:12.000000000 +0100
--- lib/c-stack.c 2009-01-24 13:54:39.000000000 +0100
***************
*** 239,259 ****
void *context __attribute__ ((unused)))
{
/* Clear SIGNO if it seems to have been a stack overflow. */
- if (0 < info->si_code)
- {
# if ! HAVE_XSI_STACK_OVERFLOW_HEURISTIC
! /* We can't easily determine whether it is a stack overflow; so
! assume that the rest of our program is perfect (!) and that
! this segmentation violation is a stack overflow.
!
! Note that although both Linux and Solaris provide
! sigaltstack, SA_ONSTACK, and SA_SIGINFO, currently only
! Solaris satisfies the XSI heueristic. This is because
! Solaris populates uc_stack with the details of the
! interrupted stack, while Linux populates it with the details
! of the current stack. */
! signo = 0;
# else
/* If the faulting address is within the stack, or within one
page of the stack end, assume that it is a stack
overflow. */
--- 239,259 ----
void *context __attribute__ ((unused)))
{
/* Clear SIGNO if it seems to have been a stack overflow. */
# if ! HAVE_XSI_STACK_OVERFLOW_HEURISTIC
! /* We can't easily determine whether it is a stack overflow; so
! assume that the rest of our program is perfect (!) and that
! this segmentation violation is a stack overflow.
!
! Note that although both Linux and Solaris provide
! sigaltstack, SA_ONSTACK, and SA_SIGINFO, currently only
! Solaris satisfies the XSI heueristic. This is because
! Solaris populates uc_stack with the details of the
! interrupted stack, while Linux populates it with the details
! of the current stack. */
! signo = 0;
# else
+ if (0 < info->si_code)
+ {
/* If the faulting address is within the stack, or within one
page of the stack end, assume that it is a stack
overflow. */
***************
*** 278,285 ****
write (STDERR_FILENO, buf, strlen (buf));
}
# endif
- # endif
}
die (signo);
}
--- 278,285 ----
write (STDERR_FILENO, buf, strlen (buf));
}
# endif
}
+ # endif
die (signo);
}
