The included program dies with a SEGV under 2.6.11.2 and 2.6.12-rc4.
It doesn't die under 2.4.25. I compiled the kernels myself. The
distribution is Fedora Core release 3, with glibc 2.3.5.
I don't have any earlier 2.6 kernels to test it on, so I have no idea
when this bug first appeared.
I only skim LKML through fa.linux.kernel, so please CC me on any reply
that I should see.
--Cliff Matthews <[email protected]>
+1 505 363 5754 Cell
#include <stdio.h>
#include <unistd.h>
#include <sys/mman.h>
#include <stdbool.h>
#include <assert.h>
#include <setjmp.h>
#include <signal.h>
static jmp_buf segv_return;
static void
segv_handler (int signum_ignored __attribute__((unused)))
{
longjmp (segv_return, 1);
}
int
main (void)
{
volatile char *volatile addr;
volatile int n_failures;
addr = (void *) 0x10000L;
n_failures = 0;
signal (SIGSEGV, segv_handler);
if (setjmp (segv_return) != 0)
++n_failures;
else
*addr;
printf ("n_failures = %d\n", n_failures);
if (setjmp (segv_return) != 0)
++n_failures;
else
*addr;
printf ("n_failures = %d\n", n_failures);
return 0;
}
* Clifford T. Matthews ([email protected]) wrote:
> The included program dies with a SEGV under 2.6.11.2 and 2.6.12-rc4.
> It doesn't die under 2.4.25. I compiled the kernels myself. The
> distribution is Fedora Core release 3, with glibc 2.3.5.
2.6 has been fixed... So your program (which happens to be slightly
buggy) no longer works as you expected. See below.
> #include <stdio.h>
> #include <unistd.h>
> #include <sys/mman.h>
> #include <stdbool.h>
> #include <assert.h>
> #include <setjmp.h>
> #include <signal.h>
>
> static jmp_buf segv_return;
>
> static void
> segv_handler (int signum_ignored __attribute__((unused)))
> {
> longjmp (segv_return, 1);
siglongjmp
> }
>
> int
> main (void)
> {
> volatile char *volatile addr;
> volatile int n_failures;
>
> addr = (void *) 0x10000L;
> n_failures = 0;
>
> signal (SIGSEGV, segv_handler);
> if (setjmp (segv_return) != 0)
sigsetjmp
> ++n_failures;
> else
> *addr;
>
> printf ("n_failures = %d\n", n_failures);
>
> if (setjmp (segv_return) != 0)
sigsetjmp
> ++n_failures;
> else
> *addr;
>
> printf ("n_failures = %d\n", n_failures);
>
> return 0;
> }
To jump out of signal handlers, you now need sigsetjmp() and
siglongjmp().
On Tue, 24 May 2005, Clifford T. Matthews wrote:
> The included program dies with a SEGV under 2.6.11.2 and 2.6.12-rc4.
> It doesn't die under 2.4.25. I compiled the kernels myself. The
> distribution is Fedora Core release 3, with glibc 2.3.5.
>
> I don't have any earlier 2.6 kernels to test it on, so I have no idea
> when this bug first appeared.
>
> I only skim LKML through fa.linux.kernel, so please CC me on any reply
> that I should see.
>
> --Cliff Matthews <[email protected]>
> +1 505 363 5754 Cell
>
> #include <stdio.h>
> #include <unistd.h>
> #include <sys/mman.h>
> #include <stdbool.h>
> #include <assert.h>
> #include <setjmp.h>
> #include <signal.h>
>
> static jmp_buf segv_return;
>
> static void
> segv_handler (int signum_ignored __attribute__((unused)))
> {
> longjmp (segv_return, 1);
> }
>
> int
> main (void)
> {
> volatile char *volatile addr;
> volatile int n_failures;
>
> addr = (void *) 0x10000L;
> n_failures = 0;
>
> signal (SIGSEGV, segv_handler);
> if (setjmp (segv_return) != 0)
> ++n_failures;
> else
> *addr;
>
> printf ("n_failures = %d\n", n_failures);
>
> if (setjmp (segv_return) != 0)
> ++n_failures;
> else
> *addr;
>
> printf ("n_failures = %d\n", n_failures);
>
> return 0;
> }
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
Cheers,
Dick Johnson
Penguin : Linux version 2.6.11.9 on an i686 machine (5537.79 BogoMips).
Notice : All mail here is now cached for review by Dictator Bush.
98.36% of all statistics are fiction.
>>>>> "Chris" == Chris Wright <[email protected]> writes:
Chris> 2.6 has been fixed... So your program (which happens to be
Chris> slightly buggy) no longer works as you expected. See
Chris> below.
Thanks for the quick response. Using sigsetjmp and siglongjmp makes
the program print two lines.
I read the setjmp / sigsetjmp documentation and misunderstood it.
I had already seen that if I inserted "signal (SIGSEGV, segv_handler)"
before the second setjmp (not sigsetjmp), the program (under 2.6
kernels) still would die.
I guess what happens there is that after coming back from the longjmp,
the error handler is still segv_handler, but the receipt of the SEGV
signal itself is blocked and if you take a SEGV when the receipt of
SEGV is blocked a program dies with a SEGV, even if you have a SEGV
handler.
--Cliff Matthews <[email protected]>