Re: Strange (or at least interesting) printf() behavior!

[Paul Goyette <paul%whooppee.com@localhost>]

On Mon, 3 Jan 2011, Joerg Sonnenberger wrote:

> On Mon, Jan 03, 2011 at 08:01:36AM -0800, Paul Goyette wrote:
> > While researching another problem, I created the following test program:
> > static sigjmp_buf sigfpe_flt_env;
> > static void
> > sigfpe_flt_action(int signo, siginfo_t *info, void *ptr)
> > {
> >
> >         flt_signal++;

I changed that to

        if (flt_signal++ != 0)
                abort();

> >         siglongjmp(sigfpe_flt_env, 1);
> > }
>
> I would suggest to add an abort here if flt_signal is already set.
> That would pin point the issue.
>
> > int main(int argc, void *argv[])
> > {
> >         struct sigaction sa;
> >         double d;
> >
> > > > > >   printf("Start\n");
>
> va_arg users have at least some minimal side effects like saving all fpu
> registers.
>
> >         if (sigsetjmp(sigfpe_flt_env, 0) == 0) {
> >                 sa.sa_flags = SA_SIGINFO;
> >                 sa.sa_sigaction = sigfpe_flt_action;
> >                 sigemptyset(&sa.sa_mask);
> >                 sigaction(SIGFPE, &sa, NULL);
> >                 fpsetmask(FP_X_INV|FP_X_DZ|FP_X_OFL|FP_X_UFL|FP_X_IMP);
> >                 d = 1.0 / strtod("0", NULL);
> >         }

And I added here (which is where the siglongjmp() should go)

                fpsetmask(0);

(I wold have placed this right in the signal handler, but fpsetmask() is 
not listed as signalsafe in the signal(7) man page!)


> >         printf("FPE signal handler invoked %d times.\n", flt_signal);
> > }
> >
> > As listed, the program runs find and produces the expected results.
> >
> > However, if I remove the printf() pointed to by the >>>> the program
> > actually hangs!  Running it under gdb and interrupting with
> > Control/C shows the following traceback:
>
> Not without a lot more guessing, but the above should tell you where
> exactly the second trap comes from. Note that the test case is likely a
> lot more stable if you actually restore the mask before calling anything
> else.


The abort() does not trigger, so the handler is being called only once.

And even though the exception mask is being reset, the program still 
hangs.


-------------------------------------------------------------------------
| Paul Goyette     | PGP Key fingerprint:     | E-mail addresses:       |
| Customer Service | FA29 0E3B 35AF E8AE 6651 | paul at whooppee.com    |
| Network Engineer | 0786 F758 55DE 53BA 7731 | pgoyette at juniper.net |
| Kernel Developer |                          | pgoyette at netbsd.org  |
-------------------------------------------------------------------------