runtime: random fatal errors on iOS

[mikioh]

/cc @ianlancetaylor

According to the dashboard, the build started to fail with signal failures with a7cad52.

Before (mostly timeouts): http://build.golang.org/log/a04d0ed55471343ddd7dec8bb8ef9459cc02588e
After (signal failures): http://build.golang.org/log/32a73620035a8c1101d15bcbbdce2cdce640d70f

[mikioh]

We can observe typically:

• fatal error: runtime: stack split at bad time on log/syslog test
• fatal error: unexpected signal during runtime execution [signal 0xb code=0x1 addr=0x0 pc=0x4066590] on net test
• fatal error: runtime: stack split at bad time on net/http test
• fatal error: unexpected signal during runtime execution [signal 0xb code=0x1 addr=0x0 pc=0x4064204] on os/signal test
• fatal error: runtime: stack split at bad time on time test

[ianlancetaylor]

Those errors do look similar to the ones we saw on NetBSD and DragonFly.

@crawshaw Is it possible for you to run this C program on Darwin/ARM, to see how signal stacks are handled for newly created threads?

#include <errno.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <pthread.h>

static void
die (const char *s)
{
  perror (s);
  exit (EXIT_FAILURE);
}

static void
printsigstack (const char *prefix)
{
  stack_t ss;

  if (sigaltstack (NULL, &ss) < 0)
die ("sigaltstack get");
  printf ("%s %p 0x%lx 0x%x\n", prefix, ss.ss_sp, (unsigned long) ss.ss_size,
      (unsigned int) ss.ss_flags);
}

static void *
thread (void *arg)
{
  stack_t ss;

  printsigstack ("thread");
  ss.ss_sp = malloc (SIGSTKSZ);
  if (ss.ss_sp == NULL)
die ("malloc");
  ss.ss_size = SIGSTKSZ;
  ss.ss_flags = 0;
  if (sigaltstack (&ss, NULL) < 0)
die ("sigaltstack set");
  return NULL;
}

int
main ()
{
  pthread_t tid;
  int i;
  stack_t ss;

  printsigstack ("main 1");

  i = pthread_create (&tid, NULL, thread, NULL);
  if (i != 0)
{
  errno = i;
  die ("pthread_create");
}

  i = pthread_join (tid, NULL);
  if (i != 0) {
errno = i;
die ("pthread_join");
  }

  printsigstack ("main 2");

  ss.ss_sp = malloc (SIGSTKSZ);
  if (ss.ss_sp == NULL)
die ("malloc");
  ss.ss_size = SIGSTKSZ;
  ss.ss_flags = 0;
  if (sigaltstack (&ss, NULL) < 0)
die ("sigaltstack set");

  printsigstack ("main 3");

  i = pthread_create (&tid, NULL, thread, NULL);
  if (i != 0)
{
  errno = i;
  die ("pthread_create");
}

  i = pthread_join (tid, NULL);
  if (i != 0) {
errno = i;
die ("pthread_join");
  }

  printsigstack ("main 4");

  return 0;
}

[minux]

I run the program on iOS 5.1.1 and the output is: main 1 0x0 0x0 0x4 thread 0x0 0x0 0x4 main 2 0x0 0x0 0x4 main 3 0x33000 0x20000 0x0 thread 0x0 0x0 0x4 main 4 0x33000 0x20000 0x0

[crawshaw]

For a few (uninteresting) reasons it would take me a couple of hours work to run this. Hopefully the run by @minux does the job.

[ianlancetaylor]

The run by minux tells me that the problem on iOS is not the same as the problem on NetBSD and DragonFly. This will require more investigation.

[gopherbot]

CL https://golang.org/cl/18940 mentions this issue.