PR remote/19496, interrupted syscall in forking-threads-plus-bkpt
authorDon Breazeal <donb@codesourcery.com>
Wed, 16 Mar 2016 22:13:44 +0000 (15:13 -0700)
committerDon Breazeal <donb@codesourcery.com>
Wed, 16 Mar 2016 22:13:44 +0000 (15:13 -0700)
commit7868401b7b63d851eea5de3b780591238d579bd1
tree74c2984e7c7b1ccf620c1365b06ce6bb010d44f8
parentbfeeb14b8468ab3f60d719f2cde697d1b537eba8
PR remote/19496, interrupted syscall in forking-threads-plus-bkpt

This patch addresses "fork:Interrupted system call" (or wait:) failures
in gdb.threads/forking-threads-plus-breakpoint.exp.

The test program spawns ten threads, each of which do ten fork/waitpid
sequences.  The cause of the problem was that when one of the fork
children exited before the corresponding fork parent could initiate its
waitpid for that child, a SIGCHLD and/or SIGSTOP was delivered and
interrupted a fork or waitpid in another thread.

The fix was to wrap the system calls in a loop to retry the call if
it was interrupted, like:

do
  {
    pid = fork ();
  }
while (pid == -1 && errno == EINTR);

Since this is a Linux-only test I figure it is OK to use errno and EINTR.
I tried a number of alternative fixes using SIG_IGN, SA_RESTART,
pthread_sigblock, and bsd_signal, but none of these worked as well.

Tested on Nios II Linux target with x86 Linux host.

gdb/testsuite/ChangeLog:
2016-03-16  Don Breazeal  <donb@codesourcery.com>

* gdb.threads/forking-threads-plus-breakpoint.c (thread_forks):
Retry fork and waitpid on interrupted system call errors.
* gdb.threads/forking-threads-plus-breakpoint.exp: (do_test):
Use with_timeout_factor to increase timeout to 90.
gdb/testsuite/ChangeLog
gdb/testsuite/gdb.threads/forking-threads-plus-breakpoint.c
gdb/testsuite/gdb.threads/forking-threads-plus-breakpoint.exp
This page took 0.035121 seconds and 4 git commands to generate.