[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH 03/10] pthread_sigmask: new module
From: |
Bruno Haible |
Subject: |
Re: [PATCH 03/10] pthread_sigmask: new module |
Date: |
Sat, 9 Jul 2011 01:40:25 +0200 |
User-agent: |
KMail/1.9.9 |
Another platform on which the pthread_sigmask unit test fails is IRIX 6.5:
test-pthread_sigmask1.c:78: assertion failed
FAIL: test-pthread_sigmask1
The cause: After pthread_sigmask returns, it needs a small usleep(1) before a
pending signal is delivered. I'm committing this workaround:
2011-07-08 Bruno Haible <address@hidden>
pthread_sigmask: Work around IRIX bug.
* m4/pthread_sigmask.m4 (gl_FUNC_PTHREAD_SIGMASK): Test for the IRIX
bug.
* lib/pthread_sigmask.c (pthread_sigmask): usleep for some time when
there may be unblocked pending signals.
* doc/posix-functions/pthread_sigmask.texi: Mention the IRIX bug.
--- doc/posix-functions/pthread_sigmask.texi.orig Sat Jul 9 01:37:09 2011
+++ doc/posix-functions/pthread_sigmask.texi Sat Jul 9 01:24:02 2011
@@ -23,6 +23,10 @@
When it fails, this functions returns -1 instead of the error number on
some platforms:
Cygwin 1.7.5.
address@hidden
+This function does not immediately raise signals that were pending before
+the call and unblocked by the call on some platforms:
+IRIX 6.5.
@end itemize
Portability problems not fixed by Gnulib:
--- lib/pthread_sigmask.c.orig Sat Jul 9 01:37:09 2011
+++ lib/pthread_sigmask.c Sat Jul 9 01:22:58 2011
@@ -47,6 +47,16 @@
if (ret == -1)
return errno;
# endif
+# if PTHREAD_SIGMASK_UNBLOCK_BUG
+ if (ret == 0
+ && new_mask != NULL
+ && (how == SIG_UNBLOCK || how == SIG_SETMASK))
+ {
+ /* Give the OS the opportunity to raise signals that were pending before
+ the pthread_sigmask call and have now been unblocked. */
+ usleep (1);
+ }
+# endif
return ret;
#else
int ret = sigprocmask (how, new_mask, old_mask);
--- m4/pthread_sigmask.m4.orig Sat Jul 9 01:37:09 2011
+++ m4/pthread_sigmask.m4 Sat Jul 9 01:36:10 2011
@@ -1,4 +1,4 @@
-# pthread_sigmask.m4 serial 8
+# pthread_sigmask.m4 serial 9
dnl Copyright (C) 2011 Free Software Foundation, Inc.
dnl This file is free software; the Free Software Foundation
dnl gives unlimited permission to copy and/or distribute it,
@@ -115,7 +115,7 @@
*no)
REPLACE_PTHREAD_SIGMASK=1
AC_DEFINE([PTHREAD_SIGMASK_INEFFECTIVE], [1],
- [Define to 1 if pthread_mask() may returns 0 and have no effect.])
+ [Define to 1 if pthread_sigmask() may returns 0 and have no
effect.])
;;
esac
fi
@@ -155,10 +155,74 @@
*no)
REPLACE_PTHREAD_SIGMASK=1
AC_DEFINE([PTHREAD_SIGMASK_FAILS_WITH_ERRNO], [1],
- [Define to 1 if pthread_mask(), when it fails, returns -1 and sets
errno.])
+ [Define to 1 if pthread_sigmask(), when it fails, returns -1 and
sets errno.])
;;
esac
+ dnl On IRIX 6.5, in a single-threaded program, pending signals are not
+ dnl immediately delivered when they are unblocked through pthread_sigmask,
+ dnl only a little while later.
+ AC_CACHE_CHECK([whether pthread_sigmask unblocks signals correctly],
+ [gl_cv_func_pthread_sigmask_unblock_works],
+ [
+ case "$host_os" in
+ irix*)
+ gl_cv_func_pthread_sigmask_unblock_works="guessing no";;
+ *)
+ gl_cv_func_pthread_sigmask_unblock_works="guessing yes";;
+ esac
+ dnl Here we link against $LIBMULTITHREAD, not only
$LIB_PTHREAD_SIGMASK,
+ dnl otherwise we get a false positive on those platforms where
+ dnl $gl_cv_func_pthread_sigmask_in_libc_works is "no".
+ gl_save_LIBS="$LIBS"
+ LIBS="$LIBS $LIBMULTITHREAD"
+ AC_RUN_IFELSE(
+ [AC_LANG_SOURCE([[
+#include <pthread.h>
+#include <signal.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+static volatile int sigint_occurred;
+static void
+sigint_handler (int sig)
+{
+ sigint_occurred++;
+}
+int main ()
+{
+ sigset_t set;
+ int pid = getpid ();
+ char command[80];
+ signal (SIGINT, sigint_handler);
+ sigemptyset (&set);
+ sigaddset (&set, SIGINT);
+ if (!(pthread_sigmask (SIG_BLOCK, &set, NULL) == 0))
+ return 1;
+ sprintf (command, "sh -c 'sleep 1; kill -%d %d' &", SIGINT, pid);
+ if (!(system (command) == 0))
+ return 2;
+ sleep (2);
+ if (!(sigint_occurred == 0))
+ return 3;
+ if (!(pthread_sigmask (SIG_UNBLOCK, &set, NULL) == 0))
+ return 4;
+ if (!(sigint_occurred == 1)) /* This fails on IRIX. */
+ return 5;
+ return 0;
+}]])],
+ [:],
+ [gl_cv_func_pthread_sigmask_unblock_works=no],
+ [:])
+ LIBS="$gl_save_LIBS"
+ ])
+ case "$gl_cv_func_pthread_sigmask_unblock_works" in
+ *no)
+ REPLACE_PTHREAD_SIGMASK=1
+ AC_DEFINE([PTHREAD_SIGMASK_UNBLOCK_BUG], [1],
+ [Define to 1 if pthread_sigmask() unblocks signals incorrectly.])
+ ;;
+ esac
fi
])
--
In memoriam Jean Moulin <http://en.wikipedia.org/wiki/Jean_Moulin>
- Re: [PATCH 03/10] pthread_sigmask: new module, (continued)
- Re: [PATCH 03/10] pthread_sigmask: new module, Bruno Haible, 2011/07/07
- Re: [PATCH 03/10] pthread_sigmask: new module, Bruno Haible, 2011/07/07
- Re: [PATCH 03/10] pthread_sigmask: new module, Bruno Haible, 2011/07/07
- Re: [PATCH 03/10] pthread_sigmask: new module, Bruno Haible, 2011/07/07
- Re: [PATCH 03/10] pthread_sigmask: new module, Bruno Haible, 2011/07/07
- Re: [PATCH 03/10] pthread_sigmask: new module, Bruno Haible, 2011/07/07
- Re: [PATCH 03/10] pthread_sigmask: new module, Bruno Haible, 2011/07/08
- Re: [PATCH 03/10] pthread_sigmask: new module, Bruno Haible, 2011/07/08
- Re: [PATCH 03/10] pthread_sigmask: new module, Bruno Haible, 2011/07/08
- Re: [PATCH 03/10] pthread_sigmask: new module,
Bruno Haible <=