Re: [PATCH glibc-2.44] Fix __pthread_timedblock for hurd

[Samuel Thibault]

Samuel Thibault, le dim. 23 août 2026 11:11:52 +0200, a ecrit:
> Samuel Thibault, le dim. 23 août 2026 10:24:52 +0200, a ecrit:
> > One thing that is a bug, however, is the program below, which stays
> > blocked because the mach_msg calling loop keeps using the same timeout.
> > 
> > We probably need a version that takes a clock_id and an absolute time,
> > and does the clock_gettime call itself on each mach_msg calling loop.
> 
> Or probably simpler: make __pthread_timedblock always pass
> MACH_RCV_INTERRUPT to __mach_msg, but loop around if it returns EINTR
> and MACH_RCV_INTERRUPT was not in MSG_OPTIONS.

Could you try this?

Samuel

Index: glibc-2.43/sysdeps/mach/htl/pt-timedblock.c
===================================================================
--- glibc-2.43.orig/sysdeps/mach/htl/pt-timedblock.c
+++ glibc-2.43/sysdeps/mach/htl/pt-timedblock.c
@@ -40,6 +40,7 @@ __pthread_timedblock (struct __pthread *
   mach_msg_timeout_t timeout;
   struct timespec now;
 
+retry:
   /* We have an absolute time and now we have to convert it to a
      relative time.  Arg.  */
 
@@ -58,13 +59,19 @@ __pthread_timedblock (struct __pthread *
     /* Need to do a carry.  */
     timeout -= (now.tv_nsec - abstime->tv_nsec + 999999) / 1000000;
 
-  err = __mach_msg (&msg, MACH_RCV_MSG | MACH_RCV_TIMEOUT | MSG_OPTIONS, 0,
+  err = __mach_msg (&msg, MACH_RCV_MSG | MACH_RCV_TIMEOUT | MSG_OPTIONS | 
MACH_RCV_INTERRUPT, 0,
                    sizeof msg, thread->wakeupmsg.msgh_remote_port,
                    timeout, MACH_PORT_NULL);
   if (err == EMACH_RCV_TIMED_OUT)
     return ETIMEDOUT;
-  if ((MSG_OPTIONS & MACH_RCV_INTERRUPT) && err == MACH_RCV_INTERRUPTED)
-    return EINTR;
+  if (err == MACH_RCV_INTERRUPTED)
+    {
+      if (!((MSG_OPTIONS & MACH_RCV_INTERRUPT)))
+       /* Re-take absolute time target. */
+       goto retry;
+      else
+       return EINTR;
+    }
 
   assert_perror (err);
   return 0;