Bug #1341: pthread_cond_timedwait failing in 1.9.1-p0 thread tests on HP-UX 11i v2

• Status: Rejected
• Assignee: -
• Target version: 2.6
• ruby -v: ruby 1.9.1p0 (2009-01-30 revision 21907) [ia64-hpux11.23]

[graza (Graham Agnew)]

=begin
I have been trying to compile and test 1.9.1-p0 on HP-UX 11i v2. When running the tests, the threads tests crash with the following bug:

[BUG] pthread_cond_timedwait: 22
ruby 1.9.1p0 (2009-01-30 revision 21907) [ia64-hpux11.23]

-- control frame ----------

-- Ruby level backtrace information-----------------------------------------

[NOTE]
You may encounter a bug of Ruby interpreter. Bug reports are welcome.
For details: http://www.ruby-lang.org/bugreport.html

The errno 22 means EINVAL. I put some print statements into the thread_pthread.c file to work out what was going on and it looks like there's a condition variable that is being initialised twice.
=end

[yugui (Yuki Sonoda)]

• Priority changed from Normal to 3
• Target version changed from 1.9.1 to 2.0.0

=begin

=end

[docwhat (Christian Höltje)]

=begin
This also happens when running "make test" in solaris but not "env RUBYLIB=./lib ./ruby test/ruby/test_thread.rb"

test_thread.rb ....bootstraptest.tmp.rb:6: [BUG] pthread_cond_timedwait: 22
ruby 1.9.1p243 (2009-07-16 revision 24175) [sparc-solaris2.8]

-- control frame ----------
c:0010 p:---- s:0028 b:0028 l:000027 d:000027 CFUNC :join
c:0009 p:0013 s:0024 b:0024 l:0018b8 d:000023 BLOCK bootstraptest.tmp.rb:6
c:0008 p:---- s:0020 b:0020 l:000019 d:000019 FINISH
c:0007 p:---- s:0018 b:0018 l:000017 d:000017 CFUNC :each
c:0006 p:0018 s:0015 b:0015 l:0018b8 d:002430 BLOCK bootstraptest.tmp.rb:3
c:0005 p:---- s:0012 b:0012 l:000011 d:000011 FINISH
c:0004 p:---- s:0010 b:0010 l:000009 d:000009 CFUNC :times
c:0003 p:0013 s:0007 b:0006 l:0018b8 d:001fd8 EVAL bootstraptest.tmp.rb:2
c:0002 p:---- s:0004 b:0004 l:000003 d:000003 FINISH
c:0001 p:0000 s:0002 b:0002 l:0018b8 d:0018b8 TOP

-- Ruby level backtrace information-----------------------------------------
bootstraptest.tmp.rb:6:in join' bootstraptest.tmp.rb:6:in block (2 levels) in '
bootstraptest.tmp.rb:3:in each' bootstraptest.tmp.rb:3:in block in '
bootstraptest.tmp.rb:2:in times' bootstraptest.tmp.rb:2:in '

[NOTE]
You may encounter a bug of Ruby interpreter. Bug reports are welcome.
For details: http://www.ruby-lang.org/bugreport.html

E......

It then hangs at this test.
=end

[mame (Yusuke Endoh)]

• Assignee set to mame (Yusuke Endoh)

=begin
Hi,

I guess this is the limitation of Solaris:

• http://bugs.opensolaris.org/view_bug.do?bug_id=4038480

• http://docs.sun.com/app/docs/doc/806-0630/6j9vkb8ct?a=view

  EINVAL
  ...
  For cond_timedwait(), the specified number of seconds, abstime,
  is greater than current_time + 100,000,000, where current_time
  is the current time, or the number of nanoseconds is greater
  than or equal to 1,000,000,000.

Maybe, HP-UX has the same limitation, though I cannot find the
evidence.

I wrote a workaround patch:

diff --git a/thread_pthread.c b/thread_pthread.c
index e6295db..7387724 100644
--- a/thread_pthread.c
+++ b/thread_pthread.c
@@ -633,6 +633,35 @@ native_sleep(rb_thread_t *th, struct timeval *tv)
(unsigned long)ts.tv_sec, ts.tv_nsec);
r = pthread_cond_timedwait(&th->native_thread_data.sleep_cond,
&th->interrupt_lock, &ts);

•  if (r == EINVAL) {
  

•      /* workaround for Solaris: wait by MEGA_SEC's.
  

•       * on Solaris, pthread_cond_timedwait fails with EINVAL
  

•       * if time is too far from now.  [Bug #1341]
  

•       * - http://docs.sun.com/app/docs/doc/806-0630/6j9vkb8ct?a=view
  

•       * - http://bugs.opensolaris.org/view_bug.do?bug_id=4038480
  

•       */
  

+#define MEGA_SEC 1000000

•      struct timeval ltv = *tv;
  

•      r = ETIMEDOUT;
  

•      while (r == ETIMEDOUT && ltv.tv_sec > MEGA_SEC) {
  

•  	ts.tv_sec = tvn.tv_sec + MEGA_SEC;
  

•  	ts.tv_nsec = tvn.tv_usec * 1000;
  

•  	ltv.tv_sec -= MEGA_SEC;
  

•  	r = pthread_cond_timedwait(&th->native_thread_data.sleep_cond,
  

•  				   &th->interrupt_lock, &ts);
  

•  	if (r && r != ETIMEDOUT) rb_bug_errno("pthread_cond_timedwait", r);
  

•      }
  

•      if (r == ETIMEDOUT) {
  

•  	ts.tv_sec = tvn.tv_sec + ltv.tv_sec;
  

•  	ts.tv_nsec = (tvn.tv_usec + ltv.tv_usec) * 1000;
  

•  	if (ts.tv_nsec >= PER_NANO){
  

•  		    ts.tv_sec += 1;
  

•  		    ts.tv_nsec -= PER_NANO;
  

•  	}
  

•  	r = pthread_cond_timedwait(&th->native_thread_data.sleep_cond,
  

•  				   &th->interrupt_lock, &ts);
  

•      }
  

•  }
   if (r && r != ETIMEDOUT) rb_bug_errno("pthread_cond_timedwait", r);
  
   thread_debug("native_sleep: pthread_cond_timedwait end (%d)\n", r);
  

--
Yusuke Endoh [e-mail]
=end

[graza (Graham Agnew)]

=begin
Today I downloaded, patched, and compiled the latest snapshot.

On HP-UX, that patch stopped the rb_bug_errno happening, although test/ruby/test_thread.rb scripts blocked indefinitely. On being interrupted it was the following test:

#874 test_thread.rb:34:in `<top (required)>':

The code in your patch doesn't look right to me. Shouldn't the code re-fetch the time using gettimeofday each time through the loop, and then add the MEGA_SEC to that? As it is, it's wrong because it adds a MEGA_SEC to tvn, so after one MEGA_SEC it will enter a hard loop.

Cheers,
Gra.
=end

[mame (Yusuke Endoh)]

=begin
Hi,

2010/5/6 Graham Agnew [e-mail]:

On HP-UX, that patch stopped the rb_bug_errno happening, although the test/ruby/test_io.rb and test/ruby/test_thread.rb scripts both blocked indefinitely.

The code in your patch doesn't look right to me. ?Shouldn't the code re-fetch the time using gettimeofday each time through the loop, and then add the MEGA_SEC to that? ?As it is, it's wrong because it adds a MEGA_SEC to tvn, so after one MEGA_SEC it will enter a hard loop.

Thank you for your testing! How about the following patch?

To tell the truth, I'm writing a patch without test because I
don't have HP-UX. If this is wrong again, It is really helpful
for you to correct the patch by yourself.

diff --git a/thread_pthread.c b/thread_pthread.c
index e6295db..3c13f72 100644
--- a/thread_pthread.c
+++ b/thread_pthread.c
@@ -631,8 +631,29 @@ native_sleep(rb_thread_t *th, struct timeval *tv)
int r;
thread_debug("native_sleep: pthread_cond_timedwait start (%ld, %ld)\n",
(unsigned long)ts.tv_sec, ts.tv_nsec);

•    again:
   r = pthread_cond_timedwait(&th->native_thread_data.sleep_cond,
   			   &th->interrupt_lock, &ts);
  

•  if (r == EINVAL) {
  

•      /* workaround for Solaris: wait by MEGA_SEC's.
  

•       * on Solaris, pthread_cond_timedwait fails with EINVAL
  

•       * if time is too far from now.  [Bug #1341]
  

•       * - http://docs.sun.com/app/docs/doc/806-0630/6j9vkb8ct?a=view
  

•       * - http://bugs.opensolaris.org/view_bug.do?bug_id=4038480
  

•       */
  

+#define MEGA_SEC 1000000

•      struct timespec lts;
  

•      r = ETIMEDOUT;
  

•      while (r == ETIMEDOUT) {
  

•  	gettimeofday(&tvn, NULL);
  

•  	lts.tv_sec = tvn.tv_sec + MEGA_SEC;
  

•  	lts.tv_nsec = tvn.tv_usec * 1000;
  

•  	if (lts.tv_sec >= ts.tv_sec) goto again;
  

•  	r = pthread_cond_timedwait(&th->native_thread_data.sleep_cond,
  

•  				   &th->interrupt_lock, &lts);
  

•  	if (r && r != ETIMEDOUT) rb_bug_errno("pthread_cond_timedwait", r);
  

•      }
  

•  }
   if (r && r != ETIMEDOUT) rb_bug_errno("pthread_cond_timedwait", r);
  
   thread_debug("native_sleep: pthread_cond_timedwait end (%d)\n", r);
  

--
Yusuke Endoh [e-mail]

=end

[graza (Graham Agnew)]

=begin
Hi Yusuke,

That code looks better although I'm still getting test/ruby/test_thread.rb blocking indefinitely. I will see if I can attach to the process with a debugger and figure out what it's blocked on.

Thanks,
Gra.
=end

[shyouhei (Shyouhei Urabe)]

• Status changed from Open to Assigned

=begin

=end

[ko1 (Koichi Sasada)]

Endo-san,

Can we close this issue?

[mame (Yusuke Endoh)]

I guess it still reproduces on Solaris.
I have no idea about HP-UX.

--
Yusuke Endoh [e-mail]

[kosaki (Motohiro KOSAKI)]

Now, following two links are dead. Do anyone know new URLs?

• • http://docs.sun.com/app/docs/doc/806-0630/6j9vkb8ct?a=view
• • http://bugs.opensolaris.org/view_bug.do?bug_id=4038480

[mame (Yusuke Endoh)]

• Status changed from Assigned to Open
• Assignee deleted (mame (Yusuke Endoh))

Hello,

Now, following two links are dead. Do anyone know new URLs?

• • http://docs.sun.com/app/docs/doc/806-0630/6j9vkb8ct?a=view
• • http://bugs.opensolaris.org/view_bug.do?bug_id=4038480

Here.

http://download.oracle.com/docs/cd/E19683-01/816-0216/6m6ngupgv/index.html

EINVAL
Invalid argument. For cond_init(), type is not a recognized type. For cond_timedwait(), the specified number of seconds, abstime, is greater than current_time + 100,000,000, where current_time is the current time, or the number of nanoseconds is greater than or equal to 1,000,000,000.

The problem that I now focus on is that pthread_cond_timedwait may fail
with EINVAL if an argument is greater than current_time + 100,000,000
on Solaris.
The patch of [ruby-core:29702] is too old and cannot be applied, so
I rewrote and committed a new patch at r32409. Now, "make test" passes
on Solaris. Congrats.

Unfortunately, the original issue that OP reported was a different problem.
But I guess that there is no hope of fixing the issue. At least I cannot.
So I resign the assignee of this ticket. Sorry for late action.

--
Yusuke Endoh [e-mail]

[naruse (Yui NARUSE)]

• Status changed from Open to Feedback

Feedback about HP-UX is welcome

[kosaki (Motohiro KOSAKI)]

• Subject changed from pthread_cond_timedwait failing in 1.9.1-p0 thread tests to pthread_cond_timedwait failing in 1.9.1-p0 thread tests on HP-UX 11i v2

[ko1 (Koichi Sasada)]

• Target version changed from 2.0.0 to 2.6

Please tell us if you have HP-UX.

[mame (Yusuke Endoh)]

• Status changed from Feedback to Rejected

HP-UX is not supported. I'm sorry, but please create a patch that works yourself.
If you provide us the patch and it looks benign to other platforms, we may apply it to trunk.

--
Yusuke Endoh [e-mail]