Subj : Re: pthread_create hangs (bis)
To   : comp.programming.threads
From : gg
Date : Fri Apr 15 2005 04:55 pm

Hi,

has anyone an idea about this problem?
Thanks in advance,

Regards

gg wrote:


> 
>> gg wrote:
>> 
>>> Hi,
>>> 
>>> I'm developping a multi-thread application on debian sarge, kernel
>>> 2.4.28. When I do heavy-load tests, pthread_create sometimes hangs and
>>> __pthread_manager thread takes all CPU (R status) :
>>> top :
>>>PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
>>>6909 vocal     16   0 17404  16m  15m R 18.4  3.4   0:53.32 cvfmon
>>> 
>>> __pthread_manager thread :
>>> (gdb) bt
>>> #0  0x4054385a in poll () from /lib/libc.so.6
>>> #1  0x40429b50 in __pthread_manager () from /lib/libpthread.so.0
>>> #2  0x4054c69a in clone () from /lib/libc.so.6
>>> 
>>> pthread_create thread :
>>> 
>>> #0  0x4042c604 in __pthread_sigsuspend () from /lib/libpthread.so.0
>>> #1  0x4042c3c8 in __pthread_wait_for_restart_signal () from
>>> /lib/libpthread.so.0
>>> #2  0x4042bc42 in pthread_create@@GLIBC_2.1 () from /lib/libpthread.so.0
>>> #3  0x0805abb2 in cvfmon_pthread_create (id=0x80959f0, bool_joinable=1,
>>> init=0xfffffffc, arg=0xfffffffc, fct_appelante=0x808746c "babySitter")
>>> at commun.c:1653
>>> #4  0x08061540 in babySitter () at canal.c:2490
>>> #5  0x40429e51 in pthread_start_thread () from /lib/libpthread.so.0
>>> #6  0x4054c69a in clone () from /lib/libc.so.6
>>> 
>>> I haven't been able to make simpler test programs that reproduce this
>>> problem.
>>> Any idea?
>>> 
>>> Thanx in advance.
>>> 
>>> Regards,
>> 
>> Hi,
>> 
>> 
>> I had a look at __pthread_manager code, it seems to be a loop, that calls
>> the poll function on a pipe (__pthread_manager_reader file descripteur)
>> in order to wait for pthread_create request.
>> 
>> A problem with the pipe could be the origin of my "pthread_create hangs"
>> and "R status for __pthread_manager" problems.
>> 
>> If the pipe is out of order, the poll function call fails, the request
>> can't be read from the pipe, and the pthread_manager thread loops without
>> sleeping.
>> 
>> What can be the problem with the pipe?
>> What do you think about this hypothesis?
>> 
>> Thanks in advance,
>> Regards,
> 
> Hi,
> 
> I tried to print errno with gdb at after a poll function call :
> the value is 4 == EINTR : interrupted system call.
> Any idea?
> 
> Thanks in advance,
> Regards,

.