Bug in pthread glibc 7.0 & 7.1

Bug in pthread glibc 7.0 & 7.1

[TWEDE,ROGER (HP-Boise,ex1)]

The following code reveals a bug in the MIPS Gnu C Libraries available via
ftp on oss.sgi.com (ftp.linux.sgi.com).

The bug is posix thread related and occurs when using the libc and
libpthread versions available at:
/pub/linux/mips/redhat/7.1/RPMS/mipsel/glibc*

The bug also appears to exist in:
/pub/linux/mips/mipsel-linux/RedHat-7.0/RPMS/mipsel/glibc*

The bug does not exist in earlier versions of the pthread library, or on the
non-MIPS (non-SGI) distributions.  (ie. RedHat 7.0 for ix86 does not exhibit
a failure with this code, but RedHat 7.0/7.1 on mips does)

Upon failure the following code sequence will core dump (SEGV) instead of
exiting normally.


Thanks,

Roger Twede
Engineer/Scientist
Hewlett Packard Company


// CODE STARTS BELOW


#include <assert.h>
#include <pthread.h>
#include <sched.h>
#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>


struct ThreadStartInfo {
    pthread_mutex_t InitCompleteMutex;
    pthread_cond_t InitCompleteCond;
    void * (*Func)(void *);
    void * Arg;
    int Priority;
    char * ThreadName;
    unsigned char InitCompleteCount;
};


static void * StartFunction(void * Arg)
{

   struct ThreadStartInfo * StartInfo = (struct ThreadStartInfo *) Arg;
   int result;
   void * retVal = 0;

   assert(Arg != NULL);

   result = pthread_mutex_lock(&(StartInfo->InitCompleteMutex));
   assert(result == 0);
   printf("pid=%d thread mutex locked at x%x\n", getpid(),
&(StartInfo->InitCompleteMutex));
   StartInfo->InitCompleteCount = 1;
   result = pthread_cond_signal(&(StartInfo->InitCompleteCond));
   assert(result == 0);
   printf("pid=%d thread cond signal sent, unlocking at 0x%x\n", getpid(),
&(StartInfo->InitCompleteMutex));
   result = pthread_mutex_unlock(&(StartInfo->InitCompleteMutex));
   assert(result == 0);
   printf("pid=%d thread unlocked\n", getpid());
   sched_yield();
   printf("pid=%d yielded and back again\n", getpid());

   return(retVal);


}




int main(void)
{
    pthread_t ThreadObject;
    struct ThreadStartInfo StartInfo;
    pthread_mutexattr_t mutexAttr;
    pthread_attr_t threadAttr;
    pthread_condattr_t condAttr;
    int result = 0;

    StartInfo.ThreadName = "mythread";

    result = pthread_mutexattr_init(&mutexAttr);
    assert(result == 0);
    printf("pid=%d Init mutex\n", getpid());
    result = pthread_mutex_init(&(StartInfo.InitCompleteMutex), &mutexAttr);
    assert(result == 0);

    result = pthread_condattr_init(&condAttr);
    assert(result == 0);
    result = pthread_cond_init(&(StartInfo.InitCompleteCond), &condAttr);
    assert(result == 0);

    StartInfo.InitCompleteCount = 0;

    pthread_mutex_lock(&(StartInfo.InitCompleteMutex));

    printf("pid=%d About to create thread: %s\n", getpid(),
StartInfo.ThreadName);
    result = pthread_attr_init(&threadAttr);
    assert(result == 0);
    result = pthread_create(&ThreadObject,
                              &threadAttr,
                              &StartFunction,
                              (void *)&StartInfo);
    assert(result == 0);
    while ((result == 0) && (StartInfo.InitCompleteCount == 0))
    {
        do
        {
	   printf("pid=%d about to cond_wait for %s init 1.\n", getpid(),
StartInfo.ThreadName);
    	   result = pthread_cond_wait(&(StartInfo.InitCompleteCond),
&(StartInfo.InitCompleteMutex));
    	   printf("pid=%d back from cond_wait for %s init 1.  result=%d\n",
getpid(), StartInfo.ThreadName, result);
        } while (result == EINTR);
    }

    result = pthread_mutex_unlock(&(StartInfo.InitCompleteMutex));
    assert(result == 0);

    getchar();
    return 0;
}


// CODE ABOVE

Re: Bug in pthread glibc 7.0 & 7.1

[H . J . Lu]

On Mon, Oct 01, 2001 at 07:52:15PM -0400, TWEDE,ROGER (HP-Boise,ex1) wrote:
> 
> 
> The following code reveals a bug in the MIPS Gnu C Libraries available via
> ftp on oss.sgi.com (ftp.linux.sgi.com).
> 
> The bug is posix thread related and occurs when using the libc and
> libpthread versions available at:
> /pub/linux/mips/redhat/7.1/RPMS/mipsel/glibc*
> 
> The bug also appears to exist in:
> /pub/linux/mips/mipsel-linux/RedHat-7.0/RPMS/mipsel/glibc*
> 
> The bug does not exist in earlier versions of the pthread library, or on the
> non-MIPS (non-SGI) distributions.  (ie. RedHat 7.0 for ix86 does not exhibit
> a failure with this code, but RedHat 7.0/7.1 on mips does)
> 
> Upon failure the following code sequence will core dump (SEGV) instead of
> exiting normally.
> 
> 

On RedHat 7.1/mips:

# gcc pthread.c -o mips -lpthread -Wall
pthread.c: In function `StartFunction':
pthread.c:64: warning: unsigned int format, pointer arg (arg 3)
pthread.c:69: warning: unsigned int format, pointer arg (arg 3)
# ./mips
pid=21905 Init mutex
pid=21905 About to create thread: mythread
pid=21905 about to cond_wait for mythread init 1.
pid=21907 thread mutex locked at x7fff79c8
pid=21907 thread cond signal sent, unlocking at 0x7fff79c8
pid=21907 thread unlocked
pid=21905 back from cond_wait for mythread init 1.  result=0
pid=21907 yielded and back again
# rpm -q glibc
glibc-2.2.4-11.2


H.J.

Re: Bug in pthread glibc 7.0 & 7.1

[Robin Humble]

H.J. wrote:
>On Mon, Oct 01, 2001 at 07:52:15PM -0400, TWEDE,ROGER (HP-Boise,ex1) wrote:
>> The following code reveals a bug in the MIPS Gnu C Libraries available via
>> ftp on oss.sgi.com (ftp.linux.sgi.com).
>> ...
>On RedHat 7.1/mips:
>
># gcc pthread.c -o mips -lpthread -Wall
>pthread.c: In function `StartFunction':
>pthread.c:64: warning: unsigned int format, pointer arg (arg 3)
>pthread.c:69: warning: unsigned int format, pointer arg (arg 3)
># ./mips
>pid=21905 Init mutex
>pid=21905 About to create thread: mythread
>pid=21905 about to cond_wait for mythread init 1.
>pid=21907 thread mutex locked at x7fff79c8
>pid=21907 thread cond signal sent, unlocking at 0x7fff79c8
>pid=21907 thread unlocked
>pid=21905 back from cond_wait for mythread init 1.  result=0
>pid=21907 yielded and back again
># rpm -q glibc
>glibc-2.2.4-11.2

On my R4600SC Indy running 7.1 (mips) I get:

% gcc blah.c -o blah -lpthread -Wall
blah.c: In function `StartFunction':
blah.c:37: warning: unsigned int format, pointer arg (arg 3)
blah.c:42: warning: unsigned int format, pointer arg (arg 3)
% ./blah
[blah:5286] Illegal instruction 0100017c at 2ad1a910 ra=2ab78ed0
Illegal instruction (core dumped)
% uname -a
Linux elan 2.4.3 #1 Sun Apr 22 23:46:19 PDT 2001 mips unknown
% rpm -q glibc gcc
glibc-2.2.4-11.2
gcc-2.96-97.2
%

This happens whether it's a native build of glibc or the one from
oss.sgi.com. The kernel is from the simple dir on oss.

With a 3rd Sept CVS 2.4.8 kernel under no load the pthreads program
runs ok, but under load (a big rm -rf over NFS for instance), it
fails like above, or fails like:

% ./blah
pid=681 Init mutex
pid=681 About to create thread: mythread
pid=681 about to cond_wait for mythread init 1.
Segmentation fault (core dumped)
%

Also I found I could only build X natively on the Indy by disabling the
pthreads parts; and maybe it's a similar 'Illegal Instruction' error in
the 'conftest' program that seems to be stopping mozilla from building??
(at least under kernel 2.4.3 which otherwise seems more stable than 2.4.8)

cheers,
robin

Re: Bug in pthread glibc 7.0 & 7.1

[H . J . Lu]

On Tue, Oct 02, 2001 at 10:09:25PM +1000, Robin Humble wrote:
> Also I found I could only build X natively on the Indy by disabling the
> pthreads parts; and maybe it's a similar 'Illegal Instruction' error in
> the 'conftest' program that seems to be stopping mozilla from building??
> (at least under kernel 2.4.3 which otherwise seems more stable than 2.4.8)
> 

My impression is 2.4.3 is more stable than 2.4.5. I haven't tried
2.4.8 yet.


H.J.