Re: possible race condition in linuxthreads-0.9 ..............

[Wolfram Gloger]

Hi,

> I am getting the segmentation fault error , when I am trying to run
> following
> 
> program on SMP .It works fine with UP system.I think , this is happening
> due to
> 
> some kind of race condition .

Yes, I have reproduced this on a dual-CPU Intel system many times with
what I consider to be a simpler test case:

http://www.malloc.de/test/fork-malloc.c

It appears to be some definitely _weird_ interaction between fork(),
threads and malloc.  Others and myself have so far failed to identify
the cause, unfortunately.  Could be a glibc or kernel bug.

> When I put a usleep () call inside manager
> thread
> 
> in glibc-2.2.2 source code , it never failed again.

Wow, this is exciting news to me, please tell us, where exactly did
you insert that usleep() to make the problem go away?

Regards,
Wolfram.