Re: [PATCH] multithreaded coredumps for elf exeecutables for O(1) scheduler

[Daniel Jacobowitz]

On Fri, May 10, 2002 at 09:24:14AM -0400, Mark Gross wrote:
> 
> Attached is my current patch for creating multithreaded core dump files,
> that works with the O(1) scheduler.
>  
> This is a continuation of the work posted by Vamsi Krishna back on 3/21/02.
> I'm sorry for the delay.  The problem of suspending the other thread
> processes for the duration of the core dump was a challenging problem with
> the O(1) scheduler.
>  
> Most of the patch is the same as that posted on 3/21/02 with some minor
> fixes and the rebasing to the 2.5.14 kernel.  The interesting bits are in
> the additions to sched.c to pause and resume the thread processes under the 
> O(1) scheduler.
>  
> Here I'm leveraging the work of Eric Foct for the process migration, to
> temporarily migrate the thread processes I need suspended to a "phantom
> runqueue".  This is just an additional run queue that has no cpu.  When I'm
> finish with the core dump I migrate them off the phantom run queue and
> continue processing whatever exit processing they do.
>  
> I tried a number of approaches to process pausing that didn't quite work 
> before I settled on the attached implementation.

That's a very interesting approach... I like it.

> This work has been unit test on a 2 way and 4 way SMP systems with no
> lockups so far.  YMMV.
>  
> Note: GDB 5.x will work with the core files created with this patch, provided
> the libpthread that gets loaded at gdb debug time is stripped of symbols. 
> 
> Run strip on your libpthread so files and things should work fine for you.  

Or use GDB 5.2.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer