Re: [PATCH] 2.5.3 remove global semaphore_lock spin lock.

[Bob Miller <rem@osdl.org>]

On Fri, Feb 01, 2002 at 01:05:22PM -0800, Andrew Morton wrote:
> Bob Miller wrote:
> > 
> > Also, at your suggestion I set wait.h:USE_RW_WAIT_QUEUE_SPINLOCK on
> > a clean 2.5.3 system to test.  The problem is that it OOPs on startup.
> 
> OK, someone broke it; possibly the scheduler changes.  Not surprising,
> really.
> 
> It'd be nice to have it fixed, but I wouldn't suggest that you
> bust a gut over it.   Certainly your patch shouldn't be held up by
> this.  An oops trace would be useful.
> 
> -
I found the problem... in kernel/sched.c around the 2.4.7 time frame
wait_for_completion() and other code was added.  It uses a new
completion structure that has a wait_queue_haed_t embedded in it.
In wait_for_completion() and other places they use spin_lock_*()
calls that cause the OOPs.

In order to do some of the clean up you suggested I needed to and
some macros to wait.h.  To fix wait_for_completion() and others
those same macros will be needed.  I'm going to fix the wait_for_completion()
stuff first and then get back to the sema stuff.

-- 
Bob Miller					Email: rem@osdl.org
Open Software Development Lab			Phone: 503.626.2455 Ext. 17