Re: [PATCH] i386 rw_semaphores fix

From: David Howells <dhowells@cambridge.redhat.com>
To: Andrew Morton <andrewm@uow.edu.au>
Cc: David Howells <dhowells@cambridge.redhat.com>,
	Linus Torvalds <torvalds@transmeta.com>,
	Ben LaHaise <bcrl@redhat.com>,
	Alan Cox <alan@lxorguk.ukuu.org.uk>,
	Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] i386 rw_semaphores fix
Date: Wed, 11 Apr 2001 22:27:02 +0100	[thread overview]
Message-ID: <17752.987024422@warthog.cambridge.redhat.com> (raw)
In-Reply-To: Your message of "Wed, 11 Apr 2001 09:56:09 PDT." <3AD48CA9.CA03B85D@uow.edu.au>

> You need sterner testing stuff :)  I hit the BUG at the end of rwsem_wake()
> in about a second running rwsem-4.  Removed the BUG and everything stops
> in D state.
> 
> Grab rwsem-4 from
> 
> 	http://www.uow.edu.au/~andrewm/linux/rwsem.tar.gz
> 
> It's very simple.  But running fully in-kernel shortens the
> code paths enormously and allows you to find those little
> timing windows.  Run rmsem-4 in two modes: one with
> the schedule() in sched() enabled, and also with it
> commented out.  If it passes that, it works.  When
> you remove the module it'll print out the number of
> read-grants versus write-grants.  If these run at 6:1
> with schedule() disabled then you've kicked butt.
> 
> Also, rwsem-4 checks that the rwsems are actually providing
> exclusion between readers and writers, and between
> writers and writers.  A useful thing to check, that.

It now works (patch to follow).

schedule() enabled:
 reads taken: 686273
 writes taken: 193414

schedule() disabled:
 reads taken: 585619
 writes taken: 292997

David