Re: spinlock requests (was RE: [Patch] don't spin with irq disabled)

[Juergen Gross <juergen.gross@fujitsu-siemens.com>]

Dan Magenheimer wrote:
> Keir (and/or others) --
> 
> If you are messing around in the spinlock code anyway,
> I have a couple of requests.  Tmem needs:
> 
> rw_is_write_locked(rwlock_t *lock)
> 
> and
> 
> write_trylock(rwlock_t *lock)
> 
> I implemented the latter by grabbing the C code from Linux
> and it seems to work, but it would be nice if it was
> consistent with the other lock code in xen and
> my asm statement understanding is too poor to try.
> 
> For rw_is_write_locked(), I had a devil of a time
> because of what appeared to be a weird code generated
> race; the obvious simple implementation failed
> periodically... apparently due to racing against
> try_readlock attempts!  (I use it in an ASSERT so it
> was a rather noticeable and spectacular failure!)
> The workaround I used below is a horrible hack
> but I haven't had problems since.
> 
> Thanks,
> Dan

Dan,

if you are planning to use rw_locks you should be aware that the current
implementation in Xen is sub-optimal on systems with high processor counts.
Read locks always succeed when other readers are already holding the lock,
even if a writer is waiting for the lock. If there are many potential readers
they might (in theory) lock out a writer for rather long times.
A better solution would be to stop further readers to acquire the lock if a
writer is waiting for it.

Juergen

-- 
Juergen Gross                             Principal Developer
IP SW OS6                      Telephone: +49 (0) 89 636 47950
Fujitsu Siemens Computers         e-mail: juergen.gross@fujitsu-siemens.com
Otto-Hahn-Ring 6                Internet: www.fujitsu-siemens.com
D-81739 Muenchen         Company details: www.fujitsu-siemens.com/imprint.html