Re: sx locks rewriting - needs testers

From: Attilio Rao <attilio_at_freebsd.org>
Date: Mon, 6 Nov 2006 03:57:50 +0100
2006/10/18, Attilio Rao <attilio_at_freebsd.org>:
> In my P4 branch: //depot/user/attilio/attilio_smpng/...
> you can find a sx locks rewriting using the optimized semantic of
> rwlocks; in the end this might result in a valuable performance
> improvement.

After have received very positive stress-test feedbacks from pho_at_
(that I would like to thank for his patience and work) I went ahead
and rather completed the implementation. Now this is ready to be
reviewed and possibly committed into the CVS.
Even if I plan a longer work on this branch (about syncronizing
primitives), I think it is time for a revision of the work done until
now from SMPng people (jhb, kmacy, etc.) and possibily an inclusion
into HEAD (patch actually is 58k...).

Some few hints about the patch:
- LOCK_DEBUG adds a dependence between sx.h and lock.h (as for rwlocks
and mutex) and a the new options SXLOCK_NOINLINE is added
- XFS locking is still disabled in the patch (I hope to do it for
tomorrow, I'm in GMT+1).
-  Possibly the sleepqueue interface modifies and new flags might be
documented in the manpages (and NOTES file too, in order to reflect
SXLOCK_NOINLINE inclusion).
- It misses still of the adaptive spinning code, but it can be
inserted after without problems.

You can download the code directly from perforce
(//depot/user/attilio/attilio_smpng/...) but patches are available
here:
http://users.gufi.org/~rookie/works/patches/smpng06112006.diff
http://users.gufi.org/~rookie/works/patches/_sx.h

I hope you will enjoy it (feedbacks, ideas, comments would be very appreciated).

Attilio


-- 
Peace can only be achieved by understanding - A. Einstein
Received on Mon Nov 06 2006 - 01:58:03 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:02 UTC