[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [PATCH] PPC lll_mutex_unlock_force write barrier



sjmunroe vnet ibm com schrieb am 26.08.03 22:01:23:
> 
> For Power4 and future processors with long store queues and out of order
> execution, lll_mutex_unlock_force needs a write barrier between the store and
> the futex syscall.

Sort of. First off, unless I'm just missing and/or misunderstanding
something, you've put it in the wrong place. Also, a write barrier 
isn't enough here. mutex.unlock also needs a read barrier. And both 
should be 'sink' barriers in this case (mutex.lock needs 'hoist' 
ones). Well, a hypothetical rdunlock operation (on a read-write 
lock) would need a 'sink-load' barrier only (no need whatsoever for 
constraining reordering of stores)... if only compiler folks would 
'catch up' on this...

regards,
alexander.

P.S. AFAICS, Intel's OP.RELEASE msync stuff can be viewed as an 
atomic operation with combined sink-load and sink-store barriers. 
Well, <http://google.com/groups?selm=3EE0CA46.593F938B%40web.de>.

__________________________________________________________________________
Die sicherste Form der Kommunikation: E-Mails verschluesseln, Spam-Filter,
Adressverifizierung, digitale Unterschrift: http://freemail.web.de




[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]