Messages in this thread | | | Date | Wed, 20 Mar 2013 11:30:37 -0700 | From | Tejun Heo <> | Subject | Re: [PATCH 00/21] workqueue: cleanups and better locking for recent changes |
| |
So, overall,
On Wed, Mar 20, 2013 at 03:28:00AM +0800, Lai Jiangshan wrote: ... > In this list, we can find that: > 1) wq_mutex protects too much different kind of things.
I don't agree with this and unless you can come up with a much better reason, won't be splitting wq_mutex further. Also, I'm not gonna rename it to wqs_mutex.
> 2) workqueue->pwqs are protected by both wq->flush_mutex and pwq_lock, > but in many read sites, they are protected by both wq->flush_mutex and pwq_lock, > in some other read sites, they are protected by pwq_lock, but can be converted > to wq->flush_mutex. it means pwq_lock and wq->flush_mutex are redundancy here. > 3) pwq_lock is global lock, but it protects only workqueue instance fields.
A global lock protecting different instances is perfectly fine unless it actually causes contention in some paths. It often actually is better to have a single global lock for cold paths as it pollutes less amount of cache, so please don't split locks for that reason.
That said, I like the fact that wq->flush_mutex can be transformed into wq->mutex and replaces pwq_lock making it go away, so, yeah, for that reason, I like pwq_lock removal, so yeah, let's replace pwq_lock with wq->mutex.
I applied the ones I can apply at this point. I'll rebase the NUMA patchset on top of the applied ones. Please update the rest on top.
Thanks.
-- tejun
| |