Re: [patch 1/3] raid5: offload stripe handle to workqueue

[Shaohua Li <shli@kernel.org>]

On Wed, Jul 31, 2013 at 06:33:32AM -0400, Tejun Heo wrote:
> Hello,
> 
> On Wed, Jul 31, 2013 at 09:24:34AM +0800, Shaohua Li wrote:
> > stripe is the work unit actually. As I said, if I queue a work for each stripe,
> > just queue_work() will make the system blast because of the pwq->pool->lock
> > contention. dispatching one work has another side effect that I can't add block
> 
> Hmmm.... I see.  I'm not familiar with the code base and could be
> missing something but how does the custom stripe dispatch queue
> synchronize?  Doesn't that make use of a lock anyway?  If so, how
> would scheduling separate work items be worse?  

It does have lock, but when a stripe is queued to handle, no lock is required.
So the workqueue lock will be high contended.

> Also, can you please
> elaborate the block plug part?

Basically I do:

blk_start_plug()
handle_stripe() //may dispatch request
blk_end_plug()

If only handle one stripe between block plug, the plug is useless, so I need
handle several stripes.

Thanks,
Shaohua