On 05/28/2012 08:18 PM, Christoph Hellwig wrote: > On Mon, May 28, 2012 at 08:15:48PM +0800, Liu Yuan wrote: >> I am unsure if there is any problem that we set sys_stat before we get >> the vdi bitmap. At least, this is one diffidence than the old logic. > > Yes - as I said in the message that is the one intentional difference. > By queing the vdi bitmap update in the block WQ we make sure it's > uptodate by the time VDI operations are processed. I'm still not overly > hapy with the way the vdi bitmap works, but I'll need some more > infrastructure changes before I can tackle that problem. > >> Also, we can rename process_request_event_queues() as >> process_request_queue since there is no event queue at all. > > I have a few more patches under development to entirely remove that > function. Given that they shouldn't take too long renaming > process_request_event_queues no seems like churn without much benefit. > Okay, I am fine with leaving it to later patch. This patch looks really nice to me, split event handling into worker and main thread sucks a long time. This patch also fixes an epoch mismatch bug that my last patch tried to address. Thanks, Yuan |