From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jens Axboe Subject: Re: [RFT PATCH] blkio: alloc per cpu data from worker thread context( Re: kvm deadlock) Date: Tue, 20 Dec 2011 13:49:14 +0100 Message-ID: <4EF0844A.4090600@kernel.dk> References: <54FC5923-2123-4BDD-A506-EA57DCE0C1F6@cpanel.net> <20111214122511.GD18317@amt.cnet> <4EE8A7ED.7060703@redhat.com> <4EE8C8EA.9070207@kernel.dk> <20111215194712.GA11194@redhat.com> <20111216202907.GH7586@redhat.com> <20111219172717.GB7175@redhat.com> <20111219173519.GL24519@google.com> <20111219182717.GC7175@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: Tejun Heo , Nate Custer , Avi Kivity , Marcelo Tosatti , kvm@vger.kernel.org, linux-kernel To: Vivek Goyal Return-path: Received: from merlin.infradead.org ([205.233.59.134]:57144 "EHLO merlin.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751125Ab1LTMtU (ORCPT ); Tue, 20 Dec 2011 07:49:20 -0500 In-Reply-To: <20111219182717.GC7175@redhat.com> Sender: kvm-owner@vger.kernel.org List-ID: On 2011-12-19 19:27, Vivek Goyal wrote: > I think reverting the previous series is not going to be simple either. It > had 13 patches. > > https://lkml.org/lkml/2011/5/19/560 > > By making stats per cpu, I was able to reduce contention on request > queue lock. Now we shall have to bring the lock back. That's not going to happen, having that lock in there was a disaster for small IO on fast devices. It's essentially the limiting factor on benchmark runs on the RHEL kernels that have it included and enabled... -- Jens Axboe