From: Jeremy Fitzhardinge Subject: Re: [PATCH RFC 0/4] Paravirtual spinlocks Date: Mon, 07 Jul 2008 17:37:54 -0700 Message-ID: <4872B6E2.5080003@goop.org> References: <20080707190749.299430659@goop.org> <200807081029.19242.rusty@rustcorp.com.au> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <200807081029.19242.rusty@rustcorp.com.au> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Rusty Russell Cc: Jens Axboe , Nick Piggin , Xen devel , Peter Zijlstra , Christoph Lameter , Petr Tesarik , LKML , virtualization@lists.linux-foundation.org, Thomas Friebel , Ingo Molnar List-Id: xen-devel@lists.xenproject.org Rusty Russell wrote: > On Tuesday 08 July 2008 05:07:49 Jeremy Fitzhardinge wrote: > >> At the most recent Xen Summit, Thomas Friebel presented a paper >> ("Preventing Guests from Spinning Around", >> http://xen.org/files/xensummitboston08/LHP.pdf) investigating the >> interactions between spinlocks and virtual machines. Specifically, he >> looked at what happens when a lock-holding VCPU gets involuntarily >> preempted. >> > > I find it interesting that gang scheduling the guest was not suggested as an > obvious solution. > It's an obvious answer, but not an obvious solution. You trade off wasting time spinning vs wasting time waiting for N vcpus to be free for scheduling. Or something; seems much more complex, particularly if you can do a small guest tweak to solve the problem. > Anyway, concept looks fine; lguest's solution is more elegant of course :) > You could remove all mutable state and call it "erlang". > A little disappointing that you can't patch your version inline. Spinlock code isn't inlined currently, so I hadn't considered it. The fast path code for both lock and unlock is nearly small enough to consider it, but it seems a bit fiddly. If the "spin_lock" and "spin_unlock" functions were inlined functions which called the out of line __raw_spin_lock/unlock functions, then after patching they would result in a direct call to the backend lock functions, which would be exactly equivalent to what happens now (since I hook __raw_spin_lock into calls via pv_lock_ops). J