From mboxrd@z Thu Jan 1 00:00:00 1970 From: Kent Overstreet Subject: Re: [PATCH 22/25] Generic dynamic per cpu refcounting Date: Thu, 29 Nov 2012 12:59:20 -0800 Message-ID: <20121129205920.GL15094@google.com> References: <20121129185720.GE15094@google.com> <20121129185953.GW16230@one.firstfloor.org> <20121129191214.GG15094@google.com> <20121129192003.GX16230@one.firstfloor.org> <20121129192925.GH15094@google.com> <20121129193452.GI19042@kvack.org> <20121129202231.GJ15094@google.com> <20121129204531.GK15094@google.com> <20121129205447.GY16230@one.firstfloor.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Benjamin LaHaise , linux-kernel@vger.kernel.org, linux-aio@kvack.org, linux-fsdevel@vger.kernel.org, zab@redhat.com, jmoyer@redhat.com, axboe@kernel.dk, viro@zeniv.linux.org.uk To: Andi Kleen Return-path: Received: from mail-da0-f46.google.com ([209.85.210.46]:36520 "EHLO mail-da0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753867Ab2K2U7Y (ORCPT ); Thu, 29 Nov 2012 15:59:24 -0500 Received: by mail-da0-f46.google.com with SMTP id p5so5427890dak.19 for ; Thu, 29 Nov 2012 12:59:23 -0800 (PST) Content-Disposition: inline In-Reply-To: <20121129205447.GY16230@one.firstfloor.org> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Thu, Nov 29, 2012 at 09:54:47PM +0100, Andi Kleen wrote: > > > The regular atomic_t is limited in ways that you are not. > > > See my original mail. > > > > I don't follow, can you explain? > > For most cases the reference count is tied to some object, which are > naturally limited by memory size or other physical resources. > > But in the assymetric CPU case with your ref count no such limiter > exists. It's got exactly the same limit as the old code which used the atomic_t - we're limited by the number of threads that can be issuing aio syscalls at a time. The assymetry you're talking about _doesn't matter_, individual cpu counters wrapping does not affect what the counters all sum to when we go to tear down. A coworker at lunch actually pointed out to me that the reason this is true is just that modular arithmatic is still associative with addition and subtraction.