Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1762112Ab0HFWba (ORCPT ); Fri, 6 Aug 2010 18:31:30 -0400 Received: from claw.goop.org ([74.207.240.146]:40709 "EHLO claw.goop.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751780Ab0HFWb1 (ORCPT ); Fri, 6 Aug 2010 18:31:27 -0400 Message-ID: <4C5C86CD.2000109@goop.org> Date: Fri, 06 Aug 2010 15:03:57 -0700 From: Jeremy Fitzhardinge User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.2.7) Gecko/20100720 Fedora/3.1.1-1.fc13 Lightning/1.0b2pre Thunderbird/3.1.1 MIME-Version: 1.0 To: "H. Peter Anvin" CC: Jan Beulich , Peter Zijlstra , Xen-devel , Avi Kivity , Nick Piggin , Linux Kernel Mailing List Subject: Re: [PATCH RFC 02/12] x86/ticketlock: convert spin loop to C References: <1280761639.1923.213.camel@laptop> <4C56E1A1.6020005@goop.org> <4C5C1F80020000780000EA6D@vpn.id2.novell.com> <4C5C21E3.1020004@goop.org> <4C5C6DCE.4020206@zytor.com> <4C5C71A5.3050500@goop.org> <4C5C7A0A.9080107@zytor.com> In-Reply-To: <4C5C7A0A.9080107@zytor.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2405 Lines: 75 On 08/06/2010 02:09 PM, H. Peter Anvin wrote: > On 08/06/2010 01:33 PM, Jeremy Fitzhardinge wrote: >> On 08/06/2010 01:17 PM, H. Peter Anvin wrote: >>> On 08/06/2010 07:53 AM, Jeremy Fitzhardinge wrote: >>>> On 08/06/2010 05:43 AM, Jan Beulich wrote: >>>>> You certainly mean "the compiler currently treats this as being:" - I >>>>> don't think there's a guarantee it'll always be doing so. >>>>> >>>>>> for (;;) { >>>>>> if (inc.tickets.head == inc.tickets.tail) >>>>>> goto out; >>>>>> ... >>>>>> } >>>>>> out: barrier(); >>>>>> } >>>>>> >>>>>> (Which would probably be a reasonable way to clarify the code.) >>>>> I therefore think it needs to be written this way. >>>> >>>> Agreed. >>>> >>> >>> A call/return to an actual out-of-line function is a barrier (and will >>> always be a barrier, as it is the fundamental ABI sequence points), >>> but to an inline function it is not. >> >> Yes. So the goto explicitly puts the barrier into the control flow which >> should stop the compiler from doing anything unexpected. >> > > In this particular case, though, I would somewhat expect the more > conventional: > > while (inc.tickets.head != inc.tickets.tail) { > cpu_relax(); > inc.tickets.head = ACCESS_ONCE(lock->tickets_head); > } Yes, that makes sense for the plain spinlock version. But the full code, including the pv-ticketlock spin timeout, ends up being: static __always_inline void arch_spin_lock(struct arch_spinlock *lock) { register struct __raw_tickets inc; inc = __ticket_spin_claim(lock); for (;;) { unsigned count = SPIN_THRESHOLD; do { if (inc.head == inc.tail) goto out; cpu_relax(); inc.head = ACCESS_ONCE(lock->tickets.head); } while (--count); __ticket_lock_spinning(lock, inc.tail); } out: barrier(); /* make sure nothing creeps before the lock is taken */ } So the goto form is closer to the final form. If it weren't for this, I'd also prefer the while() form. (If you config PARAVIRT_SPINLOCKS=n, then __ticket_lock_spinning() becomes an empty inline, which causes gcc to collapse the whole thing into a simple infinite loop (ie, it eliminates "count" and the inner loop).) J -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/