From: Keir Fraser Subject: Re: [PATCH 1 of 3] fix S3 suspend error Date: Mon, 23 Feb 2009 05:15:09 -0800 Message-ID: References: <0A882F4D99BBF6449D58E61AAFD7EDD606C4EC59@pdsmsx502.ccr.corp.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <0A882F4D99BBF6449D58E61AAFD7EDD606C4EC59@pdsmsx502.ccr.corp.intel.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: "Tian, Kevin" , "Lu, Guanqun" , "xen-devel@lists.xensource.com" List-Id: xen-devel@lists.xenproject.org On 23/02/2009 04:46, "Tian, Kevin" wrote: > I can't ensure my understanding exactly matching latest Linux > S3 implementation which evolves quickly. But generally device > drivers typically are allowed with two callbacks, one being invoked > with irq enabled and the other with irq disabled, since some actions > have such requirement. On the other side, system device class, > such as ioapic, lapic, etc. are saved with irq disabled after other > external devices. > > Xen owns a few devices with most as system class, so for > simplicity interrupt is disabled from start. It's not worthy of touching > this flow for this simple spinlock issue, imo. :-) Can you point out any core devices in modern Linux 2.6 which implement the suspend_late() or *_noirq() callbacks (those appear to be the ones handled in device_power_down() with IRQs disabled)? They seem very obsolete/legacy, and looks to me like we could unconditionally *not* disable IRQs rather than unconditionally disabling them. -- Keir