From mboxrd@z Thu Jan 1 00:00:00 1970 From: mark gross <640e9920@gmail.com> Subject: Re: [linux-pm] [PATCH] - race-free suspend. Was: Re: [PATCH 0/8] Suspend block api (version 8) Date: Wed, 2 Jun 2010 23:04:44 -0700 Message-ID: <20100603060444.GF11311@gvim.org> References: <20100601090023.788cabf4@notabene.brown> <201006010232.20263.rjw@sisk.pl> <20100601113309.609349fd@notabene.brown> <20100601122012.1edeaf48@notabene.brown> <20100602153235.340a7852@notabene.brown> <20100602180614.729246ea@notabene.brown> Reply-To: markgross@thegnar.org Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Content-Disposition: inline In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org To: Brian Swetland Cc: Neil Brown , Peter Zijlstra , "Paul@smtp1.linux-foundation.org" , LKML , Florian Mickler , James Bottomley , Linux PM , Thomas Gleixner , Linux OMAP Mailing List , Felipe Balbi , Alan Cox List-Id: linux-omap@vger.kernel.org On Wed, Jun 02, 2010 at 11:05:18AM -0700, Brian Swetland wrote: > On Wed, Jun 2, 2010 at 1:06 AM, Neil Brown wrote: > > On Wed, 2 Jun 2010 00:05:14 -0700 > > Arve Hj=F8nnev=E5g wrote: > >> > The user-space suspend daemon avoids losing wake-events by using > >> > fcntl(F_OWNER) to ensure it gets a signal whenever any important= wake-event > >> > is ready to be read by user-space. =A0This may involve: > >> > =A0- the one daemon processing all wake events > >> > >> Wake up events are not all processed by one daemon. > > > > Not with your current user-space code, no. =A0Are you saying that y= ou are not > > open to any significant change in the Android user-space code? =A0T= hat would > > make the situation a lot harder to resolve. >=20 > There are many wakeup events possible in a typical system -- > keypresses or other input events, network traffic, telephony events, > media events (fill audio buffer, fill video decoder buffer, etc), and > I think requiring that all wakeup event processing bottleneck through > a single userspace process is non-optimal here. Um doesn't the android framework bottleneck the user mode lock processing through the powermanager and any wake up event processing eventually has to grab a lock through this bottleneck anyway? >=20 > The current suspend-blocker proposal already involves userspace > changes (it's different than our existing wakelock interface), and > we're certainly not opposed to any/all userspace changes on principle= , > but on the other hand we're not interested in significant reworks of > userspace unless they actually improve the situation somehow. I thin= k > bottlenecking events through a central daemon would represent a step > backwards. I'm not sure its a step in any direction, but I do understand the avoidance of having to rework a lot of code. --mgross