From mboxrd@z Thu Jan 1 00:00:00 1970 From: Matthew Garrett Date: Thu, 19 Nov 2009 14:16:34 +0000 Subject: Re: [PATCH] [RFC] Add support for uevents on block device idle Message-Id: <20091119141634.GA311@srcf.ucam.org> List-Id: References: <20091118194053.GB12944@srcf.ucam.org> <20091118195342.GA13627@srcf.ucam.org> <20091118200712.GA14026@srcf.ucam.org> <20091118213355.GA16630@srcf.ucam.org> <20091119130107.GB20949@srcf.ucam.org> In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Kay Sievers Cc: David Zeuthen , linux-kernel@vger.kernel.org, axboe@kernel.dk, linux-hotplug@vger.kernel.org On Thu, Nov 19, 2009 at 02:29:29PM +0100, Kay Sievers wrote: > On Thu, Nov 19, 2009 at 14:01, Matthew Garrett wrote: > > On Thu, Nov 19, 2009 at 12:09:30PM +0100, Kay Sievers wrote: > >> On Wed, Nov 18, 2009 at 22:33, Matthew Garrett wrote: > >> > My use cases are on the order of a second. > >> > >> Ok, what's the specific use case, which should be triggered after a > >> second? I thought you were thinking about disk spindown or similar. > > > > The first is altering ALPM policy. ALPM will be initiated by the host if > > the number of queued requests hits zero - if there's no hysteresis > > implemented, then that can result in a significant performance hit. We > > don't need /much/ hysteresis, but it's the difference between a 50% > > performance hit and not having that. > > Can't that logic live entirely in the kernel, instead of being a > rather generic userspace event interface (with the current limitation > to a single user)? It could, but it seems a bit of a hack. It'd still also require the timer to be in the kernel, so we might as well expose that to userspace. -- Matthew Garrett | mjg59@srcf.ucam.org