Date: Sun, 27 Feb 2005 14:23:15 +1100 From: Peter Jeremy <PeterJeremy@optushome.com.au> To: Robert Watson <rwatson@freebsd.org> Cc: phk@freebsd.org Subject: Re: Excessive delays due to syncer kthread Message-ID: <20050227032315.GA96334@cirb503493.alcatel.com.au> In-Reply-To: <Pine.NEB.3.96L.1050226111418.62606C-100000@fledge.watson.org> References: <20050226071308.GN57256@cirb503493.alcatel.com.au> <Pine.NEB.3.96L.1050226111418.62606C-100000@fledge.watson.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 2005-Feb-26 11:24:26 +0000, Robert Watson wrote: >I don't have too much insight into the syncer (I've CC'd phk to victimize >him with more e-mail as this is an area he takes great interested in). A >couple of questions: > >(1) Have you tried turning on options PREEMPTION? I haven't yet. I will look at this but some of the mailing list comments made me think it wasn't totally reliable yet. >(2) Does the driver code run with Giant at all? Yes. It's marked INTR_MPSAFE and never grabs Giant. The only locking it needs is PROC_LOCK (and that's only for psignal(9)). >(3) Are you relying on callouts or taskqueues at all for processing? No. >running though. So using preemption and Giant-free code, we should be >able to get your driver code in kernel to run on short deadline, but >getting the syncer to behave better will be necessary to get the user code >running on short deadline. The userland code is less of a problem. The hardware can fit 3-6 frames (depending on depth) into a shared ring buffer - which gives me 60-120msec for the user code to wake up (100msec at the depth I will normally use). The problem is that the hardware can't autonomously move between entries in the ring buffer so the interrupt handler needs to re-write some device registers during the vertical blanking period (~1.6msec). Thanks for your input. -- Peter Jeremy
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050227032315.GA96334>