From mboxrd@z Thu Jan 1 00:00:00 1970 From: Julien Heyman Subject: Re: [Xenomai-help] Beginner's question / testsuite / latency Date: Mon, 31 Jul 2006 22:49:08 +0200 References: <442248c90607201417m24729b7cs23a8b82b719ff1cc@domain.hid> <17613.60380.820599.627459@domain.hid> <1154355564.5015.12.camel@domain.hid> In-Reply-To: <1154355564.5015.12.camel@domain.hid> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit Content-Disposition: inline Message-Id: <200607312249.09615.bidsonux@domain.hid> List-Id: Help regarding installation and common use of Xenomai List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: xenomai@xenomai.org, rpm@xenomai.org Cc: Jan Kiszka On Monday 31 July 2006 16:19, Philippe Gerum wrote: > On Mon, 2006-07-31 at 13:39 +0200, Gilles Chanteperdrix wrote: > > Philippe Gerum wrote: > > > > I rather think, also when looking at Julien's second trace, that we > > > > have some issue with X in user-space here, probably in combination > > > > with weird VIA hardware stalling IRQ delivery for a "few" > > > > microseconds. Let's see if the irqbench gives similar results. > > > > > > The problem is that I can reproduce X-related jittery (> 2 ms in a > > > row) on one of my test boxen when dragging windows over the screen, > > > without triggering the NMI watchdog set to 100 us (and guess what, the > > > chipset in question is from VIA). > > > > IMHO there are two reasons why the NMI watchdog may not trig: > > - its code is broken (we already have an unsolved bug that shows us that > > maybe it does not work as we believe) > > - the NMI watchdog only counts busy cycles, so maybe the CPU is not busy > > during these 2 ms, it is spinning, waiting for the GPU to finish the > > work using the rep; nop instruction in order to allow power > > saving. But with interrupts off. > > > > Maybe disabling hardware acceleration in your X server configuration > > would help ? > > Yes it does. Option "NoAccel" in the Screen section fixes the latency on > the test box. Still, reinitializing X while the latency test runs causes > the latter to hang, albeit LOC is still flowing properly and the box > keeps going normally. OTOH, switching from text mode to graphic mode > does not perturbate the test anymore. Some good news from my side : I did the same on my EPIA box (added Option "NoAccel" in the Device section referring to the Trident video driver) and since then the latency test works like a charm. Zero overrun, and a stable max value (around 50 microseconds), even when I launch multiple heavy applications at the same time. Display performance has obviously taken a bit hit, but this is not my main concern for now. Thanks for the tip, there is no way I would have figured that one out by myself ;) (Jan, Gilles, thx for all of your good advice too !) It's time for me to play with Xenomai now. Regards, Julien Heyman