On Sun, 2004-08-08 at 15:05, Fernando Pablo Lopez-Lezcano wrote: > On Sat, 2004-08-07 at 16:17, Lee Revell wrote: > > On Sat, 2004-08-07 at 18:50, John Check wrote: > > > On Saturday 07 August 2004 06:30 pm, Lee Revell wrote: > > > > I recommend disabling all hardware video acceleration when tuning a > > > > system for low latency (set Option "NoAccel" and commend out the "dri" > > > > line). This will prevent X from interacting directly with your hardware > > > > - it is complicated enough when only the kernel can access hardware. > > > > Then, once you get that working, re-enable 2D acceleration, then DRI/3D > > > > acceleration, if these do not cause problems. This will make it much > > > > easier to get the problematic video drivers fixed. > > > > > > Definitely a good strategy, but nobody is going to take a system with no > > > video acceleration seriously. > > > > Yes, of course this is not a long term solution. The point is to get > > the buggy video drivers fixed (this is open source after all), in order > > to do that we need to be able to say 'when i enable DRI with $FOO > > driver, I get xruns, and these go away when I disable it'. > > Here's one case. 2.6.8rc2-mm2 + voluntary O3, Still the same result on rc3, voluntary O4... Will test soon on rc3/O5. Anything else I can do to better debug these latency spikes? (they also probably happen in the mga and r128 drivers) -- Fernando > lsm module, kernel config > files derived from current default FC2's, very recent alsa CVS (post > 1.0.6rc2), FC2, xorg, gnome desktop, P4 laptop 1.7GHz/512M with Radeon > video chipset and Intel Corp. 82801CA/CAM AC'97 Audio Controller, > kernel_preempt=1, voluntary_preempt=3, acpi=on (sound does not work with > acpi=off), irq's left in default state[*] > > With "Option NoAccel" and an idle jack in an xterm (-p128 -n2), start > glxgears, move the window around and there are no xruns (any size > window). > > With acceleration enabled, start glxgears and with the default size move > the window around, no xruns. Enlarging the window (this is on a > 1600x1200 screen) eventually triggers tons of xruns when the window is > moved and is "large enough". > > Samples of output from jack: > > **** alsa_pcm: xrun of at least 7.470 msecs > **** alsa_pcm: xrun of at least 7.401 msecs > **** alsa_pcm: xrun of at least 7.438 msecs > delay of 5728.000 usecs exceeds estimated spare time of 2644.000; > restart ... > **** alsa_pcm: xrun of at least 7.482 msecs > **** alsa_pcm: xrun of at least 7.216 msecs > **** alsa_pcm: xrun of at least 7.476 msecs
