Subject: [linux-audio-dev] Re: [Jackit-devel] new low latency patch for 2.6
From: Florian Schmidt (mista.tapas_AT_gmx.net)
Date: Mon Jul 12 2004 - 15:32:22 EEST
On Sat, 10 Jul 2004 14:15:47 +0200
Florian Schmidt <mista.tapas_AT_gmx.net> wrote:
> Hi, chiming in late. I experienced the same. When i booted into the patched kernel voountary_preemption was on. But when i tried to turn on kernel_preemption, too, it immeaditly crashed.. will try your new patch now..
Ok, no more crashing, but i don't see any substantial improvement either. Actually i'm not sure i saw any improvement. I use JACK with a periodsize of 512 frames. At a framerate of 48khz this equals a time of 512/48000secs = 0.010667secs. So basically jack has about 10msecs to produce the next period while another one is playing. But: No matter which 2.6.x kernel i use i always can trigger xruns by either
1] many window manager ops [like rapid desktop wheeling in fluxbox]
2] compiling a kernel in some other term
Mind you i talk about just jackd alone, no clients connected, etc, same thing as user and as root. As root it doesn't even make a difference if i export LD_ASSUME_KERNEL in jackd's process.. And yes, jackd runs in th SCHED_FIFO scheduling class with no other SCHED_FIFO process fighting for cpu time..
mango:~# LD_ASSUME_KERNEL=2.4.19 jackd -R -d alsa -p512
jackd 0.98.1
Copyright 2001-2003 Paul Davis and others.
jackd comes with ABSOLUTELY NO WARRANTY
This is free software, and you are welcome to redistribute it
under certain conditions; see the file COPYING for details
loading driver ..
creating alsa driver ... hw:0|hw:0|512|2|48000|0|0|nomon|swmeter|-|32bit
control device hw:0
configuring for 48000Hz, period = 512 frames, buffer = 2 periods
Couldn't open hw:0 for 32bit samples trying 24bit instead
Couldn't open hw:0 for 24bit samples trying 16bit instead
Couldn't open hw:0 for 32bit samples trying 24bit instead
Couldn't open hw:0 for 24bit samples trying 16bit instead
**** alsa_pcm: xrun of at least 0.057 msecs
**** alsa_pcm: xrun of at least 0.033 msecs
**** alsa_pcm: xrun of at least 0.117 msecs
Behviour was substantially better with 2.4.x + ll + preempt. I could compile a kernel, do other stuff AND run an ardour session with 4 tracks w/o xruns at all for hours..
So, i hope that maybe you, Ingo, know a way on how to provide more useful info for you kernel development people. I read somewhere about ALSA's xrun debugging. Which kernel options and boottime parameters do i need for this? Is there a sysctl to switch it on? What file is the log dumped, too..
I also read on lkml about some latency Instrumentation patch for the 2.6.x kernels to find out where kernel latencies are happening.. Maybe you can provide a simple howto on how to use all that stuff to provide better infos for you..
Maybe the best is to assume that we start out with a clean 2.6.7 vanilla kernel source. Please tell us which patches to apply to get useful results.
thanks for all the great kernel work btw. You guys rock! :)
Florian Schmidt
-- Palimm Palimm! http://affenbande.org/~tapas/wiki
This archive was generated by hypermail 2b28 : Wed Jul 14 2004 - 20:45:25 EEST