Fwd: [pulseaudio-discuss] Distribution kernels and glitch-free (Packagers, read this!)

Dan Chen crimsunkg at yahoo.com
Mon Feb 23 06:15:05 UTC 2009


[Trimmed forwarded message for relevance]

Hi,

With regards to (1) in Lennart's e-mail below, I can easily
trigger the assertion failure[0] using the vanilla upstream
build of 2.6.29-rc5[1] on amd64. Further, like the Ubuntu
-generic kernel, the config does not use PREEMPT or HZ=1000.

With regards to (2), my timing branch has the macro
adjustments made to PA git HEAD (0.9.15). Empirical testing
has demonstrated that it performs "better" than
0.9.14-0ubuntu5, but not noticeably "less aberrantly" (for
whatever values of it that mean "no glitches or pops").

I'll work on turning on PREEMPT and HZ=1000 to test the
otherwise-identical vanilla upstream build of 2.6.29-rc5
and report further results.

Thanks,
Dan

[0]
http://git.0pointer.de/?p=pulseaudio.git;a=blob_plain;f=src/tests/alsa-time-test.c

...
15663274  15663272  15438095  4255  160     4       1       3
15673360  15673356  15448185  4699  -284    4       1       3
alsa-time-test: alsa-time-test.c:189:
main: Assertion `(unsigned) avail <= buffer_size' failed.


Codec: Conexant CX20561 (Hermosa)
Vendor Id: 0x14f15051
Subsystem Id: 0x103c30d6
Revision Id: 0x100000


[1] http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.29-rc5/


--- On Sun, 2/22/09, Luke Yelavich <themuso at ubuntu.com> wrote:

I would like to respond to Lennart about what he has said in the
below forwarded message, particularly relating to the options
that need changing to get good audio performance with glitch free
pulseaudio. I am sure there are reasons why the generic kernel
has settings other than what is listed here, so I would like to
know the kernel team's reasons why we cannot/should not change
the below mentioned settings to the suggested values.

----- Forwarded message from Lennart Poettering <lennart at poettering.net> -----
...
1) ...get your bloody kernels fixed. Enable preempt, set
   HZ to 1000. Get rid of low-quality drivers that block the
   CPU. Latencies of 210ms is *REALLY NOT NECESSARY*.

2) If you want to stick with your crap kernel, then either disable g-f
   entirely or adjust the #defines at the top of
   src/modules/alsa-sink.c and src/modules/alsa-source.c.



      




More information about the kernel-team mailing list