[BUG] Interesting race between cpufreq_ondemand and snd_atiixp

[BUG] Interesting race between cpufreq_ondemand and snd_atiixp

[S.Çağlar Onur]

[-- Attachment #1: Type: text/plain, Size: 2840 bytes --]

Hi;

One of our colleagues found following problem with his old laptop while 
testing Linus's latest git with external alsa-driver (v1.0.14). And we can 
also reproduce same problem with 2.6.18.8 so it seems not a new regression 
(if it is a regression).

As a summary "sound stops to work if cpufreq_ondemand governor is used" on 
that laptop. Problem occurs only if cpufreq_* modules are loaded and %100 
reproducable if system configured for ondemand governor. 

Here are the steps to reproduce the problem on that laptop;

1. Boot system with cpufreq_* modules blacklisted
   strace -o log -f -tttTTT mpg123 some.mp3

works without a problem [full log @ 1]

2. modprobe cpufreq_ondemand
   echo "ondemand" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
   strace -o log -f -tttTTT mpg123 some.mp3

hangs and no sound at all [full log @ 2]

3. echo "ondemand" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
   strace -o log -f -tttTTT mpg123 some.mp3

works without a problem [full log @ 3]
	
And with both 2.6.18.8 and 2.6.22-rc4 KDE (artsd) starts to gave "Operation 
not denied" errors on login, and after that error occured stracing mpg123 
from console gaves following output (the very same check of semctl fails with 
EINTR with second) [full log @ 4];
 
...
4004  1181948891.899907 semget(5678293, 1, IPC_CREAT|0660) = 1081344 
<0.000010>
4004  1181948891.899945 semctl(1081344, 0, IPC_64|IPC_STAT, 0xbfd1b498) = 0 
<0.000009>
4004  1181948891.899981 semctl(1081344, 0, IPC_64|IPC_SET, 0xbfd1b498) = -1 
EPERM (Operation not permitted) <0.000009>
4004  1181948891.900026 semop(1081344, 0xbfd1b6a0, 2) = -1 EINTR (Interrupted 
system call) <15.979441>
4004  1181948907.879574 --- SIGTERM (Terminated) @ 0 (0) ---
4004  1181948907.879782 +++ killed by SIGTERM +++

You can also find dmesg [5], lspci -vv [6], lsmod [7], /proc/cpuinfo 
[8], /proc/interrupts [9] and used .config [10] from there.

I have also a physical access to that laptop so please just ask if you want 
anything else.

[1] http://cekirdek.pardus.org.tr/~caglar/kernel/working
[2] http://cekirdek.pardus.org.tr/~caglar/kernel/notworking
[3] http://cekirdek.pardus.org.tr/~caglar/kernel/againworking
[4] http://cekirdek.pardus.org.tr/~caglar/kernel/perm
[5] http://cekirdek.pardus.org.tr/~caglar/kernel/dmesg_2.6.22
[6] http://cekirdek.pardus.org.tr/~caglar/kernel/lspci
[7] http://cekirdek.pardus.org.tr/~caglar/kernel/lsmod_2.6.22
[8] http://cekirdek.pardus.org.tr/~caglar/kernel/cpuinfo
[9] http://cekirdek.pardus.org.tr/~caglar/kernel/interrupts
[10] http://cekirdek.pardus.org.tr/~caglar/kernel/config

Cheers
-- 
S.Çağlar Onur <caglar@pardus.org.tr>
http://cekirdek.pardus.org.tr/~caglar/

Linux is like living in a teepee. No Windows, no Gates and an Apache in house!

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [BUG] Interesting race between cpufreq_ondemand and snd_atiixp

[S.Çağlar Onur]

[-- Attachment #1: Type: text/plain, Size: 460 bytes --]

16 Haz 2007 Cts tarihinde, S.Çağlar Onur şunları yazmıştı: 
...
> 3. echo "ondemand" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
		^^^ performance

Sorry for typo!

>    strace -o log -f -tttTTT mpg123 some.mp3
>
> works without a problem [full log @ 3]

Cheers
-- 
S.Çağlar Onur <caglar@pardus.org.tr>
http://cekirdek.pardus.org.tr/~caglar/

Linux is like living in a teepee. No Windows, no Gates and an Apache in house!

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [BUG] Interesting race between cpufreq_ondemand and snd_atiixp

[Dave Jones]

On Sat, Jun 16, 2007 at 02:33:41AM +0300, S.Çağlar Onur wrote:

 > One of our colleagues found following problem with his old laptop while 
 > testing Linus's latest git with external alsa-driver (v1.0.14). And we can 
 > also reproduce same problem with 2.6.18.8 so it seems not a new regression 
 > (if it is a regression).
 > 
 > As a summary "sound stops to work if cpufreq_ondemand governor is used" on 
 > that laptop. Problem occurs only if cpufreq_* modules are loaded and %100 
 > reproducable if system configured for ondemand governor. 

I'm puzzled.  The cpuinfo shows that this cpu doesn't have speedstep, so
why acpi-cpufreq successfully loads is a mystery.

What's in/sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies  ?

Maybe the acpi implementation is faking multiple speeds using throttling
a la p4-clockmod, which would be a bit loopy, but possible I guess.

	Dave

-- 
http://www.codemonkey.org.uk

Re: [BUG] Interesting race between cpufreq_ondemand and snd_atiixp

[S.Çağlar Onur]

[-- Attachment #1: Type: text/plain, Size: 1364 bytes --]

Hi;

16 Haz 2007 Cts tarihinde, Dave Jones şunları yazmıştı: 
> On Sat, Jun 16, 2007 at 02:33:41AM +0300, S.Çağlar Onur wrote:
>  > One of our colleagues found following problem with his old laptop while
>  > testing Linus's latest git with external alsa-driver (v1.0.14). And we
>  > can also reproduce same problem with 2.6.18.8 so it seems not a new
>  > regression (if it is a regression).
>  >
>  > As a summary "sound stops to work if cpufreq_ondemand governor is used"
>  > on that laptop. Problem occurs only if cpufreq_* modules are loaded and
>  > %100 reproducable if system configured for ondemand governor.
>
> I'm puzzled.  The cpuinfo shows that this cpu doesn't have speedstep, so
> why acpi-cpufreq successfully loads is a mystery.
>
> What's in/sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
>  ?

mavi ~ # 
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
2793000 1596000

> Maybe the acpi implementation is faking multiple speeds using throttling
> a la p4-clockmod, which would be a bit loopy, but possible I guess.

modprobe p4-clockmod with 2.6.18.8 ends up with "Device or resource busy"

Cheers
-- 
S.Çağlar Onur <caglar@pardus.org.tr>
http://cekirdek.pardus.org.tr/~caglar/

Linux is like living in a teepee. No Windows, no Gates and an Apache in house!

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 189 bytes --]