Skip to contentSkip to conversation

crash trace under gdb

DISCUSSION Paul Davis 3 participants 10 messages May 9, 2003 2 messages not in the archive

lore
Share & tools
lore

Paul Davis

author
#0 lore raw

well, it appears that the lock up occurs in a call to __read_lock_failed, with the PC toward the end of __kill_fasync. it alternates between two address that are both within that function, different by 9 bytes.

ctrl-alt-sysreq couldn't provide a trace more than 1 call back. not sure why.

-------------------------------------------------------
Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara The only event dedicated to issues related to Linux enterprise solutions www.enterpriselinuxforum.com

#1 lore raw

At Fri, 9 May 2003 10:05:36 -0400,

Quoted 5 lines · Paul Davis

well, it appears that the lock up occurs in a call to __read_lock_failed, with the PC toward the end of __kill_fasync. it alternates between two address that are both within that function, different by 9 bytes.

IIRC, your machine is SMP, right?
if it's related lock, it might be related with the recent change of handling of linked pcm streams...

Quoted 3 lines

ctrl-alt-sysreq couldn't provide a trace more than 1 call back. not sure why.

how about alt-sysrq-T ?

Takashi

-------------------------------------------------------
Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara The only event dedicated to issues related to Linux enterprise solutions www.enterpriselinuxforum.com

author
#2 lore raw
Quoted 3 lines

IIRC, your machine is SMP, right?
if it's related lock, it might be related with the recent change of handling of linked pcm streams...

yes, its SMP. that's what i suspected. i can't compare anything right now, because sf.net is doing its cvs-blockout stuff. when, oh when will i remember to always archive my alsa CVS tree before updating?

Quoted 4 lines
Quoted 2 lines

ctrl-alt-sysreq couldn't provide a trace more than 1 call back. not sure why.

how about alt-sysrq-T ?

ah, right, i should have tried that. i'll see if i feel like crashing my machine again rather than working on automation line editing for ardour :)

-------------------------------------------------------
Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara The only event dedicated to issues related to Linux enterprise solutions www.enterpriselinuxforum.com

author
#3 lore raw
In reply to Takashi Iwai
Quoted 1 line

how about alt-sysrq-T ?

all this "proves" is that yes, its the lock code because we're running in interrupt context, not in process context.

so something in the code jaroslav added is broken, at least for SMP.

i just hope sf.net will let me get my hands on an older copy so i can keep working ...

-------------------------------------------------------
Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara The only event dedicated to issues related to Linux enterprise solutions www.enterpriselinuxforum.com

A message that is not in the archive (sent off-list or not yet received): [email protected]

#5 lore raw

At Fri, 09 May 2003 11:39:42 -0400,

Quoted 10 lines · Paul Davis
Quoted 2 lines

how about alt-sysrq-T ?

all this "proves" is that yes, its the lock code because we're running in interrupt context, not in process context.

so something in the code jaroslav added is broken, at least for SMP.

i just hope sf.net will let me get my hands on an older copy so i can keep working ...

you can use the CVS snapshot tarball for the old versions. the changes for pcm substreams were done from Apr. 28, so Apr. 27 should be fine for this regard.
or, the CVS repositry backup would be nice to track down the whole changes, but it's big (13MB).

ciao,

Takashi

-------------------------------------------------------
Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara The only event dedicated to issues related to Linux enterprise solutions www.enterpriselinuxforum.com

#6 lore raw

Takashi Iwai ha scritto:

Quoted 14 lines

At Fri, 09 May 2003 11:39:42 -0400,

Quoted 9 lines · Paul Davis
Quoted 1 line

how about alt-sysrq-T ?

all this "proves" is that yes, its the lock code because we're running in interrupt context, not in process context.

so something in the code jaroslav added is broken, at least for SMP.

i just hope sf.net will let me get my hands on an older copy so i can keep working ...

Quoted 5 lines

you can use the CVS snapshot tarball for the old versions. the changes for pcm substreams were done from Apr. 28, so Apr. 27 should be fine for this regard.
or, the CVS repositry backup would be nice to track down the whole changes, but it's big (13MB).

However it would be nice if Paul help us to track down the source of the bug, or give a test case for someone willing to help and with access to SMP hardware.

Theoretically the new code should work, I guess there's only a trivial typo.

Signature · 12 lines
Abramo Bagnara                       mailto:[email protected]

Opera Unica                          Phone: +39.546.656023
Via Emilia Interna, 140
48014 Castel Bolognese (RA) - Italy



-------------------------------------------------------
Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara
The only event dedicated to issues related to Linux enterprise solutions
www.enterpriselinuxforum.com
author
#7 lore raw
Quoted 3 lines

However it would be nice if Paul help us to track down the source of the bug, or give a test case for someone willing to help and with access to SMP hardware.

i'm happy to help out, but there's very little information available from the system when it happens. it also turns out (unsuprisingly) not to require gdb at all: the last time it happened, i started JACK up and within a few seconds the system was locked.

if there's something i can try (adding printk's, whatever), let me know. i have both tree in place now, so its not painful to switch back and forth.

i also note that as far as i can tell, the dependency of the hdsp driver on snd-hwdep is still not in place. i had to add:

obj-$(CONFIG_SND_HDSP) += snd-pcm.o snd-timer.o snd-page-alloc.o snd.o snd-rawmidi.o snd-hwdep.o

to alsa-driver/acore/Makefile for the Apr. 27th version, although i didn't notice a problem with this for the most recent CVS version. i grepped for hwdep in the recent tree, and couldn't see where it was added, but perhaps it was. jaroslav?

--p

-------------------------------------------------------
Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara The only event dedicated to issues related to Linux enterprise solutions www.enterpriselinuxforum.com

#8 lore raw

Paul Davis ha scritto:

Quoted 7 lines
Quoted 1 line

However it would be nice if Paul help us to track down the source of the bug, or give a test case for someone willing to help and with access to SMP hardware.

i'm happy to help out, but there's very little information available from the system when it happens. it also turns out (unsuprisingly) not to require gdb at all: the last time it happened, i started JACK up and within a few seconds the system was locked.

1) Are you using linked streams?
2) What's the complete call trace on lock (alt-sysrq-T)?

Signature · 12 lines
Abramo Bagnara                       mailto:[email protected]

Opera Unica                          Phone: +39.546.656023
Via Emilia Interna, 140
48014 Castel Bolognese (RA) - Italy



-------------------------------------------------------
Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara
The only event dedicated to issues related to Linux enterprise solutions
www.enterpriselinuxforum.com
author
#9 lore raw
Quoted 2 lines

1) Are you using linked streams?
2) What's the complete call trace on lock (alt-sysrq-T)?

1) no linked streams.
2) alt-sysrq-T doesn't do anything on my system.

ctrl-alt-sysrq-P shows just one level of trackback.

-------------------------------------------------------
Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara The only event dedicated to issues related to Linux enterprise solutions www.enterpriselinuxforum.com

A message that is not in the archive (sent off-list or not yet received): [email protected]

#11 lore raw

At Fri, 09 May 2003 13:48:19 -0400,

Quoted 10 lines · Paul Davis

i also note that as far as i can tell, the dependency of the hdsp driver on snd-hwdep is still not in place. i had to add:

obj-$(CONFIG_SND_HDSP) += snd-pcm.o snd-timer.o snd-page-alloc.o snd.o snd-rawmidi.o snd-hwdep.o

to alsa-driver/acore/Makefile for the Apr. 27th version, although i didn't notice a problem with this for the most recent CVS version. i grepped for hwdep in the recent tree, and couldn't see where it was added, but perhaps it was. jaroslav?

it was very recently, namely one minute ago ;)
thanks, now fixed.

Takashi

-------------------------------------------------------
Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara The only event dedicated to issues related to Linux enterprise solutions www.enterpriselinuxforum.com