Skip to contentSkip to conversation

snd-hda-intel prevents machine to power off

DISCUSSION Luiz Fernando N. Capitulino 2 participants 8 messages Oct 28, 2008 – Oct 29, 2008 Sound

lore
Share & tools
lore
#0 lore raw

Hi Takashi,

If I have some process with the sound device opened in my eeepc
(say KDE's kmix) and try to power off the machine, for example with:

# echo o > /proc/sysrq-trigger

The machine will not power off.

But it does work if I remove the snd-hda-intel module before issuing
the echo above.

This problem seems to be pretty popular among distros, I could find
it reported for Mandriva, Ubuntu and Fedora.

Its workaround is also documented in a few wikis: people are doing
a 'rmmod snd-hda-intel' in the last halt stage (ie, halt script in etc).

I have tested with 2.6.27.4 and latest 2.7.28-rc2.

How I can I help you to debug this?

Thanks.

Signature · 1 line
Luiz Fernando N. Capitulino
#1 lore raw

At Tue, 28 Oct 2008 13:59:33 -0200,

Quoted 10 lines · Luiz Fernando N. Capitulino

Hi Takashi,

If I have some process with the sound device opened in my eeepc
(say KDE's kmix) and try to power off the machine, for example with:

# echo o > /proc/sysrq-trigger

The machine will not power off.

At which point does it stop?
What shows alt-sysrq-t (or w) output?

Quoted 5 lines

But it does work if I remove the snd-hda-intel module before issuing
the echo above.

This problem seems to be pretty popular among distros, I could find
it reported for Mandriva, Ubuntu and Fedora.

Hmm, I haven't heard of unfixed issues.

Quoted 3 lines

Its workaround is also documented in a few wikis: people are doing
a 'rmmod snd-hda-intel' in the last halt stage (ie, halt script in etc).

I hate wiki or forum just recommending a wrong workaround :)

thanks,

Takashi

#2 lore raw

Em Tue, 28 Oct 2008 17:07:42 +0100
Takashi Iwai <[email protected]> escreveu:

| >  Hi Takashi,
| > 
| >  If I have some process with the sound device opened in my eeepc
| > (say KDE's kmix) and try to power off the machine, for example with:
| > 
| > # echo o > /proc/sysrq-trigger
| > 
| >  The machine will not power off.
| 
| At which point does it stop?
| What shows alt-sysrq-t (or w) output?

I'm wondering how to get that kind of output because everything
is disabled at that point (display inclusive).

| >  But it does work if I remove the snd-hda-intel module before issuing
| > the echo above.
| > 
| >  This problem seems to be pretty popular among distros, I could find
| > it reported for Mandriva, Ubuntu and Fedora.
| 
| Hmm, I haven't heard of unfixed issues.

Well, these are the tickets I've found:

https://bugzilla.redhat.com/show_bug.cgi?id=444115
https://bugs.launchpad.net/ubuntu/+source/linux-source-2.6.22/+bug/126140
https://qa.mandriva.com/show_bug.cgi?id=44752

| >  Its workaround is also documented in a few wikis: people are doing
| > a 'rmmod snd-hda-intel' in the last halt stage (ie, halt script
| > in etc).
| 
| I hate wiki or forum just recommending a wrong workaround :)

Yeah.

Signature · 1 line
Luiz Fernando N. Capitulino
#3 lore raw

At Tue, 28 Oct 2008 14:19:16 -0200,

Quoted 16 lines · Luiz Fernando N. Capitulino

Em Tue, 28 Oct 2008 17:07:42 +0100
Takashi Iwai <[email protected]> escreveu:

| >  Hi Takashi,
| > 
| >  If I have some process with the sound device opened in my eeepc
| > (say KDE's kmix) and try to power off the machine, for example with:
| > 
| > # echo o > /proc/sysrq-trigger
| > 
| >  The machine will not power off.
| 
| At which point does it stop?
| What shows alt-sysrq-t (or w) output?
Quoted 2 lines

I'm wondering how to get that kind of output because everything
is disabled at that point (display inclusive).

OK, then it's hard to see.
A good news is that it goes at least fairly end point, thus likely no hanging task, etc.

Quoted 8 lines
| >  But it does work if I remove the snd-hda-intel module before issuing
| > the echo above.
| > 
| >  This problem seems to be pretty popular among distros, I could find
| > it reported for Mandriva, Ubuntu and Fedora.
| 
| Hmm, I haven't heard of unfixed issues.
Quoted 5 lines

Well, these are the tickets I've found:

https://bugzilla.redhat.com/show_bug.cgi?id=444115
https://bugs.launchpad.net/ubuntu/+source/linux-source-2.6.22/+bug/126140
https://qa.mandriva.com/show_bug.cgi?id=44752

Looking through the comments that the power-saving helps, the problem may disappear even without disabling the pci device. The power-saving doesn't involve with pci_disable_device() or changing the power-state of the controler. So, at least, the problem is either in the codec or the controller setup.

Suppose you already use the power-saving and it solves the shutdown problem, could you set power_save_controller=0 option for snd-hda-intel, and check whether the shutdown still works or not? If the shutdown doesn't work with this option, it means that azx_stop_chip() is what would be needed.

Anyway, I feel this is rather a BIOS issue, and not sure what is the best way to fix in the driver itself...

Takashi

#4 lore raw

Em Tue, 28 Oct 2008 17:51:30 +0100
Takashi Iwai <[email protected]> escreveu:

| At Tue, 28 Oct 2008 14:19:16 -0200,
| Luiz Fernando N. Capitulino wrote:
| > 
| > Em Tue, 28 Oct 2008 17:07:42 +0100
| > Takashi Iwai <[email protected]> escreveu:
| > 
| > | >  Hi Takashi,
| > | > 
| > | >  If I have some process with the sound device opened in my eeepc
| > | > (say KDE's kmix) and try to power off the machine, for example with:
| > | > 
| > | > # echo o > /proc/sysrq-trigger
| > | > 
| > | >  The machine will not power off.
| > | 
| > | At which point does it stop?
| > | What shows alt-sysrq-t (or w) output?
| > 
| >  I'm wondering how to get that kind of output because everything
| > is disabled at that point (display inclusive).
| 
| OK, then it's hard to see.
| A good news is that it goes at least fairly end point, thus likely no
| hanging task, etc.
| 
| 
| > | >  But it does work if I remove the snd-hda-intel module before issuing
| > | > the echo above.
| > | > 
| > | >  This problem seems to be pretty popular among distros, I could find
| > | > it reported for Mandriva, Ubuntu and Fedora.
| > | 
| > | Hmm, I haven't heard of unfixed issues.
| > 
| >  Well, these are the tickets I've found:
| > 
| > https://bugzilla.redhat.com/show_bug.cgi?id=444115
| > https://bugs.launchpad.net/ubuntu/+source/linux-source-2.6.22/+bug/126140
| > https://qa.mandriva.com/show_bug.cgi?id=44752
| 
| Looking through the comments that the power-saving helps, the problem
| may disappear even without disabling the pci device.  The power-saving
| doesn't involve with pci_disable_device() or changing the power-state
| of the controler.  So, at least, the problem is either in the codec
| or the controller setup.
| 
| Suppose you already use the power-saving and it solves the shutdown
| problem, could you set power_save_controller=0 option for
| snd-hda-intel, and check whether the shutdown still works or not?
| If the shutdown doesn't work with this option, it means that
| azx_stop_chip() is what would be needed.

Yes, power_save_controller=0 doesn't work while the power-saving
one works.

Where should I put a call to azx_stop_ship()? I'm not an alsa
hacker. :)

| Anyway, I feel this is rather a BIOS issue, and not sure what is the | best way to fix in the driver itself...

I see... I'm running with the latest BIOS update, btw.

Signature · 1 line
Luiz Fernando N. Capitulino
#5 lore raw

At Tue, 28 Oct 2008 16:44:33 -0200,

Quoted 56 lines · Luiz Fernando N. Capitulino

Em Tue, 28 Oct 2008 17:51:30 +0100
Takashi Iwai <[email protected]> escreveu:

| At Tue, 28 Oct 2008 14:19:16 -0200,
| Luiz Fernando N. Capitulino wrote:
| > 
| > Em Tue, 28 Oct 2008 17:07:42 +0100
| > Takashi Iwai <[email protected]> escreveu:
| > 
| > | >  Hi Takashi,
| > | > 
| > | >  If I have some process with the sound device opened in my eeepc
| > | > (say KDE's kmix) and try to power off the machine, for example with:
| > | > 
| > | > # echo o > /proc/sysrq-trigger
| > | > 
| > | >  The machine will not power off.
| > | 
| > | At which point does it stop?
| > | What shows alt-sysrq-t (or w) output?
| > 
| >  I'm wondering how to get that kind of output because everything
| > is disabled at that point (display inclusive).
| 
| OK, then it's hard to see.
| A good news is that it goes at least fairly end point, thus likely no
| hanging task, etc.
| 
| 
| > | >  But it does work if I remove the snd-hda-intel module before issuing
| > | > the echo above.
| > | > 
| > | >  This problem seems to be pretty popular among distros, I could find
| > | > it reported for Mandriva, Ubuntu and Fedora.
| > | 
| > | Hmm, I haven't heard of unfixed issues.
| > 
| >  Well, these are the tickets I've found:
| > 
| > https://bugzilla.redhat.com/show_bug.cgi?id=444115
| > https://bugs.launchpad.net/ubuntu/+source/linux-source-2.6.22/+bug/126140
| > https://qa.mandriva.com/show_bug.cgi?id=44752
| 
| Looking through the comments that the power-saving helps, the problem
| may disappear even without disabling the pci device.  The power-saving
| doesn't involve with pci_disable_device() or changing the power-state
| of the controler.  So, at least, the problem is either in the codec
| or the controller setup.
| 
| Suppose you already use the power-saving and it solves the shutdown
| problem, could you set power_save_controller=0 option for
| snd-hda-intel, and check whether the shutdown still works or not?
| If the shutdown doesn't work with this option, it means that
| azx_stop_chip() is what would be needed.
Quoted 2 lines

Yes, power_save_controller=0 doesn't work while the power-saving
one works.

OK, then could you check which call in azx_stop_chip() is critical? You can comment out each and check the difference.

Quoted 2 lines

Where should I put a call to azx_stop_ship()? I'm not an alsa
hacker. :)

This is the problem. If it's a fast call like azx_int_clear() or azx_int_enable(), it can be relatively easily done at each time. But, it's like freeing memories, it shouldn't be done at each operation.

Takashi

#6 lore raw

At Tue, 28 Oct 2008 20:12:09 +0100,

Quoted 65 lines · I wrote:

At Tue, 28 Oct 2008 16:44:33 -0200,

Quoted 56 lines · Luiz Fernando N. Capitulino

Em Tue, 28 Oct 2008 17:51:30 +0100
Takashi Iwai <[email protected]> escreveu:

| At Tue, 28 Oct 2008 14:19:16 -0200,
| Luiz Fernando N. Capitulino wrote:
| > 
| > Em Tue, 28 Oct 2008 17:07:42 +0100
| > Takashi Iwai <[email protected]> escreveu:
| > 
| > | >  Hi Takashi,
| > | > 
| > | >  If I have some process with the sound device opened in my eeepc
| > | > (say KDE's kmix) and try to power off the machine, for example with:
| > | > 
| > | > # echo o > /proc/sysrq-trigger
| > | > 
| > | >  The machine will not power off.
| > | 
| > | At which point does it stop?
| > | What shows alt-sysrq-t (or w) output?
| > 
| >  I'm wondering how to get that kind of output because everything
| > is disabled at that point (display inclusive).
| 
| OK, then it's hard to see.
| A good news is that it goes at least fairly end point, thus likely no
| hanging task, etc.
| 
| 
| > | >  But it does work if I remove the snd-hda-intel module before issuing
| > | > the echo above.
| > | > 
| > | >  This problem seems to be pretty popular among distros, I could find
| > | > it reported for Mandriva, Ubuntu and Fedora.
| > | 
| > | Hmm, I haven't heard of unfixed issues.
| > 
| >  Well, these are the tickets I've found:
| > 
| > https://bugzilla.redhat.com/show_bug.cgi?id=444115
| > https://bugs.launchpad.net/ubuntu/+source/linux-source-2.6.22/+bug/126140
| > https://qa.mandriva.com/show_bug.cgi?id=44752
| 
| Looking through the comments that the power-saving helps, the problem
| may disappear even without disabling the pci device.  The power-saving
| doesn't involve with pci_disable_device() or changing the power-state
| of the controler.  So, at least, the problem is either in the codec
| or the controller setup.
| 
| Suppose you already use the power-saving and it solves the shutdown
| problem, could you set power_save_controller=0 option for
| snd-hda-intel, and check whether the shutdown still works or not?
| If the shutdown doesn't work with this option, it means that
| azx_stop_chip() is what would be needed.
Quoted 2 lines

Yes, power_save_controller=0 doesn't work while the power-saving
one works.

OK, then could you check which call in azx_stop_chip() is critical? You can comment out each and check the difference.

Quoted 7 lines
Quoted 2 lines

Where should I put a call to azx_stop_ship()? I'm not an alsa
hacker. :)

This is the problem. If it's a fast call like azx_int_clear() or azx_int_enable(), it can be relatively easily done at each time. But, it's like freeing memories, it shouldn't be done at each operation.

On the second thought, we may just use a reboot notifier as a workaround for this kind of problem.
Below is a quick fix. Could you give it a try?

Anyway, could you create a kernel bugzilla entry? This will help to keep records in a commit changelog greatly, why this ugly hack is needed.
You can list the relevant URLs there as references, too.

thanks,

Takashi

---

sound/pci/hda/hda_intel.c +29 -0
index 5074ea6..042a518 100644
@@ -45,6 +45,7 @@
 #include <linux/slab.h>
 #include <linux/pci.h>
 #include <linux/mutex.h>
+#include <linux/reboot.h>
 #include <sound/core.h>
 #include <sound/initval.h>
 #include "hda_codec.h"
@@ -397,6 +398,9 @@ struct azx {
 
 	/* for pending irqs */
 	struct work_struct irq_pending_work;
+
+	/* reboot notifier (for mysterious hangup problem at power-down) */
+	struct notifier_block reboot_notifier;
 };
 
 /* driver types */
@@ -1903,12 +1907,36 @@ static int azx_resume(struct pci_dev *pci)
 
 
 /*
+ * reboot notifier for hang-up problem at power-down
+ */
+static int azx_halt(struct notifier_block *nb, unsigned long event, void *buf)
+{
+	struct azx *chip = container_of(nb, struct azx, reboot_notifier);
+	azx_stop_chip(chip);
+	return NOTIFY_OK;
+}
+
+static void azx_notifier_register(struct azx *chip)
+{
+	chip->reboot_notifier.notifier_call = azx_halt;
+	register_reboot_notifier(&chip->reboot_notifier);
+}
+
+static void azx_notifier_unregister(struct azx *chip)
+{
+	if (chip->reboot_notifier.notifier_call)
+		unregister_reboot_notifier(&chip->reboot_notifier);
+}
+
+/*
  * destructor
  */
 static int azx_free(struct azx *chip)
 {
 	int i;
 
+	azx_notifier_unregister(chip);
+
 	if (chip->initialized) {
 		azx_clear_irq_pending(chip);
 		for (i = 0; i < chip->num_streams; i++)
@@ -2272,6 +2300,7 @@ static int __devinit azx_probe(struct pci_dev *pci,
 	pci_set_drvdata(pci, card);
 	chip->running = 1;
 	power_down_all_codecs(chip);
+	azx_notifier_register(chip);
 
 	dev++;
 	return err;
#7 lore raw

Em Wed, 29 Oct 2008 08:58:32 +0100
Takashi Iwai <[email protected]> escreveu:

| At Tue, 28 Oct 2008 20:12:09 +0100,
| I wrote:
| > 
| > At Tue, 28 Oct 2008 16:44:33 -0200,
| > Luiz Fernando N. Capitulino wrote:
| > > 
| > > Em Tue, 28 Oct 2008 17:51:30 +0100
| > > Takashi Iwai <[email protected]> escreveu:
| > > 
| > > | At Tue, 28 Oct 2008 14:19:16 -0200,
| > > | Luiz Fernando N. Capitulino wrote:
| > > | > 
| > > | > Em Tue, 28 Oct 2008 17:07:42 +0100
| > > | > Takashi Iwai <[email protected]> escreveu:
| > > | > 
| > > | > | >  Hi Takashi,
| > > | > | > 
| > > | > | >  If I have some process with the sound device opened in my eeepc
| > > | > | > (say KDE's kmix) and try to power off the machine, for example with:
| > > | > | > 
| > > | > | > # echo o > /proc/sysrq-trigger
| > > | > | > 
| > > | > | >  The machine will not power off.
| > > | > | 
| > > | > | At which point does it stop?
| > > | > | What shows alt-sysrq-t (or w) output?
| > > | > 
| > > | >  I'm wondering how to get that kind of output because everything
| > > | > is disabled at that point (display inclusive).
| > > | 
| > > | OK, then it's hard to see.
| > > | A good news is that it goes at least fairly end point, thus likely no
| > > | hanging task, etc.
| > > | 
| > > | 
| > > | > | >  But it does work if I remove the snd-hda-intel module before issuing
| > > | > | > the echo above.
| > > | > | > 
| > > | > | >  This problem seems to be pretty popular among distros, I could find
| > > | > | > it reported for Mandriva, Ubuntu and Fedora.
| > > | > | 
| > > | > | Hmm, I haven't heard of unfixed issues.
| > > | > 
| > > | >  Well, these are the tickets I've found:
| > > | > 
| > > | > https://bugzilla.redhat.com/show_bug.cgi?id=444115
| > > | > https://bugs.launchpad.net/ubuntu/+source/linux-source-2.6.22/+bug/126140
| > > | > https://qa.mandriva.com/show_bug.cgi?id=44752
| > > | 
| > > | Looking through the comments that the power-saving helps, the problem
| > > | may disappear even without disabling the pci device.  The power-saving
| > > | doesn't involve with pci_disable_device() or changing the power-state
| > > | of the controler.  So, at least, the problem is either in the codec
| > > | or the controller setup.
| > > | 
| > > | Suppose you already use the power-saving and it solves the shutdown
| > > | problem, could you set power_save_controller=0 option for
| > > | snd-hda-intel, and check whether the shutdown still works or not?
| > > | If the shutdown doesn't work with this option, it means that
| > > | azx_stop_chip() is what would be needed.
| > > 
| > >  Yes, power_save_controller=0 doesn't work while the power-saving
| > > one works.
| > 
| > OK, then could you check which call in azx_stop_chip() is critical?
| > You can comment out each and check the difference.
| > 
| > >  Where should I put a call to azx_stop_ship()? I'm not an alsa
| > > hacker. :)
| > 
| > This is the problem.  If it's a fast call like azx_int_clear() or
| > azx_int_enable(), it can be relatively easily done at each time.
| > But, it's like freeing memories, it shouldn't be done at each
| > operation.
| 
| On the second thought, we may just use a reboot notifier as a
| workaround for this kind of problem.
| Below is a quick fix.  Could you give it a try?

Yeah, it fixes the problem for me.

| Anyway, could you create a kernel bugzilla entry?  This will help
| to keep records in a commit changelog greatly, why this ugly hack is
| needed.
| You can list the relevant URLs there as references, too.

Here you are:

http://bugzilla.kernel.org/show_bug.cgi?id=11889

Takashi, Thank you very much for your help.

Signature · 1 line
Luiz Fernando N. Capitulino