Skip to contentSkip to conversation

ALSA: hda - skip runtime PM if async probe is not finished

PATCH In review U. Artie Eoff 3 participants 10 messages Jul 27, 2015 – Jul 28, 2015 Sound

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 ALSA: hda - skip runtime PM if async probe is not finished

Crash can occur if runtime PM is triggered before the async probe finishes via the azx_firmware_cb when CONFIG_SND_HDA_PATCH_LOADER is defined.

Don't execute PM ops if the async probe has not completed yet.

The probe is finished when chip->running is true.

---
 sound/pci/hda/hda_intel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
sound/pci/hda/hda_intel.c +1 -1
index 735bdcb04ce8..b729b25a6ad6 100644
@@ -1023,7 +1023,7 @@ static int azx_runtime_idle(struct device *dev)
 
 	chip = card->private_data;
 	hda = container_of(chip, struct hda_intel, chip);
-	if (chip->disabled || hda->init_failed)
+	if (chip->disabled || hda->init_failed || !chip->running)
 		return 0;
 
 	if (!power_save_controller || !azx_has_pm_runtime(chip) ||
Signature · 1 line
1.9.3
#1 lore raw

PATCH 2/2 ALSA: hda - Fix race condition between HDA driver and runtime PM

Don't execute runtime suspend if HDA is not finished initializing. Otherwise, the following errors can occur in hda:

snd_hda_intel 0000:00:1b.0: CORB reset timeout#2, CORBRP = 65535 snd_hda_intel 0000:00:1b.0: no codecs initialized

Debugged and root-cause found by [email protected]

---
 sound/pci/hda/hda_intel.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
sound/pci/hda/hda_intel.c +5 -2
index b729b25a6ad6..062f2400dbc7 100644
@@ -1017,17 +1017,20 @@ static int azx_runtime_idle(struct device *dev)
 	struct snd_card *card = dev_get_drvdata(dev);
 	struct azx *chip;
 	struct hda_intel *hda;
+	struct hdac_bus *bus;
 
 	if (!card)
 		return 0;
 
 	chip = card->private_data;
 	hda = container_of(chip, struct hda_intel, chip);
-	if (chip->disabled || hda->init_failed || !chip->running)
+	bus = azx_bus(chip);
+	if (chip->disabled || hda->init_failed ||
+		(bus->chip_init && !chip->running))
 		return 0;
 
 	if (!power_save_controller || !azx_has_pm_runtime(chip) ||
-	    azx_bus(chip)->codec_powered)
+	    bus->codec_powered || !bus->chip_init)
 		return -EBUSY;
 
 	return 0;
Signature · 1 line
1.9.3
#2 lore raw

Re: [PATCH 2/2] ALSA: hda - Fix race condition between HDA driver and runtime PM

Quoted 32 lines

-----Original Message-----
From: Eoff, Ullysses A
Sent: Tuesday, July 28, 2015 5:35 AM
To: [email protected]; [email protected]
Cc: Lee, Zhuo-hao; Yang, Libin; Eoff, Ullysses A
Subject: [PATCH 2/2] ALSA: hda - Fix race condition between HDA driver and runtime PM

Don't execute runtime suspend if HDA is not finished initializing. Otherwise, the following errors can occur in hda:

snd_hda_intel 0000:00:1b.0: CORB reset timeout#2, CORBRP = 65535 snd_hda_intel 0000:00:1b.0: no codecs initialized

Debugged and root-cause found by [email protected]

---
 sound/pci/hda/hda_intel.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
sound/pci/hda/hda_intel.c +1 -0
index b729b25a6ad6..062f2400dbc7 100644
@@ -1017,17 +1017,20 @@ static int azx_runtime_idle(struct device
 *dev)
 	struct snd_card *card = dev_get_drvdata(dev);
 	struct azx *chip;
 	struct hda_intel *hda;
+	struct hdac_bus *bus;
 

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device · line 1031

Quoted 10 lines
sound/pci/hda/hda_intel.c +3 -1
 	if (!card)
 		return 0;
 
 	chip = card->private_data;
 	hda = container_of(chip, struct hda_intel, chip);
-	if (chip->disabled || hda->init_failed || !chip->running)
+	bus = azx_bus(chip);
+	if (chip->disabled || hda->init_failed ||
+		(bus->chip_init && !chip->running))
 		return 0;

If you only want not to execute runtime suspend when HDA is not finished initializing, (bus->chip_init && !chip->running) is not necessary here.

Quoted 9 lines
sound/pci/hda/hda_intel.c +1 -2
 	if (!power_save_controller || !azx_has_pm_runtime(chip) ||
-	    azx_bus(chip)->codec_powered)
+	    bus->codec_powered || !bus->chip_init)
 		return -EBUSY;
 
 	return 0;
--

1.9.3

#3 lore raw
In reply to U. Artie Eoff
Quoted 19 lines · U. Artie Eoff

Crash can occur if runtime PM is triggered before the async probe finishes via the azx_firmware_cb when CONFIG_SND_HDA_PATCH_LOADER is defined.

Don't execute PM ops if the async probe has not completed yet.

The probe is finished when chip->running is true.

---
 sound/pci/hda/hda_intel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
sound/pci/hda/hda_intel.c +0 -0
index 735bdcb04ce8..b729b25a6ad6 100644

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device *dev) · line 1027

Quoted 6 lines
sound/pci/hda/hda_intel.c +1 -1
@@ -1023,7 +1023,7 @@ static int azx_runtime_idle(struct device *dev)
 
 	chip = card->private_data;
 	hda = container_of(chip, struct hda_intel, chip);
-	if (chip->disabled || hda->init_failed)
+	if (chip->disabled || hda->init_failed || !chip->running)
 		return 0;

For !chip->running, it's better to return -EBUSY.

The disabled and init_failed flags are checked in runtime_suspend() and runtime_resume() to skip the whole procedures, too.

And I guess the check of chip->running should suffice even without the second patch. It assures that the bus is powered on.

thanks,

Takashi

#4 lore raw
Quoted 38 lines

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Takashi Iwai
Sent: Monday, July 27, 2015 10:39 PM
To: Eoff, Ullysses A
Cc: Yang, Libin; [email protected]; Lee, Zhuo-hao
Subject: Re: [alsa-devel] [PATCH 1/2] ALSA: hda - skip runtime PM if async probe is not finished

Quoted 19 lines · U. Artie Eoff

Crash can occur if runtime PM is triggered before the async probe finishes via the azx_firmware_cb when CONFIG_SND_HDA_PATCH_LOADER is defined.

Don't execute PM ops if the async probe has not completed yet.

The probe is finished when chip->running is true.

---
 sound/pci/hda/hda_intel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
sound/pci/hda/hda_intel.c +0 -0
index 735bdcb04ce8..b729b25a6ad6 100644

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device *dev) · line 1027

Quoted 6 lines
sound/pci/hda/hda_intel.c +1 -1
@@ -1023,7 +1023,7 @@ static int azx_runtime_idle(struct device *dev)
 
 	chip = card->private_data;
 	hda = container_of(chip, struct hda_intel, chip);
-	if (chip->disabled || hda->init_failed)
+	if (chip->disabled || hda->init_failed || !chip->running)
 		return 0;

For !chip->running, it's better to return -EBUSY.

Quoted 6 lines

The disabled and init_failed flags are checked in runtime_suspend() and runtime_resume() to skip the whole procedures, too.

And I guess the check of chip->running should suffice even without the second patch. It assures that the bus is powered on.

Ok, I will test your suggested change to see if it addresses both elusive race conditions and report back. It sounds like it'll be the right fix. Thanks for the feedback.

U. Artie

Quoted 8 lines

thanks,

Takashi
_______________________________________________
Alsa-devel mailing list
[email protected]
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel

#5 lore raw
In reply to Takashi Iwai
Quoted 40 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Tuesday, July 28, 2015 1:39 PM
To: Eoff, Ullysses A
Cc: [email protected]; Lee, Zhuo-hao; Yang, Libin
Subject: Re: [PATCH 1/2] ALSA: hda - skip runtime PM if async probe is not finished

Quoted 3 lines · U. Artie Eoff

Crash can occur if runtime PM is triggered before the async probe finishes via the azx_firmware_cb when

CONFIG_SND_HDA_PATCH_LOADER

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device

Quoted 16 lines

is defined.

Don't execute PM ops if the async probe has not completed yet.

The probe is finished when chip->running is true.

---
 sound/pci/hda/hda_intel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
sound/pci/hda/hda_intel.c +0 -0
index 735bdcb04ce8..b729b25a6ad6 100644
@@ -1023,7 +1023,7 @@ static int azx_runtime_idle(struct device

*dev)

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device *dev) · line 1027

Quoted 6 lines
sound/pci/hda/hda_intel.c +1 -1
 	chip = card->private_data;
 	hda = container_of(chip, struct hda_intel, chip);
-	if (chip->disabled || hda->init_failed)
+	if (chip->disabled || hda->init_failed || !chip->running)
 		return 0;

For !chip->running, it's better to return -EBUSY.

Quoted 6 lines

The disabled and init_failed flags are checked in runtime_suspend() and runtime_resume() to skip the whole procedures, too.

And I guess the check of chip->running should suffice even without the
second patch. It assures that the bus is powered on.

As, in runtime_resume will call azx_init_chip(), which may sleep, is there any race when calling azx_runtime_resume(), azx_runtime_idle is called and enter azx_runtime_suspend()? Or the ACPI subsystem will handle such situation?

Quoted 5 lines

thanks,

Takashi

#6 lore raw
Quoted 44 lines · Libin
Quoted 44 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Tuesday, July 28, 2015 1:39 PM
To: Eoff, Ullysses A
Cc: [email protected]; Lee, Zhuo-hao; Yang, Libin
Subject: Re: [PATCH 1/2] ALSA: hda - skip runtime PM if async probe is not finished

Quoted 3 lines · U. Artie Eoff

Crash can occur if runtime PM is triggered before the async probe finishes via the azx_firmware_cb when

CONFIG_SND_HDA_PATCH_LOADER

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device

Quoted 16 lines

is defined.

Don't execute PM ops if the async probe has not completed yet.

The probe is finished when chip->running is true.

---
 sound/pci/hda/hda_intel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
sound/pci/hda/hda_intel.c +0 -0
index 735bdcb04ce8..b729b25a6ad6 100644
@@ -1023,7 +1023,7 @@ static int azx_runtime_idle(struct device

*dev)

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device *dev) · line 1027

Quoted 6 lines
sound/pci/hda/hda_intel.c +1 -1
 	chip = card->private_data;
 	hda = container_of(chip, struct hda_intel, chip);
-	if (chip->disabled || hda->init_failed)
+	if (chip->disabled || hda->init_failed || !chip->running)
 		return 0;

For !chip->running, it's better to return -EBUSY.

The disabled and init_failed flags are checked in runtime_suspend() and runtime_resume() to skip the whole procedures, too.

Quoted 8 lines
Quoted 3 lines

And I guess the check of chip->running should suffice even without the
second patch. It assures that the bus is powered on.

As, in runtime_resume will call azx_init_chip(), which may sleep, is there any race when calling azx_runtime_resume(), azx_runtime_idle is called and enter azx_runtime_suspend()? Or the ACPI subsystem will handle such situation?

The reentrace is avoided in the runtime PM core side, IIRC.

Takashi

#7 lore raw
Quoted 61 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Tuesday, July 28, 2015 2:13 PM
To: Yang, Libin
Cc: Eoff, Ullysses A; [email protected]; Lee, Zhuo-hao
Subject: Re: [PATCH 1/2] ALSA: hda - skip runtime PM if async probe is not finished

Quoted 7 lines · Libin
Quoted 7 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Tuesday, July 28, 2015 1:39 PM
To: Eoff, Ullysses A
Cc: [email protected]; Lee, Zhuo-hao; Yang, Libin
Subject: Re: [PATCH 1/2] ALSA: hda - skip runtime PM if async

probe is

Quoted 12 lines
Quoted 12 lines

not finished

Quoted 3 lines · U. Artie Eoff

Crash can occur if runtime PM is triggered before the async probe finishes via the azx_firmware_cb when

CONFIG_SND_HDA_PATCH_LOADER

Quoted 4 lines

is defined.

Don't execute PM ops if the async probe has not completed yet.

Quoted 8 lines
Quoted 8 lines

Review

Quoted 8 lines

The probe is finished when chip->running is true.

---
 sound/pci/hda/hda_intel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
(quoted diff) +0 -0

b/sound/pci/hda/hda_intel.c

Quoted 21 lines
Quoted 20 lines

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device

Quoted 4 lines

index 735bdcb04ce8..b729b25a6ad6 100644

sound/pci/hda/hda_intel.c +0 -0
@@ -1023,7 +1023,7 @@ static int azx_runtime_idle(struct device

*dev)

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device *dev) · line 1027

Quoted 6 lines
sound/pci/hda/hda_intel.c +1 -1
 	chip = card->private_data;
 	hda = container_of(chip, struct hda_intel, chip);
-	if (chip->disabled || hda->init_failed)
+	if (chip->disabled || hda->init_failed || !chip->running)
 		return 0;

For !chip->running, it's better to return -EBUSY.

The disabled and init_failed flags are checked in runtime_suspend() and runtime_resume() to skip the whole procedures, too.

And I guess the check of chip->running should suffice even without the
second patch. It assures that the bus is powered on.

Quoted 6 lines
Quoted 4 lines

As, in runtime_resume will call azx_init_chip(), which may sleep, is there any race when calling azx_runtime_resume(), azx_runtime_idle is called and enter azx_runtime_suspend()? Or the ACPI subsystem will handle such situation?

The reentrace is avoided in the runtime PM core side, IIRC.

Got it. Thanks.

Quoted 3 lines

Takashi

#8 lore raw
In reply to Takashi Iwai
Quoted 38 lines

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Takashi Iwai
Sent: Monday, July 27, 2015 10:39 PM
To: Eoff, Ullysses A
Cc: Yang, Libin; [email protected]; Lee, Zhuo-hao
Subject: Re: [alsa-devel] [PATCH 1/2] ALSA: hda - skip runtime PM if async probe is not finished

Quoted 19 lines · U. Artie Eoff

Crash can occur if runtime PM is triggered before the async probe finishes via the azx_firmware_cb when CONFIG_SND_HDA_PATCH_LOADER is defined.

Don't execute PM ops if the async probe has not completed yet.

The probe is finished when chip->running is true.

---
 sound/pci/hda/hda_intel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
sound/pci/hda/hda_intel.c +0 -0
index 735bdcb04ce8..b729b25a6ad6 100644

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device *dev) · line 1027

Quoted 6 lines
sound/pci/hda/hda_intel.c +1 -1
@@ -1023,7 +1023,7 @@ static int azx_runtime_idle(struct device *dev)
 
 	chip = card->private_data;
 	hda = container_of(chip, struct hda_intel, chip);
-	if (chip->disabled || hda->init_failed)
+	if (chip->disabled || hda->init_failed || !chip->running)
 		return 0;

For !chip->running, it's better to return -EBUSY.

Quoted 6 lines

The disabled and init_failed flags are checked in runtime_suspend() and runtime_resume() to skip the whole procedures, too.

And I guess the check of chip->running should suffice even without the second patch. It assures that the bus is powered on.

Hmm... wait a second. I think we still need to check bus->chip_init as in the second patch. I just noticed that the async probe finishes (i.e. chip->running = 1) before chip init executes (i.e. before bus->chip_init is set). This allows a small window for the issue I described in my second patch to occur.

I already sent another patch to replace these two... but I didn't add the check for bus->chip_init in that patch. I'll create a v2 for that but will wait for your feedback first.

Quoted 8 lines

thanks,

Takashi
_______________________________________________
Alsa-devel mailing list
[email protected]
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel

#9 lore raw
Quoted 58 lines

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Eoff, Ullysses A
Sent: Tuesday, July 28, 2015 2:38 PM
To: Takashi Iwai
Cc: Yang, Libin; [email protected]; Lee, Zhuo-hao
Subject: Re: [alsa-devel] [PATCH 1/2] ALSA: hda - skip runtime PM if async probe is not finished

Quoted 37 lines

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Takashi Iwai
Sent: Monday, July 27, 2015 10:39 PM
To: Eoff, Ullysses A
Cc: Yang, Libin; [email protected]; Lee, Zhuo-hao
Subject: Re: [alsa-devel] [PATCH 1/2] ALSA: hda - skip runtime PM if async probe is not finished

Quoted 19 lines · U. Artie Eoff

Crash can occur if runtime PM is triggered before the async probe finishes via the azx_firmware_cb when CONFIG_SND_HDA_PATCH_LOADER is defined.

Don't execute PM ops if the async probe has not completed yet.

The probe is finished when chip->running is true.

---
 sound/pci/hda/hda_intel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
sound/pci/hda/hda_intel.c +0 -0
index 735bdcb04ce8..b729b25a6ad6 100644

sound/pci/hda/hda_intel.c static int azx_runtime_idle(struct device *dev) · line 1027

Quoted 6 lines
sound/pci/hda/hda_intel.c +1 -1
@@ -1023,7 +1023,7 @@ static int azx_runtime_idle(struct device *dev)
 
 	chip = card->private_data;
 	hda = container_of(chip, struct hda_intel, chip);
-	if (chip->disabled || hda->init_failed)
+	if (chip->disabled || hda->init_failed || !chip->running)
 		return 0;

For !chip->running, it's better to return -EBUSY.

Quoted 6 lines

The disabled and init_failed flags are checked in runtime_suspend() and runtime_resume() to skip the whole procedures, too.

And I guess the check of chip->running should suffice even without the second patch. It assures that the bus is powered on.

Hmm... wait a second. I think we still need to check bus->chip_init as in the second patch. I just noticed that the async probe finishes (i.e. chip->running = 1) before chip init executes (i.e. before bus->chip_init is set). This allows a small window for the issue I described in my second patch to occur.

Quoted 4 lines

I already sent another patch to replace these two... but I didn't add the check for bus->chip_init in that patch. I'll create a v2 for that but will wait for your feedback first.

Oh nevermind... silly me! I put prints in the wrong place when testing this theory (at top of the probe cb). Once I corrected that, I now see that chip_init=1 happens during the probe cb before running=1 gets set. I'm still quite new to this driver so bear with me ;-) Appologies for the noise.

THEREFORE, my latest patch should be good to go (pending your feedback of course :-).

Cheers,
U. Artie

Quoted 12 lines
Quoted 8 lines

thanks,

Takashi
_______________________________________________
Alsa-devel mailing list
[email protected]
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel

_______________________________________________
Alsa-devel mailing list
[email protected]
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel