Skip to contentSkip to conversation

ALSA: hda - Continue probe in work context to avoid request_module deadlock

PATCH In review Wang Xingchao 4 participants 7 messages May 23, 2013 – May 24, 2013 Sound

lore
Share & tools
lore
#0 lore raw

PATCH v2 4/4 ALSA: hda - Continue probe in work context to avoid request_module deadlock

There's deadlock when request_module(i915) in azx_probe.
It looks like:
device_lock(audio pci device) -> azx_probe -> module_request (or symbol_request) -> modprobe (userspace) -> i915 init -> drm_pci_init -> pci_register_driver -> bus_add_driver -> driver_attach -> which in turn tries all locks on pci bus, and when it tries the one on the audio device, it will deadlock.

This patch introduce a work to store remaining probe stuff, and let request_module run in safe work context.

---
 sound/pci/hda/hda_i915.c  |  13 ++++--
 sound/pci/hda/hda_intel.c | 105 +++++++++++++++++++++++++++-------------------
 2 files changed, 71 insertions(+), 47 deletions(-)
sound/pci/hda/hda_i915.c +9 -4
index 76c13d5..7547b20 100644
@@ -42,13 +42,18 @@ int hda_i915_init(void)
 {
 	int err = 0;
 
-	get_power = symbol_request(i915_request_power_well);
+	get_power = symbol_get(i915_request_power_well);
 	if (!get_power) {
-		snd_printk(KERN_WARNING "hda-i915: get_power symbol get fail\n");
-		return -ENODEV;
+		request_module("i915");
+		get_power = symbol_get(i915_request_power_well);
+		if (!get_power) {
+			snd_printk(KERN_WARNING "hda-i915: get_power symbol get fail\n");
+			return -ENODEV;
+		}
+		snd_printdd("hda-i915: get_power symbol get successful\n");
 	}
 
-	put_power = symbol_request(i915_release_power_well);
+	put_power = symbol_get(i915_release_power_well);
 	if (!put_power) {
 		symbol_put(i915_request_power_well);
 		get_power = NULL;
sound/pci/hda/hda_intel.c +62 -43

145 lines not on this page · raw message

Signature · 1 line
1.8.1.2
#1 lore raw

At Thu, 23 May 2013 09:51:07 +0800,

Quoted 28 lines · Wang Xingchao

There's deadlock when request_module(i915) in azx_probe.
It looks like:
device_lock(audio pci device) -> azx_probe -> module_request (or symbol_request) -> modprobe (userspace) -> i915 init -> drm_pci_init -> pci_register_driver -> bus_add_driver -> driver_attach -> which in turn tries all locks on pci bus, and when it tries the one on the audio device, it will deadlock.

This patch introduce a work to store remaining probe stuff, and let request_module run in safe work context.

---
 sound/pci/hda/hda_i915.c  |  13 ++++--
 sound/pci/hda/hda_intel.c | 105 +++++++++++++++++++++++++++-------------------
 2 files changed, 71 insertions(+), 47 deletions(-)
sound/pci/hda/hda_i915.c +1 -1
index 76c13d5..7547b20 100644
@@ -42,13 +42,18 @@ int hda_i915_init(void)
 {
 	int err = 0;
 
-	get_power = symbol_request(i915_request_power_well);
+	get_power = symbol_get(i915_request_power_well);

sound/pci/hda/hda_i915.c int hda_i915_init(void) · line 53

Quoted 10 lines
sound/pci/hda/hda_i915.c +7 -2
 	if (!get_power) {
-		snd_printk(KERN_WARNING "hda-i915: get_power symbol get fail\n");
-		return -ENODEV;
+		request_module("i915");
+		get_power = symbol_get(i915_request_power_well);
+		if (!get_power) {
+			snd_printk(KERN_WARNING "hda-i915: get_power symbol get fail\n");
+			return -ENODEV;
+		}
+		snd_printdd("hda-i915: get_power symbol get successful\n");

Why do you need this change?

Takashi

#2 lore raw

Hi Takashi,

Quoted 48 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, May 23, 2013 2:49 PM
To: Wang Xingchao
Cc: [email protected]; [email protected]; [email protected]; Girdwood, Liam R; Li, Jocelyn; Wang, Xingchao; Lin, Mengdong
Subject: Re: [PATCH 4/4 V2] ALSA: hda - Continue probe in work context to avoid request_module deadlock

At Thu, 23 May 2013 09:51:07 +0800,

Quoted 23 lines · Wang Xingchao

There's deadlock when request_module(i915) in azx_probe. It looks like:
device_lock(audio pci device) -> azx_probe -> module_request (or symbol_request) -> modprobe (userspace) -> i915 init -> drm_pci_init -> pci_register_driver -> bus_add_driver -> driver_attach -> which in turn tries all locks on pci bus, and when it tries the one on the audio device, it will deadlock.

This patch introduce a work to store remaining probe stuff, and let request_module run in safe work context.

---
sound/pci/hda/hda_i915.c | 13 ++++-- sound/pci/hda/hda_intel.c |
105 +++++++++++++++++++++++++++-------------------

 2 files changed, 71 insertions(+), 47 deletions(-)
sound/pci/hda/hda_i915.c index renamed from sound/pci/hda/hda_i915.c +0 -0

76c13d5..7547b20 100644

sound/pci/hda/hda_i915.c int hda_i915_init(void) { · line 46

Quoted 6 lines
sound/pci/hda/hda_i915.c +1 -2
@@ -42,13 +42,18 @@ int hda_i915_init(void)  {
 	int err = 0;
 
-	get_power = symbol_request(i915_request_power_well);
+	get_power = symbol_get(i915_request_power_well);
 	if (!get_power) {
-		snd_printk(KERN_WARNING "hda-i915: get_power symbol get

fail\n");

sound/pci/hda/hda_i915.c int hda_i915_init(void)

Quoted 5 lines
sound/pci/hda/hda_i915.c +4 -1
-		return -ENODEV;
+		request_module("i915");
+		get_power = symbol_get(i915_request_power_well);
+		if (!get_power) {
+			snd_printk(KERN_WARNING "hda-i915: get_power symbol get

fail\n");

Quoted 6 lines

sound/pci/hda/hda_i915.c int hda_i915_init(void) · line 53

Quoted 3 lines
sound/pci/hda/hda_i915.c +3 -0
+			return -ENODEV;
+		}
+		snd_printdd("hda-i915: get_power symbol get successful\n");

Why do you need this change?

symbol_request() should be the better API in such case but in my test it doesnot really load i915 module, that's why I call request_module(i915) directly here.

Please note there's parameter difference:
request_module("i915") symbol_request(i915_reauest_power_well)-->request_module("symbol:i915_request_power_well")

I donot know why the second one did not really load the module.

Thanks
--xingchao

#3 lore raw

At Thu, 23 May 2013 10:19:27 +0000,

Quoted 58 lines · Xingchao

Hi Takashi,

Quoted 48 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, May 23, 2013 2:49 PM
To: Wang Xingchao
Cc: [email protected]; [email protected]; [email protected]; Girdwood, Liam R; Li, Jocelyn; Wang, Xingchao; Lin, Mengdong
Subject: Re: [PATCH 4/4 V2] ALSA: hda - Continue probe in work context to avoid request_module deadlock

At Thu, 23 May 2013 09:51:07 +0800,

Quoted 23 lines · Wang Xingchao

There's deadlock when request_module(i915) in azx_probe. It looks like:
device_lock(audio pci device) -> azx_probe -> module_request (or symbol_request) -> modprobe (userspace) -> i915 init -> drm_pci_init -> pci_register_driver -> bus_add_driver -> driver_attach -> which in turn tries all locks on pci bus, and when it tries the one on the audio device, it will deadlock.

This patch introduce a work to store remaining probe stuff, and let request_module run in safe work context.

---
sound/pci/hda/hda_i915.c | 13 ++++-- sound/pci/hda/hda_intel.c |
105 +++++++++++++++++++++++++++-------------------

 2 files changed, 71 insertions(+), 47 deletions(-)
sound/pci/hda/hda_i915.c index renamed from sound/pci/hda/hda_i915.c +0 -0

76c13d5..7547b20 100644

sound/pci/hda/hda_i915.c int hda_i915_init(void) { · line 46

Quoted 6 lines
sound/pci/hda/hda_i915.c +1 -2
@@ -42,13 +42,18 @@ int hda_i915_init(void)  {
 	int err = 0;
 
-	get_power = symbol_request(i915_request_power_well);
+	get_power = symbol_get(i915_request_power_well);
 	if (!get_power) {
-		snd_printk(KERN_WARNING "hda-i915: get_power symbol get

fail\n");

sound/pci/hda/hda_i915.c int hda_i915_init(void)

Quoted 5 lines
sound/pci/hda/hda_i915.c +4 -1
-		return -ENODEV;
+		request_module("i915");
+		get_power = symbol_get(i915_request_power_well);
+		if (!get_power) {
+			snd_printk(KERN_WARNING "hda-i915: get_power symbol get

fail\n");

Quoted 6 lines

sound/pci/hda/hda_i915.c int hda_i915_init(void) · line 53

Quoted 3 lines
sound/pci/hda/hda_i915.c +3 -0
+			return -ENODEV;
+		}
+		snd_printdd("hda-i915: get_power symbol get successful\n");

Why do you need this change?

Quoted 8 lines

symbol_request() should be the better API in such case but in my test it doesnot really load i915 module, that's why I call request_module(i915) directly here.

Please note there's parameter difference:
request_module("i915") symbol_request(i915_reauest_power_well)-->request_module("symbol:i915_request_power_well")

I donot know why the second one did not really load the module.

Well, something is really fishy. The patch can't be accepted only because it just works by moon phase...

Takashi

#4 lore raw
Quoted 84 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, May 23, 2013 6:27 PM
To: Wang, Xingchao
Cc: Wang Xingchao; [email protected]; [email protected]; [email protected]; Girdwood, Liam R; Li, Jocelyn; Lin, Mengdong
Subject: Re: [PATCH 4/4 V2] ALSA: hda - Continue probe in work context to avoid request_module deadlock

At Thu, 23 May 2013 10:19:27 +0000,

Quoted 49 lines · Xingchao

Hi Takashi,

Quoted 46 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, May 23, 2013 2:49 PM
To: Wang Xingchao
Cc: [email protected]; [email protected]; [email protected]; Girdwood, Liam R; Li, Jocelyn; Wang, Xingchao; Lin, Mengdong
Subject: Re: [PATCH 4/4 V2] ALSA: hda - Continue probe in work context to avoid request_module deadlock

At Thu, 23 May 2013 09:51:07 +0800,

Quoted 27 lines · Wang Xingchao

There's deadlock when request_module(i915) in azx_probe. It looks like:
device_lock(audio pci device) -> azx_probe -> module_request (or symbol_request) -> modprobe (userspace) -> i915 init -> drm_pci_init
-> pci_register_driver -> bus_add_driver -> driver_attach -> which -> in
turn tries all locks on pci bus, and when it tries the one on the audio device, it will deadlock.

This patch introduce a work to store remaining probe stuff, and let request_module run in safe work context.

---
sound/pci/hda/hda_i915.c | 13 ++++-- sound/pci/hda/hda_intel.c
|
105 +++++++++++++++++++++++++++-------------------

 2 files changed, 71 insertions(+), 47 deletions(-)
sound/pci/hda/hda_i915.c +0 -0

index
76c13d5..7547b20 100644

sound/pci/hda/hda_i915.c int hda_i915_init(void) { · line 46

Quoted 6 lines
sound/pci/hda/hda_i915.c +1 -2
@@ -42,13 +42,18 @@ int hda_i915_init(void)  {
 	int err = 0;
 
-	get_power = symbol_request(i915_request_power_well);
+	get_power = symbol_get(i915_request_power_well);
 	if (!get_power) {
-		snd_printk(KERN_WARNING "hda-i915: get_power symbol get

fail\n");

Quoted 5 lines
Quoted 5 lines

sound/pci/hda/hda_i915.c int hda_i915_init(void)

Quoted 5 lines
sound/pci/hda/hda_i915.c +4 -1
-		return -ENODEV;
+		request_module("i915");
+		get_power = symbol_get(i915_request_power_well);
+		if (!get_power) {
+			snd_printk(KERN_WARNING "hda-i915: get_power symbol

get

Quoted 10 lines
Quoted 7 lines

fail\n");

sound/pci/hda/hda_i915.c int hda_i915_init(void) · line 53

Quoted 3 lines
sound/pci/hda/hda_i915.c +3 -0
+			return -ENODEV;
+		}
+		snd_printdd("hda-i915: get_power symbol get successful\n");

Why do you need this change?

symbol_request() should be the better API in such case but in my test it doesnot really load i915 module, that's why I call request_module(i915)

directly here.

Quoted 4 lines

Please note there's parameter difference:
request_module("i915")

Quoted 8 lines

symbol_request(i915_reauest_power_well)-->request_module("symbol:i915_

Quoted 3 lines

request_power_well")

I donot know why the second one did not really load the module.

Well, something is really fishy. The patch can't be accepted only because it just works by moon phase...

I will continue to figure out the reason it doesnot work. :(

--xingchao

#5 lore raw
Quoted 85 lines · Xingchao
Quoted 85 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, May 23, 2013 6:27 PM
To: Wang, Xingchao
Cc: Wang Xingchao; [email protected]; [email protected]; [email protected]; Girdwood, Liam R; Li, Jocelyn; Lin, Mengdong
Subject: Re: [PATCH 4/4 V2] ALSA: hda - Continue probe in work context to avoid request_module deadlock

At Thu, 23 May 2013 10:19:27 +0000,

Quoted 49 lines · Xingchao

Hi Takashi,

Quoted 46 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, May 23, 2013 2:49 PM
To: Wang Xingchao
Cc: [email protected]; [email protected]; [email protected]; Girdwood, Liam R; Li, Jocelyn; Wang, Xingchao; Lin, Mengdong
Subject: Re: [PATCH 4/4 V2] ALSA: hda - Continue probe in work context to avoid request_module deadlock

At Thu, 23 May 2013 09:51:07 +0800,

Quoted 27 lines · Wang Xingchao

There's deadlock when request_module(i915) in azx_probe. It looks like:
device_lock(audio pci device) -> azx_probe -> module_request (or symbol_request) -> modprobe (userspace) -> i915 init -> drm_pci_init
-> pci_register_driver -> bus_add_driver -> driver_attach -> which -> in
turn tries all locks on pci bus, and when it tries the one on the audio device, it will deadlock.

This patch introduce a work to store remaining probe stuff, and let request_module run in safe work context.

---
sound/pci/hda/hda_i915.c | 13 ++++-- sound/pci/hda/hda_intel.c
|
105 +++++++++++++++++++++++++++-------------------

  2 files changed, 71 insertions(+), 47 deletions(-)
sound/pci/hda/hda_i915.c +0 -0

index
76c13d5..7547b20 100644

sound/pci/hda/hda_i915.c int hda_i915_init(void) { · line 46

Quoted 6 lines
sound/pci/hda/hda_i915.c +1 -2
@@ -42,13 +42,18 @@ int hda_i915_init(void)  {
  	int err = 0;
 
-	get_power = symbol_request(i915_request_power_well);
+	get_power = symbol_get(i915_request_power_well);
  	if (!get_power) {
-		snd_printk(KERN_WARNING "hda-i915: get_power symbol get

fail\n");

Quoted 5 lines
Quoted 5 lines

sound/pci/hda/hda_i915.c int hda_i915_init(void)

Quoted 5 lines
sound/pci/hda/hda_i915.c +4 -1
-		return -ENODEV;
+		request_module("i915");
+		get_power = symbol_get(i915_request_power_well);
+		if (!get_power) {
+			snd_printk(KERN_WARNING "hda-i915: get_power symbol

get

Quoted 10 lines
Quoted 7 lines

fail\n");

sound/pci/hda/hda_i915.c int hda_i915_init(void) · line 53

Quoted 3 lines
sound/pci/hda/hda_i915.c +3 -0
+			return -ENODEV;
+		}
+		snd_printdd("hda-i915: get_power symbol get successful\n");

Why do you need this change?

symbol_request() should be the better API in such case but in my test it doesnot really load i915 module, that's why I call request_module(i915)

directly here.

Quoted 4 lines

Please note there's parameter difference:
request_module("i915")

symbol_request(i915_reauest_power_well)-->request_module("symbol:i915_

Quoted 8 lines
Quoted 7 lines
Quoted 3 lines

request_power_well")

I donot know why the second one did not really load the module.

Well, something is really fishy. The patch can't be accepted only because it just works by moon phase...

I will continue to figure out the reason it doesnot work. :(

To the rescue!

I've tried to track this down, and I think this is a problem in modprobe with blacklisted modules. I've just emailed the kmod maintainer (with Wang Xingchao in cc) and asked for clarification.

Signature · 2 lines
David Henningsson, Canonical Ltd.
https://launchpad.net/~diwic
#6 lore raw

At Thu, 23 May 2013 19:41:58 +0200,

Quoted 94 lines · David Henningsson
Quoted 92 lines · Xingchao
Quoted 92 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, May 23, 2013 6:27 PM
To: Wang, Xingchao
Cc: Wang Xingchao; [email protected]; [email protected]; [email protected]; Girdwood, Liam R; Li, Jocelyn; Lin, Mengdong
Subject: Re: [PATCH 4/4 V2] ALSA: hda - Continue probe in work context to avoid request_module deadlock

At Thu, 23 May 2013 10:19:27 +0000,

Quoted 49 lines · Xingchao

Hi Takashi,

Quoted 46 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, May 23, 2013 2:49 PM
To: Wang Xingchao
Cc: [email protected]; [email protected]; [email protected]; Girdwood, Liam R; Li, Jocelyn; Wang, Xingchao; Lin, Mengdong
Subject: Re: [PATCH 4/4 V2] ALSA: hda - Continue probe in work context to avoid request_module deadlock

At Thu, 23 May 2013 09:51:07 +0800,

Quoted 27 lines · Wang Xingchao

There's deadlock when request_module(i915) in azx_probe. It looks like:
device_lock(audio pci device) -> azx_probe -> module_request (or symbol_request) -> modprobe (userspace) -> i915 init -> drm_pci_init
-> pci_register_driver -> bus_add_driver -> driver_attach -> which -> in
turn tries all locks on pci bus, and when it tries the one on the audio device, it will deadlock.

This patch introduce a work to store remaining probe stuff, and let request_module run in safe work context.

---
sound/pci/hda/hda_i915.c | 13 ++++-- sound/pci/hda/hda_intel.c
|
105 +++++++++++++++++++++++++++-------------------

  2 files changed, 71 insertions(+), 47 deletions(-)
sound/pci/hda/hda_i915.c +0 -0

index
76c13d5..7547b20 100644

sound/pci/hda/hda_i915.c int hda_i915_init(void) { · line 46

Quoted 6 lines
sound/pci/hda/hda_i915.c +1 -2
@@ -42,13 +42,18 @@ int hda_i915_init(void)  {
  	int err = 0;
 
-	get_power = symbol_request(i915_request_power_well);
+	get_power = symbol_get(i915_request_power_well);
  	if (!get_power) {
-		snd_printk(KERN_WARNING "hda-i915: get_power symbol get

fail\n");

Quoted 5 lines
Quoted 5 lines

sound/pci/hda/hda_i915.c int hda_i915_init(void)

Quoted 5 lines
sound/pci/hda/hda_i915.c +4 -1
-		return -ENODEV;
+		request_module("i915");
+		get_power = symbol_get(i915_request_power_well);
+		if (!get_power) {
+			snd_printk(KERN_WARNING "hda-i915: get_power symbol

get

Quoted 10 lines
Quoted 7 lines

fail\n");

sound/pci/hda/hda_i915.c int hda_i915_init(void) · line 53

Quoted 3 lines
sound/pci/hda/hda_i915.c +3 -0
+			return -ENODEV;
+		}
+		snd_printdd("hda-i915: get_power symbol get successful\n");

Why do you need this change?

symbol_request() should be the better API in such case but in my test it doesnot really load i915 module, that's why I call request_module(i915)

directly here.

Quoted 4 lines

Please note there's parameter difference:
request_module("i915")

symbol_request(i915_reauest_power_well)-->request_module("symbol:i915_

Quoted 3 lines

request_power_well")

I donot know why the second one did not really load the module.

Well, something is really fishy. The patch can't be accepted only because it just works by moon phase...

Quoted 7 lines
Quoted 1 line

I will continue to figure out the reason it doesnot work. :(

To the rescue!

I've tried to track this down, and I think this is a problem in modprobe with blacklisted modules. I've just emailed the kmod maintainer (with Wang Xingchao in cc) and asked for clarification.

Thanks for figuring out!

Takashi