Skip to contentSkip to conversation

ALSA: HDA - Check return value to reduce useless delay

PATCH In review Wang Xingchao 2 participants 6 messages Oct 24, 2012 – Oct 26, 2012 Sound

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ALSA: HDA - Check return value to reduce useless delay

For verb 705h, it's useless to read response, so use *write api would be better. If there's error after sending cmd, just try again without continue after wrong operation.Otherwise there's long time delay.

---
 sound/pci/hda/hda_codec.c |    5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
sound/pci/hda/hda_codec.c +4 -1
index 960800b..0946eca 100644
@@ -3569,6 +3569,7 @@ static unsigned int hda_set_power_state(struct hda_codec *codec,
 	hda_nid_t fg = codec->afg ? codec->afg : codec->mfg;
 	int count;
 	unsigned int state;
+	int err;
 
 	/* this delay seems necessary to avoid click noise at power-down */
 	if (power_state == AC_PWRST_D3) {
@@ -3582,9 +3583,11 @@ static unsigned int hda_set_power_state(struct hda_codec *codec,
 			codec->patch_ops.set_power_state(codec, fg,
 							 power_state);
 		else {
-			snd_hda_codec_read(codec, fg, 0,
+			err = snd_hda_codec_write(codec, fg, 0,
 					   AC_VERB_SET_POWER_STATE,
 					   power_state);
+			if (err < 0)
+				continue;
 			snd_hda_codec_set_power_to_all(codec, fg, power_state,
 						       true);
 		}
Signature · 1 line
1.7.9.5
#1 lore raw

At Wed, 24 Oct 2012 14:53:23 +0800,

Quoted 4 lines · Wang Xingchao

For verb 705h, it's useless to read response, so use *write api would be better. If there's error after sending cmd, just try again without continue after wrong operation.Otherwise there's long time delay.

Well, this is a bit sensitive part. Did you do the good test coverage for different hardware controllers and codecs, including the non-Intel ones?

Takashi

Quoted 33 lines
---
 sound/pci/hda/hda_codec.c |    5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
sound/pci/hda/hda_codec.c +4 -1
index 960800b..0946eca 100644
@@ -3569,6 +3569,7 @@ static unsigned int hda_set_power_state(struct hda_codec *codec,
 	hda_nid_t fg = codec->afg ? codec->afg : codec->mfg;
 	int count;
 	unsigned int state;
+	int err;
 
 	/* this delay seems necessary to avoid click noise at power-down */
 	if (power_state == AC_PWRST_D3) {
@@ -3582,9 +3583,11 @@ static unsigned int hda_set_power_state(struct hda_codec *codec,
 			codec->patch_ops.set_power_state(codec, fg,
 							 power_state);
 		else {
-			snd_hda_codec_read(codec, fg, 0,
+			err = snd_hda_codec_write(codec, fg, 0,
 					   AC_VERB_SET_POWER_STATE,
 					   power_state);
+			if (err < 0)
+				continue;
 			snd_hda_codec_set_power_to_all(codec, fg, power_state,
 						       true);
 		}
Signature · 1 line
1.7.9.5
#2 lore raw
Quoted 12 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, October 25, 2012 6:01 AM
To: Wang, Xingchao
Cc: [email protected]
Subject: Re: [PATCH] ALSA: HDA - Check return value to reduce useless delay

At Wed, 24 Oct 2012 14:53:23 +0800,

Quoted 1 line · Wang Xingchao
Quoted 7 lines
Quoted 3 lines

For verb 705h, it's useless to read response, so use *write api would be better. If there's error after sending cmd, just try again without continue after wrong operation.Otherwise there's long time delay.

Well, this is a bit sensitive part. Did you do the good test coverage for different hardware controllers and codecs, including the non-Intel ones?

Well I only tested the patch on Haswell/Ivybridge platform. For Haswell I meet an issue about codec suspend/resume. There's no response from codec when there's power state transition from D3 to D0, and the patch could reduced delay time. For ivybridge the codec works well, so this patch has no active behavior.
I assume the verb 705h will not work for subnodes if it failed at first for "function group".
Is there such case when set power state for the "Audio Function Group" fail but the sub-nodes could keep on setting power?

Thanks
--xingchao

Quoted 38 lines

Takashi

sound/pci/hda/hda_codec.c static unsigned int hda_set_power_state(struct

Quoted 10 lines
---
 sound/pci/hda/hda_codec.c |    5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
sound/pci/hda/hda_codec.c +0 -0
index 960800b..0946eca 100644
@@ -3569,6 +3569,7 @@ static unsigned int hda_set_power_state(struct

hda_codec *codec,

sound/pci/hda/hda_codec.c static unsigned int hda_set_power_state(struct

Quoted 8 lines
sound/pci/hda/hda_codec.c +1 -0
 	hda_nid_t fg = codec->afg ? codec->afg : codec->mfg;
 	int count;
 	unsigned int state;
+	int err;
 
 	/* this delay seems necessary to avoid click noise at power-down */
 	if (power_state == AC_PWRST_D3) {
@@ -3582,9 +3583,11 @@ static unsigned int hda_set_power_state(struct

hda_codec *codec,

Quoted 15 lines
sound/pci/hda/hda_codec.c +3 -2
 			codec->patch_ops.set_power_state(codec, fg,
 							 power_state);
 		else {
-			snd_hda_codec_read(codec, fg, 0,
+			err = snd_hda_codec_write(codec, fg, 0,
 					   AC_VERB_SET_POWER_STATE,
 					   power_state);
+			if (err < 0)
+				continue;
 			snd_hda_codec_set_power_to_all(codec, fg, power_state,
 						       true);
 		}
--

1.7.9.5

#3 lore raw

At Thu, 25 Oct 2012 10:30:35 +0000,

Quoted 17 lines · Xingchao
Quoted 17 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, October 25, 2012 6:01 AM
To: Wang, Xingchao
Cc: [email protected]
Subject: Re: [PATCH] ALSA: HDA - Check return value to reduce useless delay

At Wed, 24 Oct 2012 14:53:23 +0800,

Quoted 4 lines · Wang Xingchao

For verb 705h, it's useless to read response, so use *write api would be better. If there's error after sending cmd, just try again without continue after wrong operation.Otherwise there's long time delay.

Quoted 5 lines
Quoted 3 lines

Well, this is a bit sensitive part. Did you do the good test coverage for different hardware controllers and codecs, including the non-Intel ones?

Well I only tested the patch on Haswell/Ivybridge platform.

Both are too new :)

Quoted 2 lines

For Haswell I meet an issue about codec suspend/resume.
There's no response from codec when there's power state transition from D3 to D0, and the patch could reduced delay time.

But does it fix the issue itself?

Quoted 3 lines

For ivybridge the codec works well, so this patch has no active behavior.
I assume the verb 705h will not work for subnodes if it failed at first for "function group".
Is there such case when set power state for the "Audio Function Group" fail but the sub-nodes could keep on setting power?

Not sure. But we need more testing obviously before breaking anything.

thanks,

Takashi

Quoted 43 lines

Thanks
--xingchao

Quoted 38 lines

Takashi

sound/pci/hda/hda_codec.c static unsigned int hda_set_power_state(struct

Quoted 10 lines
---
 sound/pci/hda/hda_codec.c |    5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
sound/pci/hda/hda_codec.c +0 -0
index 960800b..0946eca 100644
@@ -3569,6 +3569,7 @@ static unsigned int hda_set_power_state(struct

hda_codec *codec,

sound/pci/hda/hda_codec.c static unsigned int hda_set_power_state(struct

Quoted 8 lines
sound/pci/hda/hda_codec.c +1 -0
 	hda_nid_t fg = codec->afg ? codec->afg : codec->mfg;
 	int count;
 	unsigned int state;
+	int err;
 
 	/* this delay seems necessary to avoid click noise at power-down */
 	if (power_state == AC_PWRST_D3) {
@@ -3582,9 +3583,11 @@ static unsigned int hda_set_power_state(struct

hda_codec *codec,

Quoted 15 lines
sound/pci/hda/hda_codec.c +3 -2
 			codec->patch_ops.set_power_state(codec, fg,
 							 power_state);
 		else {
-			snd_hda_codec_read(codec, fg, 0,
+			err = snd_hda_codec_write(codec, fg, 0,
 					   AC_VERB_SET_POWER_STATE,
 					   power_state);
+			if (err < 0)
+				continue;
 			snd_hda_codec_set_power_to_all(codec, fg, power_state,
 						       true);
 		}
--

1.7.9.5

#4 lore raw
Quoted 37 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, October 25, 2012 7:08 PM
To: Wang, Xingchao
Cc: [email protected]
Subject: Re: [PATCH] ALSA: HDA - Check return value to reduce useless delay

At Thu, 25 Oct 2012 10:30:35 +0000,

Quoted 11 lines · Xingchao
Quoted 11 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, October 25, 2012 6:01 AM
To: Wang, Xingchao
Cc: [email protected]
Subject: Re: [PATCH] ALSA: HDA - Check return value to reduce useless delay

Quoted 6 lines
Quoted 6 lines

At Wed, 24 Oct 2012 14:53:23 +0800,

Quoted 4 lines · Wang Xingchao

For verb 705h, it's useless to read response, so use *write api would be better. If there's error after sending cmd, just try again without continue after wrong operation.Otherwise there's long

time delay.

Quoted 3 lines
Quoted 3 lines

Well, this is a bit sensitive part. Did you do the good test coverage for different hardware controllers and codecs, including the

non-Intel ones?

Quoted 3 lines
Quoted 1 line

Well I only tested the patch on Haswell/Ivybridge platform.

Quoted 7 lines

Both are too new :)

Quoted 2 lines

For Haswell I meet an issue about codec suspend/resume. There's no response from codec when there's power state transition from D3

to D0, and the patch could reduced delay time.

But does it fix the issue itself?

Actually not. :)

Quoted 8 lines
Quoted 3 lines

For ivybridge the codec works well, so this patch has no active behavior. I assume the verb 705h will not work for subnodes if it failed at first for

"function group".

Quoted 1 line

Is there such case when set power state for the "Audio Function Group" fail

but the sub-nodes could keep on setting power?

Not sure. But we need more testing obviously before breaking anything.

Okay, I will try to test it on other platforms. :)

--xingchao

Quoted 52 lines

thanks,

Takashi

Quoted 19 lines

Thanks
--xingchao

Quoted 15 lines

Takashi

Quoted 11 lines
---
 sound/pci/hda/hda_codec.c |    5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
sound/pci/hda/hda_codec.c +0 -0
index 960800b..0946eca 100644
@@ -3569,6 +3569,7 @@ static unsigned int

hda_set_power_state(struct

hda_codec *codec,

Quoted 7 lines
Quoted 7 lines

sound/pci/hda/hda_codec.c static unsigned int hda_set_power_state(struct hda_codec *codec,

Quoted 7 lines
sound/pci/hda/hda_codec.c +1 -0
 	hda_nid_t fg = codec->afg ? codec->afg : codec->mfg;
 	int count;
 	unsigned int state;
+	int err;
 
 	/* this delay seems necessary to avoid click noise at power-down */
 	if (power_state == AC_PWRST_D3) { @@ -3582,9 +3583,11 @@

static

Quoted 18 lines
Quoted 17 lines
Quoted 1 line

unsigned int hda_set_power_state(struct

hda_codec *codec,

Quoted 15 lines
sound/pci/hda/hda_codec.c +3 -2
 			codec->patch_ops.set_power_state(codec, fg,
 							 power_state);
 		else {
-			snd_hda_codec_read(codec, fg, 0,
+			err = snd_hda_codec_write(codec, fg, 0,
 					   AC_VERB_SET_POWER_STATE,
 					   power_state);
+			if (err < 0)
+				continue;
 			snd_hda_codec_set_power_to_all(codec, fg, power_state,
 						       true);
 		}
--

1.7.9.5

#5 lore raw

At Fri, 26 Oct 2012 10:45:33 +0000,

Quoted 50 lines · Xingchao
Quoted 38 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, October 25, 2012 7:08 PM
To: Wang, Xingchao
Cc: [email protected]
Subject: Re: [PATCH] ALSA: HDA - Check return value to reduce useless delay

At Thu, 25 Oct 2012 10:30:35 +0000,

Quoted 11 lines · Xingchao
Quoted 11 lines

-----Original Message-----
From: Takashi Iwai [mailto:[email protected]]
Sent: Thursday, October 25, 2012 6:01 AM
To: Wang, Xingchao
Cc: [email protected]
Subject: Re: [PATCH] ALSA: HDA - Check return value to reduce useless delay

Quoted 6 lines
Quoted 6 lines

At Wed, 24 Oct 2012 14:53:23 +0800,

Quoted 4 lines · Wang Xingchao

For verb 705h, it's useless to read response, so use *write api would be better. If there's error after sending cmd, just try again without continue after wrong operation.Otherwise there's long

time delay.

Quoted 3 lines
Quoted 3 lines

Well, this is a bit sensitive part. Did you do the good test coverage for different hardware controllers and codecs, including the

non-Intel ones?

Quoted 3 lines
Quoted 1 line

Well I only tested the patch on Haswell/Ivybridge platform.

Quoted 7 lines

Both are too new :)

Quoted 2 lines

For Haswell I meet an issue about codec suspend/resume. There's no response from codec when there's power state transition from D3

to D0, and the patch could reduced delay time.

But does it fix the issue itself?

Actually not. :)

Quoted 3 lines
Quoted 3 lines

For ivybridge the codec works well, so this patch has no active behavior. I assume the verb 705h will not work for subnodes if it failed at first for

Quoted 7 lines
Quoted 5 lines

"function group".

Quoted 1 line

Is there such case when set power state for the "Audio Function Group" fail

but the sub-nodes could keep on setting power?

Not sure. But we need more testing obviously before breaking anything.

Okay, I will try to test it on other platforms. :)

Thanks.

But thinking this twice, I think the only case it matters is that when the codec returns a bogus error at the power up. As the original code, it continues to power up all sub nodes. But with your patch, the sub node power-up will be skipped completely.

So, a compromise option would be:

- Keep using snd_hda_codec_read() to sync with the power up of FG node; snd_hda_codec_write() is asynchronous, so this can be too fast for power up of sub nodes, and...

- Check the return value from the fg node, skip the rest and continue, except for the last try. (Maybe we should add some delay for the repeated try here, though.) And in the last loop count, forcibly continue to the power up of the rest.

In this way, most of time consumed for the broken power up case you faced should be avoided.

Takashi

Quoted 54 lines

--xingchao

Quoted 52 lines

thanks,

Takashi

Quoted 19 lines

Thanks
--xingchao

Quoted 15 lines

Takashi

Quoted 11 lines
---
 sound/pci/hda/hda_codec.c |    5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
sound/pci/hda/hda_codec.c +0 -0
index 960800b..0946eca 100644
@@ -3569,6 +3569,7 @@ static unsigned int

hda_set_power_state(struct

hda_codec *codec,

Quoted 7 lines
Quoted 7 lines

sound/pci/hda/hda_codec.c static unsigned int hda_set_power_state(struct hda_codec *codec,

Quoted 7 lines
sound/pci/hda/hda_codec.c +1 -0
 	hda_nid_t fg = codec->afg ? codec->afg : codec->mfg;
 	int count;
 	unsigned int state;
+	int err;
 
 	/* this delay seems necessary to avoid click noise at power-down */
 	if (power_state == AC_PWRST_D3) { @@ -3582,9 +3583,11 @@

static

Quoted 18 lines
Quoted 17 lines
Quoted 1 line

unsigned int hda_set_power_state(struct

hda_codec *codec,

Quoted 15 lines
sound/pci/hda/hda_codec.c +3 -2
 			codec->patch_ops.set_power_state(codec, fg,
 							 power_state);
 		else {
-			snd_hda_codec_read(codec, fg, 0,
+			err = snd_hda_codec_write(codec, fg, 0,
 					   AC_VERB_SET_POWER_STATE,
 					   power_state);
+			if (err < 0)
+				continue;
 			snd_hda_codec_set_power_to_all(codec, fg, power_state,
 						       true);
 		}
--

1.7.9.5