Skip to contentSkip to conversation

cs46xx thinkpad resume fix - #bug1946

PATCH Pavel Kysilka 3 participants 7 messages Jul 14, 2007 – Jul 19, 2007 Sound

lore
Share & tools
lore
#0 lore raw

Hi,

attached patch fix resume on thinkpad notebooks with cs46xx chip. Tested with Thinkpad A21m.

Patch is opposite latest linus git 2.6.22.

Bug #1946 may be closed.

Pavel Kysilka

________ Information from NOD32 ________
This message was checked by NOD32 Antivirus System for Linux Mail Servers.
http://www.eset.com

sound/pci/cs46xx/cs46xx_lib.c +1 -0
index bef1f6d..73f66d7 100644
@@ -3715,6 +3715,7 @@ int snd_cs46xx_resume(struct pci_dev *pci)
 	chip->active_ctrl(chip, 1); /* force to on */
 
 	snd_cs46xx_chip_init(chip);
+	snd_cs46xx_start_dsp(chip); /* upload dsp code */
 
 #if 0
 	snd_cs46xx_codec_write(chip, BA0_AC97_GENERAL_PURPOSE, 

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

#1 lore raw

At Sat, 14 Jul 2007 17:55:52 +0200,

Quoted 11 lines · Pavel Kysilka
   Hi,

attached patch fix resume on thinkpad notebooks with cs46xx chip. Tested with Thinkpad A21m.

Patch is opposite latest linus git 2.6.22.

Bug #1946 may be closed.

Oh, thanks for hunting this long-standing bug!

I think, however, it cannot be applied as it is.
First, snd_cs46xx_start_dsp() calls cs46xx_dsp_load_module(), and the latter function isn't designed for the PM resume. We'd need a special resume function just for clearing area & reloading the dsp images according to the already loaded DSP modules.

Second, __devinit prefix for snd_cs46xx_start_dsp() has to be removed if it's called from the PM handler.

Takashi

#2 lore raw

At Mon, 16 Jul 2007 11:24:45 +0200,

Quoted 17 lines · I wrote:

At Sat, 14 Jul 2007 17:55:52 +0200,

Quoted 11 lines · Pavel Kysilka
   Hi,

attached patch fix resume on thinkpad notebooks with cs46xx chip. Tested with Thinkpad A21m.

Patch is opposite latest linus git 2.6.22.

Bug #1946 may be closed.

Oh, thanks for hunting this long-standing bug!

Quoted 8 lines

I think, however, it cannot be applied as it is.
First, snd_cs46xx_start_dsp() calls cs46xx_dsp_load_module(), and the latter function isn't designed for the PM resume. We'd need a special resume function just for clearing area & reloading the dsp images according to the already loaded DSP modules.

Second, __devinit prefix for snd_cs46xx_start_dsp() has to be removed if it's called from the PM handler.

Looking at the code more deeply, I suspect whether it really works when CONFIG_SND_CS46XX_NEW_DSP=y. snd_cs46xx_start_dsp() involves with many calls that add SCBs and co, which will eventually overflow.

I made a patch blindly to do the resume work in a saner way, but of course, it's totally untested. Could you check whether it works?

thanks,

Takashi

include/cs46xx.h +4 -0
diff -r 67544a207969 include/cs46xx.h
@@ -1723,6 +1723,10 @@ struct snd_cs46xx {
 	struct snd_cs46xx_pcm *playback_pcm;
 	unsigned int play_ctl;
 #endif
+
+#ifdef CONFIG_PM
+	u32 *saved_regs;
+#endif
 };
 
 int snd_cs46xx_create(struct snd_card *card,
include/cs46xx_dsp_spos.h +2 -0
diff -r 67544a207969 include/cs46xx_dsp_spos.h
@@ -107,6 +107,7 @@ struct dsp_scb_descriptor {
 	char scb_name[DSP_MAX_SCB_NAME];
 	u32 address;
 	int index;
+	u32 *data;
 
 	struct dsp_scb_descriptor * sub_list_ptr;
 	struct dsp_scb_descriptor * next_scb_ptr;
@@ -127,6 +128,7 @@ struct dsp_task_descriptor {
 	int size;
 	u32 address;
 	int index;
+	u32 *data;
 };
 
 struct dsp_pcm_channel_descriptor {
pci/cs46xx/cs46xx_lib.c +62 -15

135 lines not on this page · raw message

pci/cs46xx/cs46xx_lib.h +3 -0
diff -r 67544a207969 pci/cs46xx/cs46xx_lib.h
@@ -86,6 +86,9 @@ struct dsp_spos_instance *cs46xx_dsp_spo
 struct dsp_spos_instance *cs46xx_dsp_spos_create (struct snd_cs46xx * chip);
 void cs46xx_dsp_spos_destroy (struct snd_cs46xx * chip);
 int cs46xx_dsp_load_module (struct snd_cs46xx * chip, struct dsp_module_desc * module);
+#ifdef CONFIG_PM
+int cs46xx_dsp_resume(struct snd_cs46xx * chip);
+#endif
 struct dsp_symbol_entry *cs46xx_dsp_lookup_symbol (struct snd_cs46xx * chip, char * symbol_name,
 						   int symbol_type);
 #ifdef CONFIG_PROC_FS
pci/cs46xx/dsp_spos.c +126 -44

232 lines not on this page · raw message

#3 lore raw

On Tue Jul 17, 2007 at 12:50:21PM +0200, Takashi Iwai wrote:

Hi Takashi,

good news below.

Quoted 4 lines

At Mon, 16 Jul 2007 11:24:45 +0200,

Quoted 2 lines · I wrote:

<cut/>

Quoted 11 lines
Quoted 10 lines

Oh, thanks for hunting this long-standing bug!

I think, however, it cannot be applied as it is.
First, snd_cs46xx_start_dsp() calls cs46xx_dsp_load_module(), and the latter function isn't designed for the PM resume. We'd need a special resume function just for clearing area & reloading the dsp images according to the already loaded DSP modules.

Second, __devinit prefix for snd_cs46xx_start_dsp() has to be removed if it's called from the PM handler.

Quoted 6 lines

Looking at the code more deeply, I suspect whether it really works when CONFIG_SND_CS46XX_NEW_DSP=y. snd_cs46xx_start_dsp() involves with many calls that add SCBs and co, which will eventually overflow.

I made a patch blindly to do the resume work in a saner way, but of course, it's totally untested. Could you check whether it works?

Patch works perfectly.

Tested option:
- kernel compiled with and without CONFIG_SND_CS46XX_NEW_DSP=y
- play pcm stream, suspend, resume and pcm stream is playing
- play pcm stream, stop playing pcm stream, suspend, resume and play pcm stream

If you need log from resume from notebook, I'll send you this log.

Quoted 6 lines

thanks,

Takashi

Many thanks
Pavel

________ Information from NOD32 ________
This message was checked by NOD32 Antivirus System for Linux Mail Servers.
http://www.eset.com

#4 lore raw

Would this also apply to bug #305?

I'm happy to test this on CentOS 5 then Fedora 7 on a 600X, and write up the needed steps for ThinkWiki until the patch is included in the distributions.

Chris

Quoted 57 lines · Pavel Kysilka

On Tue Jul 17, 2007 at 12:50:21PM +0200, Takashi Iwai wrote:

Hi Takashi,

good news below.

Quoted 4 lines

At Mon, 16 Jul 2007 11:24:45 +0200,

Quoted 2 lines · I wrote:

<cut/>

Quoted 11 lines
Quoted 10 lines

Oh, thanks for hunting this long-standing bug!

I think, however, it cannot be applied as it is.
First, snd_cs46xx_start_dsp() calls cs46xx_dsp_load_module(), and the latter function isn't designed for the PM resume. We'd need a special resume function just for clearing area & reloading the dsp images according to the already loaded DSP modules.

Second, __devinit prefix for snd_cs46xx_start_dsp() has to be removed if it's called from the PM handler.

Quoted 6 lines

Looking at the code more deeply, I suspect whether it really works when CONFIG_SND_CS46XX_NEW_DSP=y. snd_cs46xx_start_dsp() involves with many calls that add SCBs and co, which will eventually overflow.

I made a patch blindly to do the resume work in a saner way, but of course, it's totally untested. Could you check whether it works?

Patch works perfectly.

Tested option:
- kernel compiled with and without CONFIG_SND_CS46XX_NEW_DSP=y
- play pcm stream, suspend, resume and pcm stream is playing
- play pcm stream, stop playing pcm stream, suspend, resume and play pcm stream

If you need log from resume from notebook, I'll send you this log.

Quoted 6 lines

thanks,

Takashi

Many thanks
Pavel

________ Information from NOD32 ________
This message was checked by NOD32 Antivirus System for Linux Mail Servers.
http://www.eset.com
_______________________________________________
Alsa-devel mailing list
[email protected]
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel

#6 lore raw
In reply to Pavel Kysilka

At Wed, 18 Jul 2007 00:26:42 +0200,

Quoted 30 lines · Pavel Kysilka

On Tue Jul 17, 2007 at 12:50:21PM +0200, Takashi Iwai wrote:

Hi Takashi,

good news below.

Quoted 4 lines

At Mon, 16 Jul 2007 11:24:45 +0200,

Quoted 2 lines · I wrote:

<cut/>

Quoted 11 lines
Quoted 10 lines

Oh, thanks for hunting this long-standing bug!

I think, however, it cannot be applied as it is.
First, snd_cs46xx_start_dsp() calls cs46xx_dsp_load_module(), and the latter function isn't designed for the PM resume. We'd need a special resume function just for clearing area & reloading the dsp images according to the already loaded DSP modules.

Second, __devinit prefix for snd_cs46xx_start_dsp() has to be removed if it's called from the PM handler.

Quoted 6 lines

Looking at the code more deeply, I suspect whether it really works when CONFIG_SND_CS46XX_NEW_DSP=y. snd_cs46xx_start_dsp() involves with many calls that add SCBs and co, which will eventually overflow.

I made a patch blindly to do the resume work in a saner way, but of course, it's totally untested. Could you check whether it works?

Quoted 8 lines

Patch works perfectly.

Tested option:
- kernel compiled with and without CONFIG_SND_CS46XX_NEW_DSP=y
- play pcm stream, suspend, resume and pcm stream is playing
- play pcm stream, stop playing pcm stream, suspend, resume and play pcm stream

If you need log from resume from notebook, I'll send you this log.

Thanks for testing! That's enough, so far. I applied the patch to HG tree now.

Takashi

#5 lore raw
In reply to [email protected]

At Wed, 18 Jul 2007 08:51:45 -0500 (CDT),

Quoted 2 lines · [email protected] wrote:

Would this also apply to bug #305?

Possible.

Quoted 3 lines

I'm happy to test this on CentOS 5 then Fedora 7 on a 600X, and write up the needed steps for ThinkWiki until the patch is included in the distributions.

You can either download the HG repo or take the latest daily snapshot from
ftp://ftp.suse.com/pub/projects/alsa/snapshot/driver

I applied the patch today (July 19), so hg20070720 snapshot will include it.

If the mirror HG repo (hg-mirror.alsa-project.org) is out of sync, you can try hg.alsa-project.org instead.

Takashi