[PATCH] ASoC: ams-delta: fix card initalization failure

[PATCH] ASoC: ams-delta: fix card initalization failure

[Janusz Krzysztofik]

Since commit 0998d0631001288a5974afc0b2a5f568bcdecb4d, 'device-core:
Ensure drvdata = NULL when no driver is bound', the Amstrad Delta sound
card no longer initializes correctly due to drvdata reset to NULL by an
upper layer before the codec device, required for successful card setup,
is registered. Fix this by moving the codec registration bits up, before
the card is probed for.

Created and tested against linux-3.6-rc3

Signed-off-by: Janusz Krzysztofik <jkrzyszt@tis.icnet.pl>
---
 sound/soc/omap/ams-delta.c |   18 ++++++++++--------
 1 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/sound/soc/omap/ams-delta.c b/sound/soc/omap/ams-delta.c
index 7d4fa8e..270de9c 100644
--- a/sound/soc/omap/ams-delta.c
+++ b/sound/soc/omap/ams-delta.c
@@ -590,20 +590,22 @@ static int __init ams_delta_module_init(void)
 	if (!ams_delta_audio_platform_device)
 		return -ENOMEM;
 
-	platform_set_drvdata(ams_delta_audio_platform_device,
-				&ams_delta_audio_card);
-
-	ret = platform_device_add(ams_delta_audio_platform_device);
-	if (ret)
-		goto err;
-
 	/*
 	 * Codec platform device could be registered from elsewhere (board?),
 	 * but I do it here as it makes sense only if used with the card.
+	 * Moreover, it must be registered before the card is probed for,
+	 * or the card setup fails due to drvdata reset by upper layers.
 	 */
 	cx20442_platform_device =
 		platform_device_register_simple("cx20442-codec", -1, NULL, 0);
-	return 0;
+
+	platform_set_drvdata(ams_delta_audio_platform_device,
+			     &ams_delta_audio_card);
+
+	ret = platform_device_add(ams_delta_audio_platform_device);
+	if (!ret)
+		return ret;
+
 err:
 	platform_device_put(ams_delta_audio_platform_device);
 	return ret;
-- 
1.7.3.4

Re: [PATCH] ASoC: ams-delta: fix card initalization failure

[Mark Brown]

On Mon, Aug 27, 2012 at 11:28:30PM +0200, Janusz Krzysztofik wrote:

> -	platform_set_drvdata(ams_delta_audio_platform_device,
> -				&ams_delta_audio_card);
> -
> -	ret = platform_device_add(ams_delta_audio_platform_device);
> -	if (ret)
> -		goto err;

The real fix here is that you should be using platform data here, not
driver data.  Is there some reason not to do that?

Re: [alsa-devel] [PATCH] ASoC: ams-delta: fix card initalization failure

[Janusz Krzysztofik]

On Mon, 27 Aug 2012 14:38:35 Mark Brown wrote:
> On Mon, Aug 27, 2012 at 11:28:30PM +0200, Janusz Krzysztofik wrote:
> 
> > -	platform_set_drvdata(ams_delta_audio_platform_device,
> > -				&ams_delta_audio_card);
> > -
> > -	ret = platform_device_add(ams_delta_audio_platform_device);
> > -	if (ret)
> > -		goto err;
> 
> The real fix here is that you should be using platform data here, not
> driver data.  Is there some reason not to do that?

Mark,

Do you think the change you propose is suitable for the rc cycle? I'm 
trying to fix a regression in the first place. Converting the ams-delta 
asoc to a platform driver is on my todo list and I'm going to take care 
of this as soon as I have enough spare time.

Thanks,
Janusz

Re: [alsa-devel] [PATCH] ASoC: ams-delta: fix card initalization failure

[Mark Brown]

On Tue, Aug 28, 2012 at 05:13:05PM +0200, Janusz Krzysztofik wrote:
> On Mon, 27 Aug 2012 14:38:35 Mark Brown wrote:
> > On Mon, Aug 27, 2012 at 11:28:30PM +0200, Janusz Krzysztofik wrote:

> > > -	platform_set_drvdata(ams_delta_audio_platform_device,
> > > -				&ams_delta_audio_card);

> > The real fix here is that you should be using platform data here, not
> > driver data.  Is there some reason not to do that?

> Do you think the change you propose is suitable for the rc cycle? I'm 
> trying to fix a regression in the first place. Converting the ams-delta 
> asoc to a platform driver is on my todo list and I'm going to take care 
> of this as soon as I have enough spare time.

The above looks like you already have a platform driver?  All I'm
suggesting is changing the above to use platform rather than driver
data.

Re: [alsa-devel] [PATCH] ASoC: ams-delta: fix card initalization failure

[Janusz Krzysztofik]

On Tue, 28 Aug 2012 11:13:39 Mark Brown wrote:
> On Tue, Aug 28, 2012 at 05:13:05PM +0200, Janusz Krzysztofik wrote:
> > On Mon, 27 Aug 2012 14:38:35 Mark Brown wrote:
> > > On Mon, Aug 27, 2012 at 11:28:30PM +0200, Janusz Krzysztofik wrote:
> 
> > > > -	platform_set_drvdata(ams_delta_audio_platform_device,
> > > > -				&ams_delta_audio_card);
> 
> > > The real fix here is that you should be using platform data here, 
not
> > > driver data.  Is there some reason not to do that?
> 
> > Do you think the change you propose is suitable for the rc cycle? 
I'm 
> > trying to fix a regression in the first place. Converting the ams-
delta 
> > asoc to a platform driver is on my todo list and I'm going to take 
care 
> > of this as soon as I have enough spare time.
> 
> The above looks like you already have a platform driver?  All I'm
> suggesting is changing the above to use platform rather than driver
> data.

The ams-delta asoc driver doesn't use snd_soc_register_card() so far, 
but relays solely on soc_probe() doing this for it, which in turn 
expects to find a snc_soc_card structure in drvdata. How is it supposed 
to find that structure if I pass it over platform data instead? Am I 
missing something?

Thanks,
Janusz

Re: [alsa-devel] [PATCH] ASoC: ams-delta: fix card initalization failure

[Mark Brown]

On Wed, Aug 29, 2012 at 07:04:48AM +0200, Janusz Krzysztofik wrote:
> On Tue, 28 Aug 2012 11:13:39 Mark Brown wrote:

> > The above looks like you already have a platform driver?  All I'm
> > suggesting is changing the above to use platform rather than driver
> > data.

> The ams-delta asoc driver doesn't use snd_soc_register_card() so far, 
> but relays solely on soc_probe() doing this for it, which in turn 
> expects to find a snc_soc_card structure in drvdata. How is it supposed 
> to find that structure if I pass it over platform data instead? Am I 
> missing something?

s/drvdata/platdata/ in the code.  If you can't do this then just
referencing the data directly in the code would be better than this
bodge, it'd be much less fragile.

Re: [alsa-devel] [PATCH] ASoC: ams-delta: fix card initalization failure

[Janusz Krzysztofik]

Dnia piątek, 31 sierpnia 2012 14:31:04 Mark Brown pisze:
> On Wed, Aug 29, 2012 at 07:04:48AM +0200, Janusz Krzysztofik wrote:
> > On Tue, 28 Aug 2012 11:13:39 Mark Brown wrote:
> 
> > > The above looks like you already have a platform driver? 

Mark,
I should have rather answered: No, ams-delta.c is not a platform driver, 
only a module which registeres a soc-audio device and provides device 
specific data and callbacks to the soc-audio platform driver.

> > > All I'm
> > > suggesting is changing the above to use platform rather than 
driver
> > > data.
> ...
> s/drvdata/platdata/ in the code. 

Taking the above into account, I would have to apply this substitue to 
the soc-core.c, which is actually the platform driver for the ams-delta 
ASoC device, and this would break a lot of other cards.

> If you can't do this then just
> referencing the data directly in the code would be better than this
> bodge, it'd be much less fragile.

I see your point, however for now I can see no better way of referencing 
the data (of type struct snd_soc_card) then passing it to 
snd_soc_register_card(). But for this to work, I would have to register 
successfully an ams-delta specific platform device first, not the soc-
audio. This, even if still done from the sound/soc/omap/ams-delta.c, not 
from an arch board file, would require now not existing ams-delta ASoC 
platform driver probe/remove callbacks at least. I'm still not convinced 
if such modification would be acceptable in the middle of the rc cycle.

If there is a simpler, less intrusive way to do this, then sorry, I 
still can't see it.

Thanks,
Janusz

Re: [alsa-devel] [PATCH] ASoC: ams-delta: fix card initalization failure

[Mark Brown]

On Sat, Sep 01, 2012 at 11:09:18AM +0200, Janusz Krzysztofik wrote:

> I see your point, however for now I can see no better way of referencing 
> the data (of type struct snd_soc_card) then passing it to 
> snd_soc_register_card(). But for this to work, I would have to register 
> successfully an ams-delta specific platform device first, not the soc-
> audio. This, even if still done from the sound/soc/omap/ams-delta.c, not 
> from an arch board file, would require now not existing ams-delta ASoC 
> platform driver probe/remove callbacks at least. I'm still not convinced 
> if such modification would be acceptable in the middle of the rc cycle.

> If there is a simpler, less intrusive way to do this, then sorry, I 
> still can't see it.

Like I already said just make it a static variable.