[PATCH] ASoC: samsung: Initialize the dma_data for secondary dai

[PATCH] ASoC: samsung: Initialize the dma_data for secondary dai

[Padmavathi Venna]

This patch initializes the dma_data for the secondary dai.
Without this patch there is crash due to NULL pointer
dereference during boot time on systems where secondary dai
is initialized.

Below is the crash log:
[    1.630000] smdk-audio-wm8894 sound.7:  wm8994-aif1 <-> 3830000.i2s mapping ok
[    1.630000] Unable to handle kernel NULL pointer dereference at virtual address 00000010
[    1.630000] pgd = c0004000
[    1.630000] [00000010] *pgd=00000000
[    1.630000] Internal error: Oops: 5 [#1] PREEMPT SMP ARM
[    1.630000] Modules linked in:
...
...
[<c0342d6c>] (dmaengine_pcm_new+0x284/0x2ac) from [<c0341ae8>] (soc_new_pcm+0x2fc/0x3f4)
[<c0341ae8>] (soc_new_pcm+0x2fc/0x3f4) from [<c0338558>] (snd_soc_register_card+0x1898/0x18d8)
[<c0338558>] (snd_soc_register_card+0x1898/0x18d8) from [<c034cc34>] (smdk_audio_probe+0xa8/0xe8)
[<c034cc34>] (smdk_audio_probe+0xa8/0xe8) from [<c0252368>] (platform_drv_probe+0x18/0x1c)
[<c0252368>] (platform_drv_probe+0x18/0x1c) from [<c02510a4>] (driver_probe_device+0xf4/0x204)
[<c02510a4>] (driver_probe_device+0xf4/0x204) from [<c0251240>] (__driver_attach+0x8c/0x90)
[<c0251240>] (__driver_attach+0x8c/0x90) from [<c024f70c>] (bus_for_each_dev+0x54/0x88)
[<c024f70c>] (bus_for_each_dev+0x54/0x88) from [<c0250724>] (bus_add_driver+0xe0/0x2b8)
[<c0250724>] (bus_add_driver+0xe0/0x2b8) from [<c0251868>] (driver_register+0x78/0xf4)
[<c0251868>] (driver_register+0x78/0xf4) from [<c00086e8>] (do_one_initcall+0x34/0x150)
[<c00086e8>] (do_one_initcall+0x34/0x150) from [<c0543c30>] (kernel_init_freeable+0x12c/0x1d0)
[<c0543c30>] (kernel_init_freeable+0x12c/0x1d0) from [<c03ebf9c>] (kernel_init+0x8/0xe8)
[<c03ebf9c>] (kernel_init+0x8/0xe8) from [<c000e978>] (ret_from_fork+0x14/0x3c)

Signed-off-by: Padmavathi Venna <padma.v@samsung.com>
---
This is made based on Mark Brown for-next branch of sound.git
This patch is based on Mark Brown below patches for
generic dma engine support for samsung audio IPs.
ASoC: samsung: Use ASoC dmaengine code where possible
ASoC: samsung: Provide helper for DMA init

 sound/soc/samsung/dmaengine.c |    3 ++-
 sound/soc/samsung/i2s.c       |    5 ++++-
 2 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/sound/soc/samsung/dmaengine.c b/sound/soc/samsung/dmaengine.c
index ad0a371..3f60479 100644
--- a/sound/soc/samsung/dmaengine.c
+++ b/sound/soc/samsung/dmaengine.c
@@ -67,7 +67,8 @@ EXPORT_SYMBOL_GPL(samsung_asoc_init_dma_data);
 int samsung_asoc_dma_platform_register(struct device *dev)
 {
 	return snd_dmaengine_pcm_register(dev, &samsung_dmaengine_pcm_config,
-					  SND_DMAENGINE_PCM_FLAG_COMPAT);
+					  SND_DMAENGINE_PCM_FLAG_COMPAT |
+				SND_DMAENGINE_PCM_FLAG_CUSTOM_CHANNEL_NAME);
 }
 EXPORT_SYMBOL_GPL(samsung_asoc_dma_platform_register);
 
diff --git a/sound/soc/samsung/i2s.c b/sound/soc/samsung/i2s.c
index 67d9fa9..0bcd7fb 100644
--- a/sound/soc/samsung/i2s.c
+++ b/sound/soc/samsung/i2s.c
@@ -946,8 +946,11 @@ static int samsung_i2s_dai_probe(struct snd_soc_dai *dai)
 	struct i2s_dai *i2s = to_info(dai);
 	struct i2s_dai *other = i2s->pri_dai ? : i2s->sec_dai;
 
-	if (other && other->clk) /* If this is probe on secondary */
+	if (other && other->clk) { /* If this is probe on secondary */
+		samsung_asoc_init_dma_data(dai, &other->sec_dai->dma_playback,
+						NULL);
 		goto probe_exit;
+	}
 
 	i2s->addr = ioremap(i2s->base, 0x100);
 	if (i2s->addr == NULL) {
-- 
1.7.4.4

Re: [PATCH] ASoC: samsung: Initialize the dma_data for secondary dai

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 819 bytes --]

On Thu, Nov 28, 2013 at 03:32:12PM +0530, Padmavathi Venna wrote:

> This is made based on Mark Brown for-next branch of sound.git
> This patch is based on Mark Brown below patches for
> generic dma engine support for samsung audio IPs.
> ASoC: samsung: Use ASoC dmaengine code where possible
> ASoC: samsung: Provide helper for DMA init

Why don't we need this fix for mainline?  The dmaengine change is just
moving the initialisation through a layer of indirection but there was
initialisation anyway.  I'm concerned that there's a bug that ought to
be fixed for v3.13 as well here.

>  	return snd_dmaengine_pcm_register(dev, &samsung_dmaengine_pcm_config,
> -					  SND_DMAENGINE_PCM_FLAG_COMPAT);
> +					  SND_DMAENGINE_PCM_FLAG_COMPAT |
> +				SND_DMAENGINE_PCM_FLAG_CUSTOM_CHANNEL_NAME);

This is already done.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH] ASoC: samsung: Initialize the dma_data for secondary dai

[Padma Venkat]

Hi Mark,

On Thu, Nov 28, 2013 at 5:08 PM, Mark Brown <broonie@kernel.org> wrote:
> On Thu, Nov 28, 2013 at 03:32:12PM +0530, Padmavathi Venna wrote:
>
>> This is made based on Mark Brown for-next branch of sound.git
>> This patch is based on Mark Brown below patches for
>> generic dma engine support for samsung audio IPs.
>> ASoC: samsung: Use ASoC dmaengine code where possible
>> ASoC: samsung: Provide helper for DMA init
>
> Why don't we need this fix for mainline?  The dmaengine change is just
> moving the initialisation through a layer of indirection but there was
> initialisation anyway.  I'm concerned that there's a bug that ought to
> be fixed for v3.13 as well here.

Yes. We need this for mainline. Not dependent on your dmaengine changes.

>
>>       return snd_dmaengine_pcm_register(dev, &samsung_dmaengine_pcm_config,
>> -                                       SND_DMAENGINE_PCM_FLAG_COMPAT);
>> +                                       SND_DMAENGINE_PCM_FLAG_COMPAT |
>> +                             SND_DMAENGINE_PCM_FLAG_CUSTOM_CHANNEL_NAME);
>
> This is already done.

Should I send another patch by removing this flag from this patch?

Thanks
Padma

Re: [PATCH] ASoC: samsung: Initialize the dma_data for secondary dai

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 1020 bytes --]

On Thu, Dec 05, 2013 at 04:33:19PM +0530, Padma Venkat wrote:
> On Thu, Nov 28, 2013 at 5:08 PM, Mark Brown <broonie@kernel.org> wrote:

> > Why don't we need this fix for mainline?  The dmaengine change is just
> > moving the initialisation through a layer of indirection but there was
> > initialisation anyway.  I'm concerned that there's a bug that ought to
> > be fixed for v3.13 as well here.

> Yes. We need this for mainline. Not dependent on your dmaengine changes.

Ah, just saw this one - I'll cook up a version of it, due to the
refactoring the change doesn't apply directly to v3.13.

> >>       return snd_dmaengine_pcm_register(dev, &samsung_dmaengine_pcm_config,
> >> -                                       SND_DMAENGINE_PCM_FLAG_COMPAT);
> >> +                                       SND_DMAENGINE_PCM_FLAG_COMPAT |
> >> +                             SND_DMAENGINE_PCM_FLAG_CUSTOM_CHANNEL_NAME);

> > This is already done.

> Should I send another patch by removing this flag from this patch?

No need.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]