RE: [PATCH v2 2/5] ASoC: samsung: i2s: add support for FSD I2S

[Padmanabhan Rajanbabu]

> -----Original Message-----
> From: Krzysztof Kozlowski [mailto:krzysztof.kozlowski@xxxxxxxxxx]
> Sent: 03 January 2023 04:39 PM
> To: Padmanabhan Rajanbabu <p.rajanbabu@xxxxxxxxxxx>;
> lgirdwood@xxxxxxxxx; broonie@xxxxxxxxxx; robh+dt@xxxxxxxxxx;
> krzysztof.kozlowski+dt@xxxxxxxxxx; s.nawrocki@xxxxxxxxxxx;
> perex@xxxxxxxx; tiwai@xxxxxxxx; pankaj.dubey@xxxxxxxxxxx;
> alim.akhtar@xxxxxxxxxxx; rcsekar@xxxxxxxxxxx;
> aswani.reddy@xxxxxxxxxxx
> Cc: alsa-devel@xxxxxxxxxxxxxxxx; devicetree@xxxxxxxxxxxxxxx; linux-
> kernel@xxxxxxxxxxxxxxx; linux-samsung-soc@xxxxxxxxxxxxxxx
> Subject: Re: [PATCH v2 2/5] ASoC: samsung: i2s: add support for FSD I2S
> 
> On 03/01/2023 05:56, Padmanabhan Rajanbabu wrote:
> > Add support for enabling I2S controller on FSD platform.
> >
> > FSD I2S controller is based on Exynos7 I2S controller, supporting 2CH
> > playback/capture in I2S mode and 7.1CH playback/capture in TDM mode.
> >
> > Signed-off-by: Padmanabhan Rajanbabu <p.rajanbabu@xxxxxxxxxxx>
> > ---
> >  sound/soc/samsung/i2s-regs.h |  1 +
> >  sound/soc/samsung/i2s.c      | 57
> ++++++++++++++++++++++++++++++++++++
> >  2 files changed, 58 insertions(+)
> >
> > diff --git a/sound/soc/samsung/i2s-regs.h
> > b/sound/soc/samsung/i2s-regs.h index b4b5d6053503..4444c857d0c0
> 100644
> > --- a/sound/soc/samsung/i2s-regs.h
> > +++ b/sound/soc/samsung/i2s-regs.h
> > @@ -132,6 +132,7 @@
> >  #define EXYNOS7_MOD_RCLK_192FS	7
> >
> >  #define PSR_PSREN		(1 << 15)
> > +#define PSR_PSVAL(x)		(((x - 1) << 8) & 0x3f00)
> >
> >  #define FIC_TX2COUNT(x)		(((x) >>  24) & 0xf)
> >  #define FIC_TX1COUNT(x)		(((x) >>  16) & 0xf)
> > diff --git a/sound/soc/samsung/i2s.c b/sound/soc/samsung/i2s.c index
> > 9505200f3d11..dcb5c438cb2f 100644
> > --- a/sound/soc/samsung/i2s.c
> > +++ b/sound/soc/samsung/i2s.c
> > @@ -50,6 +50,10 @@ struct samsung_i2s_dai_data {
> >  	u32 quirks;
> >  	unsigned int pcm_rates;
> >  	const struct samsung_i2s_variant_regs *i2s_variant_regs;
> > +	void (*fixup_early)(struct snd_pcm_substream *substream,
> > +					struct snd_soc_dai *dai);
> > +	void (*fixup_late)(struct snd_pcm_substream *substream,
> > +					struct snd_soc_dai *dai);
> >  };
> >
> >  struct i2s_dai {
> > @@ -111,6 +115,10 @@ struct samsung_i2s_priv {
> >  	u32 suspend_i2spsr;
> >
> >  	const struct samsung_i2s_variant_regs *variant_regs;
> > +	void (*fixup_early)(struct snd_pcm_substream *substream,
> > +						struct snd_soc_dai *dai);
> > +	void (*fixup_late)(struct snd_pcm_substream *substream,
> > +						struct snd_soc_dai *dai);
> >  	u32 quirks;
> >
> >  	/* The clock provider's data */
> > @@ -940,6 +948,10 @@ static int i2s_trigger(struct snd_pcm_substream
> *substream,
> >  	case SNDRV_PCM_TRIGGER_RESUME:
> >  	case SNDRV_PCM_TRIGGER_PAUSE_RELEASE:
> >  		pm_runtime_get_sync(dai->dev);
> > +
> > +		if (priv->fixup_early)
> > +			priv->fixup_early(substream, dai);
> > +
> >  		spin_lock_irqsave(&priv->lock, flags);
> >
> >  		if (config_setup(i2s)) {
> > @@ -947,6 +959,13 @@ static int i2s_trigger(struct snd_pcm_substream
> *substream,
> >  			return -EINVAL;
> >  		}
> >
> 
> Except several warnings this patch generates, this is a bit surprising:
> 
> > +		spin_unlock_irqrestore(&priv->lock, flags);
> 
> You have critical section which you now break into two. You cannot do this
> usually. How the synchronization is now kept?
> 

The actual reason behind breaking the critical section is to allow the use
of already existing functions related to configuration of CDCLK and OPCLK
source.

Based on the review comments from previous patch-set related to the actual
need of custom sound card driver, we did migrate to simple-card, where
Exynos specific configurations like RCLKSRC selection, OPCLK and CDCLK
configuration is not being handled by the simple-card. To overcome this
scenario, fixups has been added during i2s_trigger to let Exynos users to
configure the Exynos I2S specific dividers and mux.

Rather than re-implementing the routines (or) configurations already
Available in the driver, the fixup functions can call the i2s_set_sysclk
and similar functions directly (which also takes the spin lock).

But we noticed that fixup_late implemented for FSD may not require
releasing of spinlock as it involves PSR configuration and will not cause
any harm if it is still kept inside the existing critical section only. I'll
update the patch to keep the fixup_late inside critical section and
will post the same in the next patch set.

> > +
> > +		if (priv->fixup_late)
> > +			priv->fixup_late(substream, dai);
> > +
> > +		spin_lock_irqsave(&priv->lock, flags);
> > +
> >  		if (capture)
> >  			i2s_rxctrl(i2s, 1);
> >  		else
> 
> Best regards,
> Krzysztof

Thanks,
Padmanabhan R.