Re: [alsa-devel] [PATCH 2/2] ASoC: fsl_ssi: serialize AC'97 register access operations

[Nicolin Chen]

On Mon, Nov 20, 2017 at 11:16:07PM +0100, Maciej S. Szmigiero wrote:
> AC'97 register access operations (both read and write) on SSI use a one,
> shared set of SSI registers for AC'97 register address and data.
> This means that only one such access is possible at a time and so all these
> operations need to be serialized.
> 
> Since an AC'97 register access operation in this driver takes 100us+ let's
> use a mutex for this.
> 
> Use this opportunity to also change a default value returned from AC'97
> register read function from -1 to 0, since that's what AC'97 specs require
> to be returned when unknown / undefined registers are read.
> 
> Signed-off-by: Maciej S. Szmigiero <mail@xxxxxxxxxxxxxxxxxxxxx>

>  static unsigned short fsl_ssi_ac97_read(struct snd_ac97 *ac97,
> @@ -1287,16 +1295,18 @@ static unsigned short fsl_ssi_ac97_read(struct snd_ac97 *ac97,
>  {
>  	struct regmap *regs = fsl_ac97_data->regs;
>  
> -	unsigned short val = -1;
> +	unsigned short val = 0;
>  	u32 reg_val;
>  	unsigned int lreg;
>  	int ret;
>  
> +	mutex_lock(&fsl_ac97_data->ac97_reg_lock);
> +
>  	ret = clk_prepare_enable(fsl_ac97_data->clk);
>  	if (ret) {
>  		pr_err("ac97 read clk_prepare_enable failed: %d\n",
>  			ret);
> -		return -1;
> +		goto ret_unlock;

It will return val (== 0) in this case. Will this be correctly
handled by callers? I find sound/ac97/bus.c checks if ret < 0
for ops->read().

So it might be better to add "val = ret;" before goto? Or use
val instead of ret directly?

>  	}
>  
>  	lreg = (reg & 0x7f) <<  12;
> @@ -1311,6 +1321,8 @@ static unsigned short fsl_ssi_ac97_read(struct snd_ac97 *ac97,
>  
>  	clk_disable_unprepare(fsl_ac97_data->clk);
>  
> +ret_unlock:
> +	mutex_unlock(&fsl_ac97_data->ac97_reg_lock);
>  	return val;
>  }