Re: [PATCH] ASoC: qcom: add sdm845 sound card support

[Vinod <vkoul@kernel.org>]

Hi Rohit,

On 20-06-18, 13:07, Rohit Kumar wrote:
> > On 19-06-18, 19:20, Rohit Kumar wrote:
> > > On 6/19/2018 10:35 AM, Vinod wrote:
> > > > On 18-06-18, 16:46, Rohit kumar wrote:
> > > > 
> > > > > +struct sdm845_snd_data {
> > > > > +	struct snd_soc_card *card;
> > > > > +	struct regulator *vdd_supply;
> > > > > +	struct snd_soc_dai_link dai_link[];
> > > > > +};
> > > > > +
> > > > > +static struct mutex pri_mi2s_res_lock;
> > > > > +static struct mutex quat_tdm_res_lock;
> > > > any reason why the locks can't be part of sdm845_snd_data?
> > > > Also why do we need two locks ?
> > > No specific reason, I will move it to sdm845_snd_data.
> > > These locks are used to protect enable/disable of bit clocks. We have
> > > Primary MI2S RX/TX
> > > and Quaternary TDM RX/TX interfaces. For primary mi2s rx/tx, we have single
> > > clock which is
> > > synchronized with pri_mi2s_res_lock. For Quat TDM RX/TX, we are using
> > > quat_tdm_res_lock.
> > > We need two locks as we are protecting two different resources.
> > I think bigger question is why do you need any locks? What is the race
> > scenario you envision which needs protection
> > 
> 
> Below is one of the race condition:
> 
> Thread1                          |         Thread2
> ----------------------------------------------------------
> startup()                         |
> count++;                         |    startup()
> read count (count = 1) |
> enable_clock()               |    count++;   //count = 2
> shutdown()                    |
> count--;// count = 1      |
>                                          | read count (count = 1)
>                                          | enable_clock()
> 
> Here clock will be enabled twice but disable will be called only once when
> count = 0.
> 
> This will make the clock always enabled. So, I think we should keep either
> mutex lock or atomic variable to synchronize this.

we are using DPCM here right?

-- 
~Vinod