Date:   Sun, 5 Jan 2020 04:05:36 +0300
From:   Dmitry Osipenko <digetx@...il.com>
To:     Sowjanya Komatineni <skomatineni@...dia.com>,
        Mark Brown <broonie@...nel.org>
Cc:     thierry.reding@...il.com, jonathanh@...dia.com,
        lgirdwood@...il.com, perex@...ex.cz, tiwai@...e.com,
        mperttunen@...dia.com, gregkh@...uxfoundation.org,
        sboyd@...nel.org, robh+dt@...nel.org, mark.rutland@....com,
        pdeschrijver@...dia.com, pgaikwad@...dia.com, spujar@...dia.com,
        josephl@...dia.com, daniel.lezcano@...aro.org,
        mmaddireddy@...dia.com, markz@...dia.com,
        devicetree@...r.kernel.org, linux-clk@...r.kernel.org,
        linux-tegra@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v5 12/19] ASoC: tegra: Add initial parent configuration
 for audio mclk

04.01.2020 08:49, Sowjanya Komatineni пишет:
> 
> On 1/2/20 8:12 AM, Dmitry Osipenko wrote:
>> 02.01.2020 10:03, Sowjanya Komatineni пишет:
>>> On 12/27/19 1:19 PM, Sowjanya Komatineni wrote:
>>>> On 12/27/19 6:56 AM, Dmitry Osipenko wrote:
>>>>> 25.12.2019 20:57, Mark Brown пишет:
>>>>>> On Mon, Dec 23, 2019 at 12:14:34AM +0300, Dmitry Osipenko wrote:
>>>>>>> 21.12.2019 01:26, Sowjanya Komatineni пишет:
>>>>>>>> Tegra PMC clock clk_out_1 is dedicated for audio mclk from Tegra30
>>>>>>>> through Tegra210 and currently Tegra clock driver does initial parent
>>>>>>>> configuration for audio mclk "clk_out_1" and enables them by default.
>>>>>> Please delete unneeded context from mails when replying.  Doing this
>>>>>> makes it much easier to find your reply in the message, helping ensure
>>>>>> it won't be missed by people scrolling through the irrelevant quoted
>>>>>> material.
>>>>> Ok
>>>>>
>>>>>>>> -    clk_disable_unprepare(data->clk_cdev1);
>>>>>>>> -    clk_disable_unprepare(data->clk_pll_a_out0);
>>>>>>>> -    clk_disable_unprepare(data->clk_pll_a);
>>>>>>>> +    if (__clk_is_enabled(data->clk_cdev1))
>>>>>>>> +        clk_disable_unprepare(data->clk_cdev1);
>>>>>>> The root of the problem is that you removed clocks enabling from
>>>>>>> tegra_asoc_utils_init().
>>>> currently, audio mclk and its parent clocks enabling are from clock
>>>> driver init and not from tegra_asoc_utils_init.
>>>>>>> I'm not sure why clocks should be disabled during the rate-changing,
>>>>>>> probably this action is not really needed.
>>>>>> I know nothing about this particular device but this is not that
>>>>>> unusual a restriction for audio hardware, you often can't
>>>>>> robustly reconfigure the clocking for a device while it's active
>>>>>> due to issues in the hardware.  You often see issues with FIFOs
>>>>>> glitching or state machines getting stuck.  This may not be an
>>>>>> issue here but if it's something that's documented as a
>>>>>> requirement it's probably good to pay attention.
>>>>> I don't know details about that hardware either, maybe it is simply not
>>>>> safe to change PLL_A rate dynamically and then CLK_SET_RATE_GATE could
>>>>> be used.
>>>>>
>>>>> If nobody knows for sure, then will be better to keep
>>>>> tegra_asoc_utils_set_rate() unchanged.
>>>> plla rate change through tegra_asoc_utils_set_rate() happens only when
>>>> there is not active playback or record corresponding to this sound
>>>> device.
>>>>
>>>> So, I don't see reason for disabling clock during rate change and not
>>>> sure why we had this from the beginning.
>>>>
>>>> Thierry/Sameer,
>>>>
>>>> Can you please comment?
>>>>
>>>> Yes, we can use CLK_SET_RATE_GATE for PLLA and remove clock disabling
>>>> before rate change.
>>>>
>>> PLLA is used for both I2S controller clock and also for audio mclk. I2S
>>> driver suspend resume implementations takes care of enabling and
>>> disabling I2S clock but audio mclk will be enabled during that time and
>>> PLLA disable might not happen. So using CLK_SET_RATE_GATE prevents rate
>>> change to happen and as rate change happens only when there is no active
>>> audio record/playback, we can perform rate change without disable/enable
>>> during rate change.
>>>
>>> So probably below changes should be good.
>>>
>>>   * remove asoc_utils_set_rate call from asoc_utils_init as set_rate
>>>     happens during existing hw_params callback implementations in sound
>>>     drivers and there is no need to do rate change during asoc_utils_init.
>>>   * remove disable/enable clocks during rate change in asoc_utils_set_rate.
>>>   * add startup and shutdown snd_soc_ops callbacks to do enable and
>>>     disable audio mclk.
>>>
>> Sounds good, thanks. I'll be happy to review and test it.
> 
> Regarding disabling audio mclk during PLLA rate change, no need to
> explicitly disable PLLA on asoc utils as clock driver takes care of it
> properly during pll rate change.
> 
> But the downstream clock divider hardware can malfunction without
> recovery when subject to unstable PLL output during locking, unless
> clock is gated.
> 
> So it is recommended to disable downstream clocks during PLL rate change.
> 
> PLLA downstream clocks are I2S and audio mclk (cdev1/clk1 and extern1
> clocks) and I2S clock is disabled in I2S driver by PM runtime ops.

The I2S driver uses asynchronous pm_runtime_put() and thus there is no
guarantee that I2S clock is disabled at the time of changing PLLA rate.
Could this be a problem?

> For audio mclk, need to make sure mclk are disabled during rate change.
> 
> So below are the changes to audio clocks that will be in next version.
> 
>   * remove tegra_asoc_utils_set_rate call from tegra_asoc_utils_init as
>     tegra_asoc_utils_set_rate happens during hw_params callback.
>   * add shutdown snd_soc_ops callbacks to disable of audio mclk.
>   * remove disable audio mclk (cdev1) and plla clocks prior to rate
>     change in tegra_asoc_utils_set_rate (as audio mclk will always be in
>     disabled state every time hw_param callback gets executed) and keep
>     audio mclk enable after the rate change in tegra_asoc_utils_set_rate.
> 

Powered by blists - more mailing lists