Re: [PATCH v7 00/21] Move PMC clocks into Tegra PMC driver

[Dmitry Osipenko]

10.01.2020 07:47, Sowjanya Komatineni ÐÐÑÐÑ:
> 
> On 1/9/20 8:43 PM, Sameer Pujar wrote:
>> External email: Use caution opening links or attachments
>>
>>
>> On 1/10/2020 10:06 AM, Sowjanya Komatineni wrote:
>>>
>>> On 1/9/20 7:32 PM, Sowjanya Komatineni wrote:
>>>>
>>>> On 1/9/20 7:24 PM, Sowjanya Komatineni wrote:
>>>>>
>>>>> On 1/9/20 5:39 PM, Sowjanya Komatineni wrote:
>>>>>>
>>>>>> On 1/9/20 11:44 AM, Dmitry Osipenko wrote:
>>>>>>> External email: Use caution opening links or attachments
>>>>>>>
>>>>>>>
>>>>>>> 08.01.2020 07:24, Sowjanya Komatineni ÐÐÑÐÑ:
>>>>>>>> This patch series moves Tegra PMC clocks from clock driver to pmc
>>>>>>>> driver
>>>>>>>> along with the device trees changes and audio driver which uses
>>>>>>>> one of
>>>>>>>> the pmc clock for audio mclk.
>>>>>>>>
>>>>>>>> Tegra PMC has clk_out_1, clk_out_2, clk_out_3 and blink controls
>>>>>>>> which
>>>>>>>> are currently registered by Tegra clock driver using
>>>>>>>> clk_regiser_mux and
>>>>>>>> clk_register_gate which performs direct Tegra PMC register access.
>>>>>>>>
>>>>>>>> When Tegra PMC is in secure mode, any access from non-secure
>>>>>>>> world will
>>>>>>>> not go through.
>>>>>>>>
>>>>>>>> This patch series adds these Tegra PMC clocks and blink controls
>>>>>>>> to Tegra
>>>>>>>> PMC driver with PMC as clock provider and removes them from Tegra
>>>>>>>> clock
>>>>>>>> driver.
>>>>>>>>
>>>>>>>> PMC clock clk_out_1 is dedicated for audio mclk from Tegra30 thru
>>>>>>>> Tegra210
>>>>>>>> and clock driver does inital parent configuration for it and
>>>>>>>> enables them.
>>>>>>>> But this clock should be taken care by audio driver as there is
>>>>>>>> no need
>>>>>>>> to have this clock pre enabled.
>>>>>>>>
>>>>>>>> So, this series also includes patch that updates ASoC driver to
>>>>>>>> take
>>>>>>>> care of parent configuration for mclk if device tree don't specify
>>>>>>>> initial parent configuration using assigned-clock-parents and
>>>>>>>> controls
>>>>>>>> audio mclk enable/disable during ASoC machine startup and shutdown.
>>>>>>>>
>>>>>>>> DTs are also updated to use clk_out_1 as audio mclk rather than
>>>>>>>> extern1.
>>>>>>>>
>>>>>>>> This series also includes a patch for mclk fallback to extern1 when
>>>>>>>> retrieving mclk fails to have this backward compatible of new DT
>>>>>>>> with
>>>>>>>> old kernels.
>>>>>>> Suspend-resume doesn't work anymore, reverting this series helps. I
>>>>>>> don't have any other information yet, please take a look.
>>>>>> Thanks Dmitry. Will test suspend resume and check..
>>>>>
>>>>> I see if we leave audio mclk (cdev1) enabled during
>>>>> tegra_asoc_utils_init, suspend resume works.
>>>>>
>>>>> Without audio mclk enabled during tegra_asoc_utils_init, somehow it
>>>>> prevents entry to suspend on Tegra30 platform.
>>>>>
>>>>> Will look in detail..
>>>>>
>>>> audio mclk is only needed for audio and werid that having it not
>>>> enabled all the time like in current clock driver prevents suspend
>>>> entry on Tegra30
>>>>
>>>> Looks like this issue is masked earlier with having mclk enabled all
>>>> the time by clock driver.
>>>>
>>> On linux-next without this patch series, I just disabled mclk to be
>>> enabled all the time (removed set_rate from utils_init) and also
>>> disabled default enable from clock driver.
>>>
>>> So somehow disabling mclk is preventing suspend entry.
>>
>> This is strange.
>>
>>>
>>> Probably debugging suspend issue on Tegra30 when audio mclk is
>>> disabled can be done separately and will keep audio mclk enabled in
>>> asoc_utils_init with comment mentioning this issue and fix as TBD to
>>> move on with PMC clock fixes.
>>
>> Sounds fine with me as the suspend/resume issue is not introduced in the
>> current series. It can be addressed separately.
>>
>>>
> Thanks Sameer. So, will keep mclk not enabled in clock driver but will
> do mclk enable in asoc_utils_init and will remove machine startup and
> shutdown.
> 
> mclk dependency with suspend/resume and I2S and audio clocks proper
> handling in audio driver can be taken care separately out of this series.
> 
> Dimitry, I hope you too agree with this.

Yes, should be fine to fix it separately from this series.

I suppose the clocks management isn't done properly by some of the audio
drivers and machine hangs after trying to access hardware module which
has the disabled clock. That's a quite typical bug.