Re: BD71847 clk driver disables clk-32k-out causing RTC/WDT failure

[Peng Fan <peng.fan@oss.nxp.com>]

On 9/9/2022 1:06 PM, Matti Vaittinen wrote:
> Hi dee Ho peeps,
> 
> On 9/9/22 05:35, Marek Vasut wrote:
>> On 9/9/22 04:06, Peng Fan wrote:
>>>> Subject: Re: BD71847 clk driver disables clk-32k-out causing RTC/WDT 
>>>> failure
>>>>
>>>> On 9/8/22 21:25, Tim Harvey wrote:
>>>>> On Thu, Sep 8, 2022 at 9:55 AM Marek Vasut <marex@denx.de> wrote:
>>>>>>
>>>>>> On 9/8/22 18:00, Tim Harvey wrote:
>>>>>>> On Thu, Sep 1, 2022 at 9:14 PM Matti Vaittinen
>>>> <mazziesaccount@gmail.com> wrote:
>>>>>>>>
>>>>>>>> Hi Tim,
>>>>>>>>
>>>>>>>> On 9/2/22 01:23, Tim Harvey wrote:
>>>>>>>>> Greetings,
>>>>>>>>>
>>>>>>>>> I've found that the bd71847 clk driver
>>>> (CONFIG_COMMON_CLK_BD718XX
>>>>>>>>> drivers/clk/clk-bd718x7.c) disables clk-32k-out (the BD71847
>>>>>>>>> C32K_OUT
>>>>>>>>> pin) which is connected IMX8MM RTC_XTALI which ends up disabling
>>>>>>>>> the IMX RTC as well as the IMX WDOG functionality.
>>>>>>>>
>>>>>>>> //snip
>>>>>>>>
>>>>>>>>> This happens via clk_unprepare_unused() as nothing is flagging the
>>>>>>>>> clk-32k-out as being used. What should be added to the device-tree
>>>>>>>>> to signify that this clk is indeed necessary and should not be 
>>>>>>>>> disabled?
>>>>>>>>
>>>>>>>> I have seen following proposal from Marek Vasut:
>>>>>>>>
>>>>>>>>
>>>> https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fl
>>>>>>>> ore.kernel.org%2Fall%2F20220517235919.200375-1-
>>>> marex%40denx.de%2FT%
>>>>>>>>
>>>> 2F%23m52d6d0831bf43d5f293e35cb27f3021f278d0564&amp;data=05%7C0
>>>> 1%7Cp
>>>>>>>>
>>>> eng.fan%40nxp.com%7C07d48edcc47c4694e08208da91da2bf4%7C686ea1d
>>>> 3bc2b
>>>>>>>>
>>>> 4c6fa92cd99c5c301635%7C0%7C0%7C637982664162868785%7CUnknown%
>>>> 7CTWFpb
>>>>>>>>
>>>> GZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI
>>>> 6
>>>>>>>>
>>>> Mn0%3D%7C3000%7C%7C%7C&amp;sdata=uF26u9g4onuqCWzPRAvD%2F%
>>>> 2FLByaEhh5
>>>>>>>> Dtah9K8CcAOAM%3D&amp;reserved=0
>>>>>>>>
>>>>>>>> I am not sure if the discussion is completed though. I guess it was
>>>>>>>> agreed this was needed/usefull and maybe the remaining thing to
>>>>>>>> decide was just the property naming.
>>>>>>>>
>>>>>>>> Best Regards
>>>>>>>>            -- Matti
>>>>>>>>
>>>>>>>
>>>>>>> Thanks Matti,
>>>>>>>
>>>>>>> Marek - has there been any progress on determining how best to keep
>>>>>>> certain clocks from being disabled?
>>>>>>
>>>>>> No. You can read the discussion above.
>>>>>
>>>>> Marek,
>>>>>
>>>>> I wasn't on the linux-clk list at that time so can't respond to the
>>>>> thread but the discussion seems to have died out a couple of months
>>>>> ago with no agreement between you or Stephen on how to deal with it.
>>>>>
>>>>> So where do we take this from here? It looks like there are about 18
>>>>> boards with dt's using "rohm,bd718*" which would all have non working
>>>>> RTC/WDOG with CONFIG_COMMON_CLK_BD718XX enabled (which it is in
>>>>> arch/arm64/configs/defconfig) right?
> 
> Yeah. The ROHM BD71837 and BD71847 (and BD71850 - which is one of the 
> variants) are used quite a lot. I am pretty sure not fixing this in 
> upstream is increasing downstream solutions. I don't think that should 
> be preferred.
> 
>>>
>>> Is there any requirement that the bd718xx clk needs to be runtime 
>>> on/off?
>>
>> Yes, the 32kHz clock on BD71xxx should behave like any other clock, 
>> unless specified otherwise, see below.
>>
>>> I suppose the clk should always be never be off, if yes, why not have 
>>> something:
>>
>> What is needed in this specific case of BD718xx is I think clock 
>> consumer on the MX8M clock driver side which would claim the 32kHz 
>> input from the PMIC and up the clock enable count to keep the 32 kHz 
>> clock always on.
> 
> This sounds like a solution that would describe the actual HW setup. I 
> don't know the CCF of the i.MX8 well enough to tell whether this can 
> ensure the clk is not disabled before the consumer is found or when the 
> consumer is going down though. Simplest thing to me would really be to 
> just mark the clk as "do-not-touch" one on the boards where it must not 
> be touched.
> 
>   The PMIC is most likely supplying 32 kHz clock to the MX8M,
>> which if the 32 kHz clock are turned off would hang (I observed that 
>> before too).
>>
>> What I tried to address in this thread is a generic problem which 
>> commonly appears on various embedded systems, except every time anyone 
>> tried to solve it in a generic manner, it was rejected or they gave up.
> 
> I agree with Marek - generic solution would be nice. I don't think this 
> is something specific to this PMIC.
> 
>> The problem is this -- you have an arbitrary clock, and you need to 
>> keep it running always otherwise the system fails, and you do not have 
>> a clock consumer in the DT for whatever reason e.g. because the SoC is 
>> only used as a clock source for some unrelated clock net. There must 
>> be a way to mark the clock as "never disable these", i.e. 
>> critical-clock. (I feel like I keep repeating this over and over in 
>> this thread, so please read the whole thread backlog)
> 
> Thanks for the explanation and effor you did Marek.
> 
> My take on this is that from a (generic) component vendor perspective it 
> is a bad idea to hard-code the clock status (enable/disable) in the PMIC 
> driver. A vendor wants to provide a driver which allows use of the 
> component in wide variety of systems/boards. When the PMIC contains a 
> clock gate, the PMIC driver should provide the means of controlling it. 
> Some setups may want it enabled, other disabled and some want runtime 
> control. This "use-policy" must not be hard coded in the driver - it 
> needs to come from HW description which explains how the clk line is 
> wired and potentially also from the consumer drivers. This enables the 
> same PMIC driver to support all different setups with their own needs, 
> right?
> 
> I am not sure if some non email discussions have been ongoing around 
> this topic but just by reading the emails it seemed to me that Marek's 
> suggestion was acked by the DT folks - and I don't think that Stephen 
> was (at the end of the day) against that either(?). Maybe I missed 
> something.

After a thought, maybe an easier way is to add a optional property
xxx,32k-always-on to the pmic node/driver.

Regards,
Peng.

> 
> Yours
>      -- Matti
>