08.10.2019 16:24, Bartlomiej Zolnierkiewicz пишет:
>
> On 10/8/19 2:47 PM, Mark Brown wrote:
>> On Tue, Oct 08, 2019 at 02:38:55PM +0200, Marek Szyprowski wrote:
>>
>>> Then if I get it right, the issue is caused by the commit 7f93ff73f7c8
>>> ("opp: core: add regulators enable and disable"). I've checked and
>>> indeed reverting it fixes Peach Pi to boot properly.
Yes, please note that the "ww_mutex" patch didn't change the original logic and only
rearranged the code a tad.
The question is if
>>> this is desired behavior or not?
>>
>> That doesn't seem ideal - either it's redundant for regulators that need
>> to be marked as always-on anyway or it's going to force the regulators
>> on when a device could do runtime PM (eg, if the same code can run on
>> something like a GPU which can be turned off while the screen is off or
>> is displaying a static image).
>
> Commit 7f93ff73f7c8 ("opp: core: add regulators enable and disable")
> currently can be safely reverted as all affected users use always-on
> regulators. However IMHO it should be possible to enable always-on
> regulator without side-effects.
>
> When it comes to setting regulator constraints before doing enable
> operation, it also seems to be possible solution but would require
> splitting regulator_set_voltage() operation on two functions:
>
> - one for setting constraints (before regulator_enable() operation)
>
> - the other one actually setting voltage (after enable operation)
>
> Unfortunately this is much bigger task and doesn't seem to be -rc
> time material so I'm in favor of just applying Marek's fix as it is
> for now.
That OPP patch caused the same problem for the NVIDIA Tegra20 CPUFreq driver (in-progress)
and I resolved it in the coupler's code [0]. Perhaps the generic coupler could do the same
thing by assuming that min_uV=current_uV until any consumer sets the voltage, i.e. if
regulator_check_consumers(min_uV=0) returns min_uV=0.
[0] https://lkml.org/lkml/2019/7/25/892
