[PATCH v3 08/10] phy: socionext: Add UniPhier USB3 PHY driver

[Marek Vasut]

On 2/16/23 17:14, Kunihiko Hayashi wrote:
> Hi Marek,

Hello Hayashi-san,

> Sorry for late reply.
> 
> On 2023/02/14 6:06, Marek Vasut wrote:
>> On 2/13/23 04:08, Kunihiko Hayashi wrote:
>>> Hi Marek,
>>
>> Hello Hayashi-san,
>>
> 
> [...]
> 
>>> I think so, however, when I added .exit() and executed "usb stop;usb
>>> start",
>>> unfortunately the command got stuck.
>>>
>>> Currently uniphier clk doesn't support CLK_CCF and can't be nested.
>>> I think more changes are needed.
>>
>> Do you know where exactly the system hangs ?
> 
> Yes.
> 
> The controller-reset driver controls the clock/reset for the glue
> in probe(). The phy driver controls the clock/reset for the glue and
> the phy in init(). There is an inconsistency.
> 
> When submitting "usb stop", the phy driver disables all the clock/reset
> in exit().
> 
> When submitting "usb start" again, the controller-reset driver accesses
> the glue register without enabling the clock/reset, and the system hangs.

Shouldn't the PHY enable its own set of clock, while the controller 
should only enable its own (different) set of clock ?

If it is the same clock and there is no refcounting, then maybe what you 
should do for now is enable/disable all the clock in the controller 
driver only, and deal with the PHY once CCF with refcounting is in place ?

>> Do you expect to add the CCF support ?
> 
> Yes, first I thought the clock needed the CCF support to count the clock
> function calls, but there is no support to count the reset function calls.
> I need another solution.
> 
> Currently the phy driver controls all the clock/reset in init() and exit(),
> however, it should control the phy clock/reset only.

If that's possible on this hardware, that would be good, let controller 
control its clock and PHY control its (different) clock.

>> If so, then add at least a FIXME code comment that the exit callback
>> must be implemented, so that this is not forgotten once CCF is in place.
>> I don't want this series to grow much further into some "rework
>> everything at once" thing.
> 
> As above, I should put the glue clock/reset into probe(),
> and the phy clock/reset into init() without using bulk function.

OK