Re: [Bug ?] usb :typec :tcpm :fusb302

[Guenter Roeck <linux@roeck-us.net>]

On 1/20/20 6:34 AM, Markus Reichl wrote:
> Hi Guenter,
> 
> Am 20.01.20 um 15:21 schrieb Guenter Roeck:
>> On 1/20/20 3:58 AM, Heikki Krogerus wrote:
>>> Hi Markus,
>>>
>>> On Thu, Jan 09, 2020 at 05:29:07PM +0100, Markus Reichl wrote:
>>>> Hi,
>>>>
>>>> I'm working with a ROC-RK3399-PC arm64 board from firefly, circuit sheet [1].
>>>> The board is powered from an USB-C type connector via an FUSB302 PD controller.
>>>> With measured 15W+ power consumption it should use higher voltage PD modes than
>>>> the standard 5V USB-C mode.
>>>>
>>>> When I add the related connector node in DTS [2] the FUSB302 initializes
>>>> the right PD mode (e.g. 15V/3A).
>>>>
>>>> But during initialisation the PD is switched off shortly and the board has a blackout.
>>>> When I inject a backup supply voltage behind the FUSB302 (e.g. at SYS_12V line) during boot
>>>> I can remove the backup after succesfull setting up the PD and the board will run fine.
>>>>
>>>> Is it possible to change the behaviour of the fusb302 driver to not power down the PD supply
>>>> during init?
>>>
>>> I guess it's also possible that the problem is with tcpm.c instead of
>>> fusb302.c. tcpm.c provides the USB PD state matchines. Guenter! Can
>>> you take a look at this?
>>>
>>
>> There was always a problem with handoff from the bootloader. tcpm_init() calls
>> tcpm_reset_port() which turns vbus and vconn off, which I imagine can
>> trigger the situation.
>>
>> Unfortunately I was never able to solve the puzzle. The Type-C protocol does
>> not support any kind of "hand-off" from one component in the system to another.
>> If the state machine doesn't start from a clean state, there is pretty
>> much no guarantee that it ever synchronizes.
>>
>> Maybe someone can find a better solution, but when I wrote the code I just
>> could not get it to work reliably without resetting everything during
>> registration.
>>
>> Note that v4.4 did not include the upstream tcpm code, suggesting the
>> code in the vendor kernel was possibly using a different or backported
>> state machine. Impossible to say what was done there without access
>> to the code.
> 
> The vendor code for fusb302 is here:
> https://github.com/FireflyTeam/kernel/tree/rk3399/firefly/drivers/mfd
> 

AFAICS the vendor code don't reset VBUS, and selectively (only) resets the
PD state machine in the fusb302 on startup. The tcpm state machine is embedded
in the fusb302 driver, making this easier to control.

The fusb302 Linux kernel driver, on the other side, resets the entire fusb302
on initialization, not just PD (bit 0 of the reset register). Question is if
that can be changed to just reset PD (bit 1 of the reset register).
Maybe that would already fix the problem. Can you give it a try ?

Guenter