[Freedreno] [PATCH 1/2] drm/msm/dp: postpone irq_hpd event during connection pending state

[khsieh at codeaurora.org]

On 2021-01-13 16:00, Stephen Boyd wrote:
> Quoting khsieh at codeaurora.org (2021-01-13 15:44:32)
>> On 2021-01-13 12:22, Stephen Boyd wrote:
>> > Quoting khsieh at codeaurora.org (2021-01-13 09:44:24)
>> >> On 2021-01-11 11:55, Stephen Boyd wrote:
>> >> > Quoting Kuogee Hsieh (2021-01-07 12:30:24)
>> >> >> irq_hpd event can only be executed at connected state. Therefore
>> >> >> irq_hpd event should be postponed if it happened at connection
>> >> >> pending state. This patch also make sure both link rate and lane
>> >> >
>> >> > Why does it happen at connection pending state?
>> >> plug in need two state to complete it.
>> >> advance to connection pending state once link training completed and
>> >> sent uevent notification to frame work.
>> >> transition to connected state after frame work provided resolution
>> >> timing and start transmit video panel.
>> >> Therefore irq_hpd should not be handled if it occurred before
>> >> connected
>> >> state.
>> >
>> > Sure that's what's going on in the patch but you didn't answer my
>> > question. Why does irq_hpd happen before connected state?
>> 
>> I have no idea why it happen this way.
>> during debug
>> https://partnerissuetracker.corp.google.com/issues/170598152
>> I saw two different scenario
>> 1) irq_hpd followed by unplug with less than 20 ms in between. this 
>> one
>> fixed by this patch set.
>> 2) plug followed by irq_hpd around 300ms in between. it does not cause
>> problem. but it should be handled in order (after connected state).
> 
> Ok. So nobody understands why the hardware is acting this way and we're
> papering over the problem by forcing the HPD state to be what we think
> it should be? That's not great.

irq_hpd is issued from dongle.
it then go through EC ps8805 driver and reach DP driver finally.
Again, to duplicate problem #1 this at my set up, i have to 
intentionally wiggling type-c connector of dongle.
But I can not duplicate problem #2 and only saw it one time from Quantan 
provide logs.