Re: [PATCH] drm/i915/dmc: Add debugfs for dc6 counter

[Imre Deak <imre.deak@intel.com>]

On Mon, Feb 03, 2025 at 06:18:29PM -0300, Gustavo Sousa wrote:
> Quoting Vivi, Rodrigo (2025-02-03 17:59:19-03:00)
> [...]
> Perhaps Imre can explain this better, but I believe the point is that we
> want to track increments to DC5 counter when we have DC6 enabled. That
> driver-managed counter would be in dc6_allowed.
> 
> Repeating Imre's suggestions with a minor tweak:
> 
> 1. Before we tell the hardware that we are allowing DC6 (disable ->
>    DC6), we store the value of the current DC5 counter.
>
> 2. After we disable DC states from DC6 (DC6 -> disable), we read the DC5
>    counter again and subtract the value from (1). The result would then be
>    added to the current value of dc6_allowed.

Yes, with the actual delta being: DC5 counter read here - DC5 counter read
at (1).

> In (1) I think we should read the DC5 counter before we update
> DC_STATE_EN, just to be sure we avoid some sort of race (although that
> appears to be unlikely to happen).

Yes, the orders you described in both (1) and (2) are correct (since a
DC5 -> DC6 or DC6 -> DC5 transition is not possible).

The dc6_allowed counter should be also updated before returning it to
userspace via the debugfs entry, as I mentioned earlier (to account
for the case where DC6 is enabled when the read happens).

> During DC6 validation, if the test sees that dc6_allowed was
> incremented, that means that the display engine reached a state where
> the SOC would be able to put the display in DC6.

Yes, that's my understanding.

--Imre