[Intel-gfx] [RFC 0/3] Display uncore

[Chris Wilson]

Quoting Daniele Ceraolo Spurio (2019-08-08 02:44:20)
> I've been trying to identify MMIO ranges to clearly define what belongs
> to display_uncore to do a check on access, but there are lots of
> exceptions and differences across gens (with a few more coming with TGL),
> so I don't think that's a viable way. The alternative option implemented
> here is to differentiate the register by type, which should ensure we
> never mix them up, but at the cost of a more complex transition.

One thing we could try with this approach is to tag every _MMIO() as
either DE or GT and have the uncore accessors check the magic bits
before scrubbing them. (With enough magic macros to make it disappear

Huge task, but not insurmountable. The danger is that if we do this
piecemeal is that we may end up with two simultaneous accesses via the
separate uncore accessors. Hmm.

On thing though is that Jani may find the intel_de_write (or just
de_write, the frequency may be worth bending the naming rules) as being
palatable.
-Chris