[Intel-gfx] [PATCH v4] drm/i915: Try to guess PCH type even without ISA bridge

[Jani Nikula]

On Mon, 18 Jan 2021, Zhenyu Wang <zhenyuw at linux.intel.com> wrote:
> On 2021.01.15 13:01:49 +0200, Joonas Lahtinen wrote:
>> Quoting Jani Nikula (2021-01-15 12:50:54)
>> > On Thu, 14 Jan 2021, Zhenyu Wang <zhenyuw at linux.intel.com> wrote:
>> > > On 2021.01.14 08:58:19 +0800, Xiong Zhang wrote:
>> > >> From: Zhenyu Wang <zhenyuw at linux.intel.com>
>> > >> 
>> > >> Some vmm like hyperv and crosvm don't supply any ISA bridge to their guest,
>> > >> when igd passthrough is equipped on these vmm, guest i915 display may
>> > >> couldn't work as guest i915 detects PCH_NONE pch type.
>> > >> 
>> > >> When i915 runs as guest, this patch guess pch type through gpu type even
>> > >> without ISA bridge.
>> > >> 
>> > >> v2: Fix CI warning
>> > >> v3: Add HAS_DISPLAY()= true condition beforce guessing virt pch, then
>> > >>     refactori.
>> > >> v4: Fix CI warning
>> > >> 
>> > >> Signed-off-by: Zhenyu Wang <zhenyuw at linux.intel.com>
>> > >> Signed-off-by: Xiong Zhang <xiong.y.zhang at intel.com>
>> > >> ---
>> > >>  drivers/gpu/drm/i915/i915_drv.h  |  7 +++++-
>> > >>  drivers/gpu/drm/i915/intel_pch.c | 39 ++++++++++++++++++--------------
>> > >>  2 files changed, 28 insertions(+), 18 deletions(-)
>> > >>
>> > >
>> > > Good to me, thanks! I think this should change author to you. :)
>> > >
>> > > Reviewed-by: Zhenyu Wang <zhenyuw at linux.intel.com>
>> > 
>> > Pushed to drm-intel-next, thanks for the patch and review. I ended up
>> > retaining Zhenyu's authorship, and added
>> > 
>> > Co-developed-by: Xiong Zhang <xiong.y.zhang at intel.com>
>> 
>> This only works for the majority of SKUs which happen to use the PCH
>> that is most common. I NAKed very similar patch some years back asking
>> for a reliable means to detect the PCH type instead.
>> 
>> Has there been any attempt to introduce such mechanism?
>> 
>
> I think the situation is that if just passing through GPU device, as
> full device model depends on userspace VMM, we have no way to properly
> detect PCH type from device, e.g in case of one GEN device which could
> live with two or more different PCH types. It's better if either we
> have properly defined way to detect through GPU device or display arch
> always defines CPU/PCH display in pair.
>
> Currently this tries to assume PCH with best effort that fixed bunch of
> problems. On really mismatch case, how about adding i915 parameter to
> override PCH type? As that won't add dependency on other components..

In general, module parameters should not be considered a fix or a
solution.

BR,
Jani.


-- 
Jani Nikula, Intel Open Source Graphics Center