[Intel-gfx] [PATCH 3/8] drm/i915/pcode: Extend pcode functions for multiple gt's

[Jani Nikula]

On Wed, 20 Apr 2022, "Vivi, Rodrigo" <rodrigo.vivi at intel.com> wrote:
> On Tue, 2022-04-19 at 22:54 -0700, Dixit, Ashutosh wrote:
>> On Fri, 15 Apr 2022 03:21:26 -0700, Rodrigo Vivi wrote:
>> > On Thu, Apr 14, 2022 at 03:31:07PM -0700, Dixit, Ashutosh wrote:
>> > > On Thu, 14 Apr 2022 06:28:57 -0700, Jani Nikula wrote:
>> > > > 
>> > > > On Wed, 13 Apr 2022, Ashutosh Dixit <ashutosh.dixit at intel.com>
>> > > > wrote:
>> > > > > Each gt contains an independent instance of pcode. Extend
>> > > > > pcode functions
>> > > > > to interface with pcode on different gt's. Previous (GT0)
>> > > > > pcode read/write
>> > > > > interfaces are preserved.
>> > > > 
>> > > > The big problem here is that this hard couples display code to
>> > > > gt code,
>> > > > while we're trying hard to go the opposite direction. It
>> > > > doesn't matter
>> > > > that the existing interfaces are preserved as wrappers when it
>> > > > relies on
>> > > > an intel_gt being available (via i915->gt0).
>> > 
>> > I don't believe there is a big problem in here...
>> > 
>> > please note the intel_pcode.h is keeping the abstraction for
>> > display
>> > 
>> > #define snb_pcode_write_timeout(i915, mbox, val, fast_timeout_us,
>> > slow_timeout_ms) \
>> >         intel_gt_pcode_write_timeout(&(i915)->gt0, mbox, val,
>> > fast_timeout_us, slow_timeout_ms)
>> > 
>> > #define snb_pcode_write(i915, mbox, val) \
>> >         snb_pcode_write_timeout(i915, mbox, val, 500, 0)
>> > 
>> > display only uses these macros that Ashutosh didn't touch.
>> > 
>> > > > 
>> > > > Note how 'git grep intel_gt -- drivers/gpu/drm/i915/display/'
>> > > > matches
>> > > > only 1 line.
>> > 
>> > As well with the patches applied:
>> > 
>> > $ git log --oneline -1
>> > 1f58f1195478 (HEAD -> drm-tip) drm/i915/gt: Expose per-gt RPS
>> > defaults in sysfs
>> > 
>> > $ git grep intel_gt -- drivers/gpu/drm/i915/display/
>> > drivers/gpu/drm/i915/display/intel_display.c:          
>> > intel_gt_set_wedged(to_gt(dev_priv));
>> > 
>> > > 
>> > > Hi Jani, would you have suggestions about how to do this (handle
>> > > pcode on
>> > > multiple gt's)? The thinking was this patch would be a
>> > > straightforward way
>> > > to avoid code duplication. Also:
>> > 
>> > Maybe it is just a matter of renaming the macros used by display
>> > in intel_pcode.h to reflect that it should be used by display only?
>> 
>> In v2 I have added a patch ([PATCH 4/9] drm/i915/gt: Convert callers
>> to
>> user per-gt pcode functions) which correctly calls per-gt pcode
>> functions
>> where this is required. With this patch only display functions (and
>> one
>> other caller) are left calling the "global scope"
>> snb_pcode_read/write*
>> functions. So the legacy snb_pcode_read/write* are now basically
>> being used
>> only by display. Let's see if Jani is ok with this. Thanks.
>
> Jani is not happy with this abstraction because it still creates some
> dependency and also no with the name intel_gt_pcode_ in the
> functions...
>
> He has some valid points.
>
> I believe the right way to do this is to keep intel_pcode totally clean
> from intel_gt and only receive intel_uncore as the argument. Then, if
> needed we create display/intel_display_pcode and/or gt/intel_gt_pcode
> with the needed abstractions... but better with none I'd say.

I'd prefer it if you only passed uncore, not gt, to the pcode functions.

BR,
Jani.

-- 
Jani Nikula, Intel Open Source Graphics Center