Quoting Daniel Vetter (2018-07-11 10:17:09)
> On Wed, Jul 11, 2018 at 08:36:07AM +0100, Chris Wilson wrote:
> > Currently the code to reset the GPU and our state is spread widely
> > across a few files. Pull the logic together into a common file.
> > Signed-off-by: Chris Wilson <ch...@chris-wilson.co.uk>
> Hm when reading the fence code I'd suggested more to move the reset/init
> code for a given topic (like fence_reg) into one place. Imo that makes
> reviewing changes to a specific thing easier because it's all in one file.
> Ofc it makes it harder to review large-scale flows like suspend/resume and
> gpu reset, but I think the only thing that can safe us there is lots and
> lots of runtime consistency checks for each component (lockdep,
> GEM_BUG_ON). Reviewing those is doable, and for the big picture
> tldr; not sure this is the best organization we can do.
I am confident that pulling the code flow that was split across i915_irq.c
(of all, this was the silliest), i915_drv.c, intel_uncore.c and then
into i915_gem.c; into a single location is an improvement for
understanding that flow and the subsequent changes.
Whether it stands the test of time, we will see, but the immediate
question is simply whether it is an improvement that opens up further
incremental improvements, or if it is a complete dead end, a maintenance
Intel-gfx mailing list