[RFC PATCH] drm/vc4: Add a load tracker to prevent HVS underflow errors

[Daniel Vetter]

On Mon, Oct 29, 2018 at 10:10:56AM +0100, Boris Brezillon wrote:
> On Mon, 29 Oct 2018 10:03:01 +0100
> Daniel Vetter <daniel at ffwll.ch> wrote:
> > On Mon, Oct 29, 2018 at 09:41:36AM +0100, Boris Brezillon wrote:
> > > On Mon, 29 Oct 2018 09:06:40 +0100
> > > > The other upshot of a counter is that there's no problem with resetting.
> > > > Userspace simply grabs the counter before and after the test and compares.
> > > > If you only have 0/1 you need some way to reset, or otherwise automated
> > > > running in a CI farm isn't possible.  
> > > 
> > > The reset was done during atomic commit in my proposal, so no action
> > > was required on the user side apart from applying a new config.
> > > Anyway, I'll change that for a real counter.  
> > 
> > We want to measure underruns while doing full modesets too (specifically,
> > when shutting down the pipeline). At least for our hw this has
> > historically helped greatly in catching programming sequence issues.
> 
> Looks like you have particular needs for intel HW which I don't know
> about, maybe I shouldn't be the one driving this rework...

No one is expecting you do drive i915 work, we just want to make sure that
whatever you come up for igt tests is generally useful, and not just only
for the limited use case of vc4. Atm the only two drivers interested in
underrun reporting are i915 and vc4, and it looks like they span a pretty
broad spectrum alredy.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch