[Intel-gfx] [PATCH i-g-t] pm_rps: Extended testcases with checking PMINTRMSK register value

[Daniel Vetter]

On Thu, Aug 10, 2017 at 02:20:27PM +0100, Chris Wilson wrote:
> Quoting Katarzyna Dec (2017-08-10 14:06:15)
> > In addition to checking whether the frequency is in correct range
> > for certain scenario, we can also verify whether PM interrupts are
> > masked correctly.
> 
> What does correctly mean? It was a recommendation not a requirement.
> 
> You are checking internal kernel implementation details, just add the
> check to our runtime suspend that we have applied the mask (which is a
> reasonable check as the rps idling is decoupled from the rps suspend).

Yes, igt testcase cannot (well, only under very specific exceptions) make
assumptions about the kernel implementations. Doing so unecessarily ties
the test to a given implementation/platform, which destroy all the value
it provides for regression testing from one platform to the next.

But while you look into this, the testcase itself does sometimes fail in
our CI:

https://intel-gfx-ci.01.org/tree/drm-tip/shards.html

Both pm_rps at reset and pm_rps at waitboost seem to randomly fail on all
platforms we're currently testing.

Please take a look at these failures and try to fix them (either test or
kerenl), that would be an actual substantial improvement of igt.

Thanks, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch