[igt-dev] [PATCH v8 1/1 i-g-t] tests: Add a new test for driver/device hot reload

[Daniel Vetter]

On Mon, May 13, 2019 at 08:45:57AM +0200, Janusz Krzysztofik wrote:
> Hi Daniel,
> 
> On Tuesday, May 7, 2019 3:32:10 PM CEST Daniel Vetter wrote:
> > ...
> > Another option would be to first have a more minimal hotunplug test
> > without any busy batches pending, and make sure that works reliable
> > (i.e. no buffer, no context, no nothing, no open fd, absolutely
> > nothing really allocated from userspace). Then extend from there
> > step-by-step, e.g. first just open a drm device fd, then allocate a
> > buffer, then mmap, then maybe try to write to that buffer after
> > unplug, try to do an ioctl after unplug (go through all of them).
> > 
> > Only once you've gone through all the more basic things go to the much
> > more evil stuff like having a batch in-flight, or a kms request
> > in-flight. That way it should be easier to make forward progress on
> > these bugs without risking the world. Plus you can start merging the
> > igt side already, to make sure no one regresses anymore.
> 
> It looks like I might have misunderstood your idea.  Did you mean I should 
> start with a patch that adds only the minimal hotunplug test and nothing more? 
> Then extend it with more advanced subtests one by one consecutively but only 
> when driver issues revealed by those subtests already in place are resolved?

Yes that's the idea, we can't merge testcase that break the driver badly
(because that will result in other tests not being run, which isn't good).

> If that's the case, I should probably submit v10 limited to the minimal test 
> since v9 I submitted last week was not going to be taken under consideration.

I'll look at v9 still for some feedback on all the subtests, but yes need
to split it up and merge as we fix up the kernel side. So always submit
kernel and igt patches, pre-merge test them together (with the Test-with:
tag). Then first merge kernel fixes, then igt, and move on to the next set
of subtests.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch