[Intel-gfx] [PATCH] igt/gem_workarounds: igt to test workaround registers

[Damien Lespiau]

On Thu, Aug 28, 2014 at 06:55:37AM +0100, Chris Wilson wrote:
> On Wed, Aug 27, 2014 at 11:30:35PM +0100, Damien Lespiau wrote:
> > On Wed, Aug 27, 2014 at 06:52:57PM +0100, Chris Wilson wrote:
> > > > Just to clarify, he was not ok because the list we maintain in the
> > > > test can get out of sync with the workarounds we apply in the driver
> > > > which can be avoided if it is generated by the kernel itself.
> > > 
> > > Test driven development would suggest that the test itself maintains its
> > > list. Something I heard Daniel say himself before ;-)
> > >  
> > > > It may be ok to maintain the list in the test in this case
> > > > considering the list is fairly small but it is not my call.
> > > 
> > > The best thing about independent testing is that it is independent...
> > 
> > Well also depends on what you're testing I suppose. It's hard enough to
> > have a complete list of W/As, so two of them is bound to end up in
> > tears. If we are testing that the list of W/As the kernel knows about is
> > indeed applied correctly at init/reset/suspend resume, that's already a
> > good step.
> > 
> > Also, that second list in i-g-t is not going to be implemented in
> > complete independence from the kernel tree, it's likely to be the same
> > person doing both sides, ending up copy/pasting a file anyway, no value
> > in doing that. The two lists argument works well if 2 different
> > engineers/teams implement the 2 sides, effective cross-checking the list
> > of W/As as a result, but we don't quite have the people to do that.
> 
> The point of the independent test is more that you can ask people to run
> and see if it reports strange things on unknown kernels that might
> explain bugs. There has to be an external list anyway just so that you
> can check off the appropriate w/a.
> 
> Putting that second list in the kernel just leads to bugs in the kernel
> as aptly demonstrated by the patch and doesn't lead to those warm fuzzy
> feelings.

Ah, fair, for those points it'd be ok to just isolate the W/As in a
file, decide that the master copy is in the kernel and sync it from the
kernel to i-g-t when it changes. Not the full "independent" testing I
was thinking about with separate people writing code and validation.

-- 
Damien