Re: [igt-dev] [PATCH i-g-t] lib/intel_device_info: Add IS_DGFX() support

["Dixit, Ashutosh" <ashutosh.dixit@intel.com>]

On Sun, 24 Apr 2022 23:10:23 -0700, Gupta, Anshuman wrote:
>
> > > > On Fri, 22 Apr 2022 05:59:12 -0700, Anshuman Gupta wrote:
> > > > >
> > > > > Currently IGT is lacking IS_DGFX() macro support.
> > > > > There are some power features like D3Cold are only supported on
> > > > > discrete card. So IGT test/tools specific to D3Cold requires to
> > > > > consume IS_DGFX().
> > > > > Adding a is_dgfx field in intel_device_info and initializing it
> > > > > for DG1. All future discrete platform would require to initialize
> > > > > this field.
> > > >
> > > > #define IS_DGFX(fd) gem_has_lmem(fd) ?
> > > >
> > > > gem_has_lmem() is already pretty widely used so maybe we should not
> > > > introduce another way to achieve the same goal?
> > >
> > > Before introducing this , I thought on using gem_has_lmem.
> > > But I was not sure in case every discrete platform mandatory to have lmem
> > region.
> > > If it is guaranteed that every discrete platform will have lmem memory region
> > ?
> > > I will  drop  this patch.
> >
> > So not sure what will happen in the future but till now and in the forseeable
> > future all dGfx platforms have LMEM, so gem_has_lmem seems fine.
> It seems better to carry with gem_has_lmem(), as IS_DGFX() requires to populate device info fields.
> This patch spined off from series https://patchwork.freedesktop.org/series/102780/
> I will use gem_has_lmem() instead of IS_DGFX().
> @rodrigo what is your opinion about it  ?
> >
> > I think if we introduce IS_DGFX() as in this patch, we should have an additional
> > patch to convert all tests using gem_has_lmem() to IS_DGFX() so that we don't
> > have multiple ways of tests deciding if something should run for dGfx.
> >
> > Or maybe both are also ok? So lmem related tests should use gem_has_lmem()
> > and non LMEM related tests should use IS_DGFX()?
>
> Actually, both of these methods even are not sufficient, if needed prior
> to open the drm_fd.  Ex. igt_device_find_first_i915_discrete_card()
> checks the igpu b:d:f, which breaks on one of real time issue.
> https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/issues/94

In that case maybe such API's which attempt to do this and don't work in
*all* cases should be deleted and something new provided? Anyway this seems
to be a different issue to what we were discussing? Thanks.

>
> >
> > So if we want to merge this let's look at other tests which are presently using
> > gem_has_lmem() and see if any need to be converted over to IS_DGFX(). Or to
> > keep things simple just use gem_has_lmem() for now?
> >
> > Thanks.