atomic debugfs files not created

[Drew Davenport]

On Mon, Jan 22, 2018 at 1:45 PM Alex Deucher <alexdeucher at gmail.com> wrote:

> On Mon, Jan 22, 2018 at 3:20 PM, Drew Davenport <ddavenport at google.com>
> wrote:
> > I was doing some debugging and found that /sys/kernel/debug/dri/0/state
> > doesn't get created for amdgpu. After a bit of digging around I found
> that
> > the DRIVER_ATOMIC feature flag gets set in dm_early_init in amdgpu_dm.c,
> > which is run after drm_debugfs_init where the atomic debugfs files would
> > have been created.
> >
> > I tried moving DRIVER_ATOMIC to the rest of the kms_driver.driver_feature
> > flags in amdgpu_drv.c, and this allowed the debugfs to get created as
> > expected. Nothing seemed terribly broken, but I didn't test thoroughly.
> >
> > Is there any reason not to move DRIVER_ATOMIC to the rest of the feature
> > flags? Thanks.
>
> It's only available with the new dc modsetting code which doesn't
> support all asics so you'd need to selectively enable it depending on
> the asic.  You'll only want to enable it if
> amdgpu_device_asic_has_dc_support() returns true.  I think I had a
> patch to do that at one point, but I don't know what I did with it.
>
> Alex
>

Right, that makes sense. It looks like this flag could be set according to
amdgpu_device_asic_has_dc_support right after the drm_device has been
allocated in amdgpu_pci_probe. Would that be a good place to set it?

Although, from looking at vi_set_ip_blocks and similar functions,
dm_ip_block currently won't get added (and thus DRIVER_ATOMIC won't be set)
if virtual display is enabled for some asics. Would it be a problem if
DRIVER_ATOMIC is set for those asics even if virtual display is enabled?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20180123/a740232e/attachment.html>