AMD DC test results

[Daniel Drake]

Hi,

Thanks for the hard work on AMD DC development! Here are some new test
results - hope they are interesting/useful.


CONTEXT

We have been tracking DC for a while as we work with multiple products
where amdgpu previously was not able to support the HDMI audio output.
We are hoping to ship the new driver as soon as possible to fix this
Linux compatibility issue.


TEST SETUP

With an eye to shipping this ASAP, we backported DC to the current
stable release, Linux 4.14. To do this we took the amdgpu/DC commits
from:
https://cgit.freedesktop.org/~agd5f/linux/log/?h=upstream-4.14-drm-next-amd-dc-staging-chrome

And added these two pull requests:
https://www.spinics.net/lists/dri-devel/msg156828.html
https://lists.freedesktop.org/archives/dri-devel/2017-November/157687.html

At this point we diff the resultant amdgpu directories with Linus tree
as of commit f6705bf959efac87bca76d40050d342f1d212587 (when he
accepted the DC driver) and the diff is zero. This is published on the
master branch of https://github.com/endlessm/linux

There have been more AMD DC changes merged into Linus tree since then,
however we tried and the backporting seems like loads of work, so we
have not gone beyond the initial merge into Linus tree for now.

We then asked our QA team to run our standard tests with this kernel.
They tested 3 platforms:
Acer Aspire E5-553G (AMD FX-9800P RADEON R7)
Acer Aspire E5-523G (AMD E2-9010 RADEON R2)
Acer Aspire A315-21 (AMD A4-9120 RADEON R3)

CONFIG_DRM_AMD_DC_PRE_VEGA was set in order to have AMD DC support
this hardware.


TEST RESULTS

HDMI audio works now, however we have encountered some regressions. In
all cases the 3 platforms behaved the same (either all 3 pass the
individual tests, or all 3 show the same problem). The regressions
are:

1. System hangs on S3 resume.
Bug can't be reproduced when booting with amdgpu.dc=0
Also reproduced with DC on Linus master (4.15-rc), but appears fixed
on agd5f/amd-staging-drm-next
https://bugs.freedesktop.org/show_bug.cgi?id=104281

2. Display corruption when using multiple displays
Bug can't be reproduced when booting with amdgpu.dc=0
Also reproduced with DC on Linus master (4.15-rc), but appears fixed
on agd5f/amd-staging-drm-next
https://bugs.freedesktop.org/show_bug.cgi?id=104319

3. HDMI audio device still shown as present and active even after
disconnecting HDMI cable
Bug can't be reproduced when booting with amdgpu.dc=0
Appears fixed with DC on Linus master (4.15-rc).

Full test results spreadsheet (including tests passed): http://goo.gl/oqnMDx
Thanks to my colleagues: Carlo Caione for doing the backport and
Vanessa Chang for managing the testing.


NEXT STEPS

Unfortunately we can't ship the driver in current state with these
regressions, but on the positive side it looks like it will be in
better shape for Linux 4.16.

We will now focus on making 4.15 regression-free before revisiting the
idea of backporting DC to earlier kernels.

After the holiday break we will try to identify the relevant fixes in
amd-staging-drm-next and see if they can be included in Linux 4.15.
We'll send these backports upstream as soon as we find them.


Daniel