[Bug 91880] Radeonsi on Grenada cards (r9 390) exceptionally unstable and poorly performing

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=91880

--- Comment #167 from Thomas DEBESSE <dev at illwieckz.net> ---
Created attachment 134124
  --> https://bugs.freedesktop.org/attachment.cgi?id=134124&action=edit
kernel patch: set "high" default DPM profile instead of "auto" for 0x67B1
variant

So, using my 0x67B0 variant, new firmware and "auto" profile I was able to run
the vkQuake Vulkan game, the Unvanquished OpenGL game, the Unigine
Superposition OpenGL benchmark, the Luxmark Hotel OpenCL benchmark, and did
some OpenCL tasks with Darktable photo software. In fact in the past, just
running Gnome Shell (even the one run by GDM) was enough to take down the
computer.

I remember having tried the "auto" profile with new firmware in the past but I
had to to return to "low" or "high" profile because at this time the firmware
update just made the hang less immediate, but the hang was still happening
randomly, it was just happening lately. The bug appears to now be gone on my
end.

Everything looks fine on my 0x67B0 variant so it looks like the firmware and
some kernel updates did the trick for me (I'm now running Linux 4.12).

So, the only remaining variant known to be faulty is the 0x67B1 one.

So forget what I've said about the MSI 39 390X, it's now a model known to work.

Since the only affected model looks to be the 0x67B1 variant, I wrote this
small patch that must set "auto" as DPM profile on AMD GPU except for the
0x67B1 HAWAII variant which will use "high" as default. This patch does not
prevent the user to force the "auto" DPM profile by itself, but the "high" one
must now be set by default on this known variant.

I'm not able to test this patch since my GPU is the 0x67B0 one and looks to not
be affected by the bug anymore.

Before testing this patch, please check that you have the latest firmware for
your card. If not, update your firmware first as explained in previous comments
and check if it fixes the issue for you. If the latest firmware and a recent
kernel is not enough for you, well, perhaps we will have to mainline this
kernel change if AMD is not able to provide a fix.

This patch targets the 4.12 kernel tree but is so simple it must work on some
other versions too.

A review by someone at AMD like Marek Olšák, Alex Deucher or John Bridgman who
had participated in that thread would be very appreciated.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20170910/01ebe2e1/attachment-0001.html>