[RESOLVED] Anyone got AMD RX 5700 (XT) to work?

LargeDSI · 2019-09-24 06:20:17

I'm getting crashes with the latest "linux-amd-staging-drm-next-git" kernel in games like "Deus Ex:MD"(OGL) and the "Shadow of the Tomb Raider Trial"(Vulkan) with Steam, but also in Lutris with different games.

This is what shows up in the log each time:

Sep 24 07:47:37 zero-arch kernel: [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [CRTC:64:crtc-0] flip_done timed out
Sep 24 07:48:15 zero-arch kernel: [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [PLANE:57:plane-5] flip_done timed out

There is also several lines of trace output I could post if anyone is interested.

Does anybody have an idea what the problem could be?

jokeyrhyme · 2019-09-24 10:05:12

https://www.phoronix.com/scan.php?page= … rmware-Git

> AMD Navi 10 Firmware Finally Lands In The Linux-Firmware Tree

I noticed that the linux-firmware package has already been flagged as needing an update: https://www.archlinux.org/packages/core … -firmware/

> Besides needing these microcode files, for open-source Linux Navi graphics support you also should be on Linux 5.3+ and Mesa 19.2+ along with LLVM 9.0+

Whilst the kernel in core has been updated to 5.3, Mesa and LLVM still need AUR or the mesa-git repository, for now...

loqs · 2019-09-24 10:36:19

jokeyrhyme wrote:

Whilst the kernel in core has been updated to 5.3, Mesa and LLVM still need AUR or the mesa-git repository, for now...

Why is mesa-git still required? https://bbs.archlinux.org/viewtopic.php … 3#p1864863 reported success after only adding the missing firmware.

Lone_Wolf · 2019-09-24 11:31:46

The mesa 19.2 / llvm 9 requirement probably comes from this bug report , https://bugzilla.redhat.com/show_bug.cgi?id=1751884

As llvm 9 hasn't been released yet (think they're at RC3 now) people would need to build llvm from git * .

In theory they could build stable mesa against llvm-git .
However, llvm trunk changes fast and mesa frequently requires changes to succesfully build.
Those changes are added to master, but not backported.

Since llvm 8 was released I've been involved in 2 mesa-git build failures caused by changes in llvm trunk.
Those changes are also in the llvm 9 release candidates.

Until llvm 9 is released and officially supported by a stable mesa version , using mesa-git + llvm-git* seems best option ,

*
Afaik there are no llvm packages for arch that build RC versions.
For llvm-git there are several good options , same applies to mesa .

jokeyrhyme · 2019-09-24 11:35:07

LLVM 9.0 is stable now: https://releases.llvm.org/download.html#9.0.0

Lone_Wolf · 2019-09-24 12:48:11

And it appears to have entered [staging] repos, https://www.archlinux.org/packages/staging/x86_64/llvm/

NOTE : [staging] is NOT meant for end-users !

marmalade · 2019-09-25 04:39:41

Is mesa-git still a requirement? I seem to have gotten up and running without installing it and just installing the firmware drivers, however my cursor's framerate seems like it's at 30 or something, and my monitor only at 60 (its a 240hz).

Edit: Disregard, I managed to just change the refresh rate with xrandr. Looks like it's working pretty well! The cursor is a little choppy, but not by much.

Last edited by marmalade (2019-09-25 04:46:56)

loqs · 2019-09-25 11:09:44

Cursor stuttering may be fixed by 5.3.2 https://git.kernel.org/pub/scm/linux/ke … waps.patch

jokeyrhyme · 2019-09-25 19:56:50

Mesa 19.2.0 has officially been released with Navi10 support: https://mesa3d.org/relnotes/19.2.0.html

This bit confuses me:

> GL_EXT_shader_image_load_store on radeonsi (with LLVM >= 10)

On the LLVM 9.0 release notes, they have: https://releases.llvm.org/9.0.0/docs/Re … gpu-target

> Support for gfx10

So, maybe 9.0 is the minimum, but this one OpenGL extension requires LLVM 10.0 ?

Anyhow, it's great that more things are moving into stable

loqs · 2019-09-25 20:47:32

jokeyrhyme wrote:

So, maybe 9.0 is the minimum, but this one OpenGL extension requires LLVM 10.0 ?

https://github.com/mesa3d/mesa/commit/f … 87b29335e1

NivEU · 2019-09-26 09:21:47

Hi! It's nice to see that things are moving along, but I still got the same problem as @archixxx and see that he also uses a Pulse.
Is the high power consumption in idle just Pulse related? I installed already a newer version of the drm next and the problem still persists.

My system:
ASUS Z170, 5700XT Pulse, i5-6600K and an old Samsung SynchMaster 2433BW 60Hz monitor

What I did so far:
I followed the article https://wiki.archlinux.org/index.php/Navi_10.

Aka I installed the firmware from AUR, linux-firmware-agd5f-radeon-navi10.

Added mesa-git to the pacman.conf and installed these libs:

amdvlk clang-git libclc-git libdrm-git llvm-git llvm-libs-git mesa-git opencl-mesa-git vulkan-mesa-layer-git vulkan-radeon-git xf86-video-amdgpu-git lib32-mesa-git lib32-mesa-git lib32-vulkan-radeon-git lib32-vulkan-mesa-layer-git

The system does hang sometimes and the power usage is quite high in idle mode (28-38W).

Additionally if I monitor the fan speed and power usage via conky the journal is being spammed with these error messages:

Sep 25 16:54:10 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2 param 0x80
Sep 25 16:54:10 windowmaker kernel: amdgpu: [powerplay] Failed to export SMU metrics table!
Sep 25 16:54:12 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2, param 0x80
Sep 25 16:54:12 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2, param 0x80
Sep 25 16:54:14 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2 param 0x80
Sep 25 16:54:14 windowmaker kernel: amdgpu: [powerplay] Failed to export SMU metrics table!
Sep 25 16:54:14 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2 param 0x80
Sep 25 16:54:14 windowmaker kernel: amdgpu: [powerplay] Failed to export SMU metrics table!

This is how I read the fan speed and power usage:

cat /sys/class/drm/card0/device/hwmon/hwmon1/fan1_input 
cat /sys/class/drm/card0/device/hwmon/hwmon1/power1_average | awk '{print $1/1000000}'

I set the execi timer to every 5 seconds, though. On top of that the timer expires after a short while and gives me these errors:

Sep 25 16:56:43 windowmaker conky.desktop[2920]: cat: /sys/class/drm/card0/device/hwmon/hwmon1/fan1_input: Timer expired
Sep 25 16:56:43 windowmaker conky.desktop[2920]: cat: /sys/class/drm/card0/device/hwmon/hwmon1/power1_average: Timer expired

The whole system starts to hang then. The powerplay errors only happen when reading out the hardware monitor values, so as long as I don't
read them I have no error messages and the system seems to work fine except the occasional slowness and high power consumption.

Then I installed as suggested the AUR package linux-amd-staging-drm-next-git (I chose drm-next-5.4-2019-08-30) which is supposed to fix power consumption.

My results now:

Idle power usage is still 28-30W.
Still getting the error messages but in arbitrary intervals.

Sep 25 18:04:33 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0x12, response 0xfffffffb param 0x6
Sep 25 18:06:28 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xf, response 0xfffffffb, param 0x4c6000
Sep 25 18:06:33 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xf, response 0xfffffffb, param 0x4c6000
Sep 25 18:08:03 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0x12, response 0xfffffffb param 0x6
Sep 25 18:08:08 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xfffffffb param 0x80
...
Sep 25 18:12:12 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2, param 0x80
Sep 25 18:12:14 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2 param 0x80
Sep 25 18:12:14 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2 param 0x80
Sep 25 18:12:20 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2, param 0x80
Sep 25 18:12:20 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2, param 0x80
Sep 25 18:12:22 windowmaker kernel: amdgpu: [powerplay] Failed to send message 0xe, response 0xffffffc2 param 0x80

Reading the power consumption fails with

Sep 25 18:10:19 windowmaker conky.desktop[1006]: cat: /sys/class/drm/card0/device/hwmon/hwmon1/power1_average: Input/output error
Sep 25 18:11:58 windowmaker conky.desktop[1006]: cat: /sys/class/drm/card0/device/hwmon/hwmon1/power1_average: Timer expired

Opening applications (e.g. chromium) is sluggish.

Any tips on what to do next? Sorry for the long post.

blisbell · 2019-09-26 19:32:35

I'm having the same issues with monitoring system with conky and i use

${hwmon 2 temp 1}°C
${hwmon 2 fan 1} RPM

this bug report https://bugs.freedesktop.org/show_bug.cgi?id=111481
this bug report https://bugs.freedesktop.org/show_bug.cgi?id=111555
this bug report https://bugs.freedesktop.org/show_bug.cgi?id=111763

These seem to cover all the hangs that people are having, hopefully they will get them fixed soon.

archixxx · 2019-09-27 10:19:41

@NivEU Idle power consumption is fine for me now at 8W. But you need the patch set for kernel 5.4. Kernel 5.3 forces the card to stay at around 30W regardless what you do. I'm currently using this kernel: https://cgit.freedesktop.org/~agd5f/lin … 2019-08-30 . There're two newer ones but I haven't tested them: https://cgit.freedesktop.org/~agd5f/linux/

EDIT: If you're lucky maybe "aur/linux-mainline-5.4rc1" will pop up soon as merge window for kernel 5.4 comes slowly to it's end. I would assume that this one alos contains everything needed to reduce idle power consumption.

Last edited by archixxx (2019-09-27 10:22:55)

katharina · 2019-09-27 10:45:49

@archiixxx Are those changes already accepted to be merged into 5.4? It would suck if we had to wait for 5.5.

Also, is there some article detailing on how to patch and compile your own Arch kernel with a patch set like this? I'm personally familiar with building software from source but have never compiler a kernel, much less so a patched one.

Cheers

NivEU · 2019-09-27 11:04:17

@archixxx I'll give those a try then. Thanks for sharing!

EDIT I was an idiot and I forgot to set the initramfs etc in my arch.conf. Now my idle power usage is 7W (yay). For fixes for the powerplay errors I'll wait. Thanks again!

Last edited by NivEU (2019-09-27 11:34:00)

archixxx · 2019-09-27 11:31:28

@katharina I used "aur/linux-amd-staging-drm-next-git" (use your favorite AUR helper like "yay"). When asked "PKGBUILDs to edit?" I answered "[A]ll" and changed what is described here: https://bbs.archlinux.org/viewtopic.php … 9#p1860189 . Save the changes and the kernel builds. I also installed "aur/linux-amd-staging-drm-next-git-headers" because I need it for various other packages. Here I also applied the same changes. Then you may also need to adjust your boot loader to load the new kernel.

But don't be surprised: You'll get 5.3.0-rc3-amd-staging-drm-next-git when you compile the sources above.

If the changes are accepted for 5.4 I've no idea TBH ;-) I haven't had a look at the merges for kernel 5.4. I'll try 5.4rc1 when it's ready and if it doesn't work I switch back.

Last edited by archixxx (2019-09-27 11:32:04)

progandy · 2019-09-27 11:51:59

archixxx wrote:

If the changes are accepted for 5.4 I've no idea TBH ;-) I haven't had a look at the merges for kernel 5.4. I'll try 5.4rc1 when it's ready and if it doesn't work I switch back.

Those tags have been merged for 5.4:
https://git.kernel.org/pub/scm/linux/ke … 881fc751e4
https://git.kernel.org/pub/scm/linux/ke … 8898bd4c6e

archixxx · 2019-09-27 12:34:04

https://git.kernel.org/pub/scm/linux/ke … 881fc751e4 is based on drm-next-5.4-2019-08-30 and I'm using this source since a few weeks and it works quite nice. So I would assume 5.4rc1 should be in a good shape at least for Navi10 :-)

katharina · 2019-09-27 12:54:41

@archixxx thanks so much for the write up! I got it to work and my power usage issues are fixed

marmalade · 2019-09-28 04:08:33

Following the current Navi 10 wiki page, I get really terrible performance with Vulkan (Dota 2) and my OpenGL games don't even start.

Anyone managed to get this to work? I'm using the mesa-git repo, the latest stable kernel, (5.3) and the latest stable linux-firmware, again just as illustrated in the wiki. Not sure what else to do.

Thanks!

jokeyrhyme · 2019-09-28 07:44:32

marmalade wrote:

the latest stable linux-firmware

The latest linux-firmware in core does not have the correct firmware

The version in testing does, as does the specific package on AUR

Althorion · 2019-09-28 08:54:40

I’m using the AUR package and also have that terrible performance—framerate’s so low that navigating menus is difficult, so probably <5 fps. Tried adding and removing amdvlk, to no avail.

gabmus · 2019-09-28 09:47:19

Althorion wrote:

I’m using the AUR package and also have that terrible performance—framerate’s so low that navigating menus is difficult, so probably <5 fps. Tried adding and removing amdvlk, to no avail.

What DE are you using? In GNOME Wayland I haven't been seeing any issues. Also, I just installed linux-firmware from testing, everything seems smooth

katharina · 2019-09-28 22:12:54

Hello,

I've been seeing a very weird issue with the 5.4 patch set and mesa-git. If I install the Navi10 firmware files, I get very weird visual artifacts, with everything being way too bright and weird outlines around text, kinda like a sharpening filter gone rogue:

https://i.imgur.com/8yag2hf.png

But if I delete the firmware and reboot, it is always fixed:

https://i.imgur.com/RiNrXTv.png

Does anybody have an idea what this could be caused by? Thanks in advance!

Cheers,
Katharina

moderator edit -- replaced oversized images with links.
Pasting pictures and code

Edit: I am sorry for the oversized image embeds!

Last edited by katharina (2019-09-28 22:29:17)

Althorion · 2019-09-29 09:20:37

gabmus wrote:

What DE are you using? In GNOME Wayland I haven't been seeing any issues. Also, I just installed linux-firmware from testing, everything seems smooth

Tried on kwin (Plasma/KDE) and Openbox, no difference.

But here’s a funny part—it happens only to Steam games. I’ve run Battle for Wesnoth from repository without problems, same for the Tyranny from GOG and they run smoothly. I’ve tried turning off the Steam in-game overlay and using/not using native libraries instead of bundled, no change.

Any ideas what to try next?

EDIT:
It’s not Steam. I’ve tried with Wesnoth from Steam and it too is smooth. I guess it’s Vulkan, then? Never knew Stellaris uses it, though…

Last edited by Althorion (2019-09-29 09:29:53)

Arch Linux

#76 2019-09-24 06:20:17

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#77 2019-09-24 10:05:12

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#78 2019-09-24 10:36:19

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#79 2019-09-24 11:31:46

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#80 2019-09-24 11:35:07

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#81 2019-09-24 12:48:11

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#82 2019-09-25 04:39:41

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#83 2019-09-25 11:09:44

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#84 2019-09-25 19:56:50

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#85 2019-09-25 20:47:32

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#86 2019-09-26 09:21:47

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#87 2019-09-26 19:32:35

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#88 2019-09-27 10:19:41

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#89 2019-09-27 10:45:49

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#90 2019-09-27 11:04:17

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#91 2019-09-27 11:31:28

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#92 2019-09-27 11:51:59

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#93 2019-09-27 12:34:04

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#94 2019-09-27 12:54:41

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#95 2019-09-28 04:08:33

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#96 2019-09-28 07:44:32

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#97 2019-09-28 08:54:40

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#98 2019-09-28 09:47:19

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#99 2019-09-28 22:12:54

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

#100 2019-09-29 09:20:37

Re: [RESOLVED] Anyone got AMD RX 5700 (XT) to work?

Board footer