RADV Vulkan Driver Picks Up Several GFX10/Navi Fixes, Including To Address Random Hangs

[Stenka]

Rise of tomb raider hangs at 30s-1m to start on navi, hard reboot required.

[hakzsam]

Rise of tomb raider hangs at 30s-1m to start on navi, hard reboot required.

Yes, this one is a known issue. FWIW, It hangs with AMDVLK too.

[shmerl]

Yes, this one is a known issue. FWIW, It hangs with AMDVLK too.

Btw, I haven't tested this with latest master very recently, but it was still happening with radeonsi not long ago:

https://gitlab.freedesktop.org/mesa/mesa/issues/1427

[kuco]

I had freezes in Heroes of the Storm a few weeks back

Same here. RX560. Switching from the ACO-patched Mesa to Mesa-Git fixed the hangs for me.

[aufkrawall]

Same here. RX560. Switching from the ACO-patched Mesa to Mesa-Git fixed the hangs for me.

I was referring solely to the Navi 5700 XT though.
HotS runs completely stable for me on an RX 570 with or without ACO.

[donbastiano]

Speaking of random hangs, I have a 570 and I use the open driver. I had (rarely) to hard reboot but I was just using firefox, I didn't find anything in the logs, I enabled the ssh server so last time I tried to connect but without success.
Can I say it is not related to this fix, right? It hangs only while "under stress"? Have you tried to connect with ssh?
I was thinking that my case was more CPU related, 1st ryzen gen (but not the first faulty weeks of production), but maybe...

[aufkrawall]

Firefox is OpenGL and thus unrelated. Unless you got some other general issue, like the Ryzen C6 bug. But this is off topic here.

[tildearrow]

RX 580 as well. Started happening the day after the updated/reverted pkg-config stuff. Display cuts out, GPU ramps to full, hard lock, forced power-off required.

Exactly the same with Mesa 19.1 when using VA-API to encode. Exact same symptoms with the exception that the display doesn't cut out.
I mean, seriously, what causes the freezes?

[aufkrawall]

No issues with mesa-git and
ffmpeg -hwaccel vaapi -hwaccel_device /dev/dri/renderD128 -hwaccel_output_format vaapi -i input.mp4 -c:v h264_vaapi -global_quality 15 -c:a copy -map 0 output.mp4
here

[skeevy420]

Exactly the same with Mesa 19.1 when using VA-API to encode. Exact same symptoms with the exception that the display doesn't cut out.
I mean, seriously, what causes the freezes?

Don't know, for me it was perfectly fine until a day or so after the 2nd pkg-config/meson change and then my gaming went to hell in a hand-basket.

FWIW, I can tell you that it happens with and without the AMD bulk moves patch with kernel 5.2 and 5.3; with both LLVM9 and LLVM10; and randomly with proton-tkg, proton 4.11 and 4.2 -- when I get odd ass crashes, I'm thorough in testing what I've done that is non-standard....

Just compiled mesa master with llvm release/9.x branch. Gonna play Hitman 2 and I'll post back within the next 10 minutes if it hits me since that's about the longest I've played at a time in the past few days.

UPDATE: Loaded up the save that was crashing like clockwork and finally got SA/SO Master on Haven while poisoning them all. Been working on that for a few days due to the hangs effecting me. So far as of loader/dri3: do not blit outside old/new buffers it seems to be working for me. Used AMDVLK just then. Gonna restart and redo what I just did with RADV...I've poisoned these assholes enough times I have it down to an art.

Also should add that I have nir enabled and that there have been quite a few nir related commits between the hang build and now. Just realized that looking over the commit history.

UPDATE2: Worked just fine with RADV as well. Cool beans. Something committed between b3e3af0e374, when I first started getting hangs, and 0103d4747a3, my latest build from today, fixed whatever problems I've been having. My newest build that had hangs was 2236cf24a74 which is literally the commit before the commits mentioned in TFA . My previous working good commit/build before all of this was 780182f0a0c.

hakzsam Do you still need or want a bug report if it works now?