Trying To Run The AMDGPU Driver With A Hawaii GPU On Linux 4.5

This week for the public I had spun a Linux 4.5-rc1 kernel with AMDGPU PowerPlay enabled (since PP is also disabled by default in upstream now) and I also enabled the AMDGPU option for CIK GPU support. So tonight I decided to try running the AMDGPU DRM driver with a Radeon R9 290 rather than the Radeon DRM driver.
After installing that 4.5-rc1 kernel spin and then blacklisting the Radeon DRM driver (since it will still try to auto-load by default as it matches the hardware PCI ID), I booted with AMDGPU. However, I quickly realized things weren't working right when the R9 290 didn't mode-set to 4K.
Looking at the dmesg, the R9 290 Hawaii failed to initialize (early_init 3 failed -22, fatal error during GPU init). Thus my testing of AMDGPU on CIK quickly ended for now. If the AMD guys have any other ideas for getting AMDGPU to light up the card correctly, will happily give it another shot, but for now at least it's probably better just sticking to the default Radeon DRM driver.
Coming up later today though I have already prepared some Catalyst vs. Linux 4.5 benchmarks with AMDGPU (R9 Fury and R9 285) and Radeon DRM. On the Radeon/AMDGPU side is the freshest Mesa 11.2-devel + LLVM SVN, DRI3 enabled, and using this 4.5-rc1 kernel plus the kernel module parameter to ensure PowerPlay is working. Sadly there, the results show R9 Fury on open-source still leaving a lot to be desired. Additionally, on Linux 4.5 the R9 285 and R7 370 had problems with HDMI; once the system was booted, no monitor signal was being supplied to my 4K Dell display even though it worked fine on the R9 290 and R9 Fury... Fortunately, using DisplayPort between the graphics cards and the same 4K display ended up mode-setting fine and no signal loss. Anyhow, this Catalyst vs. open-source Linux 4.5 benchmarks on the R9 285 / R9 290 / R7 370 / R9 Fury tests will be out in just a few hours.
26 Comments