Created attachment 459824 [details, diff] Patch adding S.Islands support to the ebuild As of kernel version 4.9, AMDGPU supports AMD's Southern Islands chipsets (Cape Verde, Pitcairn, Tahiti, Oland, Hainan) but the corresponding firmware files are not installed by the sys-firmware/amdgpu-ucode ebuild, in contrast to all other chipsets supported by AMDGPU. This is an easy problem to work around because sys-firmware/radeon-ucode also contains the required firmware files, so people with Southern Islands chips can just install radeon-ucode instead of amdgpu-ucode. Still, it seems inconsistent, and a discussion from the forums (https://forums.gentoo.org/viewtopic-t-1057522.html) suggests that the ebuild should be changed. Could we either get that change made, or get a resolution that it shouldn't be changed and SI chipset users need to continue to use radeon-ucode instead? I'm attaching a possible patch for amdgpu-ucode-20160628.ebuild.
fixed in sys-firmware/amdgpu-ucode-20161121
The wiki (https://wiki.gentoo.org/wiki/AMDGPU#Incorporating_firmware) states that you need TAHITI_uvd.bin and TAHITI_vce.bin for the pitcairn chipset which is not included with this ebuild. "TAHITI" needs to be added to AMDGPU_LEGACY_CIK
(In reply to timmy welch from comment #2) > The wiki (https://wiki.gentoo.org/wiki/AMDGPU#Incorporating_firmware) states > that you need TAHITI_uvd.bin and TAHITI_vce.bin for the pitcairn chipset > which is not included with this ebuild. > > "TAHITI" needs to be added to AMDGPU_LEGACY_CIK You're right, but on a whim I tried compiling a version of my kernel without the TAHITI firmware files and booting it. It works fine and there's nothing showing up in dmesg or the system log about missing microcode. And on the wiki, the firmware list was just copied from the article for the (older) radeon driver, but I don't know that anyone ever explicitly checked whether those files were required. So I wonder if they're even needed. Obviously this bug report is not the place to resolve that, but this is probably as good a *time* as any to check it. I'll put something on the AMDGPU article's talk page about this.