Arch Linux

CONFIG_DRM_AMDGPU is already set, so this is really just CONFIG_DRM_AMDGPU_CIK. From what I can see, this is still experimental.

I don't know why it's still listed as experimental. It works 100% perfectly for me, honestly, I've no idea why it's still considered experimental.

Sure it works, until you try something and your system catches fire :D

No seriously, it works perfectly fine. It works perfectly fine for everyone else that's done it. I feel like this would benefit more Linux users that use AMD cards, because they have a choice instead of being stuck with radeon.

You say it works perfectly fine, but the authors don't.

I don't understand why they don't believe so. It's pretty much non-experimental at this point, actually, it should've been out of experimental for a while now I believe.

Then you might want to figure out why they believe so, rather than blindly enabling it and hoping for the best.

Blindly enabling it? I just said there's no issues with amdgpu for me, and the others that have enabled it manually, like the issues radeon has. I always got black screens that crashed my system with the radeon driver, and poor performance in games, and I haven't had any of that with the amdgpu driver. Performance is pretty solid, there's no reason this should still be experimental. It works perfectly.

Just pitching in to state my interest in having this enabled for official kernels.
Experimental support for SI cards is coming to 4.9 with CONFIG_DRM_AMDGPU_SI, and as far as I know (from my own tests), the radeon module will be used by default unless explicitly configured not to (eg, by blacklisting it).
Enabling these options would allow using the experimental code without recompiling a kernel, while preserving the current setup by default.

Note that Southern Islands (CONFIG_DRM_AMDGPU_SI) is now enabled in the Arch package (4.9-1 in [testing]). https://git.archlinux.org/svntogit/packages.git/tree/trunk/config.x86_64?h=packages/linux#n4978
(Sea Islands/CIK is still disabled, even though it is considered just as "experimental")

It does seem weird to enable AMDGPU_SI and not AMDGPU_CIK, when CIK is actually in better shape than SI. The way the kernel config stands now, AMDGPU is only enabled for cards older than Radeon 7000 and newer than r9-380, and there is a weird void where GCN 1.1 isn't enabled.

My two cents are to set AMDGPU_SI=y and AMDGPU_CIK=y, because even when both are enabled, the kernel will still default to the radeon kernel driver unless it is explicitly blacklisted. Thus, no unwitting victim will have a bad experience with an experimental driver (as they will still be defaulted to radeon), and those of us who want to test AMDGPU can do so by blacklisting the radeon driver. All that is accomplished by keeping SI and CIK disabled is forcing us to have to compile the kernel ourselves to change this one thing

@brittyazel: completely agree

Is this serious? They enabled Southern Islands, an even more experimental setting, but not Sea Islands, which is definitely working completely at this point?

Do you know if the Linux maintainers have seen this request? It would be a real shame to not have this addressed (in either one way or another) before the next kernel release, given that it is such a small change on their part

Now in Linux 4.9.3-1, both SI and CIK are enabled: https://git.archlinux.org/svntogit/packages.git/tree/trunk/config.x86_64?h=packages/linux#n4975
(fixed by this commit: https://git.archlinux.org/svntogit/packages.git/commit/trunk?h=packages/linux&id=992ffce2aa59f0102f795c92b1e9e525e777164f )

So I guess that means that this bug can be closed.

I agree, testing AMDGPU on my CIK card is yielding fantastic results. Glad to see this problem solved