Linux 4.15 Is A Huge Update For Both AMD CPU & Radeon GPU Owners

[phoronix]

Phoronix: Linux 4.15 Is A Huge Update For Both AMD CPU & Radeon GPU Owners

Linux 4.15 is shaping up to be a massive kernel release and we are just half-way through its merge window period. But for AMD Linux users especially, the 4.15 kernel release is going to be rocking...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-4.15-AMD-Mega

[Iperpido]

Ryzen and R9 fury user here. waiting for it

[xiando]

I guess that's some good news while we wait for kernel bug #196683 to be fixed. Temperature readings is kind of a moot point when Ryzen CPU systems randomly hang in kernels 4.13+ - it really makes Ryzen's unfit and unusable for GNU/Linux.

DisplayPort and HDMI audio for AMD GPUs is nice, though. I'm still (ab)using SPDIF carrying AC3 (alsa plugin) on a 16 year old surround receiver due to the lack of support for anything better.

[torsionbar28]

DisplayPort and HDMI audio for AMD GPUs is nice, though. I'm still (ab)using SPDIF carrying AC3 (alsa plugin) on a 16 year old surround receiver due to the lack of support for anything better.

Most 16 year old receivers support SPDIF audio better than HDMI audio anyways. I had an old Sony receiver where the HDMI audio would randomly drop, but the SPDIF interface worked perfectly. Others reported the same, and it was believed to be a bug in an early version of HDMI, I think v1.1.

[RavFX]

I guess that's some good news while we wait for kernel bug #196683 to be fixed. Temperature readings is kind of a moot point when Ryzen CPU systems randomly hang in kernels 4.13+ - it really makes Ryzen's unfit and unusable for GNU/Linux.

DisplayPort and HDMI audio for AMD GPUs is nice, though. I'm still (ab)using SPDIF carrying AC3 (alsa plugin) on a 16 year old surround receiver due to the lack of support for anything better.

It crash on 4.13+ thanks to Intel, but you can fix it:
First, you need to be sure RCU_NOCB_CPU enabled in the kernel
Thanks to Intel patch that then pushed an other maintainer to remove CONFIG_RCU_NOCB_CPU_ALL, automatic workaround is not possible anymore in distrib using 4.13+ but you can still apply the workaround by adding a boot parameter "rcu_nocbs=0-X" where X = number of thread-1 (so, 15 for a Ryzen 7).

Once the Fix applied, weird video artefacting on your RX if you have one and crash that append mostly on idle will got away. And you will be able to turn C6 back to enabled in the BIOS without extras crashing.

[xiando]

It crash on 4.13+ thanks to Intel

He's actually a IBM asset,

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/init/Kconfig?h=v4.14-rc3&id=44c65ff2e3b0b48250a970183ab53b0602c25764

https://paulmck.livejournal.com/47528.html

..and I do (ab)use rcu_nocbs=0-11 on my R5 1600X's. It took me a while to figure it out, I initially solved it by simply sticking with 4.12 after trying 4.13. This one is pretty serious, to me 4.12 to 4.13 was going from "works" to "hangs right after boot". It's a big deal, specially on headless boxes.

It's a total scandal, specially because nobody seems interested in fixing it which results in - as you helpfully point out - that a boot parameter is required for these CPUs to work properly. That's not exactly good. Nobody from AMD has commented or acknowledged this problem and I'm guessing they won't until some Ubuntu LTS release updates to a 4.13+ kernel.

[theriddick]

Ryzen and R9 fury user here. waiting for it

The same here.

[nanonyme]

Intel is the root cause of Ryzen hangs:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=3fc5b3b6a80b2e08a0fec0056208c5dff757e547

https://bugzilla.kernel.org/show_bug.cgi?id=196683

Seth Jennings 2017-11-11 04:19:43 UTC
I found this commit that went into 4.12-rc2

Did anyone escalate the matter to Linus?

[lucrus]

xiando wrote: He's actually a IBM asset,

Intel is the root cause of Ryzen hangs:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=3fc5b3b6a80b2e08a0fec0056208c5dff757e547

https://bugzilla.kernel.org/show_bug.cgi?id=196683

I've read through all the kernel bug #196683 comments, and finally commenters think it's neither one (Intel/IBM), but they feel like it's a Ryzen CPU bug. However they're stuck at finding a way to contact the right AMD person to report the problem, and AMD hasn't commented on that bug as of today.

I do not own any Ryzen and I don't work for AMD, so there isn't much I can do to help here, except asking bridgman to have a look, maybe he knows how get in touch with the AMD CPU team.

[Chewi]

Yes, as I said in the bug report, let's please stop with the stupid baseless FUD. I discovered the RCU_NOCB_CPU workaround myself after a lot of trial and error but didn't exactly shout about it from the rooftops so it's highly unlikely whoever removed the kernel option even knew about the issue.