AMDGPU On Linux 4.18 To Offer Greater Vega Power Savings, DisplayPort 1.4 Fixes

[Chewi]

I've seen this suggested before and I'd like to learn more about this issue.

I hope you like reading...

[boxie]

Partially implement amdgpu driver in mainline kernel again. More patches in 4.19-wip kernel: https://cgit.freedesktop.org/~agd5f/...-next-4.19-wip

I am curious what the Dentist Frequency is https://cgit.freedesktop.org/~agd5f/...a2d0038ef596d5

[shmerl]

I simply ended up replacing Ryzen 7 1700X with 2700X to get rid of those C6 related system hangs. It works fine after that. That option of "common current idle" helps for 1700X, but it's disabling package C6 so it's not the best solution.

[cybertraveler]

I am curious what the Dentist Frequency is https://cgit.freedesktop.org/~agd5f/...a2d0038ef596d5

Dentists maintain mouths and mouths emit sounds. So guess is this is something to do with some code that maintains HDMI audio for AMD cards.

[Jeff99]

What is the exact status of SR-IOV for AMD GPUs?
Is something along the lines of a Windows 10 guest on a Linux host/hypervisor with a single Vega even remotely conceivable at this point, or are we just talking about headless GPGPU device partitioning for cloud folks?

[dwagner]

My new Ryzen 2700U-powered Acer laptop freezes several times a day. I found a bug report from someone else with the same hardware just the other day and they were lucky enough to grab some kernel debug logging. I had thought it was CPU-related, much like the problems we've seen with Ryzen on the desktop, but the logging indicates a graphics-related issue so I'm praying a fix will make it in soon.

I have experienced amggpu.dc=1 related system freezes from the very moment I started to use amdgpu, so for a year now, and unluckily, they have become more frequent over time rather than getting fixed. I'm sad to say I've kind of lost hope there will be a stable driver anytime soon.

[dwagner]

AMD blamed PSU not being able to keep up with this behaviour

Given that the whole rest of the system still consumes considerable power - like > 30 Watts - even when the CPU enters C6, this is one of the most bizarre blamings I have ever heard of. PSUs failing because a system goes from like 90 to 30 Watt power consumption - seriously, whom do they want to fool?

[bridgman]

I have experienced amggpu.dc=1 related system freezes from the very moment I started to use amdgpu, so for a year now, and unluckily, they have become more frequent over time rather than getting fixed. I'm sad to say I've kind of lost hope there will be a stable driver anytime soon.

Which GPU is this happening on ?

Put differently, is it on a chip where we enable DC by default ? If yes, is there a bugzilla ticket we could look at ?

[dwagner]

Which GPU is this happening on ?

RX460

Put differently, is it on a chip where we enable DC by default?

I think dc=1 has been the default for this GPU for a while now - without it, it lacks the ability to drive a 4k display at 60Hz, which is kind of a show stopper for dc=0.

If yes, is there a bugzilla ticket we could look at ?

The original bug I posted long ago was https://bugs.freedesktop.org/show_bug.cgi?id=102322 - the same symptom (screen going one-colored, system crashes, sometimes a few processes unrelated to any graphics output seem to run for a little longer before also stopping) has also been reported by others.

Unluckily, with the most recent kernels (both mainline 4.17rc5 and drm-next) the frequency of such crashes (unluckily most often without any dmesg output) is "every few minutes" - so totally unusable.

[bridgman]

Ouch. OK, thanks - I'll ask around internally and see what we know about it.

The ticket doesn't mention anything about the problem only happening with amdgpu.dc=1... if that is the case it seems to be a really important bit of information and totally changes the way we would approach the problem.

Could I bother you to update the ticket with current information ?