AMD Laptops With Radeon dGPU Graphics Receive Fix For Poor Linux Power Management

[Azultra]

Yes it is:

And then:

What you linked to explicitly proves that what I said is true: the ACPI implementation was supposed to say that Hotplug support in D3 is supported. It didn't.

The workaround they're doing now is to just ignore what ACPI says, and go ahead and assume that it will work out fine if ACPI supports power management for that PCIe port.

I thought the description was pretty clear on what the patch does: currently the kernel only enables D3 state support for a handful of hotplug devices, this behavior is in place because of issues with SkyLake Xeons, and the patch proposes to extend the whitelist to hotplug devices power-manageable by ACPI.

The issue isn't with the ACPI tables of the AMD laptops the patch intends to help, but with the kernel's timid behavior regarding the D3 state and hotplug devices.
Not really what your original post was saying, unless I'm reading it wrong. Anyway...

[remenic]

I thought the description was pretty clear on what the patch does: currently the kernel only enables D3 state support for a handful of hotplug devices, this behavior is in place because of issues with SkyLake Xeons, and the patch proposes to extend the whitelist to hotplug devices power-manageable by ACPI.

The issue isn't with the ACPI tables of the AMD laptops the patch intends to help, but with the kernel's timid behavior regarding the D3 state and hotplug devices.
Not really what your original post was saying, unless I'm reading it wrong. Anyway...

So you're saying that if D3 support was properly announced by ACPI it still wouldn't have worked because of the skylake workaround? I don't see why the skylake workaround would even need to be applied to non skylake systems but if it does I'm sure there's a good reason.

[tildearrow]

Why do vendors not care about ACPI?
What's the deal

Or do they do this on purpose to prevent other operating systems from working properly?

[agd5f]

Yes, but still kudos to the indep. dev. Though it should be official devs doing it, but then this likely relies on laptop vendor cooperatin and HW availability to the very devs.

FWIW, I proposed the initial patch to fix it, but Lukas's patch was a better fix.

The root of the issue comes down to how each OS determines whether to put pcie bridges into d3 or not. Windows seems to assume D3 is possible if the bridge has APCI power resources regardless of whether it's a hotplug capable bridge or not. Linux previously would only allow D3 on hotplug pcie bridges in certain specific cases. Lukas' makes the behavior match so far as we can tell since we can't see the windows source code.

[agd5f]

Why do vendors not care about ACPI?
What's the deal

Or do they do this on purpose to prevent other operating systems from working properly?

It was not a vendor specific issue. It comes down to how the OS determines when to enable D3 on pcie bridges. Windows is 95% of the market so Linux testing on a lot of OEM platforms is minimal if anything.

[mazumoto]

This is great news ... I happen to have bought a Dell G5 15SE with the 5600M - since it was the only Laptop at that time with a good screen, 16GB RAM, Zen 2 and AMD graphics. As it is indeed still the only laptop with that GPU, I am a little afraid that it is so niche, that it won't be supported well in Linux. But this fix indicates it might not be too bad :-)

[Guest]

This is great news ... I happen to have bought a Dell G5 15SE with the 5600M - since it was the only Laptop at that time with a good screen, 16GB RAM, Zen 2 and AMD graphics. As it is indeed still the only laptop with that GPU, I am a little afraid that it is so niche, that it won't be supported well in Linux. But this fix indicates it might not be too bad :-)

Best way to guarantee that something continues to work well, is to maintain it yourself. If you face any big problems, report, bisect and if no one else is willing, fix it yourself.