amdgpu driver probe error

[Alex Deucher]

On Sat, Sep 14, 2024 at 6:03 AM Athul Krishna
<athul.krishna.kr at protonmail.com> wrote:
>
> Device: ASUS Zephyrus G14 GA402RJ
> CPU: AMD R7 6800HS
> GPU: Radeon RX 6700S
> Kernel: 6.10.8
>
> Hello,
>
> I have encountered a problem with this specific device. So I'll explain the backstory a little bit.
>
> \_SB_.PCI0.GPP0.SWUS.SWDS.VGA_, this is the firmware path of the discrete GPU and \_SB_.PCI0.GPP0 is the hotplug bridge. An eject notification will remove all the devices under \_SB_.PCI0.GPP0, i.e, SWUS and SWDS(PCI-PCI bridges) and VGA_ will be gone. So will be their acpi_hotplug_context.
>
> So device check notification to VGA_, with the current implementation of acpiphp module will not be handled. I have modified the code so that device check notification to VGA_ will runtime resume \_SB_.PCI0.GPP0(pci_pm_runtime_resume() callback), which then will enable interrupt to the bridge, and the rest will be handled by pciehp module.
>
> Here's my problem, when device_attach() is called on the GPU, it spews out bunch of error. Most of the time I get:
>
> amdgpu 0000:03:00.0: amdgpu: SMU: response:0xFFFFFFFF for index:2 param:0x00000000 message:GetSmuVersion?
> amdgpu 0000:03:00.0: amdgpu: GPU mode1 reset failed
> amdgpu 0000:03:00.0: amdgpu: asic reset on init failed
> amdgpu 0000:03:00.0: amdgpu: Fatal error during GPU init
>
> very rarely:
>
> amdgpu 0000:03:00.0: amdgpu: PSP create ring failed!
> amdgpu 0000:03:00.0: amdgpu: Fatal error during GPU init
>
> Any help would be really appreciated. I have attached the dmesg output.

I'm not sure I understand what you are trying to do.  The kernel can
already put the device into runtime suspend (e.g., D3cold) when it's
idle without using hotplug.  It looks like the dGPU is not properly
powered on when the drivers tries to initialize the board after
whatever you did.

Alex

>
> Thanks,
> Athul