Re: [BUG] ASUS ProArt PX13 HN7306WU: amd_pmc s2idle S0ix corrupts AMD 1022:150b root port, NVIDIA dGPU returns header type 7f

[Mario Limonciello]

On 4/3/26 4:05 PM, Mario Limonciello wrote:
> On 4/3/26 3:47 PM, Joyful Lee wrote:
> > On Fri, Apr 3, 2026 at 4:03 PM Mario Limonciello
> > <mario.limonciello@xxxxxxx> wrote:
> > > Can you do the exact same on the same branch/commit but without pciehp
> > > enabled?  It will make comparison of all the impacts (not just this root
> > > port) a lot easier.
> >
> > Here you go. Unfortunately it's not *exactly* the same because I had to
> > disable the nvidia driver to get a result.
>
> Thanks - the most obvious observation is what you already noted - 
> without hotplug support the root port skips PM all together.
>
> [24.575705] pcieport 0000:00:01.2: PCI PM: Suspend power state: D0
> [24.575707] pcieport 0000:00:01.2: PCI PM: Skipped
> [24.575709] pcieport 0000:00:01.1: PCI PM: Suspend power state: D0
> [24.575709] pcieport 0000:00:01.1: PCI PM: Skipped
>
> vs
>
> [29.279987] pcieport 0000:00:01.1: power state changed by ACPI to D3cold
> [29.279991] pcieport 0000:00:01.1: PCI PM: Suspend power state: D3cold
> [29.280214] pcieport 0000:00:01.2: power state changed by ACPI to D3cold
> [29.280219] pcieport 0000:00:01.2: PCI PM: Suspend power state: D3cold
>
> The hotplug support seems to apply to 3 ports.
>
>    0000:00:01.1 (Bus 01):
>    - Slot Capabilities: 0x00040060
>    - Flags: HotPlug+ Surprise+ LLActRep+
>
>    0000:00:01.2 (Bus 61):
>    - Slot Capabilities: 0x00040060
>    - Flags: HotPlug+ Surprise+ LLActRep+
>
>    0000:00:03.1 (Bus c4):
>    - Slot Capabilities: 0x000425e0
>    - Flags: HotPlug+ Surprise+ LLActRep+
>    - This port also has bandwidth control enabled: bwctrl: enabled with 
> IRQ 38
>
> The port stays in D0 because the child (the GPU) stays in D0 for suspend.
>
>    [24.599889] pcieport 0000:00:03.1: PCI PM: Suspend power state: D0
>
>
>    [24.599890] pcieport 0000:00:03.1: PCI PM: Skipped
>
> So I would say we should look at what the GPU isn't going to D3 for 
> suspend for the next clue.

OK - so the issue comes from acpi_pci_bridge_d3() which is called from 
platform_pci_bridge_d3():

	if (acpi_pci_disabled || !dev->is_pciehp)
		return false;