Re: [PATCH v2 02/12] PCI: Disable pcie port on specific amd devices
From: Atharva Tiwari
Date: Tue Nov 17 2026 - 13:36:09 EST
> I don't know what's going on here, but it doesn't sound like typical
> PCIe behavior. If a device is powered off, accesses to it end with
> some kind of PCIe error that is not normally fatal.
>
> It's basically the same thing that happens when the OS enumerates
> devices by reading the Vendor and Device IDs. This has to work even
> if there's no device present, which is essentially the same as a
> device being powered off.
>
> So I'm not in favor of trying to avoid some accesses to devices based
> on DMI checks. Even if that works, it doesn't cover other paths like
> sysfs.
I agree, this is not standard behaviour, this only happens when we use
GMUX to turn off the rails (which is the thing that causes power savings).
Accessing PCIe space after GMUX has cut power to the rails will cause a
CPU CATERR.
Also i dont really have ideas, im not really familiar with PCI core.
Can you give ideas, you have please?