[Question] How ASPM should be programmed for devices under VMD and TBT ports

[Kai-Heng Feng]

ASPM is not configured on devices under VMD or devices get hotplugged to TBT ports.

I know it should be done by the OS, but is it documented somewhere whey the config space can't be programmed by BIOS/firmware? We need some compelling evidences to let Linux upstream community accept such approach.

[Neo Wong (neowong-taiwan)]

Do you have existing BZ ticket discussing w/ upstream community ? That will be easy to get attention from engineering.

[Neo Wong (neowong-taiwan)]

Below should be a start to talk w/ community ?

According to PCI_Firmware_v3.3_20210120_NCB.pdf(Can be download from PCISIG), OS will report is it support ASPM or not via ACPI _OSC method, refer :

https://github.com/torvalds/linux/blob/8d025e2092e29bfd13e56c78e22af25fac83c8ec/drivers/acpi/pci_root.c#L448

If ACPI method _OSC report it support ASPM, ASPM should be take control by OS itself.

[Kai-Heng Feng (kaihengfeng)]

> If ACPI method _OSC report it support ASPM, ASPM should be take control by OS itself.
Totally.

But OS also read programmed ASPM register in config space to decide if ASPM should be enabled.

Right now NVMe driver under VMD and hotplugged PCIe devices don't have ASPM register programmed.

So I'd like to know how the VMD/TBT driver action upon unprogrammed ASPM bits under Windows.