correct link speed & width for Navi

[Umio-Yasuno]

Recent AMD GPUs have multiple endpoints (Upstream Port, Downstream Port, GPU), and the actual {current,max} link speed & width is reflected on the upstream port.

Ref: https://docs.rs/libdrm_amdgpu_sys/latest/src/libdrm_amdgpu_sys/pci/bus_info.rs.html#87-93
Ref: https://gitlab.freedesktop.org/drm/amd/-/issues/1967#note_1347631

[ilya-zlobintsev]

How does the full sysfs path look like in this scenario? Is it something like /sys/bus/pci/devices/0000:0b:00.0/0000:0b:00.1/?
I'm wondering if there's a more universal way to detect the "root" device, so this could work for generations other than navi10 without having to hardcode the device ids of each one.

[Umio-Yasuno]

How does the full sysfs path look like in this scenario? Is it something like /sys/bus/pci/devices/0000:0b:00.0/0000:0b:00.1/?

/sys/devices/pci0000:00/0000:00:01.1/0000:01:00.0/0000:02:00.0/0000:03:00.0/../..

I'm wondering if there's a more universal way to detect the "root" device, so this could work for generations other than navi10 without having to hardcode the device ids of each one.

The reason there is an "Navi 10" in the name is because of the PCI ID Repository, the same device is also used for Navi2x and Navi3x.

https://linux-hardware.org/?probe=5d96610622

LACT (amdgpu-sysfs-rs) displays the link speed & width for Navi23 on my system as 16.0 GT/s PCIe x16 (Current) and 16.0 GT/s PCIe x16 (Max), but they are actually 8.0 GT/s PCIe x8 (Current) and 16.0 GT/s PCIe x8 (Max).

Also, since the root device differs depending on the system, I think searching from the GPU is the safest way.

[ilya-zlobintsev]

the same device is also used for Navi2x and Navi3x

In that case I'm fine with the current solution.