Successful GPU passthrough, but wonder Why/How?

[meowthink]

This tutorial is amazing. I've got GPU passthrough working on a T-Bao GOD77 (7735HS, 680M/Rembrandt), despite I had to enable the device manually in Device Manager sometimes.

But the question here is, how it actually works?

As a GPU of RDNA2, there's barely nothing inside the vendor-reset kernel module, nothing newer than navi10/12/14 (RDNA1) inside actually. Another evidence is command like
echo device_specific >/sys/bus/pci/devices/0000\:e5\:00.0/reset_method leads to -bash: echo: write error: Invalid argument
(e5:00.0 is my GPU)
Besides, the pci remove-reset procedures in vreset.service did nothing help. I tried remove such service and found nothing different.

The RadeonResetBugFixService is a must.
The magic here is: when the guest shutting down, it will switch guest GPU driver to a basic one, and vice versa when booting up.

But this isn't enough. I used to think the vendor-reset kernel module worth nothing, but soon to find myself wrong.

Without any of RadeonResetBugFixService or vendor-reset module, things won't work after reset/shutdown the guest. If I got a display attached, it will turn asleep, showing no signal. Another powerup of the guest after won't let any signal on, showing that the GPU didn't reset correctly, and I can find Error 43 in Device Manager through Remote Desktop.

Thing only got work with both RadeonResetBugFixService and vendor-reset module. This is so wired. Any clue?