Created attachment 281161 [details]
acpidump from the system
When a PCIe hierarchy (connected through Thunderbolt cable) on Thinkpad X1 Carbon 6th generation (this possibly affects all such systems, such as latest Dell XPS) is runtime suspended to D3cold, there is wakeup followed by PME immediately bringing the hierarchy back to D0.
Commit 0e157e528604 ("PCI/PME: Implement runtime PM callbacks") tried to solve it but it caused a regression so current approach that seems to work is to disable Data Link Layer State Changed event as discussed in the following thread:
The way I can reproduce this is to connect Alpine/Titan Ridge Thunderbolt device to any of the TBT ports, approve it and wait for the hierarchy to runtime suspend into D3cold.
Created attachment 281163 [details]
dmesg from the system
Created attachment 281165 [details]
lspci -vv output from the system with TBT device connected