Created attachment 292541 [details] otcpl-dell-9380-cfl_freeze.html pcieport devices are taking longer than 1 second across several machines. This bug is there to monitor which machines and kernel release this happens on.
Created attachment 292543 [details] issue.def
This bug seems to have started in 5.8.0-rc4. It's affecting multiple machines, such as: otcpl-dell-7390-cmlu otcpl-dell-9380-cfl lenb-Dell-XPS-13-9300 otcpl-dell-xps-13-9300 otcpl-hp-elite-x2-kbl otcpl-dell-p5510-xeon-1 otcpl-dell-p3520 lenb-Intel-NUC8i7HVKVA otcpl-tgl-rvp-u otcpl-dell-7390-cmlu
Created attachment 292577 [details] otcpl-dell-7390-cmlu_freeze.html
Created attachment 292579 [details] otcpl-dell-p3520_freeze.html
Created attachment 292581 [details] otcpl-dell-p5510-xeon-1_freeze.html
Created attachment 292583 [details] otcpl-hp-elite-x2-kbl_mem.html
Mika, should I run a bisect on this between 5.8-rc3 and 5.8-rc4?
Yes, I think that would help. I did not see any changes in drivers/pci between v5.8-rc3 and v5.8-rc4 which is surprising.
Actually I didn't go far back enough, it apparently disappeared from 5.8-rc1-3 but reappeared in 5.7.0. I'll dig deeper to see where it really began.
I have an Icelake Dell 9300, which comes pre-installed with Ubuntu. It reproduces this issue 100% of the time, no matter what kernel is used. Dell shipped 5.0, and I've tested every kernel 5.4 - 5.10. I'll be happy to run experiments on this system if it may result in making this problem go away -- even if under limited conditions or configurations.
IIRC you guys have tools that can show where the resume time is spent? Maybe you can run that one and we can then go it over? The BIOS ASL methods for the TBT PCIe root ports have some delays but for those we cannot affect from the kernel side. Also is this with device connected to any of the TBT ports or not?
Looking at https://bugzilla.kernel.org/attachment.cgi?id=292541 (otcpl-dell-9380-cfl_freeze.html). I don't really know how to interpret this graph. Is it telling me that \_SB.PCI0.RP09._PS0 took 1583ms to execute? Is that the bulk of the delay? If so, I'm not sure what the kernel can do about that.