[Bug 221073] xHCI host controller dies on resume from s2idle on AMD Strix Halo [1022:1587]

[bugzilla-daemon@kernel.org]

https://bugzilla.kernel.org/show_bug.cgi?id=221073

--- Comment #23 from David Hubbard (david.c.hubbard@gmail.com) ---
Created attachment 309526
  --> https://bugzilla.kernel.org/attachment.cgi?id=309526&action=edit
comment16-dmesg02.gz

I'm fascinated with diagnosing the xHC USBSTS with the EINT bit set, using the
instructions in comment #5. I'm zeroing in on that. Maybe the issues with
resume taking 10 seconds in comment #19 can be considered separately in a
different bug?

My kernel has hand-applied patches I backported from
https://github.com/mkopec/linux/tree/hdmi_frl to 6.17.7 (ROCm 6.4 is still 10%
faster than newer ROCm versions). I also applied the Mediatek driver patches in
https://zbowling.github.io/mt7925/.

Anyway, I took a few interesting captures:

Cap1: used power button to resume. Can anyone else confirm this is a
workaround? The bug did not appear for this attempt.

* For some reason, debugfs files were all 0 in size?

Cap2: used space bar on keyboard to resume. Bug appeared.

* For some reason, debugfs files were all 0 in size?


# free
               total        used        free      shared  buff/cache  
available
Mem:        32636908     4008008    27282920       35664     1764500   
28628900
Swap:              0           0           0

# (this shows 32GB free because the BIOS setting is iGPU = custom, iGPU memory
= 96GB)
# echo "module xhci_hcd +p" >/proc/dynamic_debug/control
# systemctl suspend
# (wait 2 minutes, press power button to resume, no bug)
# tar czf cap01.tgz /sys/kernel/debug/usb/xhci/0000\:c2\:00.4
# dmesg | gzip > dmesg01.gz
# systelctl shutdown

# (press power button, wait for system to boot)
# echo "module xhci_hcd +p" >/proc/dynamic_debug/control
# systemctl suspend
# (wait 2 minutes, press spacebar, get this in dmesg)
[  243.002269] xhci_hcd 0000:c2:00.4: xHCI host controller not responding,
assume dead
# (unplug usb hub from c2:00.4 and plug it into c4:00.0, keyboard and mouse
alive again)
# tar czf cap02.tgz /sys/kernel/debug/usb/xhci/0000\:c2\:00.4
# dmesg | gzip > dmesg02.gz

-- 
You may reply to this email to add a comment.

You are receiving this mail because:
You are watching the assignee of the bug.