[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

Michał Pecio (michal.pecio@gmail.com) changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |michal.pecio@gmail.com

--- Comment #5 from Michał Pecio (michal.pecio@gmail.com) ---
It's intriguing that failure rate varies with kernel version, but bisecting an
intermittent problem is not going to be easy and there is risk that the result
will be boring.

Per dmesg, the crash happens a few seconds, not 22 minutes after suspend. Looks
like you have a fast and reliable repro on 6.18.

There is a warning about resume taking long time, not sure if a result of the
xhci problem or possibly both issues have a common cause elsewhere?

Is this unique to s2idle or can it also happen after suspend to RAM or
hibernation?

Do any particular devices need to be connected for this to happen? Stop
endpoint timeout implies that there exists some endpoint to be stopped in the
first place, unless the driver became completely confused.

Rebinding the driver almost always helps because it resets the HW and all
software state, so this doesn't say much about the cause. It may still be buggy
HW.

The RESET_ON_RESUME quirk may possibly work as bandaid, try
xhci_hcd.quirks=0x80 boot parameter or pass it to xhci_hcd module if using one.
All USB devices will be dropped and re-enumerated on resume.

Can you repro with dynamic debug enabled and collect debugfs data before
unbinding the driver?
mount debugfs -t debugfs /sys/kernel/debug # if not mounted already
echo "module xhci_hcd +p" >/proc/dynamic_debug/control
echo freeze >/sys/power/state or whatever you do to trigger it
tar czf debug.tgz /sys/kernel/debug/usb/xhci/0000:c1:00.4 # PCI address of the
dead chip

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

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