[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

Francis DB (francisdb@gmail.com) changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |francisdb@gmail.com

--- Comment #66 from Francis DB (francisdb@gmail.com) ---
Framework Desktop (AMD Ryzen AI Max 300, BIOS 03.05), rear USB-A xHCI
0000:c2:00.4 (1022:1587). Baseline on this machine: ~32% of s2idle cycles kill
the controller ("rtcwake -m mem -s 300" loop; 8 deaths in 25 cycles, on stock
7.1.3 and on 7.1.3 + the earlier delay patch, every run dead within 1-6
cycles).

With 7.1.3 + the v3 device link patch (attachment 310446) the link engages:

  amdgpu 0000:c2:00.0: USB controller 0000:c2:00.4 D0 power state depends on
0000:c2:00.0

and the same reproducer ran 28 consecutive clean cycles with zero controller
deaths. The v3 patch fixes the controller death on this hardware.

Re the full-system hang in comment 63: I have seen the same thing here, with
two data points that should help classify it. 
(1) It is not specific to rapid cycling: my cycles are 300 s suspends, and the
machine still hard-locked once during an overnight v3 run (no ping, no kernel
output, nothing in pstore, forced power-off). 
(2) It is not from the v3 patch: I hit the identical hard lock once on the
completely stock kernel too.
So it looks like a separate, pre-existing, rarer bug (~2 in ~35 cycles here),
not a regression from the device link.

I am testing the v4 patch (attachment 310447) now and will report back with
results.

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

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