[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 #3 from mattwork (mrh@frame.work) ---
Thanks for looking at this. A few notes on the bisect request:

The regression data in the report already narrows the timeline: the bug exists
on 6.12 LTS (intermittent, ~40%) and is deterministic by 6.18. That's a wide
bisect range — thousands of commits across multiple merge windows — and the
result would land on whichever commit pushed the failure rate from "sometimes"
to "always," not the root cause.

The root cause evidence is already in the report:

- The resume path does not fully reinitialize the xHCI controller. The bind
path does. Unbind/rebind fixes it 100% of the time.

- The controller shares a PCI root complex with the amdgpu (c1:00.0). iGPU
memory allocation correlates with failure rate, suggesting PCI power state
management during s2idle is involved.

- This reproduces on non-Framework hardware (Lenovo ThinkPad T14 Gen 6 AMD)
with identical symptoms and identical workaround, ruling out platform-specific
firmware.

A bisect would confirm which commit worsened an already-broken resume path, but
the xHCI maintainers likely already know what changed in the s2idle resume
sequence for AMD xHCI between 6.12 and 6.18. If a bisect is still needed after
review of the above, I can arrange one, but I'd like to get USB/xHCI maintainer
eyes on the dmesg and the PCI topology first.

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

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