Re: [PATCH v3] usb: typec: ucsi: do not register UCSI on Lenovo Legion Pro 7 16IAX10H

From: Huang Wei

Date: Thu Nov 19 2026 - 21:41:28 EST


On Thu, Nov 19, 2026 at 02:24:25PM -0700, Alex wrote:
> I need to correct my earlier report and withdraw my Tested-by on this
> patch. Please don't apply v3 as it stands.

Understood, and thank you for the honesty - this is exactly the right
way to handle it. I'll withdraw the v3 as well (Greg: please drop it,
the premise no longer holds).

> - The drops happen with and without ucsi_acpi bound, so the
> driver is not the cause. Skipping UCSI on this machine wouldn't
> fix the problem, and it would hide the typec/PD information.

Agreed. And the Windows comparison is the decisive data point for me:
the stock Microsoft UCSI client (UcmUsciAcpiClient) drives the same
PPM on the same firmware and charges reliably, so UCSI traffic per se
is not what disturbs the EC - which removes the entire premise of
this patch, not just the tested outcome.

What remains is an EC/ACPI-level problem where something Linux does
(or doesn't do) differs from Windows. The burst of
acpi_os_execute_deferred hogged CPU warnings lining up with the drop,
and the EC itself losing sight of the adapter (ADPT clearing in
0xA3), both point below the UCSI stack. The Q7CN44WW/Q7CN78WW BIOS
span and the "power cycle always recovers" pattern suggest an EC
state machine issue that Linux's event handling exposes.

> I'll keep working on it and will follow up on bugzilla #221142 with
> data rather than conclusions.

Sounds good - keeping the bug open and data-driven is the right call.
One suggestion if you capture more: an EC transaction trace during a
drop would show whether the storm comes from a GPE handler, a polling
query, or a buffered event burst - that would narrow the Linux-side
suspect list (acpi_ec polling vs GPE vs the lenovo_wmi/ideapad
helpers you listed) quite a bit.

Thanks again for the correction and the detailed re-test - better a
withdrawn patch than a merged wrong one.

Huang Wei