Re: [REGRESSION] v7.2: btusb SCO (HFP mic) broken on AMD Raven xHCI + Intel 9260, works on v7.1

Re: [REGRESSION] v7.2: btusb SCO (HFP mic) broken on AMD Raven xHCI + Intel 9260, works on v7.1

[Michal Pecio]

On Wed, 21 Oct 2026 02:17:04 -0300, Nahuel Clotet wrote:
> Hi,
> 
> Bluetooth HFP microphone audio stopped working in v7.2. v7.1 is fine.
>
> Hardware:
> 
>    - Lenovo ThinkPad T495, AMD Ryzen 5 3500U
>    - xHCI: AMD Raven USB 3.1 [1022:15e1] (06:00.4), BT on usb4 port 1
>      (full speed)
>    - BT: Intel Wireless-AC 9260 [8087:0025], fw intel/ibt-18-16-1.sfi
>      (201-12.24), linux-firmware 20260916
>    - Headset: Anker soundcore P30i, HFP with mSBC (also tested CVSD)
>    - Userspace: BlueZ 5.87, PipeWire 1.6.9, WirePlumber 0.5.18
>      (identical across all kernels)
> 
> Test: switch the card to headset-head-unit and record 6 s from the
> bluez_input source (pw-record, 16 kHz mono), then check the samples
> and the kernel log.
> 
> 
> *Results*:
> 
>    - 6.18.55 (CachyOS LTS): OK, real audio (rms ~1500), no SCO errors
>    - 7.1.11-arch1: OK, real audio (rms ~200), no SCO errors
>    - 7.2.11-arch1 (vanilla): BROKEN, all-zero samples, nothing in the
>    kernel log
>    - 7.2.9 (CachyOS): BROKEN, dmesg floods with "Bluetooth: hci0: corrupted
>    SCO packet" and "SCO packet for unknown connection handle 257", PipeWire
>    reports "Failure in Bluetooth audio transport"

Hi,

For the record, did you try downgrading to 6.x or 7.1 *after* finding
that 7.2 doesn't work?

Not sure where the difference between 7.2.9 and 7.2.11 comes from,
there are no xhci-hcd changes and only one btusb quirk. But there
are changes to net/bluetooth, so maybe there.

> btusb's isoc path (btusb_recv_isoc, btusb_isoc_complete,
> btusb_submit_isoc_urb, __fill_isoc_descriptor, __set_isoc_interface) is
> unchanged between v7.1 and v7.2, and btusb submits SCO URBs with
> URB_ISO_ASAP. The isoc-related xHCI changes in v7.1..v7.2 are:
> 
> d20405dcdf usb: xhci: Remove skip_isoc_td() 298ff13271 usb: xhci: Remove
> isochronous URB_SHORT_NOT_OK handling e765ab012f usb: xhci: Improve Soft
> Retries after short transfers
> 
> I have not bisected yet, so these are suspects, not a confirmed culprit.
> I'm happy to test patches or reverts on this machine.

You could start with testing this commit, which includes all those you
listed above and other USB changes that went into 7.2-rc1. If it fails,
bisection will be trivial. If it works, the bug may not be in USB.

1c2b66a7d725 ("usb: ucsi: huawei_gaokun: support mode switching")

Regards,
Michal