On 2026-10-20 11:19:53 +0200, Jeremy Bicha wrote:
> There is a lot of AI-generated text to read and this email seems to
> contradict the other emails.
The issue came from the fact that the AI chatbot did not have all
the context. To clarify:
* I intended to use the microphone of my laptop (as usual),
not the one of the headset.
* I was using webex from the web browser (not Cisco's webex
application).
* The default codec for the headset is MSBC (as shown by
pavucontrol), and with Mozilla's gum_test (microphone test),
pavucontrol shows that the codec is changed to SBC.
AI says that "mSBC is an HFP-only codec", so that WirePlumber
switches to A2DP/SBC when actual media playback starts. Indeed,
this is what I can see in pavucontrol.
AI also says that the log shows that during the Webex meeting,
WirePlumber was switching my headset into HFP, even though I
intended to use the laptop mic. I don't know why WirePlumber
did that, but this was apparently the cause of the failure,
due to some bugs (see below). AI says:
The "SCO packet for unknown connection handle 513" kernel errors
are the signature of a failed HFP SCO/eSCO audio connection —
the controller-level voice link that carries HFP audio never
established properly.
and
So the Oct 19 sequence was: Webex starts → stream with Communication
role → WirePlumber flips the XM3 to HFP → the SCO/mSBC voice link
fails on your adapter (the kernel errors) → nodes error out →
WirePlumber may flip back, Webex retriggers, flip-flop — the
repeated transport failures and the multiple node IDs (.1-79, .1-88,
.0-94) are successive profile switches, not a broken A2DP stream.
Then I asked about the mSBC/SCO failure. AI replied
The mSBC/SCO failure is actually a known, currently-active bug
cluster, and with a WH-1000XM3 + BlueZ 5.87 you're right in the
middle of it. Two distinct upstream defects match your logs:
1. BlueZ 5.87 + Sony headsets: the second profile switch breaks
[...]
2. PipeWire native HFP backend: AT+BCC during eSCO setup kills the transport
[...]
I don't know whether the analysis is correct, but at least there
are known bugs that need to be fixed (to entirely fix my problem,
or at least make things clearer).
Full conversation:
https://chat.mistral.ai/work/0c966ccb-1dd5-49c7-97b1-9a2460eeb235
> Could you summarize whether there is still anything that needs to be
> fixed in bluez?
The first bug above is
https://github.com/bluez/bluez/issues/2575
A2DP: a2dp_resume() "SEP in bad state for resume" → Acquire
NotAuthorized after A2DP↔HFP profile switching, until reconnect
(Sony WH-1000XM5, 5.87)
Note: I have the WH-1000XM3, but there shouldn't be much difference.
This bug is still open, but AI says that it is partially mitigated by
https://github.com/bluez/bluez/commit/b7d71e5067856b0daf2f1f73b3fc95483236610e
(which will be in bluez ≥ 5.88).
About the second bug,
https://github.com/bluez/bluez/issues/2545
is mentioned, but the user "solved" it by removing the mSBC from
the codec list. But it is apparently a workaround for a kernel bug.
On the Pipewire side (second bug):
https://gitlab.freedesktop.org/pipewire/pipewire/-/work_items/5506
> I also get a bit concerned about AI bugs because sometimes, the AI
> gives people suggestions on things to change in their system and it's
> unclear whether there are any AI-suggested changes that could cause
> additional bugs.
I do not want to do any change of my config (except for things
that clearly need to be changed for my own use, as I had to do
last year due to a change of behavior in WirePlumber). I was
more asking for information about known bugs.
> Is this bug reproducible from a new Debian Testing (or Unstable)
> install?
I actually don't know how often it is reproducible. There were
no issues with the same webex meeting two weeks ago, while I
have not changed anything in my config. Yesterday, disconnecting
and reconnecting to the meeting did not solve the issue.
Compared to 2 weeks ago:
* bluez got upgraded from 5.87-4 to 5.87-5
* firefox got upgraded from 157.0-1 to 158.0-1
(no pipewire / wireplumber upgrades).
--
Vincent Lefèvre <[email protected]> - Web: <https://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <https://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / Pascaline project (LIP, ENS-Lyon)
