linux-bluetooth.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: bugzilla-daemon@kernel.org
To: linux-bluetooth@vger.kernel.org
Subject: [Bug 220344] Intel AX211 bluetooth randomly disconnects and reconnects, started after kernel update
Date: Thu, 22 Oct 2026 23:23:35 +0000	[thread overview]
Message-ID: <bug-220344-62941-ANEJMXA2iI@https.bugzilla.kernel.org/> (raw)
In-Reply-To: <bug-220344-62941@https.bugzilla.kernel.org/>

https://bugzilla.kernel.org/show_bug.cgi?id=220344

Akshay (akshaym@leastaction.net) changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |akshaym@leastaction.net

--- Comment #10 from Akshay (akshaym@leastaction.net) ---
I have been encountering the same issue and it seems to be due to how the Intel
CNVi Bluetooth chips interact with headphones that have both BLE and BR/EDR
capabilities and how the kernel enabled L2CAP ecred and ertm by default. 

Setting "ControllerMode = bredr" in /etc/bluetooth/main.conf and restarting   
bluetooth.service completely and permanently eliminates all dropouts.

Here's a complete description (AI-assisted):

A recurring Bluetooth audio dropout issue affecting Intel CNVi Bluetooth
adapters (specifically the AX211, 8087:0033) paired with dual-mode (BR/EDR +
LE) headsets such as the Sennheiser Momentum 4.                                 

    Below is the detailed bug report along with evidence from                   
    btsnoop packet traces, root-cause analysis, and reproduction steps.         


    1. ENVIRONMENT & HARDWARE                                                   
    -------------------------                                                   
    - Adapter: Intel Wi-Fi 6E AX211 160MHz CNVi (8087:0033, hci0)
    - OS: Fedora Linux 44 (KDE Plasma Desktop Edition) x86_64
    - Kernel: Linux 7.2.11-200.fc44.x86_64                                      
    - Firmware: ibt-1040-0041.sfi (both production and build 82200 tested)      
    - Driver: btusb / btintel (Linux 6.11.x through 7.2.x, Fedora 44)           
    - Audio Stack: PipeWire 1.6.9 + WirePlumber 0.5.18, standard A2DP           
      (reproduced on SBC, AAC, and aptX HD)                                     
    - Headset: Sennheiser MOMENTUM 4 (80:C3:BA:4D:8A:F2)                        
      (Also reproducible on other dual-mode BR/EDR+LE headphones)               


    2. PROBLEM DESCRIPTION                                                      
    ----------------------                                                      
    When streaming Classic A2DP audio from Linux to a dual-mode headset, the    
    stream suffers brief audible audio dropouts / glitches every 5.0 to 5.4     
    seconds.                                                                    

    In extended listening sessions, cumulative starvation events eventually     
    trigger an internal controller watchdog crash:                              
      kernel: Bluetooth: hci0: Hardware error 0x0c                              

    During each glitch, PipeWire logs:                                          
      spa.bluez5.sink.media: Missing completion reports for packet:             
      Bluetooth adapter firmware bug?                                           


    3. EVIDENCE FROM BTSNOOP TRACES                                             
    ------------------------------------------                                  
    In Bluetooth Core Specifications, baseband time is partitioned into         
    discrete 625-microsecond slots.                                             

    Converting the measured stall intervals from btsnoop packet captures        
    directly into baseband slots proves the collision originates inside the     
    controller's internal LE baseband scheduling loop:                          

    Metric               Measured in btsnoop   Baseband Calculation (625 us)    
    -------------------------------------------------------------------------   
    Recurrence Period    5.118s - 5.122s       8,192 slots * 625 us = 5.1200s   
    Single-Window Stall  138.7ms - 146.5ms       224 slots * 625 us = 140.00ms  
    Double-Window Stall  281.2ms - 287.0ms       448 slots * 625 us = 280.00ms  

    Trace Snapshot (Buffer Credit Depletion):                                   
      Frame 758 | t = 5.0456s | ACL_TX     | in_flight = 0                      
      Frame 759 | t = 5.0589s | ACL_TX     | in_flight = 1                      
      Frame 760 | t = 5.0722s | ACL_TX     | in_flight = 2                      
      Frame 761 | t = 5.0856s | ACL_TX     | in_flight = 3 (BUFFER FULL: 0
credits)                                                               
      ---------- INTEL CONTROLLER CEASES ACKing FOR 145.6 ms ----------         
      Frame 762 | t = 5.2312s | EVT_ACK(1) | in_flight = 2 (Controller ACKs)    
      Frame 763 | t = 5.2313s | ACL_TX     | in_flight = 3 (Host sends next
pkt)                                                                  

    During normal streaming, ACL packet intervals are ~12.25 ms. Because the    
    Intel AX211 advertises an ACL buffer depth of only 3 packets (ACL max       
    packets: 3), a radio pause of 140 ms requires at least 11 packets of buffer 
    capacity. The 3-packet buffer fills in ~36 ms, blocking the host stack      
    for the remaining >100 ms and starving the headset DSP buffer.              


    4. ROOT CAUSE ANALYSIS                                                      
    ----------------------                                                      
    1. When BlueZ boots in default "ControllerMode = dual", the kernel issues   
       HCI_OP_WRITE_LE_SUPPORT(1) to the controller.                            
    2. Even with zero active LE connections, entering Dual-Mode activates an    
       internal firmware baseband routine (LE calibration / periodic background 
       scan reservation) scheduled on an exact 2^13 slot cadence
       (8,192 slots = 5.120 seconds).
    3. When this timer triggers, the AX211 baseband arbiter grants the radio
       exclusively to the LE state machine for 224 to 448 slots (140 to 280
ms),
       completely halting BR/EDR transmission and packet acknowledgments.
    4. Older audio architectures (PulseAudio with 250-500 ms ALSA buffers)
       masked this radio blackout in software. Modern low-latency servers
       (PipeWire) operate with tight pacing (10-20 ms), making the baseband
       blackout directly audible.
    5. In pure BR/EDR mode ("ControllerMode = bredr"), BlueZ commands
       MGMT_SET_LE(0) -> HCI_OP_WRITE_LE_SUPPORT(0). The LE state machine
       remains completely dormant, and the 5.120s timer never runs.


    5. WORKAROUND & VERIFICATION
    ----------------------------
    Setting "ControllerMode = bredr" in /etc/bluetooth/main.conf and restarting
    bluetooth.service completely and permanently eliminates all dropouts:
    - Gaps > 25ms: 0
    - Inter-packet interval: 12.25 ms +/- 0.08 ms
    - Audible stutters / dropouts: 0
    - Hardware error 0x0c resets: 0

    No kernel configuration changes (enable_ecred, disable_ertm), no PipeWire
    buffer overrides, no codec changes, and no power management overrides
    were required. Pure BR/EDR mode alone fixes the issue.


    6. PROPOSED PERMANENT SOLUTIONS
    -------------------------------
    1. Intel Firmware (ibt-*.sfi):
       The AX211 baseband scheduler should prioritize active BR/EDR ACL audio
       streams over internal LE background calibration/scanning, or preempt
       the LE routine so that BR/EDR transmission is never paused for more
       than 15-20 ms.

    2. Kernel Driver (drivers/bluetooth/btintel.c):
       If the firmware scheduler cannot be updated, btintel could provide a
       mechanism or quirk to temporarily throttle or mask background LE
       support/scheduling when high-throughput BR/EDR A2DP streams are active
       on single-radio CNVi controllers.


    I can provide btsnoop packet captures if needed.

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

You are receiving this mail because:
You are the assignee for the bug.

      parent reply	other threads:[~2026-10-22 23:23 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-07-17 10:45 [Bug 220344] New: Intel AX211 bluetooth randomly disconnects and reconnects, started after kernel update bugzilla-daemon
2025-07-17 22:45 ` [Bug 220344] " bugzilla-daemon
2025-07-19 12:06 ` bugzilla-daemon
2025-07-19 13:36 ` bugzilla-daemon
2025-07-20  9:02 ` bugzilla-daemon
2025-07-20 11:30 ` bugzilla-daemon
2025-07-24 19:29 ` bugzilla-daemon
2025-08-13  8:16 ` bugzilla-daemon
2026-03-08 14:56 ` bugzilla-daemon
2026-10-01  5:06 ` bugzilla-daemon
2026-10-22 23:23 ` bugzilla-daemon [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=bug-220344-62941-ANEJMXA2iI@https.bugzilla.kernel.org/ \
    --to=bugzilla-daemon@kernel.org \
    --cc=linux-bluetooth@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).