Auto-connect failing with bluez 5.71-1 [Failed to load link keys for hci0]

Description:

With bluez 5.71-1, my bluetooth headset is not auto-connecting correctly at startup like it usually does. Instead it goes into a connected/disconnected state loop.
I have to manually connect to the headset first (via blueman or bluetoothctl). After that it works as intended, the headset automatically reconnect if I stop and restart it (until I reboot the system, then the issue comes back).

Basically, the first auto-connect after the (re)boot of the system isn't working anymore. For what it's worth, a soft-reboot after I initially connect the headset manually is okay, the issue does not happen.

systemctl status bluetooth reports the following error:

$ systemctl status bluetooth

● bluetooth.service - Bluetooth service
Loaded: loaded (/usr/lib/systemd/system/bluetooth.service; enabled; preset: disabled)
Active: active (running) since Fri 2023-12-15 11:03:18 CET; 25s ago
Docs: man:bluetoothd(8)
Main PID: 592 (bluetoothd)
Status: "Running"
Tasks: 1 (limit: 38404)
Memory: 3.1M (peak: 3.6M)
CPU: 33ms
CGroup: /system.slice/bluetooth.service
└─592 /usr/lib/bluetooth/bluetoothd

déc. 15 11:03:18 Arch-Desktop bluetoothd[592]: Battery Provider Manager created
déc. 15 11:03:18 Arch-Desktop bluetoothd[592]: Failed to load link keys for hci0: Invalid Parameters (0x0d)
déc. 15 11:03:38 Arch-Desktop bluetoothd[592]: Endpoint registered: sender=:1.28 path=/MediaEndpoint/A2DPSink/sbc
déc. 15 11:03:38 Arch-Desktop bluetoothd[592]: Endpoint registered: sender=:1.28 path=/MediaEndpoint/A2DPSource/sbc
déc. 15 11:03:38 Arch-Desktop bluetoothd[592]: Endpoint registered: sender=:1.28 path=/MediaEndpoint/A2DPSink/sbc_xq_453
déc. 15 11:03:38 Arch-Desktop bluetoothd[592]: Endpoint registered: sender=:1.28 path=/MediaEndpoint/A2DPSource/sbc_xq_453
déc. 15 11:03:38 Arch-Desktop bluetoothd[592]: Endpoint registered: sender=:1.28 path=/MediaEndpoint/A2DPSink/sbc_xq_512
déc. 15 11:03:38 Arch-Desktop bluetoothd[592]: Endpoint registered: sender=:1.28 path=/MediaEndpoint/A2DPSource/sbc_xq_512
déc. 15 11:03:38 Arch-Desktop bluetoothd[592]: Endpoint registered: sender=:1.28 path=/MediaEndpoint/A2DPSink/sbc_xq_552
déc. 15 11:03:38 Arch-Desktop bluetoothd[592]: Endpoint registered: sender=:1.28 path=/MediaEndpoint/A2DPSource/sbc_xq_552

Downgrading to 5.70-2 solves the issue.

Additional info:

• package version(s): 5.71-1
• config and/or log files: No custom config. I'm using the linux kernel (v6.6.7 currently) and the Asus bt500 bluetooth adapter.
• link to upstream bug report, if any: https://github.com/bluez/bluez/issues/686 & https://lore.kernel.org/linux-bluetooth/55903fdb-e970-4b89-b620-daa93bad7811@archlinux.org/

Steps to reproduce:

1. (Re)boot the system
2. Start bluetooth headset
3. See it going into a connected/disconnected loop on the system side.

added priority3-normal severity4-low statusunconfirmed labels

assigned to @andyrtr, @morganamilo, @toolybird, @antiz, and @thurstylark

changed the description

Please bring this to the bluetooth ML and their github tracker. Try to bisect it.

Just going to put this here so if ever the OP brings it to the BT ML it may add some things that could help to point to the bug.

The above happens in my journalctl output while my keyboard is trying to connect on 5.71.

Unfortunately I'm unable to bisect for some reason... Each builds I'm trying to make after starting bisecting result in the following error:

[...]
configure: creating ./config.status
config.status: creating Makefile
config.status: creating src/bluetoothd.rst
config.status: creating lib/bluez.pc
config.status: creating mesh/bluetooth-meshd.rst
config.status: creating config.h
config.status: config.h is unchanged
config.status: executing depfiles commands
config.status: executing libtool commands
  GEN      ell/shared
make --no-print-directory all-am
make[1]: *** No rule to make target 'src/shared/bap-debug.c', needed by 'src/shared/libshared_glib_la-bap-debug.lo'.  Stop.
make[1]: *** Waiting for unfinished jobs....
make: *** [Makefile:4687: all] Error 2
==> ERROR: A failure occurred in build().
    Aborting...

Not sure what's wrong... Still investigating.

mentioned in issue #3 (closed)

added scoperegression statuswaiting-input labels and removed statusunconfirmed label

unassigned @morganamilo, @toolybird, and @thurstylark

Same issue here. Downgrading bluez works fine.

changed the description

I was able to bisect after all! Turned out you need to repeat the ./bootstrap step manually in sources before building depending on the changes, as indicated in this comment.

The faulty commit is d5536e0cd431e22be9a1132be6d4af2445590633:

$ git bisect log
git bisect start
# status: waiting for both good and bad commits
# good: [b05c3fbfd764a33dbaefbfb18680fa3296d4a31b] Release 5.70
git bisect good b05c3fbfd764a33dbaefbfb18680fa3296d4a31b
# status: waiting for bad commit, 1 good commit known
# bad: [04ecf635ffaa2f7f8bca89cec3a0fbdbeb016dc9] Release 5.71
git bisect bad 04ecf635ffaa2f7f8bca89cec3a0fbdbeb016dc9
# good: [e1ec086614224cb50918922dd88f71ae3652a8e5] monitor: Make use of util_debug_tlv to decode TLV entries
git bisect good e1ec086614224cb50918922dd88f71ae3652a8e5
# good: [506413dba87230b1c579eb23901bd88e52ef6825] isotest: Add presets from GMAP
git bisect good 506413dba87230b1c579eb23901bd88e52ef6825
# good: [6f957d3b6c438debefb1536a1ebb46214b592c57] test-bap: Add SCC Enable tests
git bisect good 6f957d3b6c438debefb1536a1ebb46214b592c57
# good: [b0656c2cf14da4b43ee7a446bfa9c5f1d3c044b0] Pattern match on service- and manufacturer data
git bisect good b0656c2cf14da4b43ee7a446bfa9c5f1d3c044b0
# good: [87477b7ffefcf3c64c821d9f700664d89aa2c5f7] shared/bap: Update Available context for source
git bisect good 87477b7ffefcf3c64c821d9f700664d89aa2c5f7
# bad: [ba9fda12d26bf4200339da94bc198c6641f98668] device: Notify when the bdaddr_type changes
git bisect bad ba9fda12d26bf4200339da94bc198c6641f98668
# bad: [d5536e0cd431e22be9a1132be6d4af2445590633] adapter: Fix addr_type for smp_irk/ltk_info/link_key
git bisect bad d5536e0cd431e22be9a1132be6d4af2445590633
# first bad commit: [d5536e0cd431e22be9a1132be6d4af2445590633] adapter: Fix addr_type for smp_irk/ltk_info/link_key

I confirm that reverting this commit fixes the issue.

While waiting for a proper upstream patch/fixed release that we'll be able to push to [extra], I patched the sources locally by reverting this faulty commit and built the bluez* packages against it.
The packages are available here for those interested.

I reported the issue to upstream in the GitHub repo and the ML.

changed the description

changed the description

added statuswaiting-upstream label and removed statuswaiting-input label

I tried @antiz packages and it fixed my issue.

changed the description

changed the description

assigned to @heftig

unassigned @heftig

@heftig https://github.com/bluez/bluez/issues/686#issuecomment-1858751323 - any opinion about adding this to mainline and lts kernel (if it applies)?

I compiled a kernel with the above-mentioned commit and rolled back to 5.71-1 and can confirm it fixed the issue at least with my device/s.

@bugbuddy assign @heftig

assigned to @heftig

closed

fixed in 5.71-2 by applying a not yet merged upstream MR to work again with current stable kernels