[REGRESSION] 7.2: restore from hibernation hangs while btusb boots Intel AX210 (8087:0032); 7.1 and 6.18 fine

Hi,

My first message here, hope this meets expectations—I had a bit of
help from you know who...

Since 7.2 my laptop hangs hard while resuming from hibernation, and
the hang goes away if btusb is not bound to the Bluetooth controller.

#regzbot introduced: v7.1..v7.2
#regzbot title: restore from hibernation hangs while btusb boots Intel
AX210 (8087:0032)

Hardware
  Dell XPS 13 9360 (BIOS 2.21.0, 2022-06-02), Kaby Lake, xHCI
8086:9d2f (Sunrise Point-LP).
  Wi-Fi/BT card is an Intel AX210
  Bluetooth controller is USB 8087:0032 on bus 1 port 3, driven by
btusb + btintel, firmware intel/ibt-0041-0041.sfi ("Firmware Version:
202-5.26", linux-firmware 20260810).

System
  systemd 261.2
  No BlueZ is installed: the controller is only ever set up by the
kernel's own HCI_AUTO_OFF path.
  Root on LUKS + btrfs, swapfile with resume_offset,
HibernateMode=platform, mkinitcpio busybox initramfs with the resume
hook (systemd-hibernate-resume is present in the initramfs).

Kernels (Arch Linux packages, unpatched apart from Arch's config)
  bad:  7.2.2.arch1-1, 7.2.4.arch1-2
  good: 7.1.4 to 7.1.10 (weeks of daily use), linux-lts 6.18.49 (12/12
cycles under the reproduction below)

Reproduction
  systemctl suspend-then-hibernate with HibernateDelaySec=2min, lid
open. After the timer wakes the machine from S3 it hibernates; power
on, unlock LUKS, the boot kernel loads the image and hands over. On
7.2.x the restored kernel then stops within a couple of seconds,
within the first three hibernate/restore cycles of a cold boot (it
happened on the first cycle twice, the second four times, the third
once, out of seven tests). Screen is black, backlight off, no
response. Nothing is written to the journal. With efi_pstore +
hung_task_panic=1 + softlockup_panic=1 + hardlockup_panic=1 armed and
a three-minute wait, no detector fires and nothing ends up in pstore.

  With no_console_suspend, a good cycle shows these lines on the console:

    [ 2043.009649] atkbd serio0: Failed to deactivate keyboard on isa0060/serio0
    [ 2044.588080] Bluetooth: hci0: No support for _PRR ACPI method
    [ 2050.400042] Bluetooth: hci0: HCI reset during shutdown failed

  and eventually the desktop appears. On a bad cycle, I see these 2 lines:

    [ 2750.173700] atkbd serio0: Failed to deactivate keyboard on isa0060/serio0
    [ 2750.959900] Bluetooth: hci0: No support for _PRR ACPI method

  and then a solid lockup.

Other info
  - /sys/power/disk = test_resume (write the image, read it back and
restore in-kernel, no power-off), each preceded by an S3 cycle: 8/8
successful on 7.2.4.
  - Fresh boot, `modprobe -r btusb` before the first cycle: 4/4 clean
restore. `modprobe btusb` again in the same boot: hang on the 2nd
cycle.
  - A systemd unit that unbinds the two btusb interfaces before
sleep.target and rebinds them after the thaw: 4/4 clean, Bluetooth
otherwise usable. This is my workaround for now.
  - Thermal-zone state (7.2 moved thermal zone resume earlier) was
tested and does not appear to be involved.
  - At every S3 wake on this machine btintel re-downloads the firmware
(~1.4 s); in suspend-then-hibernate the hibernate starts while that
download is still in progress.
  - Plain `systemctl hibernate` (no S3 first) has not hung in the few
tries I gave it.

Happy to provide full dmesg of a good cycle, try patches, or collect
anything else that helps.

Thanks,

David Lomas