Re: [REGRESSION][BISECTED] usb: hub: total system freeze after running adb

[Mathias Nyman <mathias.nyman@linux.intel.com>]

Hi

On 11/17/25 22:18, Forest wrote:
> #regzbot introduced 8f5b7e2bec1c
> 
> Running adb (android debug bridge) commands causes the entire system to
> freeze some time later.
> 
> The freeze usually occurs within two hours of running adb, but not always.
> Stopping adb's persistent background process with `adb kill-server` seems to
> avoid the freeze. (However, this is not always possible, since the system
> sometimes freezes just a few seconds after adb is launched.)
> 
> It occurs even when no android devices are connected to the system.
> 
> I am unsure whether user activity affects how soon the freeze happens.
> 
> Observations when it happens:
> 
> - No messages are written to dmesg or syslog.
> - Virtual console switching no longer works.
> - SysRq key commands no longer work, as far as I can tell.
> - Whatever was on the display remains intact.
> - Any already-playing audio enters a short loop, as though the motherboard's
>    sound device is playing from a ring buffer that no longer gets updated.
> 
> I have reproduced it on mainline (6.18.0-rc5) and various older kernel
> versions as early as 6.12.48, built with Debian config files and
> `make olddefconfig`. Tested using Debian Stable 13.2 (trixie), and its
> adb 1:34.0.5-12 package.
> 
> Git bisect identifies the culprit:
> commit 8f5b7e2bec1c36578fdaa74a6951833541103e27
> 2025-06-11 14:24:41 +0300
> usb: hub: fix detection of high tier USB3 devices behind suspended hubs
> 
> Reverting that commit (along with two later commits that build upon it)
> seems to fix the problem in the mainline and longterm kernels that I tried.

Could you share some info about this setup:

lspci -nnv -d ::0c03
lsusb -t

Enabling usb core and xhci dynamic debug could share some light on this:

mount -t debugfs none /sys/kernel/debug
echo 'module xhci_hcd =p' >/sys/kernel/debug/dynamic_debug/control
echo 'module usbcore =p' >/sys/kernel/debug/dynamic_debug/control
< Reproduce issue >
Send output of dmesg

If the bisected patch is the issue then disabling runtime suspend could help.
For all usb* roothubs:

echo on > /sys/bus/usb/devices/usb*/power/control

Does it affect anything?

Thanks
Mathias