Re: [PATCH net v6 2/4] net: wwan: t7xx: do not exit the TX push kthread on resume failure

From: Tim JH Chen

Date: Thu Nov 19 2026 - 21:32:21 EST


Hi Andreas,

Thanks for the report. The trace matches the sequence described in the
commit message: pm_runtime_resume_and_get() returned -EINVAL after the
runtime suspend failure left dev->power.runtime_error set, the thread
returned, and kthread_stop() in t7xx_dpmaif_tx_thread_rel() later ran on
the stale task pointer.

The fix is v7 patch 4/6: https://lore.kernel.org/netdev/20261120015556.131547-5-tim770802@xxxxxxxxx/
Its commit message now describes your case, including the remove that
hung in D state holding pci_rescan_remove_lock, and carries a
Reported-by: and Closes: for your mail. If you would rather not be
credited, tell me and I will drop the tag in the next revision.

> One question on the retry. In my case the failure is permanent:
> runtime_error is latched, so rpm_resume() returns -EINVAL from then on.
> With this patch the crash goes away, but TX stays dead and the thread
> retries every 20 ms for as long as the TX lists are non-empty. Is that
> the intended end state, or should something report the dead data path?

The intent of this patch is only that the thread stays alive until
teardown so kthread_stop() is safe; it does not try to recover the data
path. A latched runtime_error is not cleared by anything in the driver
today, and recovering from a failed runtime suspend is a separate
problem from this fix.

You are right that polling every 20 ms for as long as the lists are
non-empty is not a good steady state. In v7 the thread retries once per
second instead, and the first failure of a streak is logged, so the
dead data path is reported once rather than silently. It still does not
recover the data path from a latched runtime_error.

Thanks,
Tim