net tcp retry triggers assert in kernel/sem.c:145

[zephyrbot]

Reported by Ricardo Salveti de Araujo:

Was able to get an ASSERT FAIL in kernel/sem.c:145 when stressing the TCP stack over BT 6LoWPAN.

Assert: __ASSERT(!_is_in_isr() || timeout == K_NO_WAIT, "");

Added a few additional debug lines, and noticed that this is triggered from TCP abort_connection, when calling net_context_unref.

Abort connection is only called after giving up on the amount of allowed retries, and net_context_unref starts by calling "k_sem_take(&contexts_lock, K_FOREVER);", which ends up triggering the assert.

(Imported from Jira ZEP-2565)

[zephyrbot]

by Mark Linkmeyer:

Andrei Laperie , please set the priority. Thx.

[zephyrbot]

by Jukka Rissanen:

This looks like a cosmetic error, setting priority to low.

[zephyrbot]

by Ricardo Salveti de Araujo:

Hm, how can this be a cosmetic error? This is basically a k_sem_take with K_FOREVER in ISR, which seems to lock up the device.

Will get a better log to confirm.

[zephyrbot]

by Jukka Rissanen:

I meant that this should not cause any issue at runtime. This should be fixed but right now this looks like low priority item.

[zephyrbot]

by Ricardo Salveti de Araujo:

The problem can happen if there is already a lock in contexts_lock when tcp_retry_expired calls net_context_unref, as that will lock the device since it will call k_sem_take with K_FOREVER.

I guess the solution would be to handle the ISR case in net_context_unref and return error if there is already a lock there. Only problem is that we would also need some additional logic to try to unref it again later on, otherwise it will leak the net_context.

[nashif]

@jukka is this still valid?

[jukkar]

This is very old and no additional report has been in long time -> closing.