batman-adv 0.2-alpha: possible[tm] regression in packet aggregation

[simon]

Assume we have 2 "threads":

Thread1:

add_packet_list()

-> set_outstanding_packets_timer()

calls spin_trylock(&packets_timer_lock), acquires the lock, and calls

-> cancel_delayed_work_sync(&send_outstanding_packets_wq);

Thread 2:

Workqueue threads, calls

-> set_oustanding_packets()

Assumption:

Thread 2 jumps between the spin_trylock() and cancel_delayed_work_sync(). Then Thread 1 holds, Thread 2 tries to acquire it and waits. Thread 1 then waits for Thread 2 to complete within cancel_delayed_work_sync(). This is a deadlock/lifelock.

Its not very likely that it happens, but it might happen, so i'm opening this ticket. ;)

[marek]

has been fixed in r1247