Hi,
On 2017/01/21 6:47, Paul Goyette wrote:
> (Replying _without_ looking at the code! I can look more deeply if you
> could re-send a pointer to the code...)
In the example of gre(4) interface, if CPU#0 do "modunload if_gre" and
CPU#1 do "ifconfig gre0 create", they seems to have below race
(Replying _without_ looking at the code! I can look more deeply if you
could re-send a pointer to the code...)
Generally, the module's unload code should detach the devsw entry for
the driver first, to prevent any new accesses.
Then, the config_fini_component() call will fail if existing