On Wed, Sep 10, 2014 at 08:52:42PM +0200, Marcin Piotr Pawlowski wrote: > Hi, > > On 09/10/14 20:19, Fabian Raetz wrote: > > On Wed, Sep 10, 2014 at 02:42:43PM +0200, Marcin Piotr Pawlowski wrote: > >> Yes, I think that it could be is possible to double clean the node cache. > >> > >> Updated diff with suggestion from Stefan. > > > > Hi, > > > > this patch works for me too. > > > > But is the problem not specific to the > > iwn(4) driver or at least only specific to drivers which override the > > default state machine and not calling into the original state machine when > > IEEE80211_S_SCAN is entered? > > > > ieee80211_begin_scan() calls ieee80211_free_allnodes() which prevents > > the problem for drivers which use the default scanning infrastructure. > > > > So would it make sense to move the call to ieee80211_clean_cached() into > > iwn(4) like in the modified patch below?
Marcin pointed me to some other driver that could benefit from this and that iwn(4) is not that special as i thought, so i am ok with his original patch too (fwiw). Regards, Fabian > > > > I experimented a bit with calling ieee80211_free_allnodes() here as it > > is called in default scanning infrastructure but this does not work for > > me (yet) :) > > The ieee80211_begin_scan is used by ieee80211_newstate() but ifconfig > scan is done by ioctl(SIOCS80211SCAN) which is processes by the > ieee80211_ioctl() that then invokes ieee80211_new_state(). So both > paths are independent and this issue is not specific to the iwn driver. > > I hope that I have not overlooked something. > > Best regards, > mpp
