On Mon, Mar 16, 2009 at 10:23 AM, Quentin Armitage <Quentin@xxxxxxxxxxxxxxx> wrote: > If cfg80211 is compiled with CONFIG_WIRELESS_OLD_REGULATORY defined, if > insmod is loaded with "insmod cfg80211 ieee80211_regdom=EU" then the > execution of insmod mac80211, executing insmod ath5k causes an oops. > This is caused by wiphy_update_regulatory being called when last_request > == NULL, and the call of reg_is_world_roaming (via reg_process_beacons) > causes the oops at last_request->initiator != REGDOM_SET_BY_COUNTRY_IE. > > If reg_is_world_roaming is modified to check for last_request not being > NULL, e.g. > if (last_request && last_request->initiator != REGDOM_SET_BY_COUNTRY_IE > && > then I also get an oops in reg_device_remove where last_request is > referenced and there is only a subsequent check for it being NULL. > > Even if the above two are fixed, it will not associate with the access > point, and `iwlist wlan0 channels` produces a rather confused list. It > appears that, despite the comment, in regulatory_init it is necessary to > call regulatory_hint_core even for the pseudo country EU. Yes, good catch, the entire code in reg.c assumes this was always done. We overlooked the fact when "EU" became the only exemption. > Patch1 below covers the points above, although I am not sure how useful > the BUG_ON (!last_request) in reg_is_world_roaming is. Its better to understand the code than sprinkle random BUG_ONs, and in this particular case you've already found the issue. So lets address and fix that appropriately. > Having modified the above, I then found that the regulatory domain was > not being updated from the Country IE sent by the access point. > It > appears that the code following the comment "so we optimize an early > check ..." in regulatory_hint_11d is a little over optimized, since if > the following if statement is true, every path through that block ends > in "goto out;". Patch 2 below allows the Country IE to be processed > (note it also needs an extra check around a WARN_ON, and so the > preceeding comment to that needs considering). Thanks for reporting this as well, as you noted we didn't check for the last_request type and just made an assumption. The check should go on earlier. Will spin some patches to address this. The right solution really is to abandon OLD_REG, but that won't happen for stable right now so will spin some patches to address this for stable. Thanks, Luis -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
