2009/12/14 Johannes Berg <[email protected]>:
> On Sun, 2009-12-13 at 19:17 +0100, Albert Herranz wrote:
>
>> > Well, something in mac80211 was changed that breaks this.
>> > mac80211 currently seems to assume that the number of queues does not 
>> > change
>> > after ieee80211_register, which was not the case previously. This breaks
>> > the QoS disable, because b43 modifies the nr of queues variable in the
>> > ieee80211_hw structure.
>> >
>> >> This patch adds a config option to allow disabling QoS from the beginning.
>> >> This is the only way to workaround the described problem when building the
>> >> b43 driver in-kernel, as in that case modparam_qos can't be changed.
>> >>
>> >> Signed-off-by: Albert Herranz <[email protected]>
>> >
>> > Why are we currently adding all kind of workaround patches instead of 
>> > fixing the bugs?
>> > This bug's been there for months, so I don't see a reason to push out a 
>> > workaround now.
>> >
>>
>> I'm fine not pushing this workaround upstream. I can carry the workaround 
>> locally.
>>
>> Is someone (apart from you) aware of this bug and looking at mac80211 to fix 
>> it?
>
> I don't even consider it a bug in mac80211. Once you register your
> capabilities, they're fair game. If you need to register them based on
> firmware capabilities, load the firmware before you register, like I did
> with my [RFC] patch to ar9170 (which I'll be sending soon since the
> required patch is going into mainline).

I believe we ieee80211_register_hw in b43 once but it may happen we
reload firmware while having device still registered. Should we really
ieee80211_unregister_hw and ieee80211_register_hw on every time we
reload firmware (as we may load one with other capabilities)?

-- 
Rafał
_______________________________________________
Bcm43xx-dev mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/bcm43xx-dev

Reply via email to