Re: [RFC] ath9k: Handle interface changes properly

[Felix Fietkau <nbd@openwrt.org>]

On 2011-01-13 5:35 PM, Rajkumar Manoharan wrote:
> On Thu, Jan 13, 2011 at 07:53:27PM +0530, Felix Fietkau wrote:
>>  On 2011-01-13 6:18 AM, Rajkumar Manoharan wrote:
>>  >  On Thu, Jan 13, 2011 at 01:21:47AM +0530, Felix Fietkau wrote:
>>  >>   On 2011-01-12 10:06 AM, Björn Smedman wrote:
>>  >>   >   On Wed, Jan 12, 2011 at 3:30 PM, Rajkumar Manoharan
>>  >>   >   <rmanoharan@atheros.com>    wrote:
>>  >>   >>    The commit ""ath9k: Add change_interface callback" was failed
>>  >>   >>    to update of hw opmode, ani and interrupt mask. This leads
>>  >>   >>    to break p2p functionality on ath9k. And the existing add and
>>  >>   >>    remove interface functions are not handling hw opmode and
>>  >>   >>    ANI properly.
>>  >>   >>
>>  >>   >>    This patch combines the common code in interface callbacks
>>  >>   >>    and also takes care of multi-vif cases.
>>  >>   >
>>  >>   >   How does your patch handle the race condition between the interface
>>  >>   >   change done in process context and the beacon tasklet triggered by
>>  >>   >   SWBA?
>>  >>   >
>>  >>   >   Also, perhaps more applicable to the commit log than the patch, how
>>  >>   >   can opmode be properly handled in multi-vif cases? I mean let's say I
>>  >>   >   have two AP vifs and then change one into STA, is the opmode then STA?
>>  >>   >   Compare that to the case where I have two STA vifs and change one into
>>  >>   >   AP; so again I have one AP and one STA vif but this time opmode is AP,
>>  >>   >   right? I can see how I can be wrong about these examples but I can't
>>  >>   >   really see how the opmode concept can be properly handled in multi-vif
>>  >>   >   cases.
>>  >>   I think opmode should be handled as follows:
>>  >>   If there is at least one AP interface, opmode should be AP, regardless
>>  >>   of what the other interfaces are set to.
>>  >>   If there is no AP vif, opmode can be set to the primary vif type.
>>  >>
>>  >  Correct. this RFC patch does the same.
>>  Really? I don't see that being handled properly, it still seems to
>>  overwrite ah->opmode based on a single vif type for some types.
>>  Also, there is no reason to have a WDS opmode in ath9k_hw. WDS is
>>  typically used along with AP mode interfaces, and where it is not, the
>>  AP opmode should be used for ath9k_hw anyway.
> Instead of setting opmde as AP for WDS, it is better to handle WDS
> case in ath9k_hw.
Why? Right now I don't even see any NL80211_IFTYPE_WDS handling in 
ath9k_hw, and I can't think of anything that should be handled 
differently in ath9k_hw compared to the AP opmode.

>>  Maybe it would be a good idea to clean this up and first limit the
>>  number of different types that we pass to ath9k_hw (i.e. only AP, ADHOC,
>>  STA). Later we can make a separate enum for that to avoid passing the
>>  type as-is entirely.
> Just to stick with the currently supported interfaces list, WDS also included.
>>  I think the mesh point opmode has no place in ath9k_hw. Right now it is
>>  treated like ad-hoc, but I think that's completely wrong. Mesh should
>>  behave just like AP mode, as no ad-hoc style TSF synchronization should
>>  be done by the hardware, and 802.11s mesh nodes do not compete for
>>  beacon transmission.
> This is a different issue and it has to be addressed in separate patch.
I agree.

- Felix