implicit band switch or not

implicit band switch or not

[Arend van Spriel]

Hi Johannes,

With our mac80211-based driver I was associated with 5GHz AP and ran into  
an assert. Investigating this I found that a software scan was initiated  
and our driver got a probe request from mac80211 to be transmitted.  
However, it turned out that the transmit rate was a 2GHz rate and our  
driver was still configured to 5GHz band. The ieee80211_tx_info also  
contained the band and indeed that was for 2GHz. Given the comment for the  
band attribute I am wondering whether we should a band switch implicitly  
or not:

  * @band: the band to transmit on (use for checking for races)

Any suggestions? Are we have a race condition here?

Gr. AvS
-- 
"The most merciful thing in the world, I think, is the inability of the  
human
mind to correlate all its contents." - "The Call of Cthulhu"

Re: implicit band switch or not

[Ben Greear]

On 02/09/2011 04:59 AM, Arend van Spriel wrote:
> Hi Johannes,
>
> With our mac80211-based driver I was associated with 5GHz AP and ran into an assert. Investigating this I found that a software scan was initiated and our
> driver got a probe request from mac80211 to be transmitted. However, it turned out that the transmit rate was a 2GHz rate and our driver was still configured to
> 5GHz band. The ieee80211_tx_info also contained the band and indeed that was for 2GHz. Given the comment for the band attribute I am wondering whether we should
> a band switch implicitly or not:
>
> * @band: the band to transmit on (use for checking for races)
>
> Any suggestions? Are we have a race condition here?

What kernel revision was this happening in?  I just changed the scanning logic
a bit in wireless-testing...I don't think I would have caused this, but it's possible.

Thanks,
Ben

>
> Gr. AvS


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: implicit band switch or not

[Arend van Spriel]

Hi Ben,

On Wed, 09 Feb 2011 15:22:32 +0100, Ben Greear <greearb@candelatech.com>  
wrote:

>> Any suggestions? Are we have a race condition here?
>
> What kernel revision was this happening in?  I just changed the scanning  
> logic
> a bit in wireless-testing...I don't think I would have caused this, but  
> it's possible.
>

I am running 2.6.38-rc3 from the staging-next repository. Last commit on  
net/mac80211/ is 'eb3e554 mac80211: fix a crash in  
ieee80211_beacon_get_tim on change_interface'.

> Thanks,
> Ben
>
>>
>> Gr. AvS
>
>

Gr. AvS
-- 
"The most merciful thing in the world, I think, is the inability of the  
human
mind to correlate all its contents." - "The Call of Cthulhu"

Re: implicit band switch or not

[Ben Greear]

On 02/09/2011 06:48 AM, Arend van Spriel wrote:
> Hi Ben,
>
> On Wed, 09 Feb 2011 15:22:32 +0100, Ben Greear <greearb@candelatech.com>
> wrote:
>
>>> Any suggestions? Are we have a race condition here?
>>
>> What kernel revision was this happening in? I just changed the
>> scanning logic
>> a bit in wireless-testing...I don't think I would have caused this,
>> but it's possible.
>>
>
> I am running 2.6.38-rc3 from the staging-next repository. Last commit on
> net/mac80211/ is 'eb3e554 mac80211: fix a crash in
> ieee80211_beacon_get_tim on change_interface'.

Looks like my change went in after that, so must be something
else...

Thanks,
Ben

>
>> Thanks,
>> Ben
>>
>>>
>>> Gr. AvS
>>
>>
>
> Gr. AvS


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: implicit band switch or not

[Johannes Berg]

On Wed, 2011-02-09 at 13:59 +0100, Arend van Spriel wrote:
> Hi Johannes,
> 
> With our mac80211-based driver I was associated with 5GHz AP and ran into  
> an assert. Investigating this I found that a software scan was initiated  
> and our driver got a probe request from mac80211 to be transmitted.  
> However, it turned out that the transmit rate was a 2GHz rate and our  
> driver was still configured to 5GHz band. The ieee80211_tx_info also  
> contained the band and indeed that was for 2GHz. Given the comment for the  
> band attribute I am wondering whether we should a band switch implicitly  
> or not:
> 
>   * @band: the band to transmit on (use for checking for races)
> 
> Any suggestions? Are we have a race condition here?

Yes, there are race conditions here -- do you implement flush? If yes,
it might be interesting to hook up the tx function to tracing (in
mac80211) and analyse the race in more detail via tracing.

johannes