Re: [RFC/RFT] b43: LP-PHY: Implement channel switching for rev2+/B2063 radio

[Larry Finger <Larry.Finger@lwfinger.net>]

Michael Buesch wrote:
> On Thursday 13 August 2009 20:00:47 Gábor Stefanik wrote:
>> (List re-CC-ed.)
>>
>> 2009/8/13 Michael Buesch <mb@bu3sch.de>:
>>> On Thursday 13 August 2009 19:50:16 Gábor Stefanik wrote:
>>>> 2009/8/13 Michael Buesch <mb@bu3sch.de>:
>>>>> On Thursday 13 August 2009 17:02:21 Gábor Stefanik wrote:
>>>>>> Rev.2+/B2063 will now hopefully show some signs of life, though
>>>>>> it won't work at full performance, as calibration is still missing.
>>>>>>
>>>>>> Signed-off-by: Gábor Stefanik <netrolller.3d@gmail.com>
>>>>>>  static void lpphy_baseband_rev0_1_init(struct b43_wldev *dev)
>>>>>> @@ -1369,7 +1370,7 @@ static int b43_lpphy_op_init(struct b43_wldev *dev)
>>>>>>       lpphy_baseband_init(dev);
>>>>>>       lpphy_radio_init(dev);
>>>>>>       lpphy_calibrate_rc(dev);
>>>>>> -     //TODO set channel
>>>>>> +     b43_switch_channel(dev, dev->wl->hw->conf.channel->hw_value);
>>>>> Does dev->wl->hw->conf.channel->hw_value already have a sane value here?
>>>> I think it should - hw->conf.channel was handed to us by mac80211.
>>>> http://bcm-v4.sipsolutions.net/802.11/PHY/LP/Init says "Set channel
>>>> with current chanspec as argument", and hw->conf.channel is our
>>>> equivalent of chanspec.
>>> hw->conf.channel was not handed to us at all. You simply pick it
>>> from mac80211's conf structure and _assume_ that it is initialized.
>>> In the other PHY implementations we use phyop_default_channel (or mandatory
>>> hardcoded channel values) in the PHY init only. I think you should do the
>>> same here. Mac80211 will make sure to select the correct channel later.
>> OK, I will use the default channel (though that's not exactly what the
>> spec says - Larry, is using the default channel correct?)
> 
> What on earth could be incorrect about it? It's an as arbitrary value as
> dev->wl->hw->conf.channel->hw_value would be.

I have not found anything that says what should be used. If the
channel is valid for the selected band, it should be as good as any.
> 
>>>>> Also please call b43_lpphy_op_switch_channel() instead of b43_switch_channel().
>>>> No, that would be wrong; the generic parts of b43_switch_channel also
>>>> need to be executed. See
>>> I don't see why. The generic parts are nothing that should be done at the PHY init.
>> Well, the spec says that the generic parts need to be run, so I
>> implemented it as such.
> 
> We do understand what the generic stuff does and it is not required at this point.
> 
>> Larry, is this part of the spec correct?
> 
> I'm pretty sure it is. But that does not mean we have to implement the same
> bullshit broadcom does ;)

Amen. I have to put it on the web site just in case it is important,
but it must be done that way _ONLY_ if found to be needed for
operation or performance.

Larry