Message-ID: <43e72e890811241335j41515d9an701d76de011efdc8@mail.gmail.com> (sfid-20081124_223606_622504_3DEADC87)
Date: Mon, 24 Nov 2008 13:35:58 -0800
From: "Luis R. Rodriguez" <mcgrof@gmail.com>
To: "Richard Farina" <sidhayn@gmail.com>
Subject: Re: [Fwd: please don't regress ath5k.h]
Cc: "Jouni Malinen" <j@w1.fi>,
	"Johannes Berg" <johannes@sipsolutions.net>,
	"Michael Renzmann" <mrenzmann@madwifi-project.org>,
	wireless <linux-wireless@vger.kernel.org>
In-Reply-To: <492AF4EC.4060705@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
References: <492A14A7.4040808@gmail.com>
	 <1205.94.79.146.217.1227505499.squirrel@webmail.madwifi.org>
	 <1227507381.3599.55.camel@johannes.berg> <492AC870.3030103@gmail.com>
	 <20081124182943.GA9790@jm.kir.nu> <492AF4EC.4060705@gmail.com>
Sender: linux-wireless-owner@vger.kernel.org

On Mon, Nov 24, 2008 at 10:39 AM, Richard Farina <sidhayn@gmail.com> wrote:
> Jouni Malinen wrote:
>>
>> On Mon, Nov 24, 2008 at 10:29:52AM -0500, Richard Farina wrote:
>>
>>
>>>
>>> I actually don't have a problem with removing chan_debug, I was merely
>>>  requesting that the size hack it enables not be removed.
>>>
>>> More specifically in base.h I believe the code I specifically require is:
>>>
>>> #if CHAN_DEBUG
>>> #define ATH_CHAN_MAX    (26+26+26+200+200)
>>> #else
>>> #define ATH_CHAN_MAX    (14+14+14+252+20)
>>> #endif
>>>
>>>
>>> When removing chan_debug just please leave the higher max.
>>>
>>
>> I would actually prefer to reduce this number _way_ down to something
>> reasonable in the upstream kernel. ath5k should not really register more
>> than channels 1-14 in 2.4 GHz and 5 GHz channels that are really used
>> (i.e., only every fourth and only at ISM bands). The maximum number
>> would be somewhere closer to 40 than 500..
>>
>> If you have a license to use other frequencies, you can take care of
>> that with your own private patches, but the upstream kernel should not
>> do that. Channels in 2.3 GHz or 2.5 GHz or many of the 5 GHz areas that
>> are currently registered do not really belong here in the upstream
>> kernel no matter what the hardware might be capable of doing.
>>
>>
>
> Your response to my request seems a bit flawed in my eyes.  Drivers are
> supposed to run the hardware, we have crda to handle limiting the device
> back to to what is legal in your area.

CRDA does indeed handle proper regulatory rules for a country however
keep in mind drivers still need to use the data it is calibrated for.
So although the radio can technically use certain frequency ranges you
will find only find calibrated data for the device's targeted EEPROM
settings and for its world roaming capabilities.

>  I'm sure there are places where it
> is legal to use 2192MHz and even if not, look at the driver, it has a max of
> 2732MHz set in the driver (for the .11b/g radio), clearly the person who
> wrote that agrees that drivers should support the hardware.

Please send patches to the wireless regulatory database if such places
do exist and please provide documentation of such findings:

git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-regdb.git

Once we have listed a country with these settings then it makes sense
to add it to the database and to the card's capabilities. Keep in mind
that the EEPROM will always be respected though (patches pending).

> Honestly if you want to argue the legality I purposely posted an RFC for
> that, this change (and this email thread) is to stop a regression that would
> cause the hardware to no longer be fully and properly supported.

This part is certainly important, and is greatly appreciated.

> Yes, I can maintain my own personal patches to all of the drivers in the
> kernel (as a matter of fact I likely do) but this kind of thing belongs in
> the kernel,

What belongs in the kernel is driver components to make your 802.11
work. The arguments used here are similar to that of disagreeing with
easily allowing mac80211 to allow APs to operate without proper power
save implementation.

> we are discussing hardware support, nothing more.

Yes however the channels must be registered too and the EEPROM must be
respected. More work on that area is in the right direction.

> Limiting the
> driver to enforce some random regulatory domain is pointless and against
> kernel policy as reg domain enforcement is supposed to be handled by crda or
> oldreg.

oldreg is just that -- old, and it will be removed. It was there to
help with the migration and to account for the fact distributions may
not have CRDA packaged yet. Drivers should register channels that make
sense to support for 802.11, not every single channel a radio is
capable of tuning into. If not you'd be seeing the channel list
growing for other drivers  as well.

  Luis