Re: [PATCH v4 1/3] cfg80211: Add nl80211 antenna configuration

["Luis R. Rodriguez" <lrodriguez@atheros.com>]

On Tue, Jul 27, 2010 at 7:06 PM, Bruno Randolf <br1@einfach.org> wrote:
> On Wed July 28 2010 01:47:34 Luis R. Rodriguez wrote:
>> On Tue, Jul 27, 2010 at 9:39 AM, Felix Fietkau <nbd@openwrt.org> wrote:
>> > On 2010-07-27 6:19 PM, Luis R. Rodriguez wrote:
>> >>> + * @NL80211_ATTR_WIPHY_ANTENNA_TX: Bitmap of allowed antennas for
>> >>> transmitting. + *     Each bit represents one antenna, starting with
>> >>> antenna 1 at the first + *     bit. If the bitmap is zero (0), the TX
>> >>> antenna follows RX diversity.
>> >>
>> >> What about for 802.11n? What if you want to disable TX?
>> >
>> > Disabling tx shouldn't be handled by the antenna setting, IMHO.
>> >
>> >>> + *     If multiple antennas are selected all selected antennas have to
>> >>> be used + *     for transmitting (801.11n multiple TX chains).
>> >>
>> >> I rather call this TX / RX chainmask then.
>> >
>> > Well, for legacy hardware, these aren't really chains, as there is only
>> > one rx and one tx path, just with switching onto multiple antennas.
>> >
>> >>> + *     Drivers may reject configurations they cannot support.
>> >>> + *
>> >>> + * @NL80211_ATTR_WIPHY_ANTENNA_RX: Bitmap of allowed antennas for
>> >>> receiving. + *     Each bit represents one antenna, starting with
>> >>> antenna 1 at the first + *     bit. If multiple antennas are selected
>> >>> in the bitmap, 802.11n devices + *     should use multiple RX chains
>> >>> on these antennas, while non-802.11n + *     drivers should use
>> >>> antenna diversity between these antennas.
>> >>
>> >> What about TX beamforming, and STBC?
>> >
>> > Disabling one antenna/chain on a two-chain device would naturally
>> > disable TxBF and STBC as well, since it limits the number of available
>> > chains. The driver should simply act as if the disabled chains didn't
>> > exist.
>>
>> That would work.
>>
>> >> Unless 802.11n is completely dealt with I really prefer this patch to
>> >> only address legacy. Otherwise I see sloppyness and inconsistencies on
>> >> supporting this feature throughout different drivers. I'd like to
>> >> avoid that at all costs on nl80211. What you are trying to address is
>> >> legacy antenna setup, not 802.11n RX/TX chainmask dynamic settings so
>> >> I'd really try to avoid it unless you really want to address all
>> >> aspects of chain configuration for 802.11n and even then what I'm
>> >> leading on to say is I think you'll see if you try to address both it
>> >> just gets messy.
>> >
>> > I think 802.11n is already completely dealt with if you treat this as
>> > the chainmask on 11n devices.
>>
>> Its fine by me if the above cases are also embedded into the
>> documentation, just don't want these questions lingering. I can't
>> think of other 802.11n special cases.
>
> thanks felix :)
>
> luis, could you tell me what exactly you would want to include in the
> documentation?

Sure, but after some though I'm sticking to my preference for an API
for this, legacy and 802.11n chainmask / antenna selection should be
dealt with separately.

Reason is I just reviewed the IEEE 802.11 sections 19.19 for TX
beamforming but also happened to stumble upon section 19.20 Antenna
selection. This Antenna Selection section indicates you can support
more antennas than chains and you can use and that the antenna you use
for your chains can vary over time. You select which antenna to use
based on training/sounding tests on each antenna. Antenna selection as
per 19.19 supports up to 8 antennas and up to 4 RF chains. Section
19.20.2 deals with how a STA can initiate antenna selection training
with another STA. Now granted we don't have code for this and I don't
think this is all done explicitly in hardware so one can argue we can
deal with this when/if we do add support for this but this just
highlights how different "antenna selection" or knobs to tune antennas
is even from a standard perspective than legacy. For STBC and TX
beamforming requirements I would like to see mac80211 actually have
code which disables STBC and later TX beamforming (once we have code
for it) if we enable only one chain, I don't see why drivers should
deal with this.

So if you want to define an API I do believe its best to treat these
separately, for legacy I think your API can be fine tuned to consider
FIXED_A, FIXED_B, or DIVERSITY. If you really want some knobs to even
allow for 11n I would recommend debugfs for now which would let you do
anything you want.

  Luis