Survey data inconsistencies between drivers

[Helmut Schaa <helmut.schaa@googlemail.com>]

Hi,

currently there are at least three different survey types implemented in
different mac80211 drivers.

- ath9k always lists the survey data for all channels and always provides
  accumulated busy time data. This has the advantage that a simple scan
  followed   by a suvrey dump will give a nice overview of the different
  channel utilization.
- ath5k only lists the active channel but also accumlates the busy time
  statistics.
- rt2800 only lists the active channel and only returns relative busy time
  values since the last read.

>From a hw perspective all three drivers appear similar, there are some
registers that are cleared on read and contain the busy times since the
last read.

I really like the ath9k approach but implementing it in every driver
sounds like a lot of redundancy.

In order to have consistent survey data between drivers I'd propose the
following approach:

- The driver's get_survey callback only returns relative survey data for the
  currently active channel. That means reading the current busy time statistics
  from the hw and clearing the appropriate counters.
- mac80211 would keep track of accumulating the gathered survey data.
- mac80211 would read the survey data before each channel change and update
  the approriate channel's survey data just as ath9k does right now.

Any arguments against such a unification?

Thanks,
Helmut