On Thu, 2004-09-16 at 02:33 +0100, Jonathan Andrews wrote: > Wireless discovery in general isn't going to be easy across > distributions because for reasons purely political (as far as I can > tell???) the promisc mode was left on the cutting room floor !!! WHY > !!!! > > I would have thought that tune, and listen would be the first two things > any wireless driver would be able to do. Having looked a _lot_ at kernel wireless drivers in the past 2 months while developing NetworkManager, I'm not quite sure what the problem here is... Major wireless cards and drivers: Orinoco (Intersil, Wavelan, Airport, etc) Cisco Aironet Atmel Prism54 Intel 2100/2200 Almost all wireless drivers (with the notable exception of the Orinoco drivers, but that's upstream and currently filtering down in the kernel drivers) support wireless scanning (ie, returning a list of access points the card can see) to some degree in the latest kernels, from 2.6.6 and up. That's really what you need for good wireless support of the "it just works" type. What _isn't_ good is the inconsistency of the drivers in supporting the same types of options. For example: (1) signal strength/noise is not normalized between drivers, and the Atmel driver doesn't support signal strength data of scanned access points. Cisco drivers report totally different strength metrics for scanned access points and for the current access point too. (2) WEP/Authentication: Cisco drivers, for example, require a WEP key to be used when using Open System Authentication, which is just plain wrong. You don't need to use encryption when using Open auth. (3) Link detection. Only a few drivers (Orinoco, Intel) use the netif_carrier_on/off kernel interfaces for link detection. However, they use them mostly _wrong_. The general idea of a link-on status should be when the card has authenticated with the base station and traffic can flow (whether or not you're using the correct WEP key doesn't seem to be detectable with Open System Authentication, so that's another problem). What needs to be done? Normalize signal strength, and start using the correct idea of carrier on/off, fix freezes (Cisco drivers seem unstable on more recent kernels), and get Orinoco cards to support scanning (being worked on, albeit not fast). Dan
