On Sun, Apr 27, 2008 at 12:40 AM, Jussi Kivilinna <jussi.kivilinna@xxxxxxxx> wrote: > Quoting "Luis R. Rodriguez" <mcgrof@xxxxxxxxx>: > > > > Ok I added cdc_ether.c and it compiles now if you use > > compat-wireless.git and wireles-testing.git to pull. I tested this > > only on 2.6.24 though, 2.6.22 compile may be broken now. We'll have > > to test that. > > > > OK so regardless I ran into issues with a 13b1:0026 Linksys > > (WUSB54GSC). Either I'm really not understanding this driver well or > > it wasn't tested well. I take it its the first... The first oops I > > run into is with access to the net_device on rndis_wext_bind(). I fixed > > that (patch at the bottom) but then I run into a bug on > > generic_rndis_bind() when accessing the net_device .... > > I just didn't get that as I am sure to be setting it. > > > > I then note that somehow generic_rndis_bind() is being run first > > than rndis_wext_bind()! How is this happening? > > > > > > Ok, I'll try to explain usbnet framework. usbnet framework allows usb > netdevices to share common code. Drivers depend depend on usbnet.c as tree: > > usbnet > | > |\------------cdc_ether-----------rndis_host > | | | > drivers drivers that drivers that > that only need cdc_ether need rndis_host > need usbnet (and usbnet) functions (rndis_wlan) > functions functions > > Drivers depending on usbnet are called minidriver in usbnet framework. Each > minidriver has list of devices supported and points probe function to > usbnet:usbnet_probe(). If minidriver needs quirks over usbnet, this done in > bind function. rndis_wlan, rndis_host and cdc_ether all have own modified > bind function. So when rndis_wlan is loaded, usbnet:usbnet_probe() calls > rndis_wext:rndis_wext_bind() calls rndis_host:generic_rndis_bind() which > finally calls cdc_ether:usbnet_generic_cdc_bind(). Netdevice (and struct > usbnet) is allocated in usbnet:usbnet_probe(). > > So usbnet.c should be included to compat-wireless, but modified usbnet.c > would cause problems to other minidriver that aren't included in > compat-wireless. Maybe this could be worked around with compat.diff and > #ifdef's (?) Hm, the other strategy is just to include updates for the other usbnet drivers, the list isn't so extensive. But then this would be becoming more than just compat-wireless but more like compat-net or something, which I guess isn't such a bad idea. > Because there isn't standard way to detect wireless RNDIS device from > non-wireless, rndis_host and rndis_wext both have same usb_device_id. When > rndis_host is loaded on wireless device, rndis_host:generic_rndis_bind() > fails and lets rndis_wlan probe next. If rndis_wlan is loaded on > non-wireless device, rndis_host:generic_rndis_bind() fails and lets > rndis_host be probed. Detection is done with checking device media type with > RNDIS command. Thank you so much for your explanation. I've added usbnet into compat-wireless as well, this is a compat-wireless package after all so I expect users of it to want to get wireless drivers. So at least they'll have this driver now available. We can workout a solution for the other usbnet drivers later, it may make sense to just include them. I'm writing this e-mail on 2.6.22 using rndis_wlan.ko on a WEP network. I've tested this also on 2.6.24. [ 320.456000] usb 4-3: new high speed USB device using ehci_hcd and address 2 [ 320.588000] usb 4-3: configuration #1 chosen from 1 choice [ 320.680000] usbcore: registered new interface driver cdc_ether [ 320.688000] usbcore: registered new interface driver rndis_host [ 320.984000] wlan0: register 'rndis_wlan' at usb-0000:00:1d.7-3, Wireless RNDIS device, BCM4320b based, 00:1d:7e:9a:b3:17 [ 320.984000] usbcore: registered new interface driver rndis_wlan [ 321.024000] udev: renamed network interface wlan0 to wlan2 [ 321.168000] rndis_wlan 4-3:1.0: rndis media disconnect [ 332.492000] ADDRCONF(NETDEV_CHANGE): wlan2: link becomes ready [ 333.664000] rndis_wlan 4-3:1.0: rndis media connect [ 349.280000] wlan2: no IPv6 routers present rndis_wlan is a go! Luis -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
