=== Kernel style cleanup ===

* Move bulk messages to debugfs + relayfs

* Try not to abuse too much sysfs. For now, use it for some parameter
  easy setting.

=== protocol cleanups ===

* Now that I understand how the max RX size works, apply it to limit the
  rx queue packet sizes.

* Fix firmware loading on 3887 for 2.5.11

* Add support for the newer lmac protocol versions. Needed for Dell 1450
  USB support.

* Dabe's device has strange rx filter packets. investigate why and how
  they are used.

* ISLSM_TX_CONTROL_FILTER_MAGIC7 constants should be gotten rid of
  (lookup the firmware).

* Bug in version 2 Devices : two field in rx/tx filter, still unknown,
  have changed in those newer devices. Watch the mgmt response and try to
  tell where i can find the data. Ask Samuel what they mean.

* Understand more of the stats packets -- this one is easy enough.

* Understand more of the tx packet -- hard, but now seems somewhat
  doable.

* Understand the id.magic1 field of the received frames : it is
  influenced by the rx/tx filter packet (changes for identical
  frames). Check how/why with active testing.

* reclaim alloc'd probe/beacon on unload -- see if madwifi tolerates
  that we free the beacon. Not sure...

* Decode the set_filter packet more

=== device access cleanups ===

* Do a proper sharing of isl_romboot & al between usb and pci, then
  between ramboot and romboot.

* Get rid of all dependencies on netdev in islsm.

* drain the txq in uart_cts_cb

* precisely update all statistics and compare to the stats reports that
  we get from the device

* Maybe the hardware should maintain its own state and know when it has
  to refuse the packets ?

* Reorganize website

* Test pci driver on PPC, and get mode debug.

* Only 32-bits pad for usb version 1 device. Not sure the PCI
  appreciates this.

* segfault when islpci_alloc_queue returns enomem in islpci_alloc_memory

* tasklet in PCI code : disable it on unloading. use tasklets for usb rx
  path

* tx path in PCI sleeps

* monitor mode : allocate enough headroom so we don't have to cow the
  buffer

* tx path : ask the netdev layer to allocate enough headroom so we don't
  have to realloc the tx skb -- this actually doesn't work as madwifi is
  in the middle.

* see if bootup input is really needed as exported. Should be integrated
  into only one input function maybe.

* Write a real p54u_init that will be called from the 802.11 state machine.

* Instead of differentiating on version1/version2 in the code, write
  several functions and put them in an array to be initialized
  statically at device probe. especially usefull for usb init function
  (one function per device), transmit function (one function per device
  kind), and frequency change too, for instance.