Things that don't suck:
DS1) Everything compiles. Everything works (at least on my box).
DS2) No changes are necessary to the kernel.org source files in this build setup - which makes it easy to try it out with future versions of the Linux USB code. This is a big plus because it seems like the codebase is rewritten from scratch every few weeks, fixing one or two bugs and introducing a dozen more.
DS3) The USB 2.0 code gives us *fast* transfers. This should be nice for extraction fans.
DS4) The ax8817x driver in this package does not lock up the adapter when your packet payload is an exact multiple of 64 bytes. This is because the 2.4.20 USB code supports USB_ZERO_PACKET, and the 2.4.4 USB code does not.
Things that suck:
S1) Nothing has been tested much. If you find a bug, please submit a patch.
S2) There is a problem that may involve truncated packets in the EHCI driver. I have patched host/usb.c to fix the most obvious symptom (grep for 'WRONG WRONG WRONG WRONG WRONG') but this is definitely not a correct solution. I tried the EHCI driver from 2.4.21-rc3 and it was even worse; that driver sends bad requests to the USB chipset and starts generating nasty protocol errors. This may have something to do with the fact that I was commenting out calls to the pci driver that didn't exist in 2.4.4 (and that I do not understand at all).
S3) It would be nice to write a kernel module that re-enables wireless ioctls so that iwconfig works normally. (grep for CONFIG_NET_RADIO in the kernel tree and take a look at the code that gets lopped out when it is undefined.)
S4) These drivers generate a bunch of unaligned accesses (especially the pegasus driver). This happens because gcc does stupid things like accessing an unaligned 32-bit word in a packed structure with a single "lw" opcode. If you know of a way to override this behavior ON MIPS (no, not on i960) let me know. I was hoping that gcc-3.2 would fix the problem but it did not.
S5) We need a secure way to encrypt traffic to the TiVo for use on wireless networks, since tivoftpd, tnlited, and tivoweb are all wide open. Anyone want to build FREES/WAN for 2.4.4-TiVo? CIPE looks like another possibility, although the use of (albeit encrypted) CRC32s as a MAC leads me to question their design process.