Many, many things. Most notably:

KERNEL MODULES
==============

* linux/Documentation/devices.txt lists i2c0, i2c1 etc. instead of i2c-0, i2c-1
* LM78 detection: Tom Webster has proven that the reset bit in the 
  id register can be one (strange!)
* Check for kernel 2.0.8 warnings
* Increase use count of adapter driver if directory of chip is entered
* maxilife.c: Introduce new insmod variables
* maxilife.c: Round REG_TO_VID correctly
* icspll.c: Rewrite. The current implementation simply can't work at all.
  It needs i2c-level access (too bad for SMBus-only adapters).
* w83781d: Some features are still unsupported for the W8378[23]. For
  example, the extended beeps (bank 4 register 0x53).
  Also, alarms seem to give strange results
  sometimes, and there are some other minor problems, as indicated by
  Jonathan Bradshaw <jonathan@NrgUp.Com>
* Create wait-queues for bus locking instead of pure semaphores in the i2c
  base modules, to make it more interactive (right now one process can 
  monopolize the entire adapter)
* Find some solution for block reads in smbus-on-i2c emulation code
  We need a hack in the i2c level code for this :-(
* At least the bit-lp and bit-velle modules do no detection on loading;
  ask Simon whether this is possible to add.
* Support 10-bit addresses. At this moment, they are supported nowhere, except
  in Simon Vogl's i2c modules.
* Change the i2c modules to keep the namespace clean (EXPORT_SYMBOL does not
  work for us; many static declarations need to be added; better ask Simon
  Vogl first)
* Make it SMP-safe: there are no spinlocks yet. They are needed at many
  places, probably (everywhere where global vars are accessed). This must be
  done for the i2c modules, too.
* Better general locking, or at least a once-over to check no race-conditions
  remain. This is part of the SMP-safeness, and can better be done at once.
* Register which i2c addresses are used (a la ISA) for each adapter.
* Test SIS5595(/91/98) sensor support. Write generic i2c support. 
  Thomas Dickel <Thomas.Dickel@mni.fh-giessen.de> owns one SIS5595.
* adm9240: check whether the current voltage computations are correct.
  Probably not, as they are different from the datasheet specifications.
  Also check for the supported dallas chip.
* lm80: Check how OS alarms work. At page 17 of the sheet, it tells
  something completely different from the description at page 25.
* gl518sm: Assume that new values are close to old values, so start with
  trying to use a small range near the old values (already partially done)

LIBRARY
=======

* reload does not work (Bison/Flex problem?).
* Some library routines are pretty inefficient right now.
* Library should be split in more separate files, for better linking
  behaviour.

PROGRAMS
========

* Sensors program needs to print a + before lm75 temperatures
* Sensors program needs to be extended for many new chips

OTHERS
======

* Examine watchdog-4.3 or later of meskes@debian.org (Michael Meskes)
  tsx-11.mit.edu /pub/linux/sources/sbin or 
  sunsite.unc.edu /pub/linux/system/daemons/watchdog