
- floating point is basically not supported at all right now.  This needs to
  be fixed soon.  The compiler depends on it, and compiling panics the
  kernel currently.

- the compiler does note 8-byte align vlong/uvlongs.  Because of this
   there is some code in l.s that writes these in two-halves rather
   than a single 8-byte write.  Should be fixed when the compiler is
   updated.
- we dont make use of the FP dirty bits.  Doing so could possibly
   simplify FP active/inactive handling.
- when we go MP, are we sharing mmu ctx across processors?  If
   so we will have to be more careful during ctx dealloc!
- different ureg structure for extra fields.  Check userland
   to make sure none of the old regs (tbr, psr) are used.
- still using TSO.  should eventually support a higher performance ordering.
- need to figure out the mii stuff and add support for mii in etherhme

- modified ureg: pstate, ttype
- seperating traps and interrupts would be more efficient --
   especially fast-tracking the linear mapping of RAM above KZERO
   and the remapping of 0x60 interrupts to softints.
- traps are a mess right now.  rethink after we get everything figured out.
- the traplink is no good..  we're setting CWP for syslink and then just setting it back.
- update libmach to handle the boot image type we use. -- is this possible?
   there's no real header here.  investigate using other types of headers.  would
   be a pain if we have to support ELF.
   XXX I really hate the current boot image we're using -- the loader ends up
   loading the whole image, debug symbols and all.  However the more proper
   a.out format doesnt seem to work on the ultrasparc.
- hme ethernet keeps a fixed number of buffers for the hardware and copies
   data into and out of them.  Would be better to not allocate any xmit
   buffers and simply xfer the ones given to us in-place.  Similarly would
   be good to be smarter about the receive buffers.


TODO
- true 64bit
- graphics, keyboard, normal console.
- scsi
- SMP
