busdma(9) fixes for bfe(4):

busdma(9) fixes for bfe(4):
- Set low address of the top level dma tag to 1Gbytes, due to chipset limit.
- TX/RX rings need to be aligned on 4Kbytes.
- Correct dma map syncing direction.
- In bfe_txeof() and bfe_tx_ring_free(), unload dma map for all TX descs
  instead of only for the TX desc holding the mbuf.
- In bfe_start():
  o  If number of free TX desc is less then 2, set IFF_OACTIVE and leave
  o  Pass reference of mbuf pointer into bfe_encap(), since the passed in
     mbuf may be defragmented, i.e. changed.
- In bfe_encap():
  o  Assert that caller has make certain there is at least 2 free TX descs.
  o  Before setting up TX descs, count number of mbufs in the mbuf chain,
     and do defragmentation if there are not enough free TX descs.
  o  After above changes, number of free TX descs will never be less than 2
     when setting up TX descs; assert that.
  o  mbuf chain is always drained after setting up TX descs, nuke the goofy
     if (m != NULL)

Obtained-from: FreeBSD (w/ modification)
Tested-by: Jörg Anslik <joerg@anslik.de>
Approved-by: dillon@