It has come to my attention that the current 2.4 A20-enabling code still 
has two problems: it doesn't work on machines for which the "fast A20 
gate" bit has been hijacked for other uses (e.g. Olivetti); apparently 
it doesn't work on some more modern "legacy free" systems either 
(apparently the "Compaq 5Bw160" might be in this category - currently 
unconfirmed.)  I have ported the current A20 code from SYSLINUX (it has 
been stable there since version 1.52) and I would really appreciate it 
(PLEASE PLEASE PRETTY PLEASE) if people would try it out and see how it 
works before sending it to Linus and Alan; I'd like to know any 
combination of "works/doesn't work with the patch" and "works/doesn't 
work without the patch..."

This version first tests for the A20 gate already being enabled (e.g. 
"legacy free" systems), and secondly tries to use the BIOS function (int 
15h, AX=2401h) if it exists.  If not it tries the KBC and finally port 92h.

The patch is against 2.4.8-pre8; I have tried to test it as well as I 
can, which is somewhat hard since I don't actually have any machines 
which exhibit these pathologies; they do, however, support A20 unmasking 
via the BIOS, so that part can be considered tested; and, of course, the 
SYSLINUX code that was the porting basis has been in production use 
since version 1.52, which was released over six months ago.

	-hpa