>I used to think the way you do but a few month ago I wrote code to
>better detect BIOS hard drive parameters and found that there are modern
>BIOSes that violate the standards for things like return values and
>setting the carry flag on error, etc. And this was on BIOS calls used in
>the original AT. I guess I'm a little gun-shy these days.
It depends on the call; if it's a call that all the Windows platforms make
when they start up, it's a pretty powerful incentive for at least that call
to work.
When I wrote my original better-memory-detect patch, I didn't bother to
check more than about 10 modern machines (including Compaqs) to see if
they worked. I did, on the other hand, check between 20 and 30 older
motherboards; 386sx motherboards are dirt cheap these days, so I simply
stocked up on test subjects, all of which passed the test.
Of course I didn't realize that loadlin had intimate knowledge of the
transfer block, and thus spawned 1000 FAQs between 2.1.23 and 2.1.45 :-(
(2.1.46 has my better better-memory-detect patch, that uses a different
place to put the extended memory detect information [proof that Larry
Augustine and I share at least part of a brain, because I was on the
point of posting the b bmd p when his showed up in my mailbox], so
hopefully THAT FAQ has seen its end.) But as for reliability, I wrote
this patch when I was working for McAfee Associates, and it is being
shipped as part of their antiviral firewall WebShield (hosted these
days on top of a heavily modified 2.0.28 kernel), and it lives on
about 500 machines so far, with only two cases where it doesn't work --
one being a Compaq which persists in saying it only has 16mb, and
_possibly_ another Compaq which it crashes; if anyone has a Compaq
6000, I'd love to have you test the patch out.
____
david parsons \bi/ 088h is not my friend.
\/