From: kroe@sbwarren.cs.sunysb.edu (KiYun Roe) Subject: Re: Rebooting - an observation Date: 9 Mar 1992 02:00:31 GMT
In article <1992Mar8.233948.18770@colorado.edu> drew@cs.colorado.edu (Drew Eckhardt) writes:
>Ok, I should have qualified my statement. On the better chipsets,
>where shadowing is accomplished by remapping memory from the top
>of extended RAM into the C0000 - FFFFF address range, copying
>in the ROM, making it read only, and decreasing top of extended
>RAM acording to CMOS, etc. Lesser chipsets take 384K, run the
>same address lines into it, and select either the shadow RAM or
>real memory.
This would imply that QEMM could not get at the complete 384KB, since
some of it would conflict with ROM and video space. However, I pulled
out Manifest, and it told me that QEMM recovered all of the 384KB
(which jived with my recollection that expanded memory increased by
384KB when I disabled shadowing). Then I pulled out the QEMM manual,
and in one place in the whole manual it mentions that some computers
map the 384KB at the top of the memory space, right against (below) the
16MB boundary (I guess this is for ISA computers). So, maybe there's
hope to recover it for Linux.
On the buffer space: I'd also like to see some sort of flexible
boundary between main memory and buffer memory. Better yet, why not
abolish the boundary entirely and allocate both out of a common pool?
I/O and paging would become virtually the same thing. I've heard of
this sort of thing being done in other operating systems, but I don't
know much about it. I could imagine that it might even result in a
smaller kernel.