Re: [RFC] bootmem for 2.5

[Robert Love <rml@tech9.net>]

On Wed, 2001-11-07 at 19:44, William Lee Irwin III wrote:
> I've managed to reproduce the problem, and I heard from you elsewhere
> that you've verified the fix (although it appeared to reduce the memory
> savings to 4KB).

Fix does indeed work. Tested on:

	P3-733 i815-based, gained 4KB from 384MB
	PPro-200 i440FX-based, gained 4KB from 64MB
	Celeron-500 i440BX-based, gained 8KB from 512MB

No problem on any system -- no difference, in fact, except the gain in
total system memory.  Most importantly, however, the new design is quite
nice. :>

I bet the previous ~100KB gain came from not using APIC.  I was
comparing APIC without new bootmem to new bootmem without APIC.  The
much more realistic and modest 4KB is within range of what I would
expect, and I bet if I compared with and without bootmem on a non-APIC
kernel I would see the same results.

Would you expect problems from laptops or other things with flakey
mappings/reservations?  I can test it on a couple of laptops if you
want...

	Robert Love