Re: Ideas for reducing memory copying and zeroing times

lilo (TaRDiS@mail.utexas.edu)
Wed, 24 Apr 1996 10:34:20 -0500 (CDT)


On Tue, 23 Apr 1996, David S. Miller wrote:

> From: Alan Cox <alan@cymru.net>
> Date: Tue, 23 Apr 1996 10:16:07 +0100 (BST)
>
> > Peanuts, it takes 3mb of page tables to map all the physical pages of
> > ram into kernel space on my SMP Sparc box.
>
> Which is too much unpageable data. Page tables are meant to be paged.
>
> Yes, but this machine has 178mb of ram which the kernel has to get to
> at all times. I can use multi-page page table entries though and with
> some of the changes in the 1.3.8* era I should be able to pull it off
> and not distrurb the mm generic code.

Perhaps it's not so much of an issue in the case of the Sparc. My
impression is that page tables are forming a larger portion of available RAM
on the x86 architecture? If so, having them locked in memory is a bit of a
problem, and perhaps something should be done about providing careful
pageability....

lilo