[RFC] tabulating page->virtual on highmem

[RFC] tabulating page->virtual on highmem

[William Lee Irwin III]

The size of the kmap pool appears to dictate the number of distinct
values of page->virtual. Maintaining an index into the pool would
seem to provide superior space behavior, as the index need not be
of full machine word precision. Furthermore, no auxiliary lookup
would appear to be required as the kmap pool is virtually contiguous
and so the virtual address could be calculated from base virtual
address of the kmap pool and the index into the pool.

For architectures using page->virtual for page_address() calculation
this technique does not apply, and so page->virtual would then need
to be maintained as is, or at least retain enough precision for a full
page frame number.

I don't have my heart set on this but I thought I'd at least throw the
idea out where its desirability (and potential implementations) could
be discussed.


Cheers,
Bill
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/

Re: [RFC] tabulating page->virtual on highmem

[William Lee Irwin III]

On Wed, May 08, 2002 at 03:15:06PM -0700, William Lee Irwin III wrote:
> The size of the kmap pool appears to dictate the number of distinct
> values of page->virtual. Maintaining an index into the pool would
> seem to provide superior space behavior, as the index need not be
> of full machine word precision. Furthermore, no auxiliary lookup
> would appear to be required as the kmap pool is virtually contiguous
> and so the virtual address could be calculated from base virtual
> address of the kmap pool and the index into the pool.
> For architectures using page->virtual for page_address() calculation
> this technique does not apply, and so page->virtual would then need
> to be maintained as is, or at least retain enough precision for a full
> page frame number.
> I don't have my heart set on this but I thought I'd at least throw the
> idea out where its desirability (and potential implementations) could
> be discussed.

Since no one's screamed too loudly I'll push out an implementation of
this sometime in the next few days.


Cheers,
Bill
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/