Re: pte-highmem-5

[Andrea Arcangeli <andrea@suse.de>]

On Wed, Jan 16, 2002 at 10:04:37AM -0800, Linus Torvalds wrote:
> 
> On Wed, 16 Jan 2002, Andrea Arcangeli wrote:
> >
> > This patch in short will move pagetables into highmem, obviously it
> > breaks all the archs out there.
> 
> Hmm.. Looks ok, although I miss the "obviously". Archs have their own page
> table allocator functions, so by allocating lowmem (and most non-x86 won't
> care) the change _should_ have zero impact on them simply because they
> don't need to unmap. No?

the problem is as usual we need to work with pages or pfn somewhere (see
pte_alloc), we cannot work with virtual addresses or we'll overflow...
the pte_t * virtual address is always the result of the kmap, so the
changes required are the minimal as possible in most places, and I made
a pte_kunmap smart enough to be able to unmap any kind of kmap so the
common code changes are as simple as possible, but still pmd_populate
needs to know the whole thing (struct page * actually) and probably some
other bit of the same kind with the pmd breaks too.

Andrea