Re: [RFC] Distributed mmap API

[Daniel Phillips <phillips@arcor.de>]

On Wednesday 25 February 2004 17:16, Andrew Morton wrote:
> > but how can we legitimately get !pfn_valid there?
>
> A mapping of some I/O region?

With MAP_PRIVATE, on a distributed filesystem?  OK...

Can we recognize those I/O vmas and handle them with their own separate loop, 
saving a few cycles for the common case?  Or just:

	if (pte_present(pte)) {
		unsigned long pfn = pte_pfn(pte);
		struct page *page;
		if (unlikely(!pfn_valid(pfn))) {
			ptep_get_and_clear(ptep);
			tlb_remove_tlb_entry(tlb, ptep, address+offset);
			continue;
		}
		page = pfn_to_page(pfn);
		if (unlikely(!all) && is_anon(page))
			continue;
		pte = ptep_get_and_clear(ptep); /* get dirty bit atomically */
		tlb_remove_tlb_entry(tlb, ptep, address+offset);
		if (PageReserved(page))
			continue;
		if (pte_dirty(pte))
			set_page_dirty(page);
		if (page->mapping && pte_young(pte) && !PageSwapCache(page))
			mark_page_accessed(page);
		tlb->freed++;
		page_remove_rmap(page, ptep);
		tlb_remove_page(tlb, page);
	} else {

Regards,

Daniel