Re: [rfc] lockless get_user_pages for dio (and more)

[Dave Hansen <haveblue@us.ibm.com>]

> +static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, int write, struct page **pages, int *nr)
> +{
> +       pte_t *ptep;
> +
> +       /* XXX: this won't work for 32-bit (must map pte) */
> +       ptep = (pte_t *)pmd_page_vaddr(pmd) + pte_index(addr);
> +       do {
> +               pte_t pte = *ptep;
> +               unsigned long pfn;
> +               struct page *page;
> +
> +               if ((pte_val(pte) & (_PAGE_PRESENT|_PAGE_USER)) != (_PAGE_PRESENT|_PAGE_USER))
> +                       return 0;
> +
> +               if (write && !pte_write(pte))
> +                       return 0;
> +
> +               /* XXX: really need new bit in pte to denote normal page */
> +               pfn = pte_pfn(pte);
> +               if (unlikely(!pfn_valid(pfn)))
> +                       return 0;

Is that little pfn_valid() nugget to help detect VM_IO and VM_PFNMAP
areas?  Does that work 100% of the time?  Is it for anything else?

If that is all that you want a bit in the pte for, I guess we could get
a bitfield or a simple flag in the mm to say whether there are any
VM_IO/PFNMAP areas around.  If we used the same IPI/RCU rules as
pagetables to manage such a flag, I think it would be sufficient to dump
us into the slow path when we hit those areas.  

-- Dave

--
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/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>