On Sun, May 18, 2014 at 07:45:59PM -0400, Matthew Wilcox wrote:
> On Sat, May 17, 2014 at 03:33:05AM +0300, Kirill A. Shutemov wrote:
> > Below is my attempt to play with the problem. I've took one function --
> > page_referenced_one() -- which looks ugly because of different APIs for
> > PTE/PMD and convert it to use vpte_t. vpte_t is union for pte_t, pmd_t
> > and pud_t.
> >
> > Basically, the idea is instead of having different helpers to handle
> > PTE/PMD/PUD, we have one, which take pair of vpte_t + pglevel.
>
> I can't find my original attempt at this now (I am lost in a maze of
> twisted git trees, all subtly different), but I called it a vpe (Virtual
> Page Entry).
>
> Rather than using a pair of vpte_t and pglevel, the vpe_t contained
> enough information to discern what level it was; that's only two bits
> and I think all the architectures have enough space to squeeze in two
> more bits to the PTE (the PMD and PUD obviously have plenty of space).
I'm not sure if it's possible to find a single free bit on all
architectures. Two is near impossible.
And what about 5-level page tables in future? Will we need 3 bits there?
No way.
> > +static inline unsigned long vpte_size(vpte_t vptep, enum ptlevel ptlvl)
> > +{
> > + switch (ptlvl) {
> > + case PTE:
> > + return PAGE_SIZE;
> > +#ifdef PMD_SIZE
> > + case PMD:
> > + return PMD_SIZE;
> > +#endif
> > +#ifdef PUD_SIZE
> > + case PUD:
> > + return PUD_SIZE;
> > +#endif
> > + default:
> > + return 0; /* XXX */
>
> As you say, XXX. This needs to be an error ... perhaps VM_BUG_ON(1)
> in this case?
Yeah. But include <linux/bug.h> is problematic here...
Anyway the only purpose of the code is to start discussion.
--
Kirill A. Shutemov
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>