2008/1/23, Peter Zijlstra <[EMAIL PROTECTED]>:
>
> On Wed, 2008-01-23 at 02:21 +0300, Anton Salikhmetov wrote:
>
> > +static void vma_wrprotect_pmd_range(struct vm_area_struct *vma, pmd_t *pmd,
> > + unsigned long start, unsigned long end)
> > +{
> > + while (start < end) {
> > + spinlock_t *ptl;
> > + pte_t *pte = pte_offset_map_lock(vma->vm_mm, pmd, start,
> > &ptl);
> > +
> > + if (pte_dirty(*pte) && pte_write(*pte)) {
> > + pte_t entry = ptep_clear_flush(vma, start, pte);
> > +
> > + entry = pte_wrprotect(entry);
> > + set_pte_at(vma->vm_mm, start, pte, entry);
> > + }
> > +
> > + pte_unmap_unlock(pte, ptl);
> > + start += PAGE_SIZE;
> > + }
> > +}
>
> You've had two examples on how to write this loop, one from git commit
> 204ec841fbea3e5138168edbc3a76d46747cc987, and one from my draft, but
> this one looks like neither and is much less efficient. Take the lock
> only once per pmd, not once per pte please.
>
> > +static void vma_wrprotect_pud_range(struct vm_area_struct *vma, pud_t *pud,
> > + unsigned long start, unsigned long end)
> > +{
> > + pmd_t *pmd = pmd_offset(pud, start);
> > +
> > + while (start < end) {
> > + unsigned long next = pmd_addr_end(start, end);
> > +
> > + if (!pmd_none_or_clear_bad(pmd))
> > + vma_wrprotect_pmd_range(vma, pmd, start, next);
> > +
> > + ++pmd;
> > + start = next;
> > + }
> > +}
> > +
> > +static void vma_wrprotect_pgd_range(struct vm_area_struct *vma, pgd_t *pgd,
> > + unsigned long start, unsigned long end)
> > +{
> > + pud_t *pud = pud_offset(pgd, start);
> > +
> > + while (start < end) {
> > + unsigned long next = pud_addr_end(start, end);
> > +
> > + if (!pud_none_or_clear_bad(pud))
> > + vma_wrprotect_pud_range(vma, pud, start, next);
> > +
> > + ++pud;
> > + start = next;
> > + }
> > +}
> > +
> > +static void vma_wrprotect(struct vm_area_struct *vma)
> > +{
> > + unsigned long addr = vma->vm_start;
> > + pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
> > +
> > + while (addr < vma->vm_end) {
> > + unsigned long next = pgd_addr_end(addr, vma->vm_end);
> > +
> > + if (!pgd_none_or_clear_bad(pgd))
> > + vma_wrprotect_pgd_range(vma, pgd, addr, next);
> > +
> > + ++pgd;
> > + addr = next;
> > + }
> > +}
>
> I think you want to pass start, end here too, you might not need to
> sweep the whole vma.
Thanks for you feedback, Peter!
I will redesign the vma_wrprotect_pmd_range() routine the way it
acquires the spinlock outside of the loop. I will also rewrite the
vma_wrprotect() function to process only the specified range.
>
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
