Re: [PATCH][RFC][0/4] InfiniBand userspace verbs implementation

[Andrew Morton <akpm@osdl.org>]

Roland Dreier <roland@topspin.com> wrote:
>
> OK, I'm by no means an expert on this, but Libor and I looked at
> rmap.c a little more, and there is code:
> 
> 	if ((vma->vm_flags & (VM_LOCKED|VM_RESERVED)) ||
> 			ptep_clear_flush_young(vma, address, pte)) {
> 		ret = SWAP_FAIL;
> 		goto out_unmap;
> 	}
> 
> before the check
> 
> 	if (PageSwapCache(page) &&
> 	    page_count(page) != page_mapcount(page) + 2) {
> 		ret = SWAP_FAIL;
> 		goto out_unmap;
> 	}
> 
> If userspace allocates some memory but doesn't touch it aside from
> passing the address in to the kernel, which does get_user_pages(), the
> PTE will be young in that first test, right?

If get_user_pages() was called with write=1, get_user_pages() will fault in
a real page and yes, I guess it'll be pte_young.

If get_user_pages() was called with write=0, get_user_pages() will fault
in a mapping of the zero page and we'd never get this far.

> Does that mean that
> the userspace mapping will be cleared and userspace will get a
> different physical page if it faults that address back in? 
>

We won't try to unmap a page's ptes until that page has file-or-swapcache
backing.

If the pte is then cleared, a subsequent minor fault will reestablish the
mapping to the same physical page.  A major fault cannot happen because the
page was pinned by get_user_pages().

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/