Re: [patch 12/13] GRU Driver V3 - export is_uv_system(), zap_page_range() & follow_page()

[Nick Piggin]

On Friday 11 July 2008 03:20, Jack Steiner wrote:
> On Fri, Jul 11, 2008 at 02:52:00AM +1000, Nick Piggin wrote:

> > lockless gup checks for struct page by checking a bit in the pte.
> > This should be enough to guarantee  it is cacheable memory (unless
> > another driver has done something tricky like set the the page's
> > cache attributes to UC or WC -- I don't know if there is a way to
> > completely avoid all corner cases).
>
> The GRU itself has no need to reference the page struct.
> However, it WILL reference valid ptes that represent pages imported from
> other SSIs via xpmem. These will have cacheable ptes but no page structs.

Oh, I'm sorry Jack, I misread the patch and thought you still had
the page_to_phys thing in there... OK, then it probably isn't
broken. And in which case you would have to add a little code to
gup.c...


> Maybe checking the pte attributes is the best way to do the check.
>
> If we take this approach, what is a good API for the gup.c walker?
> Return the pte attributes?
>
> 	int get_user_pte(struct mm_struct *mm, unsigned long address,
> 	        int write, unsigned long *paddr, int *pageshift, pgprot_t *prot)
>
> The GRU would enforce the check for cacheable access.

Yeah that wouldn't be a bad API, although being a lockless, may-fail,
not available on all archs kind of thing, I would prefer a different
name, maybe get_user_pte_fast() to match?

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