On 9/20/19 12:50 PM, Leonardo Bras wrote:
> As decribed, gup_pgd_range is a lockless pagetable walk. So, in order to
> monitor against THP split/collapse with the couting method, it's necessary
> to bound it with {start,end}_lockless_pgtbl_walk.
>
> There are dummy functions, so it is not going to add any overhead on archs
> that don't use this method.
>
> Signed-off-by: Leonardo Bras <leonardo@xxxxxxxxxxxxx>
> ---
> mm/gup.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/mm/gup.c b/mm/gup.c
> index 98f13ab37bac..675e4be27082 100644
> --- a/mm/gup.c
> +++ b/mm/gup.c
> @@ -2404,6 +2404,7 @@ int get_user_pages_fast(unsigned long start, int nr_pages,
> unsigned int gup_flags, struct page **pages)
> {
> unsigned long addr, len, end;
> + struct mm_struct *mm;
> int nr = 0, ret = 0;
>
> if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM)))
> @@ -2421,9 +2422,12 @@ int get_user_pages_fast(unsigned long start, int nr_pages,
>
> if (IS_ENABLED(CONFIG_HAVE_FAST_GUP) &&
> gup_fast_permitted(start, end)) {
> + mm = current->mm;
> + start_lockless_pgtbl_walk(mm);
> local_irq_disable();
I'd also like a second opinion from the "core" -mm maintainers, but it seems like
there is now too much code around the gup_pgd_range() call. Especially since there
are two places where it's called--did you forget the other one in
__get_user_pages_fast(), btw??
Maybe the irq handling and atomic counting should be moved into start/finish
calls, like this:
start_gup_fast_walk()
gup_pgd_range()
finish_gup_fast_walk()
> gup_pgd_range(addr, end, gup_flags, pages, &nr);
> local_irq_enable();
> + end_lockless_pgtbl_walk(mm);
> ret = nr;
> }
>
>
thanks,
--
John Hubbard
NVIDIA
