On Sun 23-08-15 16:23:37, Tetsuo Handa wrote:
> >From 08a638e04351386ab03cd1223988ac7940d4d3aa Mon Sep 17 00:00:00 2001
> From: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
> Date: Sat, 1 Aug 2015 22:46:12 +0900
> Subject: [PATCH 2/2] mm: Fix potentially scheduling in GFP_ATOMIC
> allocations.
>
> Currently, if somebody does GFP_ATOMIC | __GFP_NOFAIL allocation,
This combination of flags is broken by definition and I fail to see it
being used anywhere in the kernel.
> wait_iff_congested() might be called via __alloc_pages_high_priority()
> before reaching
>
> if (!wait) {
> WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
> goto nopage;
> }
>
> because gfp_to_alloc_flags() includes ALLOC_NO_WATERMARKS if TIF_MEMDIE
> was set.
>
> We need to check for __GFP_WAIT flag at __alloc_pages_high_priority()
> in order to make sure that we won't schedule.
I do not think this is an improvement. It is true we are already failing
__GFP_NOFAIL & ~__GFP_WAIT but I believe it doesn't make much sense
to replace one buggy behavior (sleeping in atomic context) by another
(failing __GFP_NOFAIL). It is the caller which should be fixed here. We
should get "scheduling while atomic:" and the trace with the current
code so we are not loosing any debugging options.
> Signed-off-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
> ---
> mm/page_alloc.c | 13 ++++++-------
> 1 file changed, 6 insertions(+), 7 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 37a0390..f9f09fa 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -2917,16 +2917,15 @@ __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
> {
> struct page *page;
>
> - do {
> + for (;;) {
> page = get_page_from_freelist(gfp_mask, order,
> ALLOC_NO_WATERMARKS, ac);
>
> - if (!page && gfp_mask & __GFP_NOFAIL)
> - wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
> - HZ/50);
> - } while (!page && (gfp_mask & __GFP_NOFAIL));
> -
> - return page;
> + if (page || (gfp_mask & (__GFP_NOFAIL | __GFP_WAIT)) !=
> + (__GFP_NOFAIL | __GFP_WAIT))
> + return page;
> + wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
> + }
> }
>
> static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
> --
> 1.8.3.1
--
Michal Hocko
SUSE Labs
--
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>