Re: [patch v2] mm, hugetlb: abort __get_user_pages if current has been oom killed

[Greg Thelen <gthelen@google.com>]

On Mon, Mar 09 2015, David Rientjes wrote:

> If __get_user_pages() is faulting a significant number of hugetlb pages,
> usually as the result of mmap(MAP_LOCKED), it can potentially allocate a
> very large amount of memory.
>
> If the process has been oom killed, this will cause a lot of memory to
> be overcharged to its memcg since it has access to memory reserves or
> could potentially deplete all system memory reserves.

s/memcg/hugetlb_cgroup/ but I don't think hugetlb has any
fatal_signal_pending() based overcharging.  I no objection to the patch,
but this doesn't seems like a cgroup thing, so the commit log could
stand a tweak.

> In the same way that commit 4779280d1ea4 ("mm: make get_user_pages() 
> interruptible") aborted for pending SIGKILLs when faulting non-hugetlb
> memory, based on the premise of commit 462e00cc7151 ("oom: stop
> allocating user memory if TIF_MEMDIE is set"), hugetlb page faults now
> terminate when the process has been oom killed.
>
> Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
> Cc: Davidlohr Bueso <dave@stgolabs.net>
> Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
> Signed-off-by: David Rientjes <rientjes@google.com>
> ---
>  v2: check signal inside follow_huegtlb_page() loop per Kirill
>
>  mm/hugetlb.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
>
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -3276,6 +3276,15 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
>  		struct page *page;
>  
>  		/*
> +		 * If we have a pending SIGKILL, don't keep faulting pages and
> +		 * potentially allocating memory.
> +		 */
> +		if (unlikely(fatal_signal_pending(current))) {
> +			remainder = 0;
> +			break;
> +		}
> +
> +		/*
>  		 * Some archs (sparc64, sh*) have multiple pte_ts to
>  		 * each hugepage.  We have to make sure we get the
>  		 * first, for the page indexing below to work.