Re: [PATCH] mm, hugetlb: use memory policy when available

2015-11-05 Thread Vlastimil Babka
On 10/20/2015 09:53 PM, Dave Hansen wrote: > From: Dave Hansen > > I have a hugetlbfs user which is never explicitly allocating huge pages > with 'nr_hugepages'. They only set 'nr_overcommit_hugepages' and then let > the pages be allocated from the buddy allocator at fault time. > > This works,

Re: [PATCH] mm, hugetlb: use memory policy when available

2015-11-03 Thread Sasha Levin
On 10/22/2015 05:42 PM, Dave Hansen wrote: > On 10/22/2015 02:39 PM, Sasha Levin wrote: >> > Trinity seems to be able to hit the newly added warnings pretty easily: > Kirill reported the same thing. Is it fixed with this applied? > >> > http://ozlabs.org/~akpm/mmots/broken-out/mm-hugetlb-use-memo

Re: [PATCH] mm, hugetlb: use memory policy when available

2015-10-22 Thread Dave Hansen
On 10/22/2015 02:39 PM, Sasha Levin wrote: > Trinity seems to be able to hit the newly added warnings pretty easily: Kirill reported the same thing. Is it fixed with this applied? > http://ozlabs.org/~akpm/mmots/broken-out/mm-hugetlb-use-memory-policy-when-available-fix.patch -- To unsubscribe

Re: [PATCH] mm, hugetlb: use memory policy when available

2015-10-22 Thread Sasha Levin
On 10/20/2015 03:53 PM, Dave Hansen wrote: > From: Dave Hansen > > I have a hugetlbfs user which is never explicitly allocating huge pages > with 'nr_hugepages'. They only set 'nr_overcommit_hugepages' and then let > the pages be allocated from the buddy allocator at fault time. > > This works,

Re: [PATCH] mm, hugetlb: use memory policy when available

2015-10-21 Thread Kirill A. Shutemov
On Tue, Oct 20, 2015 at 12:53:17PM -0700, Dave Hansen wrote: > @@ -1445,6 +1514,10 @@ static struct page *alloc_buddy_huge_pag > if (hstate_is_gigantic(h)) > return NULL; > > + if (vma || addr) { > + WARN_ON_ONCE(!addr || addr == -1); Trinity triggered the WAR

Re: [PATCH] mm, hugetlb: use memory policy when available

2015-10-20 Thread Andrew Morton
On Tue, 20 Oct 2015 12:53:17 -0700 Dave Hansen wrote: > > From: Dave Hansen > > I have a hugetlbfs user which is never explicitly allocating huge pages > with 'nr_hugepages'. They only set 'nr_overcommit_hugepages' and then let > the pages be allocated from the buddy allocator at fault time.

[PATCH] mm, hugetlb: use memory policy when available

2015-10-20 Thread Dave Hansen
From: Dave Hansen I have a hugetlbfs user which is never explicitly allocating huge pages with 'nr_hugepages'. They only set 'nr_overcommit_hugepages' and then let the pages be allocated from the buddy allocator at fault time. This works, but they noticed that mbind() was not doing them any go