lkml.org 
[lkml]   [2020]   [Jul]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v4 04/11] mm/hugetlb: make hugetlb migration callback CMA aware
On Tue, Jul 07, 2020 at 01:31:16PM +0200, Michal Hocko wrote:
> On Tue 07-07-20 16:44:42, Joonsoo Kim wrote:
> > From: Joonsoo Kim <iamjoonsoo.kim@lge.com>
> >
> > new_non_cma_page() in gup.c which try to allocate migration target page
> > requires to allocate the new page that is not on the CMA area.
> > new_non_cma_page() implements it by removing __GFP_MOVABLE flag. This way
> > works well for THP page or normal page but not for hugetlb page.
> >
> > hugetlb page allocation process consists of two steps. First is dequeing
> > from the pool. Second is, if there is no available page on the queue,
> > allocating from the page allocator.
> >
> > new_non_cma_page() can control allocation from the page allocator by
> > specifying correct gfp flag. However, dequeing cannot be controlled until
> > now, so, new_non_cma_page() skips dequeing completely. It is a suboptimal
> > since new_non_cma_page() cannot utilize hugetlb pages on the queue so this
> > patch tries to fix this situation.
> >
> > This patch makes the deque function on hugetlb CMA aware and skip CMA
> > pages if newly added skip_cma argument is passed as true.
>
> I really dislike this as already mentioned in the previous version of
> the patch. You are making hugetlb and only one part of its allocator a
> special snowflake which is almost always a bad idea. Your changelog
> lacks any real justification for this inconsistency.
>
> Also by doing so you are keeping an existing bug that the hugetlb
> allocator doesn't respect scope gfp flags as I have mentioned when
> reviewing the previous version. That bug likely doesn't matter now but
> it might in future and as soon as it is fixed all this is just a
> pointless exercise.
>
> I do not have energy and time to burn to repeat that argumentation to I
> will let Mike to have a final word. Btw. you are keeping his acks even
> after considerable changes to patches which I am not really sure he is
> ok with.

As you replied a minute ago, Mike acked.

> > Acked-by: Mike Kravetz <mike.kravetz@oracle.com>
> > Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
>
> To this particular patch.
> [...]
>
> > diff --git a/mm/gup.c b/mm/gup.c
> > index 5daadae..2c3dab4 100644
> > --- a/mm/gup.c
> > +++ b/mm/gup.c
> > @@ -1630,11 +1630,12 @@ static struct page *new_non_cma_page(struct page *page, unsigned long private)
> > #ifdef CONFIG_HUGETLB_PAGE
> > if (PageHuge(page)) {
> > struct hstate *h = page_hstate(page);
> > +
> > /*
> > * We don't want to dequeue from the pool because pool pages will
> > * mostly be from the CMA region.
> > */
> > - return alloc_migrate_huge_page(h, gfp_mask, nid, NULL);
> > + return alloc_huge_page_nodemask(h, nid, NULL, gfp_mask, true);
>
> Let me repeat that this whole thing is running under
> memalloc_nocma_save. So additional parameter is bogus.

As Vlasimil said in other reply, we are not under
memalloc_nocma_save(). Anyway, now, I also think that additional parameter
isn't need.

> [...]
> > -static struct page *dequeue_huge_page_node_exact(struct hstate *h, int nid)
> > +static struct page *dequeue_huge_page_node_exact(struct hstate *h, int nid, bool skip_cma)
>
> If you really insist on an additional parameter at this layer than it
> should be checking for the PF_MEMALLOC_NOCMA instead.

I will change the patch to check PF_MEMALLOC_NOCMA instead of
introducing and checking skip_cma.

> [...]
> > @@ -1971,21 +1977,29 @@ struct page *alloc_buddy_huge_page_with_mpol(struct hstate *h,
> >
> > /* page migration callback function */
> > struct page *alloc_huge_page_nodemask(struct hstate *h, int preferred_nid,
> > - nodemask_t *nmask, gfp_t gfp_mask)
> > + nodemask_t *nmask, gfp_t gfp_mask, bool skip_cma)
> > {
> > + unsigned int flags = 0;
> > + struct page *page = NULL;
> > +
> > + if (skip_cma)
> > + flags = memalloc_nocma_save();
>
> This is pointless for a scope that is already defined up in the call
> chain and fundamentally this is breaking the expected use of the scope
> API. The primary reason for that API to exist is to define the scope and
> have it sticky for _all_ allocation contexts. So if you have to use it
> deep in the allocator then you are doing something wrong.

As mentioned above, we are not under memalloc_nocma_save(). Anyway, I
will rework the patch and attach it to Vlasimil's reply. It's appreciate
if you check it again.

Thanks.

\
 
 \ /
  Last update: 2020-07-08 09:14    [W:0.930 / U:0.224 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site