Received-SPF: pass (google.com: domain of linux-kernel+bounces-166787-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Precedence: bulk
MIME-Version: 1.0
References: <20240430161437.2100295-1-fvdl@google.com> <f44e3b1b-d8f7-4b5c-a66b-13ae3f3d53bd@redhat.com>
In-Reply-To: <f44e3b1b-d8f7-4b5c-a66b-13ae3f3d53bd@redhat.com>
From: Frank van der Linden <fvdl@google.com>
Date: Thu, 2 May 2024 10:27:25 -0700
Message-ID: <CAPTztWZB1kYCETiwEAdemrqgPC+KmRMHTvb945BPjCo4QgWQvQ@mail.gmail.com>
Subject: Re: [PATCH] mm/hugetlb: align cma on allocation order, not demotion order
To: David Hildenbrand <david@redhat.com>
Cc: linux-mm@kvack.org, muchun.song@linux.dev, akpm@linux-foundation.org, 
	linux-kernel@vger.kernel.org, Roman Gushchin <roman.gushchin@linux.dev>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Thu, May 2, 2024 at 6:15=E2=80=AFAM David Hildenbrand <david@redhat.com>=
 wrote:
>
> On 30.04.24 18:14, Frank van der Linden wrote:
> > Align the CMA area for hugetlb gigantic pages to their size, not the
> > size that they can be demoted to. Otherwise there might be misaligned
> > sections at the start and end of the CMA area that will never be used
> > for hugetlb page allocations.
> >
> > Signed-off-by: Frank van der Linden <fvdl@google.com>
> > Cc: Roman Gushchin <roman.gushchin@linux.dev>
> > Fixes: a01f43901cfb ("hugetlb: be sure to free demoted CMA pages to CMA=
")
> > ---
> >   mm/hugetlb.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index 5dc3f5ea3a2e..cfe7b025c576 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -7794,7 +7794,7 @@ void __init hugetlb_cma_reserve(int order)
> >                * huge page demotion.
> >                */
> >               res =3D cma_declare_contiguous_nid(0, size, 0,
> > -                                     PAGE_SIZE << HUGETLB_PAGE_ORDER,
> > +                                     PAGE_SIZE << order,
> >                                       HUGETLB_PAGE_ORDER, false, name,
> >                                       &hugetlb_cma[nid], nid);
> >               if (res) {
>
> I was wondering how that worked when reviewing your other patch.
> Wondering why we never got a BUG report, maybe we were always lucky
> about the alignment we actually got?

I think this issue was probably masked by the hugetlb allocator
falling back to direct alloc_contig_pages allocation if cma_alloc
fails. So if you're not under memory pressure, the failure to allocate
from the misaligned areas might not have been noticed.

I noticed it, because I was working with change I made: a flag that
prevents the fallback to straight alloc_contig_pages, as that behavior
may not be desired - you don't want to potentially eat in to
non-movable space that the kernel needs, it might be better to fail if
there's no CMA available.
>
> We round up size to PAGE_SIZE << order, so that's the alignment we need.
>
> Reviewed-by: David Hildenbrand <david@redhat.com>

Thanks!