Received-SPF: pass (google.com: domain of linux-kernel+bounces-132171-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Precedence: bulk
MIME-Version: 1.0
References: <20240404162515.527802-1-fvdl@google.com> <20240404162515.527802-2-fvdl@google.com>
 <e74cfee3-565f-4c69-bb7b-bdd40d01d212@redhat.com>
In-Reply-To: <e74cfee3-565f-4c69-bb7b-bdd40d01d212@redhat.com>
From: Frank van der Linden <fvdl@google.com>
Date: Thu, 4 Apr 2024 14:44:21 -0700
Message-ID: <CAPTztWbpDizLVk14jfxtS8mzQo3KvR+3KHiAiksxCXo=SQ4HrQ@mail.gmail.com>
Subject: Re: [PATCH 2/2] mm/hugetlb: pass correct order_per_bit to cma_declare_contiguous_nid
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, Apr 4, 2024 at 1:13=E2=80=AFPM David Hildenbrand <david@redhat.com>=
 wrote:
>
> On 04.04.24 18:25, Frank van der Linden wrote:
> > The hugetlb_cma code passes 0 in the order_per_bit argument to
> > cma_declare_contiguous_nid (the alignment, computed using the
> > page order, is correctly passed in).
> >
> > This causes a bit in the cma allocation bitmap to always represent
> > a 4k page, making the bitmaps potentially very large, and slower.
> >
> > So, correctly pass in the order instead.
> >
> > Signed-off-by: Frank van der Linden <fvdl@google.com>
> > Cc: Roman Gushchin <roman.gushchin@linux.dev>
> > Fixes: cf11e85fc08c ("mm: hugetlb: optionally allocate gigantic hugepag=
es using cma")
>
> It might be subopimal, but do we call it a "BUG" that needs "fixing". I
> know, controversial :)
>
> > ---
> >   mm/hugetlb.c | 6 +++---
> >   1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index 23ef240ba48a..6dc62d8b2a3a 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -7873,9 +7873,9 @@ void __init hugetlb_cma_reserve(int order)
> >                * huge page demotion.
> >                */
> >               res =3D cma_declare_contiguous_nid(0, size, 0,
> > -                                             PAGE_SIZE << HUGETLB_PAGE=
_ORDER,
> > -                                              0, false, name,
> > -                                              &hugetlb_cma[nid], nid);
> > +                                     PAGE_SIZE << HUGETLB_PAGE_ORDER,
> > +                                     HUGETLB_PAGE_ORDER, false, name,
> > +                                     &hugetlb_cma[nid], nid);
> >               if (res) {
> >                       pr_warn("hugetlb_cma: reservation failed: err %d,=
 node %d",
> >                               res, nid);
>
> ... I'm afraid this is not completely correct.
>
> For example, on arm64, HUGETLB_PAGE_ORDER is essentially PMD_ORDER.
>
> ... but we do support smaller hugetlb sizes than that (cont-pte hugetlb
> size is 64 KiB, not 2 MiB -- PMD -- on a 4k kernel)

Right,  smaller hugetlb page sizes exist. But, the value here is not
intended to represent the minimum hugetlb page size - it's the minimum
hugetlb page size that we can demote a CMA-allocated hugetlb page to.
See:

a01f43901cfb ("hugetlb: be sure to free demoted CMA pages to CMA")

So, this just restricts demotion of the gigantic, CMA-allocated pages
to HUGETLB_PAGE_ORDER-sized chunks.

- Frank