Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:5 as permitted sender) client-ip=2620:137:e000::3:5;
MIME-Version: 1.0
References: <20231027054154.2935054-1-lizhijian@fujitsu.com>
 <4da48f85-a72f-4f6b-900f-fc293d63b5ae@fujitsu.com> <20231030124041.GE691768@ziepe.ca>
 <7c5dd149-395a-46a2-96a0-89c182105eaf@linux.dev> <20231031131928.GH691768@ziepe.ca>
In-Reply-To: <20231031131928.GH691768@ziepe.ca>
From:   Greg Sword <gregsword0@gmail.com>
Date:   Wed, 1 Nov 2023 08:58:27 +0800
Message-ID: <CAEz=LcsoDbGmTmVmGyPbcoFjahyf-ruuddjbSXE2W5EH9KDtmA@mail.gmail.com>
Subject: Re: [PATCH RFC 1/2] RDMA/rxe: don't allow registering !PAGE_SIZE mr
To:     Jason Gunthorpe <jgg@ziepe.ca>
Cc:     Zhu Yanjun <yanjun.zhu@linux.dev>,
        "Zhijian Li (Fujitsu)" <lizhijian@fujitsu.com>,
        "zyjzyj2000@gmail.com" <zyjzyj2000@gmail.com>,
        "leon@kernel.org" <leon@kernel.org>,
        "linux-rdma@vger.kernel.org" <linux-rdma@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "rpearsonhpe@gmail.com" <rpearsonhpe@gmail.com>,
        "Daisuke Matsuda (Fujitsu)" <matsuda-daisuke@fujitsu.com>,
        "bvanassche@acm.org" <bvanassche@acm.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Tue, Oct 31, 2023 at 9:19=E2=80=AFPM Jason Gunthorpe <jgg@ziepe.ca> wrot=
e:
>
> On Tue, Oct 31, 2023 at 04:52:23PM +0800, Zhu Yanjun wrote:
> > =E5=9C=A8 2023/10/30 20:40, Jason Gunthorpe =E5=86=99=E9=81=93:
> > > On Mon, Oct 30, 2023 at 07:51:41AM +0000, Zhijian Li (Fujitsu) wrote:
> > > >
> > > >
> > > > On 27/10/2023 13:41, Li Zhijian wrote:
> > > > > mr->page_list only encodes *page without page offset, when
> > > > > page_size !=3D PAGE_SIZE, we cannot restore the address with a wr=
ong
> > > > > page_offset.
> > > > >
> > > > > Note that this patch will break some ULPs that try to register 4K
> > > > > MR when PAGE_SIZE is not 4K.
> > > > > SRP and nvme over RXE is known to be impacted.
> > > > >
> > > > > Signed-off-by: Li Zhijian <lizhijian@fujitsu.com>
> > > > > ---
> > > > >    drivers/infiniband/sw/rxe/rxe_mr.c | 6 ++++++
> > > > >    1 file changed, 6 insertions(+)
> > > > >
> > > > > diff --git a/drivers/infiniband/sw/rxe/rxe_mr.c b/drivers/infinib=
and/sw/rxe/rxe_mr.c
> > > > > index f54042e9aeb2..61a136ea1d91 100644
> > > > > --- a/drivers/infiniband/sw/rxe/rxe_mr.c
> > > > > +++ b/drivers/infiniband/sw/rxe/rxe_mr.c
> > > > > @@ -234,6 +234,12 @@ int rxe_map_mr_sg(struct ib_mr *ibmr, struct=
 scatterlist *sgl,
> > > > >         struct rxe_mr *mr =3D to_rmr(ibmr);
> > > > >         unsigned int page_size =3D mr_page_size(mr);
> > > > > +       if (page_size !=3D PAGE_SIZE) {
> > > >
> > > > It seems this condition is too strict, it should be:
> > > >   if (!IS_ALIGNED(page_size, PAGE_SIZE))
> > > >
> > > > So that, page_size with (N * PAGE_SIZE) can work as previously.
> > > > Because the offset(mr.iova & page_mask) will get lost only when !IS=
_ALIGNED(page_size, PAGE_SIZE)
> > >
> > > That makes sense
> >
> > I read all the discussions very carefully.
> >
> > Thanks, Greg.
> >
> > Because RXE only supports PAGE_SIZE, when CONFIG_ARM64_64K_PAGES is ena=
bled,
> > the PAGE_SIZE is 64K, when CONFIG_ARM64_64K_PAGES is disabled, PAGE_SIZ=
E is
> > 4K.
> >
> > But NVMe calls ib_map_mr_sg with a fixed size SZ_4K. When
> > CONFIG_ARM64_64K_PAGES is enabled, it is still 4K. This is not a proble=
m in
> > RXE. This problem is in NVMe.
>
> Maybe, but no real RDMA devices don't support 4K.
>
> The xarray conversion may need revision to use physical addresses
> instead of storing struct pages so it can handle this kind of
> segmentation.

This problem can not be fixed until rxe supports multiple page sizes
including 4K page size.
Now it is not fixed. It is an intermediate way.

>
> Certainly in the mean time it should be rejected.
>
> Jason