Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 8 Feb 2021 12:04:45 +0200
From:   Eli Cohen <elic@nvidia.com>
To:     Jason Wang <jasowang@redhat.com>
CC:     Si-Wei Liu <si-wei.liu@oracle.com>, <mst@redhat.com>,
        <virtualization@lists.linux-foundation.org>,
        <netdev@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
        <lulu@redhat.com>
Subject: Re: [PATCH v1] vdpa/mlx5: Restore the hardware used index after
 change map
Message-ID: <20210208100445.GA173340@mtl-vdi-166.wap.labs.mlnx>
References: <20210204073618.36336-1-elic@nvidia.com>
 <81f5ce4f-cdb0-26cd-0dce-7ada824b1b86@oracle.com>
 <f2206fa2-0ddc-1858-54e7-71614b142e46@redhat.com>
 <20210208063736.GA166546@mtl-vdi-166.wap.labs.mlnx>
 <0d592ed0-3cea-cfb0-9b7b-9d2755da3f12@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable
In-Reply-To: <0d592ed0-3cea-cfb0-9b7b-9d2755da3f12@redhat.com>
User-Agent: Mutt/1.9.5 (bf161cf53efb) (2018-04-13)
Precedence: bulk

On Mon, Feb 08, 2021 at 05:04:27PM +0800, Jason Wang wrote:
>=20
> On 2021/2/8 =E4=B8=8B=E5=8D=882:37, Eli Cohen wrote:
> > On Mon, Feb 08, 2021 at 12:27:18PM +0800, Jason Wang wrote:
> > > On 2021/2/6 =E4=B8=8A=E5=8D=887:07, Si-Wei Liu wrote:
> > > >=20
> > > > On 2/3/2021 11:36 PM, Eli Cohen wrote:
> > > > > When a change of memory map occurs, the hardware resources are de=
stroyed
> > > > > and then re-created again with the new memory map. In such case, =
we need
> > > > > to restore the hardware available and used indices. The driver fa=
iled to
> > > > > restore the used index which is added here.
> > > > >=20
> > > > > Also, since the driver also fails to reset the available and used
> > > > > indices upon device reset, fix this here to avoid regression caus=
ed by
> > > > > the fact that used index may not be zero upon device reset.
> > > > >=20
> > > > > Fixes: 1a86b377aa21 ("vdpa/mlx5: Add VDPA driver for supported ml=
x5
> > > > > devices")
> > > > > Signed-off-by: Eli Cohen <elic@nvidia.com>
> > > > > ---
> > > > > v0 -> v1:
> > > > > Clear indices upon device reset
> > > > >=20
> > > > >  =C2=A0 drivers/vdpa/mlx5/net/mlx5_vnet.c | 18 ++++++++++++++++++
> > > > >  =C2=A0 1 file changed, 18 insertions(+)
> > > > >=20
> > > > > diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > index 88dde3455bfd..b5fe6d2ad22f 100644
> > > > > --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > @@ -87,6 +87,7 @@ struct mlx5_vq_restore_info {
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 u64 device_addr;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 u64 driver_addr;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 u16 avail_index;
> > > > > +=C2=A0=C2=A0=C2=A0 u16 used_index;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 bool ready;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 struct vdpa_callback cb;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 bool restore;
> > > > > @@ -121,6 +122,7 @@ struct mlx5_vdpa_virtqueue {
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 u32 virtq_id;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 struct mlx5_vdpa_net *ndev;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 u16 avail_idx;
> > > > > +=C2=A0=C2=A0=C2=A0 u16 used_idx;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 int fw_state;
> > > > >  =C2=A0 =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 /* keep last in the struct=
 */
> > > > > @@ -804,6 +806,7 @@ static int create_virtqueue(struct mlx5_vdpa_=
net
> > > > > *ndev, struct mlx5_vdpa_virtque
> > > > >  =C2=A0 =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 obj_context =3D MLX5_ADDR_=
OF(create_virtio_net_q_in, in,
> > > > > obj_context);
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 MLX5_SET(virtio_net_q_object, obj=
_context, hw_available_index,
> > > > > mvq->avail_idx);
> > > > > +=C2=A0=C2=A0=C2=A0 MLX5_SET(virtio_net_q_object, obj_context, hw=
_used_index,
> > > > > mvq->used_idx);
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 MLX5_SET(virtio_net_q_object, obj=
_context,
> > > > > queue_feature_bit_mask_12_3,
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 get=
_features_12_3(ndev->mvdev.actual_features));
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 vq_ctx =3D MLX5_ADDR_OF(virtio_ne=
t_q_object, obj_context,
> > > > > virtio_q_context);
> > > > > @@ -1022,6 +1025,7 @@ static int connect_qps(struct mlx5_vdpa_net
> > > > > *ndev, struct mlx5_vdpa_virtqueue *m
> > > > >  =C2=A0 struct mlx5_virtq_attr {
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 u8 state;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 u16 available_index;
> > > > > +=C2=A0=C2=A0=C2=A0 u16 used_index;
> > > > >  =C2=A0 };
> > > > >  =C2=A0 =C2=A0 static int query_virtqueue(struct mlx5_vdpa_net *n=
dev, struct
> > > > > mlx5_vdpa_virtqueue *mvq,
> > > > > @@ -1052,6 +1056,7 @@ static int query_virtqueue(struct
> > > > > mlx5_vdpa_net *ndev, struct mlx5_vdpa_virtqueu
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 memset(attr, 0, sizeof(*attr));
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 attr->state =3D MLX5_GET(virtio_n=
et_q_object, obj_context, state);
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 attr->available_index =3D MLX5_GE=
T(virtio_net_q_object,
> > > > > obj_context, hw_available_index);
> > > > > +=C2=A0=C2=A0=C2=A0 attr->used_index =3D MLX5_GET(virtio_net_q_ob=
ject, obj_context,
> > > > > hw_used_index);
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 kfree(out);
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 return 0;
> > > > >  =C2=A0 @@ -1535,6 +1540,16 @@ static void teardown_virtqueues(st=
ruct
> > > > > mlx5_vdpa_net *ndev)
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 }
> > > > >  =C2=A0 }
> > > > >  =C2=A0 +static void clear_virtqueues(struct mlx5_vdpa_net *ndev)
> > > > > +{
> > > > > +=C2=A0=C2=A0=C2=A0 int i;
> > > > > +
> > > > > +=C2=A0=C2=A0=C2=A0 for (i =3D ndev->mvdev.max_vqs - 1; i >=3D 0;=
 i--) {
> > > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 ndev->vqs[i].avail_id=
x =3D 0;
> > > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 ndev->vqs[i].used_idx=
 =3D 0;
> > > > > +=C2=A0=C2=A0=C2=A0 }
> > > > > +}
> > > > > +
> > > > >  =C2=A0 /* TODO: cross-endian support */
> > > > >  =C2=A0 static inline bool mlx5_vdpa_is_little_endian(struct mlx5=
_vdpa_dev
> > > > > *mvdev)
> > > > >  =C2=A0 {
> > > > > @@ -1610,6 +1625,7 @@ static int save_channel_info(struct
> > > > > mlx5_vdpa_net *ndev, struct mlx5_vdpa_virtqu
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 return er=
r;
> > > > >  =C2=A0 =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 ri->avail_index =3D attr.a=
vailable_index;
> > > > > +=C2=A0=C2=A0=C2=A0 ri->used_index =3D attr.used_index;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 ri->ready =3D mvq->ready;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 ri->num_ent =3D mvq->num_ent;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 ri->desc_addr =3D mvq->desc_addr;
> > > > > @@ -1654,6 +1670,7 @@ static void restore_channels_info(struct
> > > > > mlx5_vdpa_net *ndev)
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=
=A0=C2=A0=C2=A0 continue;
> > > > >  =C2=A0 =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 mv=
q->avail_idx =3D ri->avail_index;
> > > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 mvq->used_idx =3D ri-=
>used_index;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 mvq->read=
y =3D ri->ready;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 mvq->num_=
ent =3D ri->num_ent;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 mvq->desc=
_addr =3D ri->desc_addr;
> > > > > @@ -1768,6 +1785,7 @@ static void mlx5_vdpa_set_status(struct
> > > > > vdpa_device *vdev, u8 status)
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 if (!status) {
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 mlx5_vdpa=
_info(mvdev, "performing device reset\n");
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 teardown_=
driver(ndev);
> > > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 clear_virtqueues(ndev=
);
> > > > The clearing looks fine at the first glance, as it aligns with the =
other
> > > > state cleanups floating around at the same place. However, the thin=
g is
> > > > get_vq_state() is supposed to be called right after to get sync'ed =
with
> > > > the latest internal avail_index from device while vq is stopped. Th=
e
> > > > index was saved in the driver software at vq suspension, but before=
 the
> > > > virtq object is destroyed. We shouldn't clear the avail_index too e=
arly.
> > >=20
> > > Good point.
> > >=20
> > > There's a limitation on the virtio spec and vDPA framework that we ca=
n not
> > > simply differ device suspending from device reset.
> > >=20
> > Are you talking about live migration where you reset the device but
> > still want to know how far it progressed in order to continue from the
> > same place in the new VM?
>=20
>=20
> Yes. So if we want to support live migration at we need:
>=20
> in src node:
> 1) suspend the device
> 2) get last_avail_idx via get_vq_state()
>=20
> in the dst node:
> 3) set last_avail_idx via set_vq_state()
> 4) resume the device
>=20
> So you can see, step 2 requires the device/driver not to forget the
> last_avail_idx.
>=20

Just to be sure, what really matters here is the used index. Becuase the
vriqtueue itself is copied from the src VM to the dest VM. The available
index is alreay there and we know the hardware reads it from there.

So it puzzles me why is set_vq_state() we do not communicate the saved
used index.

> The annoying thing is that, in the virtio spec there's no definition of
> device suspending. So we reuse set_status(0) right now for vq suspending.
> Then if we forget last_avail_idx in set_status(0), it will break the
> assumption of step 2).
>=20
>=20
> >=20
> > > Need to think about that. I suggest a new state in [1], the issue is =
that
> > > people doesn't like the asynchronous API that it introduces.
> > >=20
> > >=20
> > > > Possibly it can be postponed to where VIRTIO_CONFIG_S_DRIVER_OK get=
s set
> > > > again, i.e. right before the setup_driver() in mlx5_vdpa_set_status=
()?
> > >=20
> > > Looks like a good workaround.
>=20
>=20
> Rethink of this, this won't work for the step 4), if we reuse the S_DRING=
_OK
> for resuming.
>=20
> The most clean way is to invent the feature in virtio spec and implement
> that in the driver.
>=20
> Thanks
>=20
>=20
>=20
> > >=20
> > > Thanks
> > >=20
> > >=20
> > > > -Siwei
> > >=20
> > > [1]
> > > https://lists.oasis-open.org/archives/virtio-comment/202012/msg00029.=
html
> > >=20
> > >=20
> > > > > mlx5_vdpa_destroy_mr(&ndev->mvdev);
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 ndev->mvd=
ev.status =3D 0;
> > > > >  =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 ndev->mvd=
ev.mlx_features =3D 0;
>=20