2021-06-03 08:13:35

by Eli Cohen

Subject: [PATCH v1] vdpa/mlx5: Clear vq ready indication upon device reset

After device reset, the virtqueues are not ready so clear the ready
field.

Failing to do so can result in virtio_vdpa failing to load if the device
was previously used by vhost_vdpa and the old values are ready.
virtio_vdpa expects to find VQs in "not ready" state.

Fixes: 1a86b377aa21 ("vdpa/mlx5: Add VDPA driver for supported mlx5 devices")
Signed-off-by: Eli Cohen <[email protected]>
---
v0 --> v1:
Make sure clear of ready is done only in reset flow

drivers/vdpa/mlx5/net/mlx5_vnet.c | 9 +++++++++
1 file changed, 9 insertions(+)

diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
index 02a05492204c..eaeae67e0ff0 100644
--- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
+++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
@@ -1766,6 +1766,14 @@ static void teardown_driver(struct mlx5_vdpa_net *ndev)
mutex_unlock(&ndev->reslock);
}

+static void clear_vqs_ready(struct mlx5_vdpa_net *ndev)
+{
+ int i;
+
+ for (i = ndev->mvdev.max_vqs - 1; i >= 0; i--)
+ ndev->vqs[i].ready = false;
+}
+
static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
{
struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
@@ -1776,6 +1784,7 @@ static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
if (!status) {
mlx5_vdpa_info(mvdev, "performing device reset\n");
teardown_driver(ndev);
+ clear_vqs_ready(ndev);
mlx5_vdpa_destroy_mr(&ndev->mvdev);
ndev->mvdev.status = 0;
ndev->mvdev.mlx_features = 0;
--
2.31.1

2021-06-04 02:37:53

by Jason Wang

Subject: Re: [PATCH v1] vdpa/mlx5: Clear vq ready indication upon device reset

?? 2021/6/3 ????4:10, Eli Cohen д??:
> After device reset, the virtqueues are not ready so clear the ready
> field.
>
> Failing to do so can result in virtio_vdpa failing to load if the device
> was previously used by vhost_vdpa and the old values are ready.
> virtio_vdpa expects to find VQs in "not ready" state.
>
> Fixes: 1a86b377aa21 ("vdpa/mlx5: Add VDPA driver for supported mlx5 devices")
> Signed-off-by: Eli Cohen <[email protected]>
> ---
> v0 --> v1:
> Make sure clear of ready is done only in reset flow
>
> drivers/vdpa/mlx5/net/mlx5_vnet.c | 9 +++++++++
> 1 file changed, 9 insertions(+)
>
> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> index 02a05492204c..eaeae67e0ff0 100644
> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> @@ -1766,6 +1766,14 @@ static void teardown_driver(struct mlx5_vdpa_net *ndev)
> mutex_unlock(&ndev->reslock);
> }
>
> +static void clear_vqs_ready(struct mlx5_vdpa_net *ndev)
> +{
> + int i;
> +
> + for (i = ndev->mvdev.max_vqs - 1; i >= 0; i--)
> + ndev->vqs[i].ready = false;


The patch looks correct but I wonder the reason for the decreasing used
in the loop.

Usually, it means it has some reason that must be done in that way.

Thanks


> +}
> +
> static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
> {
> struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> @@ -1776,6 +1784,7 @@ static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
> if (!status) {
> mlx5_vdpa_info(mvdev, "performing device reset\n");
> teardown_driver(ndev);
> + clear_vqs_ready(ndev);
> mlx5_vdpa_destroy_mr(&ndev->mvdev);
> ndev->mvdev.status = 0;
> ndev->mvdev.mlx_features = 0;

2021-06-06 05:09:05

by Eli Cohen

Subject: Re: [PATCH v1] vdpa/mlx5: Clear vq ready indication upon device reset

On Fri, Jun 04, 2021 at 10:34:29AM +0800, Jason Wang wrote:
>
> 在 2021/6/3 下午4:10, Eli Cohen 写道:
> > After device reset, the virtqueues are not ready so clear the ready
> > field.
> >
> > Failing to do so can result in virtio_vdpa failing to load if the device
> > was previously used by vhost_vdpa and the old values are ready.
> > virtio_vdpa expects to find VQs in "not ready" state.
> >
> > Fixes: 1a86b377aa21 ("vdpa/mlx5: Add VDPA driver for supported mlx5 devices")
> > Signed-off-by: Eli Cohen <[email protected]>
> > ---
> > v0 --> v1:
> > Make sure clear of ready is done only in reset flow
> >
> > drivers/vdpa/mlx5/net/mlx5_vnet.c | 9 +++++++++
> > 1 file changed, 9 insertions(+)
> >
> > diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > index 02a05492204c..eaeae67e0ff0 100644
> > --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > @@ -1766,6 +1766,14 @@ static void teardown_driver(struct mlx5_vdpa_net *ndev)
> > mutex_unlock(&ndev->reslock);
> > }
> > +static void clear_vqs_ready(struct mlx5_vdpa_net *ndev)
> > +{
> > + int i;
> > +
> > + for (i = ndev->mvdev.max_vqs - 1; i >= 0; i--)
> > + ndev->vqs[i].ready = false;
>
>
> The patch looks correct but I wonder the reason for the decreasing used in
> the loop.

No special reason. I just copy pasted it from somewhere else in the
code. Will post another version.
>
> Usually, it means it has some reason that must be done in that way.
>
> Thanks
>
>
> > +}
> > +
> > static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
> > {
> > struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> > @@ -1776,6 +1784,7 @@ static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
> > if (!status) {
> > mlx5_vdpa_info(mvdev, "performing device reset\n");
> > teardown_driver(ndev);
> > + clear_vqs_ready(ndev);
> > mlx5_vdpa_destroy_mr(&ndev->mvdev);
> > ndev->mvdev.status = 0;
> > ndev->mvdev.mlx_features = 0;
>