On Fri, Jun 9, 2023 at 3:45 AM Michael S. Tsirkin <mst@xxxxxxxxxx> wrote:
>
> On Thu, Jun 08, 2023 at 05:01:24PM +0800, Angus Chen wrote:
> > When add virtio_pci vdpa device,check the vqs number of device cap
> > and max_vq_pairs from add_config.
> > Simply starting from failing if the provisioned #qp is not
> > equal to the one that hardware has.
I think I kind of agree with Michael, I don't see any obvious
advantages to allow usersapce to configure max_vqp if it can't be
provisioned dynamically. What's wrong if we just stick the current
approach that doesn't accept max_vqp?
A better approach is to tweak the vdpa tool to display the legal
attributes that can be provisioned.
> >
> > Signed-off-by: Angus Chen <angus.chen@xxxxxxxxxxxxxxx>
>
> I am not sure about this one. How does userspace know
> which values are legal?
vdpa mgmtdev show can gives hints like:
max_supported_vqs 3
>
> If there's no way then maybe we should just cap the value
> to what device can support but otherwise keep the device
> working.
This seems conflict to how other drivers (like mlx5) did:
if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MAX_VQP)) {
if (add_config->net.max_vq_pairs > max_vqs / 2)
return -EINVAL;
max_vqs = min_t(u32, max_vqs, 2 * add_config->net.max_vq_pairs);
} else {
max_vqs = 2;
}
Thanks
>
> > ---
> > v1: Use max_vqs from add_config
> > v2: Just return fail if max_vqs from add_config is not same as device
> > cap. Suggested by jason.
> >
> > drivers/vdpa/virtio_pci/vp_vdpa.c | 35 ++++++++++++++++++-------------
> > 1 file changed, 21 insertions(+), 14 deletions(-)
> >
> > diff --git a/drivers/vdpa/virtio_pci/vp_vdpa.c b/drivers/vdpa/virtio_pci/vp_vdpa.c
> > index 281287fae89f..c1fb6963da12 100644
> > --- a/drivers/vdpa/virtio_pci/vp_vdpa.c
> > +++ b/drivers/vdpa/virtio_pci/vp_vdpa.c
> > @@ -480,32 +480,39 @@ static int vp_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
> > u64 device_features;
> > int ret, i;
> >
> > - vp_vdpa = vdpa_alloc_device(struct vp_vdpa, vdpa,
> > - dev, &vp_vdpa_ops, 1, 1, name, false);
> > -
> > - if (IS_ERR(vp_vdpa)) {
> > - dev_err(dev, "vp_vdpa: Failed to allocate vDPA structure\n");
> > - return PTR_ERR(vp_vdpa);
> > + if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MAX_VQP)) {
> > + if (add_config->net.max_vq_pairs != (v_mdev->max_supported_vqs / 2)) {
> > + dev_err(&pdev->dev, "max vqs 0x%x should be equal to 0x%x which device has\n",
> > + add_config->net.max_vq_pairs*2, v_mdev->max_supported_vqs);
> > + return -EINVAL;
> > + }
> > }
> >
> > - vp_vdpa_mgtdev->vp_vdpa = vp_vdpa;
> > -
> > - vp_vdpa->vdpa.dma_dev = &pdev->dev;
> > - vp_vdpa->queues = vp_modern_get_num_queues(mdev);
> > - vp_vdpa->mdev = mdev;
> > -
> > device_features = vp_modern_get_features(mdev);
> > if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_FEATURES)) {
> > if (add_config->device_features & ~device_features) {
> > - ret = -EINVAL;
> > dev_err(&pdev->dev, "Try to provision features "
> > "that are not supported by the device: "
> > "device_features 0x%llx provisioned 0x%llx\n",
> > device_features, add_config->device_features);
> > - goto err;
> > + return -EINVAL;
> > }
> > device_features = add_config->device_features;
> > }
> > +
> > + vp_vdpa = vdpa_alloc_device(struct vp_vdpa, vdpa,
> > + dev, &vp_vdpa_ops, 1, 1, name, false);
> > +
> > + if (IS_ERR(vp_vdpa)) {
> > + dev_err(dev, "vp_vdpa: Failed to allocate vDPA structure\n");
> > + return PTR_ERR(vp_vdpa);
> > + }
> > +
> > + vp_vdpa_mgtdev->vp_vdpa = vp_vdpa;
> > +
> > + vp_vdpa->vdpa.dma_dev = &pdev->dev;
> > + vp_vdpa->queues = v_mdev->max_supported_vqs;
> > + vp_vdpa->mdev = mdev;
> > vp_vdpa->device_features = device_features;
> >
> > ret = devm_add_action_or_reset(dev, vp_vdpa_free_irq_vectors, pdev);
> > --
> > 2.25.1
>
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
