Re: [PATCH v1] vdpa/mlx5: Add support for doorbell bypassing

[Eli Cohen]

On Thu, Jun 03, 2021 at 03:49:04PM +0800, Jason Wang wrote:
> 
> 在 2021/6/3 下午3:38, Eli Cohen 写道:
> > On Thu, Jun 03, 2021 at 03:11:51PM +0800, Jason Wang wrote:
> > > 在 2021/6/2 下午5:53, Eli Cohen 写道:
> > > > Implement mlx5_get_vq_notification() to return the doorbell address.
> > > > Since the notification area is mapped to userspace, make sure that the
> > > > BAR size is at least PAGE_SIZE large.
> > > > 
> > > > Signed-off-by: Eli Cohen <elic@xxxxxxxxxx>
> > > > ---
> > > > v0 --> v1:
> > > >     Make sure SF bar size is not smaller than PAGE_SIZE
> > > > 
> > > >    drivers/vdpa/mlx5/core/mlx5_vdpa.h |  1 +
> > > >    drivers/vdpa/mlx5/core/resources.c |  1 +
> > > >    drivers/vdpa/mlx5/net/mlx5_vnet.c  | 17 +++++++++++++++++
> > > >    3 files changed, 19 insertions(+)
> > > > 
> > > > diff --git a/drivers/vdpa/mlx5/core/mlx5_vdpa.h b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> > > > index 09a16a3d1b2a..0002b2136b48 100644
> > > > --- a/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> > > > +++ b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> > > > @@ -42,6 +42,7 @@ struct mlx5_vdpa_resources {
> > > >    	u32 pdn;
> > > >    	struct mlx5_uars_page *uar;
> > > >    	void __iomem *kick_addr;
> > > > +	u64 phys_kick_addr;
> > > >    	u16 uid;
> > > >    	u32 null_mkey;
> > > >    	bool valid;
> > > > diff --git a/drivers/vdpa/mlx5/core/resources.c b/drivers/vdpa/mlx5/core/resources.c
> > > > index 836ab9ef0fa6..d4606213f88a 100644
> > > > --- a/drivers/vdpa/mlx5/core/resources.c
> > > > +++ b/drivers/vdpa/mlx5/core/resources.c
> > > > @@ -253,6 +253,7 @@ int mlx5_vdpa_alloc_resources(struct mlx5_vdpa_dev *mvdev)
> > > >    		goto err_key;
> > > >    	kick_addr = mdev->bar_addr + offset;
> > > > +	res->phys_kick_addr = kick_addr;
> > > >    	res->kick_addr = ioremap(kick_addr, PAGE_SIZE);
> > > >    	if (!res->kick_addr) {
> > > > diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > index 5500bcfe84b4..1936039e05bd 100644
> > > > --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > @@ -1871,8 +1871,25 @@ static void mlx5_vdpa_free(struct vdpa_device *vdev)
> > > >    static struct vdpa_notification_area mlx5_get_vq_notification(struct vdpa_device *vdev, u16 idx)
> > > >    {
> > > > +	struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> > > >    	struct vdpa_notification_area ret = {};
> > > > +	struct mlx5_vdpa_net *ndev;
> > > > +	phys_addr_t addr;
> > > > +
> > > > +	/* If SF BAR size is smaller than PAGE_SIZE, do not use direct
> > > > +	 * notification to avoid the risk of mapping pages that contain BAR of more
> > > > +	 * than one SF
> > > > +	 */
> > > > +	if (MLX5_CAP_GEN(mvdev->mdev, log_min_sf_size) + 12 < PAGE_SHIFT)
> > > > +		return ret;
> > > > +
> > > > +	ndev = to_mlx5_vdpa_ndev(mvdev);
> > > > +	addr = (phys_addr_t)ndev->mvdev.res.phys_kick_addr;
> > > > +	if (addr & ~PAGE_MASK)
> > > > +		return ret;
> > > 
> > > This has been checked by vhost-vDPA, and it's better to leave those policy
> > > checking to them driver instead of checking it in the parent.
> > > 
> > Not in all invocations of get_vq_notification(). For example, in
> > vhost_vdpa_fault() you call remap_pfn_range() with notify.addr >>
> > PAGE_SHIFT so it it was not aligned you mask this misalignment.
> 
> 
> In order to have vhost_vdpa_fault() works, it should first pass the check of
> vhost_vdpa_mmap().
> 
> Othewise we won't install vma->vm_ops so there won't be a page fault for the
> doorbell.

Right, thanks for pointing this out.
Will send another version.

> 
> Thanks
> 
> 
> > 
> > > Thanks
> > > 
> > > 
> > > > +	ret.addr = (phys_addr_t)ndev->mvdev.res.phys_kick_addr;
> > > > +	ret.size = PAGE_SIZE;
> > > >    	return ret;
> > > >    }
>