[lkml]   [2020]   [Sep]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH] vfio: add a singleton check for vfio_group_pin_pages
On Tue, 15 Sep 2020 08:30:42 +0800
Yan Zhao <> wrote:

> vfio_pin_pages() and vfio_group_pin_pages() are used purely to mark dirty
> pages to devices with IOMMU backend as they already have all VM pages
> pinned at VM startup.

This is wrong. The entire initial basis of mdev devices is for
non-IOMMU backed devices which provide mediation outside of the scope
of the IOMMU. That mediation includes interpreting device DMA
programming and making use of the vfio_pin_pages() interface to
translate and pin IOVA address to HPA. Marking pages dirty is a
secondary feature.

> when there're multiple devices in the vfio group, the dirty pages
> marked through pin_pages interface by one device is not useful as the
> other devices may access and dirty any VM pages.

I don't know of any cases where there are multiple devices in a group
that would make use of this interface, however, all devices within a
group necessarily share an IOMMU context and any one device dirtying a
page will dirty that page for all devices, so I don't see that this is
a valid statement either.

> So added a check such that only singleton IOMMU groups can pin pages
> in vfio_group_pin_pages. for mdevs, there's always only one dev in a
> vfio group.
> This is a fix to the commit below that added a singleton IOMMU group
> check in vfio_pin_pages.

None of the justification above is accurate, please try again. Thanks,


> Fixes: 95fc87b44104 (vfio: Selective dirty page tracking if IOMMU backed
> device pins pages)
> Signed-off-by: Yan Zhao <>
> ---
> drivers/vfio/vfio.c | 3 +++
> 1 file changed, 3 insertions(+)
> diff --git a/drivers/vfio/vfio.c b/drivers/vfio/vfio.c
> index 5e6e0511b5aa..2f0fa272ebf2 100644
> --- a/drivers/vfio/vfio.c
> +++ b/drivers/vfio/vfio.c
> @@ -2053,6 +2053,9 @@ int vfio_group_pin_pages(struct vfio_group *group,
> if (!group || !user_iova_pfn || !phys_pfn || !npage)
> return -EINVAL;
> + if (group->dev_counter > 1)
> + return -EINVAL;
> +
> return -E2BIG;

 \ /
  Last update: 2020-09-15 21:05    [W:0.043 / U:1.044 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site