[Intel-gfx] [PATCH v3 05/12] vfio/pci: Allow passing zero-length fd array in VFIO_DEVICE_PCI_HOT_RESET
Liu, Yi L
yi.l.liu at intel.com
Wed Apr 5 08:09:46 UTC 2023
Hi Eric,
> From: Eric Auger <eric.auger at redhat.com>
> Sent: Wednesday, April 5, 2023 4:02 PM
>
> On 4/4/23 22:18, Alex Williamson wrote:
> > On Sat, 1 Apr 2023 07:44:22 -0700
> > Yi Liu <yi.l.liu at intel.com> wrote:
> >
> >> as an alternative method for ownership check when iommufd is used. In
> >> this case all opened devices in the affected dev_set are verified to
> >> be bound to a same valid iommufd value to allow reset. It's simpler
> >> and faster as user does not need to pass a set of fds and kernel no
> >> need to search the device within the given fds.
> >>
> >> a device in noiommu mode doesn't have a valid iommufd, so this method
> >> should not be used in a dev_set which contains multiple devices and one
> >> of them is in noiommu. The only allowed noiommu scenario is that the
> >> calling device is noiommu and it's in a singleton dev_set.
> >>
> >> Suggested-by: Jason Gunthorpe <jgg at nvidia.com>
> >> Signed-off-by: Jason Gunthorpe <jgg at nvidia.com>
> >> Reviewed-by: Jason Gunthorpe <jgg at nvidia.com>
> >> Tested-by: Yanting Jiang <yanting.jiang at intel.com>
> >> Signed-off-by: Yi Liu <yi.l.liu at intel.com>
> >> ---
> >> drivers/vfio/pci/vfio_pci_core.c | 42 +++++++++++++++++++++++++++-----
> >> include/uapi/linux/vfio.h | 9 ++++++-
> >> 2 files changed, 44 insertions(+), 7 deletions(-)
> >>
> >> diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c
> >> index 3696b8e58445..b68fcba67a4b 100644
> >> --- a/drivers/vfio/pci/vfio_pci_core.c
> >> +++ b/drivers/vfio/pci/vfio_pci_core.c
> >> @@ -180,7 +180,8 @@ static void vfio_pci_probe_mmaps(struct
> vfio_pci_core_device *vdev)
> >> struct vfio_pci_group_info;
> >> static void vfio_pci_dev_set_try_reset(struct vfio_device_set *dev_set);
> >> static int vfio_pci_dev_set_hot_reset(struct vfio_device_set *dev_set,
> >> - struct vfio_pci_group_info *groups);
> >> + struct vfio_pci_group_info *groups,
> >> + struct iommufd_ctx *iommufd_ctx);
> >>
> >> /*
> >> * INTx masking requires the ability to disable INTx signaling via PCI_COMMAND
> >> @@ -1277,7 +1278,7 @@ vfio_pci_ioctl_pci_hot_reset_groups(struct
> vfio_pci_core_device *vdev,
> >> return ret;
> >>
> >> /* Somewhere between 1 and count is OK */
> >> - if (!hdr->count || hdr->count > count)
> >> + if (hdr->count > count)
> >> return -EINVAL;
> >>
> >> group_fds = kcalloc(hdr->count, sizeof(*group_fds), GFP_KERNEL);
> >> @@ -1326,7 +1327,7 @@ vfio_pci_ioctl_pci_hot_reset_groups(struct
> vfio_pci_core_device *vdev,
> >> info.count = hdr->count;
> >> info.files = files;
> >>
> >> - ret = vfio_pci_dev_set_hot_reset(vdev->vdev.dev_set, &info);
> >> + ret = vfio_pci_dev_set_hot_reset(vdev->vdev.dev_set, &info, NULL);
> >>
> >> hot_reset_release:
> >> for (file_idx--; file_idx >= 0; file_idx--)
> >> @@ -1341,6 +1342,7 @@ static int vfio_pci_ioctl_pci_hot_reset(struct
> vfio_pci_core_device *vdev,
> >> {
> >> unsigned long minsz = offsetofend(struct vfio_pci_hot_reset, count);
> >> struct vfio_pci_hot_reset hdr;
> >> + struct iommufd_ctx *iommufd;
> >> bool slot = false;
> >>
> >> if (copy_from_user(&hdr, arg, minsz))
> >> @@ -1355,7 +1357,12 @@ static int vfio_pci_ioctl_pci_hot_reset(struct
> vfio_pci_core_device *vdev,
> >> else if (pci_probe_reset_bus(vdev->pdev->bus))
> >> return -ENODEV;
> >>
> >> - return vfio_pci_ioctl_pci_hot_reset_groups(vdev, &hdr, slot, arg);
> >> + if (hdr.count)
> >> + return vfio_pci_ioctl_pci_hot_reset_groups(vdev, &hdr, slot, arg);
> >> +
> >> + iommufd = vfio_iommufd_physical_ictx(&vdev->vdev);
> >> +
> >> + return vfio_pci_dev_set_hot_reset(vdev->vdev.dev_set, NULL, iommufd);
> >> }
> >>
> >> static int vfio_pci_ioctl_ioeventfd(struct vfio_pci_core_device *vdev,
> >> @@ -2327,6 +2334,9 @@ static bool vfio_dev_in_groups(struct
> vfio_pci_core_device *vdev,
> >> {
> >> unsigned int i;
> >>
> >> + if (!groups)
> >> + return false;
> >> +
> >> for (i = 0; i < groups->count; i++)
> >> if (vfio_file_has_dev(groups->files[i], &vdev->vdev))
> >> return true;
> >> @@ -2402,13 +2412,25 @@ static int vfio_pci_dev_set_pm_runtime_get(struct
> vfio_device_set *dev_set)
> >> return ret;
> >> }
> >>
> >> +static bool vfio_dev_in_iommufd_ctx(struct vfio_pci_core_device *vdev,
> >> + struct iommufd_ctx *iommufd_ctx)
> >> +{
> >> + struct iommufd_ctx *iommufd = vfio_iommufd_physical_ictx(&vdev->vdev);
> >> +
> >> + if (!iommufd)
> >> + return false;
> >> +
> >> + return iommufd == iommufd_ctx;
> >> +}
> >> +
> >> /*
> >> * We need to get memory_lock for each device, but devices can share mmap_lock,
> >> * therefore we need to zap and hold the vma_lock for each device, and only then
> >> * get each memory_lock.
> >> */
> >> static int vfio_pci_dev_set_hot_reset(struct vfio_device_set *dev_set,
> >> - struct vfio_pci_group_info *groups)
> >> + struct vfio_pci_group_info *groups,
> >> + struct iommufd_ctx *iommufd_ctx)
> >> {
> >> struct vfio_pci_core_device *cur_mem;
> >> struct vfio_pci_core_device *cur_vma;
> >> @@ -2448,9 +2470,17 @@ static int vfio_pci_dev_set_hot_reset(struct
> vfio_device_set *dev_set,
> >> *
> >> * Otherwise all opened devices in the dev_set must be
> >> * contained by the set of groups provided by the user.
> >> + *
> >> + * If user provides a zero-length array, then all the
> >> + * opened devices must be bound to a same iommufd_ctx.
> >> + *
> >> + * If all above checks are failed, reset is allowed only if
> >> + * the calling device is in a singleton dev_set.
> >> */
> >> if (cur_vma->vdev.open_count &&
> >> - !vfio_dev_in_groups(cur_vma, groups)) {
> >> + !vfio_dev_in_groups(cur_vma, groups) &&
> >> + !vfio_dev_in_iommufd_ctx(cur_vma, iommufd_ctx) &&
> >> + (dev_set->device_count > 1)) {
> > This last condition looks buggy to me, we need all conditions to be
> > true to generate an error here, which means that for a singleton
> > dev_set, it doesn't matter what group fds are passed, if any, or whether
> > the iommufd context matches. I think in fact this means that the empty
> > array path is equally available for group use cases with a singleton
> > dev_set, but we don't enable it for multiple device dev_sets like we do
> > iommufd.
> >
> > You pointed out a previous issue with hot-reset info and no-iommu where
> > if other affected devices are not bound to vfio-pci the info ioctl
> > returns error. That's handled in the hot-reset ioctl by the fact that
> > all affected devices must be in the dev_set and therefore bound to
> > vfio-pci drivers. So it seems to me that aside from the spurious error
> > because we can't report an iommu group when none exists, and didn't
> > spot it to invent an invalid group for debugging, hot-reset otherwise
> > works with no-iommu just like it does for iommu backed devices. We
> > don't currently require singleton no-iommu dev_sets afaict.
> >
> > I'll also note that if the dev_set is singleton, this suggests that
> > pci_reset_function() can make use of bus reset, so a hot-reset is
> > accessible via VFIO_DEVICE_RESET if the appropriate reset method is
> > selected.
> >
> > Therefore, I think as written, the singleton dev_set hot-reset is
> > enabled for iommufd and (unintentionally?) for the group path, while
> > also negating a requirement for a group fd or that a provided group fd
> > actually matches the device in this latter case. The null-array
> > approach is not however extended to groups for more general use.
> > Additionally, limiting no-iommu hot-reset to singleton dev_sets
> > provides only a marginal functional difference vs VFIO_DEVICE_RESET.
> > Thanks,
> >
> > Alex
> What bout introducing a helper
> static bool is_reset_ok(pdev, groups, ctx) {
> if (!pdev->vdev.open_count)
> return true;
> if (groups && vfio_dev_in_groups(pdev, groups))
> return true;
> if (ctx && vfio_dev_in_iommufd_ctx(pdev, ctx)
> return true;
> return false;
> }
>
> Assuming the above logic is correct I think this would make the code
> more readable
this logic may fail the noiommu devices in the cdev path as the
cdev path binds the devices to iommufd==-1. The ctx would be
NULL. So we agreed to allow the reset if the dev_set is sigletion.
Detail can be found in below paragraph. As I replied in another
email. Maybe this singleton support can be dropped since singleton
dev_set may just do reset with VFIO_DEVICE_RESET. Alex may
correct me if userspace is not so intelligent.
"However the iommufd method has difficulty working with noiommu devices
since those devices don't have a valid iommufd, unless the noiommu device
is in a singleton dev_set hence no ownership check is required. [3]
[3] https://lore.kernel.org/kvm/ZACX+Np%2FIY7ygqL5@nvidia.com/"
Regards,
Yi Liu
> Thanks
>
> Eric
> >> ret = -EINVAL;
> >> goto err_undo;
> >> }
> >> diff --git a/include/uapi/linux/vfio.h b/include/uapi/linux/vfio.h
> >> index f96e5689cffc..17aa5d09db41 100644
> >> --- a/include/uapi/linux/vfio.h
> >> +++ b/include/uapi/linux/vfio.h
> >> @@ -679,7 +679,14 @@ struct vfio_pci_hot_reset_info {
> >> * the calling user must ensure all affected devices, if opened, are
> >> * owned by itself.
> >> *
> >> - * The ownership is proved by an array of group fds.
> >> + * The ownership can be proved by:
> >> + * - An array of group fds
> >> + * - A zero-length array
> >> + *
> >> + * In the last case all affected devices which are opened by this user
> >> + * must have been bound to a same iommufd. If the calling device is in
> >> + * noiommu mode (no valid iommufd) then it can be reset only if the reset
> >> + * doesn't affect other devices.
> >> *
> >> * Return: 0 on success, -errno on failure.
> >> */
More information about the Intel-gfx
mailing list