Pull VFIO updates from Alex Williamson:
- Remove redundant resource check in vfio-platform (Angus Chen)
- Use GFP_KERNEL_ACCOUNT for persistent userspace allocations, allowing
removal of arbitrary kernel limits in favor of cgroup control (Yishai
Hadas)
- mdev tidy-ups, including removing the module-only build restriction
for sample drivers, Kconfig changes to select mdev support,
documentation movement to keep sample driver usage instructions with
sample drivers rather than with API docs, remove references to
out-of-tree drivers in docs (Christoph Hellwig)
- Fix collateral breakages from mdev Kconfig changes (Arnd Bergmann)
- Make mlx5 migration support match device support, improve source and
target flows to improve pre-copy support and reduce downtime (Yishai
Hadas)
- Convert additional mdev sysfs case to use sysfs_emit() (Bo Liu)
- Resolve copy-paste error in mdev mbochs sample driver Kconfig (Ye
Xingchen)
- Avoid propagating missing reset error in vfio-platform if reset
requirement is relaxed by module option (Tomasz Duszynski)
- Range size fixes in mlx5 variant driver for missed last byte and
stricter range calculation (Yishai Hadas)
- Fixes to suspended vaddr support and locked_vm accounting, excluding
mdev configurations from the former due to potential to indefinitely
block kernel threads, fix underflow and restore locked_vm on new mm
(Steve Sistare)
- Update outdated vfio documentation due to new IOMMUFD interfaces in
recent kernels (Yi Liu)
- Resolve deadlock between group_lock and kvm_lock, finally (Matthew
Rosato)
- Fix NULL pointer in group initialization error path with IOMMUFD (Yan
Zhao)
* tag 'vfio-v6.3-rc1' of https://github.com/awilliam/linux-vfio: (32 commits)
vfio: Fix NULL pointer dereference caused by uninitialized group->iommufd
docs: vfio: Update vfio.rst per latest interfaces
vfio: Update the kdoc for vfio_device_ops
vfio/mlx5: Fix range size calculation upon tracker creation
vfio: no need to pass kvm pointer during device open
vfio: fix deadlock between group lock and kvm lock
vfio: revert "iommu driver notify callback"
vfio/type1: revert "implement notify callback"
vfio/type1: revert "block on invalid vaddr"
vfio/type1: restore locked_vm
vfio/type1: track locked_vm per dma
vfio/type1: prevent underflow of locked_vm via exec()
vfio/type1: exclude mdevs from VFIO_UPDATE_VADDR
vfio: platform: ignore missing reset if disabled at module init
vfio/mlx5: Improve the target side flow to reduce downtime
vfio/mlx5: Improve the source side flow upon pre_copy
vfio/mlx5: Check whether VF is migratable
samples: fix the prompt about SAMPLE_VFIO_MDEV_MBOCHS
vfio/mdev: Use sysfs_emit() to instead of sprintf()
vfio-mdev: add back CONFIG_VFIO dependency
...
iommufd = iommufd_ctx_from_file(f.file);
if (!IS_ERR(iommufd)) {
- u32 ioas_id;
+ if (IS_ENABLED(CONFIG_VFIO_NOIOMMU) &&
+ group->type == VFIO_NO_IOMMU)
+ ret = iommufd_vfio_compat_set_no_iommu(iommufd);
+ else
+ ret = iommufd_vfio_compat_ioas_create(iommufd);
- ret = iommufd_vfio_compat_ioas_id(iommufd, &ioas_id);
if (ret) {
- iommufd_ctx_put(group->iommufd);
+ iommufd_ctx_put(iommufd);
goto out_unlock;
}
int __init vfio_group_init(void);
void vfio_group_cleanup(void);
+static inline bool vfio_device_is_noiommu(struct vfio_device *vdev)
+{
+ return IS_ENABLED(CONFIG_VFIO_NOIOMMU) &&
+ vdev->group->type == VFIO_NO_IOMMU;
+}
+
#if IS_ENABLED(CONFIG_VFIO_CONTAINER)
- /* events for the backend driver notify callback */
- enum vfio_iommu_notify_type {
- VFIO_IOMMU_CONTAINER_CLOSE = 0,
- };
-
/**
* struct vfio_iommu_driver_ops - VFIO IOMMU driver callbacks
*/