Message ID | 58b67ca78925186b246ce8ed3639a1c92e56f98a.1678284812.git.nicolinc@nvidia.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | Add IO page table replacement support | expand |
> From: Nicolin Chen <nicolinc@nvidia.com> > Sent: Wednesday, March 8, 2023 10:26 PM > > Support an access->ioas replacement in iommufd_access_set_ioas(), which > sets the access->ioas to NULL provisionally so that any further incoming > iommufd_access_pin_pages() callback can be blocked. > > Then, call access->ops->unmap() to clean up the entire iopt. To allow an > iommufd_access_unpin_pages() callback to happen via this unmap() call, > add an ioas_unpin pointer so the unpin routine won't be affected by the > "access->ioas = NULL" trick above. > > Suggested-by: Jason Gunthorpe <jgg@nvidia.com> > Signed-off-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Kevin Tian <kevin.tian@intel.com>
diff --git a/drivers/iommu/iommufd/device.c b/drivers/iommu/iommufd/device.c index 612a5f424bd7..c10e02f6a0be 100644 --- a/drivers/iommu/iommufd/device.c +++ b/drivers/iommu/iommufd/device.c @@ -727,11 +727,24 @@ int iommufd_access_set_ioas(struct iommufd_access *access, u32 ioas_id) iommufd_ref_to_users(obj); } + /* + * Set ioas to NULL to block any further iommufd_access_pin_pages(). + * iommufd_access_unpin_pages() can continue using access->ioas_unpin. + */ + access->ioas = NULL; + + if (access->ops->unmap) { + mutex_unlock(&access->ioas_lock); + access->ops->unmap(access->data, 0, ULONG_MAX); + mutex_lock(&access->ioas_lock); + } + if (cur_ioas) { iopt_remove_access(&cur_ioas->iopt, access); refcount_dec(&cur_ioas->obj.users); } + access->ioas_unpin = new_ioas; access->ioas = new_ioas; mutex_unlock(&access->ioas_lock); @@ -805,11 +818,11 @@ void iommufd_access_unpin_pages(struct iommufd_access *access, return; mutex_lock(&access->ioas_lock); - if (!access->ioas) { + if (!access->ioas_unpin) { mutex_unlock(&access->ioas_lock); return; } - iopt = &access->ioas->iopt; + iopt = &access->ioas_unpin->iopt; down_read(&iopt->iova_rwsem); iopt_for_each_contig_area(&iter, area, iopt, iova, last_iova) diff --git a/drivers/iommu/iommufd/iommufd_private.h b/drivers/iommu/iommufd/iommufd_private.h index db1ead8ff9d0..b18f843ad6a4 100644 --- a/drivers/iommu/iommufd/iommufd_private.h +++ b/drivers/iommu/iommufd/iommufd_private.h @@ -312,6 +312,7 @@ struct iommufd_access { struct iommufd_object obj; struct iommufd_ctx *ictx; struct iommufd_ioas *ioas; + struct iommufd_ioas *ioas_unpin; struct mutex ioas_lock; const struct iommufd_access_ops *ops; void *data;
Support an access->ioas replacement in iommufd_access_set_ioas(), which sets the access->ioas to NULL provisionally so that any further incoming iommufd_access_pin_pages() callback can be blocked. Then, call access->ops->unmap() to clean up the entire iopt. To allow an iommufd_access_unpin_pages() callback to happen via this unmap() call, add an ioas_unpin pointer so the unpin routine won't be affected by the "access->ioas = NULL" trick above. Suggested-by: Jason Gunthorpe <jgg@nvidia.com> Signed-off-by: Nicolin Chen <nicolinc@nvidia.com> --- drivers/iommu/iommufd/device.c | 17 +++++++++++++++-- drivers/iommu/iommufd/iommufd_private.h | 1 + 2 files changed, 16 insertions(+), 2 deletions(-)