diff options
| author |   Jason Gunthorpe <jgg@mellanox.com> | 2019-10-09 13:09:35 -0300 |
|---|---|---|
| committer | Jason Gunthorpe <jgg@mellanox.com> | 2019-10-28 16:41:14 -0300 |
| commit | 46870b2391d5163e84180e051fdabadd502d8b44 (patch) | |
| tree | 422ea77709dc782164d7c09a9cc3a0593a7a4fc9 /drivers/infiniband/core | |
| parent | RDMA/mlx5: Do not race with mlx5_ib_invalidate_range during create and destroy (diff) | |
| download | linux-dev-46870b2391d5163e84180e051fdabadd502d8b44.tar.xz linux-dev-46870b2391d5163e84180e051fdabadd502d8b44.zip | |
RDMA/odp: Remove broken debugging call to invalidate_range
invalidate_range() also obtains the umem_mutex which is being held at this
point, so if this path were was ever called it would deadlock. Thus
conclude the debugging never triggers and rework it into a simple WARN_ON
and leave things as they are.
While here add a note to explain how we could possibly get inconsistent
page pointers.
Link: https://lore.kernel.org/r/20191009160934.3143-16-jgg@ziepe.ca
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Diffstat (limited to 'drivers/infiniband/core')
| -rw-r--r-- | drivers/infiniband/core/umem_odp.c | 38 |
1 files changed, 19 insertions, 19 deletions
diff --git a/drivers/infiniband/core/umem_odp.c b/drivers/infiniband/core/umem_odp.c index 163ff7ba92b7..d7d5fadf0899 100644 --- a/drivers/infiniband/core/umem_odp.c +++ b/drivers/infiniband/core/umem_odp.c @@ -508,7 +508,6 @@ static int ib_umem_odp_map_dma_single_page( { struct ib_device *dev = umem_odp->umem.ibdev; dma_addr_t dma_addr; - int remove_existing_mapping = 0; int ret = 0; /* @@ -534,28 +533,29 @@ static int ib_umem_odp_map_dma_single_page( } else if (umem_odp->page_list[page_index] == page) { umem_odp->dma_list[page_index] |= access_mask; } else { - pr_err("error: got different pages in IB device and from get_user_pages. IB device page: %p, gup page: %p\n", - umem_odp->page_list[page_index], page); - /* Better remove the mapping now, to prevent any further - * damage. */ - remove_existing_mapping = 1; + /* + * This is a race here where we could have done: + * + * CPU0 CPU1 + * get_user_pages() + * invalidate() + * page_fault() + * mutex_lock(umem_mutex) + * page from GUP != page in ODP + * + * It should be prevented by the retry test above as reading + * the seq number should be reliable under the + * umem_mutex. Thus something is really not working right if + * things get here. + */ + WARN(true, + "Got different pages in IB device and from get_user_pages. IB device page: %p, gup page: %p\n", + umem_odp->page_list[page_index], page); + ret = -EAGAIN; } out: put_user_page(page); - - if (remove_existing_mapping) { - ib_umem_notifier_start_account(umem_odp); - dev->ops.invalidate_range( - umem_odp, - ib_umem_start(umem_odp) + - (page_index << umem_odp->page_shift), - ib_umem_start(umem_odp) + - ((page_index + 1) << umem_odp->page_shift)); - ib_umem_notifier_end_account(umem_odp); - ret = -EAGAIN; - } - return ret; } |