IB/core: Add might_sleep() annotation to ib_init_ah_from_wc()
authorRoland Dreier <roland@purestorage.com>
Tue, 29 Aug 2017 17:34:44 +0000 (10:34 -0700)
committerDoug Ledford <dledford@redhat.com>
Thu, 31 Aug 2017 12:35:07 +0000 (08:35 -0400)
For RoCE, ib_init_ah_from_wc() can follow the path

    ib_init_ah_from_wc() ->
      rdma_addr_find_l2_eth_by_grh() ->
        rdma_resolve_ip()

and rdma_resolve_ip() will sleep in kzalloc() and wait_for_completion().

However, developers will not see any warnings if they use ib_init_ah_from_wc()
in an atomic context and test only on IB, because the function doesn't
sleep in that case.

Add a might_sleep() so that lockdep will catch bugs no matter what hardware is
used to test.

Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
drivers/infiniband/core/verbs.c

index ecb6c395f19b95168790ecdd76f8bebd265fddab..ee9e27dc799b50ead4663cd5fd13c94a3ac020d4 100644 (file)
@@ -468,6 +468,8 @@ int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
        union ib_gid dgid;
        union ib_gid sgid;
 
+       might_sleep();
+
        memset(ah_attr, 0, sizeof *ah_attr);
        ah_attr->type = rdma_ah_find_type(device, port_num);
        if (rdma_cap_eth_ah(device, port_num)) {