[PATCH] RDMA/hfi1: shut down the trap timer before unregistering

[Hongyan Xu]

MAD processing arms a trap timer embedded in each hfi1_ibport. Device
unregistration does not stop the timer, so its callback can run after
the port and its containing hfi1_devdata have been released.

Shut down every port timer before unregistering the IB device. The
shutdown form also prevents a concurrent path from rearming the timer.

Fixes: bf90aadd630c ("IB/hfi1: Send MAD traps until repressed")
Signed-off-by: Hongyan Xu <getshell@xxxxxxxxxx>
---
 drivers/infiniband/hw/hfi1/verbs.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/infiniband/hw/hfi1/verbs.c b/drivers/infiniband/hw/hfi1/verbs.c
index e569b647d611..0ef3b1a89a53 100644
--- a/drivers/infiniband/hw/hfi1/verbs.c
+++ b/drivers/infiniband/hw/hfi1/verbs.c
@@ -1888,6 +1888,10 @@ err_verbs_txreq:
 void hfi1_unregister_ib_device(struct hfi1_devdata *dd)
 {
 	struct hfi1_ibdev *dev = &dd->verbs_dev;
+	int i;
+
+	for (i = 0; i < dd->num_pports; i++)
+		timer_shutdown_sync(&dd->pport[i].ibport_data.rvp.trap_timer);
 
 	hfi1_verbs_unregister_sysfs(dd);
 
-- 
2.50.1.windows.1