IB/ipath: Better comment for rmb() in ipath_intr()

An internal code review found the comment here lacking -- update it with
more specifics of how and why the rmb() is there.

Signed-off-by: Arthur Jones <arthur.jones@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

diff --git a/drivers/infiniband/hw/ipath/ipath_intr.c b/drivers/infiniband/hw/ipath/ipath_intr.c
index eac2e9c55281d9b8c18390a7f710dcb236b6de7d..4795cb895f8577100fb8126a2aaa5bc8fd59769e 100644 (file)
--- a/drivers/infiniband/hw/ipath/ipath_intr.c
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c
@@ -954,7 +954,15 @@ static void handle_urcv(struct ipath_devdata *dd, u32 istat)
        int i;
        int rcvdint = 0;
 
-       /* test_bit below needs this... */
+       /*
+        * test_and_clear_bit(IPATH_PORT_WAITING_RCV) and
+        * test_and_clear_bit(IPATH_PORT_WAITING_URG) below
+        * would both like timely updates of the bits so that
+        * we don't pass them by unnecessarily.  the rmb()
+        * here ensures that we see them promptly -- the
+        * corresponding wmb()'s are in ipath_poll_urgent()
+        * and ipath_poll_next()...
+        */
        rmb();
        portr = ((istat >> INFINIPATH_I_RCVAVAIL_SHIFT) &
                 dd->ipath_i_rcvavail_mask)