staging/rdma/hfi1: Fix downgrade race

A link downgrade can race with link up. Avoid the race
in two ways. First, by having the downgrade application logic
take the link state mutex for all of its checking. Second, by
waiting for the link to move out of the going up state.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
index dc69159..017d439 100644 (file)
--- a/drivers/staging/rdma/hfi1/chip.c
+++ b/drivers/staging/rdma/hfi1/chip.c
@@ -3907,18 +3907,32 @@ void handle_verify_cap(struct work_struct *work)
  */
 void apply_link_downgrade_policy(struct hfi1_pportdata *ppd, int refresh_widths)
 {
-       int skip = 1;
        int do_bounce = 0;
-       u16 lwde = ppd->link_width_downgrade_enabled;
+       int tries;
+       u16 lwde;
        u16 tx, rx;
 
+       /* use the hls lock to avoid a race with actual link up */
+       tries = 0;
+retry:
        mutex_lock(&ppd->hls_lock);
        /* only apply if the link is up */
-       if (ppd->host_link_state & HLS_UP)
-               skip = 0;
-       mutex_unlock(&ppd->hls_lock);
-       if (skip)
-               return;
+       if (!(ppd->host_link_state & HLS_UP)) {
+               /* still going up..wait and retry */
+               if (ppd->host_link_state & HLS_GOING_UP) {
+                       if (++tries < 1000) {
+                               mutex_unlock(&ppd->hls_lock);
+                               usleep_range(100, 120); /* arbitrary */
+                               goto retry;
+                       }
+                       dd_dev_err(ppd->dd,
+                                  "%s: giving up waiting for link state change\n",
+                                  __func__);
+               }
+               goto done;
+       }
+
+       lwde = ppd->link_width_downgrade_enabled;
 
        if (refresh_widths) {
                get_link_widths(ppd->dd, &tx, &rx);
@@ -3956,6 +3970,9 @@ void apply_link_downgrade_policy(struct hfi1_pportdata *ppd, int refresh_widths)
                do_bounce = 1;
        }
 
+done:
+       mutex_unlock(&ppd->hls_lock);
+
        if (do_bounce) {
                set_link_down_reason(ppd, OPA_LINKDOWN_REASON_WIDTH_POLICY, 0,
                  OPA_LINKDOWN_REASON_WIDTH_POLICY);