ipmi: Fix a memory ordering issue

From a locking point of view it is safe to check waiting_msg without
a lock, but there is a memory ordering issue that causes it to
possibly not be set right when viewed from another processor.  We are
already claiming a lock right after that, move the check to inside
the lock to enforce the memory ordering.

Signed-off-by: Corey Minyard <cminyard@mvista.com>

diff --git a/drivers/char/ipmi/ipmi_si_intf.c b/drivers/char/ipmi/ipmi_si_intf.c
index 321ecb2..f6646ed 100644 (file)
--- a/drivers/char/ipmi/ipmi_si_intf.c
+++ b/drivers/char/ipmi/ipmi_si_intf.c
@@ -932,9 +932,6 @@ static void sender(void                *send_info,
        enum si_sm_result result;
        unsigned long     flags;
 
-       BUG_ON(smi_info->waiting_msg);
-       smi_info->waiting_msg = msg;
-
        debug_timestamp("Enqueue");
 
        if (smi_info->run_to_completion) {
@@ -942,7 +939,7 @@ static void sender(void                *send_info,
                 * If we are running to completion, start it and run
                 * transactions until everything is clear.
                 */
-               smi_info->curr_msg = smi_info->waiting_msg;
+               smi_info->curr_msg = msg;
                smi_info->waiting_msg = NULL;
 
                /*
@@ -960,6 +957,15 @@ static void sender(void                *send_info,
        }
 
        spin_lock_irqsave(&smi_info->si_lock, flags);
+       /*
+        * The following two lines don't need to be under the lock for
+        * the lock's sake, but they do need SMP memory barriers to
+        * avoid getting things out of order.  We are already claiming
+        * the lock, anyway, so just do it under the lock to avoid the
+        * ordering problem.
+        */
+       BUG_ON(smi_info->waiting_msg);
+       smi_info->waiting_msg = msg;
        check_start_timer_thread(smi_info);
        spin_unlock_irqrestore(&smi_info->si_lock, flags);
 }