[Linux1394-cvslog] rev 840 - trunk

[SVN U. <ben...@li...>]

Author: bencollins
Date: 2003-03-14 16:56:41 -0500 (Fri, 14 Mar 2003)
New Revision: 840

Modified:
   trunk/nodemgr.c
Log:
Handle back-to-back resets better, and fix what appears to be a race in
the generation check. Also handle forced IRM reset better.

Patch from Jim Radford.


Modified: trunk/nodemgr.c
==============================================================================
--- trunk/nodemgr.c	(original)
+++ trunk/nodemgr.c	2003-03-14 16:56:43.000000000 -0500
@@ -1455,27 +1455,12 @@
 }
 
 
-static void nodemgr_node_probe(struct host_info *hi)
+static void nodemgr_node_probe(struct host_info *hi, int generation)
 {
 	int count;
 	struct hpsb_host *host = hi->host;
 	struct selfid *sid = (struct selfid *)host->topology_map;
 	nodeid_t nodeid = LOCAL_BUS;
-	unsigned int generation;
-
-	/* Pause for 1/4 second, to make sure things settle down. If
-	 * schedule_timeout returns non-zero, it means we caught a signal
-	 * and need to return. */
-	set_current_state(TASK_INTERRUPTIBLE);
-	if (schedule_timeout (HZ/4))
-		return;
-
-	/* Now get the generation in which the node ID's we collect
-	 * are valid.  During the bus scan we will use this generation
-	 * for the read transactions, so that if another reset occurs
-	 * during the scan the transactions will fail instead of
-	 * returning bogus data. */
-	generation = get_hpsb_generation(host);
 
 	/* Scan each node on the bus */
 	for (count = host->selfid_count; count; count--, sid++) {
@@ -1588,6 +1573,29 @@
 	 * happens when we get a bus reset. */
 	while (!down_interruptible(&hi->reset_sem) &&
 	       !down_interruptible(&nodemgr_serialize)) {
+		unsigned int generation;
+
+		/* Many resets can happen while we are waiting below.
+		 * Make sure we respond only to the most recent one. */
+		do {
+			/* Pause for 1/4 second, to make sure things
+			 * settle down. If schedule_timeout returns
+			 * non-zero, it means we caught a signal and that
+			 * we should exit. */
+			set_current_state(TASK_INTERRUPTIBLE);
+			if (schedule_timeout (HZ/4)) {
+				up(&nodemgr_serialize);
+				goto caught_signal;
+			}
+
+			/* Now get the generation in which the node ID's
+			 * we collect are valid.  During the bus scan we
+			 * will use this generation for the read
+			 * transactions, so that if another reset occurs
+			 * during the scan the transactions will fail
+			 * instead of returning bogus data. */
+			generation = get_hpsb_generation(hi->host);
+		} while (!down_trylock(&hi->reset_sem));
 
 		if (!nodemgr_check_root_capability(host)) {
 			/* Do nothing, we are resetting */
@@ -1595,7 +1603,7 @@
 			continue;
 		}
 
-		nodemgr_node_probe(hi);
+		nodemgr_node_probe(hi, generation);
 		nodemgr_do_irm_duties(host);
 
 		up(&nodemgr_serialize);