[Linux1394-cvslog] rev 839 - branches/linux-2.4

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

Author: bencollins
Date: 2003-03-14 16:49:00 -0500 (Fri, 14 Mar 2003)
New Revision: 839

Modified:
   branches/linux-2.4/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: branches/linux-2.4/nodemgr.c
==============================================================================
--- branches/linux-2.4/nodemgr.c	(original)
+++ branches/linux-2.4/nodemgr.c	2003-03-14 16:49:01.000000000 -0500
@@ -1124,26 +1124,11 @@
 	return;
 }
 
-static void nodemgr_node_probe(struct hpsb_host *host)
+static void nodemgr_node_probe(struct hpsb_host *host, int generation)
 {
 	int count;
 	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++) {
@@ -1158,15 +1143,6 @@
 		nodemgr_node_probe_one(host, nodeid++, generation);
 	}
 
-	/* If we had a bus reset while we were scanning the bus, it is
-	 * possible that we did not probe all nodes.  In that case, we
-	 * skip the clean up for now, since we could remove nodes that
-	 * were still on the bus.  The bus reset increased
-	 * hi->reset_sem, so there's a bus scan pending which will do
-	 * the clean up eventually. */
-	if (generation == get_hpsb_generation(host))
-		nodemgr_node_probe_cleanup(host, generation);
-
 	return;
 }
 
@@ -1229,6 +1205,27 @@
 	 * 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(hi->host)) {
 			/* Do nothing, we are resetting */
@@ -1236,11 +1233,23 @@
 			continue;
 		}
 
-		nodemgr_node_probe(hi->host);
+		nodemgr_node_probe(hi->host, generation);
+
+		/* If we had a bus reset while we were scanning the bus, it is
+		 * possible that we did not probe all nodes.  In that case, we
+		 * skip the clean up for now, since we could remove nodes that
+		 * were still on the bus.  The bus reset increased
+		 * hi->reset_sem, so there's a bus scan pending which will do
+		 * the clean up eventually. */
+		if (generation == get_hpsb_generation(hi->host))
+			nodemgr_node_probe_cleanup(hi->host, generation);
+
 		nodemgr_do_irm_duties(hi->host);
 
 		up(&nodemgr_serialize);
 	}
+
+ caught_signal:
 #ifdef CONFIG_IEEE1394_VERBOSEDEBUG
 	HPSB_DEBUG ("NodeMgr: Exiting thread for %s", hi->host->driver->name);
 #endif