Merge pull request #136 from yeyunfeng-dev/master-dev

irqbalance: prevent NULL pointer dereference in do_one_cpu()

cputree.c

@@ -176,8 +176,10 @@ static struct topo_obj* add_cache_domain_to_package(struct topo_obj *cache,
 
 	if (!entry) {
 		package = calloc(1, sizeof(struct topo_obj));
-		if (!package)
+		if (!package) {
+			need_rebuild = 1;
 			return NULL;
+		}
 		package->mask = package_mask;
 		package->obj_type = OBJ_TYPE_PACKAGE;
 		package->obj_type_list = &packages;
@@ -214,8 +216,10 @@ static struct topo_obj* add_cpu_to_cache_domain(struct topo_obj *cpu,
 
 	if (!entry) {
 		cache = calloc(1, sizeof(struct topo_obj));
-		if (!cache)
+		if (!cache) {
+			need_rebuild = 1;
 			return NULL;
+		}
 		cache->obj_type = OBJ_TYPE_CACHE;
 		cache->mask = cache_mask;
 		cache->number = cache_domain_count;
@@ -271,8 +275,10 @@ static void do_one_cpu(char *path)
 		return;
 
 	cpu = calloc(1, sizeof(struct topo_obj));
-	if (!cpu)
+	if (!cpu) {
+		need_rebuild = 1;
 		return;
+	}
 
 	cpu->obj_type = OBJ_TYPE_CPU;
 
@@ -372,8 +378,8 @@ static void do_one_cpu(char *path)
 	cpus_and(package_mask, package_mask, unbanned_cpus);
 
 	cache = add_cpu_to_cache_domain(cpu, cache_mask, nodeid);
-	add_cache_domain_to_package(cache, packageid, package_mask,
-	    nodeid);
+	if (cache)
+		add_cache_domain_to_package(cache, packageid, package_mask, nodeid);
 
 	cpu->obj_type_list = &cpus;
 	cpus = g_list_append(cpus, cpu);

irqbalance.c

@@ -52,6 +52,7 @@ int foreground_mode;
 int numa_avail;
 int journal_logging = 0;
 int need_rescan;
+int need_rebuild;
 unsigned int log_mask = TO_ALL;
 const char *log_indent;
 unsigned long power_thresh = ULONG_MAX;
@@ -260,14 +261,25 @@ gboolean scan(gpointer data __attribute__((unused)))
 
 
 	/* cope with cpu hotplug -- detected during /proc/interrupts parsing */
-	if (need_rescan) {
+	if (need_rescan || need_rebuild) {
+		int try_times = 0;
+
 		need_rescan = 0;
 		cycle_count = 0;
 		log(TO_CONSOLE, LOG_INFO, "Rescanning cpu topology \n");
 		clear_work_stats();
 
-		free_object_tree();
-		build_object_tree();
+		do {
+			free_object_tree();
+			if (++try_times > 3) {
+				log(TO_CONSOLE, LOG_WARNING, "Rescanning cpu topology: fail\n");
+				goto out;
+			}
+
+			need_rebuild = 0;
+			build_object_tree();
+		} while (need_rebuild);
+
 		for_each_irq(NULL, force_rebalance_irq, NULL);
 		parse_proc_interrupts();
 		parse_proc_stat();
@@ -284,6 +296,7 @@ gboolean scan(gpointer data __attribute__((unused)))
 	calculate_placement();
 	activate_mappings();
 
+out:
 	if (debug_mode)
 		dump_tree();
 	if (one_shot_mode)

irqbalance.h

@@ -67,6 +67,7 @@ extern int debug_mode;
 extern int journal_logging;
 extern int one_shot_mode;
 extern int need_rescan;
+extern int need_rebuild;
 extern unsigned long long cycle_count;
 extern unsigned long power_thresh;
 extern unsigned long deepest_cache;

numa.c

@@ -58,8 +58,10 @@ static void add_one_node(const char *nodename)
 	struct topo_obj *new;
 
 	new = calloc(1, sizeof(struct topo_obj));
-	if (!new)
+	if (!new) {
+		need_rebuild = 1;
 		return;
+	}
 
 	cpus_clear(new->mask);
 	sprintf(path, "%s/%s/cpumap", SYSFS_NODE_PATH, nodename);