[DNM] Add log for packing

packing

cputree.c

@@ -224,11 +224,13 @@ static struct topo_obj* add_cache_domain_to_package(struct topo_obj *cache,
 		package->obj_type = OBJ_TYPE_PACKAGE;
 		package->obj_type_list = &packages;
 		package->number = packageid;
+		package->max_load = 0;
 		packages = g_list_append(packages, package);
 	}
 
 	entry = g_list_find(package->children, cache);
 	if (!entry) {
+		package->max_load += cache->max_load;
 		package->children = g_list_append(package->children, cache);
 		cache->parent = package;
 	}
@@ -264,12 +266,14 @@ static struct topo_obj* add_cpu_to_cache_domain(struct topo_obj *cpu,
 		cache->mask = cache_mask;
 		cache->number = cache_domain_count;
 		cache->obj_type_list = &cache_domains;
+		cache->max_load = 0;
 		cache_domains = g_list_append(cache_domains, cache);
 		cache_domain_count++;
 	}
 
 	entry = g_list_find(cache->children, cpu);
 	if (!entry) {
+		cache->max_load += cpu->max_load;
 		cache->children = g_list_append(cache->children, cpu);
 		cpu->parent = (struct topo_obj *)cache;
 	}
@@ -317,6 +321,8 @@ static void do_one_cpu(char *path)
 
 	cpu->number = strtoul(&path[27], NULL, 10);
 
+	cpu->max_load = SLEEP_INTERVAL * 1000 * 1000 * 1000; // SLEEP_INTERVAL in nanoseconds, should be a good enough approximation
+
 	cpu_set(cpu->number, cpu_online_map);
 	
 	cpu_set(cpu->number, cpu->mask);
@@ -398,6 +404,8 @@ static void do_one_cpu(char *path)
 		 * we override package_mask with node mask.
 		 */
 		node = get_numa_node(nodeid);
+		if (node)
+			node->max_load += cpu->max_load;
 		if (node && (cpus_weight(package_mask) > cpus_weight(node->mask)))
 			cpus_and(package_mask, package_mask, node->mask);
 	}
@@ -445,7 +453,7 @@ static void dump_balance_obj(struct topo_obj *d, void *data __attribute__((unuse
 	log(TO_CONSOLE, LOG_INFO, "%s%s%s%sCPU number %i  numa_node is ",
 	    log_indent, log_indent, log_indent, log_indent, c->number);
 	for_each_object(cpu_numa_node(c), dump_numa_node_num, NULL);
-	log(TO_CONSOLE, LOG_INFO, "(load %lu)\n", (unsigned long)c->load);
+	log(TO_CONSOLE, LOG_INFO, "(load %lu, max %lu)\n", (unsigned long)c->load, (unsigned long)c->max_load);
 	if (c->interrupts)
 		for_each_irq(c->interrupts, dump_irq, (void *)18);
 }
@@ -457,8 +465,8 @@ static void dump_cache_domain(struct topo_obj *d, void *data)
 	log(TO_CONSOLE, LOG_INFO, "%s%sCache domain %i:  numa_node is ",
 	    log_indent, log_indent, d->number);
 	for_each_object(d->numa_nodes, dump_numa_node_num, NULL);
-	log(TO_CONSOLE, LOG_INFO, "cpu mask is %s  (load %lu) \n", buffer,
-	    (unsigned long)d->load);
+	log(TO_CONSOLE, LOG_INFO, "cpu mask is %s  (load %lu, max %lu) \n", buffer,
+	    (unsigned long)d->load, (unsigned long)d->max_load);
 	if (d->children)
 		for_each_object(d->children, dump_balance_obj, NULL);
 	if (g_list_length(d->interrupts) > 0)
@@ -471,8 +479,8 @@ static void dump_package(struct topo_obj *d, void *data)
 	cpumask_scnprintf(buffer, 4096, d->mask);
 	log(TO_CONSOLE, LOG_INFO, "Package %i:  numa_node ", d->number);
 	for_each_object(d->numa_nodes, dump_numa_node_num, NULL);
-	log(TO_CONSOLE, LOG_INFO, "cpu mask is %s (load %lu)\n",
-	    buffer, (unsigned long)d->load);
+	log(TO_CONSOLE, LOG_INFO, "cpu mask is %s (load %lu, max %lu)\n",
+	    buffer, (unsigned long)d->load, (unsigned long)d->max_load);
 	if (d->children)
 		for_each_object(d->children, dump_cache_domain, buffer);
 	if (g_list_length(d->interrupts) > 0)

irqlist.c

@@ -79,7 +79,7 @@ static void move_candidate_irqs(struct irq_info *info, void *data)
 	unsigned long delta_load = 0;
 
 	/* Don't rebalance irqs that don't want it */
-	if (info->level == BALANCE_NONE)
+	if (info->level == BALANCE_NONE || info->level == BALANCE_CORE)
 		return;
 
 	/* Don't move cpus that only have one irq, regardless of load */
@@ -113,6 +113,28 @@ static void move_candidate_irqs(struct irq_info *info, void *data)
 	info->assigned_obj = NULL;
 }
 
+static void move_candidate_irqs_packing(GList *irqs, struct topo_obj *obj, struct load_balance_info *lb_info)
+{
+	GList *entry;
+	entry = g_list_last(irqs);
+	while (entry) {
+		struct irq_info *info = entry->data;
+
+		// Only pack BALANCE_CORE irqs
+		if (info->level != BALANCE_CORE)
+			continue;
+
+
+		if (lb_info->adjustment_load >= obj->max_load / 10 * 9) {
+			lb_info->adjustment_load -= info->load;
+			log(TO_CONSOLE, LOG_INFO, "Selecting irq %d for rebalancing via packing\n", info->irq);
+			migrate_irq(&info->assigned_obj->interrupts, &rebalance_irq_list, info);
+		}
+
+		entry = g_list_previous(entry);
+	}
+}
+
 static void migrate_overloaded_irqs(struct topo_obj *obj, void *data)
 {
 	struct load_balance_info *info = data;
@@ -143,6 +165,7 @@ static void migrate_overloaded_irqs(struct topo_obj *obj, void *data)
 		 */
 		info->adjustment_load = obj->load;
 		for_each_irq(obj->interrupts, move_candidate_irqs, info);
+		move_candidate_irqs_packing(obj->interrupts, obj, info);
 	}
 }
 

numa.c

@@ -58,9 +58,10 @@ static void add_one_node(int nodeid)
 		process_one_line(path, get_mask_from_bitmap, &new->mask);
 	}
 
-	new->obj_type = OBJ_TYPE_NODE;	
+	new->obj_type = OBJ_TYPE_NODE;
 	new->number = nodeid;
 	new->obj_type_list = &numa_nodes;
+	new->max_load = 0;
 	numa_nodes = g_list_append(numa_nodes, new);
 }
 

placement.c

@@ -76,7 +76,22 @@ static void find_best_object_packing(GList *objs, struct irq_info *info, struct
 	entry = g_list_first(objs);
 	while (entry) {
 		struct topo_obj *d = entry->data;
-		if (d->load + info->load < 0.9e9) {
+
+		/*
+		 * Don't consider the unspecified numa node here
+		 */
+		if (numa_avail && (d->obj_type == OBJ_TYPE_NODE) && (d->number == NUMA_NO_NODE))
+			continue;
+
+		/*
+		 * also don't consider any node that doesn't have at least one cpu in
+		 * the unbanned list
+		 */
+		if ((d->obj_type == OBJ_TYPE_NODE) &&
+			(!cpus_intersects(d->mask, unbanned_cpus)))
+			continue;
+
+		if (d->load + info->load < d->max_load / 10 * 9) {
 			place->best = d;
 			place->best_cost = d->load + info->load;
 			return;

types.h

@@ -46,6 +46,7 @@ enum obj_type_e {
 struct topo_obj {
 	uint64_t load;
 	uint64_t last_load;
+	uint64_t max_load;
 	uint64_t irq_count;
 	enum obj_type_e obj_type;
 	int number;