Fix up load estimator

Found a few bugs in the load estimator - we're were attributing load to multiple
irqs errneously, and in the process of fixing that I found that we had a bogus
topology map - the same package was getting added multiple times to a given numa
node, since we didn't already detect that was part of that nodes child list.

cputree.c

@@ -308,7 +308,6 @@ static void clear_irq_stats(struct irq_info *info, void *data __attribute__((unu
 
 static void clear_obj_stats(struct topo_obj *d, void *data __attribute__((unused)))
 {
-	d->load = 0;
 	for_each_object(d->children, clear_obj_stats, NULL);
 	for_each_irq(d->interrupts, clear_irq_stats, NULL);
 }
@@ -318,7 +317,7 @@ static void clear_obj_stats(struct topo_obj *d, void *data __attribute__((unused
  * which level does how much work and the actual lists of interrupts 
  * assigned to each component
  */
-void clear_work_stats(void)
+void clear_work_stats()
 {
 	for_each_object(numa_nodes, clear_obj_stats, NULL);
 }

irqbalance.c

@@ -220,12 +220,16 @@ int main(int argc, char** argv)
 
 	for_each_irq(NULL, force_rebalance_irq, NULL);
 
+	parse_proc_interrupts();
+	parse_proc_stat();
+
 	while (1) {
 		sleep_approx(SLEEP_INTERVAL);
 		if (debug_mode)
 			printf("\n\n\n-----------------------------------------------------------------------------\n");
 
 
+		clear_work_stats();
 		parse_proc_interrupts();
 		parse_proc_stat();
 
@@ -242,6 +246,12 @@ int main(int argc, char** argv)
 			free_object_tree();
 			build_object_tree();
 			for_each_irq(NULL, force_rebalance_irq, NULL);
+			parse_proc_interrupts();
+			parse_proc_stat();
+			sleep_approx(SLEEP_INTERVAL);
+			clear_work_stats();
+			parse_proc_interrupts();
+			parse_proc_stat();
 			cycle_count=0;
 		} 
 
@@ -255,7 +265,6 @@ int main(int argc, char** argv)
 			dump_tree();
 		if (one_shot_mode)
 			break;
-		clear_work_stats();
 		cycle_count++;
 
 	}

numa.c

@@ -130,8 +130,10 @@ void add_package_to_node(struct topo_obj *p, int nodeid)
 
 	node = entry->data;
 
-	node->children = g_list_append(node->children, p);
-	p->parent = node;
+	if (!p->parent) {
+		node->children = g_list_append(node->children, p);
+		p->parent = node;
+	}
 }
 
 void dump_numa_node_info(struct topo_obj *d, void *unused __attribute__((unused)))

procinterrupts.c

@@ -154,24 +154,19 @@ static uint64_t get_parent_branch_irq_count_share(struct topo_obj *d)
 
 static void compute_irq_branch_load_share(struct topo_obj *d, void *data __attribute__((unused)))
 {
-	uint64_t total_irq_counts = 0;
 	uint64_t local_irq_counts = 0;
 
 	uint64_t load_slice;
 
-	total_irq_counts = get_parent_branch_irq_count_share(d);
-
-	load_slice = local_irq_counts ? (d->load / local_irq_counts) : 1;
 
 	if (g_list_length(d->interrupts) > 0) {
-		for_each_irq(d->interrupts, accumulate_irq_count, &local_irq_counts);
+		local_irq_counts = get_parent_branch_irq_count_share(d);
+		load_slice = local_irq_counts ? (d->load / local_irq_counts) : 1;
 		for_each_irq(d->interrupts, assign_load_slice, &load_slice);
 	}
 
-	if (d->parent) {
-		load_slice = total_irq_counts ? (d->load / total_irq_counts) : 1;
-		d->parent->load += (total_irq_counts - local_irq_counts) * load_slice;
-	}
+	if (d->parent)
+		d->parent->load += d->load;
 }
 
 void parse_proc_stat()
@@ -222,7 +217,9 @@ void parse_proc_stat()
  		 * For each cpu add the irq and softirq load and propagate that
  		 * all the way up the device tree
  		 */
-		cpu->load = irq_load + softirq_load;
+		if (cycle_count)
+			cpu->load = (irq_load + softirq_load) - (cpu->last_load);
+		cpu->last_load = (irq_load + softirq_load);
 	}
 
 	fclose(file);

types.h

@@ -35,6 +35,7 @@ enum obj_type_e {
 
 struct topo_obj {
 	uint64_t load;
+	uint64_t last_load;
 	enum obj_type_e obj_type;
 	int number;
 	int powersave_mode;