From: Andrew Theurer <habanero@us.ibm.com>
This patch makes nr_running o(1) for non NUMA and o(n) n=num_nodes for NUMA
systems instead of o(n) n=nr_cpus (in both cases).
We were already doing 90% of this by keeping track of nr_running per node,
so we simply do the same for a nonNUMA system (1 node) and we sum up the
node_nr_running[nodes] for nr_running(). An added bonus, we removed nasty
ifdefs from sched.c. Actually, I this patch removes more code than it
adds!
I tested the patch on a dual node x440 with kernel compiles. With NUMA
support enabled, I didn't see any performance difference, but when running
with out NUMA support I did see a consistent 2% reduction on kernel compile
times.
include/linux/sched.h | 5 ++---
kernel/sched.c | 18 +++---------------
2 files changed, 5 insertions(+), 18 deletions(-)
diff -puN kernel/sched.c~nr_running-speedup kernel/sched.c
--- 25/kernel/sched.c~nr_running-speedup 2003-06-23 23:02:19.000000000 -0700
+++ 25-akpm/kernel/sched.c 2003-06-23 23:02:19.000000000 -0700
@@ -162,10 +162,8 @@ struct runqueue {
struct mm_struct *prev_mm;
prio_array_t *active, *expired, arrays[2];
int prev_cpu_load[NR_CPUS];
-#ifdef CONFIG_NUMA
atomic_t *node_nr_running;
int prev_node_load[MAX_NUMNODES];
-#endif
task_t *migration_thread;
struct list_head migration_queue;
@@ -189,8 +187,6 @@ static struct runqueue runqueues[NR_CPUS
# define task_running(rq, p) ((rq)->curr == (p))
#endif
-#ifdef CONFIG_NUMA
-
/*
* Keep track of running tasks.
*/
@@ -215,7 +211,7 @@ static inline void nr_running_dec(runque
rq->nr_running--;
}
-__init void node_nr_running_init(void)
+void __init node_nr_running_init(void)
{
int i;
@@ -226,14 +222,6 @@ __init void node_nr_running_init(void)
}
}
-#else /* !CONFIG_NUMA */
-
-# define nr_running_init(rq) do { } while (0)
-# define nr_running_inc(rq) do { (rq)->nr_running++; } while (0)
-# define nr_running_dec(rq) do { (rq)->nr_running--; } while (0)
-
-#endif /* CONFIG_NUMA */
-
/*
* task_rq_lock - lock the runqueue a given task resides on and disable
* interrupts. Note the ordering: we can safely lookup the task_rq without
@@ -672,8 +660,8 @@ unsigned long nr_running(void)
{
unsigned long i, sum = 0;
- for (i = 0; i < NR_CPUS; i++)
- sum += cpu_rq(i)->nr_running;
+ for (i = 0; i < MAX_NUMNODES; i++)
+ sum += atomic_read(&node_nr_running[i]);
return sum;
}
diff -puN include/linux/sched.h~nr_running-speedup include/linux/sched.h
--- 25/include/linux/sched.h~nr_running-speedup 2003-06-23 23:02:19.000000000 -0700
+++ 25-akpm/include/linux/sched.h 2003-06-23 23:02:19.000000000 -0700
@@ -521,11 +521,10 @@ static inline int set_cpus_allowed(task_
#ifdef CONFIG_NUMA
extern void sched_balance_exec(void);
-extern void node_nr_running_init(void);
#else
-#define sched_balance_exec() {}
-#define node_nr_running_init() {}
+#define sched_balance_exec() do {} while (0)
#endif
+extern void node_nr_running_init(void);
extern void set_user_nice(task_t *p, long nice);
extern int task_prio(task_t *p);
_