[PATCH v1 18/19] sched: update shares distribution to use u64

[Nikhil Rao]

Update the shares distribution code to use u64. We still maintain tg->shares as
an unsigned long since sched entity weights can't exceed MAX_SHARES (2^28). This
patch updates all the calculations required to estimate shares to use u64.

Signed-off-by: Nikhil Rao <ncrao@xxxxxxxxxx>
---
 kernel/sched.c       |    2 +-
 kernel/sched_debug.c |    6 +++---
 kernel/sched_fair.c  |   17 ++++++++++++-----
 3 files changed, 16 insertions(+), 9 deletions(-)

diff --git a/kernel/sched.c b/kernel/sched.c
index 6b9b02a..e131225 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -367,7 +367,7 @@ struct cfs_rq {
 	u64 load_period;
 	u64 load_stamp, load_last, load_unacc_exec_time;
 
-	unsigned long load_contribution;
+	u64 load_contribution;
 #endif
 #endif
 };
diff --git a/kernel/sched_debug.c b/kernel/sched_debug.c
index d22b666..b809651 100644
--- a/kernel/sched_debug.c
+++ b/kernel/sched_debug.c
@@ -204,11 +204,11 @@ void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 	SEQ_printf(m, "  .%-30s: %lld\n", "load", cfs_rq->load.weight);
 #ifdef CONFIG_FAIR_GROUP_SCHED
 #ifdef CONFIG_SMP
-	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "load_avg",
+	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", "load_avg",
 			SPLIT_NS(cfs_rq->load_avg));
-	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "load_period",
+	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", "load_period",
 			SPLIT_NS(cfs_rq->load_period));
-	SEQ_printf(m, "  .%-30s: %ld\n", "load_contrib",
+	SEQ_printf(m, "  .%-30s: %lld\n", "load_contrib",
 			cfs_rq->load_contribution);
 	SEQ_printf(m, "  .%-30s: %d\n", "load_tg",
 			atomic_read(&cfs_rq->tg->load_weight));
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index bf9bbaa..3f56410 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -708,12 +708,13 @@ static void update_cfs_rq_load_contribution(struct cfs_rq *cfs_rq,
 					    int global_update)
 {
 	struct task_group *tg = cfs_rq->tg;
-	long load_avg;
+	s64 load_avg;
 
 	load_avg = div64_u64(cfs_rq->load_avg, cfs_rq->load_period+1);
 	load_avg -= cfs_rq->load_contribution;
 
 	if (global_update || abs(load_avg) > cfs_rq->load_contribution / 8) {
+		/* TODO: fix atomics for 64-bit additions */
 		atomic_add(load_avg, &tg->load_weight);
 		cfs_rq->load_contribution += load_avg;
 	}
@@ -723,7 +724,7 @@ static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update)
 {
 	u64 period = sysctl_sched_shares_window;
 	u64 now, delta;
-	unsigned long load = cfs_rq->load.weight;
+	u64 load = cfs_rq->load.weight;
 
 	if (cfs_rq->tg == &root_task_group)
 		return;
@@ -743,8 +744,13 @@ static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update)
 	cfs_rq->load_unacc_exec_time = 0;
 	cfs_rq->load_period += delta;
 	if (load) {
+		u64 tmp = cfs_rq->load_avg;
 		cfs_rq->load_last = now;
 		cfs_rq->load_avg += delta * load;
+
+		/* Detect overflow and set load_avg to max */
+		if (unlikely(cfs_rq->load_avg < tmp))
+			cfs_rq->load_avg = ~0ULL;
 	}
 
 	/* consider updating load contribution on each fold or truncate */
@@ -769,24 +775,25 @@ static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update)
 
 static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
 {
-	long load_weight, load, shares;
+	s64 load_weight, load, shares;
 
 	load = cfs_rq->load.weight;
 
+	/* TODO: fixup atomics to handle u64 in 32-bit */
 	load_weight = atomic_read(&tg->load_weight);
 	load_weight += load;
 	load_weight -= cfs_rq->load_contribution;
 
 	shares = (tg->shares * load);
 	if (load_weight)
-		shares /= load_weight;
+		shares = div64_u64(shares, load_weight);
 
 	if (shares < MIN_SHARES)
 		shares = MIN_SHARES;
 	if (shares > tg->shares)
 		shares = tg->shares;
 
-	return shares;
+	return (long)shares;
 }
 
 static void update_entity_shares_tick(struct cfs_rq *cfs_rq)
-- 
1.7.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/