On AMD Family17h-based (EPYC) system, a NUMA node can contain
upto 8 cores (16 threads) with the following topology.
----------------------------
C0 | T0 T1 | || | T0 T1 | C4
--------| || |--------
C1 | T0 T1 | L3 || L3 | T0 T1 | C5
--------| || |--------
C2 | T0 T1 | #0 || #1 | T0 T1 | C6
--------| || |--------
C3 | T0 T1 | || | T0 T1 | C7
----------------------------
Here, there are 2 last-level (L3) caches per NUMA node. A socket can
contain upto 4 NUMA nodes, and a system can support upto 2 sockets.
With full system configuration, current scheduler creates 4 sched
domains:
domain0 SMT (span a core)
domain1 MC (span a last-level-cache)
domain2 NUMA (span a socket: 4 nodes)
domain3 NUMA (span a system: 8 nodes)
Note that there is no domain to represent cpus spaning a NUMA node.
With this hierarchy of sched domains, the scheduler does not balance
properly in the following cases:
Case1:
When running 8 tasks, a properly balanced system should
schedule a task per NUMA node. This is not the case for
the current scheduler.
Case2:
Sometimes, threads are scheduled on the same cpu, while other
cpus are idle. This results in run-to-run inconsistency. For example:
taskset -c 0-7 sysbench --num-threads=8 --test=cpu \
--cpu-max-prime=100000 run
Total execution time ranges from 25.1s to 33.5s depending on threads
placement, where 25.1s is when all 8 threads are balanced properly
across 8 cpus.
Introducing NUMA identity node sched domain, which is based on how
SRAT/SLIT table define a NUMA node. This results in the following
hierarchy of sched domains on the same system described above.
domain0 SMT (span a core)
domain1 MC (span a last-level-cache)
domain2 NODE (span a NUMA node)
domain3 NUMA (span a socket: 4 nodes)
domain4 NUMA (span a system: 8 nodes)
This fixes the improper load balancing cases mentioned above.
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx>
---
Changes from V1 (https://lkml.org/lkml/2017/8/10/540)
* Update commit message to include performance number.
* Change from NUMA_IDEN to NODE.
* Fix code styling and update comments.
kernel/sched/topology.c | 26 +++++++++++++++++++++++---
1 file changed, 23 insertions(+), 3 deletions(-)
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index 79895ae..2dd5b11 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -1335,6 +1335,10 @@ void sched_init_numa(void)
if (!sched_domains_numa_distance)
return;
+ /* Includes NUMA identity node at level 0. */
+ sched_domains_numa_distance[level++] = curr_distance;
+ sched_domains_numa_levels = level;
+
/*
* O(nr_nodes^2) deduplicating selection sort -- in order to find the
* unique distances in the node_distance() table.
@@ -1382,8 +1386,7 @@ void sched_init_numa(void)
return;
/*
- * 'level' contains the number of unique distances, excluding the
- * identity distance node_distance(i,i).
+ * 'level' contains the number of unique distances
*
* The sched_domains_numa_distance[] array includes the actual distance
* numbers.
@@ -1445,9 +1448,26 @@ void sched_init_numa(void)
tl[i] = sched_domain_topology[i];
/*
+ * Do not setup NUMA node level if it has the same cpumask
+ * as sched domain at previous level:
+ * This is the case for system with:
+ * - LLC == NODE : LLC (MC) sched domain span a NUMA node.
+ * - DIE == NODE : DIE sched domain span a NUMA node.
+ *
+ * Assume all NUMA nodes are identical, so only check node 0.
+ */
+ if (!cpumask_equal(sched_domains_numa_masks[0][0], tl[i-1].mask(0))) {
+ tl[i++] = (struct sched_domain_topology_level){
+ .mask = sd_numa_mask,
+ .numa_level = 0,
+ SD_INIT_NAME(NODE)
+ };
+ }
+
+ /*
* .. and append 'j' levels of NUMA goodness.
*/
- for (j = 0; j < level; i++, j++) {
+ for (j = 1; j < level; i++, j++) {
tl[i] = (struct sched_domain_topology_level){
.mask = sd_numa_mask,
.sd_flags = cpu_numa_flags,
--
2.7.4