* [PATCH v5 1/2] sched: Create architecture specific sched domain distances
2025-10-03 19:31 [PATCH v5 0/2] Fix NUMA sched domain build errors for GNR and CWF Tim Chen
@ 2025-10-03 19:31 ` Tim Chen
2025-10-03 19:31 ` [PATCH v5 2/2] sched/topology: Fix sched domain build error for GNR, CWF in SNC-3 mode Tim Chen
2025-10-06 19:34 ` [PATCH v5 0/2] Fix NUMA sched domain build errors for GNR and CWF Peter Zijlstra
2 siblings, 0 replies; 4+ messages in thread
From: Tim Chen @ 2025-10-03 19:31 UTC (permalink / raw)
To: Peter Zijlstra, Ingo Molnar
Cc: Tim Chen, Juri Lelli, Dietmar Eggemann, Ben Segall, Mel Gorman,
Valentin Schneider, Tim Chen, Vincent Guittot, Len Brown,
linux-kernel, Chen Yu, K Prateek Nayak, Gautham R . Shenoy,
Zhao Liu, Vinicius Costa Gomes, Arjan Van De Ven
Allow architecture specific sched domain NUMA distances that are
modified from actual NUMA node distances for the purpose of building
NUMA sched domains.
Keep actual NUMA distances separately if modified distances
are used for building sched domains. Such distances
are still needed as NUMA balancing benefits from finding the
NUMA nodes that are actually closer to a task numa_group.
Consolidate the recording of unique NUMA distances in an array to
sched_record_numa_dist() so the function can be reused to record NUMA
distances when the NUMA distance metric is changed.
No functional change and additional distance array
allocated if there're no arch specific NUMA distances
being defined.
Reviewed-by: Chen Yu <yu.c.chen@intel.com>
Co-developed-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
include/linux/sched/topology.h | 1 +
kernel/sched/topology.c | 108 ++++++++++++++++++++++++++-------
2 files changed, 87 insertions(+), 22 deletions(-)
diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h
index 5263746b63e8..2d2d29553df8 100644
--- a/include/linux/sched/topology.h
+++ b/include/linux/sched/topology.h
@@ -56,6 +56,7 @@ static inline int cpu_numa_flags(void)
{
return SD_NUMA;
}
+extern int arch_sched_node_distance(int from, int to);
#endif
extern int arch_asym_cpu_priority(int cpu);
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index 6e2f54169e66..3bbe48176dbd 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -1590,11 +1590,18 @@ static void claim_allocations(int cpu, struct sched_domain *sd)
#ifdef CONFIG_NUMA
enum numa_topology_type sched_numa_topology_type;
+/*
+ * sched_domains_numa_distance is derived from sched_numa_node_distance
+ * and provides a simplified view of NUMA distances used specifically
+ * for building NUMA scheduling domains.
+ */
static int sched_domains_numa_levels;
+static int sched_numa_node_levels;
static int sched_domains_curr_level;
int sched_max_numa_distance;
static int *sched_domains_numa_distance;
+static int *sched_numa_node_distance;
static struct cpumask ***sched_domains_numa_masks;
#endif /* CONFIG_NUMA */
@@ -1808,10 +1815,10 @@ bool find_numa_distance(int distance)
return true;
rcu_read_lock();
- distances = rcu_dereference(sched_domains_numa_distance);
+ distances = rcu_dereference(sched_numa_node_distance);
if (!distances)
goto unlock;
- for (i = 0; i < sched_domains_numa_levels; i++) {
+ for (i = 0; i < sched_numa_node_levels; i++) {
if (distances[i] == distance) {
found = true;
break;
@@ -1887,14 +1894,34 @@ static void init_numa_topology_type(int offline_node)
#define NR_DISTANCE_VALUES (1 << DISTANCE_BITS)
-void sched_init_numa(int offline_node)
+/*
+ * An architecture could modify its NUMA distance, to change
+ * grouping of NUMA nodes and number of NUMA levels when creating
+ * NUMA level sched domains.
+ *
+ * A NUMA level is created for each unique
+ * arch_sched_node_distance.
+ */
+static int numa_node_dist(int i, int j)
{
- struct sched_domain_topology_level *tl;
- unsigned long *distance_map;
+ return node_distance(i, j);
+}
+
+int arch_sched_node_distance(int from, int to)
+ __weak __alias(numa_node_dist);
+
+static bool modified_sched_node_distance(void)
+{
+ return numa_node_dist != arch_sched_node_distance;
+}
+
+static int sched_record_numa_dist(int offline_node, int (*n_dist)(int, int),
+ int **dist, int *levels)
+{
+ unsigned long *distance_map __free(bitmap) = NULL;
int nr_levels = 0;
int i, j;
int *distances;
- struct cpumask ***masks;
/*
* O(nr_nodes^2) de-duplicating selection sort -- in order to find the
@@ -1902,17 +1929,16 @@ void sched_init_numa(int offline_node)
*/
distance_map = bitmap_alloc(NR_DISTANCE_VALUES, GFP_KERNEL);
if (!distance_map)
- return;
+ return -ENOMEM;
bitmap_zero(distance_map, NR_DISTANCE_VALUES);
for_each_cpu_node_but(i, offline_node) {
for_each_cpu_node_but(j, offline_node) {
- int distance = node_distance(i, j);
+ int distance = n_dist(i, j);
if (distance < LOCAL_DISTANCE || distance >= NR_DISTANCE_VALUES) {
sched_numa_warn("Invalid distance value range");
- bitmap_free(distance_map);
- return;
+ return -EINVAL;
}
bitmap_set(distance_map, distance, 1);
@@ -1925,18 +1951,46 @@ void sched_init_numa(int offline_node)
nr_levels = bitmap_weight(distance_map, NR_DISTANCE_VALUES);
distances = kcalloc(nr_levels, sizeof(int), GFP_KERNEL);
- if (!distances) {
- bitmap_free(distance_map);
- return;
- }
+ if (!distances)
+ return -ENOMEM;
for (i = 0, j = 0; i < nr_levels; i++, j++) {
j = find_next_bit(distance_map, NR_DISTANCE_VALUES, j);
distances[i] = j;
}
- rcu_assign_pointer(sched_domains_numa_distance, distances);
+ *dist = distances;
+ *levels = nr_levels;
+
+ return 0;
+}
+
+void sched_init_numa(int offline_node)
+{
+ struct sched_domain_topology_level *tl;
+ int nr_levels, nr_node_levels;
+ int i, j;
+ int *distances, *domain_distances;
+ struct cpumask ***masks;
- bitmap_free(distance_map);
+ /* Record the NUMA distances from SLIT table */
+ if (sched_record_numa_dist(offline_node, numa_node_dist, &distances,
+ &nr_node_levels))
+ return;
+
+ /* Record modified NUMA distances for building sched domains */
+ if (modified_sched_node_distance()) {
+ if (sched_record_numa_dist(offline_node, arch_sched_node_distance,
+ &domain_distances, &nr_levels)) {
+ kfree(distances);
+ return;
+ }
+ } else {
+ domain_distances = distances;
+ nr_levels = nr_node_levels;
+ }
+ rcu_assign_pointer(sched_numa_node_distance, distances);
+ WRITE_ONCE(sched_max_numa_distance, distances[nr_node_levels - 1]);
+ WRITE_ONCE(sched_numa_node_levels, nr_node_levels);
/*
* 'nr_levels' contains the number of unique distances
@@ -1954,6 +2008,8 @@ void sched_init_numa(int offline_node)
*
* We reset it to 'nr_levels' at the end of this function.
*/
+ rcu_assign_pointer(sched_domains_numa_distance, domain_distances);
+
sched_domains_numa_levels = 0;
masks = kzalloc(sizeof(void *) * nr_levels, GFP_KERNEL);
@@ -1979,10 +2035,13 @@ void sched_init_numa(int offline_node)
masks[i][j] = mask;
for_each_cpu_node_but(k, offline_node) {
- if (sched_debug() && (node_distance(j, k) != node_distance(k, j)))
+ if (sched_debug() &&
+ (arch_sched_node_distance(j, k) !=
+ arch_sched_node_distance(k, j)))
sched_numa_warn("Node-distance not symmetric");
- if (node_distance(j, k) > sched_domains_numa_distance[i])
+ if (arch_sched_node_distance(j, k) >
+ sched_domains_numa_distance[i])
continue;
cpumask_or(mask, mask, cpumask_of_node(k));
@@ -2022,7 +2081,6 @@ void sched_init_numa(int offline_node)
sched_domain_topology = tl;
sched_domains_numa_levels = nr_levels;
- WRITE_ONCE(sched_max_numa_distance, sched_domains_numa_distance[nr_levels - 1]);
init_numa_topology_type(offline_node);
}
@@ -2030,14 +2088,18 @@ void sched_init_numa(int offline_node)
static void sched_reset_numa(void)
{
- int nr_levels, *distances;
+ int nr_levels, *distances, *dom_distances = NULL;
struct cpumask ***masks;
nr_levels = sched_domains_numa_levels;
+ sched_numa_node_levels = 0;
sched_domains_numa_levels = 0;
sched_max_numa_distance = 0;
sched_numa_topology_type = NUMA_DIRECT;
- distances = sched_domains_numa_distance;
+ distances = sched_numa_node_distance;
+ if (sched_numa_node_distance != sched_domains_numa_distance)
+ dom_distances = sched_domains_numa_distance;
+ rcu_assign_pointer(sched_numa_node_distance, NULL);
rcu_assign_pointer(sched_domains_numa_distance, NULL);
masks = sched_domains_numa_masks;
rcu_assign_pointer(sched_domains_numa_masks, NULL);
@@ -2046,6 +2108,7 @@ static void sched_reset_numa(void)
synchronize_rcu();
kfree(distances);
+ kfree(dom_distances);
for (i = 0; i < nr_levels && masks; i++) {
if (!masks[i])
continue;
@@ -2092,7 +2155,8 @@ void sched_domains_numa_masks_set(unsigned int cpu)
continue;
/* Set ourselves in the remote node's masks */
- if (node_distance(j, node) <= sched_domains_numa_distance[i])
+ if (arch_sched_node_distance(j, node) <=
+ sched_domains_numa_distance[i])
cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]);
}
}
--
2.32.0
^ permalink raw reply related [flat|nested] 4+ messages in thread* [PATCH v5 2/2] sched/topology: Fix sched domain build error for GNR, CWF in SNC-3 mode
2025-10-03 19:31 [PATCH v5 0/2] Fix NUMA sched domain build errors for GNR and CWF Tim Chen
2025-10-03 19:31 ` [PATCH v5 1/2] sched: Create architecture specific sched domain distances Tim Chen
@ 2025-10-03 19:31 ` Tim Chen
2025-10-06 19:34 ` [PATCH v5 0/2] Fix NUMA sched domain build errors for GNR and CWF Peter Zijlstra
2 siblings, 0 replies; 4+ messages in thread
From: Tim Chen @ 2025-10-03 19:31 UTC (permalink / raw)
To: Peter Zijlstra, Ingo Molnar
Cc: Tim Chen, Juri Lelli, Dietmar Eggemann, Ben Segall, Mel Gorman,
Valentin Schneider, Tim Chen, Vincent Guittot, Len Brown,
linux-kernel, Chen Yu, K Prateek Nayak, Gautham R . Shenoy,
Zhao Liu, Vinicius Costa Gomes, Arjan Van De Ven
It is possible for Granite Rapids (GNR) and Clearwater Forest
(CWF) to have up to 3 dies per package. When sub-numa cluster (SNC-3)
is enabled, each die will become a separate NUMA node in the package
with different distances between dies within the same package.
For example, on GNR, we see the following numa distances for a 2 socket
system with 3 dies per socket:
package 1 package2
----------------
| |
--------- ---------
| 0 | | 3 |
--------- ---------
| |
--------- ---------
| 1 | | 4 |
--------- ---------
| |
--------- ---------
| 2 | | 5 |
--------- ---------
| |
----------------
node distances:
node 0 1 2 3 4 5
0: 10 15 17 21 28 26
1: 15 10 15 23 26 23
2: 17 15 10 26 23 21
3: 21 28 26 10 15 17
4: 23 26 23 15 10 15
5: 26 23 21 17 15 10
The node distances above led to 2 problems:
1. Asymmetric routes taken between nodes in different packages led to
asymmetric scheduler domain perspective depending on which node you
are on. Current scheduler code failed to build domains properly with
asymmetric distances.
2. Multiple remote distances to respective tiles on remote package create
too many levels of domain hierarchies grouping different nodes between
remote packages.
For example, the above GNR topology lead to NUMA domains below:
Sched domains from the perspective of a CPU in node 0, where the number
in bracket represent node number.
NUMA-level 1 [0,1] [2]
NUMA-level 2 [0,1,2] [3]
NUMA-level 3 [0,1,2,3] [5]
NUMA-level 4 [0,1,2,3,5] [4]
Sched domains from the perspective of a CPU in node 4
NUMA-level 1 [4] [3,5]
NUMA-level 2 [3,4,5] [0,2]
NUMA-level 3 [0,2,3,4,5] [1]
Scheduler group peers for load balancing from the perspective of CPU 0
and 4 are different. Improper task could be chosen for load balancing
between groups such as [0,2,3,4,5] [1]. Ideally you should choose nodes
in 0 or 2 that are in same package as node 1 first. But instead tasks
in the remote package node 3, 4, 5 could be chosen with an equal chance
and could lead to excessive remote package migrations and imbalance of
load between packages. We should not group partial remote nodes and
local nodes together.
Simplify the remote distances for CWF and GNR for the purpose of
sched domains building, which maintains symmetry and leads to a more
reasonable load balance hierarchy.
The sched domains from the perspective of a CPU in node 0 NUMA-level 1
is now
NUMA-level 1 [0,1] [2]
NUMA-level 2 [0,1,2] [3,4,5]
The sched domains from the perspective of a CPU in node 4 NUMA-level 1
is now
NUMA-level 1 [4] [3,5]
NUMA-level 2 [3,4,5] [0,1,2]
We have the same balancing perspective from node 0 or node 4. Loads are
now balanced equally between packages.
Tested-by: Zhao Liu <zhao1.liu@intel.com>
Reviewed-by: Chen Yu <yu.c.chen@intel.com>
Co-developed-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/kernel/smpboot.c | 70 +++++++++++++++++++++++++++++++++++++++
1 file changed, 70 insertions(+)
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 33e166f6ab12..1feca897c6f6 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -515,6 +515,76 @@ static void __init build_sched_topology(void)
set_sched_topology(topology);
}
+#ifdef CONFIG_NUMA
+static int sched_avg_remote_distance;
+static int avg_remote_numa_distance(void)
+{
+ int i, j;
+ int distance, nr_remote, total_distance;
+
+ if (sched_avg_remote_distance > 0)
+ return sched_avg_remote_distance;
+
+ nr_remote = 0;
+ total_distance = 0;
+ for_each_node_state(i, N_CPU) {
+ for_each_node_state(j, N_CPU) {
+ distance = node_distance(i, j);
+
+ if (distance >= REMOTE_DISTANCE) {
+ nr_remote++;
+ total_distance += distance;
+ }
+ }
+ }
+ if (nr_remote)
+ sched_avg_remote_distance = total_distance / nr_remote;
+ else
+ sched_avg_remote_distance = REMOTE_DISTANCE;
+
+ return sched_avg_remote_distance;
+}
+
+int arch_sched_node_distance(int from, int to)
+{
+ int d = node_distance(from, to);
+
+ switch (boot_cpu_data.x86_vfm) {
+ case INTEL_GRANITERAPIDS_X:
+ case INTEL_ATOM_DARKMONT_X:
+
+ if (!x86_has_numa_in_package || topology_max_packages() == 1 ||
+ d < REMOTE_DISTANCE)
+ return d;
+
+ /*
+ * With SNC enabled, there could be too many levels of remote
+ * NUMA node distances, creating NUMA domain levels
+ * including local nodes and partial remote nodes.
+ *
+ * Trim finer distance tuning for NUMA nodes in remote package
+ * for the purpose of building sched domains. Group NUMA nodes
+ * in the remote package in the same sched group.
+ * Simplify NUMA domains and avoid extra NUMA levels including
+ * different remote NUMA nodes and local nodes.
+ *
+ * GNR and CWF don't expect systems with more than 2 packages
+ * and more than 2 hops between packages. Single average remote
+ * distance won't be appropriate if there are more than 2
+ * packages as average distance to different remote packages
+ * could be different.
+ */
+ WARN_ONCE(topology_max_packages() > 2,
+ "sched: Expect only up to 2 packages for GNR or CWF, "
+ "but saw %d packages when building sched domains.",
+ topology_max_packages());
+
+ d = avg_remote_numa_distance();
+ }
+ return d;
+}
+#endif /* CONFIG_NUMA */
+
void set_cpu_sibling_map(int cpu)
{
bool has_smt = __max_threads_per_core > 1;
--
2.32.0
^ permalink raw reply related [flat|nested] 4+ messages in thread