sched_ext: idle: Prioritize idle SMT sibling

In the default built-in idle CPU selection policy, when @prev_cpu is
busy and no fully idle core is available, try to place the task on its
SMT sibling if that sibling is idle, before searching any other idle CPU
in the same LLC.

Migration to the sibling is cheap and keeps the task on the same core,
preserving L1 cache and reducing wakeup latency.

On large SMT systems this appears to consistently boost throughput by
roughly 2-3% on CPU-bound workloads (running a number of tasks equal to
the number of SMT cores).

Cc: Cheng-Yang Chou <yphbchou0911@gmail.com>
Signed-off-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

1 files changed, 22 insertions, 4 deletions

diff --git a/kernel/sched/ext_idle.c b/kernel/sched/ext_idle.c
index c7e405262697..d9596427b5aa 100644
--- a/kernel/sched/ext_idle.c
+++ b/kernel/sched/ext_idle.c
@@ -424,18 +424,24 @@ static inline bool task_affinity_all(const struct task_struct *p)
  *   - prefer the last used CPU to take advantage of cached data (L1, L2) and
  *     branch prediction optimizations.
  *
- * 3. Pick a CPU within the same LLC (Last-Level Cache):
+ * 3. Prefer @prev_cpu's SMT sibling:
+ *   - if @prev_cpu is busy and no fully idle core is available, try to
+ *     place the task on an idle SMT sibling of @prev_cpu; keeping the
+ *     task on the same core makes migration cheaper, preserves L1 cache
+ *     locality and reduces wakeup latency.
+ *
+ * 4. Pick a CPU within the same LLC (Last-Level Cache):
  *   - if the above conditions aren't met, pick a CPU that shares the same
  *     LLC, if the LLC domain is a subset of @cpus_allowed, to maintain
  *     cache locality.
  *
- * 4. Pick a CPU within the same NUMA node, if enabled:
+ * 5. Pick a CPU within the same NUMA node, if enabled:
  *   - choose a CPU from the same NUMA node, if the node cpumask is a
  *     subset of @cpus_allowed, to reduce memory access latency.
  *
- * 5. Pick any idle CPU within the @cpus_allowed domain.
+ * 6. Pick any idle CPU within the @cpus_allowed domain.
  *
- * Step 3 and 4 are performed only if the system has, respectively,
+ * Step 4 and 5 are performed only if the system has, respectively,
  * multiple LLCs / multiple NUMA nodes (see scx_selcpu_topo_llc and
  * scx_selcpu_topo_numa) and they don't contain the same subset of CPUs.
  *
@@ -617,6 +623,18 @@ s32 scx_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags,
 	}
 
 	/*
+	 * Use @prev_cpu's sibling if it's idle.
+	 */
+	if (sched_smt_active()) {
+		for_each_cpu_and(cpu, cpu_smt_mask(prev_cpu), allowed) {
+			if (cpu == prev_cpu)
+				continue;
+			if (scx_idle_test_and_clear_cpu(cpu))
+				goto out_unlock;
+		}
+	}
+
+	/*
 	 * Search for any idle CPU in the same LLC domain.
 	 */
 	if (llc_cpus) {