From: Li Chen <chenl311@chinatelecom.cn>

Currently, the SMT domain is added into sched_domain_topology by default.

If cpu_attach_domain() finds that the CPU SMT domain’s cpumask_weight
is just 1, it will destroy it.

On a large machine, such as one with 512 cores, this results in
512 redundant domain attach/destroy operations.

Avoid these unnecessary operations by simply checking
cpu_smt_num_threads and skip SMT domain if the SMT domain is not
enabled.

Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Li Chen <chenl311@chinatelecom.cn>
---
 arch/x86/kernel/smpboot.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 309338033356f..33e166f6ab122 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -492,6 +492,8 @@ static struct sched_domain_topology_level x86_topology[] = {
 
 static void __init build_sched_topology(void)
 {
+	struct sched_domain_topology_level *topology = x86_topology;
+
 	/*
 	 * When there is NUMA topology inside the package invalidate the
 	 * PKG domain since the NUMA domains will auto-magically create the
@@ -502,7 +504,15 @@ static void __init build_sched_topology(void)
 
 		memset(&x86_topology[pkgdom], 0, sizeof(x86_topology[pkgdom]));
 	}
-	set_sched_topology(x86_topology);
+
+	/*
+	 * Drop the SMT domains if there is only one thread per-core
+	 * since it'll get degenerated by the scheduler anyways.
+	 */
+	if (cpu_smt_num_threads <= 1)
+		++topology;
+
+	set_sched_topology(topology);
 }
 
 void set_cpu_sibling_map(int cpu)
-- 
2.50.0

The following commit has been merged into the sched/core branch of tip:

Commit-ID:     f79c9aa446d638190578515afcd06d6c9d72da55
Gitweb:        https://git.kernel.org/tip/f79c9aa446d638190578515afcd06d6c9d72da55
Author:        Li Chen <chenl311@chinatelecom.cn>
AuthorDate:    Thu, 10 Jul 2025 18:57:10 +08:00
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Mon, 14 Jul 2025 10:59:34 +02:00

x86/smpboot: avoid SMT domain attach/destroy if SMT is not enabled

Currently, the SMT domain is added into sched_domain_topology by default.

If cpu_attach_domain() finds that the CPU SMT domain’s cpumask_weight
is just 1, it will destroy it.

On a large machine, such as one with 512 cores, this results in
512 redundant domain attach/destroy operations.

Avoid these unnecessary operations by simply checking
cpu_smt_num_threads and skip SMT domain if the SMT domain is not
enabled.

Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Li Chen <chenl311@chinatelecom.cn>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Link: https://lore.kernel.org/r/20250710105715.66594-5-me@linux.beauty
---
 arch/x86/kernel/smpboot.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 8038286..412c813 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -492,6 +492,8 @@ static struct sched_domain_topology_level x86_topology[] = {
 
 static void __init build_sched_topology(void)
 {
+	struct sched_domain_topology_level *topology = x86_topology;
+
 	/*
 	 * When there is NUMA topology inside the package invalidate the
 	 * PKG domain since the NUMA domains will auto-magically create the
@@ -502,7 +504,15 @@ static void __init build_sched_topology(void)
 
 		memset(&x86_topology[pkgdom], 0, sizeof(x86_topology[pkgdom]));
 	}
-	set_sched_topology(x86_topology);
+
+	/*
+	 * Drop the SMT domains if there is only one thread per-core
+	 * since it'll get degenerated by the scheduler anyways.
+	 */
+	if (cpu_smt_num_threads <= 1)
+		++topology;
+
+	set_sched_topology(topology);
 }
 
 void set_cpu_sibling_map(int cpu)