[PATCH v5 0/4] x86/smpboot: tidy sched-topology and drop useless SMT level

[PATCH v5 0/4] x86/smpboot: tidy sched-topology and drop useless SMT level

[Li Chen]

From: Li Chen <chenl311@chinatelecom.cn>

This series cleans up sched-domain topology handling and
eliminates hundreds of pointless attach/destroy cycles for large
machines when SMT is not available.

Patch 1, 2, and 3 do some cleanup and refactor.

Patch 4 is a follow-up that simply skip SMT domain when
cpu_smt_num_threads <= 1, so the SMT level never gets created.

Tested on Qemu.

changelog:
v2: fix wording issue as suggested by Thomas [1]
v3: remove pointless memset and adjust PKG index accordingly, as
    suggested by Thomas [2], and refine some other wording issues.
v4: v4: Split refactor patche into three parts (as suggested by Peter [3])
    and refined patch 4 logic (as done by K. [4]).
v5: fix some style issues and do some improvements as pointed out by K. [5][6]

[1]: https://lore.kernel.org/all/87msa2r018.ffs@tglx/
[2]: https://lore.kernel.org/all/875xglntx1.ffs@tglx/
[3]: https://lkml.org/lkml/2025/6/25/584
[4]: https://lore.kernel.org/lkml/1b706790-2fec-4582-a425-55eeff36c32e@amd.com/
[5]: https://lore.kernel.org/all/f391491d-f886-4579-9b40-78a57f2ed1b5@amd.com/
[6]: https://lore.kernel.org/all/7e7f686b-1f5a-46dc-8b16-c6d491160d1c@amd.com/

Li Chen (4):
  smpboot: introduce SDTL() helper to tidy sched topology setup
  x86/smpboot: remove redundant CONFIG_SCHED_SMT
  x86/smpboot: moves x86_topology to static initialize and truncate
  x86/smpboot: avoid SMT domain attach/destroy if SMT is not enabled

 arch/powerpc/kernel/smp.c      | 34 +++++++++--------------
 arch/s390/kernel/topology.c    | 10 +++----
 arch/x86/kernel/smpboot.c      | 51 ++++++++++++++++------------------
 include/linux/sched/topology.h |  4 +--
 kernel/sched/topology.c        | 24 ++++++----------
 5 files changed, 52 insertions(+), 71 deletions(-)

-- 
2.50.0

[PATCH v5 1/4] smpboot: introduce SDTL_INIT() helper to tidy sched topology setup

[Li Chen]

From: Li Chen <chenl311@chinatelecom.cn>

Define a small SDTL_INIT(maskfn, flagsfn, name) macro and use it to build the
sched_domain_topology_level array. Purely a cleanup; behaviour is unchanged.

Signed-off-by: Li Chen <chenl311@chinatelecom.cn>
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
---
 arch/powerpc/kernel/smp.c      | 34 +++++++++++++---------------------
 arch/s390/kernel/topology.c    | 10 +++++-----
 arch/x86/kernel/smpboot.c      | 21 ++++++---------------
 include/linux/sched/topology.h |  4 ++--
 kernel/sched/topology.c        | 24 ++++++++----------------
 5 files changed, 34 insertions(+), 59 deletions(-)

diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
index 5ac7084eebc0b..0b7ab7d2eb142 100644
--- a/arch/powerpc/kernel/smp.c
+++ b/arch/powerpc/kernel/smp.c
@@ -1700,28 +1700,20 @@ static void __init build_sched_topology(void)
 #ifdef CONFIG_SCHED_SMT
 	if (has_big_cores) {
 		pr_info("Big cores detected but using small core scheduling\n");
-		powerpc_topology[i++] = (struct sched_domain_topology_level){
-			smallcore_smt_mask, powerpc_smt_flags, SD_INIT_NAME(SMT)
-		};
-	} else {
-		powerpc_topology[i++] = (struct sched_domain_topology_level){
-			cpu_smt_mask, powerpc_smt_flags, SD_INIT_NAME(SMT)
-		};
-	}
+		powerpc_topology[i++] =
+			SDTL_INIT(smallcore_smt_mask, powerpc_smt_flags, SMT);
+	} else
+		powerpc_topology[i++] = SDTL_INIT(cpu_smt_mask, powerpc_smt_flags, SMT);
 #endif
-	if (shared_caches) {
-		powerpc_topology[i++] = (struct sched_domain_topology_level){
-			shared_cache_mask, powerpc_shared_cache_flags, SD_INIT_NAME(CACHE)
-		};
-	}
-	if (has_coregroup_support()) {
-		powerpc_topology[i++] = (struct sched_domain_topology_level){
-			cpu_mc_mask, powerpc_shared_proc_flags, SD_INIT_NAME(MC)
-		};
-	}
-	powerpc_topology[i++] = (struct sched_domain_topology_level){
-		cpu_cpu_mask, powerpc_shared_proc_flags, SD_INIT_NAME(PKG)
-	};
+	if (shared_caches)
+		powerpc_topology[i++] =
+			SDTL_INIT(shared_cache_mask, powerpc_shared_cache_flags, CACHE);
+
+	if (has_coregroup_support())
+		powerpc_topology[i++] =
+			SDTL_INIT(cpu_mc_mask, powerpc_shared_proc_flags, MC);
+
+	powerpc_topology[i++] = SDTL_INIT(cpu_cpu_mask, powerpc_shared_proc_flags, PKG);
 
 	/* There must be one trailing NULL entry left.  */
 	BUG_ON(i >= ARRAY_SIZE(powerpc_topology) - 1);
diff --git a/arch/s390/kernel/topology.c b/arch/s390/kernel/topology.c
index 3df048e190b11..46569b8e47dde 100644
--- a/arch/s390/kernel/topology.c
+++ b/arch/s390/kernel/topology.c
@@ -531,11 +531,11 @@ static const struct cpumask *cpu_drawer_mask(int cpu)
 }
 
 static struct sched_domain_topology_level s390_topology[] = {
-	{ cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
-	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
-	{ cpu_book_mask, SD_INIT_NAME(BOOK) },
-	{ cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
-	{ cpu_cpu_mask, SD_INIT_NAME(PKG) },
+	SDTL_INIT(cpu_thread_mask, cpu_smt_flags, SMT),
+	SDTL_INIT(cpu_coregroup_mask, cpu_core_flags, MC),
+	SDTL_INIT(cpu_book_mask, NULL, BOOK),
+	SDTL_INIT(cpu_drawer_mask, NULL, DRAWER),
+	SDTL_INIT(cpu_cpu_mask, NULL, PKG),
 	{ NULL, },
 };
 
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 58ede3fa6a75b..445127df2cb19 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -485,35 +485,26 @@ static void __init build_sched_topology(void)
 	int i = 0;
 
 #ifdef CONFIG_SCHED_SMT
-	x86_topology[i++] = (struct sched_domain_topology_level){
-		cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT)
-	};
+	x86_topology[i++] = SDTL_INIT(cpu_smt_mask, cpu_smt_flags, SMT);
 #endif
 #ifdef CONFIG_SCHED_CLUSTER
-	x86_topology[i++] = (struct sched_domain_topology_level){
-		cpu_clustergroup_mask, x86_cluster_flags, SD_INIT_NAME(CLS)
-	};
+	x86_topology[i++] = SDTL_INIT(cpu_clustergroup_mask, x86_cluster_flags, CLS);
 #endif
 #ifdef CONFIG_SCHED_MC
-	x86_topology[i++] = (struct sched_domain_topology_level){
-		cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC)
-	};
+	x86_topology[i++] = SDTL_INIT(cpu_coregroup_mask, x86_core_flags, MC);
 #endif
 	/*
 	 * When there is NUMA topology inside the package skip the PKG domain
 	 * since the NUMA domains will auto-magically create the right spanning
 	 * domains based on the SLIT.
 	 */
-	if (!x86_has_numa_in_package) {
-		x86_topology[i++] = (struct sched_domain_topology_level){
-			cpu_cpu_mask, x86_sched_itmt_flags, SD_INIT_NAME(PKG)
-		};
-	}
+	if (!x86_has_numa_in_package)
+		x86_topology[i++] = SDTL_INIT(cpu_cpu_mask, x86_sched_itmt_flags, PKG);
 
 	/*
 	 * There must be one trailing NULL entry left.
 	 */
-	BUG_ON(i >= ARRAY_SIZE(x86_topology)-1);
+	BUG_ON(i >= ARRAY_SIZE(x86_topology) - 1);
 
 	set_sched_topology(x86_topology);
 }
diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h
index e54e7fa76ba63..0d5daaa277b75 100644
--- a/include/linux/sched/topology.h
+++ b/include/linux/sched/topology.h
@@ -196,8 +196,8 @@ struct sched_domain_topology_level {
 extern void __init set_sched_topology(struct sched_domain_topology_level *tl);
 extern void sched_update_asym_prefer_cpu(int cpu, int old_prio, int new_prio);
 
-
-# define SD_INIT_NAME(type)		.name = #type
+#define SDTL_INIT(maskfn, flagsfn, dname) ((struct sched_domain_topology_level) \
+	    { .mask = maskfn, .sd_flags = flagsfn, .name = #dname })
 
 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
 extern void rebuild_sched_domains_energy(void);
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index 8e06b1d22e91e..d01f5a49f2e7a 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -1737,17 +1737,17 @@ sd_init(struct sched_domain_topology_level *tl,
  */
 static struct sched_domain_topology_level default_topology[] = {
 #ifdef CONFIG_SCHED_SMT
-	{ cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
+	SDTL_INIT(cpu_smt_mask, cpu_smt_flags, SMT),
 #endif
 
 #ifdef CONFIG_SCHED_CLUSTER
-	{ cpu_clustergroup_mask, cpu_cluster_flags, SD_INIT_NAME(CLS) },
+	SDTL_INIT(cpu_clustergroup_mask, cpu_cluster_flags, CLS),
 #endif
 
 #ifdef CONFIG_SCHED_MC
-	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
+	SDTL_INIT(cpu_coregroup_mask, cpu_core_flags, MC),
 #endif
-	{ cpu_cpu_mask, SD_INIT_NAME(PKG) },
+	SDTL_INIT(cpu_cpu_mask, NULL, PKG),
 	{ NULL, },
 };
 
@@ -2008,23 +2008,15 @@ void sched_init_numa(int offline_node)
 	/*
 	 * Add the NUMA identity distance, aka single NODE.
 	 */
-	tl[i++] = (struct sched_domain_topology_level){
-		.mask = sd_numa_mask,
-		.numa_level = 0,
-		SD_INIT_NAME(NODE)
-	};
+	tl[i++] = SDTL_INIT(sd_numa_mask, NULL, NODE);
 
 	/*
 	 * .. and append 'j' levels of NUMA goodness.
 	 */
 	for (j = 1; j < nr_levels; i++, j++) {
-		tl[i] = (struct sched_domain_topology_level){
-			.mask = sd_numa_mask,
-			.sd_flags = cpu_numa_flags,
-			.flags = SDTL_OVERLAP,
-			.numa_level = j,
-			SD_INIT_NAME(NUMA)
-		};
+		tl[i] = SDTL_INIT(sd_numa_mask, cpu_numa_flags, NUMA);
+		tl[i].numa_level = j;
+		tl[i].flags = SDTL_OVERLAP;
 	}
 
 	sched_domain_topology_saved = sched_domain_topology;
-- 
2.50.0

[PATCH v5 2/4] x86/smpboot: remove redundant CONFIG_SCHED_SMT

[Li Chen]

From: Li Chen <chenl311@chinatelecom.cn>

CONFIG_SCHED_SMT is default y if SMP is enabled,
so let's simply drop CONFIG_SCHED_SMT.

Signed-off-by: Li Chen <chenl311@chinatelecom.cn>
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
---
 arch/x86/kernel/smpboot.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 445127df2cb19..3d0f5737a157c 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -484,9 +484,7 @@ static void __init build_sched_topology(void)
 {
 	int i = 0;
 
-#ifdef CONFIG_SCHED_SMT
 	x86_topology[i++] = SDTL_INIT(cpu_smt_mask, cpu_smt_flags, SMT);
-#endif
 #ifdef CONFIG_SCHED_CLUSTER
 	x86_topology[i++] = SDTL_INIT(cpu_clustergroup_mask, x86_cluster_flags, CLS);
 #endif
-- 
2.50.0

[PATCH v5 3/4] x86/smpboot: moves x86_topology to static initialize and truncate

[Li Chen]

From: Li Chen <chenl311@chinatelecom.cn>

The #ifdeffery and the initializers in build_sched_topology() are just
disgusting.

Statically initialize the domain levels in the topology array and let
build_sched_topology() invalidate the package domain level when NUMA in
package is available.

Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Li Chen <chenl311@chinatelecom.cn>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
---
 arch/x86/kernel/smpboot.c | 34 ++++++++++++++++------------------
 1 file changed, 16 insertions(+), 18 deletions(-)

diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 3d0f5737a157c..309338033356f 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -478,32 +478,30 @@ static int x86_cluster_flags(void)
  */
 static bool x86_has_numa_in_package;
 
-static struct sched_domain_topology_level x86_topology[6];
-
-static void __init build_sched_topology(void)
-{
-	int i = 0;
-
-	x86_topology[i++] = SDTL_INIT(cpu_smt_mask, cpu_smt_flags, SMT);
+static struct sched_domain_topology_level x86_topology[] = {
+	SDTL_INIT(cpu_smt_mask, cpu_smt_flags, SMT),
 #ifdef CONFIG_SCHED_CLUSTER
-	x86_topology[i++] = SDTL_INIT(cpu_clustergroup_mask, x86_cluster_flags, CLS);
+	SDTL_INIT(cpu_clustergroup_mask, x86_cluster_flags, CLS),
 #endif
 #ifdef CONFIG_SCHED_MC
-	x86_topology[i++] = SDTL_INIT(cpu_coregroup_mask, x86_core_flags, MC);
+	SDTL_INIT(cpu_coregroup_mask, x86_core_flags, MC),
 #endif
-	/*
-	 * When there is NUMA topology inside the package skip the PKG domain
-	 * since the NUMA domains will auto-magically create the right spanning
-	 * domains based on the SLIT.
-	 */
-	if (!x86_has_numa_in_package)
-		x86_topology[i++] = SDTL_INIT(cpu_cpu_mask, x86_sched_itmt_flags, PKG);
+	SDTL_INIT(cpu_cpu_mask, x86_sched_itmt_flags, PKG),
+	{ NULL },
+};
 
+static void __init build_sched_topology(void)
+{
 	/*
-	 * There must be one trailing NULL entry left.
+	 * When there is NUMA topology inside the package invalidate the
+	 * PKG domain since the NUMA domains will auto-magically create the
+	 * right spanning domains based on the SLIT.
 	 */
-	BUG_ON(i >= ARRAY_SIZE(x86_topology) - 1);
+	if (x86_has_numa_in_package) {
+		unsigned int pkgdom = ARRAY_SIZE(x86_topology) - 2;
 
+		memset(&x86_topology[pkgdom], 0, sizeof(x86_topology[pkgdom]));
+	}
 	set_sched_topology(x86_topology);
 }
 
-- 
2.50.0

[PATCH v5 4/4] x86/smpboot: avoid SMT domain attach/destroy if SMT is not enabled

[Li Chen]

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

Re: [PATCH v5 1/4] smpboot: introduce SDTL_INIT() helper to tidy sched topology setup

[K Prateek Nayak]

On 7/10/2025 4:27 PM, Li Chen wrote:
>  	/*
>  	 * .. and append 'j' levels of NUMA goodness.
>  	 */
>  	for (j = 1; j < nr_levels; i++, j++) {
> -		tl[i] = (struct sched_domain_topology_level){
> -			.mask = sd_numa_mask,
> -			.sd_flags = cpu_numa_flags,
> -			.flags = SDTL_OVERLAP,
> -			.numa_level = j,
> -			SD_INIT_NAME(NUMA)
> -		};
> +		tl[i] = SDTL_INIT(sd_numa_mask, cpu_numa_flags, NUMA);
> +		tl[i].numa_level = j;
> +		tl[i].flags = SDTL_OVERLAP;

Tangential discussion: I was looking at this and was wondering why we
need a "tl->flags" when there is already sd_flags() function and we can
simply add SD_OVERLAP to sd_numa_flags().

I think "tl->flags" was needed when the idea of overlap domains was
added in commit e3589f6c81e4 ("sched: Allow for overlapping sched_domain
spans") when it depended on "FORCE_SD_OVERLAP" sched_feat() which
allowed toggling this off but that was done away with in commit
af85596c74de ("sched/topology: Remove FORCE_SD_OVERLAP") so perhaps we
can get rid of it now?

Relying on SD_NUMA should be enough currently. Peter, Valentin, what do
you think of something like below?

(Build and boot tested on top of this series on tip:sched/core)

diff --git a/include/linux/sched/sd_flags.h b/include/linux/sched/sd_flags.h
index b04a5d04dee9..42839cfa2778 100644
--- a/include/linux/sched/sd_flags.h
+++ b/include/linux/sched/sd_flags.h
@@ -153,14 +153,6 @@ SD_FLAG(SD_ASYM_PACKING, SDF_NEEDS_GROUPS)
  */
 SD_FLAG(SD_PREFER_SIBLING, SDF_NEEDS_GROUPS)
 
-/*
- * sched_groups of this level overlap
- *
- * SHARED_PARENT: Set for all NUMA levels above NODE.
- * NEEDS_GROUPS: Overlaps can only exist with more than one group.
- */
-SD_FLAG(SD_OVERLAP, SDF_SHARED_PARENT | SDF_NEEDS_GROUPS)
-
 /*
  * Cross-node balancing
  *
diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h
index 0d5daaa277b7..5263746b63e8 100644
--- a/include/linux/sched/topology.h
+++ b/include/linux/sched/topology.h
@@ -175,8 +175,6 @@ bool cpus_share_resources(int this_cpu, int that_cpu);
 typedef const struct cpumask *(*sched_domain_mask_f)(int cpu);
 typedef int (*sched_domain_flags_f)(void);
 
-#define SDTL_OVERLAP	0x01
-
 struct sd_data {
 	struct sched_domain *__percpu *sd;
 	struct sched_domain_shared *__percpu *sds;
@@ -187,7 +185,6 @@ struct sd_data {
 struct sched_domain_topology_level {
 	sched_domain_mask_f mask;
 	sched_domain_flags_f sd_flags;
-	int		    flags;
 	int		    numa_level;
 	struct sd_data      data;
 	char                *name;
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 20a845697c1d..b9b4bbbf0af6 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -9926,9 +9926,9 @@ void update_group_capacity(struct sched_domain *sd, int cpu)
 	min_capacity = ULONG_MAX;
 	max_capacity = 0;
 
-	if (child->flags & SD_OVERLAP) {
+	if (child->flags & SD_NUMA) {
 		/*
-		 * SD_OVERLAP domains cannot assume that child groups
+		 * SD_NUMA domains cannot assume that child groups
 		 * span the current group.
 		 */
 
@@ -9941,7 +9941,7 @@ void update_group_capacity(struct sched_domain *sd, int cpu)
 		}
 	} else  {
 		/*
-		 * !SD_OVERLAP domains can assume that child groups
+		 * !SD_NUMA domains can assume that child groups
 		 * span the current group.
 		 */
 
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index d01f5a49f2e7..977e133bb8a4 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -89,7 +89,7 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 			break;
 		}
 
-		if (!(sd->flags & SD_OVERLAP) &&
+		if (!(sd->flags & SD_NUMA) &&
 		    cpumask_intersects(groupmask, sched_group_span(group))) {
 			printk(KERN_CONT "\n");
 			printk(KERN_ERR "ERROR: repeated CPUs\n");
@@ -102,7 +102,7 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 				group->sgc->id,
 				cpumask_pr_args(sched_group_span(group)));
 
-		if ((sd->flags & SD_OVERLAP) &&
+		if ((sd->flags & SD_NUMA) &&
 		    !cpumask_equal(group_balance_mask(group), sched_group_span(group))) {
 			printk(KERN_CONT " mask=%*pbl",
 				cpumask_pr_args(group_balance_mask(group)));
@@ -1344,7 +1344,7 @@ void sched_update_asym_prefer_cpu(int cpu, int old_prio, int new_prio)
 		 * "sg->asym_prefer_cpu" to "sg->sgc->asym_prefer_cpu"
 		 * which is shared by all the overlapping groups.
 		 */
-		WARN_ON_ONCE(sd->flags & SD_OVERLAP);
+		WARN_ON_ONCE(sd->flags & SD_NUMA);
 
 		sg = sd->groups;
 		if (cpu != sg->asym_prefer_cpu) {
@@ -2016,7 +2016,6 @@ void sched_init_numa(int offline_node)
 	for (j = 1; j < nr_levels; i++, j++) {
 		tl[i] = SDTL_INIT(sd_numa_mask, cpu_numa_flags, NUMA);
 		tl[i].numa_level = j;
-		tl[i].flags = SDTL_OVERLAP;
 	}
 
 	sched_domain_topology_saved = sched_domain_topology;
@@ -2327,7 +2326,7 @@ static void __sdt_free(const struct cpumask *cpu_map)
 
 			if (sdd->sd) {
 				sd = *per_cpu_ptr(sdd->sd, j);
-				if (sd && (sd->flags & SD_OVERLAP))
+				if (sd && (sd->flags & SD_NUMA))
 					free_sched_groups(sd->groups, 0);
 				kfree(*per_cpu_ptr(sdd->sd, j));
 			}
@@ -2393,9 +2392,13 @@ static bool topology_span_sane(const struct cpumask *cpu_map)
 	id_seen = sched_domains_tmpmask2;
 
 	for_each_sd_topology(tl) {
+		int tl_common_flags = 0;
+
+		if (tl->sd_flags)
+			tl_common_flags = (*tl->sd_flags)();
 
 		/* NUMA levels are allowed to overlap */
-		if (tl->flags & SDTL_OVERLAP)
+		if (tl_common_flags & SD_NUMA)
 			continue;
 
 		cpumask_clear(covered);
@@ -2466,8 +2469,6 @@ build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *att
 
 			if (tl == sched_domain_topology)
 				*per_cpu_ptr(d.sd, i) = sd;
-			if (tl->flags & SDTL_OVERLAP)
-				sd->flags |= SD_OVERLAP;
 			if (cpumask_equal(cpu_map, sched_domain_span(sd)))
 				break;
 		}
@@ -2480,7 +2481,7 @@ build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *att
 	for_each_cpu(i, cpu_map) {
 		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
 			sd->span_weight = cpumask_weight(sched_domain_span(sd));
-			if (sd->flags & SD_OVERLAP) {
+			if (sd->flags & SD_NUMA) {
 				if (build_overlap_sched_groups(sd, i))
 					goto error;
 			} else {
---

We can also keep SD_OVERLAP and only remove SDTL_OVERLAP, tl->flags if
that is preferred or just have them both if you see a future !NUMA
usecases for overlapping domains.

>  	}
>  
>  	sched_domain_topology_saved = sched_domain_topology;

-- 
Thanks and Regards,
Prateek

Re: [PATCH v5 0/4] x86/smpboot: tidy sched-topology and drop useless SMT level

[K Prateek Nayak]

Hello Li,

On 7/10/2025 4:27 PM, Li Chen wrote:
> From: Li Chen <chenl311@chinatelecom.cn>
> 
> This series cleans up sched-domain topology handling and
> eliminates hundreds of pointless attach/destroy cycles for large
> machines when SMT is not available.
> 
> Patch 1, 2, and 3 do some cleanup and refactor.
> 
> Patch 4 is a follow-up that simply skip SMT domain when
> cpu_smt_num_threads <= 1, so the SMT level never gets created.
> 
> Tested on Qemu.

Feel free to include:

Reviewed-by: K Prateek Nayak <kprateek.nayak@amd.com>

I've given this series a spin again when looking at "tl->flags"
and everything runs as expected on x86 :)

> changelog:
> v2: fix wording issue as suggested by Thomas [1]
> v3: remove pointless memset and adjust PKG index accordingly, as
>     suggested by Thomas [2], and refine some other wording issues.
> v4: v4: Split refactor patche into three parts (as suggested by Peter [3])
>     and refined patch 4 logic (as done by K. [4]).
> v5: fix some style issues and do some improvements as pointed out by K. [5][6]
> 
> [1]: https://lore.kernel.org/all/87msa2r018.ffs@tglx/
> [2]: https://lore.kernel.org/all/875xglntx1.ffs@tglx/
> [3]: https://lkml.org/lkml/2025/6/25/584
> [4]: https://lore.kernel.org/lkml/1b706790-2fec-4582-a425-55eeff36c32e@amd.com/
> [5]: https://lore.kernel.org/all/f391491d-f886-4579-9b40-78a57f2ed1b5@amd.com/
> [6]: https://lore.kernel.org/all/7e7f686b-1f5a-46dc-8b16-c6d491160d1c@amd.com/
> 
> Li Chen (4):
>   smpboot: introduce SDTL() helper to tidy sched topology setup
>   x86/smpboot: remove redundant CONFIG_SCHED_SMT
>   x86/smpboot: moves x86_topology to static initialize and truncate
>   x86/smpboot: avoid SMT domain attach/destroy if SMT is not enabled
> 
>  arch/powerpc/kernel/smp.c      | 34 +++++++++--------------
>  arch/s390/kernel/topology.c    | 10 +++----
>  arch/x86/kernel/smpboot.c      | 51 ++++++++++++++++------------------
>  include/linux/sched/topology.h |  4 +--
>  kernel/sched/topology.c        | 24 ++++++----------
>  5 files changed, 52 insertions(+), 71 deletions(-)
> 

-- 
Thanks and Regards,
Prateek

Re: [PATCH v5 1/4] smpboot: introduce SDTL_INIT() helper to tidy sched topology setup

[Peter Zijlstra]

On Thu, Jul 10, 2025 at 06:57:07PM +0800, Li Chen wrote:
> From: Li Chen <chenl311@chinatelecom.cn>
> 
> Define a small SDTL_INIT(maskfn, flagsfn, name) macro and use it to build the
> sched_domain_topology_level array. Purely a cleanup; behaviour is unchanged.
> 
> Signed-off-by: Li Chen <chenl311@chinatelecom.cn>
> Suggested-by: Thomas Gleixner <tglx@linutronix.de>
> Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
> ---
>  arch/powerpc/kernel/smp.c      | 34 +++++++++++++---------------------
>  arch/s390/kernel/topology.c    | 10 +++++-----
>  arch/x86/kernel/smpboot.c      | 21 ++++++---------------
>  include/linux/sched/topology.h |  4 ++--
>  kernel/sched/topology.c        | 24 ++++++++----------------
>  5 files changed, 34 insertions(+), 59 deletions(-)
> 
> diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
> index 5ac7084eebc0b..0b7ab7d2eb142 100644
> --- a/arch/powerpc/kernel/smp.c
> +++ b/arch/powerpc/kernel/smp.c
> @@ -1700,28 +1700,20 @@ static void __init build_sched_topology(void)
>  #ifdef CONFIG_SCHED_SMT
>  	if (has_big_cores) {
>  		pr_info("Big cores detected but using small core scheduling\n");
> -		powerpc_topology[i++] = (struct sched_domain_topology_level){
> -			smallcore_smt_mask, powerpc_smt_flags, SD_INIT_NAME(SMT)
> -		};
> -	} else {
> -		powerpc_topology[i++] = (struct sched_domain_topology_level){
> -			cpu_smt_mask, powerpc_smt_flags, SD_INIT_NAME(SMT)
> -		};
> -	}
> +		powerpc_topology[i++] =
> +			SDTL_INIT(smallcore_smt_mask, powerpc_smt_flags, SMT);
> +	} else
> +		powerpc_topology[i++] = SDTL_INIT(cpu_smt_mask, powerpc_smt_flags, SMT);
>  #endif
> -	if (shared_caches) {
> -		powerpc_topology[i++] = (struct sched_domain_topology_level){
> -			shared_cache_mask, powerpc_shared_cache_flags, SD_INIT_NAME(CACHE)
> -		};
> -	}
> -	if (has_coregroup_support()) {
> -		powerpc_topology[i++] = (struct sched_domain_topology_level){
> -			cpu_mc_mask, powerpc_shared_proc_flags, SD_INIT_NAME(MC)
> -		};
> -	}
> -	powerpc_topology[i++] = (struct sched_domain_topology_level){
> -		cpu_cpu_mask, powerpc_shared_proc_flags, SD_INIT_NAME(PKG)
> -	};
> +	if (shared_caches)
> +		powerpc_topology[i++] =
> +			SDTL_INIT(shared_cache_mask, powerpc_shared_cache_flags, CACHE);
> +
> +	if (has_coregroup_support())
> +		powerpc_topology[i++] =
> +			SDTL_INIT(cpu_mc_mask, powerpc_shared_proc_flags, MC);
> +
> +	powerpc_topology[i++] = SDTL_INIT(cpu_cpu_mask, powerpc_shared_proc_flags, PKG);

You wrecked coding-style here and lost a bunch of curlies. I've fixed
that up for you.

Re: [PATCH v5 3/4] x86/smpboot: moves x86_topology to static initialize and truncate

[Peter Zijlstra]

On Thu, Jul 10, 2025 at 06:57:09PM +0800, Li Chen wrote:

> +		memset(&x86_topology[pkgdom], 0, sizeof(x86_topology[pkgdom]));

$ git grep "memset(&\([^,]*\), 0, sizeof(\1))" | wc -l
6439
$ git grep "memset(\([^,]*\), 0, sizeof(\*\1))" | wc -l
3319

But we don't have anything like:

#define zero_ref(var) memset(&(var), 0, sizeof(var))
#define zero_ptr(ptr) memset((ptr), 0, sizeof(*(ptr)))

Oh well..

Re: [PATCH v5 1/4] smpboot: introduce SDTL_INIT() helper to tidy sched topology setup

[Peter Zijlstra]

On Fri, Jul 11, 2025 at 11:20:30AM +0530, K Prateek Nayak wrote:
> On 7/10/2025 4:27 PM, Li Chen wrote:
> >  	/*
> >  	 * .. and append 'j' levels of NUMA goodness.
> >  	 */
> >  	for (j = 1; j < nr_levels; i++, j++) {
> > -		tl[i] = (struct sched_domain_topology_level){
> > -			.mask = sd_numa_mask,
> > -			.sd_flags = cpu_numa_flags,
> > -			.flags = SDTL_OVERLAP,
> > -			.numa_level = j,
> > -			SD_INIT_NAME(NUMA)
> > -		};
> > +		tl[i] = SDTL_INIT(sd_numa_mask, cpu_numa_flags, NUMA);
> > +		tl[i].numa_level = j;
> > +		tl[i].flags = SDTL_OVERLAP;
> 
> Tangential discussion: I was looking at this and was wondering why we
> need a "tl->flags" when there is already sd_flags() function and we can
> simply add SD_OVERLAP to sd_numa_flags().
> 
> I think "tl->flags" was needed when the idea of overlap domains was
> added in commit e3589f6c81e4 ("sched: Allow for overlapping sched_domain
> spans") when it depended on "FORCE_SD_OVERLAP" sched_feat() which
> allowed toggling this off but that was done away with in commit
> af85596c74de ("sched/topology: Remove FORCE_SD_OVERLAP") so perhaps we
> can get rid of it now?
> 
> Relying on SD_NUMA should be enough currently. Peter, Valentin, what do
> you think of something like below?

I think you're right. SD_NUMA appears to be the one and only case that
also has SDTL_OVERLAP which then results in setting SD_OVERLAP, making
SD_NUMA and SD_OVERLAP equivalent and SDTL_OVERLAP redundant.

I'll presume you're okay with me adding your SoB to things, and I'll
push out all 5 patches to queue/sched/core to let the robots have a go
at things.

Re: [PATCH v5 1/4] smpboot: introduce SDTL_INIT() helper to tidy sched topology setup

[Valentin Schneider]

On 11/07/25 11:20, K Prateek Nayak wrote:
> Tangential discussion: I was looking at this and was wondering why we
> need a "tl->flags" when there is already sd_flags() function and we can
> simply add SD_OVERLAP to sd_numa_flags().
>
> I think "tl->flags" was needed when the idea of overlap domains was
> added in commit e3589f6c81e4 ("sched: Allow for overlapping sched_domain
> spans") when it depended on "FORCE_SD_OVERLAP" sched_feat() which
> allowed toggling this off but that was done away with in commit
> af85596c74de ("sched/topology: Remove FORCE_SD_OVERLAP") so perhaps we
> can get rid of it now?
>
> Relying on SD_NUMA should be enough currently. Peter, Valentin, what do
> you think of something like below?
>

I remember asking myself the same question when I mucked about the SD
flags; I ended up convincing myself to let it be but I couldn't find any
note as to why. Looking at things in their current state, I agree with you,
we could just bin it.

Re: [PATCH v5 1/4] smpboot: introduce SDTL_INIT() helper to tidy sched topology setup

[K Prateek Nayak]

(trimming the cc to only kernel/sched folks to reduce the noise)

On 7/11/2025 6:36 PM, Peter Zijlstra wrote:
> On Fri, Jul 11, 2025 at 11:20:30AM +0530, K Prateek Nayak wrote:
>> On 7/10/2025 4:27 PM, Li Chen wrote:
>>>  	/*
>>>  	 * .. and append 'j' levels of NUMA goodness.
>>>  	 */
>>>  	for (j = 1; j < nr_levels; i++, j++) {
>>> -		tl[i] = (struct sched_domain_topology_level){
>>> -			.mask = sd_numa_mask,
>>> -			.sd_flags = cpu_numa_flags,
>>> -			.flags = SDTL_OVERLAP,
>>> -			.numa_level = j,
>>> -			SD_INIT_NAME(NUMA)
>>> -		};
>>> +		tl[i] = SDTL_INIT(sd_numa_mask, cpu_numa_flags, NUMA);
>>> +		tl[i].numa_level = j;
>>> +		tl[i].flags = SDTL_OVERLAP;
>>
>> Tangential discussion: I was looking at this and was wondering why we
>> need a "tl->flags" when there is already sd_flags() function and we can
>> simply add SD_OVERLAP to sd_numa_flags().
>>
>> I think "tl->flags" was needed when the idea of overlap domains was
>> added in commit e3589f6c81e4 ("sched: Allow for overlapping sched_domain
>> spans") when it depended on "FORCE_SD_OVERLAP" sched_feat() which
>> allowed toggling this off but that was done away with in commit
>> af85596c74de ("sched/topology: Remove FORCE_SD_OVERLAP") so perhaps we
>> can get rid of it now?
>>
>> Relying on SD_NUMA should be enough currently. Peter, Valentin, what do
>> you think of something like below?
> 
> I think you're right. SD_NUMA appears to be the one and only case that
> also has SDTL_OVERLAP which then results in setting SD_OVERLAP, making
> SD_NUMA and SD_OVERLAP equivalent and SDTL_OVERLAP redundant.
> 
> I'll presume you're okay with me adding your SoB to things, and I'll
> push out all 5 patches to queue/sched/core to let the robots have a go
> at things.

Works for me! If you need a formal commit message:

Support for overlapping domains added in commit e3589f6c81e4 ("sched:
Allow for overlapping sched_domain spans") also allowed forcefully
setting SD_OVERLAP for !NUMA domains via FORCE_SD_OVERLAP sched_feat().

Since NUMA domains had to be presumed overlapping to ensure correct
behavior, "sched_domain_topology_level::flags" was introduced. NUMA
domains added the SDTL_OVERLAP flag would ensure SD_OVERLAP was always
added during build_sched_domains() for these domains, even when
FORCE_SD_OVERLAP was off.

Condition for adding the SD_OVERLAP flag at the aforementioned commit
was as follows:

    if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP))
            sd->flags |= SD_OVERLAP; 

The FORCE_SD_OVERLAP debug feature was removed in commit af85596c74de
("sched/topology: Remove FORCE_SD_OVERLAP") which left the NUMA domains
as the exclusive users of SDTL_OVERLAP, SD_OVERLAP, and SD_NUMA flags.

Get rid of SDTL_OVERLAP and SD_OVERLAP as they have become redundant
and instead rely on SD_NUMA to detect the only overlapping domain
currently supported. Since SDTL_OVERLAP was the only user of
"tl->flags", get rid of "sched_domain_topology_level::flags" too.

Signed-off-by: K Prateek Nayak <kprateek.nayak@amd.com>
---

P.S. Are we still considering the following for v6.16 cycle?
https://lore.kernel.org/lkml/20250709161917.14298-1-kprateek.nayak@amd.com/

If not, I can rebase it on top of queue:sched/core and send it out with
the conflicts resolved to save you a couple of edits :)

-- 
Thanks and Regards,
Prateek

Re: [PATCH v5 1/4] smpboot: introduce SDTL_INIT() helper to tidy sched topology setup

[Peter Zijlstra]

On Mon, Jul 14, 2025 at 09:33:42AM +0530, K Prateek Nayak wrote:

> Works for me! If you need a formal commit message:

Thanks, much better than the badly edited thing I put in place.

> P.S. Are we still considering the following for v6.16 cycle?
> https://lore.kernel.org/lkml/20250709161917.14298-1-kprateek.nayak@amd.com/
> 
> If not, I can rebase it on top of queue:sched/core and send it out with
> the conflicts resolved to save you a couple of edits :)

Already done, about to push out.

[tip: sched/core] sched/topology: Remove sched_domain_topology_level::flags

[tip-bot2 for K Prateek Nayak]

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

Commit-ID:     1eec89a671413ce38df9fe9e70f5130a9eb79a59
Gitweb:        https://git.kernel.org/tip/1eec89a671413ce38df9fe9e70f5130a9eb79a59
Author:        K Prateek Nayak <kprateek.nayak@amd.com>
AuthorDate:    Fri, 11 Jul 2025 11:20:30 +05:30
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Mon, 14 Jul 2025 10:59:35 +02:00

sched/topology: Remove sched_domain_topology_level::flags

Support for overlapping domains added in commit e3589f6c81e4 ("sched:
Allow for overlapping sched_domain spans") also allowed forcefully
setting SD_OVERLAP for !NUMA domains via FORCE_SD_OVERLAP sched_feat().

Since NUMA domains had to be presumed overlapping to ensure correct
behavior, "sched_domain_topology_level::flags" was introduced. NUMA
domains added the SDTL_OVERLAP flag would ensure SD_OVERLAP was always
added during build_sched_domains() for these domains, even when
FORCE_SD_OVERLAP was off.

Condition for adding the SD_OVERLAP flag at the aforementioned commit
was as follows:

    if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP))
            sd->flags |= SD_OVERLAP;

The FORCE_SD_OVERLAP debug feature was removed in commit af85596c74de
("sched/topology: Remove FORCE_SD_OVERLAP") which left the NUMA domains
as the exclusive users of SDTL_OVERLAP, SD_OVERLAP, and SD_NUMA flags.

Get rid of SDTL_OVERLAP and SD_OVERLAP as they have become redundant
and instead rely on SD_NUMA to detect the only overlapping domain
currently supported. Since SDTL_OVERLAP was the only user of
"tl->flags", get rid of "sched_domain_topology_level::flags" too.

Signed-off-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/ba4dbdf8-bc37-493d-b2e0-2efb00ea3e19@amd.com
---
 include/linux/sched/sd_flags.h |  8 --------
 include/linux/sched/topology.h |  3 ---
 kernel/sched/fair.c            |  6 +++---
 kernel/sched/topology.c        | 19 ++++++++++---------
 4 files changed, 13 insertions(+), 23 deletions(-)

diff --git a/include/linux/sched/sd_flags.h b/include/linux/sched/sd_flags.h
index b04a5d0..42839cf 100644
--- a/include/linux/sched/sd_flags.h
+++ b/include/linux/sched/sd_flags.h
@@ -154,14 +154,6 @@ SD_FLAG(SD_ASYM_PACKING, SDF_NEEDS_GROUPS)
 SD_FLAG(SD_PREFER_SIBLING, SDF_NEEDS_GROUPS)
 
 /*
- * sched_groups of this level overlap
- *
- * SHARED_PARENT: Set for all NUMA levels above NODE.
- * NEEDS_GROUPS: Overlaps can only exist with more than one group.
- */
-SD_FLAG(SD_OVERLAP, SDF_SHARED_PARENT | SDF_NEEDS_GROUPS)
-
-/*
  * Cross-node balancing
  *
  * SHARED_PARENT: Set for all NUMA levels above NODE.
diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h
index 0d5daaa..5263746 100644
--- a/include/linux/sched/topology.h
+++ b/include/linux/sched/topology.h
@@ -175,8 +175,6 @@ bool cpus_share_resources(int this_cpu, int that_cpu);
 typedef const struct cpumask *(*sched_domain_mask_f)(int cpu);
 typedef int (*sched_domain_flags_f)(void);
 
-#define SDTL_OVERLAP	0x01
-
 struct sd_data {
 	struct sched_domain *__percpu *sd;
 	struct sched_domain_shared *__percpu *sds;
@@ -187,7 +185,6 @@ struct sd_data {
 struct sched_domain_topology_level {
 	sched_domain_mask_f mask;
 	sched_domain_flags_f sd_flags;
-	int		    flags;
 	int		    numa_level;
 	struct sd_data      data;
 	char                *name;
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 20a8456..b9b4bbb 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -9926,9 +9926,9 @@ void update_group_capacity(struct sched_domain *sd, int cpu)
 	min_capacity = ULONG_MAX;
 	max_capacity = 0;
 
-	if (child->flags & SD_OVERLAP) {
+	if (child->flags & SD_NUMA) {
 		/*
-		 * SD_OVERLAP domains cannot assume that child groups
+		 * SD_NUMA domains cannot assume that child groups
 		 * span the current group.
 		 */
 
@@ -9941,7 +9941,7 @@ void update_group_capacity(struct sched_domain *sd, int cpu)
 		}
 	} else  {
 		/*
-		 * !SD_OVERLAP domains can assume that child groups
+		 * !SD_NUMA domains can assume that child groups
 		 * span the current group.
 		 */
 
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index d01f5a4..977e133 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -89,7 +89,7 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 			break;
 		}
 
-		if (!(sd->flags & SD_OVERLAP) &&
+		if (!(sd->flags & SD_NUMA) &&
 		    cpumask_intersects(groupmask, sched_group_span(group))) {
 			printk(KERN_CONT "\n");
 			printk(KERN_ERR "ERROR: repeated CPUs\n");
@@ -102,7 +102,7 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 				group->sgc->id,
 				cpumask_pr_args(sched_group_span(group)));
 
-		if ((sd->flags & SD_OVERLAP) &&
+		if ((sd->flags & SD_NUMA) &&
 		    !cpumask_equal(group_balance_mask(group), sched_group_span(group))) {
 			printk(KERN_CONT " mask=%*pbl",
 				cpumask_pr_args(group_balance_mask(group)));
@@ -1344,7 +1344,7 @@ void sched_update_asym_prefer_cpu(int cpu, int old_prio, int new_prio)
 		 * "sg->asym_prefer_cpu" to "sg->sgc->asym_prefer_cpu"
 		 * which is shared by all the overlapping groups.
 		 */
-		WARN_ON_ONCE(sd->flags & SD_OVERLAP);
+		WARN_ON_ONCE(sd->flags & SD_NUMA);
 
 		sg = sd->groups;
 		if (cpu != sg->asym_prefer_cpu) {
@@ -2016,7 +2016,6 @@ void sched_init_numa(int offline_node)
 	for (j = 1; j < nr_levels; i++, j++) {
 		tl[i] = SDTL_INIT(sd_numa_mask, cpu_numa_flags, NUMA);
 		tl[i].numa_level = j;
-		tl[i].flags = SDTL_OVERLAP;
 	}
 
 	sched_domain_topology_saved = sched_domain_topology;
@@ -2327,7 +2326,7 @@ static void __sdt_free(const struct cpumask *cpu_map)
 
 			if (sdd->sd) {
 				sd = *per_cpu_ptr(sdd->sd, j);
-				if (sd && (sd->flags & SD_OVERLAP))
+				if (sd && (sd->flags & SD_NUMA))
 					free_sched_groups(sd->groups, 0);
 				kfree(*per_cpu_ptr(sdd->sd, j));
 			}
@@ -2393,9 +2392,13 @@ static bool topology_span_sane(const struct cpumask *cpu_map)
 	id_seen = sched_domains_tmpmask2;
 
 	for_each_sd_topology(tl) {
+		int tl_common_flags = 0;
+
+		if (tl->sd_flags)
+			tl_common_flags = (*tl->sd_flags)();
 
 		/* NUMA levels are allowed to overlap */
-		if (tl->flags & SDTL_OVERLAP)
+		if (tl_common_flags & SD_NUMA)
 			continue;
 
 		cpumask_clear(covered);
@@ -2466,8 +2469,6 @@ build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *att
 
 			if (tl == sched_domain_topology)
 				*per_cpu_ptr(d.sd, i) = sd;
-			if (tl->flags & SDTL_OVERLAP)
-				sd->flags |= SD_OVERLAP;
 			if (cpumask_equal(cpu_map, sched_domain_span(sd)))
 				break;
 		}
@@ -2480,7 +2481,7 @@ build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *att
 	for_each_cpu(i, cpu_map) {
 		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
 			sd->span_weight = cpumask_weight(sched_domain_span(sd));
-			if (sd->flags & SD_OVERLAP) {
+			if (sd->flags & SD_NUMA) {
 				if (build_overlap_sched_groups(sd, i))
 					goto error;
 			} else {

[tip: sched/core] x86/smpboot: avoid SMT domain attach/destroy if SMT is not enabled

[tip-bot2 for Li Chen]

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)

[tip: sched/core] x86/smpboot: moves x86_topology to static initialize and truncate

[tip-bot2 for Li Chen]

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

Commit-ID:     fbc2010d92e595dc13d8048db2419f963c8cb25e
Gitweb:        https://git.kernel.org/tip/fbc2010d92e595dc13d8048db2419f963c8cb25e
Author:        Li Chen <chenl311@chinatelecom.cn>
AuthorDate:    Thu, 10 Jul 2025 18:57:09 +08:00
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Mon, 14 Jul 2025 10:59:34 +02:00

x86/smpboot: moves x86_topology to static initialize and truncate

The #ifdeffery and the initializers in build_sched_topology() are just
disgusting.

Statically initialize the domain levels in the topology array and let
build_sched_topology() invalidate the package domain level when NUMA in
package is available.

Suggested-by: Thomas Gleixner <tglx@linutronix.de>
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/20250710105715.66594-4-me@linux.beauty
---
 arch/x86/kernel/smpboot.c | 34 ++++++++++++++++------------------
 1 file changed, 16 insertions(+), 18 deletions(-)

diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 30e6df2..8038286 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -478,32 +478,30 @@ static int x86_cluster_flags(void)
  */
 static bool x86_has_numa_in_package;
 
-static struct sched_domain_topology_level x86_topology[6];
-
-static void __init build_sched_topology(void)
-{
-	int i = 0;
-
-	x86_topology[i++] = SDTL_INIT(cpu_smt_mask, cpu_smt_flags, SMT);
+static struct sched_domain_topology_level x86_topology[] = {
+	SDTL_INIT(cpu_smt_mask, cpu_smt_flags, SMT),
 #ifdef CONFIG_SCHED_CLUSTER
-	x86_topology[i++] = SDTL_INIT(cpu_clustergroup_mask, x86_cluster_flags, CLS);
+	SDTL_INIT(cpu_clustergroup_mask, x86_cluster_flags, CLS),
 #endif
 #ifdef CONFIG_SCHED_MC
-	x86_topology[i++] = SDTL_INIT(cpu_coregroup_mask, x86_core_flags, MC);
+	SDTL_INIT(cpu_coregroup_mask, x86_core_flags, MC),
 #endif
-	/*
-	 * When there is NUMA topology inside the package skip the PKG domain
-	 * since the NUMA domains will auto-magically create the right spanning
-	 * domains based on the SLIT.
-	 */
-	if (!x86_has_numa_in_package)
-		x86_topology[i++] = SDTL_INIT(cpu_cpu_mask, x86_sched_itmt_flags, PKG);
+	SDTL_INIT(cpu_cpu_mask, x86_sched_itmt_flags, PKG),
+	{ NULL },
+};
 
+static void __init build_sched_topology(void)
+{
 	/*
-	 * There must be one trailing NULL entry left.
+	 * When there is NUMA topology inside the package invalidate the
+	 * PKG domain since the NUMA domains will auto-magically create the
+	 * right spanning domains based on the SLIT.
 	 */
-	BUG_ON(i >= ARRAY_SIZE(x86_topology) - 1);
+	if (x86_has_numa_in_package) {
+		unsigned int pkgdom = ARRAY_SIZE(x86_topology) - 2;
 
+		memset(&x86_topology[pkgdom], 0, sizeof(x86_topology[pkgdom]));
+	}
 	set_sched_topology(x86_topology);
 }
 

[tip: sched/core] x86/smpboot: remove redundant CONFIG_SCHED_SMT

[tip-bot2 for Li Chen]

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

Commit-ID:     992de2b02509bed68f693ea5a68b07cd586197b7
Gitweb:        https://git.kernel.org/tip/992de2b02509bed68f693ea5a68b07cd586197b7
Author:        Li Chen <chenl311@chinatelecom.cn>
AuthorDate:    Thu, 10 Jul 2025 18:57:08 +08:00
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Mon, 14 Jul 2025 10:59:34 +02:00

x86/smpboot: remove redundant CONFIG_SCHED_SMT

On x86 CONFIG_SCHED_SMT is default y if SMP is enabled, so let's
simply drop CONFIG_SCHED_SMT.

Suggested-by: Thomas Gleixner <tglx@linutronix.de>
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/20250710105715.66594-3-me@linux.beauty
---
 arch/x86/kernel/smpboot.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index e0adf75..30e6df2 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -484,9 +484,7 @@ static void __init build_sched_topology(void)
 {
 	int i = 0;
 
-#ifdef CONFIG_SCHED_SMT
 	x86_topology[i++] = SDTL_INIT(cpu_smt_mask, cpu_smt_flags, SMT);
-#endif
 #ifdef CONFIG_SCHED_CLUSTER
 	x86_topology[i++] = SDTL_INIT(cpu_clustergroup_mask, x86_cluster_flags, CLS);
 #endif

[tip: sched/core] smpboot: introduce SDTL_INIT() helper to tidy sched topology setup

[tip-bot2 for Li Chen]

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

Commit-ID:     e075f4360931263f5ec006ea5dadc065e5e98eb8
Gitweb:        https://git.kernel.org/tip/e075f4360931263f5ec006ea5dadc065e5e98eb8
Author:        Li Chen <chenl311@chinatelecom.cn>
AuthorDate:    Thu, 10 Jul 2025 18:57:07 +08:00
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Mon, 14 Jul 2025 10:59:34 +02:00

smpboot: introduce SDTL_INIT() helper to tidy sched topology setup

Define a small SDTL_INIT(maskfn, flagsfn, name) macro and use it to build the
sched_domain_topology_level array. Purely a cleanup; behaviour is unchanged.

Suggested-by: Thomas Gleixner <tglx@linutronix.de>
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/20250710105715.66594-2-me@linux.beauty
---
 arch/powerpc/kernel/smp.c      | 25 ++++++++++---------------
 arch/s390/kernel/topology.c    | 10 +++++-----
 arch/x86/kernel/smpboot.c      | 21 ++++++---------------
 include/linux/sched/topology.h |  4 ++--
 kernel/sched/topology.c        | 24 ++++++++----------------
 5 files changed, 31 insertions(+), 53 deletions(-)

diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
index 5ac7084..f59e4b9 100644
--- a/arch/powerpc/kernel/smp.c
+++ b/arch/powerpc/kernel/smp.c
@@ -1700,28 +1700,23 @@ static void __init build_sched_topology(void)
 #ifdef CONFIG_SCHED_SMT
 	if (has_big_cores) {
 		pr_info("Big cores detected but using small core scheduling\n");
-		powerpc_topology[i++] = (struct sched_domain_topology_level){
-			smallcore_smt_mask, powerpc_smt_flags, SD_INIT_NAME(SMT)
-		};
+		powerpc_topology[i++] =
+			SDTL_INIT(smallcore_smt_mask, powerpc_smt_flags, SMT);
 	} else {
-		powerpc_topology[i++] = (struct sched_domain_topology_level){
-			cpu_smt_mask, powerpc_smt_flags, SD_INIT_NAME(SMT)
-		};
+		powerpc_topology[i++] = SDTL_INIT(cpu_smt_mask, powerpc_smt_flags, SMT);
 	}
 #endif
 	if (shared_caches) {
-		powerpc_topology[i++] = (struct sched_domain_topology_level){
-			shared_cache_mask, powerpc_shared_cache_flags, SD_INIT_NAME(CACHE)
-		};
+		powerpc_topology[i++] =
+			SDTL_INIT(shared_cache_mask, powerpc_shared_cache_flags, CACHE);
 	}
+
 	if (has_coregroup_support()) {
-		powerpc_topology[i++] = (struct sched_domain_topology_level){
-			cpu_mc_mask, powerpc_shared_proc_flags, SD_INIT_NAME(MC)
-		};
+		powerpc_topology[i++] =
+			SDTL_INIT(cpu_mc_mask, powerpc_shared_proc_flags, MC);
 	}
-	powerpc_topology[i++] = (struct sched_domain_topology_level){
-		cpu_cpu_mask, powerpc_shared_proc_flags, SD_INIT_NAME(PKG)
-	};
+
+	powerpc_topology[i++] = SDTL_INIT(cpu_cpu_mask, powerpc_shared_proc_flags, PKG);
 
 	/* There must be one trailing NULL entry left.  */
 	BUG_ON(i >= ARRAY_SIZE(powerpc_topology) - 1);
diff --git a/arch/s390/kernel/topology.c b/arch/s390/kernel/topology.c
index 3df048e..46569b8 100644
--- a/arch/s390/kernel/topology.c
+++ b/arch/s390/kernel/topology.c
@@ -531,11 +531,11 @@ static const struct cpumask *cpu_drawer_mask(int cpu)
 }
 
 static struct sched_domain_topology_level s390_topology[] = {
-	{ cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
-	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
-	{ cpu_book_mask, SD_INIT_NAME(BOOK) },
-	{ cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
-	{ cpu_cpu_mask, SD_INIT_NAME(PKG) },
+	SDTL_INIT(cpu_thread_mask, cpu_smt_flags, SMT),
+	SDTL_INIT(cpu_coregroup_mask, cpu_core_flags, MC),
+	SDTL_INIT(cpu_book_mask, NULL, BOOK),
+	SDTL_INIT(cpu_drawer_mask, NULL, DRAWER),
+	SDTL_INIT(cpu_cpu_mask, NULL, PKG),
 	{ NULL, },
 };
 
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index fc78c23..e0adf75 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -485,35 +485,26 @@ static void __init build_sched_topology(void)
 	int i = 0;
 
 #ifdef CONFIG_SCHED_SMT
-	x86_topology[i++] = (struct sched_domain_topology_level){
-		cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT)
-	};
+	x86_topology[i++] = SDTL_INIT(cpu_smt_mask, cpu_smt_flags, SMT);
 #endif
 #ifdef CONFIG_SCHED_CLUSTER
-	x86_topology[i++] = (struct sched_domain_topology_level){
-		cpu_clustergroup_mask, x86_cluster_flags, SD_INIT_NAME(CLS)
-	};
+	x86_topology[i++] = SDTL_INIT(cpu_clustergroup_mask, x86_cluster_flags, CLS);
 #endif
 #ifdef CONFIG_SCHED_MC
-	x86_topology[i++] = (struct sched_domain_topology_level){
-		cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC)
-	};
+	x86_topology[i++] = SDTL_INIT(cpu_coregroup_mask, x86_core_flags, MC);
 #endif
 	/*
 	 * When there is NUMA topology inside the package skip the PKG domain
 	 * since the NUMA domains will auto-magically create the right spanning
 	 * domains based on the SLIT.
 	 */
-	if (!x86_has_numa_in_package) {
-		x86_topology[i++] = (struct sched_domain_topology_level){
-			cpu_cpu_mask, x86_sched_itmt_flags, SD_INIT_NAME(PKG)
-		};
-	}
+	if (!x86_has_numa_in_package)
+		x86_topology[i++] = SDTL_INIT(cpu_cpu_mask, x86_sched_itmt_flags, PKG);
 
 	/*
 	 * There must be one trailing NULL entry left.
 	 */
-	BUG_ON(i >= ARRAY_SIZE(x86_topology)-1);
+	BUG_ON(i >= ARRAY_SIZE(x86_topology) - 1);
 
 	set_sched_topology(x86_topology);
 }
diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h
index e54e7fa..0d5daaa 100644
--- a/include/linux/sched/topology.h
+++ b/include/linux/sched/topology.h
@@ -196,8 +196,8 @@ struct sched_domain_topology_level {
 extern void __init set_sched_topology(struct sched_domain_topology_level *tl);
 extern void sched_update_asym_prefer_cpu(int cpu, int old_prio, int new_prio);
 
-
-# define SD_INIT_NAME(type)		.name = #type
+#define SDTL_INIT(maskfn, flagsfn, dname) ((struct sched_domain_topology_level) \
+	    { .mask = maskfn, .sd_flags = flagsfn, .name = #dname })
 
 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
 extern void rebuild_sched_domains_energy(void);
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index 8e06b1d..d01f5a4 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -1737,17 +1737,17 @@ sd_init(struct sched_domain_topology_level *tl,
  */
 static struct sched_domain_topology_level default_topology[] = {
 #ifdef CONFIG_SCHED_SMT
-	{ cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
+	SDTL_INIT(cpu_smt_mask, cpu_smt_flags, SMT),
 #endif
 
 #ifdef CONFIG_SCHED_CLUSTER
-	{ cpu_clustergroup_mask, cpu_cluster_flags, SD_INIT_NAME(CLS) },
+	SDTL_INIT(cpu_clustergroup_mask, cpu_cluster_flags, CLS),
 #endif
 
 #ifdef CONFIG_SCHED_MC
-	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
+	SDTL_INIT(cpu_coregroup_mask, cpu_core_flags, MC),
 #endif
-	{ cpu_cpu_mask, SD_INIT_NAME(PKG) },
+	SDTL_INIT(cpu_cpu_mask, NULL, PKG),
 	{ NULL, },
 };
 
@@ -2008,23 +2008,15 @@ void sched_init_numa(int offline_node)
 	/*
 	 * Add the NUMA identity distance, aka single NODE.
 	 */
-	tl[i++] = (struct sched_domain_topology_level){
-		.mask = sd_numa_mask,
-		.numa_level = 0,
-		SD_INIT_NAME(NODE)
-	};
+	tl[i++] = SDTL_INIT(sd_numa_mask, NULL, NODE);
 
 	/*
 	 * .. and append 'j' levels of NUMA goodness.
 	 */
 	for (j = 1; j < nr_levels; i++, j++) {
-		tl[i] = (struct sched_domain_topology_level){
-			.mask = sd_numa_mask,
-			.sd_flags = cpu_numa_flags,
-			.flags = SDTL_OVERLAP,
-			.numa_level = j,
-			SD_INIT_NAME(NUMA)
-		};
+		tl[i] = SDTL_INIT(sd_numa_mask, cpu_numa_flags, NUMA);
+		tl[i].numa_level = j;
+		tl[i].flags = SDTL_OVERLAP;
 	}
 
 	sched_domain_topology_saved = sched_domain_topology;