lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [day] [month] [year] [list]
Message-ID: <aXjvLjTCRe8d3UFD@hpe.com>
Date: Tue, 27 Jan 2026 11:02:54 -0600
From: Kyle Meyer <kyle.meyer@....com>
To: bp@...en8.de, dave.hansen@...ux.intel.com, mingo@...hat.com,
        peterz@...radead.org, tglx@...nel.org, tim.c.chen@...ux.intel.com,
        vinicius.gomes@...el.com
Cc: brgerst@...il.com, hpa@...or.com, kprateek.nayak@....com,
        linux-kernel@...r.kernel.org, patryk.wlazlyn@...ux.intel.com,
        rafael.j.wysocki@...el.com, russ.anderson@....com, x86@...nel.org,
        yu.c.chen@...el.com, zhao1.liu@...el.com
Subject: [PATCH] sched/topology: Check average distances to remote packages

Granite Rapids (GNR) and Clearwater Forest (CWF) average distances to
remote packages to fix scheduler domains, see [1] for more information.

A warning and backtrace are printed when sub-NUMA clustering (SNC) is
enabled and there are more than 2 packages because the average distances
to remote packages could be different, skewing the single average remote
distance.

This is unnecessary when the average distances to remote packages are
the same.

Support single average remote distance on systems with more than 2
packages, preventing unnecessary warnings and backtraces by checking if
average distances to remote packages are the same.

[1] commit 4d6dd05d07d0 ("sched/topology: Fix sched domain build error for GNR, CWF in SNC-3 mode").

Signed-off-by: Kyle Meyer <kyle.meyer@....com>
---

The warning and backtrace were noticed on a 16 socket GNR system with SNC-2 enabled.

---
 arch/x86/kernel/smpboot.c | 70 ++++++++++++++++++++++++++++-----------
 1 file changed, 51 insertions(+), 19 deletions(-)

diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 5cd6950ab672..4467716f4054 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -518,27 +518,70 @@ static int avg_remote_numa_distance(void)
 {
 	int i, j;
 	int distance, nr_remote, total_distance;
+	int max_pkgs = topology_max_packages();
+	int cpu, pkg, pkg_avg_distance;
+	int *pkg_total_distance;
+	int *pkg_nr_remote;
 
 	if (sched_avg_remote_distance > 0)
 		return sched_avg_remote_distance;
 
+	sched_avg_remote_distance = REMOTE_DISTANCE;
+
 	nr_remote = 0;
 	total_distance = 0;
+
+	pkg_total_distance = kcalloc(max_pkgs, sizeof(int), GFP_KERNEL);
+	if (!pkg_total_distance)
+		goto cleanup;
+
+	pkg_nr_remote = kcalloc(max_pkgs, sizeof(int), GFP_KERNEL);
+	if (!pkg_nr_remote)
+		goto cleanup;
+
 	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 (distance < REMOTE_DISTANCE)
+				continue;
+
+			nr_remote++;
+			total_distance += distance;
+
+			cpu = cpumask_first(cpumask_of_node(j));
+			if (cpu >= nr_cpu_ids)
+				continue;
+
+			pkg = topology_physical_package_id(cpu);
+			pkg_total_distance[pkg] += distance;
+			pkg_nr_remote[pkg]++;
 		}
 	}
-	if (nr_remote)
-		sched_avg_remote_distance = total_distance / nr_remote;
-	else
-		sched_avg_remote_distance = REMOTE_DISTANCE;
 
+	if (!nr_remote)
+		goto cleanup;
+
+	sched_avg_remote_distance = total_distance / nr_remote;
+
+	/*
+	 * Single average remote distance won't be appropriate if different
+	 * packages have different distances to remote packages.
+	 */
+	for (i = 0; i < max_pkgs; i++) {
+		if (!pkg_nr_remote[i])
+			continue;
+
+		pkg_avg_distance = pkg_total_distance[i] / pkg_nr_remote[i];
+
+		pr_debug("sched: Avg. distance to remote package %d: %d\n", i, pkg_avg_distance);
+
+		if (pkg_avg_distance != sched_avg_remote_distance)
+			WARN_ONCE(1, "sched: Avg. distances to remote packages are different\n");
+	}
+cleanup:
+	kfree(pkg_nr_remote);
+	kfree(pkg_total_distance);
 	return sched_avg_remote_distance;
 }
 
@@ -564,18 +607,7 @@ int arch_sched_node_distance(int from, int to)
 		 * 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;
-- 
2.52.0


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ