linux-kernel - Re: [PATCH] sched/topology: Check average distances to remote packages

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>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <af42256c10975b2abed577244f4556863fab656d.camel@linux.intel.com>
Date: Wed, 04 Feb 2026 13:42:49 -0800
From: Tim Chen <tim.c.chen@...ux.intel.com>
To: Kyle Meyer <kyle.meyer@....com>, bp@...en8.de,
 dave.hansen@...ux.intel.com, 	mingo@...hat.com, peterz@...radead.org,
 tglx@...nel.org, 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: Re: [PATCH] sched/topology: Check average distances to remote
 packages

On Tue, 2026-01-27 at 11:02 -0600, Kyle Meyer wrote:
> 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.

Yes, I agree that if there are more than 2 packages but the packages are the
same average distance with each other (symmetric package topology), that
should be okay and no warnings is necessary.

Reviewed-by: Tim Chen <tim.c.chen@...ux.intel.com>

> 
> [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;