mirror of
https://github.com/torvalds/linux.git
synced 2026-03-08 05:44:45 +01:00
x86/topo: Fix SNC topology mess
Per4d6dd05d07("sched/topology: Fix sched domain build error for GNR, CWF in SNC-3 mode"), the original crazy SNC-3 SLIT table was: node distances: node 0 1 2 3 4 5 0: 10 15 17 21 28 26 1: 15 10 15 23 26 23 2: 17 15 10 26 23 21 3: 21 28 26 10 15 17 4: 23 26 23 15 10 15 5: 26 23 21 17 15 10 And per: https://lore.kernel.org/lkml/20250825075642.GQ3245006@noisy.programming.kicks-ass.net/ The suggestion was to average the off-trace clusters to restore sanity. However,4d6dd05d07implements this under various assumptions: - anything GNR/CWF with numa_in_package; - there will never be more than 2 packages; - the off-trace cluster will have distance >20 And then HPE shows up with a machine that matches the Vendor-Family-Model checks but looks like this: Here's an 8 socket (2 chassis) HPE system with SNC enabled: node 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0: 10 12 16 16 16 16 18 18 40 40 40 40 40 40 40 40 1: 12 10 16 16 16 16 18 18 40 40 40 40 40 40 40 40 2: 16 16 10 12 18 18 16 16 40 40 40 40 40 40 40 40 3: 16 16 12 10 18 18 16 16 40 40 40 40 40 40 40 40 4: 16 16 18 18 10 12 16 16 40 40 40 40 40 40 40 40 5: 16 16 18 18 12 10 16 16 40 40 40 40 40 40 40 40 6: 18 18 16 16 16 16 10 12 40 40 40 40 40 40 40 40 7: 18 18 16 16 16 16 12 10 40 40 40 40 40 40 40 40 8: 40 40 40 40 40 40 40 40 10 12 16 16 16 16 18 18 9: 40 40 40 40 40 40 40 40 12 10 16 16 16 16 18 18 10: 40 40 40 40 40 40 40 40 16 16 10 12 18 18 16 16 11: 40 40 40 40 40 40 40 40 16 16 12 10 18 18 16 16 12: 40 40 40 40 40 40 40 40 16 16 18 18 10 12 16 16 13: 40 40 40 40 40 40 40 40 16 16 18 18 12 10 16 16 14: 40 40 40 40 40 40 40 40 18 18 16 16 16 16 10 12 15: 40 40 40 40 40 40 40 40 18 18 16 16 16 16 12 10 10 = Same chassis and socket 12 = Same chassis and socket (SNC) 16 = Same chassis and adjacent socket 18 = Same chassis and non-adjacent socket 40 = Different chassis Turns out, the 'max 2 packages' thing is only relevant to the SNC-3 parts, the smaller parts do 8 sockets (like usual). The above SLIT table is sane, but violates the previous assumptions and trips a WARN. Now that the topology code has a sensible measure of nodes-per-package, we can use that to divinate the SNC mode at hand, and only fix up SNC-3 topologies. There is a 'healthy' amount of paranoia code validating the assumptions on the SLIT table, a simple pr_err(FW_BUG) print on failure and a fallback to using the regular table. Lets see how long this lasts :-) Fixes:4d6dd05d07("sched/topology: Fix sched domain build error for GNR, CWF in SNC-3 mode") Reported-by: Kyle Meyer <kyle.meyer@hpe.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Ingo Molnar <mingo@kernel.org> Tested-by: K Prateek Nayak <kprateek.nayak@amd.com> Tested-by: Zhang Rui <rui.zhang@intel.com> Tested-by: Chen Yu <yu.c.chen@intel.com> Tested-by: Kyle Meyer <kyle.meyer@hpe.com> Link: https://patch.msgid.link/20260303110100.238361290@infradead.org
This commit is contained in:
Peter Zijlstra
parent
717b64d58c
commit
528d89a470
1 changed files with 142 additions and 46 deletions
|
|
@ -506,33 +506,149 @@ static void __init build_sched_topology(void)
|
|||
}
|
||||
|
||||
#ifdef CONFIG_NUMA
|
||||
static int sched_avg_remote_distance;
|
||||
static int avg_remote_numa_distance(void)
|
||||
/*
|
||||
* Test if the on-trace cluster at (N,N) is symmetric.
|
||||
* Uses upper triangle iteration to avoid obvious duplicates.
|
||||
*/
|
||||
static bool slit_cluster_symmetric(int N)
|
||||
{
|
||||
int i, j;
|
||||
int distance, nr_remote, total_distance;
|
||||
int u = topology_num_nodes_per_package();
|
||||
|
||||
if (sched_avg_remote_distance > 0)
|
||||
return sched_avg_remote_distance;
|
||||
|
||||
nr_remote = 0;
|
||||
total_distance = 0;
|
||||
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;
|
||||
}
|
||||
for (int k = 0; k < u; k++) {
|
||||
for (int l = k; l < u; l++) {
|
||||
if (node_distance(N + k, N + l) !=
|
||||
node_distance(N + l, N + k))
|
||||
return false;
|
||||
}
|
||||
}
|
||||
if (nr_remote)
|
||||
sched_avg_remote_distance = total_distance / nr_remote;
|
||||
else
|
||||
sched_avg_remote_distance = REMOTE_DISTANCE;
|
||||
|
||||
return sched_avg_remote_distance;
|
||||
return true;
|
||||
}
|
||||
|
||||
/*
|
||||
* Return the package-id of the cluster, or ~0 if indeterminate.
|
||||
* Each node in the on-trace cluster should have the same package-id.
|
||||
*/
|
||||
static u32 slit_cluster_package(int N)
|
||||
{
|
||||
int u = topology_num_nodes_per_package();
|
||||
u32 pkg_id = ~0;
|
||||
|
||||
for (int n = 0; n < u; n++) {
|
||||
const struct cpumask *cpus = cpumask_of_node(N + n);
|
||||
int cpu;
|
||||
|
||||
for_each_cpu(cpu, cpus) {
|
||||
u32 id = topology_logical_package_id(cpu);
|
||||
|
||||
if (pkg_id == ~0)
|
||||
pkg_id = id;
|
||||
if (pkg_id != id)
|
||||
return ~0;
|
||||
}
|
||||
}
|
||||
|
||||
return pkg_id;
|
||||
}
|
||||
|
||||
/*
|
||||
* Validate the SLIT table is of the form expected for SNC, specifically:
|
||||
*
|
||||
* - each on-trace cluster should be symmetric,
|
||||
* - each on-trace cluster should have a unique package-id.
|
||||
*
|
||||
* If you NUMA_EMU on top of SNC, you get to keep the pieces.
|
||||
*/
|
||||
static bool slit_validate(void)
|
||||
{
|
||||
int u = topology_num_nodes_per_package();
|
||||
u32 pkg_id, prev_pkg_id = ~0;
|
||||
|
||||
for (int pkg = 0; pkg < topology_max_packages(); pkg++) {
|
||||
int n = pkg * u;
|
||||
|
||||
/*
|
||||
* Ensure the on-trace cluster is symmetric and each cluster
|
||||
* has a different package id.
|
||||
*/
|
||||
if (!slit_cluster_symmetric(n))
|
||||
return false;
|
||||
pkg_id = slit_cluster_package(n);
|
||||
if (pkg_id == ~0)
|
||||
return false;
|
||||
if (pkg && pkg_id == prev_pkg_id)
|
||||
return false;
|
||||
|
||||
prev_pkg_id = pkg_id;
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
/*
|
||||
* Compute a sanitized SLIT table for SNC; notably SNC-3 can end up with
|
||||
* asymmetric off-trace clusters, reflecting physical assymmetries. However
|
||||
* this leads to 'unfortunate' sched_domain configurations.
|
||||
*
|
||||
* For example dual socket GNR with SNC-3:
|
||||
*
|
||||
* node distances:
|
||||
* node 0 1 2 3 4 5
|
||||
* 0: 10 15 17 21 28 26
|
||||
* 1: 15 10 15 23 26 23
|
||||
* 2: 17 15 10 26 23 21
|
||||
* 3: 21 28 26 10 15 17
|
||||
* 4: 23 26 23 15 10 15
|
||||
* 5: 26 23 21 17 15 10
|
||||
*
|
||||
* Fix things up by averaging out the off-trace clusters; resulting in:
|
||||
*
|
||||
* node 0 1 2 3 4 5
|
||||
* 0: 10 15 17 24 24 24
|
||||
* 1: 15 10 15 24 24 24
|
||||
* 2: 17 15 10 24 24 24
|
||||
* 3: 24 24 24 10 15 17
|
||||
* 4: 24 24 24 15 10 15
|
||||
* 5: 24 24 24 17 15 10
|
||||
*/
|
||||
static int slit_cluster_distance(int i, int j)
|
||||
{
|
||||
static int slit_valid = -1;
|
||||
int u = topology_num_nodes_per_package();
|
||||
long d = 0;
|
||||
int x, y;
|
||||
|
||||
if (slit_valid < 0) {
|
||||
slit_valid = slit_validate();
|
||||
if (!slit_valid)
|
||||
pr_err(FW_BUG "SLIT table doesn't have the expected form for SNC -- fixup disabled!\n");
|
||||
else
|
||||
pr_info("Fixing up SNC SLIT table.\n");
|
||||
}
|
||||
|
||||
/*
|
||||
* Is this a unit cluster on the trace?
|
||||
*/
|
||||
if ((i / u) == (j / u) || !slit_valid)
|
||||
return node_distance(i, j);
|
||||
|
||||
/*
|
||||
* Off-trace cluster.
|
||||
*
|
||||
* Notably average out the symmetric pair of off-trace clusters to
|
||||
* ensure the resulting SLIT table is symmetric.
|
||||
*/
|
||||
x = i - (i % u);
|
||||
y = j - (j % u);
|
||||
|
||||
for (i = x; i < x + u; i++) {
|
||||
for (j = y; j < y + u; j++) {
|
||||
d += node_distance(i, j);
|
||||
d += node_distance(j, i);
|
||||
}
|
||||
}
|
||||
|
||||
return d / (2*u*u);
|
||||
}
|
||||
|
||||
int arch_sched_node_distance(int from, int to)
|
||||
|
|
@ -542,34 +658,14 @@ int arch_sched_node_distance(int from, int to)
|
|||
switch (boot_cpu_data.x86_vfm) {
|
||||
case INTEL_GRANITERAPIDS_X:
|
||||
case INTEL_ATOM_DARKMONT_X:
|
||||
|
||||
if (topology_max_packages() == 1 || topology_num_nodes_per_package() == 1 ||
|
||||
d < REMOTE_DISTANCE)
|
||||
if (topology_max_packages() == 1 ||
|
||||
topology_num_nodes_per_package() < 3)
|
||||
return d;
|
||||
|
||||
/*
|
||||
* With SNC enabled, there could be too many levels of remote
|
||||
* NUMA node distances, creating NUMA domain levels
|
||||
* including local nodes and partial remote nodes.
|
||||
*
|
||||
* Trim finer distance tuning for NUMA nodes in remote package
|
||||
* for the purpose of building sched domains. Group NUMA nodes
|
||||
* 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.
|
||||
* Handle SNC-3 asymmetries.
|
||||
*/
|
||||
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 slit_cluster_distance(from, to);
|
||||
}
|
||||
return d;
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue