1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
|
From c4b023618252ad8c03b7ae2cc411718af285bf66 Mon Sep 17 00:00:00 2001
From: Robin Murphy <robin.murphy@arm.com>
Date: Fri, 3 Dec 2021 11:44:51 +0000
Subject: [PATCH 03/14] perf/arm-cmn: Account for NUMA affinity
On a system with multiple CMN meshes, ideally we'd want to access each
PMU from within its own mesh, rather than with a long CML round-trip,
wherever feasible. Since such a system is likely to be presented as
multiple NUMA nodes, let's also hope a proximity domain is specified
for each CMN programming interface, and use that to guide our choice
of IRQ affinity to favour a node-local CPU where possible.
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/32438b0d016e0649d882d47d30ac2000484287b9.1638530442.git.robin.murphy@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
Upstream-Status: Backport [https://lore.kernel.org/r/32438b0d016e0649d882d47d30ac2000484287b9.1638530442.git.robin.murphy@arm.com]
Signed-off-by: Rupinderjit Singh <rupinderjit.singh@arm.com>
---
drivers/perf/arm-cmn.c | 51 +++++++++++++++++++++++++++++++-----------
1 file changed, 38 insertions(+), 13 deletions(-)
diff --git a/drivers/perf/arm-cmn.c b/drivers/perf/arm-cmn.c
index fe7f3e945481..2146d1c0103f 100644
--- a/drivers/perf/arm-cmn.c
+++ b/drivers/perf/arm-cmn.c
@@ -1147,23 +1147,47 @@ static int arm_cmn_commit_txn(struct pmu *pmu)
return 0;
}
-static int arm_cmn_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+static void arm_cmn_migrate(struct arm_cmn *cmn, unsigned int cpu)
+{
+ unsigned int i;
+
+ perf_pmu_migrate_context(&cmn->pmu, cmn->cpu, cpu);
+ for (i = 0; i < cmn->num_dtcs; i++)
+ irq_set_affinity(cmn->dtc[i].irq, cpumask_of(cpu));
+ cmn->cpu = cpu;
+}
+
+static int arm_cmn_pmu_online_cpu(unsigned int cpu, struct hlist_node *cpuhp_node)
{
struct arm_cmn *cmn;
- unsigned int i, target;
+ int node;
- cmn = hlist_entry_safe(node, struct arm_cmn, cpuhp_node);
- if (cpu != cmn->cpu)
- return 0;
+ cmn = hlist_entry_safe(cpuhp_node, struct arm_cmn, cpuhp_node);
+ node = dev_to_node(cmn->dev);
+ if (node != NUMA_NO_NODE && cpu_to_node(cmn->cpu) != node && cpu_to_node(cpu) == node)
+ arm_cmn_migrate(cmn, cpu);
+ return 0;
+}
+
+static int arm_cmn_pmu_offline_cpu(unsigned int cpu, struct hlist_node *cpuhp_node)
+{
+ struct arm_cmn *cmn;
+ unsigned int target;
+ int node;
+ cpumask_t mask;
- target = cpumask_any_but(cpu_online_mask, cpu);
- if (target >= nr_cpu_ids)
+ cmn = hlist_entry_safe(cpuhp_node, struct arm_cmn, cpuhp_node);
+ if (cpu != cmn->cpu)
return 0;
- perf_pmu_migrate_context(&cmn->pmu, cpu, target);
- for (i = 0; i < cmn->num_dtcs; i++)
- irq_set_affinity(cmn->dtc[i].irq, cpumask_of(target));
- cmn->cpu = target;
+ node = dev_to_node(cmn->dev);
+ if (cpumask_and(&mask, cpumask_of_node(node), cpu_online_mask) &&
+ cpumask_andnot(&mask, &mask, cpumask_of(cpu)))
+ target = cpumask_any(&mask);
+ else
+ target = cpumask_any_but(cpu_online_mask, cpu);
+ if (target < nr_cpu_ids)
+ arm_cmn_migrate(cmn, target);
return 0;
}
@@ -1532,7 +1556,7 @@ static int arm_cmn_probe(struct platform_device *pdev)
if (err)
return err;
- cmn->cpu = raw_smp_processor_id();
+ cmn->cpu = cpumask_local_spread(0, dev_to_node(cmn->dev));
cmn->pmu = (struct pmu) {
.module = THIS_MODULE,
.attr_groups = arm_cmn_attr_groups,
@@ -1608,7 +1632,8 @@ static int __init arm_cmn_init(void)
int ret;
ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
- "perf/arm/cmn:online", NULL,
+ "perf/arm/cmn:online",
+ arm_cmn_pmu_online_cpu,
arm_cmn_pmu_offline_cpu);
if (ret < 0)
return ret;
--
2.25.1
|
