ktstr 0.5.2

Test harness for Linux process schedulers
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
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275

use anyhow::Result;
use ktstr::assert::AssertResult;
use ktstr::scenario::Ctx;
use ktstr::scenario::ops::{
    CgroupDef, CpusetSpec, HoldSpec, Step, execute_steps, execute_steps_with,
};
use ktstr::test_support::{KtstrTestEntry, NumaDistance, NumaNode, Topology, TopologyConstraints};
use ktstr::workload::{MemPolicy, MpolFlags};

// All NUMA tests use auto_repro: false — these verify topology plumbing,
// not scheduler behavior, so BPF crash probes add no diagnostic value.

// ---------------------------------------------------------------------------
// Multi-NUMA boot: 2 NUMA nodes, uniform memory, default 10/20 distances
// ---------------------------------------------------------------------------

fn scenario_multi_numa_boot(ctx: &Ctx) -> Result<AssertResult> {
    let topo = &ctx.topo;
    assert!(
        topo.num_numa_nodes() >= 2,
        "expected >= 2 NUMA nodes, got {}",
        topo.num_numa_nodes()
    );
    assert_eq!(topo.numa_distance(0, 0), 10);
    assert_eq!(topo.numa_distance(1, 1), 10);
    assert_eq!(topo.numa_distance(0, 1), 20);
    assert_eq!(topo.numa_distance(1, 0), 20);

    for &nid in topo.numa_node_ids() {
        let mi = topo
            .node_meminfo(nid)
            .unwrap_or_else(|| panic!("node {nid} missing meminfo"));
        assert!(mi.total_kb > 0, "node {nid} has zero memory");
    }

    let steps = vec![Step {
        setup: vec![CgroupDef::named("cg_0").workers(ctx.workers_per_cgroup)].into(),
        ops: vec![],
        hold: HoldSpec::FULL,
    }];
    execute_steps(ctx, steps)
}

#[ktstr::__private::linkme::distributed_slice(ktstr::test_support::KTSTR_TESTS)]
#[linkme(crate = ktstr::__private::linkme)]
static __KTSTR_ENTRY_MULTI_NUMA_BOOT: KtstrTestEntry = KtstrTestEntry {
    name: "numa_multi_node_boot",
    func: scenario_multi_numa_boot,
    topology: Topology::new(2, 4, 2, 1),
    constraints: TopologyConstraints {
        min_numa_nodes: 2,
        max_numa_nodes: Some(2),
        ..TopologyConstraints::DEFAULT
    },
    auto_repro: false,
    duration: std::time::Duration::from_secs(3),
    ..KtstrTestEntry::DEFAULT
};

// ---------------------------------------------------------------------------
// CXL memory-only node: node 2 has llcs=0 (no CPUs), only memory.
// Manual distributed_slice: #[ktstr_test] cannot express with_nodes/with_distances.
// ---------------------------------------------------------------------------

static CXL_NODES: [NumaNode; 3] = [
    NumaNode::new(2, 256),
    NumaNode::new(2, 256),
    NumaNode::new(0, 128),
];

static CXL_DIST: NumaDistance = NumaDistance::new(3, &[10, 20, 30, 20, 10, 25, 30, 25, 10]);

fn scenario_cxl_memory_only(ctx: &Ctx) -> Result<AssertResult> {
    let topo = &ctx.topo;
    assert_eq!(topo.num_numa_nodes(), 3, "expected 3 NUMA nodes");

    assert!(topo.is_memory_only(2), "node 2 must be memory-only");
    assert!(!topo.is_memory_only(0), "node 0 has CPUs");
    assert!(!topo.is_memory_only(1), "node 1 has CPUs");

    assert_eq!(topo.numa_distance(0, 2), 30);
    assert_eq!(topo.numa_distance(1, 2), 25);

    let mi = topo.node_meminfo(2).expect("CXL node 2 must have meminfo");
    assert!(mi.total_kb > 0, "CXL node must have memory");

    let steps = vec![Step {
        setup: vec![CgroupDef::named("cg_0").workers(ctx.workers_per_cgroup)].into(),
        ops: vec![],
        hold: HoldSpec::FULL,
    }];
    execute_steps(ctx, steps)
}

#[ktstr::__private::linkme::distributed_slice(ktstr::test_support::KTSTR_TESTS)]
#[linkme(crate = ktstr::__private::linkme)]
static __KTSTR_ENTRY_CXL_MEM_ONLY: KtstrTestEntry = KtstrTestEntry {
    name: "numa_cxl_memory_only_node",
    func: scenario_cxl_memory_only,
    topology: Topology::with_nodes(2, 1, &CXL_NODES).with_distances(&CXL_DIST),
    constraints: TopologyConstraints {
        min_numa_nodes: 3,
        max_numa_nodes: Some(3),
        ..TopologyConstraints::DEFAULT
    },
    memory_mb: 640,
    auto_repro: false,
    duration: std::time::Duration::from_secs(3),
    ..KtstrTestEntry::DEFAULT
};

// ---------------------------------------------------------------------------
// MemPolicy::Bind + min_page_locality assertion
// ---------------------------------------------------------------------------

fn scenario_mempolicy_bind_locality(ctx: &Ctx) -> Result<AssertResult> {
    let checks = ktstr::assert::Assert::default_checks().min_page_locality(0.5);
    let steps = vec![Step {
        setup: vec![
            CgroupDef::named("cg_0")
                .workers(ctx.workers_per_cgroup)
                .with_cpuset(CpusetSpec::Numa(0))
                .mem_policy(MemPolicy::bind([0])),
        ]
        .into(),
        ops: vec![],
        hold: HoldSpec::FULL,
    }];
    execute_steps_with(ctx, steps, Some(&checks))
}

#[ktstr::__private::linkme::distributed_slice(ktstr::test_support::KTSTR_TESTS)]
#[linkme(crate = ktstr::__private::linkme)]
static __KTSTR_ENTRY_MEMPOLICY_BIND: KtstrTestEntry = KtstrTestEntry {
    name: "numa_mempolicy_bind_locality",
    func: scenario_mempolicy_bind_locality,
    topology: Topology::new(2, 4, 2, 1),
    constraints: TopologyConstraints {
        min_numa_nodes: 2,
        max_numa_nodes: Some(2),
        ..TopologyConstraints::DEFAULT
    },
    auto_repro: false,
    duration: std::time::Duration::from_secs(5),
    ..KtstrTestEntry::DEFAULT
};

// ---------------------------------------------------------------------------
// vmstat cross-node migration tracking
// ---------------------------------------------------------------------------

fn scenario_vmstat_migration(ctx: &Ctx) -> Result<AssertResult> {
    let checks = ktstr::assert::Assert::default_checks().max_cross_node_migration_ratio(0.5);
    let steps = vec![Step {
        setup: vec![
            CgroupDef::named("cg_0")
                .workers(ctx.workers_per_cgroup)
                .with_cpuset(CpusetSpec::Numa(0))
                .mem_policy(MemPolicy::bind([0])),
        ]
        .into(),
        ops: vec![],
        hold: HoldSpec::FULL,
    }];
    execute_steps_with(ctx, steps, Some(&checks))
}

#[ktstr::__private::linkme::distributed_slice(ktstr::test_support::KTSTR_TESTS)]
#[linkme(crate = ktstr::__private::linkme)]
static __KTSTR_ENTRY_VMSTAT_MIGRATION: KtstrTestEntry = KtstrTestEntry {
    name: "numa_vmstat_migration_tracking",
    func: scenario_vmstat_migration,
    topology: Topology::new(2, 4, 2, 1),
    constraints: TopologyConstraints {
        min_numa_nodes: 2,
        max_numa_nodes: Some(2),
        ..TopologyConstraints::DEFAULT
    },
    auto_repro: false,
    duration: std::time::Duration::from_secs(5),
    ..KtstrTestEntry::DEFAULT
};

// ---------------------------------------------------------------------------
// MemPolicy::Interleave round-robins allocations across BOTH nodes, so the
// expected page locality on the binding-cpuset half is roughly 50%. The
// scenario pins workers to node 0 via cpuset but interleaves their pages
// across nodes 0 and 1, so the locality assertion sets a low minimum.
// Exercises the round-robin nodemask path that Bind/Preferred don't reach.
//
// `MpolFlags::STATIC_NODES` is required because without it the kernel
// silently intersects the interleave nodemask with the task's cpuset —
// which would degenerate to "interleave across {0} only" and defeat the
// cross-node intent.
// ---------------------------------------------------------------------------

fn scenario_mempolicy_interleave_cross_node(ctx: &Ctx) -> Result<AssertResult> {
    let checks = ktstr::assert::Assert::default_checks().min_page_locality(0.3);
    let steps = vec![Step {
        setup: vec![
            CgroupDef::named("cg_0")
                .workers(ctx.workers_per_cgroup)
                .with_cpuset(CpusetSpec::Numa(0))
                .mem_policy(MemPolicy::interleave([0, 1]))
                .mpol_flags(MpolFlags::STATIC_NODES),
        ]
        .into(),
        ops: vec![],
        hold: HoldSpec::FULL,
    }];
    execute_steps_with(ctx, steps, Some(&checks))
}

#[ktstr::__private::linkme::distributed_slice(ktstr::test_support::KTSTR_TESTS)]
#[linkme(crate = ktstr::__private::linkme)]
static __KTSTR_ENTRY_MEMPOLICY_INTERLEAVE: KtstrTestEntry = KtstrTestEntry {
    name: "numa_mempolicy_interleave_cross_node",
    func: scenario_mempolicy_interleave_cross_node,
    topology: Topology::new(2, 4, 2, 1),
    constraints: TopologyConstraints {
        min_numa_nodes: 2,
        max_numa_nodes: Some(2),
        ..TopologyConstraints::DEFAULT
    },
    auto_repro: false,
    duration: std::time::Duration::from_secs(5),
    ..KtstrTestEntry::DEFAULT
};

// ---------------------------------------------------------------------------
// MemPolicy::PreferredMany lists multiple nodes that the kernel may use for
// allocation, falling back across the set rather than a single fallback path.
// Workers are pinned to node 0; the policy prefers nodes 0 and 1, so most
// pages should land on node 0 with the minority spilling to node 1.
// Exercises the MPOL_PREFERRED_MANY (kernel 5.15+) path that
// `MemPolicy::Preferred` cannot express.
//
// `MpolFlags::STATIC_NODES` is required so node 1 stays in the preferred
// set — without it the kernel narrows the nodemask to the cpuset (node 0
// only), collapsing the PreferredMany semantics into a single-node
// preferred that this test is not exercising.
// ---------------------------------------------------------------------------

fn scenario_mempolicy_preferred_many_locality(ctx: &Ctx) -> Result<AssertResult> {
    let checks = ktstr::assert::Assert::default_checks().min_page_locality(0.5);
    let steps = vec![Step {
        setup: vec![
            CgroupDef::named("cg_0")
                .workers(ctx.workers_per_cgroup)
                .with_cpuset(CpusetSpec::Numa(0))
                .mem_policy(MemPolicy::preferred_many([0, 1]))
                .mpol_flags(MpolFlags::STATIC_NODES),
        ]
        .into(),
        ops: vec![],
        hold: HoldSpec::FULL,
    }];
    execute_steps_with(ctx, steps, Some(&checks))
}

#[ktstr::__private::linkme::distributed_slice(ktstr::test_support::KTSTR_TESTS)]
#[linkme(crate = ktstr::__private::linkme)]
static __KTSTR_ENTRY_MEMPOLICY_PREFERRED_MANY: KtstrTestEntry = KtstrTestEntry {
    name: "numa_mempolicy_preferred_many_locality",
    func: scenario_mempolicy_preferred_many_locality,
    topology: Topology::new(2, 4, 2, 1),
    constraints: TopologyConstraints {
        min_numa_nodes: 2,
        max_numa_nodes: Some(2),
        ..TopologyConstraints::DEFAULT
    },
    auto_repro: false,
    duration: std::time::Duration::from_secs(5),
    ..KtstrTestEntry::DEFAULT
};