1.. SPDX-License-Identifier: GPL-2.0
2
3.. _cpumasks-header-label:
4
5==================
6BPF cpumask kfuncs
7==================
8
91. Introduction
10===============
11
12``struct cpumask`` is a bitmap data structure in the kernel whose indices
13reflect the CPUs on the system. Commonly, cpumasks are used to track which CPUs
14a task is affinitized to, but they can also be used to e.g. track which cores
15are associated with a scheduling domain, which cores on a machine are idle,
16etc.
17
18BPF provides programs with a set of :ref:`kfuncs-header-label` that can be
19used to allocate, mutate, query, and free cpumasks.
20
212. BPF cpumask objects
22======================
23
24There are two different types of cpumasks that can be used by BPF programs.
25
262.1 ``struct bpf_cpumask *``
27----------------------------
28
29``struct bpf_cpumask *`` is a cpumask that is allocated by BPF, on behalf of a
30BPF program, and whose lifecycle is entirely controlled by BPF. These cpumasks
31are RCU-protected, can be mutated, can be used as kptrs, and can be safely cast
32to a ``struct cpumask *``.
33
342.1.1 ``struct bpf_cpumask *`` lifecycle
35----------------------------------------
36
37A ``struct bpf_cpumask *`` is allocated, acquired, and released, using the
38following functions:
39
40.. kernel-doc:: kernel/bpf/cpumask.c
41  :identifiers: bpf_cpumask_create
42
43.. kernel-doc:: kernel/bpf/cpumask.c
44  :identifiers: bpf_cpumask_acquire
45
46.. kernel-doc:: kernel/bpf/cpumask.c
47  :identifiers: bpf_cpumask_release
48
49For example:
50
51.. code-block:: c
52
53        struct cpumask_map_value {
54                struct bpf_cpumask __kptr * cpumask;
55        };
56
57        struct array_map {
58                __uint(type, BPF_MAP_TYPE_ARRAY);
59                __type(key, int);
60                __type(value, struct cpumask_map_value);
61                __uint(max_entries, 65536);
62        } cpumask_map SEC(".maps");
63
64        static int cpumask_map_insert(struct bpf_cpumask *mask, u32 pid)
65        {
66                struct cpumask_map_value local, *v;
67                long status;
68                struct bpf_cpumask *old;
69                u32 key = pid;
70
71                local.cpumask = NULL;
72                status = bpf_map_update_elem(&cpumask_map, &key, &local, 0);
73                if (status) {
74                        bpf_cpumask_release(mask);
75                        return status;
76                }
77
78                v = bpf_map_lookup_elem(&cpumask_map, &key);
79                if (!v) {
80                        bpf_cpumask_release(mask);
81                        return -ENOENT;
82                }
83
84                old = bpf_kptr_xchg(&v->cpumask, mask);
85                if (old)
86                        bpf_cpumask_release(old);
87
88                return 0;
89        }
90
91        /**
92         * A sample tracepoint showing how a task's cpumask can be queried and
93         * recorded as a kptr.
94         */
95        SEC("tp_btf/task_newtask")
96        int BPF_PROG(record_task_cpumask, struct task_struct *task, u64 clone_flags)
97        {
98                struct bpf_cpumask *cpumask;
99                int ret;
100
101                cpumask = bpf_cpumask_create();
102                if (!cpumask)
103                        return -ENOMEM;
104
105                if (!bpf_cpumask_full(task->cpus_ptr))
106                        bpf_printk("task %s has CPU affinity", task->comm);
107
108                bpf_cpumask_copy(cpumask, task->cpus_ptr);
109                return cpumask_map_insert(cpumask, task->pid);
110        }
111
112----
113
1142.1.1 ``struct bpf_cpumask *`` as kptrs
115---------------------------------------
116
117As mentioned and illustrated above, these ``struct bpf_cpumask *`` objects can
118also be stored in a map and used as kptrs. If a ``struct bpf_cpumask *`` is in
119a map, the reference can be removed from the map with bpf_kptr_xchg(), or
120opportunistically acquired using RCU:
121
122.. code-block:: c
123
124	/* struct containing the struct bpf_cpumask kptr which is stored in the map. */
125	struct cpumasks_kfunc_map_value {
126		struct bpf_cpumask __kptr * bpf_cpumask;
127	};
128
129	/* The map containing struct cpumasks_kfunc_map_value entries. */
130	struct {
131		__uint(type, BPF_MAP_TYPE_ARRAY);
132		__type(key, int);
133		__type(value, struct cpumasks_kfunc_map_value);
134		__uint(max_entries, 1);
135	} cpumasks_kfunc_map SEC(".maps");
136
137	/* ... */
138
139	/**
140	 * A simple example tracepoint program showing how a
141	 * struct bpf_cpumask * kptr that is stored in a map can
142	 * be passed to kfuncs using RCU protection.
143	 */
144	SEC("tp_btf/cgroup_mkdir")
145	int BPF_PROG(cgrp_ancestor_example, struct cgroup *cgrp, const char *path)
146	{
147		struct bpf_cpumask *kptr;
148		struct cpumasks_kfunc_map_value *v;
149		u32 key = 0;
150
151		/* Assume a bpf_cpumask * kptr was previously stored in the map. */
152		v = bpf_map_lookup_elem(&cpumasks_kfunc_map, &key);
153		if (!v)
154			return -ENOENT;
155
156		bpf_rcu_read_lock();
157		/* Acquire a reference to the bpf_cpumask * kptr that's already stored in the map. */
158		kptr = v->cpumask;
159		if (!kptr) {
160			/* If no bpf_cpumask was present in the map, it's because
161			 * we're racing with another CPU that removed it with
162			 * bpf_kptr_xchg() between the bpf_map_lookup_elem()
163			 * above, and our load of the pointer from the map.
164			 */
165			bpf_rcu_read_unlock();
166			return -EBUSY;
167		}
168
169		bpf_cpumask_setall(kptr);
170		bpf_rcu_read_unlock();
171
172		return 0;
173	}
174
175----
176
1772.2 ``struct cpumask``
178----------------------
179
180``struct cpumask`` is the object that actually contains the cpumask bitmap
181being queried, mutated, etc. A ``struct bpf_cpumask`` wraps a ``struct
182cpumask``, which is why it's safe to cast it as such (note however that it is
183**not** safe to cast a ``struct cpumask *`` to a ``struct bpf_cpumask *``, and
184the verifier will reject any program that tries to do so).
185
186As we'll see below, any kfunc that mutates its cpumask argument will take a
187``struct bpf_cpumask *`` as that argument. Any argument that simply queries the
188cpumask will instead take a ``struct cpumask *``.
189
1903. cpumask kfuncs
191=================
192
193Above, we described the kfuncs that can be used to allocate, acquire, release,
194etc a ``struct bpf_cpumask *``. This section of the document will describe the
195kfuncs for mutating and querying cpumasks.
196
1973.1 Mutating cpumasks
198---------------------
199
200Some cpumask kfuncs are "read-only" in that they don't mutate any of their
201arguments, whereas others mutate at least one argument (which means that the
202argument must be a ``struct bpf_cpumask *``, as described above).
203
204This section will describe all of the cpumask kfuncs which mutate at least one
205argument. :ref:`cpumasks-querying-label` below describes the read-only kfuncs.
206
2073.1.1 Setting and clearing CPUs
208-------------------------------
209
210bpf_cpumask_set_cpu() and bpf_cpumask_clear_cpu() can be used to set and clear
211a CPU in a ``struct bpf_cpumask`` respectively:
212
213.. kernel-doc:: kernel/bpf/cpumask.c
214   :identifiers: bpf_cpumask_set_cpu bpf_cpumask_clear_cpu
215
216These kfuncs are pretty straightforward, and can be used, for example, as
217follows:
218
219.. code-block:: c
220
221        /**
222         * A sample tracepoint showing how a cpumask can be queried.
223         */
224        SEC("tp_btf/task_newtask")
225        int BPF_PROG(test_set_clear_cpu, struct task_struct *task, u64 clone_flags)
226        {
227                struct bpf_cpumask *cpumask;
228
229                cpumask = bpf_cpumask_create();
230                if (!cpumask)
231                        return -ENOMEM;
232
233                bpf_cpumask_set_cpu(0, cpumask);
234                if (!bpf_cpumask_test_cpu(0, cast(cpumask)))
235                        /* Should never happen. */
236                        goto release_exit;
237
238                bpf_cpumask_clear_cpu(0, cpumask);
239                if (bpf_cpumask_test_cpu(0, cast(cpumask)))
240                        /* Should never happen. */
241                        goto release_exit;
242
243                /* struct cpumask * pointers such as task->cpus_ptr can also be queried. */
244                if (bpf_cpumask_test_cpu(0, task->cpus_ptr))
245                        bpf_printk("task %s can use CPU %d", task->comm, 0);
246
247        release_exit:
248                bpf_cpumask_release(cpumask);
249                return 0;
250        }
251
252----
253
254bpf_cpumask_test_and_set_cpu() and bpf_cpumask_test_and_clear_cpu() are
255complementary kfuncs that allow callers to atomically test and set (or clear)
256CPUs:
257
258.. kernel-doc:: kernel/bpf/cpumask.c
259   :identifiers: bpf_cpumask_test_and_set_cpu bpf_cpumask_test_and_clear_cpu
260
261----
262
263We can also set and clear entire ``struct bpf_cpumask *`` objects in one
264operation using bpf_cpumask_setall() and bpf_cpumask_clear():
265
266.. kernel-doc:: kernel/bpf/cpumask.c
267   :identifiers: bpf_cpumask_setall bpf_cpumask_clear
268
2693.1.2 Operations between cpumasks
270---------------------------------
271
272In addition to setting and clearing individual CPUs in a single cpumask,
273callers can also perform bitwise operations between multiple cpumasks using
274bpf_cpumask_and(), bpf_cpumask_or(), and bpf_cpumask_xor():
275
276.. kernel-doc:: kernel/bpf/cpumask.c
277   :identifiers: bpf_cpumask_and bpf_cpumask_or bpf_cpumask_xor
278
279The following is an example of how they may be used. Note that some of the
280kfuncs shown in this example will be covered in more detail below.
281
282.. code-block:: c
283
284        /**
285         * A sample tracepoint showing how a cpumask can be mutated using
286           bitwise operators (and queried).
287         */
288        SEC("tp_btf/task_newtask")
289        int BPF_PROG(test_and_or_xor, struct task_struct *task, u64 clone_flags)
290        {
291                struct bpf_cpumask *mask1, *mask2, *dst1, *dst2;
292
293                mask1 = bpf_cpumask_create();
294                if (!mask1)
295                        return -ENOMEM;
296
297                mask2 = bpf_cpumask_create();
298                if (!mask2) {
299                        bpf_cpumask_release(mask1);
300                        return -ENOMEM;
301                }
302
303                // ...Safely create the other two masks... */
304
305                bpf_cpumask_set_cpu(0, mask1);
306                bpf_cpumask_set_cpu(1, mask2);
307                bpf_cpumask_and(dst1, (const struct cpumask *)mask1, (const struct cpumask *)mask2);
308                if (!bpf_cpumask_empty((const struct cpumask *)dst1))
309                        /* Should never happen. */
310                        goto release_exit;
311
312                bpf_cpumask_or(dst1, (const struct cpumask *)mask1, (const struct cpumask *)mask2);
313                if (!bpf_cpumask_test_cpu(0, (const struct cpumask *)dst1))
314                        /* Should never happen. */
315                        goto release_exit;
316
317                if (!bpf_cpumask_test_cpu(1, (const struct cpumask *)dst1))
318                        /* Should never happen. */
319                        goto release_exit;
320
321                bpf_cpumask_xor(dst2, (const struct cpumask *)mask1, (const struct cpumask *)mask2);
322                if (!bpf_cpumask_equal((const struct cpumask *)dst1,
323                                       (const struct cpumask *)dst2))
324                        /* Should never happen. */
325                        goto release_exit;
326
327         release_exit:
328                bpf_cpumask_release(mask1);
329                bpf_cpumask_release(mask2);
330                bpf_cpumask_release(dst1);
331                bpf_cpumask_release(dst2);
332                return 0;
333        }
334
335----
336
337The contents of an entire cpumask may be copied to another using
338bpf_cpumask_copy():
339
340.. kernel-doc:: kernel/bpf/cpumask.c
341   :identifiers: bpf_cpumask_copy
342
343----
344
345.. _cpumasks-querying-label:
346
3473.2 Querying cpumasks
348---------------------
349
350In addition to the above kfuncs, there is also a set of read-only kfuncs that
351can be used to query the contents of cpumasks.
352
353.. kernel-doc:: kernel/bpf/cpumask.c
354   :identifiers: bpf_cpumask_first bpf_cpumask_first_zero bpf_cpumask_first_and
355                 bpf_cpumask_test_cpu
356
357.. kernel-doc:: kernel/bpf/cpumask.c
358   :identifiers: bpf_cpumask_equal bpf_cpumask_intersects bpf_cpumask_subset
359                 bpf_cpumask_empty bpf_cpumask_full
360
361.. kernel-doc:: kernel/bpf/cpumask.c
362   :identifiers: bpf_cpumask_any_distribute bpf_cpumask_any_and_distribute
363
364----
365
366Some example usages of these querying kfuncs were shown above. We will not
367replicate those exmaples here. Note, however, that all of the aforementioned
368kfuncs are tested in `tools/testing/selftests/bpf/progs/cpumask_success.c`_, so
369please take a look there if you're looking for more examples of how they can be
370used.
371
372.. _tools/testing/selftests/bpf/progs/cpumask_success.c:
373   https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/tools/testing/selftests/bpf/progs/cpumask_success.c
374
375
3764. Adding BPF cpumask kfuncs
377============================
378
379The set of supported BPF cpumask kfuncs are not (yet) a 1-1 match with the
380cpumask operations in include/linux/cpumask.h. Any of those cpumask operations
381could easily be encapsulated in a new kfunc if and when required. If you'd like
382to support a new cpumask operation, please feel free to submit a patch. If you
383do add a new cpumask kfunc, please document it here, and add any relevant
384selftest testcases to the cpumask selftest suite.
385