new file mode 100644
@@ -0,0 +1,137 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2023 Yafang Shao <laoar.shao@gmail.com> */
+
+#include <test_progs.h>
+#include <bpf/libbpf.h>
+#include "cgroup_helpers.h"
+#include "test_for_each_cpu.skel.h"
+
+static void verify_percpu_psi_value(struct test_for_each_cpu *skel, int fd, __u32 running, int res)
+{
+ DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts);
+ union bpf_iter_link_info linfo;
+ int len, iter_fd, result;
+ struct bpf_link *link;
+ static char buf[128];
+ __u32 nr_running;
+ size_t left;
+ char *p;
+
+ memset(&linfo, 0, sizeof(linfo));
+ linfo.cgroup.cgroup_fd = fd;
+ linfo.cgroup.order = BPF_CGROUP_ITER_SELF_ONLY;
+ opts.link_info = &linfo;
+ opts.link_info_len = sizeof(linfo);
+
+ link = bpf_program__attach_iter(skel->progs.psi_cgroup, &opts);
+ if (!ASSERT_OK_PTR(link, "attach_iter"))
+ return;
+
+ iter_fd = bpf_iter_create(bpf_link__fd(link));
+ if (!ASSERT_GE(iter_fd, 0, "iter_fd"))
+ goto free_link;
+
+ memset(buf, 0, sizeof(buf));
+ left = ARRAY_SIZE(buf);
+ p = buf;
+ while ((len = read(iter_fd, p, left)) > 0) {
+ p += len;
+ left -= len;
+ }
+
+ ASSERT_EQ(sscanf(buf, "nr_running %u ret %d\n", &nr_running, &result), 2, "seq_format");
+ ASSERT_EQ(result, res, "for_each_cpu_result");
+ if (running)
+ ASSERT_GE(nr_running, running, "nr_running");
+ else
+ ASSERT_EQ(nr_running, running, "nr_running");
+
+ /* read() after iter finishes should be ok. */
+ if (len == 0)
+ ASSERT_OK(read(iter_fd, buf, sizeof(buf)), "second_read");
+ close(iter_fd);
+free_link:
+ bpf_link__destroy(link);
+}
+
+void test_root_cgroup(struct test_for_each_cpu *skel)
+{
+ int cgrp_fd, nr_cpus;
+
+ cgrp_fd = get_root_cgroup();
+ if (!ASSERT_GE(cgrp_fd, 0, "create cgrp"))
+ return;
+
+ skel->bss->cpu_mask = CPU_MASK_POSSIBLE;
+ skel->bss->pid = 0;
+ nr_cpus = bpf_num_possible_cpus();
+ /* At least current is running */
+ verify_percpu_psi_value(skel, cgrp_fd, 1, nr_cpus);
+ close(cgrp_fd);
+}
+
+void test_child_cgroup(struct test_for_each_cpu *skel)
+{
+ int cgrp_fd, nr_cpus;
+
+ cgrp_fd = create_and_get_cgroup("for_each_cpu");
+ if (!ASSERT_GE(cgrp_fd, 0, "create cgrp"))
+ return;
+
+ skel->bss->cpu_mask = CPU_MASK_POSSIBLE;
+ skel->bss->pid = 0;
+ nr_cpus = bpf_num_possible_cpus();
+ /* No tasks in the cgroup */
+ verify_percpu_psi_value(skel, cgrp_fd, 0, nr_cpus);
+ close(cgrp_fd);
+ remove_cgroup("for_each_cpu");
+}
+
+void verify_invalid_cpumask(struct test_for_each_cpu *skel, int fd, __u32 cpumask, __u32 pid)
+{
+ DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts);
+
+ skel->bss->cpu_mask = cpumask;
+ skel->bss->pid = pid;
+ verify_percpu_psi_value(skel, fd, 0, -EINVAL);
+}
+
+void test_invalid_cpumask(struct test_for_each_cpu *skel)
+{
+ int cgrp_fd;
+
+ cgrp_fd = create_and_get_cgroup("for_each_cpu");
+ if (!ASSERT_GE(cgrp_fd, 0, "create cgrp"))
+ return;
+
+ verify_invalid_cpumask(skel, cgrp_fd, CPU_MASK_POSSIBLE, 1);
+ verify_invalid_cpumask(skel, cgrp_fd, CPU_MASK_PRESENT, 1);
+ verify_invalid_cpumask(skel, cgrp_fd, CPU_MASK_ONLINE, 1);
+ verify_invalid_cpumask(skel, cgrp_fd, CPU_MASK_TASK, 0);
+ verify_invalid_cpumask(skel, cgrp_fd, -1, 0);
+ verify_invalid_cpumask(skel, cgrp_fd, -1, 1);
+ close(cgrp_fd);
+ remove_cgroup("for_each_cpu");
+}
+
+void test_for_each_cpu(void)
+{
+ struct test_for_each_cpu *skel = NULL;
+
+ skel = test_for_each_cpu__open_and_load();
+ if (!ASSERT_OK_PTR(skel, "test_for_each_cpu__open_and_load"))
+ return;
+
+ if (setup_cgroup_environment())
+ return;
+
+ if (test__start_subtest("psi_system"))
+ test_root_cgroup(skel);
+ if (test__start_subtest("psi_cgroup"))
+ test_child_cgroup(skel);
+ if (test__start_subtest("invalid_cpumask"))
+ test_invalid_cpumask(skel);
+
+ test_for_each_cpu__destroy(skel);
+ cleanup_cgroup_environment();
+}
new file mode 100644
@@ -0,0 +1,63 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2023 Yafang Shao <laoar.shao@gmail.com> */
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+#define __percpu __attribute__((btf_type_tag("percpu")))
+
+enum bpf_cpu_mask_type cpu_mask;
+__u32 pid;
+
+struct callback_ctx {
+ __u32 nr_running;
+ __u32 id;
+};
+
+static uint64_t cgroup_id(struct cgroup *cgrp)
+{
+ return cgrp->kn->id;
+}
+
+static int callback(__u32 cpu, void *ctx, const void *ptr)
+{
+ unsigned int tasks[NR_PSI_TASK_COUNTS];
+ const struct psi_group_cpu *groupc = ptr;
+ struct callback_ctx *data = ctx;
+
+ bpf_probe_read_kernel(&tasks, sizeof(tasks), &groupc->tasks);
+ data->nr_running += tasks[NR_RUNNING];
+ return 0;
+}
+
+SEC("iter.s/cgroup")
+int BPF_PROG(psi_cgroup, struct bpf_iter_meta *meta, struct cgroup *cgrp)
+{
+ struct seq_file *seq = (struct seq_file *)meta->seq;
+ struct psi_group_cpu __percpu *pcpu_ptr;
+ struct callback_ctx data;
+ struct psi_group *psi;
+ __u64 cg_id;
+ int ret;
+
+ cg_id = cgrp ? cgroup_id(cgrp) : 0;
+ if (!cg_id)
+ return 1;
+
+ psi = cgrp->psi;
+ if (!psi)
+ return 1;
+
+ pcpu_ptr = psi->pcpu;
+ if (!pcpu_ptr)
+ return 1;
+
+ data.nr_running = 0;
+ data.id = cg_id;
+ ret = bpf_for_each_cpu(callback, &data, pcpu_ptr, cpu_mask, pid);
+ BPF_SEQ_PRINTF(seq, "nr_running %d ret %d\n", data.nr_running, ret);
+
+ return ret ? 1 : 0;
+}
+
+char _license[] SEC("license") = "GPL";
Add selftest for the new for_each_cpu helper. The result: $ tools/testing/selftests/bpf/test_progs --name=for_each_cpu #84/1 for_each_cpu/psi_system:OK #84/2 for_each_cpu/psi_cgroup:OK #84/3 for_each_cpu/invalid_cpumask:OK #84 for_each_cpu:OK Summary: 1/3 PASSED, 0 SKIPPED, 0 FAILED Signed-off-by: Yafang Shao <laoar.shao@gmail.com> --- .../selftests/bpf/prog_tests/for_each_cpu.c | 137 +++++++++++++++++++++ .../selftests/bpf/progs/test_for_each_cpu.c | 63 ++++++++++ 2 files changed, 200 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/for_each_cpu.c create mode 100644 tools/testing/selftests/bpf/progs/test_for_each_cpu.c