hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICA1GK -------------------------------- Implement the bpf prog for the 'stat' interface. Signed-off-by: GONG Ruiqi <gongruiqi1@huawei.com> --- samples/bpf/bpf_rvi_stat.bpf.c | 219 +++++++++++++++++++++++++++++++++ 1 file changed, 219 insertions(+) create mode 100644 samples/bpf/bpf_rvi_stat.bpf.c diff --git a/samples/bpf/bpf_rvi_stat.bpf.c b/samples/bpf/bpf_rvi_stat.bpf.c new file mode 100644 index 000000000000..c0b5fd9d3eab --- /dev/null +++ b/samples/bpf/bpf_rvi_stat.bpf.c @@ -0,0 +1,219 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2025 Huawei Technologies Co., Ltd */ +#include <vmlinux.h> +#include <bpf/bpf_helpers.h> +#include <bpf/bpf_tracing.h> + +#define READ_ONCE(x) (*(volatile typeof(x) *)&(x)) + +unsigned int bpf_kstat_softirqs_cpu(unsigned int irq, int cpu) __ksym; +unsigned long bpf_kstat_cpu_irqs_sum(unsigned int cpu) __ksym; +void bpf_kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu) __ksym; +u64 bpf_get_idle_time(struct kernel_cpustat *kcs, int cpu) __ksym; +u64 bpf_get_iowait_time(struct kernel_cpustat *kcs, int cpu) __ksym; +struct task_struct *bpf_current_level1_reaper(void) __ksym; +void bpf_task_release(struct task_struct *p) __ksym; +struct cpumask *bpf_task_allowed_cpus(struct task_struct *p) __ksym; +u32 bpf_cpumask_next_idx(int n, const struct cpumask *mask) __ksym; +u64 bpf_cpuacct_stat_from_task(struct task_struct *p, int cpu, enum cpuacct_stat_index idx) __ksym; +void bpf_cpuacct_kcpustat_cpu_fetch(struct kernel_cpustat *dst, struct cpuacct *ca, int cpu) __ksym; +void bpf_seq_file_append(struct seq_file *dst, struct seq_file *src) __ksym; +void bpf_get_boottime_timens(struct task_struct *tsk, struct timespec64 *boottime) __ksym; +unsigned long bpf_get_total_forks(void) __ksym; +unsigned int bpf_nr_running(void) __ksym; +unsigned long long bpf_nr_context_switches(void) __ksym; +unsigned int bpf_nr_iowait(void) __ksym; +void bpf_show_all_irqs(struct seq_file *p) __ksym; + +#define NSEC_PER_SEC 1000000000L +#define USER_HZ 100 +/* Take the `#if (NSEC_PER_SEC % USER_HZ) == 0` case */ +static inline u64 nsec_to_clock_t(u64 x) +{ + return x / (NSEC_PER_SEC / USER_HZ); +} + +struct stat_sum_data { + u64 user, nice, system, idle, iowait, irq, softirq, steal; + u64 guest, guest_nice; + u64 sum; + u64 sum_softirq; + unsigned int per_softirq_sums[NR_SOFTIRQS]; + u64 nr_context_switches, nr_running, nr_iowait; +}; + +struct stat_sum_data_map { + __uint(type, BPF_MAP_TYPE_TASK_STORAGE); + __uint(map_flags, BPF_F_NO_PREALLOC); + __type(key, int); + __type(value, struct stat_sum_data); +} collect_map SEC(".maps"); + +#define RET_OK 0 +#define RET_FAIL 1 + +SEC("iter/stat") +s64 dump_stat(struct bpf_iter__stat *ctx) +{ + struct seq_file *m = ctx->meta->seq; + u64 cpuid = ctx->cpuid; + struct cpuacct *cpuacct = ctx->cpuacct; + bool print_all = ctx->print_all; + struct seq_file *seqf_pcpu = ctx->seqf_pcpu; + struct task_struct *current = bpf_get_current_task_btf(); // just for bpf map management + struct stat_sum_data *collect; + int j; + + collect = bpf_task_storage_get(&collect_map, current, NULL, BPF_LOCAL_STORAGE_GET_F_CREATE); + if (!collect) + return RET_FAIL; + + if (!print_all) { + u64 user, nice, system, idle, iowait, irq, softirq, steal; + u64 guest, guest_nice; + struct kernel_cpustat kcpustat = {}; + u64 *cpustat = kcpustat.cpustat; + + bpf_kcpustat_cpu_fetch(&kcpustat, cpuid); + user = cpustat[CPUTIME_USER]; + nice = cpustat[CPUTIME_NICE]; + system = cpustat[CPUTIME_SYSTEM]; + idle = bpf_get_idle_time(&kcpustat, cpuid); + iowait = bpf_get_iowait_time(&kcpustat, cpuid); + irq = cpustat[CPUTIME_IRQ]; + softirq = cpustat[CPUTIME_SOFTIRQ]; + steal = cpustat[CPUTIME_STEAL]; + guest = cpustat[CPUTIME_GUEST]; + guest_nice = cpustat[CPUTIME_GUEST_NICE]; + + collect->sum += bpf_kstat_cpu_irqs_sum(cpuid); + collect->sum += ctx->arch_irq_stat_cpu; + for (j = 0; j < NR_SOFTIRQS; j++) { + unsigned int softirq_stat = bpf_kstat_softirqs_cpu(j, cpuid); + + collect->per_softirq_sums[j] += softirq_stat; + collect->sum_softirq += softirq_stat; + } + + // don't print cpuid to avoid leaking host info + BPF_SEQ_PRINTF(seqf_pcpu, "cpu%d", ctx->meta->seq_num); + + if (cpuacct) { + struct kernel_cpustat kcpustat = {}; + u64 *cpustat = kcpustat.cpustat; + + bpf_cpuacct_kcpustat_cpu_fetch(&kcpustat, cpuacct, cpuid); + + user = cpustat[CPUTIME_USER]; + nice = cpustat[CPUTIME_NICE]; + system = cpustat[CPUTIME_SYSTEM]; + irq = cpustat[CPUTIME_IRQ]; + softirq = cpustat[CPUTIME_SOFTIRQ]; + idle = cpustat[CPUTIME_IDLE]; + iowait = cpustat[CPUTIME_IOWAIT]; + steal = cpustat[CPUTIME_STEAL]; + guest = cpustat[CPUTIME_GUEST]; + guest_nice = cpustat[CPUTIME_GUEST_NICE]; + + collect->user += user; + collect->nice += nice; + collect->system += system; + collect->idle += idle; + collect->iowait += iowait; + collect->irq += irq; + collect->softirq += softirq; + collect->steal += steal; + collect->guest += guest; + collect->guest_nice += guest_nice; + + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(user)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(nice)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(system)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(idle)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(iowait)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(irq)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(softirq)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(steal)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(guest)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(guest_nice)); + } else { + collect->user += user; + collect->nice += nice; + collect->system += system; + collect->idle += idle; + collect->iowait += iowait; + collect->irq += irq; + collect->softirq += softirq; + collect->steal += steal; + collect->guest += guest; + collect->guest_nice += guest_nice; + + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(user)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(nice)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(system)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(idle)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(iowait)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(irq)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(softirq)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(steal)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(guest)); + BPF_SEQ_PRINTF(seqf_pcpu, " %ld", nsec_to_clock_t(guest_nice)); + } + + BPF_SEQ_PRINTF(seqf_pcpu, "\n"); + } else { + struct timespec64 boottime; + + // Add only once + collect->sum += ctx->arch_irq_stat; + + BPF_SEQ_PRINTF(m, "cpu %ld", nsec_to_clock_t(collect->user)); + BPF_SEQ_PRINTF(m, " %ld", nsec_to_clock_t(collect->nice)); + BPF_SEQ_PRINTF(m, " %ld", nsec_to_clock_t(collect->system)); + BPF_SEQ_PRINTF(m, " %ld", nsec_to_clock_t(collect->idle)); + BPF_SEQ_PRINTF(m, " %ld", nsec_to_clock_t(collect->iowait)); + BPF_SEQ_PRINTF(m, " %ld", nsec_to_clock_t(collect->irq)); + BPF_SEQ_PRINTF(m, " %ld", nsec_to_clock_t(collect->softirq)); + BPF_SEQ_PRINTF(m, " %ld", nsec_to_clock_t(collect->steal)); + BPF_SEQ_PRINTF(m, " %ld", nsec_to_clock_t(collect->guest)); + BPF_SEQ_PRINTF(m, " %ld", nsec_to_clock_t(collect->guest_nice)); + BPF_SEQ_PRINTF(m, "\n"); + + // ************************************ + // Dump percpu printing + // Don't do this: + // BPF_SEQ_PRINTF(m, "%s", seqf_pcpu->buf); + // as it prints at most 512 bytes each time + bpf_seq_file_append(m, seqf_pcpu); + // ************************************ + + BPF_SEQ_PRINTF(m, "intr %ld\n", collect->sum); + + bpf_show_all_irqs(m); + + bpf_get_boottime_timens(current, &boottime); + BPF_SEQ_PRINTF(m, + "\nctxt %llu\n" + "btime %llu\n" + "processes %lu\n" + "procs_running %u\n" + "procs_blocked %u\n", + bpf_nr_context_switches(), + (unsigned long long)boottime.tv_sec, + bpf_get_total_forks(), + bpf_nr_running(), + bpf_nr_iowait()); + + BPF_SEQ_PRINTF(m, "softirq %ld", collect->sum_softirq); + + for (j = 0; j < NR_SOFTIRQS; j++) + BPF_SEQ_PRINTF(m, " %d", collect->per_softirq_sums[j]); + BPF_SEQ_PRINTF(m, "\n"); + + bpf_task_storage_delete(&collect_map, current); + } + + return RET_OK; +} + +char _license[] SEC("license") = "GPL"; -- 2.25.1