hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICA1GK -------------------------------- Implement the bpf prog for the 'uptime' interface. Signed-off-by: GONG Ruiqi <gongruiqi1@huawei.com> --- kernel/bpf-rvi/generic_single_iter.c | 1 + samples/bpf/Makefile | 1 + samples/bpf/bpf_rvi_uptime.bpf.c | 122 +++++++++++++++++++++++++++ 3 files changed, 124 insertions(+) create mode 100644 samples/bpf/bpf_rvi_uptime.bpf.c diff --git a/kernel/bpf-rvi/generic_single_iter.c b/kernel/bpf-rvi/generic_single_iter.c index 542241aed597..88ced6d8fabd 100644 --- a/kernel/bpf-rvi/generic_single_iter.c +++ b/kernel/bpf-rvi/generic_single_iter.c @@ -51,6 +51,7 @@ static const struct seq_operations generic_single_seq_ops = { * Users of "generic_single" iter type: * - cpu_online * - loadavg + * - uptime */ DEFINE_BPF_ITER_FUNC(generic_single, struct bpf_iter_meta *meta) diff --git a/samples/bpf/Makefile b/samples/bpf/Makefile index b7afc09514d9..4aeb40711241 100644 --- a/samples/bpf/Makefile +++ b/samples/bpf/Makefile @@ -159,6 +159,7 @@ always-$(CONFIG_BPF_RVI) += bpf_rvi_cpu_online.bpf.o always-$(CONFIG_BPF_RVI) += bpf_rvi_diskstats.bpf.o always-$(CONFIG_BPF_RVI) += bpf_rvi_partitions.bpf.o always-$(CONFIG_BPF_RVI) += bpf_rvi_loadavg.bpf.o +always-$(CONFIG_BPF_RVI) += bpf_rvi_uptime.bpf.o ifeq ($(ARCH), arm) # Strip all except -D__LINUX_ARM_ARCH__ option needed to handle linux diff --git a/samples/bpf/bpf_rvi_uptime.bpf.c b/samples/bpf/bpf_rvi_uptime.bpf.c new file mode 100644 index 000000000000..2c6d539476b0 --- /dev/null +++ b/samples/bpf/bpf_rvi_uptime.bpf.c @@ -0,0 +1,122 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2025 Huawei Technologies Co., Ltd */ +#include <vmlinux.h> +#include <bpf/bpf_helpers.h> + +#define NSEC_PER_SEC 1000000000L + +void bpf_rcu_read_lock(void) __ksym; +void bpf_rcu_read_unlock(void) __ksym; +void bpf_task_release(struct task_struct *p) __ksym; +struct task_struct *bpf_current_level1_reaper(void) __ksym; +struct cpuset *bpf_cpuset_from_task(struct task_struct *p) __ksym; +void cgroup_rstat_flush_atomic(struct cgroup *cgrp) __ksym; +unsigned int bpf_cpumask_weight(struct cpumask *pmask) __ksym; +u64 bpf_task_ca_cpuusage(struct task_struct *p) __ksym; + +char _license[] SEC("license") = "GPL"; + +static int task_effective_cpus_num(struct task_struct *reaper) +{ + struct cpuset *cpuset; + + cpuset = bpf_cpuset_from_task(reaper); + if (!cpuset) + return -1; + + return bpf_cpumask_weight(cpuset->effective_cpus); +} + +/* cpuacct.usage of cgroup v1. See cpuusage_read(). */ +static int cgroup_v1_get_cpuusage(struct task_struct *reaper, u64 *usage) +{ + *usage = bpf_task_ca_cpuusage(reaper); + return 0; +} + +/* "usage_usec" of cpu.stat of cgroup v2. See cgroup_base_stat_cputime_show(). */ +static int cgroup_v2_get_cpuusage(struct task_struct *reaper, u64 *usage) +{ + struct cgroup *cgroup; + + cgroup = reaper->cgroups->dfl_cgrp; + if (!cgroup) + return -1; + cgroup_rstat_flush_atomic(cgroup); + *usage = cgroup->bstat.cputime.sum_exec_runtime; + + return 0; +} + +/* + * What LXCFS uses is cpuacct.usage, which, as LXCFS's code comment says, might overestimate the + * container's busy time if the container doesn't have its own cpuacct cgroup. + */ +static int cgroup_get_cpuusage(struct task_struct *reaper, u64 *usage) +{ + int err; + + bpf_rcu_read_lock(); + if (reaper->cgroups->dfl_cgrp) + err = cgroup_v2_get_cpuusage(reaper, usage); + else + err = cgroup_v1_get_cpuusage(reaper, usage); + bpf_rcu_read_unlock(); + return err; +} + +#define RET_OK 0 +#define RET_FAIL 1 +#define RET_SKIP -1 + +SEC("iter/generic_single") +s64 dump_uptime(struct bpf_iter__generic_single *ctx) +{ + struct seq_file *m = ctx->meta->seq; + struct task_struct *reaper; + u64 cur_timestamp; + u64 runtime, totaltime, idletime = 0, cpuusage = 0; + u64 run_sec, run_nsec, idle_sec, idle_nsec; + unsigned int cpu_count; + int err, ret = RET_FAIL; + + reaper = bpf_current_level1_reaper(); + if (!reaper) + return RET_FAIL; + err = cgroup_get_cpuusage(reaper, &cpuusage); + if (err) + goto err; + cpu_count = task_effective_cpus_num(reaper); + if (cpu_count == -1) + goto err; + + cur_timestamp = bpf_ktime_get_boot_ns(); // "ns": nanosecond, not namespace + /* + * LXCFS takes the 22nd column from /proc/<pid>/stat, which is task->start_boottime (with + * transforming unit and adding timens bias). See do_task_stat() for details. + */ + if (cur_timestamp < reaper->start_boottime) + runtime = 0; + else + runtime = cur_timestamp - reaper->start_boottime; + /* + * As implemented in uptime_proc_show(), idle time of the original /proc/uptime is the sum + * of each cpu's idle time. Here we calculate it the other way around: subtract the total + * amount of cpu time by cpu usage. + */ + totaltime = runtime * cpu_count; + if (totaltime > cpuusage) + idletime = totaltime - cpuusage; + + run_sec = runtime / NSEC_PER_SEC; + run_nsec = runtime % NSEC_PER_SEC; + idle_sec = idletime / NSEC_PER_SEC; + idle_nsec = idletime % NSEC_PER_SEC; + BPF_SEQ_PRINTF(m, "%llu.%02llu %llu.%02llu\n", run_sec, run_nsec / (NSEC_PER_SEC / 100), + idle_sec, idle_nsec / (NSEC_PER_SEC / 100)); + + ret = RET_OK; +err: + bpf_task_release(reaper); + return ret; +} -- 2.25.1