0001
0002
0003 #include <argp.h>
0004 #include <stdio.h>
0005 #include <stdlib.h>
0006 #include <string.h>
0007 #include <time.h>
0008 #include <bpf/libbpf.h>
0009 #include <bpf/bpf.h>
0010 #include "runqslower.h"
0011 #include "runqslower.skel.h"
0012
0013 struct env {
0014 pid_t pid;
0015 __u64 min_us;
0016 bool verbose;
0017 } env = {
0018 .min_us = 10000,
0019 };
0020
0021 const char *argp_program_version = "runqslower 0.1";
0022 const char *argp_program_bug_address = "<bpf@vger.kernel.org>";
0023 const char argp_program_doc[] =
0024 "runqslower Trace long process scheduling delays.\n"
0025 " For Linux, uses eBPF, BPF CO-RE, libbpf, BTF.\n"
0026 "\n"
0027 "This script traces high scheduling delays between tasks being\n"
0028 "ready to run and them running on CPU after that.\n"
0029 "\n"
0030 "USAGE: runqslower [-p PID] [min_us]\n"
0031 "\n"
0032 "EXAMPLES:\n"
0033 " runqslower # trace run queue latency higher than 10000 us (default)\n"
0034 " runqslower 1000 # trace run queue latency higher than 1000 us\n"
0035 " runqslower -p 123 # trace pid 123 only\n";
0036
0037 static const struct argp_option opts[] = {
0038 { "pid", 'p', "PID", 0, "Process PID to trace"},
0039 { "verbose", 'v', NULL, 0, "Verbose debug output" },
0040 {},
0041 };
0042
0043 static error_t parse_arg(int key, char *arg, struct argp_state *state)
0044 {
0045 static int pos_args;
0046 int pid;
0047 long long min_us;
0048
0049 switch (key) {
0050 case 'v':
0051 env.verbose = true;
0052 break;
0053 case 'p':
0054 errno = 0;
0055 pid = strtol(arg, NULL, 10);
0056 if (errno || pid <= 0) {
0057 fprintf(stderr, "Invalid PID: %s\n", arg);
0058 argp_usage(state);
0059 }
0060 env.pid = pid;
0061 break;
0062 case ARGP_KEY_ARG:
0063 if (pos_args++) {
0064 fprintf(stderr,
0065 "Unrecognized positional argument: %s\n", arg);
0066 argp_usage(state);
0067 }
0068 errno = 0;
0069 min_us = strtoll(arg, NULL, 10);
0070 if (errno || min_us <= 0) {
0071 fprintf(stderr, "Invalid delay (in us): %s\n", arg);
0072 argp_usage(state);
0073 }
0074 env.min_us = min_us;
0075 break;
0076 default:
0077 return ARGP_ERR_UNKNOWN;
0078 }
0079 return 0;
0080 }
0081
0082 int libbpf_print_fn(enum libbpf_print_level level,
0083 const char *format, va_list args)
0084 {
0085 if (level == LIBBPF_DEBUG && !env.verbose)
0086 return 0;
0087 return vfprintf(stderr, format, args);
0088 }
0089
0090 void handle_event(void *ctx, int cpu, void *data, __u32 data_sz)
0091 {
0092 const struct runq_event *e = data;
0093 struct tm *tm;
0094 char ts[32];
0095 time_t t;
0096
0097 time(&t);
0098 tm = localtime(&t);
0099 strftime(ts, sizeof(ts), "%H:%M:%S", tm);
0100 printf("%-8s %-16s %-6d %14llu\n", ts, e->task, e->pid, e->delta_us);
0101 }
0102
0103 void handle_lost_events(void *ctx, int cpu, __u64 lost_cnt)
0104 {
0105 printf("Lost %llu events on CPU #%d!\n", lost_cnt, cpu);
0106 }
0107
0108 int main(int argc, char **argv)
0109 {
0110 static const struct argp argp = {
0111 .options = opts,
0112 .parser = parse_arg,
0113 .doc = argp_program_doc,
0114 };
0115 struct perf_buffer *pb = NULL;
0116 struct runqslower_bpf *obj;
0117 int err;
0118
0119 err = argp_parse(&argp, argc, argv, 0, NULL, NULL);
0120 if (err)
0121 return err;
0122
0123 libbpf_set_print(libbpf_print_fn);
0124
0125
0126 libbpf_set_strict_mode(LIBBPF_STRICT_ALL);
0127
0128 obj = runqslower_bpf__open();
0129 if (!obj) {
0130 fprintf(stderr, "failed to open and/or load BPF object\n");
0131 return 1;
0132 }
0133
0134
0135 obj->rodata->targ_pid = env.pid;
0136 obj->rodata->min_us = env.min_us;
0137
0138 err = runqslower_bpf__load(obj);
0139 if (err) {
0140 fprintf(stderr, "failed to load BPF object: %d\n", err);
0141 goto cleanup;
0142 }
0143
0144 err = runqslower_bpf__attach(obj);
0145 if (err) {
0146 fprintf(stderr, "failed to attach BPF programs\n");
0147 goto cleanup;
0148 }
0149
0150 printf("Tracing run queue latency higher than %llu us\n", env.min_us);
0151 printf("%-8s %-16s %-6s %14s\n", "TIME", "COMM", "PID", "LAT(us)");
0152
0153 pb = perf_buffer__new(bpf_map__fd(obj->maps.events), 64,
0154 handle_event, handle_lost_events, NULL, NULL);
0155 err = libbpf_get_error(pb);
0156 if (err) {
0157 pb = NULL;
0158 fprintf(stderr, "failed to open perf buffer: %d\n", err);
0159 goto cleanup;
0160 }
0161
0162 while ((err = perf_buffer__poll(pb, 100)) >= 0)
0163 ;
0164 printf("Error polling perf buffer: %d\n", err);
0165
0166 cleanup:
0167 perf_buffer__free(pb);
0168 runqslower_bpf__destroy(obj);
0169
0170 return err != 0;
0171 }