Arnaldo Carvalho de Melo | 514f1c6 | 2012-09-26 20:05:56 -0300 | [diff] [blame^] | 1 | #include "builtin.h" |
| 2 | #include "util/evlist.h" |
| 3 | #include "util/parse-options.h" |
| 4 | #include "util/thread_map.h" |
| 5 | #include "event-parse.h" |
| 6 | |
| 7 | #include <libaudit.h> |
| 8 | #include <stdlib.h> |
| 9 | |
| 10 | static struct syscall_fmt { |
| 11 | const char *name; |
| 12 | bool errmsg; |
| 13 | bool timeout; |
| 14 | } syscall_fmts[] = { |
| 15 | { .name = "futex", .errmsg = true, }, |
| 16 | { .name = "poll", .errmsg = true, .timeout = true, }, |
| 17 | { .name = "ppoll", .errmsg = true, .timeout = true, }, |
| 18 | { .name = "read", .errmsg = true, }, |
| 19 | { .name = "recvfrom", .errmsg = true, }, |
| 20 | { .name = "select", .errmsg = true, .timeout = true, }, |
| 21 | }; |
| 22 | |
| 23 | static int syscall_fmt__cmp(const void *name, const void *fmtp) |
| 24 | { |
| 25 | const struct syscall_fmt *fmt = fmtp; |
| 26 | return strcmp(name, fmt->name); |
| 27 | } |
| 28 | |
| 29 | static struct syscall_fmt *syscall_fmt__find(const char *name) |
| 30 | { |
| 31 | const int nmemb = ARRAY_SIZE(syscall_fmts); |
| 32 | return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp); |
| 33 | } |
| 34 | |
| 35 | struct syscall { |
| 36 | struct event_format *tp_format; |
| 37 | const char *name; |
| 38 | struct syscall_fmt *fmt; |
| 39 | }; |
| 40 | |
| 41 | struct trace { |
| 42 | int audit_machine; |
| 43 | struct { |
| 44 | int max; |
| 45 | struct syscall *table; |
| 46 | } syscalls; |
| 47 | struct perf_record_opts opts; |
| 48 | }; |
| 49 | |
| 50 | static int trace__read_syscall_info(struct trace *trace, int id) |
| 51 | { |
| 52 | char tp_name[128]; |
| 53 | struct syscall *sc; |
| 54 | |
| 55 | if (id > trace->syscalls.max) { |
| 56 | struct syscall *nsyscalls = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc)); |
| 57 | |
| 58 | if (nsyscalls == NULL) |
| 59 | return -1; |
| 60 | |
| 61 | if (trace->syscalls.max != -1) { |
| 62 | memset(nsyscalls + trace->syscalls.max + 1, 0, |
| 63 | (id - trace->syscalls.max) * sizeof(*sc)); |
| 64 | } else { |
| 65 | memset(nsyscalls, 0, (id + 1) * sizeof(*sc)); |
| 66 | } |
| 67 | |
| 68 | trace->syscalls.table = nsyscalls; |
| 69 | trace->syscalls.max = id; |
| 70 | } |
| 71 | |
| 72 | sc = trace->syscalls.table + id; |
| 73 | sc->name = audit_syscall_to_name(id, trace->audit_machine); |
| 74 | if (sc->name == NULL) |
| 75 | return -1; |
| 76 | |
| 77 | snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name); |
| 78 | |
| 79 | sc->tp_format = event_format__new("syscalls", tp_name); |
| 80 | sc->fmt = syscall_fmt__find(sc->name); |
| 81 | |
| 82 | return sc->tp_format != NULL ? 0 : -1; |
| 83 | } |
| 84 | |
| 85 | static size_t syscall__fprintf_args(struct syscall *sc, unsigned long *args, FILE *fp) |
| 86 | { |
| 87 | int i = 0; |
| 88 | size_t printed = 0; |
| 89 | |
| 90 | if (sc->tp_format != NULL) { |
| 91 | struct format_field *field; |
| 92 | |
| 93 | for (field = sc->tp_format->format.fields->next; field; field = field->next) { |
| 94 | printed += fprintf(fp, "%s%s: %ld", printed ? ", " : "", |
| 95 | field->name, args[i++]); |
| 96 | } |
| 97 | } else { |
| 98 | while (i < 6) { |
| 99 | printed += fprintf(fp, "%sarg%d: %ld", printed ? ", " : "", i, args[i]); |
| 100 | ++i; |
| 101 | } |
| 102 | } |
| 103 | |
| 104 | return printed; |
| 105 | } |
| 106 | |
| 107 | static int trace__run(struct trace *trace) |
| 108 | { |
| 109 | struct perf_evlist *evlist = perf_evlist__new(NULL, NULL); |
| 110 | struct perf_evsel *evsel, *evsel_enter, *evsel_exit; |
| 111 | int err = -1, i, nr_events = 0, before; |
| 112 | |
| 113 | if (evlist == NULL) { |
| 114 | printf("Not enough memory to run!\n"); |
| 115 | goto out; |
| 116 | } |
| 117 | |
| 118 | evsel_enter = perf_evsel__newtp("raw_syscalls", "sys_enter", 0); |
| 119 | if (evsel_enter == NULL) { |
| 120 | printf("Couldn't read the raw_syscalls:sys_enter tracepoint information!\n"); |
| 121 | goto out_delete_evlist; |
| 122 | } |
| 123 | |
| 124 | perf_evlist__add(evlist, evsel_enter); |
| 125 | |
| 126 | evsel_exit = perf_evsel__newtp("raw_syscalls", "sys_exit", 1); |
| 127 | if (evsel_exit == NULL) { |
| 128 | printf("Couldn't read the raw_syscalls:sys_exit tracepoint information!\n"); |
| 129 | goto out_delete_evlist; |
| 130 | } |
| 131 | |
| 132 | perf_evlist__add(evlist, evsel_exit); |
| 133 | |
| 134 | err = perf_evlist__create_maps(evlist, &trace->opts.target); |
| 135 | if (err < 0) { |
| 136 | printf("Problems parsing the target to trace, check your options!\n"); |
| 137 | goto out_delete_evlist; |
| 138 | } |
| 139 | |
| 140 | perf_evlist__config_attrs(evlist, &trace->opts); |
| 141 | |
| 142 | err = perf_evlist__open(evlist); |
| 143 | if (err < 0) { |
| 144 | printf("Couldn't create the events: %s\n", strerror(errno)); |
| 145 | goto out_delete_evlist; |
| 146 | } |
| 147 | |
| 148 | err = perf_evlist__mmap(evlist, UINT_MAX, false); |
| 149 | if (err < 0) { |
| 150 | printf("Couldn't mmap the events: %s\n", strerror(errno)); |
| 151 | goto out_delete_evlist; |
| 152 | } |
| 153 | |
| 154 | perf_evlist__enable(evlist); |
| 155 | again: |
| 156 | before = nr_events; |
| 157 | |
| 158 | for (i = 0; i < evlist->nr_mmaps; i++) { |
| 159 | union perf_event *event; |
| 160 | |
| 161 | while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) { |
| 162 | const u32 type = event->header.type; |
| 163 | struct syscall *sc; |
| 164 | struct perf_sample sample; |
| 165 | int id; |
| 166 | |
| 167 | ++nr_events; |
| 168 | |
| 169 | switch (type) { |
| 170 | case PERF_RECORD_SAMPLE: |
| 171 | break; |
| 172 | case PERF_RECORD_LOST: |
| 173 | printf("LOST %" PRIu64 " events!\n", event->lost.lost); |
| 174 | continue; |
| 175 | default: |
| 176 | printf("Unexpected %s event, skipping...\n", |
| 177 | perf_event__name(type)); |
| 178 | continue; |
| 179 | } |
| 180 | |
| 181 | err = perf_evlist__parse_sample(evlist, event, &sample); |
| 182 | if (err) { |
| 183 | printf("Can't parse sample, err = %d, skipping...\n", err); |
| 184 | continue; |
| 185 | } |
| 186 | |
| 187 | evsel = perf_evlist__id2evsel(evlist, sample.id); |
| 188 | if (evsel == NULL) { |
| 189 | printf("Unknown tp ID %" PRIu64 ", skipping...\n", sample.id); |
| 190 | continue; |
| 191 | } |
| 192 | |
| 193 | id = perf_evsel__intval(evsel, &sample, "id"); |
| 194 | if (id < 0) { |
| 195 | printf("Invalid syscall %d id, skipping...\n", id); |
| 196 | continue; |
| 197 | } |
| 198 | |
| 199 | if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL) && |
| 200 | trace__read_syscall_info(trace, id)) |
| 201 | continue; |
| 202 | |
| 203 | if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL)) |
| 204 | continue; |
| 205 | |
| 206 | sc = &trace->syscalls.table[id]; |
| 207 | |
| 208 | if (evlist->threads->map[0] == -1 || evlist->threads->nr > 1) |
| 209 | printf("%d ", sample.tid); |
| 210 | |
| 211 | if (evsel == evsel_enter) { |
| 212 | void *args = perf_evsel__rawptr(evsel, &sample, "args"); |
| 213 | |
| 214 | printf("%s(", sc->name); |
| 215 | syscall__fprintf_args(sc, args, stdout); |
| 216 | } else if (evsel == evsel_exit) { |
| 217 | int ret = perf_evsel__intval(evsel, &sample, "ret"); |
| 218 | |
| 219 | if (ret < 0 && sc->fmt && sc->fmt->errmsg) { |
| 220 | char bf[256]; |
| 221 | const char *emsg = strerror_r(-ret, bf, sizeof(bf)), |
| 222 | *e = audit_errno_to_name(-ret); |
| 223 | |
| 224 | printf(") = -1 %s %s", e, emsg); |
| 225 | } else if (ret == 0 && sc->fmt && sc->fmt->timeout) |
| 226 | printf(") = 0 Timeout"); |
| 227 | else |
| 228 | printf(") = %d", ret); |
| 229 | |
| 230 | putchar('\n'); |
| 231 | } |
| 232 | } |
| 233 | } |
| 234 | |
| 235 | if (nr_events == before) |
| 236 | poll(evlist->pollfd, evlist->nr_fds, -1); |
| 237 | |
| 238 | goto again; |
| 239 | |
| 240 | out_delete_evlist: |
| 241 | perf_evlist__delete(evlist); |
| 242 | out: |
| 243 | return err; |
| 244 | } |
| 245 | |
| 246 | int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused) |
| 247 | { |
| 248 | const char * const trace_usage[] = { |
| 249 | "perf trace [<options>]", |
| 250 | NULL |
| 251 | }; |
| 252 | struct trace trace = { |
| 253 | .audit_machine = audit_detect_machine(), |
| 254 | .syscalls = { |
| 255 | . max = -1, |
| 256 | }, |
| 257 | .opts = { |
| 258 | .target = { |
| 259 | .uid = UINT_MAX, |
| 260 | .uses_mmap = true, |
| 261 | }, |
| 262 | .user_freq = UINT_MAX, |
| 263 | .user_interval = ULLONG_MAX, |
| 264 | .no_delay = true, |
| 265 | .mmap_pages = 1024, |
| 266 | }, |
| 267 | }; |
| 268 | const struct option trace_options[] = { |
| 269 | OPT_STRING('p', "pid", &trace.opts.target.pid, "pid", |
| 270 | "trace events on existing process id"), |
| 271 | OPT_STRING(0, "tid", &trace.opts.target.tid, "tid", |
| 272 | "trace events on existing thread id"), |
| 273 | OPT_BOOLEAN(0, "all-cpus", &trace.opts.target.system_wide, |
| 274 | "system-wide collection from all CPUs"), |
| 275 | OPT_STRING(0, "cpu", &trace.opts.target.cpu_list, "cpu", |
| 276 | "list of cpus to monitor"), |
| 277 | OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit, |
| 278 | "child tasks do not inherit counters"), |
| 279 | OPT_UINTEGER(0, "mmap-pages", &trace.opts.mmap_pages, |
| 280 | "number of mmap data pages"), |
| 281 | OPT_STRING(0, "uid", &trace.opts.target.uid_str, "user", |
| 282 | "user to profile"), |
| 283 | OPT_END() |
| 284 | }; |
| 285 | int err; |
| 286 | |
| 287 | argc = parse_options(argc, argv, trace_options, trace_usage, 0); |
| 288 | if (argc) |
| 289 | usage_with_options(trace_usage, trace_options); |
| 290 | |
| 291 | err = perf_target__parse_uid(&trace.opts.target); |
| 292 | if (err) { |
| 293 | char bf[BUFSIZ]; |
| 294 | perf_target__strerror(&trace.opts.target, err, bf, sizeof(bf)); |
| 295 | printf("%s", bf); |
| 296 | return err; |
| 297 | } |
| 298 | |
| 299 | return trace__run(&trace); |
| 300 | } |