]>
Commit | Line | Data |
---|---|---|
acddc0ed | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
718e3744 | 2 | /* Thread management routine |
3 | * Copyright (C) 1998, 2000 Kunihiro Ishiguro <kunihiro@zebra.org> | |
718e3744 | 4 | */ |
5 | ||
6 | /* #define DEBUG */ | |
7 | ||
8 | #include <zebra.h> | |
308d14ae | 9 | #include <sys/resource.h> |
718e3744 | 10 | |
11 | #include "thread.h" | |
12 | #include "memory.h" | |
3e41733f | 13 | #include "frrcu.h" |
718e3744 | 14 | #include "log.h" |
e04ab74d | 15 | #include "hash.h" |
16 | #include "command.h" | |
05c447dd | 17 | #include "sigevent.h" |
3bf2673b | 18 | #include "network.h" |
bd74dc61 | 19 | #include "jhash.h" |
fbcac826 | 20 | #include "frratomic.h" |
00dffa8c | 21 | #include "frr_pthread.h" |
9ef9495e | 22 | #include "lib_errors.h" |
912d45a1 | 23 | #include "libfrr_trace.h" |
1a9f340b | 24 | #include "libfrr.h" |
d6be5fb9 | 25 | |
bf8d3d6a DL |
26 | DEFINE_MTYPE_STATIC(LIB, THREAD, "Thread"); |
27 | DEFINE_MTYPE_STATIC(LIB, THREAD_MASTER, "Thread master"); | |
28 | DEFINE_MTYPE_STATIC(LIB, THREAD_POLL, "Thread Poll Info"); | |
29 | DEFINE_MTYPE_STATIC(LIB, THREAD_STATS, "Thread stats"); | |
4a1ab8e4 | 30 | |
960b9a53 | 31 | DECLARE_LIST(thread_list, struct thread, threaditem); |
c284542b | 32 | |
aea25d1e MS |
33 | struct cancel_req { |
34 | int flags; | |
35 | struct thread *thread; | |
36 | void *eventobj; | |
37 | struct thread **threadref; | |
38 | }; | |
39 | ||
40 | /* Flags for task cancellation */ | |
41 | #define THREAD_CANCEL_FLAG_READY 0x01 | |
42 | ||
27d29ced DL |
43 | static int thread_timer_cmp(const struct thread *a, const struct thread *b) |
44 | { | |
45 | if (a->u.sands.tv_sec < b->u.sands.tv_sec) | |
46 | return -1; | |
47 | if (a->u.sands.tv_sec > b->u.sands.tv_sec) | |
48 | return 1; | |
49 | if (a->u.sands.tv_usec < b->u.sands.tv_usec) | |
50 | return -1; | |
51 | if (a->u.sands.tv_usec > b->u.sands.tv_usec) | |
52 | return 1; | |
53 | return 0; | |
54 | } | |
55 | ||
960b9a53 | 56 | DECLARE_HEAP(thread_timer_list, struct thread, timeritem, thread_timer_cmp); |
27d29ced | 57 | |
3b96b781 HT |
58 | #if defined(__APPLE__) |
59 | #include <mach/mach.h> | |
60 | #include <mach/mach_time.h> | |
61 | #endif | |
62 | ||
d62a17ae | 63 | #define AWAKEN(m) \ |
64 | do { \ | |
2b64873d | 65 | const unsigned char wakebyte = 0x01; \ |
d62a17ae | 66 | write(m->io_pipe[1], &wakebyte, 1); \ |
67 | } while (0); | |
3bf2673b | 68 | |
62f44022 | 69 | /* control variable for initializer */ |
c17faa4b | 70 | static pthread_once_t init_once = PTHREAD_ONCE_INIT; |
e0bebc7c | 71 | pthread_key_t thread_current; |
6b0655a2 | 72 | |
c17faa4b | 73 | static pthread_mutex_t masters_mtx = PTHREAD_MUTEX_INITIALIZER; |
62f44022 QY |
74 | static struct list *masters; |
75 | ||
6655966d | 76 | static void thread_free(struct thread_master *master, struct thread *thread); |
6b0655a2 | 77 | |
45f01188 DL |
78 | #ifndef EXCLUDE_CPU_TIME |
79 | #define EXCLUDE_CPU_TIME 0 | |
80 | #endif | |
81 | #ifndef CONSUMED_TIME_CHECK | |
82 | #define CONSUMED_TIME_CHECK 0 | |
83 | #endif | |
84 | ||
85 | bool cputime_enabled = !EXCLUDE_CPU_TIME; | |
86 | unsigned long cputime_threshold = CONSUMED_TIME_CHECK; | |
87 | unsigned long walltime_threshold = CONSUMED_TIME_CHECK; | |
88 | ||
62f44022 | 89 | /* CLI start ---------------------------------------------------------------- */ |
45f01188 | 90 | #include "lib/thread_clippy.c" |
45f01188 | 91 | |
d8b87afe | 92 | static unsigned int cpu_record_hash_key(const struct cpu_thread_history *a) |
e04ab74d | 93 | { |
883cc51d | 94 | int size = sizeof(a->func); |
bd74dc61 DS |
95 | |
96 | return jhash(&a->func, size, 0); | |
e04ab74d | 97 | } |
98 | ||
74df8d6d | 99 | static bool cpu_record_hash_cmp(const struct cpu_thread_history *a, |
d62a17ae | 100 | const struct cpu_thread_history *b) |
e04ab74d | 101 | { |
d62a17ae | 102 | return a->func == b->func; |
e04ab74d | 103 | } |
104 | ||
d62a17ae | 105 | static void *cpu_record_hash_alloc(struct cpu_thread_history *a) |
e04ab74d | 106 | { |
d62a17ae | 107 | struct cpu_thread_history *new; |
108 | new = XCALLOC(MTYPE_THREAD_STATS, sizeof(struct cpu_thread_history)); | |
109 | new->func = a->func; | |
110 | new->funcname = a->funcname; | |
111 | return new; | |
e04ab74d | 112 | } |
113 | ||
d62a17ae | 114 | static void cpu_record_hash_free(void *a) |
228da428 | 115 | { |
d62a17ae | 116 | struct cpu_thread_history *hist = a; |
117 | ||
118 | XFREE(MTYPE_THREAD_STATS, hist); | |
228da428 CC |
119 | } |
120 | ||
d62a17ae | 121 | static void vty_out_cpu_thread_history(struct vty *vty, |
122 | struct cpu_thread_history *a) | |
e04ab74d | 123 | { |
1dd08c22 DS |
124 | vty_out(vty, |
125 | "%5zu %10zu.%03zu %9zu %8zu %9zu %8zu %9zu %9zu %9zu %10zu", | |
72327cf3 | 126 | a->total_active, a->cpu.total / 1000, a->cpu.total % 1000, |
9b8e01ca DS |
127 | a->total_calls, (a->cpu.total / a->total_calls), a->cpu.max, |
128 | (a->real.total / a->total_calls), a->real.max, | |
1dd08c22 | 129 | a->total_cpu_warn, a->total_wall_warn, a->total_starv_warn); |
9b8e01ca | 130 | vty_out(vty, " %c%c%c%c%c %s\n", |
d62a17ae | 131 | a->types & (1 << THREAD_READ) ? 'R' : ' ', |
132 | a->types & (1 << THREAD_WRITE) ? 'W' : ' ', | |
133 | a->types & (1 << THREAD_TIMER) ? 'T' : ' ', | |
134 | a->types & (1 << THREAD_EVENT) ? 'E' : ' ', | |
135 | a->types & (1 << THREAD_EXECUTE) ? 'X' : ' ', a->funcname); | |
e04ab74d | 136 | } |
137 | ||
e3b78da8 | 138 | static void cpu_record_hash_print(struct hash_bucket *bucket, void *args[]) |
e04ab74d | 139 | { |
d62a17ae | 140 | struct cpu_thread_history *totals = args[0]; |
fbcac826 | 141 | struct cpu_thread_history copy; |
d62a17ae | 142 | struct vty *vty = args[1]; |
fbcac826 | 143 | uint8_t *filter = args[2]; |
d62a17ae | 144 | |
145 | struct cpu_thread_history *a = bucket->data; | |
146 | ||
fbcac826 QY |
147 | copy.total_active = |
148 | atomic_load_explicit(&a->total_active, memory_order_seq_cst); | |
149 | copy.total_calls = | |
150 | atomic_load_explicit(&a->total_calls, memory_order_seq_cst); | |
9b8e01ca DS |
151 | copy.total_cpu_warn = |
152 | atomic_load_explicit(&a->total_cpu_warn, memory_order_seq_cst); | |
153 | copy.total_wall_warn = | |
154 | atomic_load_explicit(&a->total_wall_warn, memory_order_seq_cst); | |
1dd08c22 DS |
155 | copy.total_starv_warn = atomic_load_explicit(&a->total_starv_warn, |
156 | memory_order_seq_cst); | |
fbcac826 QY |
157 | copy.cpu.total = |
158 | atomic_load_explicit(&a->cpu.total, memory_order_seq_cst); | |
159 | copy.cpu.max = atomic_load_explicit(&a->cpu.max, memory_order_seq_cst); | |
160 | copy.real.total = | |
161 | atomic_load_explicit(&a->real.total, memory_order_seq_cst); | |
162 | copy.real.max = | |
163 | atomic_load_explicit(&a->real.max, memory_order_seq_cst); | |
164 | copy.types = atomic_load_explicit(&a->types, memory_order_seq_cst); | |
165 | copy.funcname = a->funcname; | |
166 | ||
167 | if (!(copy.types & *filter)) | |
d62a17ae | 168 | return; |
fbcac826 QY |
169 | |
170 | vty_out_cpu_thread_history(vty, ©); | |
171 | totals->total_active += copy.total_active; | |
172 | totals->total_calls += copy.total_calls; | |
9b8e01ca DS |
173 | totals->total_cpu_warn += copy.total_cpu_warn; |
174 | totals->total_wall_warn += copy.total_wall_warn; | |
1dd08c22 | 175 | totals->total_starv_warn += copy.total_starv_warn; |
fbcac826 QY |
176 | totals->real.total += copy.real.total; |
177 | if (totals->real.max < copy.real.max) | |
178 | totals->real.max = copy.real.max; | |
179 | totals->cpu.total += copy.cpu.total; | |
180 | if (totals->cpu.max < copy.cpu.max) | |
181 | totals->cpu.max = copy.cpu.max; | |
e04ab74d | 182 | } |
183 | ||
fbcac826 | 184 | static void cpu_record_print(struct vty *vty, uint8_t filter) |
e04ab74d | 185 | { |
d62a17ae | 186 | struct cpu_thread_history tmp; |
187 | void *args[3] = {&tmp, vty, &filter}; | |
188 | struct thread_master *m; | |
189 | struct listnode *ln; | |
190 | ||
45f01188 DL |
191 | if (!cputime_enabled) |
192 | vty_out(vty, | |
193 | "\n" | |
194 | "Collecting CPU time statistics is currently disabled. Following statistics\n" | |
195 | "will be zero or may display data from when collection was enabled. Use the\n" | |
196 | " \"service cputime-stats\" command to start collecting data.\n" | |
197 | "\nCounters and wallclock times are always maintained and should be accurate.\n"); | |
198 | ||
0d6f7fd6 | 199 | memset(&tmp, 0, sizeof(tmp)); |
d62a17ae | 200 | tmp.funcname = "TOTAL"; |
201 | tmp.types = filter; | |
202 | ||
cb1991af | 203 | frr_with_mutex (&masters_mtx) { |
d62a17ae | 204 | for (ALL_LIST_ELEMENTS_RO(masters, ln, m)) { |
205 | const char *name = m->name ? m->name : "main"; | |
206 | ||
207 | char underline[strlen(name) + 1]; | |
208 | memset(underline, '-', sizeof(underline)); | |
4f113d60 | 209 | underline[sizeof(underline) - 1] = '\0'; |
d62a17ae | 210 | |
211 | vty_out(vty, "\n"); | |
212 | vty_out(vty, "Showing statistics for pthread %s\n", | |
213 | name); | |
214 | vty_out(vty, "-------------------------------%s\n", | |
215 | underline); | |
84d951d0 | 216 | vty_out(vty, "%30s %18s %18s\n", "", |
d62a17ae | 217 | "CPU (user+system):", "Real (wall-clock):"); |
218 | vty_out(vty, | |
219 | "Active Runtime(ms) Invoked Avg uSec Max uSecs"); | |
220 | vty_out(vty, " Avg uSec Max uSecs"); | |
1dd08c22 DS |
221 | vty_out(vty, |
222 | " CPU_Warn Wall_Warn Starv_Warn Type Thread\n"); | |
d62a17ae | 223 | |
224 | if (m->cpu_record->count) | |
225 | hash_iterate( | |
226 | m->cpu_record, | |
e3b78da8 | 227 | (void (*)(struct hash_bucket *, |
d62a17ae | 228 | void *))cpu_record_hash_print, |
229 | args); | |
230 | else | |
231 | vty_out(vty, "No data to display yet.\n"); | |
232 | ||
233 | vty_out(vty, "\n"); | |
234 | } | |
235 | } | |
d62a17ae | 236 | |
237 | vty_out(vty, "\n"); | |
238 | vty_out(vty, "Total thread statistics\n"); | |
239 | vty_out(vty, "-------------------------\n"); | |
84d951d0 | 240 | vty_out(vty, "%30s %18s %18s\n", "", |
d62a17ae | 241 | "CPU (user+system):", "Real (wall-clock):"); |
242 | vty_out(vty, "Active Runtime(ms) Invoked Avg uSec Max uSecs"); | |
9b8e01ca | 243 | vty_out(vty, " Avg uSec Max uSecs CPU_Warn Wall_Warn"); |
d62a17ae | 244 | vty_out(vty, " Type Thread\n"); |
245 | ||
246 | if (tmp.total_calls > 0) | |
247 | vty_out_cpu_thread_history(vty, &tmp); | |
e04ab74d | 248 | } |
249 | ||
e3b78da8 | 250 | static void cpu_record_hash_clear(struct hash_bucket *bucket, void *args[]) |
e276eb82 | 251 | { |
fbcac826 | 252 | uint8_t *filter = args[0]; |
d62a17ae | 253 | struct hash *cpu_record = args[1]; |
254 | ||
255 | struct cpu_thread_history *a = bucket->data; | |
62f44022 | 256 | |
d62a17ae | 257 | if (!(a->types & *filter)) |
258 | return; | |
f48f65d2 | 259 | |
d62a17ae | 260 | hash_release(cpu_record, bucket->data); |
e276eb82 PJ |
261 | } |
262 | ||
fbcac826 | 263 | static void cpu_record_clear(uint8_t filter) |
e276eb82 | 264 | { |
fbcac826 | 265 | uint8_t *tmp = &filter; |
d62a17ae | 266 | struct thread_master *m; |
267 | struct listnode *ln; | |
268 | ||
cb1991af | 269 | frr_with_mutex (&masters_mtx) { |
d62a17ae | 270 | for (ALL_LIST_ELEMENTS_RO(masters, ln, m)) { |
cb1991af | 271 | frr_with_mutex (&m->mtx) { |
d62a17ae | 272 | void *args[2] = {tmp, m->cpu_record}; |
273 | hash_iterate( | |
274 | m->cpu_record, | |
e3b78da8 | 275 | (void (*)(struct hash_bucket *, |
d62a17ae | 276 | void *))cpu_record_hash_clear, |
277 | args); | |
278 | } | |
d62a17ae | 279 | } |
280 | } | |
62f44022 QY |
281 | } |
282 | ||
fbcac826 | 283 | static uint8_t parse_filter(const char *filterstr) |
62f44022 | 284 | { |
d62a17ae | 285 | int i = 0; |
286 | int filter = 0; | |
287 | ||
288 | while (filterstr[i] != '\0') { | |
289 | switch (filterstr[i]) { | |
290 | case 'r': | |
291 | case 'R': | |
292 | filter |= (1 << THREAD_READ); | |
293 | break; | |
294 | case 'w': | |
295 | case 'W': | |
296 | filter |= (1 << THREAD_WRITE); | |
297 | break; | |
298 | case 't': | |
299 | case 'T': | |
300 | filter |= (1 << THREAD_TIMER); | |
301 | break; | |
302 | case 'e': | |
303 | case 'E': | |
304 | filter |= (1 << THREAD_EVENT); | |
305 | break; | |
306 | case 'x': | |
307 | case 'X': | |
308 | filter |= (1 << THREAD_EXECUTE); | |
309 | break; | |
310 | default: | |
311 | break; | |
312 | } | |
313 | ++i; | |
314 | } | |
315 | return filter; | |
62f44022 QY |
316 | } |
317 | ||
ee4dcee8 DL |
318 | DEFUN_NOSH (show_thread_cpu, |
319 | show_thread_cpu_cmd, | |
320 | "show thread cpu [FILTER]", | |
321 | SHOW_STR | |
322 | "Thread information\n" | |
323 | "Thread CPU usage\n" | |
324 | "Display filter (rwtex)\n") | |
62f44022 | 325 | { |
fbcac826 | 326 | uint8_t filter = (uint8_t)-1U; |
d62a17ae | 327 | int idx = 0; |
328 | ||
329 | if (argv_find(argv, argc, "FILTER", &idx)) { | |
330 | filter = parse_filter(argv[idx]->arg); | |
331 | if (!filter) { | |
332 | vty_out(vty, | |
3efd0893 | 333 | "Invalid filter \"%s\" specified; must contain at leastone of 'RWTEXB'\n", |
d62a17ae | 334 | argv[idx]->arg); |
335 | return CMD_WARNING; | |
336 | } | |
337 | } | |
338 | ||
339 | cpu_record_print(vty, filter); | |
340 | return CMD_SUCCESS; | |
e276eb82 | 341 | } |
45f01188 DL |
342 | |
343 | DEFPY (service_cputime_stats, | |
344 | service_cputime_stats_cmd, | |
345 | "[no] service cputime-stats", | |
346 | NO_STR | |
347 | "Set up miscellaneous service\n" | |
348 | "Collect CPU usage statistics\n") | |
349 | { | |
350 | cputime_enabled = !no; | |
351 | return CMD_SUCCESS; | |
352 | } | |
353 | ||
354 | DEFPY (service_cputime_warning, | |
355 | service_cputime_warning_cmd, | |
356 | "[no] service cputime-warning (1-4294967295)", | |
357 | NO_STR | |
358 | "Set up miscellaneous service\n" | |
359 | "Warn for tasks exceeding CPU usage threshold\n" | |
360 | "Warning threshold in milliseconds\n") | |
361 | { | |
362 | if (no) | |
363 | cputime_threshold = 0; | |
364 | else | |
365 | cputime_threshold = cputime_warning * 1000; | |
366 | return CMD_SUCCESS; | |
367 | } | |
368 | ||
369 | ALIAS (service_cputime_warning, | |
370 | no_service_cputime_warning_cmd, | |
371 | "no service cputime-warning", | |
372 | NO_STR | |
373 | "Set up miscellaneous service\n" | |
374 | "Warn for tasks exceeding CPU usage threshold\n") | |
375 | ||
376 | DEFPY (service_walltime_warning, | |
377 | service_walltime_warning_cmd, | |
378 | "[no] service walltime-warning (1-4294967295)", | |
379 | NO_STR | |
380 | "Set up miscellaneous service\n" | |
381 | "Warn for tasks exceeding total wallclock threshold\n" | |
382 | "Warning threshold in milliseconds\n") | |
383 | { | |
384 | if (no) | |
385 | walltime_threshold = 0; | |
386 | else | |
387 | walltime_threshold = walltime_warning * 1000; | |
388 | return CMD_SUCCESS; | |
389 | } | |
390 | ||
391 | ALIAS (service_walltime_warning, | |
392 | no_service_walltime_warning_cmd, | |
393 | "no service walltime-warning", | |
394 | NO_STR | |
395 | "Set up miscellaneous service\n" | |
396 | "Warn for tasks exceeding total wallclock threshold\n") | |
e276eb82 | 397 | |
8872626b DS |
398 | static void show_thread_poll_helper(struct vty *vty, struct thread_master *m) |
399 | { | |
400 | const char *name = m->name ? m->name : "main"; | |
401 | char underline[strlen(name) + 1]; | |
a0b36ae6 | 402 | struct thread *thread; |
8872626b DS |
403 | uint32_t i; |
404 | ||
405 | memset(underline, '-', sizeof(underline)); | |
406 | underline[sizeof(underline) - 1] = '\0'; | |
407 | ||
408 | vty_out(vty, "\nShowing poll FD's for %s\n", name); | |
409 | vty_out(vty, "----------------------%s\n", underline); | |
6c19478a DS |
410 | vty_out(vty, "Count: %u/%d\n", (uint32_t)m->handler.pfdcount, |
411 | m->fd_limit); | |
a0b36ae6 DS |
412 | for (i = 0; i < m->handler.pfdcount; i++) { |
413 | vty_out(vty, "\t%6d fd:%6d events:%2d revents:%2d\t\t", i, | |
414 | m->handler.pfds[i].fd, m->handler.pfds[i].events, | |
8872626b | 415 | m->handler.pfds[i].revents); |
a0b36ae6 DS |
416 | |
417 | if (m->handler.pfds[i].events & POLLIN) { | |
418 | thread = m->read[m->handler.pfds[i].fd]; | |
419 | ||
420 | if (!thread) | |
421 | vty_out(vty, "ERROR "); | |
422 | else | |
60a3efec | 423 | vty_out(vty, "%s ", thread->xref->funcname); |
a0b36ae6 DS |
424 | } else |
425 | vty_out(vty, " "); | |
426 | ||
427 | if (m->handler.pfds[i].events & POLLOUT) { | |
428 | thread = m->write[m->handler.pfds[i].fd]; | |
429 | ||
430 | if (!thread) | |
431 | vty_out(vty, "ERROR\n"); | |
432 | else | |
60a3efec | 433 | vty_out(vty, "%s\n", thread->xref->funcname); |
a0b36ae6 DS |
434 | } else |
435 | vty_out(vty, "\n"); | |
436 | } | |
8872626b DS |
437 | } |
438 | ||
ee4dcee8 DL |
439 | DEFUN_NOSH (show_thread_poll, |
440 | show_thread_poll_cmd, | |
441 | "show thread poll", | |
442 | SHOW_STR | |
443 | "Thread information\n" | |
444 | "Show poll FD's and information\n") | |
8872626b DS |
445 | { |
446 | struct listnode *node; | |
447 | struct thread_master *m; | |
448 | ||
cb1991af | 449 | frr_with_mutex (&masters_mtx) { |
8872626b DS |
450 | for (ALL_LIST_ELEMENTS_RO(masters, node, m)) { |
451 | show_thread_poll_helper(vty, m); | |
452 | } | |
453 | } | |
8872626b DS |
454 | |
455 | return CMD_SUCCESS; | |
456 | } | |
457 | ||
458 | ||
49d41a26 DS |
459 | DEFUN (clear_thread_cpu, |
460 | clear_thread_cpu_cmd, | |
461 | "clear thread cpu [FILTER]", | |
62f44022 | 462 | "Clear stored data in all pthreads\n" |
49d41a26 DS |
463 | "Thread information\n" |
464 | "Thread CPU usage\n" | |
465 | "Display filter (rwtexb)\n") | |
e276eb82 | 466 | { |
fbcac826 | 467 | uint8_t filter = (uint8_t)-1U; |
d62a17ae | 468 | int idx = 0; |
469 | ||
470 | if (argv_find(argv, argc, "FILTER", &idx)) { | |
471 | filter = parse_filter(argv[idx]->arg); | |
472 | if (!filter) { | |
473 | vty_out(vty, | |
3efd0893 | 474 | "Invalid filter \"%s\" specified; must contain at leastone of 'RWTEXB'\n", |
d62a17ae | 475 | argv[idx]->arg); |
476 | return CMD_WARNING; | |
477 | } | |
478 | } | |
479 | ||
480 | cpu_record_clear(filter); | |
481 | return CMD_SUCCESS; | |
e276eb82 | 482 | } |
6b0655a2 | 483 | |
22f31b8c DS |
484 | static void show_thread_timers_helper(struct vty *vty, struct thread_master *m) |
485 | { | |
486 | const char *name = m->name ? m->name : "main"; | |
487 | char underline[strlen(name) + 1]; | |
488 | struct thread *thread; | |
489 | ||
490 | memset(underline, '-', sizeof(underline)); | |
491 | underline[sizeof(underline) - 1] = '\0'; | |
492 | ||
493 | vty_out(vty, "\nShowing timers for %s\n", name); | |
494 | vty_out(vty, "-------------------%s\n", underline); | |
495 | ||
496 | frr_each (thread_timer_list, &m->timer, thread) { | |
497 | vty_out(vty, " %-50s%pTH\n", thread->hist->funcname, thread); | |
498 | } | |
499 | } | |
500 | ||
501 | DEFPY_NOSH (show_thread_timers, | |
502 | show_thread_timers_cmd, | |
503 | "show thread timers", | |
504 | SHOW_STR | |
505 | "Thread information\n" | |
506 | "Show all timers and how long they have in the system\n") | |
507 | { | |
508 | struct listnode *node; | |
509 | struct thread_master *m; | |
510 | ||
511 | frr_with_mutex (&masters_mtx) { | |
512 | for (ALL_LIST_ELEMENTS_RO(masters, node, m)) | |
513 | show_thread_timers_helper(vty, m); | |
514 | } | |
515 | ||
516 | return CMD_SUCCESS; | |
517 | } | |
518 | ||
d62a17ae | 519 | void thread_cmd_init(void) |
0b84f294 | 520 | { |
d62a17ae | 521 | install_element(VIEW_NODE, &show_thread_cpu_cmd); |
8872626b | 522 | install_element(VIEW_NODE, &show_thread_poll_cmd); |
d62a17ae | 523 | install_element(ENABLE_NODE, &clear_thread_cpu_cmd); |
45f01188 DL |
524 | |
525 | install_element(CONFIG_NODE, &service_cputime_stats_cmd); | |
526 | install_element(CONFIG_NODE, &service_cputime_warning_cmd); | |
527 | install_element(CONFIG_NODE, &no_service_cputime_warning_cmd); | |
528 | install_element(CONFIG_NODE, &service_walltime_warning_cmd); | |
529 | install_element(CONFIG_NODE, &no_service_walltime_warning_cmd); | |
22f31b8c DS |
530 | |
531 | install_element(VIEW_NODE, &show_thread_timers_cmd); | |
0b84f294 | 532 | } |
62f44022 QY |
533 | /* CLI end ------------------------------------------------------------------ */ |
534 | ||
0b84f294 | 535 | |
d62a17ae | 536 | static void cancelreq_del(void *cr) |
63ccb9cb | 537 | { |
d62a17ae | 538 | XFREE(MTYPE_TMP, cr); |
63ccb9cb QY |
539 | } |
540 | ||
e0bebc7c | 541 | /* initializer, only ever called once */ |
4d762f26 | 542 | static void initializer(void) |
e0bebc7c | 543 | { |
d62a17ae | 544 | pthread_key_create(&thread_current, NULL); |
e0bebc7c QY |
545 | } |
546 | ||
d62a17ae | 547 | struct thread_master *thread_master_create(const char *name) |
718e3744 | 548 | { |
d62a17ae | 549 | struct thread_master *rv; |
550 | struct rlimit limit; | |
551 | ||
552 | pthread_once(&init_once, &initializer); | |
553 | ||
554 | rv = XCALLOC(MTYPE_THREAD_MASTER, sizeof(struct thread_master)); | |
d62a17ae | 555 | |
556 | /* Initialize master mutex */ | |
557 | pthread_mutex_init(&rv->mtx, NULL); | |
558 | pthread_cond_init(&rv->cancel_cond, NULL); | |
559 | ||
560 | /* Set name */ | |
7ffcd8bd QY |
561 | name = name ? name : "default"; |
562 | rv->name = XSTRDUP(MTYPE_THREAD_MASTER, name); | |
d62a17ae | 563 | |
564 | /* Initialize I/O task data structures */ | |
1a9f340b MS |
565 | |
566 | /* Use configured limit if present, ulimit otherwise. */ | |
567 | rv->fd_limit = frr_get_fd_limit(); | |
568 | if (rv->fd_limit == 0) { | |
569 | getrlimit(RLIMIT_NOFILE, &limit); | |
570 | rv->fd_limit = (int)limit.rlim_cur; | |
571 | } | |
572 | ||
a6f235f3 DS |
573 | rv->read = XCALLOC(MTYPE_THREAD_POLL, |
574 | sizeof(struct thread *) * rv->fd_limit); | |
575 | ||
576 | rv->write = XCALLOC(MTYPE_THREAD_POLL, | |
577 | sizeof(struct thread *) * rv->fd_limit); | |
d62a17ae | 578 | |
7ffcd8bd QY |
579 | char tmhashname[strlen(name) + 32]; |
580 | snprintf(tmhashname, sizeof(tmhashname), "%s - threadmaster event hash", | |
581 | name); | |
bd74dc61 | 582 | rv->cpu_record = hash_create_size( |
d8b87afe | 583 | 8, (unsigned int (*)(const void *))cpu_record_hash_key, |
74df8d6d | 584 | (bool (*)(const void *, const void *))cpu_record_hash_cmp, |
7ffcd8bd | 585 | tmhashname); |
d62a17ae | 586 | |
c284542b DL |
587 | thread_list_init(&rv->event); |
588 | thread_list_init(&rv->ready); | |
589 | thread_list_init(&rv->unuse); | |
27d29ced | 590 | thread_timer_list_init(&rv->timer); |
d62a17ae | 591 | |
592 | /* Initialize thread_fetch() settings */ | |
593 | rv->spin = true; | |
594 | rv->handle_signals = true; | |
595 | ||
596 | /* Set pthread owner, should be updated by actual owner */ | |
597 | rv->owner = pthread_self(); | |
598 | rv->cancel_req = list_new(); | |
599 | rv->cancel_req->del = cancelreq_del; | |
600 | rv->canceled = true; | |
601 | ||
602 | /* Initialize pipe poker */ | |
603 | pipe(rv->io_pipe); | |
604 | set_nonblocking(rv->io_pipe[0]); | |
605 | set_nonblocking(rv->io_pipe[1]); | |
606 | ||
607 | /* Initialize data structures for poll() */ | |
608 | rv->handler.pfdsize = rv->fd_limit; | |
609 | rv->handler.pfdcount = 0; | |
610 | rv->handler.pfds = XCALLOC(MTYPE_THREAD_MASTER, | |
611 | sizeof(struct pollfd) * rv->handler.pfdsize); | |
612 | rv->handler.copy = XCALLOC(MTYPE_THREAD_MASTER, | |
613 | sizeof(struct pollfd) * rv->handler.pfdsize); | |
614 | ||
eff09c66 | 615 | /* add to list of threadmasters */ |
cb1991af | 616 | frr_with_mutex (&masters_mtx) { |
eff09c66 QY |
617 | if (!masters) |
618 | masters = list_new(); | |
619 | ||
d62a17ae | 620 | listnode_add(masters, rv); |
621 | } | |
d62a17ae | 622 | |
623 | return rv; | |
718e3744 | 624 | } |
625 | ||
d8a8a8de QY |
626 | void thread_master_set_name(struct thread_master *master, const char *name) |
627 | { | |
cb1991af | 628 | frr_with_mutex (&master->mtx) { |
0a22ddfb | 629 | XFREE(MTYPE_THREAD_MASTER, master->name); |
d8a8a8de QY |
630 | master->name = XSTRDUP(MTYPE_THREAD_MASTER, name); |
631 | } | |
d8a8a8de QY |
632 | } |
633 | ||
6ed04aa2 DS |
634 | #define THREAD_UNUSED_DEPTH 10 |
635 | ||
718e3744 | 636 | /* Move thread to unuse list. */ |
d62a17ae | 637 | static void thread_add_unuse(struct thread_master *m, struct thread *thread) |
718e3744 | 638 | { |
6655966d RZ |
639 | pthread_mutex_t mtxc = thread->mtx; |
640 | ||
d62a17ae | 641 | assert(m != NULL && thread != NULL); |
d62a17ae | 642 | |
d62a17ae | 643 | thread->hist->total_active--; |
6ed04aa2 DS |
644 | memset(thread, 0, sizeof(struct thread)); |
645 | thread->type = THREAD_UNUSED; | |
646 | ||
6655966d RZ |
647 | /* Restore the thread mutex context. */ |
648 | thread->mtx = mtxc; | |
649 | ||
c284542b DL |
650 | if (thread_list_count(&m->unuse) < THREAD_UNUSED_DEPTH) { |
651 | thread_list_add_tail(&m->unuse, thread); | |
6655966d RZ |
652 | return; |
653 | } | |
654 | ||
655 | thread_free(m, thread); | |
718e3744 | 656 | } |
657 | ||
658 | /* Free all unused thread. */ | |
c284542b DL |
659 | static void thread_list_free(struct thread_master *m, |
660 | struct thread_list_head *list) | |
718e3744 | 661 | { |
d62a17ae | 662 | struct thread *t; |
d62a17ae | 663 | |
c284542b | 664 | while ((t = thread_list_pop(list))) |
6655966d | 665 | thread_free(m, t); |
718e3744 | 666 | } |
667 | ||
d62a17ae | 668 | static void thread_array_free(struct thread_master *m, |
669 | struct thread **thread_array) | |
308d14ae | 670 | { |
d62a17ae | 671 | struct thread *t; |
672 | int index; | |
673 | ||
674 | for (index = 0; index < m->fd_limit; ++index) { | |
675 | t = thread_array[index]; | |
676 | if (t) { | |
677 | thread_array[index] = NULL; | |
6655966d | 678 | thread_free(m, t); |
d62a17ae | 679 | } |
680 | } | |
a6f235f3 | 681 | XFREE(MTYPE_THREAD_POLL, thread_array); |
308d14ae DV |
682 | } |
683 | ||
495f0b13 DS |
684 | /* |
685 | * thread_master_free_unused | |
686 | * | |
687 | * As threads are finished with they are put on the | |
688 | * unuse list for later reuse. | |
689 | * If we are shutting down, Free up unused threads | |
690 | * So we can see if we forget to shut anything off | |
691 | */ | |
d62a17ae | 692 | void thread_master_free_unused(struct thread_master *m) |
495f0b13 | 693 | { |
cb1991af | 694 | frr_with_mutex (&m->mtx) { |
d62a17ae | 695 | struct thread *t; |
c284542b | 696 | while ((t = thread_list_pop(&m->unuse))) |
6655966d | 697 | thread_free(m, t); |
d62a17ae | 698 | } |
495f0b13 DS |
699 | } |
700 | ||
718e3744 | 701 | /* Stop thread scheduler. */ |
d62a17ae | 702 | void thread_master_free(struct thread_master *m) |
718e3744 | 703 | { |
27d29ced DL |
704 | struct thread *t; |
705 | ||
cb1991af | 706 | frr_with_mutex (&masters_mtx) { |
d62a17ae | 707 | listnode_delete(masters, m); |
eff09c66 | 708 | if (masters->count == 0) { |
6a154c88 | 709 | list_delete(&masters); |
eff09c66 | 710 | } |
d62a17ae | 711 | } |
d62a17ae | 712 | |
713 | thread_array_free(m, m->read); | |
714 | thread_array_free(m, m->write); | |
27d29ced DL |
715 | while ((t = thread_timer_list_pop(&m->timer))) |
716 | thread_free(m, t); | |
d62a17ae | 717 | thread_list_free(m, &m->event); |
718 | thread_list_free(m, &m->ready); | |
719 | thread_list_free(m, &m->unuse); | |
720 | pthread_mutex_destroy(&m->mtx); | |
33844bbe | 721 | pthread_cond_destroy(&m->cancel_cond); |
d62a17ae | 722 | close(m->io_pipe[0]); |
723 | close(m->io_pipe[1]); | |
6a154c88 | 724 | list_delete(&m->cancel_req); |
1a0a92ea | 725 | m->cancel_req = NULL; |
d62a17ae | 726 | |
727 | hash_clean(m->cpu_record, cpu_record_hash_free); | |
728 | hash_free(m->cpu_record); | |
729 | m->cpu_record = NULL; | |
730 | ||
0a22ddfb | 731 | XFREE(MTYPE_THREAD_MASTER, m->name); |
d62a17ae | 732 | XFREE(MTYPE_THREAD_MASTER, m->handler.pfds); |
733 | XFREE(MTYPE_THREAD_MASTER, m->handler.copy); | |
734 | XFREE(MTYPE_THREAD_MASTER, m); | |
718e3744 | 735 | } |
736 | ||
94202742 | 737 | /* Return remain time in milliseconds. */ |
78ca0342 | 738 | unsigned long thread_timer_remain_msec(struct thread *thread) |
718e3744 | 739 | { |
d62a17ae | 740 | int64_t remain; |
1189d95f | 741 | |
13bcc010 DA |
742 | if (!thread_is_scheduled(thread)) |
743 | return 0; | |
744 | ||
cb1991af | 745 | frr_with_mutex (&thread->mtx) { |
78ca0342 | 746 | remain = monotime_until(&thread->u.sands, NULL) / 1000LL; |
d62a17ae | 747 | } |
1189d95f | 748 | |
d62a17ae | 749 | return remain < 0 ? 0 : remain; |
718e3744 | 750 | } |
751 | ||
78ca0342 CF |
752 | /* Return remain time in seconds. */ |
753 | unsigned long thread_timer_remain_second(struct thread *thread) | |
754 | { | |
755 | return thread_timer_remain_msec(thread) / 1000LL; | |
756 | } | |
757 | ||
d62a17ae | 758 | struct timeval thread_timer_remain(struct thread *thread) |
6ac44687 | 759 | { |
d62a17ae | 760 | struct timeval remain; |
cb1991af | 761 | frr_with_mutex (&thread->mtx) { |
d62a17ae | 762 | monotime_until(&thread->u.sands, &remain); |
763 | } | |
d62a17ae | 764 | return remain; |
6ac44687 CF |
765 | } |
766 | ||
0447957e AK |
767 | static int time_hhmmss(char *buf, int buf_size, long sec) |
768 | { | |
769 | long hh; | |
770 | long mm; | |
771 | int wr; | |
772 | ||
642ac49d | 773 | assert(buf_size >= 8); |
0447957e AK |
774 | |
775 | hh = sec / 3600; | |
776 | sec %= 3600; | |
777 | mm = sec / 60; | |
778 | sec %= 60; | |
779 | ||
780 | wr = snprintf(buf, buf_size, "%02ld:%02ld:%02ld", hh, mm, sec); | |
781 | ||
782 | return wr != 8; | |
783 | } | |
784 | ||
785 | char *thread_timer_to_hhmmss(char *buf, int buf_size, | |
786 | struct thread *t_timer) | |
787 | { | |
788 | if (t_timer) { | |
789 | time_hhmmss(buf, buf_size, | |
790 | thread_timer_remain_second(t_timer)); | |
791 | } else { | |
792 | snprintf(buf, buf_size, "--:--:--"); | |
793 | } | |
794 | return buf; | |
795 | } | |
796 | ||
718e3744 | 797 | /* Get new thread. */ |
d7c0a89a | 798 | static struct thread *thread_get(struct thread_master *m, uint8_t type, |
cc9f21da | 799 | void (*func)(struct thread *), void *arg, |
60a3efec | 800 | const struct xref_threadsched *xref) |
718e3744 | 801 | { |
c284542b | 802 | struct thread *thread = thread_list_pop(&m->unuse); |
d62a17ae | 803 | struct cpu_thread_history tmp; |
804 | ||
805 | if (!thread) { | |
806 | thread = XCALLOC(MTYPE_THREAD, sizeof(struct thread)); | |
807 | /* mutex only needs to be initialized at struct creation. */ | |
808 | pthread_mutex_init(&thread->mtx, NULL); | |
809 | m->alloc++; | |
810 | } | |
811 | ||
812 | thread->type = type; | |
813 | thread->add_type = type; | |
814 | thread->master = m; | |
815 | thread->arg = arg; | |
d62a17ae | 816 | thread->yield = THREAD_YIELD_TIME_SLOT; /* default */ |
817 | thread->ref = NULL; | |
e8b3a2f7 | 818 | thread->ignore_timer_late = false; |
d62a17ae | 819 | |
820 | /* | |
821 | * So if the passed in funcname is not what we have | |
822 | * stored that means the thread->hist needs to be | |
823 | * updated. We keep the last one around in unused | |
824 | * under the assumption that we are probably | |
825 | * going to immediately allocate the same | |
826 | * type of thread. | |
827 | * This hopefully saves us some serious | |
828 | * hash_get lookups. | |
829 | */ | |
60a3efec DL |
830 | if ((thread->xref && thread->xref->funcname != xref->funcname) |
831 | || thread->func != func) { | |
d62a17ae | 832 | tmp.func = func; |
60a3efec | 833 | tmp.funcname = xref->funcname; |
d62a17ae | 834 | thread->hist = |
835 | hash_get(m->cpu_record, &tmp, | |
836 | (void *(*)(void *))cpu_record_hash_alloc); | |
837 | } | |
838 | thread->hist->total_active++; | |
839 | thread->func = func; | |
60a3efec | 840 | thread->xref = xref; |
d62a17ae | 841 | |
842 | return thread; | |
718e3744 | 843 | } |
844 | ||
6655966d RZ |
845 | static void thread_free(struct thread_master *master, struct thread *thread) |
846 | { | |
847 | /* Update statistics. */ | |
848 | assert(master->alloc > 0); | |
849 | master->alloc--; | |
850 | ||
851 | /* Free allocated resources. */ | |
852 | pthread_mutex_destroy(&thread->mtx); | |
853 | XFREE(MTYPE_THREAD, thread); | |
854 | } | |
855 | ||
d81ca9a3 MS |
856 | static int fd_poll(struct thread_master *m, const struct timeval *timer_wait, |
857 | bool *eintr_p) | |
209a72a6 | 858 | { |
d81ca9a3 MS |
859 | sigset_t origsigs; |
860 | unsigned char trash[64]; | |
861 | nfds_t count = m->handler.copycount; | |
862 | ||
d279ef57 DS |
863 | /* |
864 | * If timer_wait is null here, that means poll() should block | |
865 | * indefinitely, unless the thread_master has overridden it by setting | |
d62a17ae | 866 | * ->selectpoll_timeout. |
d279ef57 | 867 | * |
d62a17ae | 868 | * If the value is positive, it specifies the maximum number of |
d279ef57 DS |
869 | * milliseconds to wait. If the timeout is -1, it specifies that |
870 | * we should never wait and always return immediately even if no | |
871 | * event is detected. If the value is zero, the behavior is default. | |
872 | */ | |
d62a17ae | 873 | int timeout = -1; |
874 | ||
875 | /* number of file descriptors with events */ | |
876 | int num; | |
877 | ||
878 | if (timer_wait != NULL | |
879 | && m->selectpoll_timeout == 0) // use the default value | |
880 | timeout = (timer_wait->tv_sec * 1000) | |
881 | + (timer_wait->tv_usec / 1000); | |
882 | else if (m->selectpoll_timeout > 0) // use the user's timeout | |
883 | timeout = m->selectpoll_timeout; | |
884 | else if (m->selectpoll_timeout | |
885 | < 0) // effect a poll (return immediately) | |
886 | timeout = 0; | |
887 | ||
0bdeb5e5 | 888 | zlog_tls_buffer_flush(); |
3e41733f DL |
889 | rcu_read_unlock(); |
890 | rcu_assert_read_unlocked(); | |
891 | ||
d62a17ae | 892 | /* add poll pipe poker */ |
d81ca9a3 MS |
893 | assert(count + 1 < m->handler.pfdsize); |
894 | m->handler.copy[count].fd = m->io_pipe[0]; | |
895 | m->handler.copy[count].events = POLLIN; | |
896 | m->handler.copy[count].revents = 0x00; | |
897 | ||
898 | /* We need to deal with a signal-handling race here: we | |
899 | * don't want to miss a crucial signal, such as SIGTERM or SIGINT, | |
900 | * that may arrive just before we enter poll(). We will block the | |
901 | * key signals, then check whether any have arrived - if so, we return | |
902 | * before calling poll(). If not, we'll re-enable the signals | |
903 | * in the ppoll() call. | |
904 | */ | |
905 | ||
906 | sigemptyset(&origsigs); | |
907 | if (m->handle_signals) { | |
908 | /* Main pthread that handles the app signals */ | |
909 | if (frr_sigevent_check(&origsigs)) { | |
910 | /* Signal to process - restore signal mask and return */ | |
911 | pthread_sigmask(SIG_SETMASK, &origsigs, NULL); | |
912 | num = -1; | |
913 | *eintr_p = true; | |
914 | goto done; | |
915 | } | |
916 | } else { | |
917 | /* Don't make any changes for the non-main pthreads */ | |
918 | pthread_sigmask(SIG_SETMASK, NULL, &origsigs); | |
919 | } | |
d62a17ae | 920 | |
d81ca9a3 MS |
921 | #if defined(HAVE_PPOLL) |
922 | struct timespec ts, *tsp; | |
923 | ||
924 | if (timeout >= 0) { | |
925 | ts.tv_sec = timeout / 1000; | |
926 | ts.tv_nsec = (timeout % 1000) * 1000000; | |
927 | tsp = &ts; | |
928 | } else | |
929 | tsp = NULL; | |
930 | ||
931 | num = ppoll(m->handler.copy, count + 1, tsp, &origsigs); | |
932 | pthread_sigmask(SIG_SETMASK, &origsigs, NULL); | |
933 | #else | |
934 | /* Not ideal - there is a race after we restore the signal mask */ | |
935 | pthread_sigmask(SIG_SETMASK, &origsigs, NULL); | |
936 | num = poll(m->handler.copy, count + 1, timeout); | |
937 | #endif | |
d62a17ae | 938 | |
d81ca9a3 MS |
939 | done: |
940 | ||
941 | if (num < 0 && errno == EINTR) | |
942 | *eintr_p = true; | |
943 | ||
944 | if (num > 0 && m->handler.copy[count].revents != 0 && num--) | |
d62a17ae | 945 | while (read(m->io_pipe[0], &trash, sizeof(trash)) > 0) |
946 | ; | |
947 | ||
3e41733f DL |
948 | rcu_read_lock(); |
949 | ||
d62a17ae | 950 | return num; |
209a72a6 DS |
951 | } |
952 | ||
718e3744 | 953 | /* Add new read thread. */ |
ee1455dd IR |
954 | void _thread_add_read_write(const struct xref_threadsched *xref, |
955 | struct thread_master *m, | |
cc9f21da | 956 | void (*func)(struct thread *), void *arg, int fd, |
ee1455dd | 957 | struct thread **t_ptr) |
718e3744 | 958 | { |
60a3efec | 959 | int dir = xref->thread_type; |
d62a17ae | 960 | struct thread *thread = NULL; |
1ef14bee | 961 | struct thread **thread_array; |
d62a17ae | 962 | |
abf96a87 | 963 | if (dir == THREAD_READ) |
6c3aa850 DL |
964 | frrtrace(9, frr_libfrr, schedule_read, m, |
965 | xref->funcname, xref->xref.file, xref->xref.line, | |
966 | t_ptr, fd, 0, arg, 0); | |
abf96a87 | 967 | else |
6c3aa850 DL |
968 | frrtrace(9, frr_libfrr, schedule_write, m, |
969 | xref->funcname, xref->xref.file, xref->xref.line, | |
970 | t_ptr, fd, 0, arg, 0); | |
abf96a87 | 971 | |
188acbb9 DS |
972 | assert(fd >= 0); |
973 | if (fd >= m->fd_limit) | |
974 | assert(!"Number of FD's open is greater than FRR currently configured to handle, aborting"); | |
975 | ||
cb1991af | 976 | frr_with_mutex (&m->mtx) { |
00dffa8c DL |
977 | if (t_ptr && *t_ptr) |
978 | // thread is already scheduled; don't reschedule | |
979 | break; | |
d62a17ae | 980 | |
981 | /* default to a new pollfd */ | |
982 | nfds_t queuepos = m->handler.pfdcount; | |
983 | ||
1ef14bee DS |
984 | if (dir == THREAD_READ) |
985 | thread_array = m->read; | |
986 | else | |
987 | thread_array = m->write; | |
988 | ||
d62a17ae | 989 | /* if we already have a pollfd for our file descriptor, find and |
990 | * use it */ | |
991 | for (nfds_t i = 0; i < m->handler.pfdcount; i++) | |
992 | if (m->handler.pfds[i].fd == fd) { | |
993 | queuepos = i; | |
1ef14bee DS |
994 | |
995 | #ifdef DEV_BUILD | |
996 | /* | |
997 | * What happens if we have a thread already | |
998 | * created for this event? | |
999 | */ | |
1000 | if (thread_array[fd]) | |
1001 | assert(!"Thread already scheduled for file descriptor"); | |
1002 | #endif | |
d62a17ae | 1003 | break; |
1004 | } | |
1005 | ||
1006 | /* make sure we have room for this fd + pipe poker fd */ | |
1007 | assert(queuepos + 1 < m->handler.pfdsize); | |
1008 | ||
60a3efec | 1009 | thread = thread_get(m, dir, func, arg, xref); |
d62a17ae | 1010 | |
1011 | m->handler.pfds[queuepos].fd = fd; | |
1012 | m->handler.pfds[queuepos].events |= | |
1013 | (dir == THREAD_READ ? POLLIN : POLLOUT); | |
1014 | ||
1015 | if (queuepos == m->handler.pfdcount) | |
1016 | m->handler.pfdcount++; | |
1017 | ||
1018 | if (thread) { | |
cb1991af | 1019 | frr_with_mutex (&thread->mtx) { |
d62a17ae | 1020 | thread->u.fd = fd; |
1ef14bee | 1021 | thread_array[thread->u.fd] = thread; |
d62a17ae | 1022 | } |
d62a17ae | 1023 | |
1024 | if (t_ptr) { | |
1025 | *t_ptr = thread; | |
1026 | thread->ref = t_ptr; | |
1027 | } | |
1028 | } | |
1029 | ||
1030 | AWAKEN(m); | |
1031 | } | |
718e3744 | 1032 | } |
1033 | ||
ee1455dd IR |
1034 | static void _thread_add_timer_timeval(const struct xref_threadsched *xref, |
1035 | struct thread_master *m, | |
cc9f21da | 1036 | void (*func)(struct thread *), void *arg, |
ee1455dd IR |
1037 | struct timeval *time_relative, |
1038 | struct thread **t_ptr) | |
718e3744 | 1039 | { |
d62a17ae | 1040 | struct thread *thread; |
96fe578a | 1041 | struct timeval t; |
d62a17ae | 1042 | |
1043 | assert(m != NULL); | |
1044 | ||
d62a17ae | 1045 | assert(time_relative); |
1046 | ||
6c3aa850 DL |
1047 | frrtrace(9, frr_libfrr, schedule_timer, m, |
1048 | xref->funcname, xref->xref.file, xref->xref.line, | |
c7bb4f00 | 1049 | t_ptr, 0, 0, arg, (long)time_relative->tv_sec); |
abf96a87 | 1050 | |
96fe578a MS |
1051 | /* Compute expiration/deadline time. */ |
1052 | monotime(&t); | |
1053 | timeradd(&t, time_relative, &t); | |
1054 | ||
cb1991af | 1055 | frr_with_mutex (&m->mtx) { |
00dffa8c | 1056 | if (t_ptr && *t_ptr) |
d279ef57 | 1057 | /* thread is already scheduled; don't reschedule */ |
ee1455dd | 1058 | return; |
d62a17ae | 1059 | |
4322dea7 | 1060 | thread = thread_get(m, THREAD_TIMER, func, arg, xref); |
d62a17ae | 1061 | |
cb1991af | 1062 | frr_with_mutex (&thread->mtx) { |
96fe578a | 1063 | thread->u.sands = t; |
27d29ced | 1064 | thread_timer_list_add(&m->timer, thread); |
d62a17ae | 1065 | if (t_ptr) { |
1066 | *t_ptr = thread; | |
1067 | thread->ref = t_ptr; | |
1068 | } | |
1069 | } | |
d62a17ae | 1070 | |
96fe578a MS |
1071 | /* The timer list is sorted - if this new timer |
1072 | * might change the time we'll wait for, give the pthread | |
1073 | * a chance to re-compute. | |
1074 | */ | |
1075 | if (thread_timer_list_first(&m->timer) == thread) | |
1076 | AWAKEN(m); | |
d62a17ae | 1077 | } |
e2eff5c3 DS |
1078 | #define ONEYEAR2SEC (60 * 60 * 24 * 365) |
1079 | if (time_relative->tv_sec > ONEYEAR2SEC) | |
1080 | flog_err( | |
1081 | EC_LIB_TIMER_TOO_LONG, | |
1082 | "Timer: %pTHD is created with an expiration that is greater than 1 year", | |
1083 | thread); | |
9e867fe6 | 1084 | } |
1085 | ||
98c91ac6 | 1086 | |
1087 | /* Add timer event thread. */ | |
ee1455dd | 1088 | void _thread_add_timer(const struct xref_threadsched *xref, |
cc9f21da | 1089 | struct thread_master *m, void (*func)(struct thread *), |
ee1455dd | 1090 | void *arg, long timer, struct thread **t_ptr) |
9e867fe6 | 1091 | { |
d62a17ae | 1092 | struct timeval trel; |
9e867fe6 | 1093 | |
d62a17ae | 1094 | assert(m != NULL); |
9e867fe6 | 1095 | |
d62a17ae | 1096 | trel.tv_sec = timer; |
1097 | trel.tv_usec = 0; | |
9e867fe6 | 1098 | |
ee1455dd | 1099 | _thread_add_timer_timeval(xref, m, func, arg, &trel, t_ptr); |
98c91ac6 | 1100 | } |
9e867fe6 | 1101 | |
98c91ac6 | 1102 | /* Add timer event thread with "millisecond" resolution */ |
ee1455dd IR |
1103 | void _thread_add_timer_msec(const struct xref_threadsched *xref, |
1104 | struct thread_master *m, | |
cc9f21da DS |
1105 | void (*func)(struct thread *), void *arg, |
1106 | long timer, struct thread **t_ptr) | |
98c91ac6 | 1107 | { |
d62a17ae | 1108 | struct timeval trel; |
9e867fe6 | 1109 | |
d62a17ae | 1110 | assert(m != NULL); |
718e3744 | 1111 | |
d62a17ae | 1112 | trel.tv_sec = timer / 1000; |
1113 | trel.tv_usec = 1000 * (timer % 1000); | |
98c91ac6 | 1114 | |
ee1455dd | 1115 | _thread_add_timer_timeval(xref, m, func, arg, &trel, t_ptr); |
a48b4e6d | 1116 | } |
1117 | ||
4322dea7 | 1118 | /* Add timer event thread with "timeval" resolution */ |
ee1455dd | 1119 | void _thread_add_timer_tv(const struct xref_threadsched *xref, |
cc9f21da DS |
1120 | struct thread_master *m, |
1121 | void (*func)(struct thread *), void *arg, | |
1122 | struct timeval *tv, struct thread **t_ptr) | |
d03c4cbd | 1123 | { |
ee1455dd | 1124 | _thread_add_timer_timeval(xref, m, func, arg, tv, t_ptr); |
d03c4cbd DL |
1125 | } |
1126 | ||
718e3744 | 1127 | /* Add simple event thread. */ |
ee1455dd | 1128 | void _thread_add_event(const struct xref_threadsched *xref, |
cc9f21da | 1129 | struct thread_master *m, void (*func)(struct thread *), |
ee1455dd | 1130 | void *arg, int val, struct thread **t_ptr) |
718e3744 | 1131 | { |
00dffa8c | 1132 | struct thread *thread = NULL; |
d62a17ae | 1133 | |
6c3aa850 DL |
1134 | frrtrace(9, frr_libfrr, schedule_event, m, |
1135 | xref->funcname, xref->xref.file, xref->xref.line, | |
c7bb4f00 | 1136 | t_ptr, 0, val, arg, 0); |
abf96a87 | 1137 | |
d62a17ae | 1138 | assert(m != NULL); |
1139 | ||
cb1991af | 1140 | frr_with_mutex (&m->mtx) { |
00dffa8c | 1141 | if (t_ptr && *t_ptr) |
d279ef57 | 1142 | /* thread is already scheduled; don't reschedule */ |
00dffa8c | 1143 | break; |
d62a17ae | 1144 | |
60a3efec | 1145 | thread = thread_get(m, THREAD_EVENT, func, arg, xref); |
cb1991af | 1146 | frr_with_mutex (&thread->mtx) { |
d62a17ae | 1147 | thread->u.val = val; |
c284542b | 1148 | thread_list_add_tail(&m->event, thread); |
d62a17ae | 1149 | } |
d62a17ae | 1150 | |
1151 | if (t_ptr) { | |
1152 | *t_ptr = thread; | |
1153 | thread->ref = t_ptr; | |
1154 | } | |
1155 | ||
1156 | AWAKEN(m); | |
1157 | } | |
718e3744 | 1158 | } |
1159 | ||
63ccb9cb QY |
1160 | /* Thread cancellation ------------------------------------------------------ */ |
1161 | ||
8797240e QY |
1162 | /** |
1163 | * NOT's out the .events field of pollfd corresponding to the given file | |
1164 | * descriptor. The event to be NOT'd is passed in the 'state' parameter. | |
1165 | * | |
1166 | * This needs to happen for both copies of pollfd's. See 'thread_fetch' | |
1167 | * implementation for details. | |
1168 | * | |
1169 | * @param master | |
1170 | * @param fd | |
1171 | * @param state the event to cancel. One or more (OR'd together) of the | |
1172 | * following: | |
1173 | * - POLLIN | |
1174 | * - POLLOUT | |
1175 | */ | |
a9318a32 MS |
1176 | static void thread_cancel_rw(struct thread_master *master, int fd, short state, |
1177 | int idx_hint) | |
0a95a0d0 | 1178 | { |
42d74538 QY |
1179 | bool found = false; |
1180 | ||
d62a17ae | 1181 | /* find the index of corresponding pollfd */ |
1182 | nfds_t i; | |
1183 | ||
a9318a32 MS |
1184 | /* Cancel POLLHUP too just in case some bozo set it */ |
1185 | state |= POLLHUP; | |
1186 | ||
1187 | /* Some callers know the index of the pfd already */ | |
1188 | if (idx_hint >= 0) { | |
1189 | i = idx_hint; | |
1190 | found = true; | |
1191 | } else { | |
1192 | /* Have to look for the fd in the pfd array */ | |
1193 | for (i = 0; i < master->handler.pfdcount; i++) | |
1194 | if (master->handler.pfds[i].fd == fd) { | |
1195 | found = true; | |
1196 | break; | |
1197 | } | |
1198 | } | |
42d74538 QY |
1199 | |
1200 | if (!found) { | |
1201 | zlog_debug( | |
1202 | "[!] Received cancellation request for nonexistent rw job"); | |
1203 | zlog_debug("[!] threadmaster: %s | fd: %d", | |
996c9314 | 1204 | master->name ? master->name : "", fd); |
42d74538 QY |
1205 | return; |
1206 | } | |
d62a17ae | 1207 | |
1208 | /* NOT out event. */ | |
1209 | master->handler.pfds[i].events &= ~(state); | |
1210 | ||
1211 | /* If all events are canceled, delete / resize the pollfd array. */ | |
1212 | if (master->handler.pfds[i].events == 0) { | |
1213 | memmove(master->handler.pfds + i, master->handler.pfds + i + 1, | |
1214 | (master->handler.pfdcount - i - 1) | |
1215 | * sizeof(struct pollfd)); | |
1216 | master->handler.pfdcount--; | |
e985cda0 S |
1217 | master->handler.pfds[master->handler.pfdcount].fd = 0; |
1218 | master->handler.pfds[master->handler.pfdcount].events = 0; | |
d62a17ae | 1219 | } |
1220 | ||
1221 | /* If we have the same pollfd in the copy, perform the same operations, | |
1222 | * otherwise return. */ | |
1223 | if (i >= master->handler.copycount) | |
1224 | return; | |
1225 | ||
1226 | master->handler.copy[i].events &= ~(state); | |
1227 | ||
1228 | if (master->handler.copy[i].events == 0) { | |
1229 | memmove(master->handler.copy + i, master->handler.copy + i + 1, | |
1230 | (master->handler.copycount - i - 1) | |
1231 | * sizeof(struct pollfd)); | |
1232 | master->handler.copycount--; | |
e985cda0 S |
1233 | master->handler.copy[master->handler.copycount].fd = 0; |
1234 | master->handler.copy[master->handler.copycount].events = 0; | |
d62a17ae | 1235 | } |
0a95a0d0 DS |
1236 | } |
1237 | ||
a9318a32 MS |
1238 | /* |
1239 | * Process task cancellation given a task argument: iterate through the | |
1240 | * various lists of tasks, looking for any that match the argument. | |
1241 | */ | |
1242 | static void cancel_arg_helper(struct thread_master *master, | |
1243 | const struct cancel_req *cr) | |
1244 | { | |
1245 | struct thread *t; | |
1246 | nfds_t i; | |
1247 | int fd; | |
1248 | struct pollfd *pfd; | |
1249 | ||
1250 | /* We're only processing arg-based cancellations here. */ | |
1251 | if (cr->eventobj == NULL) | |
1252 | return; | |
1253 | ||
1254 | /* First process the ready lists. */ | |
1255 | frr_each_safe(thread_list, &master->event, t) { | |
1256 | if (t->arg != cr->eventobj) | |
1257 | continue; | |
1258 | thread_list_del(&master->event, t); | |
1259 | if (t->ref) | |
1260 | *t->ref = NULL; | |
1261 | thread_add_unuse(master, t); | |
1262 | } | |
1263 | ||
1264 | frr_each_safe(thread_list, &master->ready, t) { | |
1265 | if (t->arg != cr->eventobj) | |
1266 | continue; | |
1267 | thread_list_del(&master->ready, t); | |
1268 | if (t->ref) | |
1269 | *t->ref = NULL; | |
1270 | thread_add_unuse(master, t); | |
1271 | } | |
1272 | ||
1273 | /* If requested, stop here and ignore io and timers */ | |
1274 | if (CHECK_FLAG(cr->flags, THREAD_CANCEL_FLAG_READY)) | |
1275 | return; | |
1276 | ||
1277 | /* Check the io tasks */ | |
1278 | for (i = 0; i < master->handler.pfdcount;) { | |
1279 | pfd = master->handler.pfds + i; | |
1280 | ||
1281 | if (pfd->events & POLLIN) | |
1282 | t = master->read[pfd->fd]; | |
1283 | else | |
1284 | t = master->write[pfd->fd]; | |
1285 | ||
1286 | if (t && t->arg == cr->eventobj) { | |
1287 | fd = pfd->fd; | |
1288 | ||
1289 | /* Found a match to cancel: clean up fd arrays */ | |
1290 | thread_cancel_rw(master, pfd->fd, pfd->events, i); | |
1291 | ||
1292 | /* Clean up thread arrays */ | |
1293 | master->read[fd] = NULL; | |
1294 | master->write[fd] = NULL; | |
1295 | ||
1296 | /* Clear caller's ref */ | |
1297 | if (t->ref) | |
1298 | *t->ref = NULL; | |
1299 | ||
1300 | thread_add_unuse(master, t); | |
1301 | ||
1302 | /* Don't increment 'i' since the cancellation will have | |
1303 | * removed the entry from the pfd array | |
1304 | */ | |
1305 | } else | |
1306 | i++; | |
1307 | } | |
1308 | ||
1309 | /* Check the timer tasks */ | |
1310 | t = thread_timer_list_first(&master->timer); | |
1311 | while (t) { | |
1312 | struct thread *t_next; | |
1313 | ||
1314 | t_next = thread_timer_list_next(&master->timer, t); | |
1315 | ||
1316 | if (t->arg == cr->eventobj) { | |
1317 | thread_timer_list_del(&master->timer, t); | |
1318 | if (t->ref) | |
1319 | *t->ref = NULL; | |
1320 | thread_add_unuse(master, t); | |
1321 | } | |
1322 | ||
1323 | t = t_next; | |
1324 | } | |
1325 | } | |
1326 | ||
1189d95f | 1327 | /** |
63ccb9cb | 1328 | * Process cancellation requests. |
1189d95f | 1329 | * |
63ccb9cb QY |
1330 | * This may only be run from the pthread which owns the thread_master. |
1331 | * | |
1332 | * @param master the thread master to process | |
1333 | * @REQUIRE master->mtx | |
1189d95f | 1334 | */ |
d62a17ae | 1335 | static void do_thread_cancel(struct thread_master *master) |
718e3744 | 1336 | { |
c284542b | 1337 | struct thread_list_head *list = NULL; |
d62a17ae | 1338 | struct thread **thread_array = NULL; |
1339 | struct thread *thread; | |
d62a17ae | 1340 | struct cancel_req *cr; |
1341 | struct listnode *ln; | |
7e93a54c | 1342 | |
d62a17ae | 1343 | for (ALL_LIST_ELEMENTS_RO(master->cancel_req, ln, cr)) { |
d279ef57 | 1344 | /* |
a9318a32 MS |
1345 | * If this is an event object cancellation, search |
1346 | * through task lists deleting any tasks which have the | |
1347 | * specified argument - use this handy helper function. | |
d279ef57 | 1348 | */ |
d62a17ae | 1349 | if (cr->eventobj) { |
a9318a32 | 1350 | cancel_arg_helper(master, cr); |
d62a17ae | 1351 | continue; |
1352 | } | |
1353 | ||
d279ef57 DS |
1354 | /* |
1355 | * The pointer varies depending on whether the cancellation | |
1356 | * request was made asynchronously or not. If it was, we | |
1357 | * need to check whether the thread even exists anymore | |
1358 | * before cancelling it. | |
1359 | */ | |
d62a17ae | 1360 | thread = (cr->thread) ? cr->thread : *cr->threadref; |
1361 | ||
1362 | if (!thread) | |
1363 | continue; | |
1364 | ||
7e93a54c MS |
1365 | list = NULL; |
1366 | thread_array = NULL; | |
1367 | ||
d62a17ae | 1368 | /* Determine the appropriate queue to cancel the thread from */ |
1369 | switch (thread->type) { | |
1370 | case THREAD_READ: | |
a9318a32 | 1371 | thread_cancel_rw(master, thread->u.fd, POLLIN, -1); |
d62a17ae | 1372 | thread_array = master->read; |
1373 | break; | |
1374 | case THREAD_WRITE: | |
a9318a32 | 1375 | thread_cancel_rw(master, thread->u.fd, POLLOUT, -1); |
d62a17ae | 1376 | thread_array = master->write; |
1377 | break; | |
1378 | case THREAD_TIMER: | |
27d29ced | 1379 | thread_timer_list_del(&master->timer, thread); |
d62a17ae | 1380 | break; |
1381 | case THREAD_EVENT: | |
1382 | list = &master->event; | |
1383 | break; | |
1384 | case THREAD_READY: | |
1385 | list = &master->ready; | |
1386 | break; | |
1387 | default: | |
1388 | continue; | |
1389 | break; | |
1390 | } | |
1391 | ||
27d29ced | 1392 | if (list) { |
c284542b | 1393 | thread_list_del(list, thread); |
d62a17ae | 1394 | } else if (thread_array) { |
1395 | thread_array[thread->u.fd] = NULL; | |
d62a17ae | 1396 | } |
1397 | ||
1398 | if (thread->ref) | |
1399 | *thread->ref = NULL; | |
1400 | ||
1401 | thread_add_unuse(thread->master, thread); | |
1402 | } | |
1403 | ||
1404 | /* Delete and free all cancellation requests */ | |
41b21bfa MS |
1405 | if (master->cancel_req) |
1406 | list_delete_all_node(master->cancel_req); | |
d62a17ae | 1407 | |
1408 | /* Wake up any threads which may be blocked in thread_cancel_async() */ | |
1409 | master->canceled = true; | |
1410 | pthread_cond_broadcast(&master->cancel_cond); | |
718e3744 | 1411 | } |
1412 | ||
a9318a32 MS |
1413 | /* |
1414 | * Helper function used for multiple flavors of arg-based cancellation. | |
1415 | */ | |
1416 | static void cancel_event_helper(struct thread_master *m, void *arg, int flags) | |
1417 | { | |
1418 | struct cancel_req *cr; | |
1419 | ||
1420 | assert(m->owner == pthread_self()); | |
1421 | ||
1422 | /* Only worth anything if caller supplies an arg. */ | |
1423 | if (arg == NULL) | |
1424 | return; | |
1425 | ||
1426 | cr = XCALLOC(MTYPE_TMP, sizeof(struct cancel_req)); | |
1427 | ||
1428 | cr->flags = flags; | |
1429 | ||
cb1991af | 1430 | frr_with_mutex (&m->mtx) { |
a9318a32 MS |
1431 | cr->eventobj = arg; |
1432 | listnode_add(m->cancel_req, cr); | |
1433 | do_thread_cancel(m); | |
1434 | } | |
1435 | } | |
1436 | ||
63ccb9cb QY |
1437 | /** |
1438 | * Cancel any events which have the specified argument. | |
1439 | * | |
1440 | * MT-Unsafe | |
1441 | * | |
1442 | * @param m the thread_master to cancel from | |
1443 | * @param arg the argument passed when creating the event | |
1444 | */ | |
d62a17ae | 1445 | void thread_cancel_event(struct thread_master *master, void *arg) |
718e3744 | 1446 | { |
a9318a32 MS |
1447 | cancel_event_helper(master, arg, 0); |
1448 | } | |
d62a17ae | 1449 | |
a9318a32 MS |
1450 | /* |
1451 | * Cancel ready tasks with an arg matching 'arg' | |
1452 | * | |
1453 | * MT-Unsafe | |
1454 | * | |
1455 | * @param m the thread_master to cancel from | |
1456 | * @param arg the argument passed when creating the event | |
1457 | */ | |
1458 | void thread_cancel_event_ready(struct thread_master *m, void *arg) | |
1459 | { | |
1460 | ||
1461 | /* Only cancel ready/event tasks */ | |
1462 | cancel_event_helper(m, arg, THREAD_CANCEL_FLAG_READY); | |
63ccb9cb | 1463 | } |
1189d95f | 1464 | |
63ccb9cb QY |
1465 | /** |
1466 | * Cancel a specific task. | |
1467 | * | |
1468 | * MT-Unsafe | |
1469 | * | |
1470 | * @param thread task to cancel | |
1471 | */ | |
b3d6bc6e | 1472 | void thread_cancel(struct thread **thread) |
63ccb9cb | 1473 | { |
b3d6bc6e MS |
1474 | struct thread_master *master; |
1475 | ||
1476 | if (thread == NULL || *thread == NULL) | |
1477 | return; | |
1478 | ||
1479 | master = (*thread)->master; | |
d62a17ae | 1480 | |
6c3aa850 DL |
1481 | frrtrace(9, frr_libfrr, thread_cancel, master, |
1482 | (*thread)->xref->funcname, (*thread)->xref->xref.file, | |
1483 | (*thread)->xref->xref.line, NULL, (*thread)->u.fd, | |
b4d6e855 | 1484 | (*thread)->u.val, (*thread)->arg, (*thread)->u.sands.tv_sec); |
abf96a87 | 1485 | |
6ed04aa2 DS |
1486 | assert(master->owner == pthread_self()); |
1487 | ||
cb1991af | 1488 | frr_with_mutex (&master->mtx) { |
d62a17ae | 1489 | struct cancel_req *cr = |
1490 | XCALLOC(MTYPE_TMP, sizeof(struct cancel_req)); | |
b3d6bc6e | 1491 | cr->thread = *thread; |
6ed04aa2 DS |
1492 | listnode_add(master->cancel_req, cr); |
1493 | do_thread_cancel(master); | |
d62a17ae | 1494 | } |
b3d6bc6e MS |
1495 | |
1496 | *thread = NULL; | |
63ccb9cb | 1497 | } |
1189d95f | 1498 | |
63ccb9cb QY |
1499 | /** |
1500 | * Asynchronous cancellation. | |
1501 | * | |
8797240e QY |
1502 | * Called with either a struct thread ** or void * to an event argument, |
1503 | * this function posts the correct cancellation request and blocks until it is | |
1504 | * serviced. | |
63ccb9cb QY |
1505 | * |
1506 | * If the thread is currently running, execution blocks until it completes. | |
1507 | * | |
8797240e QY |
1508 | * The last two parameters are mutually exclusive, i.e. if you pass one the |
1509 | * other must be NULL. | |
1510 | * | |
1511 | * When the cancellation procedure executes on the target thread_master, the | |
1512 | * thread * provided is checked for nullity. If it is null, the thread is | |
1513 | * assumed to no longer exist and the cancellation request is a no-op. Thus | |
1514 | * users of this API must pass a back-reference when scheduling the original | |
1515 | * task. | |
1516 | * | |
63ccb9cb QY |
1517 | * MT-Safe |
1518 | * | |
8797240e QY |
1519 | * @param master the thread master with the relevant event / task |
1520 | * @param thread pointer to thread to cancel | |
1521 | * @param eventobj the event | |
63ccb9cb | 1522 | */ |
d62a17ae | 1523 | void thread_cancel_async(struct thread_master *master, struct thread **thread, |
1524 | void *eventobj) | |
63ccb9cb | 1525 | { |
d62a17ae | 1526 | assert(!(thread && eventobj) && (thread || eventobj)); |
abf96a87 QY |
1527 | |
1528 | if (thread && *thread) | |
c7bb4f00 | 1529 | frrtrace(9, frr_libfrr, thread_cancel_async, master, |
6c3aa850 DL |
1530 | (*thread)->xref->funcname, (*thread)->xref->xref.file, |
1531 | (*thread)->xref->xref.line, NULL, (*thread)->u.fd, | |
c7bb4f00 QY |
1532 | (*thread)->u.val, (*thread)->arg, |
1533 | (*thread)->u.sands.tv_sec); | |
abf96a87 | 1534 | else |
c7bb4f00 QY |
1535 | frrtrace(9, frr_libfrr, thread_cancel_async, master, NULL, NULL, |
1536 | 0, NULL, 0, 0, eventobj, 0); | |
abf96a87 | 1537 | |
d62a17ae | 1538 | assert(master->owner != pthread_self()); |
1539 | ||
cb1991af | 1540 | frr_with_mutex (&master->mtx) { |
d62a17ae | 1541 | master->canceled = false; |
1542 | ||
1543 | if (thread) { | |
1544 | struct cancel_req *cr = | |
1545 | XCALLOC(MTYPE_TMP, sizeof(struct cancel_req)); | |
1546 | cr->threadref = thread; | |
1547 | listnode_add(master->cancel_req, cr); | |
1548 | } else if (eventobj) { | |
1549 | struct cancel_req *cr = | |
1550 | XCALLOC(MTYPE_TMP, sizeof(struct cancel_req)); | |
1551 | cr->eventobj = eventobj; | |
1552 | listnode_add(master->cancel_req, cr); | |
1553 | } | |
1554 | AWAKEN(master); | |
1555 | ||
1556 | while (!master->canceled) | |
1557 | pthread_cond_wait(&master->cancel_cond, &master->mtx); | |
1558 | } | |
50478845 MS |
1559 | |
1560 | if (thread) | |
1561 | *thread = NULL; | |
718e3744 | 1562 | } |
63ccb9cb | 1563 | /* ------------------------------------------------------------------------- */ |
718e3744 | 1564 | |
27d29ced | 1565 | static struct timeval *thread_timer_wait(struct thread_timer_list_head *timers, |
d62a17ae | 1566 | struct timeval *timer_val) |
718e3744 | 1567 | { |
27d29ced DL |
1568 | if (!thread_timer_list_count(timers)) |
1569 | return NULL; | |
1570 | ||
1571 | struct thread *next_timer = thread_timer_list_first(timers); | |
1572 | monotime_until(&next_timer->u.sands, timer_val); | |
1573 | return timer_val; | |
718e3744 | 1574 | } |
718e3744 | 1575 | |
d62a17ae | 1576 | static struct thread *thread_run(struct thread_master *m, struct thread *thread, |
1577 | struct thread *fetch) | |
718e3744 | 1578 | { |
d62a17ae | 1579 | *fetch = *thread; |
1580 | thread_add_unuse(m, thread); | |
1581 | return fetch; | |
718e3744 | 1582 | } |
1583 | ||
d62a17ae | 1584 | static int thread_process_io_helper(struct thread_master *m, |
45f3d590 DS |
1585 | struct thread *thread, short state, |
1586 | short actual_state, int pos) | |
5d4ccd4e | 1587 | { |
d62a17ae | 1588 | struct thread **thread_array; |
1589 | ||
45f3d590 DS |
1590 | /* |
1591 | * poll() clears the .events field, but the pollfd array we | |
1592 | * pass to poll() is a copy of the one used to schedule threads. | |
1593 | * We need to synchronize state between the two here by applying | |
1594 | * the same changes poll() made on the copy of the "real" pollfd | |
1595 | * array. | |
1596 | * | |
1597 | * This cleans up a possible infinite loop where we refuse | |
1598 | * to respond to a poll event but poll is insistent that | |
1599 | * we should. | |
1600 | */ | |
1601 | m->handler.pfds[pos].events &= ~(state); | |
1602 | ||
1603 | if (!thread) { | |
1604 | if ((actual_state & (POLLHUP|POLLIN)) != POLLHUP) | |
1605 | flog_err(EC_LIB_NO_THREAD, | |
1d5453d6 | 1606 | "Attempting to process an I/O event but for fd: %d(%d) no thread to handle this!", |
45f3d590 | 1607 | m->handler.pfds[pos].fd, actual_state); |
d62a17ae | 1608 | return 0; |
45f3d590 | 1609 | } |
d62a17ae | 1610 | |
1611 | if (thread->type == THREAD_READ) | |
1612 | thread_array = m->read; | |
1613 | else | |
1614 | thread_array = m->write; | |
1615 | ||
1616 | thread_array[thread->u.fd] = NULL; | |
c284542b | 1617 | thread_list_add_tail(&m->ready, thread); |
d62a17ae | 1618 | thread->type = THREAD_READY; |
45f3d590 | 1619 | |
d62a17ae | 1620 | return 1; |
5d4ccd4e DS |
1621 | } |
1622 | ||
8797240e QY |
1623 | /** |
1624 | * Process I/O events. | |
1625 | * | |
1626 | * Walks through file descriptor array looking for those pollfds whose .revents | |
1627 | * field has something interesting. Deletes any invalid file descriptors. | |
1628 | * | |
1629 | * @param m the thread master | |
1630 | * @param num the number of active file descriptors (return value of poll()) | |
1631 | */ | |
d62a17ae | 1632 | static void thread_process_io(struct thread_master *m, unsigned int num) |
0a95a0d0 | 1633 | { |
d62a17ae | 1634 | unsigned int ready = 0; |
1635 | struct pollfd *pfds = m->handler.copy; | |
1636 | ||
1637 | for (nfds_t i = 0; i < m->handler.copycount && ready < num; ++i) { | |
1638 | /* no event for current fd? immediately continue */ | |
1639 | if (pfds[i].revents == 0) | |
1640 | continue; | |
1641 | ||
1642 | ready++; | |
1643 | ||
d279ef57 DS |
1644 | /* |
1645 | * Unless someone has called thread_cancel from another | |
1646 | * pthread, the only thing that could have changed in | |
1647 | * m->handler.pfds while we were asleep is the .events | |
1648 | * field in a given pollfd. Barring thread_cancel() that | |
1649 | * value should be a superset of the values we have in our | |
1650 | * copy, so there's no need to update it. Similarily, | |
1651 | * barring deletion, the fd should still be a valid index | |
1652 | * into the master's pfds. | |
d142453d DS |
1653 | * |
1654 | * We are including POLLERR here to do a READ event | |
1655 | * this is because the read should fail and the | |
1656 | * read function should handle it appropriately | |
d279ef57 | 1657 | */ |
d142453d | 1658 | if (pfds[i].revents & (POLLIN | POLLHUP | POLLERR)) { |
d62a17ae | 1659 | thread_process_io_helper(m, m->read[pfds[i].fd], POLLIN, |
45f3d590 DS |
1660 | pfds[i].revents, i); |
1661 | } | |
d62a17ae | 1662 | if (pfds[i].revents & POLLOUT) |
1663 | thread_process_io_helper(m, m->write[pfds[i].fd], | |
45f3d590 | 1664 | POLLOUT, pfds[i].revents, i); |
d62a17ae | 1665 | |
1666 | /* if one of our file descriptors is garbage, remove the same | |
1667 | * from | |
1668 | * both pfds + update sizes and index */ | |
1669 | if (pfds[i].revents & POLLNVAL) { | |
1670 | memmove(m->handler.pfds + i, m->handler.pfds + i + 1, | |
1671 | (m->handler.pfdcount - i - 1) | |
1672 | * sizeof(struct pollfd)); | |
1673 | m->handler.pfdcount--; | |
e985cda0 S |
1674 | m->handler.pfds[m->handler.pfdcount].fd = 0; |
1675 | m->handler.pfds[m->handler.pfdcount].events = 0; | |
d62a17ae | 1676 | |
1677 | memmove(pfds + i, pfds + i + 1, | |
1678 | (m->handler.copycount - i - 1) | |
1679 | * sizeof(struct pollfd)); | |
1680 | m->handler.copycount--; | |
e985cda0 S |
1681 | m->handler.copy[m->handler.copycount].fd = 0; |
1682 | m->handler.copy[m->handler.copycount].events = 0; | |
d62a17ae | 1683 | |
1684 | i--; | |
1685 | } | |
1686 | } | |
718e3744 | 1687 | } |
1688 | ||
8b70d0b0 | 1689 | /* Add all timers that have popped to the ready list. */ |
e7d9e44b | 1690 | static unsigned int thread_process_timers(struct thread_master *m, |
d62a17ae | 1691 | struct timeval *timenow) |
a48b4e6d | 1692 | { |
ab01a001 DS |
1693 | struct timeval prev = *timenow; |
1694 | bool displayed = false; | |
d62a17ae | 1695 | struct thread *thread; |
1696 | unsigned int ready = 0; | |
1697 | ||
e7d9e44b | 1698 | while ((thread = thread_timer_list_first(&m->timer))) { |
d62a17ae | 1699 | if (timercmp(timenow, &thread->u.sands, <)) |
e7d9e44b | 1700 | break; |
ab01a001 DS |
1701 | prev = thread->u.sands; |
1702 | prev.tv_sec += 4; | |
1703 | /* | |
1704 | * If the timer would have popped 4 seconds in the | |
1705 | * past then we are in a situation where we are | |
1706 | * really getting behind on handling of events. | |
1707 | * Let's log it and do the right thing with it. | |
1708 | */ | |
1dd08c22 DS |
1709 | if (timercmp(timenow, &prev, >)) { |
1710 | atomic_fetch_add_explicit( | |
1711 | &thread->hist->total_starv_warn, 1, | |
1712 | memory_order_seq_cst); | |
1713 | if (!displayed && !thread->ignore_timer_late) { | |
1714 | flog_warn( | |
1715 | EC_LIB_STARVE_THREAD, | |
1716 | "Thread Starvation: %pTHD was scheduled to pop greater than 4s ago", | |
1717 | thread); | |
1718 | displayed = true; | |
1719 | } | |
ab01a001 DS |
1720 | } |
1721 | ||
e7d9e44b | 1722 | thread_timer_list_pop(&m->timer); |
d62a17ae | 1723 | thread->type = THREAD_READY; |
e7d9e44b | 1724 | thread_list_add_tail(&m->ready, thread); |
d62a17ae | 1725 | ready++; |
1726 | } | |
e7d9e44b | 1727 | |
d62a17ae | 1728 | return ready; |
a48b4e6d | 1729 | } |
1730 | ||
2613abe6 | 1731 | /* process a list en masse, e.g. for event thread lists */ |
c284542b | 1732 | static unsigned int thread_process(struct thread_list_head *list) |
2613abe6 | 1733 | { |
d62a17ae | 1734 | struct thread *thread; |
d62a17ae | 1735 | unsigned int ready = 0; |
1736 | ||
c284542b | 1737 | while ((thread = thread_list_pop(list))) { |
d62a17ae | 1738 | thread->type = THREAD_READY; |
c284542b | 1739 | thread_list_add_tail(&thread->master->ready, thread); |
d62a17ae | 1740 | ready++; |
1741 | } | |
1742 | return ready; | |
2613abe6 PJ |
1743 | } |
1744 | ||
1745 | ||
718e3744 | 1746 | /* Fetch next ready thread. */ |
d62a17ae | 1747 | struct thread *thread_fetch(struct thread_master *m, struct thread *fetch) |
718e3744 | 1748 | { |
d62a17ae | 1749 | struct thread *thread = NULL; |
1750 | struct timeval now; | |
1751 | struct timeval zerotime = {0, 0}; | |
1752 | struct timeval tv; | |
1753 | struct timeval *tw = NULL; | |
d81ca9a3 | 1754 | bool eintr_p = false; |
d62a17ae | 1755 | int num = 0; |
1756 | ||
1757 | do { | |
1758 | /* Handle signals if any */ | |
1759 | if (m->handle_signals) | |
7cc91e67 | 1760 | frr_sigevent_process(); |
d62a17ae | 1761 | |
1762 | pthread_mutex_lock(&m->mtx); | |
1763 | ||
1764 | /* Process any pending cancellation requests */ | |
1765 | do_thread_cancel(m); | |
1766 | ||
e3c9529e QY |
1767 | /* |
1768 | * Attempt to flush ready queue before going into poll(). | |
1769 | * This is performance-critical. Think twice before modifying. | |
1770 | */ | |
c284542b | 1771 | if ((thread = thread_list_pop(&m->ready))) { |
e3c9529e QY |
1772 | fetch = thread_run(m, thread, fetch); |
1773 | if (fetch->ref) | |
1774 | *fetch->ref = NULL; | |
1775 | pthread_mutex_unlock(&m->mtx); | |
5e822957 DS |
1776 | if (!m->ready_run_loop) |
1777 | GETRUSAGE(&m->last_getrusage); | |
1778 | m->ready_run_loop = true; | |
e3c9529e QY |
1779 | break; |
1780 | } | |
1781 | ||
5e822957 | 1782 | m->ready_run_loop = false; |
e3c9529e QY |
1783 | /* otherwise, tick through scheduling sequence */ |
1784 | ||
bca37d17 QY |
1785 | /* |
1786 | * Post events to ready queue. This must come before the | |
1787 | * following block since events should occur immediately | |
1788 | */ | |
d62a17ae | 1789 | thread_process(&m->event); |
1790 | ||
bca37d17 QY |
1791 | /* |
1792 | * If there are no tasks on the ready queue, we will poll() | |
1793 | * until a timer expires or we receive I/O, whichever comes | |
1794 | * first. The strategy for doing this is: | |
d62a17ae | 1795 | * |
1796 | * - If there are events pending, set the poll() timeout to zero | |
1797 | * - If there are no events pending, but there are timers | |
d279ef57 DS |
1798 | * pending, set the timeout to the smallest remaining time on |
1799 | * any timer. | |
d62a17ae | 1800 | * - If there are neither timers nor events pending, but there |
d279ef57 | 1801 | * are file descriptors pending, block indefinitely in poll() |
d62a17ae | 1802 | * - If nothing is pending, it's time for the application to die |
1803 | * | |
1804 | * In every case except the last, we need to hit poll() at least | |
bca37d17 QY |
1805 | * once per loop to avoid starvation by events |
1806 | */ | |
c284542b | 1807 | if (!thread_list_count(&m->ready)) |
27d29ced | 1808 | tw = thread_timer_wait(&m->timer, &tv); |
d62a17ae | 1809 | |
c284542b DL |
1810 | if (thread_list_count(&m->ready) || |
1811 | (tw && !timercmp(tw, &zerotime, >))) | |
d62a17ae | 1812 | tw = &zerotime; |
1813 | ||
1814 | if (!tw && m->handler.pfdcount == 0) { /* die */ | |
1815 | pthread_mutex_unlock(&m->mtx); | |
1816 | fetch = NULL; | |
1817 | break; | |
1818 | } | |
1819 | ||
bca37d17 QY |
1820 | /* |
1821 | * Copy pollfd array + # active pollfds in it. Not necessary to | |
1822 | * copy the array size as this is fixed. | |
1823 | */ | |
d62a17ae | 1824 | m->handler.copycount = m->handler.pfdcount; |
1825 | memcpy(m->handler.copy, m->handler.pfds, | |
1826 | m->handler.copycount * sizeof(struct pollfd)); | |
1827 | ||
e3c9529e QY |
1828 | pthread_mutex_unlock(&m->mtx); |
1829 | { | |
d81ca9a3 MS |
1830 | eintr_p = false; |
1831 | num = fd_poll(m, tw, &eintr_p); | |
e3c9529e QY |
1832 | } |
1833 | pthread_mutex_lock(&m->mtx); | |
d764d2cc | 1834 | |
e3c9529e QY |
1835 | /* Handle any errors received in poll() */ |
1836 | if (num < 0) { | |
d81ca9a3 | 1837 | if (eintr_p) { |
d62a17ae | 1838 | pthread_mutex_unlock(&m->mtx); |
e3c9529e QY |
1839 | /* loop around to signal handler */ |
1840 | continue; | |
d62a17ae | 1841 | } |
1842 | ||
e3c9529e | 1843 | /* else die */ |
450971aa | 1844 | flog_err(EC_LIB_SYSTEM_CALL, "poll() error: %s", |
9ef9495e | 1845 | safe_strerror(errno)); |
e3c9529e QY |
1846 | pthread_mutex_unlock(&m->mtx); |
1847 | fetch = NULL; | |
1848 | break; | |
bca37d17 | 1849 | } |
d62a17ae | 1850 | |
1851 | /* Post timers to ready queue. */ | |
1852 | monotime(&now); | |
e7d9e44b | 1853 | thread_process_timers(m, &now); |
d62a17ae | 1854 | |
1855 | /* Post I/O to ready queue. */ | |
1856 | if (num > 0) | |
1857 | thread_process_io(m, num); | |
1858 | ||
d62a17ae | 1859 | pthread_mutex_unlock(&m->mtx); |
1860 | ||
1861 | } while (!thread && m->spin); | |
1862 | ||
1863 | return fetch; | |
718e3744 | 1864 | } |
1865 | ||
d62a17ae | 1866 | static unsigned long timeval_elapsed(struct timeval a, struct timeval b) |
62f44022 | 1867 | { |
d62a17ae | 1868 | return (((a.tv_sec - b.tv_sec) * TIMER_SECOND_MICRO) |
1869 | + (a.tv_usec - b.tv_usec)); | |
62f44022 QY |
1870 | } |
1871 | ||
d62a17ae | 1872 | unsigned long thread_consumed_time(RUSAGE_T *now, RUSAGE_T *start, |
1873 | unsigned long *cputime) | |
718e3744 | 1874 | { |
6418e2d3 | 1875 | #ifdef HAVE_CLOCK_THREAD_CPUTIME_ID |
4e2839de DS |
1876 | |
1877 | #ifdef __FreeBSD__ | |
1878 | /* | |
1879 | * FreeBSD appears to have an issue when calling clock_gettime | |
1880 | * with CLOCK_THREAD_CPUTIME_ID really close to each other | |
1881 | * occassionally the now time will be before the start time. | |
1882 | * This is not good and FRR is ending up with CPU HOG's | |
1883 | * when the subtraction wraps to very large numbers | |
1884 | * | |
1885 | * What we are going to do here is cheat a little bit | |
1886 | * and notice that this is a problem and just correct | |
1887 | * it so that it is impossible to happen | |
1888 | */ | |
1889 | if (start->cpu.tv_sec == now->cpu.tv_sec && | |
1890 | start->cpu.tv_nsec > now->cpu.tv_nsec) | |
1891 | now->cpu.tv_nsec = start->cpu.tv_nsec + 1; | |
1892 | else if (start->cpu.tv_sec > now->cpu.tv_sec) { | |
1893 | now->cpu.tv_sec = start->cpu.tv_sec; | |
1894 | now->cpu.tv_nsec = start->cpu.tv_nsec + 1; | |
1895 | } | |
1896 | #endif | |
6418e2d3 DL |
1897 | *cputime = (now->cpu.tv_sec - start->cpu.tv_sec) * TIMER_SECOND_MICRO |
1898 | + (now->cpu.tv_nsec - start->cpu.tv_nsec) / 1000; | |
1899 | #else | |
d62a17ae | 1900 | /* This is 'user + sys' time. */ |
1901 | *cputime = timeval_elapsed(now->cpu.ru_utime, start->cpu.ru_utime) | |
1902 | + timeval_elapsed(now->cpu.ru_stime, start->cpu.ru_stime); | |
6418e2d3 | 1903 | #endif |
d62a17ae | 1904 | return timeval_elapsed(now->real, start->real); |
8b70d0b0 | 1905 | } |
1906 | ||
50596be0 DS |
1907 | /* We should aim to yield after yield milliseconds, which defaults |
1908 | to THREAD_YIELD_TIME_SLOT . | |
8b70d0b0 | 1909 | Note: we are using real (wall clock) time for this calculation. |
1910 | It could be argued that CPU time may make more sense in certain | |
1911 | contexts. The things to consider are whether the thread may have | |
1912 | blocked (in which case wall time increases, but CPU time does not), | |
1913 | or whether the system is heavily loaded with other processes competing | |
d62a17ae | 1914 | for CPU time. On balance, wall clock time seems to make sense. |
8b70d0b0 | 1915 | Plus it has the added benefit that gettimeofday should be faster |
1916 | than calling getrusage. */ | |
d62a17ae | 1917 | int thread_should_yield(struct thread *thread) |
718e3744 | 1918 | { |
d62a17ae | 1919 | int result; |
cb1991af | 1920 | frr_with_mutex (&thread->mtx) { |
d62a17ae | 1921 | result = monotime_since(&thread->real, NULL) |
1922 | > (int64_t)thread->yield; | |
1923 | } | |
d62a17ae | 1924 | return result; |
50596be0 DS |
1925 | } |
1926 | ||
d62a17ae | 1927 | void thread_set_yield_time(struct thread *thread, unsigned long yield_time) |
50596be0 | 1928 | { |
cb1991af | 1929 | frr_with_mutex (&thread->mtx) { |
d62a17ae | 1930 | thread->yield = yield_time; |
1931 | } | |
718e3744 | 1932 | } |
1933 | ||
d62a17ae | 1934 | void thread_getrusage(RUSAGE_T *r) |
db9c0df9 | 1935 | { |
6418e2d3 DL |
1936 | monotime(&r->real); |
1937 | if (!cputime_enabled) { | |
1938 | memset(&r->cpu, 0, sizeof(r->cpu)); | |
1939 | return; | |
1940 | } | |
1941 | ||
1942 | #ifdef HAVE_CLOCK_THREAD_CPUTIME_ID | |
1943 | /* not currently implemented in Linux's vDSO, but maybe at some point | |
1944 | * in the future? | |
1945 | */ | |
1946 | clock_gettime(CLOCK_THREAD_CPUTIME_ID, &r->cpu); | |
1947 | #else /* !HAVE_CLOCK_THREAD_CPUTIME_ID */ | |
231db9a6 DS |
1948 | #if defined RUSAGE_THREAD |
1949 | #define FRR_RUSAGE RUSAGE_THREAD | |
1950 | #else | |
1951 | #define FRR_RUSAGE RUSAGE_SELF | |
1952 | #endif | |
6418e2d3 DL |
1953 | getrusage(FRR_RUSAGE, &(r->cpu)); |
1954 | #endif | |
db9c0df9 PJ |
1955 | } |
1956 | ||
fbcac826 QY |
1957 | /* |
1958 | * Call a thread. | |
1959 | * | |
1960 | * This function will atomically update the thread's usage history. At present | |
1961 | * this is the only spot where usage history is written. Nevertheless the code | |
1962 | * has been written such that the introduction of writers in the future should | |
1963 | * not need to update it provided the writers atomically perform only the | |
1964 | * operations done here, i.e. updating the total and maximum times. In | |
1965 | * particular, the maximum real and cpu times must be monotonically increasing | |
1966 | * or this code is not correct. | |
1967 | */ | |
d62a17ae | 1968 | void thread_call(struct thread *thread) |
718e3744 | 1969 | { |
d62a17ae | 1970 | RUSAGE_T before, after; |
cc8b13a0 | 1971 | |
45f01188 DL |
1972 | /* if the thread being called is the CLI, it may change cputime_enabled |
1973 | * ("service cputime-stats" command), which can result in nonsensical | |
1974 | * and very confusing warnings | |
1975 | */ | |
1976 | bool cputime_enabled_here = cputime_enabled; | |
1977 | ||
5e822957 DS |
1978 | if (thread->master->ready_run_loop) |
1979 | before = thread->master->last_getrusage; | |
1980 | else | |
1981 | GETRUSAGE(&before); | |
1982 | ||
d62a17ae | 1983 | thread->real = before.real; |
718e3744 | 1984 | |
6c3aa850 DL |
1985 | frrtrace(9, frr_libfrr, thread_call, thread->master, |
1986 | thread->xref->funcname, thread->xref->xref.file, | |
1987 | thread->xref->xref.line, NULL, thread->u.fd, | |
c7bb4f00 | 1988 | thread->u.val, thread->arg, thread->u.sands.tv_sec); |
abf96a87 | 1989 | |
d62a17ae | 1990 | pthread_setspecific(thread_current, thread); |
1991 | (*thread->func)(thread); | |
1992 | pthread_setspecific(thread_current, NULL); | |
718e3744 | 1993 | |
d62a17ae | 1994 | GETRUSAGE(&after); |
5e822957 | 1995 | thread->master->last_getrusage = after; |
718e3744 | 1996 | |
45f01188 DL |
1997 | unsigned long walltime, cputime; |
1998 | unsigned long exp; | |
fbcac826 | 1999 | |
45f01188 DL |
2000 | walltime = thread_consumed_time(&after, &before, &cputime); |
2001 | ||
2002 | /* update walltime */ | |
2003 | atomic_fetch_add_explicit(&thread->hist->real.total, walltime, | |
fbcac826 QY |
2004 | memory_order_seq_cst); |
2005 | exp = atomic_load_explicit(&thread->hist->real.max, | |
2006 | memory_order_seq_cst); | |
45f01188 | 2007 | while (exp < walltime |
fbcac826 | 2008 | && !atomic_compare_exchange_weak_explicit( |
45f01188 DL |
2009 | &thread->hist->real.max, &exp, walltime, |
2010 | memory_order_seq_cst, memory_order_seq_cst)) | |
fbcac826 QY |
2011 | ; |
2012 | ||
45f01188 DL |
2013 | if (cputime_enabled_here && cputime_enabled) { |
2014 | /* update cputime */ | |
2015 | atomic_fetch_add_explicit(&thread->hist->cpu.total, cputime, | |
2016 | memory_order_seq_cst); | |
2017 | exp = atomic_load_explicit(&thread->hist->cpu.max, | |
2018 | memory_order_seq_cst); | |
2019 | while (exp < cputime | |
2020 | && !atomic_compare_exchange_weak_explicit( | |
2021 | &thread->hist->cpu.max, &exp, cputime, | |
2022 | memory_order_seq_cst, memory_order_seq_cst)) | |
2023 | ; | |
2024 | } | |
fbcac826 QY |
2025 | |
2026 | atomic_fetch_add_explicit(&thread->hist->total_calls, 1, | |
2027 | memory_order_seq_cst); | |
2028 | atomic_fetch_or_explicit(&thread->hist->types, 1 << thread->add_type, | |
2029 | memory_order_seq_cst); | |
718e3744 | 2030 | |
45f01188 DL |
2031 | if (cputime_enabled_here && cputime_enabled && cputime_threshold |
2032 | && cputime > cputime_threshold) { | |
d62a17ae | 2033 | /* |
45f01188 DL |
2034 | * We have a CPU Hog on our hands. The time FRR has spent |
2035 | * doing actual work (not sleeping) is greater than 5 seconds. | |
d62a17ae | 2036 | * Whinge about it now, so we're aware this is yet another task |
2037 | * to fix. | |
2038 | */ | |
9b8e01ca DS |
2039 | atomic_fetch_add_explicit(&thread->hist->total_cpu_warn, |
2040 | 1, memory_order_seq_cst); | |
9ef9495e | 2041 | flog_warn( |
039d547f DS |
2042 | EC_LIB_SLOW_THREAD_CPU, |
2043 | "CPU HOG: task %s (%lx) ran for %lums (cpu time %lums)", | |
2044 | thread->xref->funcname, (unsigned long)thread->func, | |
45f01188 DL |
2045 | walltime / 1000, cputime / 1000); |
2046 | ||
2047 | } else if (walltime_threshold && walltime > walltime_threshold) { | |
039d547f | 2048 | /* |
45f01188 DL |
2049 | * The runtime for a task is greater than 5 seconds, but the |
2050 | * cpu time is under 5 seconds. Let's whine about this because | |
2051 | * this could imply some sort of scheduling issue. | |
039d547f | 2052 | */ |
9b8e01ca DS |
2053 | atomic_fetch_add_explicit(&thread->hist->total_wall_warn, |
2054 | 1, memory_order_seq_cst); | |
039d547f DS |
2055 | flog_warn( |
2056 | EC_LIB_SLOW_THREAD_WALL, | |
2057 | "STARVATION: task %s (%lx) ran for %lums (cpu time %lums)", | |
60a3efec | 2058 | thread->xref->funcname, (unsigned long)thread->func, |
45f01188 | 2059 | walltime / 1000, cputime / 1000); |
d62a17ae | 2060 | } |
718e3744 | 2061 | } |
2062 | ||
2063 | /* Execute thread */ | |
60a3efec | 2064 | void _thread_execute(const struct xref_threadsched *xref, |
cc9f21da | 2065 | struct thread_master *m, void (*func)(struct thread *), |
60a3efec | 2066 | void *arg, int val) |
718e3744 | 2067 | { |
c4345fbf | 2068 | struct thread *thread; |
718e3744 | 2069 | |
c4345fbf | 2070 | /* Get or allocate new thread to execute. */ |
cb1991af | 2071 | frr_with_mutex (&m->mtx) { |
60a3efec | 2072 | thread = thread_get(m, THREAD_EVENT, func, arg, xref); |
9c7753e4 | 2073 | |
c4345fbf | 2074 | /* Set its event value. */ |
cb1991af | 2075 | frr_with_mutex (&thread->mtx) { |
c4345fbf RZ |
2076 | thread->add_type = THREAD_EXECUTE; |
2077 | thread->u.val = val; | |
2078 | thread->ref = &thread; | |
2079 | } | |
c4345fbf | 2080 | } |
f7c62e11 | 2081 | |
c4345fbf RZ |
2082 | /* Execute thread doing all accounting. */ |
2083 | thread_call(thread); | |
9c7753e4 | 2084 | |
c4345fbf RZ |
2085 | /* Give back or free thread. */ |
2086 | thread_add_unuse(m, thread); | |
718e3744 | 2087 | } |
1543c387 MS |
2088 | |
2089 | /* Debug signal mask - if 'sigs' is NULL, use current effective mask. */ | |
2090 | void debug_signals(const sigset_t *sigs) | |
2091 | { | |
2092 | int i, found; | |
2093 | sigset_t tmpsigs; | |
2094 | char buf[300]; | |
2095 | ||
2096 | /* | |
2097 | * We're only looking at the non-realtime signals here, so we need | |
2098 | * some limit value. Platform differences mean at some point we just | |
2099 | * need to pick a reasonable value. | |
2100 | */ | |
2101 | #if defined SIGRTMIN | |
2102 | # define LAST_SIGNAL SIGRTMIN | |
2103 | #else | |
2104 | # define LAST_SIGNAL 32 | |
2105 | #endif | |
2106 | ||
2107 | ||
2108 | if (sigs == NULL) { | |
2109 | sigemptyset(&tmpsigs); | |
2110 | pthread_sigmask(SIG_BLOCK, NULL, &tmpsigs); | |
2111 | sigs = &tmpsigs; | |
2112 | } | |
2113 | ||
2114 | found = 0; | |
2115 | buf[0] = '\0'; | |
2116 | ||
2117 | for (i = 0; i < LAST_SIGNAL; i++) { | |
2118 | char tmp[20]; | |
2119 | ||
2120 | if (sigismember(sigs, i) > 0) { | |
2121 | if (found > 0) | |
2122 | strlcat(buf, ",", sizeof(buf)); | |
2123 | snprintf(tmp, sizeof(tmp), "%d", i); | |
2124 | strlcat(buf, tmp, sizeof(buf)); | |
2125 | found++; | |
2126 | } | |
2127 | } | |
2128 | ||
2129 | if (found == 0) | |
2130 | snprintf(buf, sizeof(buf), "<none>"); | |
2131 | ||
2132 | zlog_debug("%s: %s", __func__, buf); | |
2133 | } | |
a505383d | 2134 | |
f59e6882 DL |
2135 | static ssize_t printfrr_thread_dbg(struct fbuf *buf, struct printfrr_eargs *ea, |
2136 | const struct thread *thread) | |
2137 | { | |
2138 | static const char * const types[] = { | |
2139 | [THREAD_READ] = "read", | |
2140 | [THREAD_WRITE] = "write", | |
2141 | [THREAD_TIMER] = "timer", | |
2142 | [THREAD_EVENT] = "event", | |
2143 | [THREAD_READY] = "ready", | |
2144 | [THREAD_UNUSED] = "unused", | |
2145 | [THREAD_EXECUTE] = "exec", | |
2146 | }; | |
2147 | ssize_t rv = 0; | |
2148 | char info[16] = ""; | |
2149 | ||
2150 | if (!thread) | |
2151 | return bputs(buf, "{(thread *)NULL}"); | |
2152 | ||
2153 | rv += bprintfrr(buf, "{(thread *)%p arg=%p", thread, thread->arg); | |
2154 | ||
2155 | if (thread->type < array_size(types) && types[thread->type]) | |
2156 | rv += bprintfrr(buf, " %-6s", types[thread->type]); | |
2157 | else | |
2158 | rv += bprintfrr(buf, " INVALID(%u)", thread->type); | |
2159 | ||
2160 | switch (thread->type) { | |
2161 | case THREAD_READ: | |
2162 | case THREAD_WRITE: | |
2163 | snprintfrr(info, sizeof(info), "fd=%d", thread->u.fd); | |
2164 | break; | |
2165 | ||
2166 | case THREAD_TIMER: | |
2167 | snprintfrr(info, sizeof(info), "r=%pTVMud", &thread->u.sands); | |
2168 | break; | |
2169 | } | |
2170 | ||
2171 | rv += bprintfrr(buf, " %-12s %s() %s from %s:%d}", info, | |
2172 | thread->xref->funcname, thread->xref->dest, | |
2173 | thread->xref->xref.file, thread->xref->xref.line); | |
2174 | return rv; | |
2175 | } | |
2176 | ||
54929fd3 | 2177 | printfrr_ext_autoreg_p("TH", printfrr_thread); |
f59e6882 DL |
2178 | static ssize_t printfrr_thread(struct fbuf *buf, struct printfrr_eargs *ea, |
2179 | const void *ptr) | |
2180 | { | |
2181 | const struct thread *thread = ptr; | |
2182 | struct timespec remain = {}; | |
2183 | ||
2184 | if (ea->fmt[0] == 'D') { | |
2185 | ea->fmt++; | |
2186 | return printfrr_thread_dbg(buf, ea, thread); | |
2187 | } | |
2188 | ||
2189 | if (!thread) { | |
2190 | /* need to jump over time formatting flag characters in the | |
2191 | * input format string, i.e. adjust ea->fmt! | |
2192 | */ | |
2193 | printfrr_time(buf, ea, &remain, | |
2194 | TIMEFMT_TIMER_DEADLINE | TIMEFMT_SKIP); | |
2195 | return bputch(buf, '-'); | |
2196 | } | |
2197 | ||
2198 | TIMEVAL_TO_TIMESPEC(&thread->u.sands, &remain); | |
2199 | return printfrr_time(buf, ea, &remain, TIMEFMT_TIMER_DEADLINE); | |
2200 | } |