]>
Commit | Line | Data |
---|---|---|
718e3744 | 1 | /* Thread management routine |
2 | * Copyright (C) 1998, 2000 Kunihiro Ishiguro <kunihiro@zebra.org> | |
3 | * | |
4 | * This file is part of GNU Zebra. | |
5 | * | |
6 | * GNU Zebra is free software; you can redistribute it and/or modify it | |
7 | * under the terms of the GNU General Public License as published by the | |
8 | * Free Software Foundation; either version 2, or (at your option) any | |
9 | * later version. | |
10 | * | |
11 | * GNU Zebra is distributed in the hope that it will be useful, but | |
12 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 | * General Public License for more details. | |
15 | * | |
896014f4 DL |
16 | * You should have received a copy of the GNU General Public License along |
17 | * with this program; see the file COPYING; if not, write to the Free Software | |
18 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
718e3744 | 19 | */ |
20 | ||
21 | /* #define DEBUG */ | |
22 | ||
23 | #include <zebra.h> | |
308d14ae | 24 | #include <sys/resource.h> |
718e3744 | 25 | |
26 | #include "thread.h" | |
27 | #include "memory.h" | |
3e41733f | 28 | #include "frrcu.h" |
718e3744 | 29 | #include "log.h" |
e04ab74d | 30 | #include "hash.h" |
31 | #include "command.h" | |
05c447dd | 32 | #include "sigevent.h" |
3bf2673b | 33 | #include "network.h" |
bd74dc61 | 34 | #include "jhash.h" |
fbcac826 | 35 | #include "frratomic.h" |
00dffa8c | 36 | #include "frr_pthread.h" |
9ef9495e | 37 | #include "lib_errors.h" |
912d45a1 | 38 | #include "libfrr_trace.h" |
1a9f340b | 39 | #include "libfrr.h" |
d6be5fb9 | 40 | |
bf8d3d6a DL |
41 | DEFINE_MTYPE_STATIC(LIB, THREAD, "Thread"); |
42 | DEFINE_MTYPE_STATIC(LIB, THREAD_MASTER, "Thread master"); | |
43 | DEFINE_MTYPE_STATIC(LIB, THREAD_POLL, "Thread Poll Info"); | |
44 | DEFINE_MTYPE_STATIC(LIB, THREAD_STATS, "Thread stats"); | |
4a1ab8e4 | 45 | |
960b9a53 | 46 | DECLARE_LIST(thread_list, struct thread, threaditem); |
c284542b | 47 | |
aea25d1e MS |
48 | struct cancel_req { |
49 | int flags; | |
50 | struct thread *thread; | |
51 | void *eventobj; | |
52 | struct thread **threadref; | |
53 | }; | |
54 | ||
55 | /* Flags for task cancellation */ | |
56 | #define THREAD_CANCEL_FLAG_READY 0x01 | |
57 | ||
27d29ced DL |
58 | static int thread_timer_cmp(const struct thread *a, const struct thread *b) |
59 | { | |
60 | if (a->u.sands.tv_sec < b->u.sands.tv_sec) | |
61 | return -1; | |
62 | if (a->u.sands.tv_sec > b->u.sands.tv_sec) | |
63 | return 1; | |
64 | if (a->u.sands.tv_usec < b->u.sands.tv_usec) | |
65 | return -1; | |
66 | if (a->u.sands.tv_usec > b->u.sands.tv_usec) | |
67 | return 1; | |
68 | return 0; | |
69 | } | |
70 | ||
960b9a53 | 71 | DECLARE_HEAP(thread_timer_list, struct thread, timeritem, thread_timer_cmp); |
27d29ced | 72 | |
3b96b781 HT |
73 | #if defined(__APPLE__) |
74 | #include <mach/mach.h> | |
75 | #include <mach/mach_time.h> | |
76 | #endif | |
77 | ||
d62a17ae | 78 | #define AWAKEN(m) \ |
79 | do { \ | |
2b64873d | 80 | const unsigned char wakebyte = 0x01; \ |
d62a17ae | 81 | write(m->io_pipe[1], &wakebyte, 1); \ |
82 | } while (0); | |
3bf2673b | 83 | |
62f44022 | 84 | /* control variable for initializer */ |
c17faa4b | 85 | static pthread_once_t init_once = PTHREAD_ONCE_INIT; |
e0bebc7c | 86 | pthread_key_t thread_current; |
6b0655a2 | 87 | |
c17faa4b | 88 | static pthread_mutex_t masters_mtx = PTHREAD_MUTEX_INITIALIZER; |
62f44022 QY |
89 | static struct list *masters; |
90 | ||
6655966d | 91 | static void thread_free(struct thread_master *master, struct thread *thread); |
6b0655a2 | 92 | |
45f01188 DL |
93 | #ifndef EXCLUDE_CPU_TIME |
94 | #define EXCLUDE_CPU_TIME 0 | |
95 | #endif | |
96 | #ifndef CONSUMED_TIME_CHECK | |
97 | #define CONSUMED_TIME_CHECK 0 | |
98 | #endif | |
99 | ||
100 | bool cputime_enabled = !EXCLUDE_CPU_TIME; | |
101 | unsigned long cputime_threshold = CONSUMED_TIME_CHECK; | |
102 | unsigned long walltime_threshold = CONSUMED_TIME_CHECK; | |
103 | ||
62f44022 | 104 | /* CLI start ---------------------------------------------------------------- */ |
45f01188 DL |
105 | #ifndef VTYSH_EXTRACT_PL |
106 | #include "lib/thread_clippy.c" | |
107 | #endif | |
108 | ||
d8b87afe | 109 | static unsigned int cpu_record_hash_key(const struct cpu_thread_history *a) |
e04ab74d | 110 | { |
883cc51d | 111 | int size = sizeof(a->func); |
bd74dc61 DS |
112 | |
113 | return jhash(&a->func, size, 0); | |
e04ab74d | 114 | } |
115 | ||
74df8d6d | 116 | static bool cpu_record_hash_cmp(const struct cpu_thread_history *a, |
d62a17ae | 117 | const struct cpu_thread_history *b) |
e04ab74d | 118 | { |
d62a17ae | 119 | return a->func == b->func; |
e04ab74d | 120 | } |
121 | ||
d62a17ae | 122 | static void *cpu_record_hash_alloc(struct cpu_thread_history *a) |
e04ab74d | 123 | { |
d62a17ae | 124 | struct cpu_thread_history *new; |
125 | new = XCALLOC(MTYPE_THREAD_STATS, sizeof(struct cpu_thread_history)); | |
126 | new->func = a->func; | |
127 | new->funcname = a->funcname; | |
128 | return new; | |
e04ab74d | 129 | } |
130 | ||
d62a17ae | 131 | static void cpu_record_hash_free(void *a) |
228da428 | 132 | { |
d62a17ae | 133 | struct cpu_thread_history *hist = a; |
134 | ||
135 | XFREE(MTYPE_THREAD_STATS, hist); | |
228da428 CC |
136 | } |
137 | ||
d62a17ae | 138 | static void vty_out_cpu_thread_history(struct vty *vty, |
139 | struct cpu_thread_history *a) | |
e04ab74d | 140 | { |
1dd08c22 DS |
141 | vty_out(vty, |
142 | "%5zu %10zu.%03zu %9zu %8zu %9zu %8zu %9zu %9zu %9zu %10zu", | |
72327cf3 | 143 | a->total_active, a->cpu.total / 1000, a->cpu.total % 1000, |
9b8e01ca DS |
144 | a->total_calls, (a->cpu.total / a->total_calls), a->cpu.max, |
145 | (a->real.total / a->total_calls), a->real.max, | |
1dd08c22 | 146 | a->total_cpu_warn, a->total_wall_warn, a->total_starv_warn); |
9b8e01ca | 147 | vty_out(vty, " %c%c%c%c%c %s\n", |
d62a17ae | 148 | a->types & (1 << THREAD_READ) ? 'R' : ' ', |
149 | a->types & (1 << THREAD_WRITE) ? 'W' : ' ', | |
150 | a->types & (1 << THREAD_TIMER) ? 'T' : ' ', | |
151 | a->types & (1 << THREAD_EVENT) ? 'E' : ' ', | |
152 | a->types & (1 << THREAD_EXECUTE) ? 'X' : ' ', a->funcname); | |
e04ab74d | 153 | } |
154 | ||
e3b78da8 | 155 | static void cpu_record_hash_print(struct hash_bucket *bucket, void *args[]) |
e04ab74d | 156 | { |
d62a17ae | 157 | struct cpu_thread_history *totals = args[0]; |
fbcac826 | 158 | struct cpu_thread_history copy; |
d62a17ae | 159 | struct vty *vty = args[1]; |
fbcac826 | 160 | uint8_t *filter = args[2]; |
d62a17ae | 161 | |
162 | struct cpu_thread_history *a = bucket->data; | |
163 | ||
fbcac826 QY |
164 | copy.total_active = |
165 | atomic_load_explicit(&a->total_active, memory_order_seq_cst); | |
166 | copy.total_calls = | |
167 | atomic_load_explicit(&a->total_calls, memory_order_seq_cst); | |
9b8e01ca DS |
168 | copy.total_cpu_warn = |
169 | atomic_load_explicit(&a->total_cpu_warn, memory_order_seq_cst); | |
170 | copy.total_wall_warn = | |
171 | atomic_load_explicit(&a->total_wall_warn, memory_order_seq_cst); | |
1dd08c22 DS |
172 | copy.total_starv_warn = atomic_load_explicit(&a->total_starv_warn, |
173 | memory_order_seq_cst); | |
fbcac826 QY |
174 | copy.cpu.total = |
175 | atomic_load_explicit(&a->cpu.total, memory_order_seq_cst); | |
176 | copy.cpu.max = atomic_load_explicit(&a->cpu.max, memory_order_seq_cst); | |
177 | copy.real.total = | |
178 | atomic_load_explicit(&a->real.total, memory_order_seq_cst); | |
179 | copy.real.max = | |
180 | atomic_load_explicit(&a->real.max, memory_order_seq_cst); | |
181 | copy.types = atomic_load_explicit(&a->types, memory_order_seq_cst); | |
182 | copy.funcname = a->funcname; | |
183 | ||
184 | if (!(copy.types & *filter)) | |
d62a17ae | 185 | return; |
fbcac826 QY |
186 | |
187 | vty_out_cpu_thread_history(vty, ©); | |
188 | totals->total_active += copy.total_active; | |
189 | totals->total_calls += copy.total_calls; | |
9b8e01ca DS |
190 | totals->total_cpu_warn += copy.total_cpu_warn; |
191 | totals->total_wall_warn += copy.total_wall_warn; | |
1dd08c22 | 192 | totals->total_starv_warn += copy.total_starv_warn; |
fbcac826 QY |
193 | totals->real.total += copy.real.total; |
194 | if (totals->real.max < copy.real.max) | |
195 | totals->real.max = copy.real.max; | |
196 | totals->cpu.total += copy.cpu.total; | |
197 | if (totals->cpu.max < copy.cpu.max) | |
198 | totals->cpu.max = copy.cpu.max; | |
e04ab74d | 199 | } |
200 | ||
fbcac826 | 201 | static void cpu_record_print(struct vty *vty, uint8_t filter) |
e04ab74d | 202 | { |
d62a17ae | 203 | struct cpu_thread_history tmp; |
204 | void *args[3] = {&tmp, vty, &filter}; | |
205 | struct thread_master *m; | |
206 | struct listnode *ln; | |
207 | ||
45f01188 DL |
208 | if (!cputime_enabled) |
209 | vty_out(vty, | |
210 | "\n" | |
211 | "Collecting CPU time statistics is currently disabled. Following statistics\n" | |
212 | "will be zero or may display data from when collection was enabled. Use the\n" | |
213 | " \"service cputime-stats\" command to start collecting data.\n" | |
214 | "\nCounters and wallclock times are always maintained and should be accurate.\n"); | |
215 | ||
0d6f7fd6 | 216 | memset(&tmp, 0, sizeof(tmp)); |
d62a17ae | 217 | tmp.funcname = "TOTAL"; |
218 | tmp.types = filter; | |
219 | ||
00dffa8c | 220 | frr_with_mutex(&masters_mtx) { |
d62a17ae | 221 | for (ALL_LIST_ELEMENTS_RO(masters, ln, m)) { |
222 | const char *name = m->name ? m->name : "main"; | |
223 | ||
224 | char underline[strlen(name) + 1]; | |
225 | memset(underline, '-', sizeof(underline)); | |
4f113d60 | 226 | underline[sizeof(underline) - 1] = '\0'; |
d62a17ae | 227 | |
228 | vty_out(vty, "\n"); | |
229 | vty_out(vty, "Showing statistics for pthread %s\n", | |
230 | name); | |
231 | vty_out(vty, "-------------------------------%s\n", | |
232 | underline); | |
84d951d0 | 233 | vty_out(vty, "%30s %18s %18s\n", "", |
d62a17ae | 234 | "CPU (user+system):", "Real (wall-clock):"); |
235 | vty_out(vty, | |
236 | "Active Runtime(ms) Invoked Avg uSec Max uSecs"); | |
237 | vty_out(vty, " Avg uSec Max uSecs"); | |
1dd08c22 DS |
238 | vty_out(vty, |
239 | " CPU_Warn Wall_Warn Starv_Warn Type Thread\n"); | |
d62a17ae | 240 | |
241 | if (m->cpu_record->count) | |
242 | hash_iterate( | |
243 | m->cpu_record, | |
e3b78da8 | 244 | (void (*)(struct hash_bucket *, |
d62a17ae | 245 | void *))cpu_record_hash_print, |
246 | args); | |
247 | else | |
248 | vty_out(vty, "No data to display yet.\n"); | |
249 | ||
250 | vty_out(vty, "\n"); | |
251 | } | |
252 | } | |
d62a17ae | 253 | |
254 | vty_out(vty, "\n"); | |
255 | vty_out(vty, "Total thread statistics\n"); | |
256 | vty_out(vty, "-------------------------\n"); | |
84d951d0 | 257 | vty_out(vty, "%30s %18s %18s\n", "", |
d62a17ae | 258 | "CPU (user+system):", "Real (wall-clock):"); |
259 | vty_out(vty, "Active Runtime(ms) Invoked Avg uSec Max uSecs"); | |
9b8e01ca | 260 | vty_out(vty, " Avg uSec Max uSecs CPU_Warn Wall_Warn"); |
d62a17ae | 261 | vty_out(vty, " Type Thread\n"); |
262 | ||
263 | if (tmp.total_calls > 0) | |
264 | vty_out_cpu_thread_history(vty, &tmp); | |
e04ab74d | 265 | } |
266 | ||
e3b78da8 | 267 | static void cpu_record_hash_clear(struct hash_bucket *bucket, void *args[]) |
e276eb82 | 268 | { |
fbcac826 | 269 | uint8_t *filter = args[0]; |
d62a17ae | 270 | struct hash *cpu_record = args[1]; |
271 | ||
272 | struct cpu_thread_history *a = bucket->data; | |
62f44022 | 273 | |
d62a17ae | 274 | if (!(a->types & *filter)) |
275 | return; | |
f48f65d2 | 276 | |
d62a17ae | 277 | hash_release(cpu_record, bucket->data); |
e276eb82 PJ |
278 | } |
279 | ||
fbcac826 | 280 | static void cpu_record_clear(uint8_t filter) |
e276eb82 | 281 | { |
fbcac826 | 282 | uint8_t *tmp = &filter; |
d62a17ae | 283 | struct thread_master *m; |
284 | struct listnode *ln; | |
285 | ||
00dffa8c | 286 | frr_with_mutex(&masters_mtx) { |
d62a17ae | 287 | for (ALL_LIST_ELEMENTS_RO(masters, ln, m)) { |
00dffa8c | 288 | frr_with_mutex(&m->mtx) { |
d62a17ae | 289 | void *args[2] = {tmp, m->cpu_record}; |
290 | hash_iterate( | |
291 | m->cpu_record, | |
e3b78da8 | 292 | (void (*)(struct hash_bucket *, |
d62a17ae | 293 | void *))cpu_record_hash_clear, |
294 | args); | |
295 | } | |
d62a17ae | 296 | } |
297 | } | |
62f44022 QY |
298 | } |
299 | ||
fbcac826 | 300 | static uint8_t parse_filter(const char *filterstr) |
62f44022 | 301 | { |
d62a17ae | 302 | int i = 0; |
303 | int filter = 0; | |
304 | ||
305 | while (filterstr[i] != '\0') { | |
306 | switch (filterstr[i]) { | |
307 | case 'r': | |
308 | case 'R': | |
309 | filter |= (1 << THREAD_READ); | |
310 | break; | |
311 | case 'w': | |
312 | case 'W': | |
313 | filter |= (1 << THREAD_WRITE); | |
314 | break; | |
315 | case 't': | |
316 | case 'T': | |
317 | filter |= (1 << THREAD_TIMER); | |
318 | break; | |
319 | case 'e': | |
320 | case 'E': | |
321 | filter |= (1 << THREAD_EVENT); | |
322 | break; | |
323 | case 'x': | |
324 | case 'X': | |
325 | filter |= (1 << THREAD_EXECUTE); | |
326 | break; | |
327 | default: | |
328 | break; | |
329 | } | |
330 | ++i; | |
331 | } | |
332 | return filter; | |
62f44022 QY |
333 | } |
334 | ||
ee4dcee8 DL |
335 | DEFUN_NOSH (show_thread_cpu, |
336 | show_thread_cpu_cmd, | |
337 | "show thread cpu [FILTER]", | |
338 | SHOW_STR | |
339 | "Thread information\n" | |
340 | "Thread CPU usage\n" | |
341 | "Display filter (rwtex)\n") | |
62f44022 | 342 | { |
fbcac826 | 343 | uint8_t filter = (uint8_t)-1U; |
d62a17ae | 344 | int idx = 0; |
345 | ||
346 | if (argv_find(argv, argc, "FILTER", &idx)) { | |
347 | filter = parse_filter(argv[idx]->arg); | |
348 | if (!filter) { | |
349 | vty_out(vty, | |
3efd0893 | 350 | "Invalid filter \"%s\" specified; must contain at leastone of 'RWTEXB'\n", |
d62a17ae | 351 | argv[idx]->arg); |
352 | return CMD_WARNING; | |
353 | } | |
354 | } | |
355 | ||
356 | cpu_record_print(vty, filter); | |
357 | return CMD_SUCCESS; | |
e276eb82 | 358 | } |
45f01188 DL |
359 | |
360 | DEFPY (service_cputime_stats, | |
361 | service_cputime_stats_cmd, | |
362 | "[no] service cputime-stats", | |
363 | NO_STR | |
364 | "Set up miscellaneous service\n" | |
365 | "Collect CPU usage statistics\n") | |
366 | { | |
367 | cputime_enabled = !no; | |
368 | return CMD_SUCCESS; | |
369 | } | |
370 | ||
371 | DEFPY (service_cputime_warning, | |
372 | service_cputime_warning_cmd, | |
373 | "[no] service cputime-warning (1-4294967295)", | |
374 | NO_STR | |
375 | "Set up miscellaneous service\n" | |
376 | "Warn for tasks exceeding CPU usage threshold\n" | |
377 | "Warning threshold in milliseconds\n") | |
378 | { | |
379 | if (no) | |
380 | cputime_threshold = 0; | |
381 | else | |
382 | cputime_threshold = cputime_warning * 1000; | |
383 | return CMD_SUCCESS; | |
384 | } | |
385 | ||
386 | ALIAS (service_cputime_warning, | |
387 | no_service_cputime_warning_cmd, | |
388 | "no service cputime-warning", | |
389 | NO_STR | |
390 | "Set up miscellaneous service\n" | |
391 | "Warn for tasks exceeding CPU usage threshold\n") | |
392 | ||
393 | DEFPY (service_walltime_warning, | |
394 | service_walltime_warning_cmd, | |
395 | "[no] service walltime-warning (1-4294967295)", | |
396 | NO_STR | |
397 | "Set up miscellaneous service\n" | |
398 | "Warn for tasks exceeding total wallclock threshold\n" | |
399 | "Warning threshold in milliseconds\n") | |
400 | { | |
401 | if (no) | |
402 | walltime_threshold = 0; | |
403 | else | |
404 | walltime_threshold = walltime_warning * 1000; | |
405 | return CMD_SUCCESS; | |
406 | } | |
407 | ||
408 | ALIAS (service_walltime_warning, | |
409 | no_service_walltime_warning_cmd, | |
410 | "no service walltime-warning", | |
411 | NO_STR | |
412 | "Set up miscellaneous service\n" | |
413 | "Warn for tasks exceeding total wallclock threshold\n") | |
e276eb82 | 414 | |
8872626b DS |
415 | static void show_thread_poll_helper(struct vty *vty, struct thread_master *m) |
416 | { | |
417 | const char *name = m->name ? m->name : "main"; | |
418 | char underline[strlen(name) + 1]; | |
a0b36ae6 | 419 | struct thread *thread; |
8872626b DS |
420 | uint32_t i; |
421 | ||
422 | memset(underline, '-', sizeof(underline)); | |
423 | underline[sizeof(underline) - 1] = '\0'; | |
424 | ||
425 | vty_out(vty, "\nShowing poll FD's for %s\n", name); | |
426 | vty_out(vty, "----------------------%s\n", underline); | |
6c19478a DS |
427 | vty_out(vty, "Count: %u/%d\n", (uint32_t)m->handler.pfdcount, |
428 | m->fd_limit); | |
a0b36ae6 DS |
429 | for (i = 0; i < m->handler.pfdcount; i++) { |
430 | vty_out(vty, "\t%6d fd:%6d events:%2d revents:%2d\t\t", i, | |
431 | m->handler.pfds[i].fd, m->handler.pfds[i].events, | |
8872626b | 432 | m->handler.pfds[i].revents); |
a0b36ae6 DS |
433 | |
434 | if (m->handler.pfds[i].events & POLLIN) { | |
435 | thread = m->read[m->handler.pfds[i].fd]; | |
436 | ||
437 | if (!thread) | |
438 | vty_out(vty, "ERROR "); | |
439 | else | |
60a3efec | 440 | vty_out(vty, "%s ", thread->xref->funcname); |
a0b36ae6 DS |
441 | } else |
442 | vty_out(vty, " "); | |
443 | ||
444 | if (m->handler.pfds[i].events & POLLOUT) { | |
445 | thread = m->write[m->handler.pfds[i].fd]; | |
446 | ||
447 | if (!thread) | |
448 | vty_out(vty, "ERROR\n"); | |
449 | else | |
60a3efec | 450 | vty_out(vty, "%s\n", thread->xref->funcname); |
a0b36ae6 DS |
451 | } else |
452 | vty_out(vty, "\n"); | |
453 | } | |
8872626b DS |
454 | } |
455 | ||
ee4dcee8 DL |
456 | DEFUN_NOSH (show_thread_poll, |
457 | show_thread_poll_cmd, | |
458 | "show thread poll", | |
459 | SHOW_STR | |
460 | "Thread information\n" | |
461 | "Show poll FD's and information\n") | |
8872626b DS |
462 | { |
463 | struct listnode *node; | |
464 | struct thread_master *m; | |
465 | ||
00dffa8c | 466 | frr_with_mutex(&masters_mtx) { |
8872626b DS |
467 | for (ALL_LIST_ELEMENTS_RO(masters, node, m)) { |
468 | show_thread_poll_helper(vty, m); | |
469 | } | |
470 | } | |
8872626b DS |
471 | |
472 | return CMD_SUCCESS; | |
473 | } | |
474 | ||
475 | ||
49d41a26 DS |
476 | DEFUN (clear_thread_cpu, |
477 | clear_thread_cpu_cmd, | |
478 | "clear thread cpu [FILTER]", | |
62f44022 | 479 | "Clear stored data in all pthreads\n" |
49d41a26 DS |
480 | "Thread information\n" |
481 | "Thread CPU usage\n" | |
482 | "Display filter (rwtexb)\n") | |
e276eb82 | 483 | { |
fbcac826 | 484 | uint8_t filter = (uint8_t)-1U; |
d62a17ae | 485 | int idx = 0; |
486 | ||
487 | if (argv_find(argv, argc, "FILTER", &idx)) { | |
488 | filter = parse_filter(argv[idx]->arg); | |
489 | if (!filter) { | |
490 | vty_out(vty, | |
3efd0893 | 491 | "Invalid filter \"%s\" specified; must contain at leastone of 'RWTEXB'\n", |
d62a17ae | 492 | argv[idx]->arg); |
493 | return CMD_WARNING; | |
494 | } | |
495 | } | |
496 | ||
497 | cpu_record_clear(filter); | |
498 | return CMD_SUCCESS; | |
e276eb82 | 499 | } |
6b0655a2 | 500 | |
22f31b8c DS |
501 | static void show_thread_timers_helper(struct vty *vty, struct thread_master *m) |
502 | { | |
503 | const char *name = m->name ? m->name : "main"; | |
504 | char underline[strlen(name) + 1]; | |
505 | struct thread *thread; | |
506 | ||
507 | memset(underline, '-', sizeof(underline)); | |
508 | underline[sizeof(underline) - 1] = '\0'; | |
509 | ||
510 | vty_out(vty, "\nShowing timers for %s\n", name); | |
511 | vty_out(vty, "-------------------%s\n", underline); | |
512 | ||
513 | frr_each (thread_timer_list, &m->timer, thread) { | |
514 | vty_out(vty, " %-50s%pTH\n", thread->hist->funcname, thread); | |
515 | } | |
516 | } | |
517 | ||
518 | DEFPY_NOSH (show_thread_timers, | |
519 | show_thread_timers_cmd, | |
520 | "show thread timers", | |
521 | SHOW_STR | |
522 | "Thread information\n" | |
523 | "Show all timers and how long they have in the system\n") | |
524 | { | |
525 | struct listnode *node; | |
526 | struct thread_master *m; | |
527 | ||
528 | frr_with_mutex (&masters_mtx) { | |
529 | for (ALL_LIST_ELEMENTS_RO(masters, node, m)) | |
530 | show_thread_timers_helper(vty, m); | |
531 | } | |
532 | ||
533 | return CMD_SUCCESS; | |
534 | } | |
535 | ||
d62a17ae | 536 | void thread_cmd_init(void) |
0b84f294 | 537 | { |
d62a17ae | 538 | install_element(VIEW_NODE, &show_thread_cpu_cmd); |
8872626b | 539 | install_element(VIEW_NODE, &show_thread_poll_cmd); |
d62a17ae | 540 | install_element(ENABLE_NODE, &clear_thread_cpu_cmd); |
45f01188 DL |
541 | |
542 | install_element(CONFIG_NODE, &service_cputime_stats_cmd); | |
543 | install_element(CONFIG_NODE, &service_cputime_warning_cmd); | |
544 | install_element(CONFIG_NODE, &no_service_cputime_warning_cmd); | |
545 | install_element(CONFIG_NODE, &service_walltime_warning_cmd); | |
546 | install_element(CONFIG_NODE, &no_service_walltime_warning_cmd); | |
22f31b8c DS |
547 | |
548 | install_element(VIEW_NODE, &show_thread_timers_cmd); | |
0b84f294 | 549 | } |
62f44022 QY |
550 | /* CLI end ------------------------------------------------------------------ */ |
551 | ||
0b84f294 | 552 | |
d62a17ae | 553 | static void cancelreq_del(void *cr) |
63ccb9cb | 554 | { |
d62a17ae | 555 | XFREE(MTYPE_TMP, cr); |
63ccb9cb QY |
556 | } |
557 | ||
e0bebc7c | 558 | /* initializer, only ever called once */ |
4d762f26 | 559 | static void initializer(void) |
e0bebc7c | 560 | { |
d62a17ae | 561 | pthread_key_create(&thread_current, NULL); |
e0bebc7c QY |
562 | } |
563 | ||
d62a17ae | 564 | struct thread_master *thread_master_create(const char *name) |
718e3744 | 565 | { |
d62a17ae | 566 | struct thread_master *rv; |
567 | struct rlimit limit; | |
568 | ||
569 | pthread_once(&init_once, &initializer); | |
570 | ||
571 | rv = XCALLOC(MTYPE_THREAD_MASTER, sizeof(struct thread_master)); | |
d62a17ae | 572 | |
573 | /* Initialize master mutex */ | |
574 | pthread_mutex_init(&rv->mtx, NULL); | |
575 | pthread_cond_init(&rv->cancel_cond, NULL); | |
576 | ||
577 | /* Set name */ | |
7ffcd8bd QY |
578 | name = name ? name : "default"; |
579 | rv->name = XSTRDUP(MTYPE_THREAD_MASTER, name); | |
d62a17ae | 580 | |
581 | /* Initialize I/O task data structures */ | |
1a9f340b MS |
582 | |
583 | /* Use configured limit if present, ulimit otherwise. */ | |
584 | rv->fd_limit = frr_get_fd_limit(); | |
585 | if (rv->fd_limit == 0) { | |
586 | getrlimit(RLIMIT_NOFILE, &limit); | |
587 | rv->fd_limit = (int)limit.rlim_cur; | |
588 | } | |
589 | ||
a6f235f3 DS |
590 | rv->read = XCALLOC(MTYPE_THREAD_POLL, |
591 | sizeof(struct thread *) * rv->fd_limit); | |
592 | ||
593 | rv->write = XCALLOC(MTYPE_THREAD_POLL, | |
594 | sizeof(struct thread *) * rv->fd_limit); | |
d62a17ae | 595 | |
7ffcd8bd QY |
596 | char tmhashname[strlen(name) + 32]; |
597 | snprintf(tmhashname, sizeof(tmhashname), "%s - threadmaster event hash", | |
598 | name); | |
bd74dc61 | 599 | rv->cpu_record = hash_create_size( |
d8b87afe | 600 | 8, (unsigned int (*)(const void *))cpu_record_hash_key, |
74df8d6d | 601 | (bool (*)(const void *, const void *))cpu_record_hash_cmp, |
7ffcd8bd | 602 | tmhashname); |
d62a17ae | 603 | |
c284542b DL |
604 | thread_list_init(&rv->event); |
605 | thread_list_init(&rv->ready); | |
606 | thread_list_init(&rv->unuse); | |
27d29ced | 607 | thread_timer_list_init(&rv->timer); |
d62a17ae | 608 | |
609 | /* Initialize thread_fetch() settings */ | |
610 | rv->spin = true; | |
611 | rv->handle_signals = true; | |
612 | ||
613 | /* Set pthread owner, should be updated by actual owner */ | |
614 | rv->owner = pthread_self(); | |
615 | rv->cancel_req = list_new(); | |
616 | rv->cancel_req->del = cancelreq_del; | |
617 | rv->canceled = true; | |
618 | ||
619 | /* Initialize pipe poker */ | |
620 | pipe(rv->io_pipe); | |
621 | set_nonblocking(rv->io_pipe[0]); | |
622 | set_nonblocking(rv->io_pipe[1]); | |
623 | ||
624 | /* Initialize data structures for poll() */ | |
625 | rv->handler.pfdsize = rv->fd_limit; | |
626 | rv->handler.pfdcount = 0; | |
627 | rv->handler.pfds = XCALLOC(MTYPE_THREAD_MASTER, | |
628 | sizeof(struct pollfd) * rv->handler.pfdsize); | |
629 | rv->handler.copy = XCALLOC(MTYPE_THREAD_MASTER, | |
630 | sizeof(struct pollfd) * rv->handler.pfdsize); | |
631 | ||
eff09c66 | 632 | /* add to list of threadmasters */ |
00dffa8c | 633 | frr_with_mutex(&masters_mtx) { |
eff09c66 QY |
634 | if (!masters) |
635 | masters = list_new(); | |
636 | ||
d62a17ae | 637 | listnode_add(masters, rv); |
638 | } | |
d62a17ae | 639 | |
640 | return rv; | |
718e3744 | 641 | } |
642 | ||
d8a8a8de QY |
643 | void thread_master_set_name(struct thread_master *master, const char *name) |
644 | { | |
00dffa8c | 645 | frr_with_mutex(&master->mtx) { |
0a22ddfb | 646 | XFREE(MTYPE_THREAD_MASTER, master->name); |
d8a8a8de QY |
647 | master->name = XSTRDUP(MTYPE_THREAD_MASTER, name); |
648 | } | |
d8a8a8de QY |
649 | } |
650 | ||
6ed04aa2 DS |
651 | #define THREAD_UNUSED_DEPTH 10 |
652 | ||
718e3744 | 653 | /* Move thread to unuse list. */ |
d62a17ae | 654 | static void thread_add_unuse(struct thread_master *m, struct thread *thread) |
718e3744 | 655 | { |
6655966d RZ |
656 | pthread_mutex_t mtxc = thread->mtx; |
657 | ||
d62a17ae | 658 | assert(m != NULL && thread != NULL); |
d62a17ae | 659 | |
d62a17ae | 660 | thread->hist->total_active--; |
6ed04aa2 DS |
661 | memset(thread, 0, sizeof(struct thread)); |
662 | thread->type = THREAD_UNUSED; | |
663 | ||
6655966d RZ |
664 | /* Restore the thread mutex context. */ |
665 | thread->mtx = mtxc; | |
666 | ||
c284542b DL |
667 | if (thread_list_count(&m->unuse) < THREAD_UNUSED_DEPTH) { |
668 | thread_list_add_tail(&m->unuse, thread); | |
6655966d RZ |
669 | return; |
670 | } | |
671 | ||
672 | thread_free(m, thread); | |
718e3744 | 673 | } |
674 | ||
675 | /* Free all unused thread. */ | |
c284542b DL |
676 | static void thread_list_free(struct thread_master *m, |
677 | struct thread_list_head *list) | |
718e3744 | 678 | { |
d62a17ae | 679 | struct thread *t; |
d62a17ae | 680 | |
c284542b | 681 | while ((t = thread_list_pop(list))) |
6655966d | 682 | thread_free(m, t); |
718e3744 | 683 | } |
684 | ||
d62a17ae | 685 | static void thread_array_free(struct thread_master *m, |
686 | struct thread **thread_array) | |
308d14ae | 687 | { |
d62a17ae | 688 | struct thread *t; |
689 | int index; | |
690 | ||
691 | for (index = 0; index < m->fd_limit; ++index) { | |
692 | t = thread_array[index]; | |
693 | if (t) { | |
694 | thread_array[index] = NULL; | |
6655966d | 695 | thread_free(m, t); |
d62a17ae | 696 | } |
697 | } | |
a6f235f3 | 698 | XFREE(MTYPE_THREAD_POLL, thread_array); |
308d14ae DV |
699 | } |
700 | ||
495f0b13 DS |
701 | /* |
702 | * thread_master_free_unused | |
703 | * | |
704 | * As threads are finished with they are put on the | |
705 | * unuse list for later reuse. | |
706 | * If we are shutting down, Free up unused threads | |
707 | * So we can see if we forget to shut anything off | |
708 | */ | |
d62a17ae | 709 | void thread_master_free_unused(struct thread_master *m) |
495f0b13 | 710 | { |
00dffa8c | 711 | frr_with_mutex(&m->mtx) { |
d62a17ae | 712 | struct thread *t; |
c284542b | 713 | while ((t = thread_list_pop(&m->unuse))) |
6655966d | 714 | thread_free(m, t); |
d62a17ae | 715 | } |
495f0b13 DS |
716 | } |
717 | ||
718e3744 | 718 | /* Stop thread scheduler. */ |
d62a17ae | 719 | void thread_master_free(struct thread_master *m) |
718e3744 | 720 | { |
27d29ced DL |
721 | struct thread *t; |
722 | ||
00dffa8c | 723 | frr_with_mutex(&masters_mtx) { |
d62a17ae | 724 | listnode_delete(masters, m); |
eff09c66 | 725 | if (masters->count == 0) { |
6a154c88 | 726 | list_delete(&masters); |
eff09c66 | 727 | } |
d62a17ae | 728 | } |
d62a17ae | 729 | |
730 | thread_array_free(m, m->read); | |
731 | thread_array_free(m, m->write); | |
27d29ced DL |
732 | while ((t = thread_timer_list_pop(&m->timer))) |
733 | thread_free(m, t); | |
d62a17ae | 734 | thread_list_free(m, &m->event); |
735 | thread_list_free(m, &m->ready); | |
736 | thread_list_free(m, &m->unuse); | |
737 | pthread_mutex_destroy(&m->mtx); | |
33844bbe | 738 | pthread_cond_destroy(&m->cancel_cond); |
d62a17ae | 739 | close(m->io_pipe[0]); |
740 | close(m->io_pipe[1]); | |
6a154c88 | 741 | list_delete(&m->cancel_req); |
1a0a92ea | 742 | m->cancel_req = NULL; |
d62a17ae | 743 | |
744 | hash_clean(m->cpu_record, cpu_record_hash_free); | |
745 | hash_free(m->cpu_record); | |
746 | m->cpu_record = NULL; | |
747 | ||
0a22ddfb | 748 | XFREE(MTYPE_THREAD_MASTER, m->name); |
d62a17ae | 749 | XFREE(MTYPE_THREAD_MASTER, m->handler.pfds); |
750 | XFREE(MTYPE_THREAD_MASTER, m->handler.copy); | |
751 | XFREE(MTYPE_THREAD_MASTER, m); | |
718e3744 | 752 | } |
753 | ||
94202742 | 754 | /* Return remain time in milliseconds. */ |
78ca0342 | 755 | unsigned long thread_timer_remain_msec(struct thread *thread) |
718e3744 | 756 | { |
d62a17ae | 757 | int64_t remain; |
1189d95f | 758 | |
00dffa8c | 759 | frr_with_mutex(&thread->mtx) { |
78ca0342 | 760 | remain = monotime_until(&thread->u.sands, NULL) / 1000LL; |
d62a17ae | 761 | } |
1189d95f | 762 | |
d62a17ae | 763 | return remain < 0 ? 0 : remain; |
718e3744 | 764 | } |
765 | ||
78ca0342 CF |
766 | /* Return remain time in seconds. */ |
767 | unsigned long thread_timer_remain_second(struct thread *thread) | |
768 | { | |
769 | return thread_timer_remain_msec(thread) / 1000LL; | |
770 | } | |
771 | ||
d62a17ae | 772 | struct timeval thread_timer_remain(struct thread *thread) |
6ac44687 | 773 | { |
d62a17ae | 774 | struct timeval remain; |
00dffa8c | 775 | frr_with_mutex(&thread->mtx) { |
d62a17ae | 776 | monotime_until(&thread->u.sands, &remain); |
777 | } | |
d62a17ae | 778 | return remain; |
6ac44687 CF |
779 | } |
780 | ||
0447957e AK |
781 | static int time_hhmmss(char *buf, int buf_size, long sec) |
782 | { | |
783 | long hh; | |
784 | long mm; | |
785 | int wr; | |
786 | ||
642ac49d | 787 | assert(buf_size >= 8); |
0447957e AK |
788 | |
789 | hh = sec / 3600; | |
790 | sec %= 3600; | |
791 | mm = sec / 60; | |
792 | sec %= 60; | |
793 | ||
794 | wr = snprintf(buf, buf_size, "%02ld:%02ld:%02ld", hh, mm, sec); | |
795 | ||
796 | return wr != 8; | |
797 | } | |
798 | ||
799 | char *thread_timer_to_hhmmss(char *buf, int buf_size, | |
800 | struct thread *t_timer) | |
801 | { | |
802 | if (t_timer) { | |
803 | time_hhmmss(buf, buf_size, | |
804 | thread_timer_remain_second(t_timer)); | |
805 | } else { | |
806 | snprintf(buf, buf_size, "--:--:--"); | |
807 | } | |
808 | return buf; | |
809 | } | |
810 | ||
718e3744 | 811 | /* Get new thread. */ |
d7c0a89a | 812 | static struct thread *thread_get(struct thread_master *m, uint8_t type, |
cc9f21da | 813 | void (*func)(struct thread *), void *arg, |
60a3efec | 814 | const struct xref_threadsched *xref) |
718e3744 | 815 | { |
c284542b | 816 | struct thread *thread = thread_list_pop(&m->unuse); |
d62a17ae | 817 | struct cpu_thread_history tmp; |
818 | ||
819 | if (!thread) { | |
820 | thread = XCALLOC(MTYPE_THREAD, sizeof(struct thread)); | |
821 | /* mutex only needs to be initialized at struct creation. */ | |
822 | pthread_mutex_init(&thread->mtx, NULL); | |
823 | m->alloc++; | |
824 | } | |
825 | ||
826 | thread->type = type; | |
827 | thread->add_type = type; | |
828 | thread->master = m; | |
829 | thread->arg = arg; | |
d62a17ae | 830 | thread->yield = THREAD_YIELD_TIME_SLOT; /* default */ |
831 | thread->ref = NULL; | |
e8b3a2f7 | 832 | thread->ignore_timer_late = false; |
d62a17ae | 833 | |
834 | /* | |
835 | * So if the passed in funcname is not what we have | |
836 | * stored that means the thread->hist needs to be | |
837 | * updated. We keep the last one around in unused | |
838 | * under the assumption that we are probably | |
839 | * going to immediately allocate the same | |
840 | * type of thread. | |
841 | * This hopefully saves us some serious | |
842 | * hash_get lookups. | |
843 | */ | |
60a3efec DL |
844 | if ((thread->xref && thread->xref->funcname != xref->funcname) |
845 | || thread->func != func) { | |
d62a17ae | 846 | tmp.func = func; |
60a3efec | 847 | tmp.funcname = xref->funcname; |
d62a17ae | 848 | thread->hist = |
849 | hash_get(m->cpu_record, &tmp, | |
850 | (void *(*)(void *))cpu_record_hash_alloc); | |
851 | } | |
852 | thread->hist->total_active++; | |
853 | thread->func = func; | |
60a3efec | 854 | thread->xref = xref; |
d62a17ae | 855 | |
856 | return thread; | |
718e3744 | 857 | } |
858 | ||
6655966d RZ |
859 | static void thread_free(struct thread_master *master, struct thread *thread) |
860 | { | |
861 | /* Update statistics. */ | |
862 | assert(master->alloc > 0); | |
863 | master->alloc--; | |
864 | ||
865 | /* Free allocated resources. */ | |
866 | pthread_mutex_destroy(&thread->mtx); | |
867 | XFREE(MTYPE_THREAD, thread); | |
868 | } | |
869 | ||
d81ca9a3 MS |
870 | static int fd_poll(struct thread_master *m, const struct timeval *timer_wait, |
871 | bool *eintr_p) | |
209a72a6 | 872 | { |
d81ca9a3 MS |
873 | sigset_t origsigs; |
874 | unsigned char trash[64]; | |
875 | nfds_t count = m->handler.copycount; | |
876 | ||
d279ef57 DS |
877 | /* |
878 | * If timer_wait is null here, that means poll() should block | |
879 | * indefinitely, unless the thread_master has overridden it by setting | |
d62a17ae | 880 | * ->selectpoll_timeout. |
d279ef57 | 881 | * |
d62a17ae | 882 | * If the value is positive, it specifies the maximum number of |
d279ef57 DS |
883 | * milliseconds to wait. If the timeout is -1, it specifies that |
884 | * we should never wait and always return immediately even if no | |
885 | * event is detected. If the value is zero, the behavior is default. | |
886 | */ | |
d62a17ae | 887 | int timeout = -1; |
888 | ||
889 | /* number of file descriptors with events */ | |
890 | int num; | |
891 | ||
892 | if (timer_wait != NULL | |
893 | && m->selectpoll_timeout == 0) // use the default value | |
894 | timeout = (timer_wait->tv_sec * 1000) | |
895 | + (timer_wait->tv_usec / 1000); | |
896 | else if (m->selectpoll_timeout > 0) // use the user's timeout | |
897 | timeout = m->selectpoll_timeout; | |
898 | else if (m->selectpoll_timeout | |
899 | < 0) // effect a poll (return immediately) | |
900 | timeout = 0; | |
901 | ||
0bdeb5e5 | 902 | zlog_tls_buffer_flush(); |
3e41733f DL |
903 | rcu_read_unlock(); |
904 | rcu_assert_read_unlocked(); | |
905 | ||
d62a17ae | 906 | /* add poll pipe poker */ |
d81ca9a3 MS |
907 | assert(count + 1 < m->handler.pfdsize); |
908 | m->handler.copy[count].fd = m->io_pipe[0]; | |
909 | m->handler.copy[count].events = POLLIN; | |
910 | m->handler.copy[count].revents = 0x00; | |
911 | ||
912 | /* We need to deal with a signal-handling race here: we | |
913 | * don't want to miss a crucial signal, such as SIGTERM or SIGINT, | |
914 | * that may arrive just before we enter poll(). We will block the | |
915 | * key signals, then check whether any have arrived - if so, we return | |
916 | * before calling poll(). If not, we'll re-enable the signals | |
917 | * in the ppoll() call. | |
918 | */ | |
919 | ||
920 | sigemptyset(&origsigs); | |
921 | if (m->handle_signals) { | |
922 | /* Main pthread that handles the app signals */ | |
923 | if (frr_sigevent_check(&origsigs)) { | |
924 | /* Signal to process - restore signal mask and return */ | |
925 | pthread_sigmask(SIG_SETMASK, &origsigs, NULL); | |
926 | num = -1; | |
927 | *eintr_p = true; | |
928 | goto done; | |
929 | } | |
930 | } else { | |
931 | /* Don't make any changes for the non-main pthreads */ | |
932 | pthread_sigmask(SIG_SETMASK, NULL, &origsigs); | |
933 | } | |
d62a17ae | 934 | |
d81ca9a3 MS |
935 | #if defined(HAVE_PPOLL) |
936 | struct timespec ts, *tsp; | |
937 | ||
938 | if (timeout >= 0) { | |
939 | ts.tv_sec = timeout / 1000; | |
940 | ts.tv_nsec = (timeout % 1000) * 1000000; | |
941 | tsp = &ts; | |
942 | } else | |
943 | tsp = NULL; | |
944 | ||
945 | num = ppoll(m->handler.copy, count + 1, tsp, &origsigs); | |
946 | pthread_sigmask(SIG_SETMASK, &origsigs, NULL); | |
947 | #else | |
948 | /* Not ideal - there is a race after we restore the signal mask */ | |
949 | pthread_sigmask(SIG_SETMASK, &origsigs, NULL); | |
950 | num = poll(m->handler.copy, count + 1, timeout); | |
951 | #endif | |
d62a17ae | 952 | |
d81ca9a3 MS |
953 | done: |
954 | ||
955 | if (num < 0 && errno == EINTR) | |
956 | *eintr_p = true; | |
957 | ||
958 | if (num > 0 && m->handler.copy[count].revents != 0 && num--) | |
d62a17ae | 959 | while (read(m->io_pipe[0], &trash, sizeof(trash)) > 0) |
960 | ; | |
961 | ||
3e41733f DL |
962 | rcu_read_lock(); |
963 | ||
d62a17ae | 964 | return num; |
209a72a6 DS |
965 | } |
966 | ||
718e3744 | 967 | /* Add new read thread. */ |
ee1455dd IR |
968 | void _thread_add_read_write(const struct xref_threadsched *xref, |
969 | struct thread_master *m, | |
cc9f21da | 970 | void (*func)(struct thread *), void *arg, int fd, |
ee1455dd | 971 | struct thread **t_ptr) |
718e3744 | 972 | { |
60a3efec | 973 | int dir = xref->thread_type; |
d62a17ae | 974 | struct thread *thread = NULL; |
1ef14bee | 975 | struct thread **thread_array; |
d62a17ae | 976 | |
abf96a87 | 977 | if (dir == THREAD_READ) |
6c3aa850 DL |
978 | frrtrace(9, frr_libfrr, schedule_read, m, |
979 | xref->funcname, xref->xref.file, xref->xref.line, | |
980 | t_ptr, fd, 0, arg, 0); | |
abf96a87 | 981 | else |
6c3aa850 DL |
982 | frrtrace(9, frr_libfrr, schedule_write, m, |
983 | xref->funcname, xref->xref.file, xref->xref.line, | |
984 | t_ptr, fd, 0, arg, 0); | |
abf96a87 | 985 | |
188acbb9 DS |
986 | assert(fd >= 0); |
987 | if (fd >= m->fd_limit) | |
988 | assert(!"Number of FD's open is greater than FRR currently configured to handle, aborting"); | |
989 | ||
00dffa8c DL |
990 | frr_with_mutex(&m->mtx) { |
991 | if (t_ptr && *t_ptr) | |
992 | // thread is already scheduled; don't reschedule | |
993 | break; | |
d62a17ae | 994 | |
995 | /* default to a new pollfd */ | |
996 | nfds_t queuepos = m->handler.pfdcount; | |
997 | ||
1ef14bee DS |
998 | if (dir == THREAD_READ) |
999 | thread_array = m->read; | |
1000 | else | |
1001 | thread_array = m->write; | |
1002 | ||
d62a17ae | 1003 | /* if we already have a pollfd for our file descriptor, find and |
1004 | * use it */ | |
1005 | for (nfds_t i = 0; i < m->handler.pfdcount; i++) | |
1006 | if (m->handler.pfds[i].fd == fd) { | |
1007 | queuepos = i; | |
1ef14bee DS |
1008 | |
1009 | #ifdef DEV_BUILD | |
1010 | /* | |
1011 | * What happens if we have a thread already | |
1012 | * created for this event? | |
1013 | */ | |
1014 | if (thread_array[fd]) | |
1015 | assert(!"Thread already scheduled for file descriptor"); | |
1016 | #endif | |
d62a17ae | 1017 | break; |
1018 | } | |
1019 | ||
1020 | /* make sure we have room for this fd + pipe poker fd */ | |
1021 | assert(queuepos + 1 < m->handler.pfdsize); | |
1022 | ||
60a3efec | 1023 | thread = thread_get(m, dir, func, arg, xref); |
d62a17ae | 1024 | |
1025 | m->handler.pfds[queuepos].fd = fd; | |
1026 | m->handler.pfds[queuepos].events |= | |
1027 | (dir == THREAD_READ ? POLLIN : POLLOUT); | |
1028 | ||
1029 | if (queuepos == m->handler.pfdcount) | |
1030 | m->handler.pfdcount++; | |
1031 | ||
1032 | if (thread) { | |
00dffa8c | 1033 | frr_with_mutex(&thread->mtx) { |
d62a17ae | 1034 | thread->u.fd = fd; |
1ef14bee | 1035 | thread_array[thread->u.fd] = thread; |
d62a17ae | 1036 | } |
d62a17ae | 1037 | |
1038 | if (t_ptr) { | |
1039 | *t_ptr = thread; | |
1040 | thread->ref = t_ptr; | |
1041 | } | |
1042 | } | |
1043 | ||
1044 | AWAKEN(m); | |
1045 | } | |
718e3744 | 1046 | } |
1047 | ||
ee1455dd IR |
1048 | static void _thread_add_timer_timeval(const struct xref_threadsched *xref, |
1049 | struct thread_master *m, | |
cc9f21da | 1050 | void (*func)(struct thread *), void *arg, |
ee1455dd IR |
1051 | struct timeval *time_relative, |
1052 | struct thread **t_ptr) | |
718e3744 | 1053 | { |
d62a17ae | 1054 | struct thread *thread; |
96fe578a | 1055 | struct timeval t; |
d62a17ae | 1056 | |
1057 | assert(m != NULL); | |
1058 | ||
d62a17ae | 1059 | assert(time_relative); |
1060 | ||
6c3aa850 DL |
1061 | frrtrace(9, frr_libfrr, schedule_timer, m, |
1062 | xref->funcname, xref->xref.file, xref->xref.line, | |
c7bb4f00 | 1063 | t_ptr, 0, 0, arg, (long)time_relative->tv_sec); |
abf96a87 | 1064 | |
96fe578a MS |
1065 | /* Compute expiration/deadline time. */ |
1066 | monotime(&t); | |
1067 | timeradd(&t, time_relative, &t); | |
1068 | ||
00dffa8c DL |
1069 | frr_with_mutex(&m->mtx) { |
1070 | if (t_ptr && *t_ptr) | |
d279ef57 | 1071 | /* thread is already scheduled; don't reschedule */ |
ee1455dd | 1072 | return; |
d62a17ae | 1073 | |
4322dea7 | 1074 | thread = thread_get(m, THREAD_TIMER, func, arg, xref); |
d62a17ae | 1075 | |
00dffa8c | 1076 | frr_with_mutex(&thread->mtx) { |
96fe578a | 1077 | thread->u.sands = t; |
27d29ced | 1078 | thread_timer_list_add(&m->timer, thread); |
d62a17ae | 1079 | if (t_ptr) { |
1080 | *t_ptr = thread; | |
1081 | thread->ref = t_ptr; | |
1082 | } | |
1083 | } | |
d62a17ae | 1084 | |
96fe578a MS |
1085 | /* The timer list is sorted - if this new timer |
1086 | * might change the time we'll wait for, give the pthread | |
1087 | * a chance to re-compute. | |
1088 | */ | |
1089 | if (thread_timer_list_first(&m->timer) == thread) | |
1090 | AWAKEN(m); | |
d62a17ae | 1091 | } |
e2eff5c3 DS |
1092 | #define ONEYEAR2SEC (60 * 60 * 24 * 365) |
1093 | if (time_relative->tv_sec > ONEYEAR2SEC) | |
1094 | flog_err( | |
1095 | EC_LIB_TIMER_TOO_LONG, | |
1096 | "Timer: %pTHD is created with an expiration that is greater than 1 year", | |
1097 | thread); | |
9e867fe6 | 1098 | } |
1099 | ||
98c91ac6 | 1100 | |
1101 | /* Add timer event thread. */ | |
ee1455dd | 1102 | void _thread_add_timer(const struct xref_threadsched *xref, |
cc9f21da | 1103 | struct thread_master *m, void (*func)(struct thread *), |
ee1455dd | 1104 | void *arg, long timer, struct thread **t_ptr) |
9e867fe6 | 1105 | { |
d62a17ae | 1106 | struct timeval trel; |
9e867fe6 | 1107 | |
d62a17ae | 1108 | assert(m != NULL); |
9e867fe6 | 1109 | |
d62a17ae | 1110 | trel.tv_sec = timer; |
1111 | trel.tv_usec = 0; | |
9e867fe6 | 1112 | |
ee1455dd | 1113 | _thread_add_timer_timeval(xref, m, func, arg, &trel, t_ptr); |
98c91ac6 | 1114 | } |
9e867fe6 | 1115 | |
98c91ac6 | 1116 | /* Add timer event thread with "millisecond" resolution */ |
ee1455dd IR |
1117 | void _thread_add_timer_msec(const struct xref_threadsched *xref, |
1118 | struct thread_master *m, | |
cc9f21da DS |
1119 | void (*func)(struct thread *), void *arg, |
1120 | long timer, struct thread **t_ptr) | |
98c91ac6 | 1121 | { |
d62a17ae | 1122 | struct timeval trel; |
9e867fe6 | 1123 | |
d62a17ae | 1124 | assert(m != NULL); |
718e3744 | 1125 | |
d62a17ae | 1126 | trel.tv_sec = timer / 1000; |
1127 | trel.tv_usec = 1000 * (timer % 1000); | |
98c91ac6 | 1128 | |
ee1455dd | 1129 | _thread_add_timer_timeval(xref, m, func, arg, &trel, t_ptr); |
a48b4e6d | 1130 | } |
1131 | ||
4322dea7 | 1132 | /* Add timer event thread with "timeval" resolution */ |
ee1455dd | 1133 | void _thread_add_timer_tv(const struct xref_threadsched *xref, |
cc9f21da DS |
1134 | struct thread_master *m, |
1135 | void (*func)(struct thread *), void *arg, | |
1136 | struct timeval *tv, struct thread **t_ptr) | |
d03c4cbd | 1137 | { |
ee1455dd | 1138 | _thread_add_timer_timeval(xref, m, func, arg, tv, t_ptr); |
d03c4cbd DL |
1139 | } |
1140 | ||
718e3744 | 1141 | /* Add simple event thread. */ |
ee1455dd | 1142 | void _thread_add_event(const struct xref_threadsched *xref, |
cc9f21da | 1143 | struct thread_master *m, void (*func)(struct thread *), |
ee1455dd | 1144 | void *arg, int val, struct thread **t_ptr) |
718e3744 | 1145 | { |
00dffa8c | 1146 | struct thread *thread = NULL; |
d62a17ae | 1147 | |
6c3aa850 DL |
1148 | frrtrace(9, frr_libfrr, schedule_event, m, |
1149 | xref->funcname, xref->xref.file, xref->xref.line, | |
c7bb4f00 | 1150 | t_ptr, 0, val, arg, 0); |
abf96a87 | 1151 | |
d62a17ae | 1152 | assert(m != NULL); |
1153 | ||
00dffa8c DL |
1154 | frr_with_mutex(&m->mtx) { |
1155 | if (t_ptr && *t_ptr) | |
d279ef57 | 1156 | /* thread is already scheduled; don't reschedule */ |
00dffa8c | 1157 | break; |
d62a17ae | 1158 | |
60a3efec | 1159 | thread = thread_get(m, THREAD_EVENT, func, arg, xref); |
00dffa8c | 1160 | frr_with_mutex(&thread->mtx) { |
d62a17ae | 1161 | thread->u.val = val; |
c284542b | 1162 | thread_list_add_tail(&m->event, thread); |
d62a17ae | 1163 | } |
d62a17ae | 1164 | |
1165 | if (t_ptr) { | |
1166 | *t_ptr = thread; | |
1167 | thread->ref = t_ptr; | |
1168 | } | |
1169 | ||
1170 | AWAKEN(m); | |
1171 | } | |
718e3744 | 1172 | } |
1173 | ||
63ccb9cb QY |
1174 | /* Thread cancellation ------------------------------------------------------ */ |
1175 | ||
8797240e QY |
1176 | /** |
1177 | * NOT's out the .events field of pollfd corresponding to the given file | |
1178 | * descriptor. The event to be NOT'd is passed in the 'state' parameter. | |
1179 | * | |
1180 | * This needs to happen for both copies of pollfd's. See 'thread_fetch' | |
1181 | * implementation for details. | |
1182 | * | |
1183 | * @param master | |
1184 | * @param fd | |
1185 | * @param state the event to cancel. One or more (OR'd together) of the | |
1186 | * following: | |
1187 | * - POLLIN | |
1188 | * - POLLOUT | |
1189 | */ | |
a9318a32 MS |
1190 | static void thread_cancel_rw(struct thread_master *master, int fd, short state, |
1191 | int idx_hint) | |
0a95a0d0 | 1192 | { |
42d74538 QY |
1193 | bool found = false; |
1194 | ||
d62a17ae | 1195 | /* find the index of corresponding pollfd */ |
1196 | nfds_t i; | |
1197 | ||
a9318a32 MS |
1198 | /* Cancel POLLHUP too just in case some bozo set it */ |
1199 | state |= POLLHUP; | |
1200 | ||
1201 | /* Some callers know the index of the pfd already */ | |
1202 | if (idx_hint >= 0) { | |
1203 | i = idx_hint; | |
1204 | found = true; | |
1205 | } else { | |
1206 | /* Have to look for the fd in the pfd array */ | |
1207 | for (i = 0; i < master->handler.pfdcount; i++) | |
1208 | if (master->handler.pfds[i].fd == fd) { | |
1209 | found = true; | |
1210 | break; | |
1211 | } | |
1212 | } | |
42d74538 QY |
1213 | |
1214 | if (!found) { | |
1215 | zlog_debug( | |
1216 | "[!] Received cancellation request for nonexistent rw job"); | |
1217 | zlog_debug("[!] threadmaster: %s | fd: %d", | |
996c9314 | 1218 | master->name ? master->name : "", fd); |
42d74538 QY |
1219 | return; |
1220 | } | |
d62a17ae | 1221 | |
1222 | /* NOT out event. */ | |
1223 | master->handler.pfds[i].events &= ~(state); | |
1224 | ||
1225 | /* If all events are canceled, delete / resize the pollfd array. */ | |
1226 | if (master->handler.pfds[i].events == 0) { | |
1227 | memmove(master->handler.pfds + i, master->handler.pfds + i + 1, | |
1228 | (master->handler.pfdcount - i - 1) | |
1229 | * sizeof(struct pollfd)); | |
1230 | master->handler.pfdcount--; | |
e985cda0 S |
1231 | master->handler.pfds[master->handler.pfdcount].fd = 0; |
1232 | master->handler.pfds[master->handler.pfdcount].events = 0; | |
d62a17ae | 1233 | } |
1234 | ||
1235 | /* If we have the same pollfd in the copy, perform the same operations, | |
1236 | * otherwise return. */ | |
1237 | if (i >= master->handler.copycount) | |
1238 | return; | |
1239 | ||
1240 | master->handler.copy[i].events &= ~(state); | |
1241 | ||
1242 | if (master->handler.copy[i].events == 0) { | |
1243 | memmove(master->handler.copy + i, master->handler.copy + i + 1, | |
1244 | (master->handler.copycount - i - 1) | |
1245 | * sizeof(struct pollfd)); | |
1246 | master->handler.copycount--; | |
e985cda0 S |
1247 | master->handler.copy[master->handler.copycount].fd = 0; |
1248 | master->handler.copy[master->handler.copycount].events = 0; | |
d62a17ae | 1249 | } |
0a95a0d0 DS |
1250 | } |
1251 | ||
a9318a32 MS |
1252 | /* |
1253 | * Process task cancellation given a task argument: iterate through the | |
1254 | * various lists of tasks, looking for any that match the argument. | |
1255 | */ | |
1256 | static void cancel_arg_helper(struct thread_master *master, | |
1257 | const struct cancel_req *cr) | |
1258 | { | |
1259 | struct thread *t; | |
1260 | nfds_t i; | |
1261 | int fd; | |
1262 | struct pollfd *pfd; | |
1263 | ||
1264 | /* We're only processing arg-based cancellations here. */ | |
1265 | if (cr->eventobj == NULL) | |
1266 | return; | |
1267 | ||
1268 | /* First process the ready lists. */ | |
1269 | frr_each_safe(thread_list, &master->event, t) { | |
1270 | if (t->arg != cr->eventobj) | |
1271 | continue; | |
1272 | thread_list_del(&master->event, t); | |
1273 | if (t->ref) | |
1274 | *t->ref = NULL; | |
1275 | thread_add_unuse(master, t); | |
1276 | } | |
1277 | ||
1278 | frr_each_safe(thread_list, &master->ready, t) { | |
1279 | if (t->arg != cr->eventobj) | |
1280 | continue; | |
1281 | thread_list_del(&master->ready, t); | |
1282 | if (t->ref) | |
1283 | *t->ref = NULL; | |
1284 | thread_add_unuse(master, t); | |
1285 | } | |
1286 | ||
1287 | /* If requested, stop here and ignore io and timers */ | |
1288 | if (CHECK_FLAG(cr->flags, THREAD_CANCEL_FLAG_READY)) | |
1289 | return; | |
1290 | ||
1291 | /* Check the io tasks */ | |
1292 | for (i = 0; i < master->handler.pfdcount;) { | |
1293 | pfd = master->handler.pfds + i; | |
1294 | ||
1295 | if (pfd->events & POLLIN) | |
1296 | t = master->read[pfd->fd]; | |
1297 | else | |
1298 | t = master->write[pfd->fd]; | |
1299 | ||
1300 | if (t && t->arg == cr->eventobj) { | |
1301 | fd = pfd->fd; | |
1302 | ||
1303 | /* Found a match to cancel: clean up fd arrays */ | |
1304 | thread_cancel_rw(master, pfd->fd, pfd->events, i); | |
1305 | ||
1306 | /* Clean up thread arrays */ | |
1307 | master->read[fd] = NULL; | |
1308 | master->write[fd] = NULL; | |
1309 | ||
1310 | /* Clear caller's ref */ | |
1311 | if (t->ref) | |
1312 | *t->ref = NULL; | |
1313 | ||
1314 | thread_add_unuse(master, t); | |
1315 | ||
1316 | /* Don't increment 'i' since the cancellation will have | |
1317 | * removed the entry from the pfd array | |
1318 | */ | |
1319 | } else | |
1320 | i++; | |
1321 | } | |
1322 | ||
1323 | /* Check the timer tasks */ | |
1324 | t = thread_timer_list_first(&master->timer); | |
1325 | while (t) { | |
1326 | struct thread *t_next; | |
1327 | ||
1328 | t_next = thread_timer_list_next(&master->timer, t); | |
1329 | ||
1330 | if (t->arg == cr->eventobj) { | |
1331 | thread_timer_list_del(&master->timer, t); | |
1332 | if (t->ref) | |
1333 | *t->ref = NULL; | |
1334 | thread_add_unuse(master, t); | |
1335 | } | |
1336 | ||
1337 | t = t_next; | |
1338 | } | |
1339 | } | |
1340 | ||
1189d95f | 1341 | /** |
63ccb9cb | 1342 | * Process cancellation requests. |
1189d95f | 1343 | * |
63ccb9cb QY |
1344 | * This may only be run from the pthread which owns the thread_master. |
1345 | * | |
1346 | * @param master the thread master to process | |
1347 | * @REQUIRE master->mtx | |
1189d95f | 1348 | */ |
d62a17ae | 1349 | static void do_thread_cancel(struct thread_master *master) |
718e3744 | 1350 | { |
c284542b | 1351 | struct thread_list_head *list = NULL; |
d62a17ae | 1352 | struct thread **thread_array = NULL; |
1353 | struct thread *thread; | |
1354 | ||
1355 | struct cancel_req *cr; | |
1356 | struct listnode *ln; | |
1357 | for (ALL_LIST_ELEMENTS_RO(master->cancel_req, ln, cr)) { | |
d279ef57 | 1358 | /* |
a9318a32 MS |
1359 | * If this is an event object cancellation, search |
1360 | * through task lists deleting any tasks which have the | |
1361 | * specified argument - use this handy helper function. | |
d279ef57 | 1362 | */ |
d62a17ae | 1363 | if (cr->eventobj) { |
a9318a32 | 1364 | cancel_arg_helper(master, cr); |
d62a17ae | 1365 | continue; |
1366 | } | |
1367 | ||
d279ef57 DS |
1368 | /* |
1369 | * The pointer varies depending on whether the cancellation | |
1370 | * request was made asynchronously or not. If it was, we | |
1371 | * need to check whether the thread even exists anymore | |
1372 | * before cancelling it. | |
1373 | */ | |
d62a17ae | 1374 | thread = (cr->thread) ? cr->thread : *cr->threadref; |
1375 | ||
1376 | if (!thread) | |
1377 | continue; | |
1378 | ||
1379 | /* Determine the appropriate queue to cancel the thread from */ | |
1380 | switch (thread->type) { | |
1381 | case THREAD_READ: | |
a9318a32 | 1382 | thread_cancel_rw(master, thread->u.fd, POLLIN, -1); |
d62a17ae | 1383 | thread_array = master->read; |
1384 | break; | |
1385 | case THREAD_WRITE: | |
a9318a32 | 1386 | thread_cancel_rw(master, thread->u.fd, POLLOUT, -1); |
d62a17ae | 1387 | thread_array = master->write; |
1388 | break; | |
1389 | case THREAD_TIMER: | |
27d29ced | 1390 | thread_timer_list_del(&master->timer, thread); |
d62a17ae | 1391 | break; |
1392 | case THREAD_EVENT: | |
1393 | list = &master->event; | |
1394 | break; | |
1395 | case THREAD_READY: | |
1396 | list = &master->ready; | |
1397 | break; | |
1398 | default: | |
1399 | continue; | |
1400 | break; | |
1401 | } | |
1402 | ||
27d29ced | 1403 | if (list) { |
c284542b | 1404 | thread_list_del(list, thread); |
d62a17ae | 1405 | } else if (thread_array) { |
1406 | thread_array[thread->u.fd] = NULL; | |
d62a17ae | 1407 | } |
1408 | ||
1409 | if (thread->ref) | |
1410 | *thread->ref = NULL; | |
1411 | ||
1412 | thread_add_unuse(thread->master, thread); | |
1413 | } | |
1414 | ||
1415 | /* Delete and free all cancellation requests */ | |
41b21bfa MS |
1416 | if (master->cancel_req) |
1417 | list_delete_all_node(master->cancel_req); | |
d62a17ae | 1418 | |
1419 | /* Wake up any threads which may be blocked in thread_cancel_async() */ | |
1420 | master->canceled = true; | |
1421 | pthread_cond_broadcast(&master->cancel_cond); | |
718e3744 | 1422 | } |
1423 | ||
a9318a32 MS |
1424 | /* |
1425 | * Helper function used for multiple flavors of arg-based cancellation. | |
1426 | */ | |
1427 | static void cancel_event_helper(struct thread_master *m, void *arg, int flags) | |
1428 | { | |
1429 | struct cancel_req *cr; | |
1430 | ||
1431 | assert(m->owner == pthread_self()); | |
1432 | ||
1433 | /* Only worth anything if caller supplies an arg. */ | |
1434 | if (arg == NULL) | |
1435 | return; | |
1436 | ||
1437 | cr = XCALLOC(MTYPE_TMP, sizeof(struct cancel_req)); | |
1438 | ||
1439 | cr->flags = flags; | |
1440 | ||
1441 | frr_with_mutex(&m->mtx) { | |
1442 | cr->eventobj = arg; | |
1443 | listnode_add(m->cancel_req, cr); | |
1444 | do_thread_cancel(m); | |
1445 | } | |
1446 | } | |
1447 | ||
63ccb9cb QY |
1448 | /** |
1449 | * Cancel any events which have the specified argument. | |
1450 | * | |
1451 | * MT-Unsafe | |
1452 | * | |
1453 | * @param m the thread_master to cancel from | |
1454 | * @param arg the argument passed when creating the event | |
1455 | */ | |
d62a17ae | 1456 | void thread_cancel_event(struct thread_master *master, void *arg) |
718e3744 | 1457 | { |
a9318a32 MS |
1458 | cancel_event_helper(master, arg, 0); |
1459 | } | |
d62a17ae | 1460 | |
a9318a32 MS |
1461 | /* |
1462 | * Cancel ready tasks with an arg matching 'arg' | |
1463 | * | |
1464 | * MT-Unsafe | |
1465 | * | |
1466 | * @param m the thread_master to cancel from | |
1467 | * @param arg the argument passed when creating the event | |
1468 | */ | |
1469 | void thread_cancel_event_ready(struct thread_master *m, void *arg) | |
1470 | { | |
1471 | ||
1472 | /* Only cancel ready/event tasks */ | |
1473 | cancel_event_helper(m, arg, THREAD_CANCEL_FLAG_READY); | |
63ccb9cb | 1474 | } |
1189d95f | 1475 | |
63ccb9cb QY |
1476 | /** |
1477 | * Cancel a specific task. | |
1478 | * | |
1479 | * MT-Unsafe | |
1480 | * | |
1481 | * @param thread task to cancel | |
1482 | */ | |
b3d6bc6e | 1483 | void thread_cancel(struct thread **thread) |
63ccb9cb | 1484 | { |
b3d6bc6e MS |
1485 | struct thread_master *master; |
1486 | ||
1487 | if (thread == NULL || *thread == NULL) | |
1488 | return; | |
1489 | ||
1490 | master = (*thread)->master; | |
d62a17ae | 1491 | |
6c3aa850 DL |
1492 | frrtrace(9, frr_libfrr, thread_cancel, master, |
1493 | (*thread)->xref->funcname, (*thread)->xref->xref.file, | |
1494 | (*thread)->xref->xref.line, NULL, (*thread)->u.fd, | |
b4d6e855 | 1495 | (*thread)->u.val, (*thread)->arg, (*thread)->u.sands.tv_sec); |
abf96a87 | 1496 | |
6ed04aa2 DS |
1497 | assert(master->owner == pthread_self()); |
1498 | ||
00dffa8c | 1499 | frr_with_mutex(&master->mtx) { |
d62a17ae | 1500 | struct cancel_req *cr = |
1501 | XCALLOC(MTYPE_TMP, sizeof(struct cancel_req)); | |
b3d6bc6e | 1502 | cr->thread = *thread; |
6ed04aa2 DS |
1503 | listnode_add(master->cancel_req, cr); |
1504 | do_thread_cancel(master); | |
d62a17ae | 1505 | } |
b3d6bc6e MS |
1506 | |
1507 | *thread = NULL; | |
63ccb9cb | 1508 | } |
1189d95f | 1509 | |
63ccb9cb QY |
1510 | /** |
1511 | * Asynchronous cancellation. | |
1512 | * | |
8797240e QY |
1513 | * Called with either a struct thread ** or void * to an event argument, |
1514 | * this function posts the correct cancellation request and blocks until it is | |
1515 | * serviced. | |
63ccb9cb QY |
1516 | * |
1517 | * If the thread is currently running, execution blocks until it completes. | |
1518 | * | |
8797240e QY |
1519 | * The last two parameters are mutually exclusive, i.e. if you pass one the |
1520 | * other must be NULL. | |
1521 | * | |
1522 | * When the cancellation procedure executes on the target thread_master, the | |
1523 | * thread * provided is checked for nullity. If it is null, the thread is | |
1524 | * assumed to no longer exist and the cancellation request is a no-op. Thus | |
1525 | * users of this API must pass a back-reference when scheduling the original | |
1526 | * task. | |
1527 | * | |
63ccb9cb QY |
1528 | * MT-Safe |
1529 | * | |
8797240e QY |
1530 | * @param master the thread master with the relevant event / task |
1531 | * @param thread pointer to thread to cancel | |
1532 | * @param eventobj the event | |
63ccb9cb | 1533 | */ |
d62a17ae | 1534 | void thread_cancel_async(struct thread_master *master, struct thread **thread, |
1535 | void *eventobj) | |
63ccb9cb | 1536 | { |
d62a17ae | 1537 | assert(!(thread && eventobj) && (thread || eventobj)); |
abf96a87 QY |
1538 | |
1539 | if (thread && *thread) | |
c7bb4f00 | 1540 | frrtrace(9, frr_libfrr, thread_cancel_async, master, |
6c3aa850 DL |
1541 | (*thread)->xref->funcname, (*thread)->xref->xref.file, |
1542 | (*thread)->xref->xref.line, NULL, (*thread)->u.fd, | |
c7bb4f00 QY |
1543 | (*thread)->u.val, (*thread)->arg, |
1544 | (*thread)->u.sands.tv_sec); | |
abf96a87 | 1545 | else |
c7bb4f00 QY |
1546 | frrtrace(9, frr_libfrr, thread_cancel_async, master, NULL, NULL, |
1547 | 0, NULL, 0, 0, eventobj, 0); | |
abf96a87 | 1548 | |
d62a17ae | 1549 | assert(master->owner != pthread_self()); |
1550 | ||
00dffa8c | 1551 | frr_with_mutex(&master->mtx) { |
d62a17ae | 1552 | master->canceled = false; |
1553 | ||
1554 | if (thread) { | |
1555 | struct cancel_req *cr = | |
1556 | XCALLOC(MTYPE_TMP, sizeof(struct cancel_req)); | |
1557 | cr->threadref = thread; | |
1558 | listnode_add(master->cancel_req, cr); | |
1559 | } else if (eventobj) { | |
1560 | struct cancel_req *cr = | |
1561 | XCALLOC(MTYPE_TMP, sizeof(struct cancel_req)); | |
1562 | cr->eventobj = eventobj; | |
1563 | listnode_add(master->cancel_req, cr); | |
1564 | } | |
1565 | AWAKEN(master); | |
1566 | ||
1567 | while (!master->canceled) | |
1568 | pthread_cond_wait(&master->cancel_cond, &master->mtx); | |
1569 | } | |
50478845 MS |
1570 | |
1571 | if (thread) | |
1572 | *thread = NULL; | |
718e3744 | 1573 | } |
63ccb9cb | 1574 | /* ------------------------------------------------------------------------- */ |
718e3744 | 1575 | |
27d29ced | 1576 | static struct timeval *thread_timer_wait(struct thread_timer_list_head *timers, |
d62a17ae | 1577 | struct timeval *timer_val) |
718e3744 | 1578 | { |
27d29ced DL |
1579 | if (!thread_timer_list_count(timers)) |
1580 | return NULL; | |
1581 | ||
1582 | struct thread *next_timer = thread_timer_list_first(timers); | |
1583 | monotime_until(&next_timer->u.sands, timer_val); | |
1584 | return timer_val; | |
718e3744 | 1585 | } |
718e3744 | 1586 | |
d62a17ae | 1587 | static struct thread *thread_run(struct thread_master *m, struct thread *thread, |
1588 | struct thread *fetch) | |
718e3744 | 1589 | { |
d62a17ae | 1590 | *fetch = *thread; |
1591 | thread_add_unuse(m, thread); | |
1592 | return fetch; | |
718e3744 | 1593 | } |
1594 | ||
d62a17ae | 1595 | static int thread_process_io_helper(struct thread_master *m, |
45f3d590 DS |
1596 | struct thread *thread, short state, |
1597 | short actual_state, int pos) | |
5d4ccd4e | 1598 | { |
d62a17ae | 1599 | struct thread **thread_array; |
1600 | ||
45f3d590 DS |
1601 | /* |
1602 | * poll() clears the .events field, but the pollfd array we | |
1603 | * pass to poll() is a copy of the one used to schedule threads. | |
1604 | * We need to synchronize state between the two here by applying | |
1605 | * the same changes poll() made on the copy of the "real" pollfd | |
1606 | * array. | |
1607 | * | |
1608 | * This cleans up a possible infinite loop where we refuse | |
1609 | * to respond to a poll event but poll is insistent that | |
1610 | * we should. | |
1611 | */ | |
1612 | m->handler.pfds[pos].events &= ~(state); | |
1613 | ||
1614 | if (!thread) { | |
1615 | if ((actual_state & (POLLHUP|POLLIN)) != POLLHUP) | |
1616 | flog_err(EC_LIB_NO_THREAD, | |
1d5453d6 | 1617 | "Attempting to process an I/O event but for fd: %d(%d) no thread to handle this!", |
45f3d590 | 1618 | m->handler.pfds[pos].fd, actual_state); |
d62a17ae | 1619 | return 0; |
45f3d590 | 1620 | } |
d62a17ae | 1621 | |
1622 | if (thread->type == THREAD_READ) | |
1623 | thread_array = m->read; | |
1624 | else | |
1625 | thread_array = m->write; | |
1626 | ||
1627 | thread_array[thread->u.fd] = NULL; | |
c284542b | 1628 | thread_list_add_tail(&m->ready, thread); |
d62a17ae | 1629 | thread->type = THREAD_READY; |
45f3d590 | 1630 | |
d62a17ae | 1631 | return 1; |
5d4ccd4e DS |
1632 | } |
1633 | ||
8797240e QY |
1634 | /** |
1635 | * Process I/O events. | |
1636 | * | |
1637 | * Walks through file descriptor array looking for those pollfds whose .revents | |
1638 | * field has something interesting. Deletes any invalid file descriptors. | |
1639 | * | |
1640 | * @param m the thread master | |
1641 | * @param num the number of active file descriptors (return value of poll()) | |
1642 | */ | |
d62a17ae | 1643 | static void thread_process_io(struct thread_master *m, unsigned int num) |
0a95a0d0 | 1644 | { |
d62a17ae | 1645 | unsigned int ready = 0; |
1646 | struct pollfd *pfds = m->handler.copy; | |
1647 | ||
1648 | for (nfds_t i = 0; i < m->handler.copycount && ready < num; ++i) { | |
1649 | /* no event for current fd? immediately continue */ | |
1650 | if (pfds[i].revents == 0) | |
1651 | continue; | |
1652 | ||
1653 | ready++; | |
1654 | ||
d279ef57 DS |
1655 | /* |
1656 | * Unless someone has called thread_cancel from another | |
1657 | * pthread, the only thing that could have changed in | |
1658 | * m->handler.pfds while we were asleep is the .events | |
1659 | * field in a given pollfd. Barring thread_cancel() that | |
1660 | * value should be a superset of the values we have in our | |
1661 | * copy, so there's no need to update it. Similarily, | |
1662 | * barring deletion, the fd should still be a valid index | |
1663 | * into the master's pfds. | |
d142453d DS |
1664 | * |
1665 | * We are including POLLERR here to do a READ event | |
1666 | * this is because the read should fail and the | |
1667 | * read function should handle it appropriately | |
d279ef57 | 1668 | */ |
d142453d | 1669 | if (pfds[i].revents & (POLLIN | POLLHUP | POLLERR)) { |
d62a17ae | 1670 | thread_process_io_helper(m, m->read[pfds[i].fd], POLLIN, |
45f3d590 DS |
1671 | pfds[i].revents, i); |
1672 | } | |
d62a17ae | 1673 | if (pfds[i].revents & POLLOUT) |
1674 | thread_process_io_helper(m, m->write[pfds[i].fd], | |
45f3d590 | 1675 | POLLOUT, pfds[i].revents, i); |
d62a17ae | 1676 | |
1677 | /* if one of our file descriptors is garbage, remove the same | |
1678 | * from | |
1679 | * both pfds + update sizes and index */ | |
1680 | if (pfds[i].revents & POLLNVAL) { | |
1681 | memmove(m->handler.pfds + i, m->handler.pfds + i + 1, | |
1682 | (m->handler.pfdcount - i - 1) | |
1683 | * sizeof(struct pollfd)); | |
1684 | m->handler.pfdcount--; | |
e985cda0 S |
1685 | m->handler.pfds[m->handler.pfdcount].fd = 0; |
1686 | m->handler.pfds[m->handler.pfdcount].events = 0; | |
d62a17ae | 1687 | |
1688 | memmove(pfds + i, pfds + i + 1, | |
1689 | (m->handler.copycount - i - 1) | |
1690 | * sizeof(struct pollfd)); | |
1691 | m->handler.copycount--; | |
e985cda0 S |
1692 | m->handler.copy[m->handler.copycount].fd = 0; |
1693 | m->handler.copy[m->handler.copycount].events = 0; | |
d62a17ae | 1694 | |
1695 | i--; | |
1696 | } | |
1697 | } | |
718e3744 | 1698 | } |
1699 | ||
8b70d0b0 | 1700 | /* Add all timers that have popped to the ready list. */ |
e7d9e44b | 1701 | static unsigned int thread_process_timers(struct thread_master *m, |
d62a17ae | 1702 | struct timeval *timenow) |
a48b4e6d | 1703 | { |
ab01a001 DS |
1704 | struct timeval prev = *timenow; |
1705 | bool displayed = false; | |
d62a17ae | 1706 | struct thread *thread; |
1707 | unsigned int ready = 0; | |
1708 | ||
e7d9e44b | 1709 | while ((thread = thread_timer_list_first(&m->timer))) { |
d62a17ae | 1710 | if (timercmp(timenow, &thread->u.sands, <)) |
e7d9e44b | 1711 | break; |
ab01a001 DS |
1712 | prev = thread->u.sands; |
1713 | prev.tv_sec += 4; | |
1714 | /* | |
1715 | * If the timer would have popped 4 seconds in the | |
1716 | * past then we are in a situation where we are | |
1717 | * really getting behind on handling of events. | |
1718 | * Let's log it and do the right thing with it. | |
1719 | */ | |
1dd08c22 DS |
1720 | if (timercmp(timenow, &prev, >)) { |
1721 | atomic_fetch_add_explicit( | |
1722 | &thread->hist->total_starv_warn, 1, | |
1723 | memory_order_seq_cst); | |
1724 | if (!displayed && !thread->ignore_timer_late) { | |
1725 | flog_warn( | |
1726 | EC_LIB_STARVE_THREAD, | |
1727 | "Thread Starvation: %pTHD was scheduled to pop greater than 4s ago", | |
1728 | thread); | |
1729 | displayed = true; | |
1730 | } | |
ab01a001 DS |
1731 | } |
1732 | ||
e7d9e44b | 1733 | thread_timer_list_pop(&m->timer); |
d62a17ae | 1734 | thread->type = THREAD_READY; |
e7d9e44b | 1735 | thread_list_add_tail(&m->ready, thread); |
d62a17ae | 1736 | ready++; |
1737 | } | |
e7d9e44b | 1738 | |
d62a17ae | 1739 | return ready; |
a48b4e6d | 1740 | } |
1741 | ||
2613abe6 | 1742 | /* process a list en masse, e.g. for event thread lists */ |
c284542b | 1743 | static unsigned int thread_process(struct thread_list_head *list) |
2613abe6 | 1744 | { |
d62a17ae | 1745 | struct thread *thread; |
d62a17ae | 1746 | unsigned int ready = 0; |
1747 | ||
c284542b | 1748 | while ((thread = thread_list_pop(list))) { |
d62a17ae | 1749 | thread->type = THREAD_READY; |
c284542b | 1750 | thread_list_add_tail(&thread->master->ready, thread); |
d62a17ae | 1751 | ready++; |
1752 | } | |
1753 | return ready; | |
2613abe6 PJ |
1754 | } |
1755 | ||
1756 | ||
718e3744 | 1757 | /* Fetch next ready thread. */ |
d62a17ae | 1758 | struct thread *thread_fetch(struct thread_master *m, struct thread *fetch) |
718e3744 | 1759 | { |
d62a17ae | 1760 | struct thread *thread = NULL; |
1761 | struct timeval now; | |
1762 | struct timeval zerotime = {0, 0}; | |
1763 | struct timeval tv; | |
1764 | struct timeval *tw = NULL; | |
d81ca9a3 | 1765 | bool eintr_p = false; |
d62a17ae | 1766 | int num = 0; |
1767 | ||
1768 | do { | |
1769 | /* Handle signals if any */ | |
1770 | if (m->handle_signals) | |
7cc91e67 | 1771 | frr_sigevent_process(); |
d62a17ae | 1772 | |
1773 | pthread_mutex_lock(&m->mtx); | |
1774 | ||
1775 | /* Process any pending cancellation requests */ | |
1776 | do_thread_cancel(m); | |
1777 | ||
e3c9529e QY |
1778 | /* |
1779 | * Attempt to flush ready queue before going into poll(). | |
1780 | * This is performance-critical. Think twice before modifying. | |
1781 | */ | |
c284542b | 1782 | if ((thread = thread_list_pop(&m->ready))) { |
e3c9529e QY |
1783 | fetch = thread_run(m, thread, fetch); |
1784 | if (fetch->ref) | |
1785 | *fetch->ref = NULL; | |
1786 | pthread_mutex_unlock(&m->mtx); | |
5e822957 DS |
1787 | if (!m->ready_run_loop) |
1788 | GETRUSAGE(&m->last_getrusage); | |
1789 | m->ready_run_loop = true; | |
e3c9529e QY |
1790 | break; |
1791 | } | |
1792 | ||
5e822957 | 1793 | m->ready_run_loop = false; |
e3c9529e QY |
1794 | /* otherwise, tick through scheduling sequence */ |
1795 | ||
bca37d17 QY |
1796 | /* |
1797 | * Post events to ready queue. This must come before the | |
1798 | * following block since events should occur immediately | |
1799 | */ | |
d62a17ae | 1800 | thread_process(&m->event); |
1801 | ||
bca37d17 QY |
1802 | /* |
1803 | * If there are no tasks on the ready queue, we will poll() | |
1804 | * until a timer expires or we receive I/O, whichever comes | |
1805 | * first. The strategy for doing this is: | |
d62a17ae | 1806 | * |
1807 | * - If there are events pending, set the poll() timeout to zero | |
1808 | * - If there are no events pending, but there are timers | |
d279ef57 DS |
1809 | * pending, set the timeout to the smallest remaining time on |
1810 | * any timer. | |
d62a17ae | 1811 | * - If there are neither timers nor events pending, but there |
d279ef57 | 1812 | * are file descriptors pending, block indefinitely in poll() |
d62a17ae | 1813 | * - If nothing is pending, it's time for the application to die |
1814 | * | |
1815 | * In every case except the last, we need to hit poll() at least | |
bca37d17 QY |
1816 | * once per loop to avoid starvation by events |
1817 | */ | |
c284542b | 1818 | if (!thread_list_count(&m->ready)) |
27d29ced | 1819 | tw = thread_timer_wait(&m->timer, &tv); |
d62a17ae | 1820 | |
c284542b DL |
1821 | if (thread_list_count(&m->ready) || |
1822 | (tw && !timercmp(tw, &zerotime, >))) | |
d62a17ae | 1823 | tw = &zerotime; |
1824 | ||
1825 | if (!tw && m->handler.pfdcount == 0) { /* die */ | |
1826 | pthread_mutex_unlock(&m->mtx); | |
1827 | fetch = NULL; | |
1828 | break; | |
1829 | } | |
1830 | ||
bca37d17 QY |
1831 | /* |
1832 | * Copy pollfd array + # active pollfds in it. Not necessary to | |
1833 | * copy the array size as this is fixed. | |
1834 | */ | |
d62a17ae | 1835 | m->handler.copycount = m->handler.pfdcount; |
1836 | memcpy(m->handler.copy, m->handler.pfds, | |
1837 | m->handler.copycount * sizeof(struct pollfd)); | |
1838 | ||
e3c9529e QY |
1839 | pthread_mutex_unlock(&m->mtx); |
1840 | { | |
d81ca9a3 MS |
1841 | eintr_p = false; |
1842 | num = fd_poll(m, tw, &eintr_p); | |
e3c9529e QY |
1843 | } |
1844 | pthread_mutex_lock(&m->mtx); | |
d764d2cc | 1845 | |
e3c9529e QY |
1846 | /* Handle any errors received in poll() */ |
1847 | if (num < 0) { | |
d81ca9a3 | 1848 | if (eintr_p) { |
d62a17ae | 1849 | pthread_mutex_unlock(&m->mtx); |
e3c9529e QY |
1850 | /* loop around to signal handler */ |
1851 | continue; | |
d62a17ae | 1852 | } |
1853 | ||
e3c9529e | 1854 | /* else die */ |
450971aa | 1855 | flog_err(EC_LIB_SYSTEM_CALL, "poll() error: %s", |
9ef9495e | 1856 | safe_strerror(errno)); |
e3c9529e QY |
1857 | pthread_mutex_unlock(&m->mtx); |
1858 | fetch = NULL; | |
1859 | break; | |
bca37d17 | 1860 | } |
d62a17ae | 1861 | |
1862 | /* Post timers to ready queue. */ | |
1863 | monotime(&now); | |
e7d9e44b | 1864 | thread_process_timers(m, &now); |
d62a17ae | 1865 | |
1866 | /* Post I/O to ready queue. */ | |
1867 | if (num > 0) | |
1868 | thread_process_io(m, num); | |
1869 | ||
d62a17ae | 1870 | pthread_mutex_unlock(&m->mtx); |
1871 | ||
1872 | } while (!thread && m->spin); | |
1873 | ||
1874 | return fetch; | |
718e3744 | 1875 | } |
1876 | ||
d62a17ae | 1877 | static unsigned long timeval_elapsed(struct timeval a, struct timeval b) |
62f44022 | 1878 | { |
d62a17ae | 1879 | return (((a.tv_sec - b.tv_sec) * TIMER_SECOND_MICRO) |
1880 | + (a.tv_usec - b.tv_usec)); | |
62f44022 QY |
1881 | } |
1882 | ||
d62a17ae | 1883 | unsigned long thread_consumed_time(RUSAGE_T *now, RUSAGE_T *start, |
1884 | unsigned long *cputime) | |
718e3744 | 1885 | { |
6418e2d3 | 1886 | #ifdef HAVE_CLOCK_THREAD_CPUTIME_ID |
4e2839de DS |
1887 | |
1888 | #ifdef __FreeBSD__ | |
1889 | /* | |
1890 | * FreeBSD appears to have an issue when calling clock_gettime | |
1891 | * with CLOCK_THREAD_CPUTIME_ID really close to each other | |
1892 | * occassionally the now time will be before the start time. | |
1893 | * This is not good and FRR is ending up with CPU HOG's | |
1894 | * when the subtraction wraps to very large numbers | |
1895 | * | |
1896 | * What we are going to do here is cheat a little bit | |
1897 | * and notice that this is a problem and just correct | |
1898 | * it so that it is impossible to happen | |
1899 | */ | |
1900 | if (start->cpu.tv_sec == now->cpu.tv_sec && | |
1901 | start->cpu.tv_nsec > now->cpu.tv_nsec) | |
1902 | now->cpu.tv_nsec = start->cpu.tv_nsec + 1; | |
1903 | else if (start->cpu.tv_sec > now->cpu.tv_sec) { | |
1904 | now->cpu.tv_sec = start->cpu.tv_sec; | |
1905 | now->cpu.tv_nsec = start->cpu.tv_nsec + 1; | |
1906 | } | |
1907 | #endif | |
6418e2d3 DL |
1908 | *cputime = (now->cpu.tv_sec - start->cpu.tv_sec) * TIMER_SECOND_MICRO |
1909 | + (now->cpu.tv_nsec - start->cpu.tv_nsec) / 1000; | |
1910 | #else | |
d62a17ae | 1911 | /* This is 'user + sys' time. */ |
1912 | *cputime = timeval_elapsed(now->cpu.ru_utime, start->cpu.ru_utime) | |
1913 | + timeval_elapsed(now->cpu.ru_stime, start->cpu.ru_stime); | |
6418e2d3 | 1914 | #endif |
d62a17ae | 1915 | return timeval_elapsed(now->real, start->real); |
8b70d0b0 | 1916 | } |
1917 | ||
50596be0 DS |
1918 | /* We should aim to yield after yield milliseconds, which defaults |
1919 | to THREAD_YIELD_TIME_SLOT . | |
8b70d0b0 | 1920 | Note: we are using real (wall clock) time for this calculation. |
1921 | It could be argued that CPU time may make more sense in certain | |
1922 | contexts. The things to consider are whether the thread may have | |
1923 | blocked (in which case wall time increases, but CPU time does not), | |
1924 | or whether the system is heavily loaded with other processes competing | |
d62a17ae | 1925 | for CPU time. On balance, wall clock time seems to make sense. |
8b70d0b0 | 1926 | Plus it has the added benefit that gettimeofday should be faster |
1927 | than calling getrusage. */ | |
d62a17ae | 1928 | int thread_should_yield(struct thread *thread) |
718e3744 | 1929 | { |
d62a17ae | 1930 | int result; |
00dffa8c | 1931 | frr_with_mutex(&thread->mtx) { |
d62a17ae | 1932 | result = monotime_since(&thread->real, NULL) |
1933 | > (int64_t)thread->yield; | |
1934 | } | |
d62a17ae | 1935 | return result; |
50596be0 DS |
1936 | } |
1937 | ||
d62a17ae | 1938 | void thread_set_yield_time(struct thread *thread, unsigned long yield_time) |
50596be0 | 1939 | { |
00dffa8c | 1940 | frr_with_mutex(&thread->mtx) { |
d62a17ae | 1941 | thread->yield = yield_time; |
1942 | } | |
718e3744 | 1943 | } |
1944 | ||
d62a17ae | 1945 | void thread_getrusage(RUSAGE_T *r) |
db9c0df9 | 1946 | { |
6418e2d3 DL |
1947 | monotime(&r->real); |
1948 | if (!cputime_enabled) { | |
1949 | memset(&r->cpu, 0, sizeof(r->cpu)); | |
1950 | return; | |
1951 | } | |
1952 | ||
1953 | #ifdef HAVE_CLOCK_THREAD_CPUTIME_ID | |
1954 | /* not currently implemented in Linux's vDSO, but maybe at some point | |
1955 | * in the future? | |
1956 | */ | |
1957 | clock_gettime(CLOCK_THREAD_CPUTIME_ID, &r->cpu); | |
1958 | #else /* !HAVE_CLOCK_THREAD_CPUTIME_ID */ | |
231db9a6 DS |
1959 | #if defined RUSAGE_THREAD |
1960 | #define FRR_RUSAGE RUSAGE_THREAD | |
1961 | #else | |
1962 | #define FRR_RUSAGE RUSAGE_SELF | |
1963 | #endif | |
6418e2d3 DL |
1964 | getrusage(FRR_RUSAGE, &(r->cpu)); |
1965 | #endif | |
db9c0df9 PJ |
1966 | } |
1967 | ||
fbcac826 QY |
1968 | /* |
1969 | * Call a thread. | |
1970 | * | |
1971 | * This function will atomically update the thread's usage history. At present | |
1972 | * this is the only spot where usage history is written. Nevertheless the code | |
1973 | * has been written such that the introduction of writers in the future should | |
1974 | * not need to update it provided the writers atomically perform only the | |
1975 | * operations done here, i.e. updating the total and maximum times. In | |
1976 | * particular, the maximum real and cpu times must be monotonically increasing | |
1977 | * or this code is not correct. | |
1978 | */ | |
d62a17ae | 1979 | void thread_call(struct thread *thread) |
718e3744 | 1980 | { |
d62a17ae | 1981 | RUSAGE_T before, after; |
cc8b13a0 | 1982 | |
45f01188 DL |
1983 | /* if the thread being called is the CLI, it may change cputime_enabled |
1984 | * ("service cputime-stats" command), which can result in nonsensical | |
1985 | * and very confusing warnings | |
1986 | */ | |
1987 | bool cputime_enabled_here = cputime_enabled; | |
1988 | ||
5e822957 DS |
1989 | if (thread->master->ready_run_loop) |
1990 | before = thread->master->last_getrusage; | |
1991 | else | |
1992 | GETRUSAGE(&before); | |
1993 | ||
d62a17ae | 1994 | thread->real = before.real; |
718e3744 | 1995 | |
6c3aa850 DL |
1996 | frrtrace(9, frr_libfrr, thread_call, thread->master, |
1997 | thread->xref->funcname, thread->xref->xref.file, | |
1998 | thread->xref->xref.line, NULL, thread->u.fd, | |
c7bb4f00 | 1999 | thread->u.val, thread->arg, thread->u.sands.tv_sec); |
abf96a87 | 2000 | |
d62a17ae | 2001 | pthread_setspecific(thread_current, thread); |
2002 | (*thread->func)(thread); | |
2003 | pthread_setspecific(thread_current, NULL); | |
718e3744 | 2004 | |
d62a17ae | 2005 | GETRUSAGE(&after); |
5e822957 | 2006 | thread->master->last_getrusage = after; |
718e3744 | 2007 | |
45f01188 DL |
2008 | unsigned long walltime, cputime; |
2009 | unsigned long exp; | |
fbcac826 | 2010 | |
45f01188 DL |
2011 | walltime = thread_consumed_time(&after, &before, &cputime); |
2012 | ||
2013 | /* update walltime */ | |
2014 | atomic_fetch_add_explicit(&thread->hist->real.total, walltime, | |
fbcac826 QY |
2015 | memory_order_seq_cst); |
2016 | exp = atomic_load_explicit(&thread->hist->real.max, | |
2017 | memory_order_seq_cst); | |
45f01188 | 2018 | while (exp < walltime |
fbcac826 | 2019 | && !atomic_compare_exchange_weak_explicit( |
45f01188 DL |
2020 | &thread->hist->real.max, &exp, walltime, |
2021 | memory_order_seq_cst, memory_order_seq_cst)) | |
fbcac826 QY |
2022 | ; |
2023 | ||
45f01188 DL |
2024 | if (cputime_enabled_here && cputime_enabled) { |
2025 | /* update cputime */ | |
2026 | atomic_fetch_add_explicit(&thread->hist->cpu.total, cputime, | |
2027 | memory_order_seq_cst); | |
2028 | exp = atomic_load_explicit(&thread->hist->cpu.max, | |
2029 | memory_order_seq_cst); | |
2030 | while (exp < cputime | |
2031 | && !atomic_compare_exchange_weak_explicit( | |
2032 | &thread->hist->cpu.max, &exp, cputime, | |
2033 | memory_order_seq_cst, memory_order_seq_cst)) | |
2034 | ; | |
2035 | } | |
fbcac826 QY |
2036 | |
2037 | atomic_fetch_add_explicit(&thread->hist->total_calls, 1, | |
2038 | memory_order_seq_cst); | |
2039 | atomic_fetch_or_explicit(&thread->hist->types, 1 << thread->add_type, | |
2040 | memory_order_seq_cst); | |
718e3744 | 2041 | |
45f01188 DL |
2042 | if (cputime_enabled_here && cputime_enabled && cputime_threshold |
2043 | && cputime > cputime_threshold) { | |
d62a17ae | 2044 | /* |
45f01188 DL |
2045 | * We have a CPU Hog on our hands. The time FRR has spent |
2046 | * doing actual work (not sleeping) is greater than 5 seconds. | |
d62a17ae | 2047 | * Whinge about it now, so we're aware this is yet another task |
2048 | * to fix. | |
2049 | */ | |
9b8e01ca DS |
2050 | atomic_fetch_add_explicit(&thread->hist->total_cpu_warn, |
2051 | 1, memory_order_seq_cst); | |
9ef9495e | 2052 | flog_warn( |
039d547f DS |
2053 | EC_LIB_SLOW_THREAD_CPU, |
2054 | "CPU HOG: task %s (%lx) ran for %lums (cpu time %lums)", | |
2055 | thread->xref->funcname, (unsigned long)thread->func, | |
45f01188 DL |
2056 | walltime / 1000, cputime / 1000); |
2057 | ||
2058 | } else if (walltime_threshold && walltime > walltime_threshold) { | |
039d547f | 2059 | /* |
45f01188 DL |
2060 | * The runtime for a task is greater than 5 seconds, but the |
2061 | * cpu time is under 5 seconds. Let's whine about this because | |
2062 | * this could imply some sort of scheduling issue. | |
039d547f | 2063 | */ |
9b8e01ca DS |
2064 | atomic_fetch_add_explicit(&thread->hist->total_wall_warn, |
2065 | 1, memory_order_seq_cst); | |
039d547f DS |
2066 | flog_warn( |
2067 | EC_LIB_SLOW_THREAD_WALL, | |
2068 | "STARVATION: task %s (%lx) ran for %lums (cpu time %lums)", | |
60a3efec | 2069 | thread->xref->funcname, (unsigned long)thread->func, |
45f01188 | 2070 | walltime / 1000, cputime / 1000); |
d62a17ae | 2071 | } |
718e3744 | 2072 | } |
2073 | ||
2074 | /* Execute thread */ | |
60a3efec | 2075 | void _thread_execute(const struct xref_threadsched *xref, |
cc9f21da | 2076 | struct thread_master *m, void (*func)(struct thread *), |
60a3efec | 2077 | void *arg, int val) |
718e3744 | 2078 | { |
c4345fbf | 2079 | struct thread *thread; |
718e3744 | 2080 | |
c4345fbf | 2081 | /* Get or allocate new thread to execute. */ |
00dffa8c | 2082 | frr_with_mutex(&m->mtx) { |
60a3efec | 2083 | thread = thread_get(m, THREAD_EVENT, func, arg, xref); |
9c7753e4 | 2084 | |
c4345fbf | 2085 | /* Set its event value. */ |
00dffa8c | 2086 | frr_with_mutex(&thread->mtx) { |
c4345fbf RZ |
2087 | thread->add_type = THREAD_EXECUTE; |
2088 | thread->u.val = val; | |
2089 | thread->ref = &thread; | |
2090 | } | |
c4345fbf | 2091 | } |
f7c62e11 | 2092 | |
c4345fbf RZ |
2093 | /* Execute thread doing all accounting. */ |
2094 | thread_call(thread); | |
9c7753e4 | 2095 | |
c4345fbf RZ |
2096 | /* Give back or free thread. */ |
2097 | thread_add_unuse(m, thread); | |
718e3744 | 2098 | } |
1543c387 MS |
2099 | |
2100 | /* Debug signal mask - if 'sigs' is NULL, use current effective mask. */ | |
2101 | void debug_signals(const sigset_t *sigs) | |
2102 | { | |
2103 | int i, found; | |
2104 | sigset_t tmpsigs; | |
2105 | char buf[300]; | |
2106 | ||
2107 | /* | |
2108 | * We're only looking at the non-realtime signals here, so we need | |
2109 | * some limit value. Platform differences mean at some point we just | |
2110 | * need to pick a reasonable value. | |
2111 | */ | |
2112 | #if defined SIGRTMIN | |
2113 | # define LAST_SIGNAL SIGRTMIN | |
2114 | #else | |
2115 | # define LAST_SIGNAL 32 | |
2116 | #endif | |
2117 | ||
2118 | ||
2119 | if (sigs == NULL) { | |
2120 | sigemptyset(&tmpsigs); | |
2121 | pthread_sigmask(SIG_BLOCK, NULL, &tmpsigs); | |
2122 | sigs = &tmpsigs; | |
2123 | } | |
2124 | ||
2125 | found = 0; | |
2126 | buf[0] = '\0'; | |
2127 | ||
2128 | for (i = 0; i < LAST_SIGNAL; i++) { | |
2129 | char tmp[20]; | |
2130 | ||
2131 | if (sigismember(sigs, i) > 0) { | |
2132 | if (found > 0) | |
2133 | strlcat(buf, ",", sizeof(buf)); | |
2134 | snprintf(tmp, sizeof(tmp), "%d", i); | |
2135 | strlcat(buf, tmp, sizeof(buf)); | |
2136 | found++; | |
2137 | } | |
2138 | } | |
2139 | ||
2140 | if (found == 0) | |
2141 | snprintf(buf, sizeof(buf), "<none>"); | |
2142 | ||
2143 | zlog_debug("%s: %s", __func__, buf); | |
2144 | } | |
a505383d DS |
2145 | |
2146 | bool thread_is_scheduled(struct thread *thread) | |
2147 | { | |
2148 | if (thread == NULL) | |
2149 | return false; | |
2150 | ||
2151 | return true; | |
2152 | } | |
f59e6882 DL |
2153 | |
2154 | static ssize_t printfrr_thread_dbg(struct fbuf *buf, struct printfrr_eargs *ea, | |
2155 | const struct thread *thread) | |
2156 | { | |
2157 | static const char * const types[] = { | |
2158 | [THREAD_READ] = "read", | |
2159 | [THREAD_WRITE] = "write", | |
2160 | [THREAD_TIMER] = "timer", | |
2161 | [THREAD_EVENT] = "event", | |
2162 | [THREAD_READY] = "ready", | |
2163 | [THREAD_UNUSED] = "unused", | |
2164 | [THREAD_EXECUTE] = "exec", | |
2165 | }; | |
2166 | ssize_t rv = 0; | |
2167 | char info[16] = ""; | |
2168 | ||
2169 | if (!thread) | |
2170 | return bputs(buf, "{(thread *)NULL}"); | |
2171 | ||
2172 | rv += bprintfrr(buf, "{(thread *)%p arg=%p", thread, thread->arg); | |
2173 | ||
2174 | if (thread->type < array_size(types) && types[thread->type]) | |
2175 | rv += bprintfrr(buf, " %-6s", types[thread->type]); | |
2176 | else | |
2177 | rv += bprintfrr(buf, " INVALID(%u)", thread->type); | |
2178 | ||
2179 | switch (thread->type) { | |
2180 | case THREAD_READ: | |
2181 | case THREAD_WRITE: | |
2182 | snprintfrr(info, sizeof(info), "fd=%d", thread->u.fd); | |
2183 | break; | |
2184 | ||
2185 | case THREAD_TIMER: | |
2186 | snprintfrr(info, sizeof(info), "r=%pTVMud", &thread->u.sands); | |
2187 | break; | |
2188 | } | |
2189 | ||
2190 | rv += bprintfrr(buf, " %-12s %s() %s from %s:%d}", info, | |
2191 | thread->xref->funcname, thread->xref->dest, | |
2192 | thread->xref->xref.file, thread->xref->xref.line); | |
2193 | return rv; | |
2194 | } | |
2195 | ||
54929fd3 | 2196 | printfrr_ext_autoreg_p("TH", printfrr_thread); |
f59e6882 DL |
2197 | static ssize_t printfrr_thread(struct fbuf *buf, struct printfrr_eargs *ea, |
2198 | const void *ptr) | |
2199 | { | |
2200 | const struct thread *thread = ptr; | |
2201 | struct timespec remain = {}; | |
2202 | ||
2203 | if (ea->fmt[0] == 'D') { | |
2204 | ea->fmt++; | |
2205 | return printfrr_thread_dbg(buf, ea, thread); | |
2206 | } | |
2207 | ||
2208 | if (!thread) { | |
2209 | /* need to jump over time formatting flag characters in the | |
2210 | * input format string, i.e. adjust ea->fmt! | |
2211 | */ | |
2212 | printfrr_time(buf, ea, &remain, | |
2213 | TIMEFMT_TIMER_DEADLINE | TIMEFMT_SKIP); | |
2214 | return bputch(buf, '-'); | |
2215 | } | |
2216 | ||
2217 | TIMEVAL_TO_TIMESPEC(&thread->u.sands, &remain); | |
2218 | return printfrr_time(buf, ea, &remain, TIMEFMT_TIMER_DEADLINE); | |
2219 | } |