xref: /titanic_50/usr/src/cmd/stat/common/acquire.c (revision 1aae5cd0fbf4c2622e76b8be40ca23c7c1f78e2b)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 #include "statcommon.h"
26 #include "dsr.h"
27 
28 #include <stdlib.h>
29 #include <unistd.h>
30 #include <strings.h>
31 #include <errno.h>
32 #include <limits.h>
33 #include <poll.h>
34 
35 #define	ARRAY_SIZE(a)	(sizeof (a) / sizeof (*a))
36 
37 /*
38  * The time we delay before retrying after an allocation
39  * failure, in milliseconds
40  */
41 #define	RETRY_DELAY 200
42 
43 static char *cpu_states[] = {
44 	"cpu_ticks_idle",
45 	"cpu_ticks_user",
46 	"cpu_ticks_kernel",
47 	"cpu_ticks_wait"
48 };
49 
50 static kstat_t *
51 kstat_lookup_read(kstat_ctl_t *kc, char *module,
52 		int instance, char *name)
53 {
54 	kstat_t *ksp = kstat_lookup(kc, module, instance, name);
55 	if (ksp == NULL)
56 		return (NULL);
57 	if (kstat_read(kc, ksp, NULL) == -1)
58 		return (NULL);
59 	return (ksp);
60 }
61 
62 /*
63  * Note: the following helpers do not clean up on the failure case,
64  * because it is left to the free_snapshot() in the acquire_snapshot()
65  * failure path.
66  */
67 
68 static int
69 acquire_cpus(struct snapshot *ss, kstat_ctl_t *kc)
70 {
71 	size_t i;
72 
73 	ss->s_nr_cpus = sysconf(_SC_CPUID_MAX) + 1;
74 	ss->s_cpus = calloc(ss->s_nr_cpus, sizeof (struct cpu_snapshot));
75 	if (ss->s_cpus == NULL)
76 		goto out;
77 
78 	for (i = 0; i < ss->s_nr_cpus; i++) {
79 		kstat_t *ksp;
80 
81 		ss->s_cpus[i].cs_id = ID_NO_CPU;
82 		ss->s_cpus[i].cs_state = p_online(i, P_STATUS);
83 		/* If no valid CPU is present, move on to the next one */
84 		if (ss->s_cpus[i].cs_state == -1)
85 			continue;
86 		ss->s_cpus[i].cs_id = i;
87 
88 		if ((ksp = kstat_lookup_read(kc, "cpu_info", i, NULL)) == NULL)
89 			goto out;
90 
91 		(void) pset_assign(PS_QUERY, i, &ss->s_cpus[i].cs_pset_id);
92 		if (ss->s_cpus[i].cs_pset_id == PS_NONE)
93 			ss->s_cpus[i].cs_pset_id = ID_NO_PSET;
94 
95 		if (!CPU_ACTIVE(&ss->s_cpus[i]))
96 			continue;
97 
98 		if ((ksp = kstat_lookup_read(kc, "cpu", i, "vm")) == NULL)
99 			goto out;
100 
101 		if (kstat_copy(ksp, &ss->s_cpus[i].cs_vm))
102 			goto out;
103 
104 		if ((ksp = kstat_lookup_read(kc, "cpu", i, "sys")) == NULL)
105 			goto out;
106 
107 		if (kstat_copy(ksp, &ss->s_cpus[i].cs_sys))
108 			goto out;
109 	}
110 
111 	errno = 0;
112 out:
113 	return (errno);
114 }
115 
116 static int
117 acquire_psets(struct snapshot *ss)
118 {
119 	psetid_t *pids = NULL;
120 	struct pset_snapshot *ps;
121 	size_t pids_nr;
122 	size_t i, j;
123 
124 	/*
125 	 * Careful in this code. We have to use pset_list
126 	 * twice, but inbetween pids_nr can change at will.
127 	 * We delay the setting of s_nr_psets until we have
128 	 * the "final" value of pids_nr.
129 	 */
130 
131 	if (pset_list(NULL, &pids_nr) < 0)
132 		return (errno);
133 
134 	if ((pids = calloc(pids_nr, sizeof (psetid_t))) == NULL)
135 		goto out;
136 
137 	if (pset_list(pids, &pids_nr) < 0)
138 		goto out;
139 
140 	ss->s_psets = calloc(pids_nr + 1, sizeof (struct pset_snapshot));
141 	if (ss->s_psets == NULL)
142 		goto out;
143 	ss->s_nr_psets = pids_nr + 1;
144 
145 	/* CPUs not in any actual pset */
146 	ps = &ss->s_psets[0];
147 	ps->ps_id = 0;
148 	ps->ps_cpus = calloc(ss->s_nr_cpus, sizeof (struct cpu_snapshot *));
149 	if (ps->ps_cpus == NULL)
150 		goto out;
151 
152 	/* CPUs in a a pset */
153 	for (i = 1; i < ss->s_nr_psets; i++) {
154 		ps = &ss->s_psets[i];
155 
156 		ps->ps_id = pids[i - 1];
157 		ps->ps_cpus =
158 		    calloc(ss->s_nr_cpus, sizeof (struct cpu_snapshot *));
159 		if (ps->ps_cpus == NULL)
160 			goto out;
161 	}
162 
163 	for (i = 0; i < ss->s_nr_psets; i++) {
164 		ps = &ss->s_psets[i];
165 
166 		for (j = 0; j < ss->s_nr_cpus; j++) {
167 			if (!CPU_ACTIVE(&ss->s_cpus[j]))
168 				continue;
169 			if (ss->s_cpus[j].cs_pset_id != ps->ps_id)
170 				continue;
171 
172 			ps->ps_cpus[ps->ps_nr_cpus++] = &ss->s_cpus[j];
173 		}
174 	}
175 
176 	errno = 0;
177 out:
178 	free(pids);
179 	return (errno);
180 }
181 
182 static int
183 acquire_intrs(struct snapshot *ss, kstat_ctl_t *kc)
184 {
185 	kstat_t *ksp;
186 	size_t i = 0;
187 	kstat_t *sys_misc;
188 	kstat_named_t *clock;
189 
190 	/* clock interrupt */
191 	ss->s_nr_intrs = 1;
192 
193 	for (ksp = kc->kc_chain; ksp; ksp = ksp->ks_next) {
194 		if (ksp->ks_type == KSTAT_TYPE_INTR)
195 			ss->s_nr_intrs++;
196 	}
197 
198 	ss->s_intrs = calloc(ss->s_nr_intrs, sizeof (struct intr_snapshot));
199 	if (ss->s_intrs == NULL)
200 		return (errno);
201 
202 	sys_misc = kstat_lookup_read(kc, "unix", 0, "system_misc");
203 	if (sys_misc == NULL)
204 		goto out;
205 
206 	clock = (kstat_named_t *)kstat_data_lookup(sys_misc, "clk_intr");
207 	if (clock == NULL)
208 		goto out;
209 
210 	(void) strlcpy(ss->s_intrs[0].is_name, "clock", KSTAT_STRLEN);
211 	ss->s_intrs[0].is_total = clock->value.ui32;
212 
213 	i = 1;
214 
215 	for (ksp = kc->kc_chain; ksp; ksp = ksp->ks_next) {
216 		kstat_intr_t *ki;
217 		int j;
218 
219 		if (ksp->ks_type != KSTAT_TYPE_INTR)
220 			continue;
221 		if (kstat_read(kc, ksp, NULL) == -1)
222 			goto out;
223 
224 		ki = KSTAT_INTR_PTR(ksp);
225 
226 		(void) strlcpy(ss->s_intrs[i].is_name, ksp->ks_name,
227 		    KSTAT_STRLEN);
228 		ss->s_intrs[i].is_total = 0;
229 
230 		for (j = 0; j < KSTAT_NUM_INTRS; j++)
231 			ss->s_intrs[i].is_total += ki->intrs[j];
232 
233 		i++;
234 	}
235 
236 	errno = 0;
237 out:
238 	return (errno);
239 }
240 
241 int
242 acquire_sys(struct snapshot *ss, kstat_ctl_t *kc)
243 {
244 	size_t i;
245 	kstat_named_t *knp;
246 	kstat_t *ksp;
247 
248 	if ((ksp = kstat_lookup(kc, "unix", 0, "sysinfo")) == NULL)
249 		return (errno);
250 
251 	if (kstat_read(kc, ksp, &ss->s_sys.ss_sysinfo) == -1)
252 		return (errno);
253 
254 	if ((ksp = kstat_lookup(kc, "unix", 0, "vminfo")) == NULL)
255 		return (errno);
256 
257 	if (kstat_read(kc, ksp, &ss->s_sys.ss_vminfo) == -1)
258 		return (errno);
259 
260 	if ((ksp = kstat_lookup(kc, "unix", 0, "dnlcstats")) == NULL)
261 		return (errno);
262 
263 	if (kstat_read(kc, ksp, &ss->s_sys.ss_nc) == -1)
264 		return (errno);
265 
266 	if ((ksp = kstat_lookup(kc, "unix", 0, "system_misc")) == NULL)
267 		return (errno);
268 
269 	if (kstat_read(kc, ksp, NULL) == -1)
270 		return (errno);
271 
272 	knp = (kstat_named_t *)kstat_data_lookup(ksp, "clk_intr");
273 	if (knp == NULL)
274 		return (errno);
275 
276 	ss->s_sys.ss_ticks = knp->value.l;
277 
278 	knp = (kstat_named_t *)kstat_data_lookup(ksp, "deficit");
279 	if (knp == NULL)
280 		return (errno);
281 
282 	ss->s_sys.ss_deficit = knp->value.l;
283 
284 	for (i = 0; i < ss->s_nr_cpus; i++) {
285 		if (!CPU_ACTIVE(&ss->s_cpus[i]))
286 			continue;
287 
288 		if (kstat_add(&ss->s_cpus[i].cs_sys, &ss->s_sys.ss_agg_sys))
289 			return (errno);
290 		if (kstat_add(&ss->s_cpus[i].cs_vm, &ss->s_sys.ss_agg_vm))
291 			return (errno);
292 	}
293 
294 	return (0);
295 }
296 
297 struct snapshot *
298 acquire_snapshot(kstat_ctl_t *kc, int types, struct iodev_filter *iodev_filter)
299 {
300 	struct snapshot *ss = NULL;
301 	int err;
302 
303 retry:
304 	err = 0;
305 	/* ensure any partial resources are freed on a retry */
306 	free_snapshot(ss);
307 
308 	ss = safe_alloc(sizeof (struct snapshot));
309 
310 	(void) memset(ss, 0, sizeof (struct snapshot));
311 
312 	ss->s_types = types;
313 
314 	/* wait for a possibly up-to-date chain */
315 	while (kstat_chain_update(kc) == -1) {
316 		if (errno == EAGAIN)
317 			(void) poll(NULL, 0, RETRY_DELAY);
318 		else
319 			fail(1, "kstat_chain_update failed");
320 	}
321 
322 	if (!err && (types & SNAP_INTERRUPTS))
323 		err = acquire_intrs(ss, kc);
324 
325 	if (!err && (types & (SNAP_CPUS | SNAP_SYSTEM | SNAP_PSETS)))
326 		err = acquire_cpus(ss, kc);
327 
328 	if (!err && (types & SNAP_PSETS))
329 		err = acquire_psets(ss);
330 
331 	if (!err && (types & (SNAP_IODEVS | SNAP_CONTROLLERS |
332 	    SNAP_IOPATHS_LI | SNAP_IOPATHS_LTI)))
333 		err = acquire_iodevs(ss, kc, iodev_filter);
334 
335 	if (!err && (types & SNAP_SYSTEM))
336 		err = acquire_sys(ss, kc);
337 
338 	switch (err) {
339 		case 0:
340 			break;
341 		case EAGAIN:
342 			(void) poll(NULL, 0, RETRY_DELAY);
343 		/* a kstat disappeared from under us */
344 		/*FALLTHRU*/
345 		case ENXIO:
346 		case ENOENT:
347 			goto retry;
348 		default:
349 			fail(1, "acquiring snapshot failed");
350 	}
351 
352 	return (ss);
353 }
354 
355 void
356 free_snapshot(struct snapshot *ss)
357 {
358 	size_t i;
359 
360 	if (ss == NULL)
361 		return;
362 
363 	while (ss->s_iodevs) {
364 		struct iodev_snapshot *tmp = ss->s_iodevs;
365 		ss->s_iodevs = ss->s_iodevs->is_next;
366 		free_iodev(tmp);
367 	}
368 
369 	if (ss->s_cpus) {
370 		for (i = 0; i < ss->s_nr_cpus; i++) {
371 			free(ss->s_cpus[i].cs_vm.ks_data);
372 			free(ss->s_cpus[i].cs_sys.ks_data);
373 		}
374 		free(ss->s_cpus);
375 	}
376 
377 	if (ss->s_psets) {
378 		for (i = 0; i < ss->s_nr_psets; i++)
379 			free(ss->s_psets[i].ps_cpus);
380 		free(ss->s_psets);
381 	}
382 
383 	free(ss->s_sys.ss_agg_sys.ks_data);
384 	free(ss->s_sys.ss_agg_vm.ks_data);
385 	free(ss);
386 }
387 
388 kstat_ctl_t *
389 open_kstat(void)
390 {
391 	kstat_ctl_t *kc;
392 
393 	while ((kc = kstat_open()) == NULL) {
394 		if (errno == EAGAIN)
395 			(void) poll(NULL, 0, RETRY_DELAY);
396 		else
397 			fail(1, "kstat_open failed");
398 	}
399 
400 	return (kc);
401 }
402 
403 void *
404 safe_alloc(size_t size)
405 {
406 	void *ptr;
407 
408 	while ((ptr = malloc(size)) == NULL) {
409 		if (errno == EAGAIN)
410 			(void) poll(NULL, 0, RETRY_DELAY);
411 		else
412 			fail(1, "malloc failed");
413 	}
414 	return (ptr);
415 }
416 
417 char *
418 safe_strdup(char *str)
419 {
420 	char *ret;
421 
422 	if (str == NULL)
423 		return (NULL);
424 
425 	while ((ret = strdup(str)) == NULL) {
426 		if (errno == EAGAIN)
427 			(void) poll(NULL, 0, RETRY_DELAY);
428 		else
429 			fail(1, "malloc failed");
430 	}
431 	return (ret);
432 }
433 
434 uint64_t
435 kstat_delta(kstat_t *old, kstat_t *new, char *name)
436 {
437 	kstat_named_t *knew = kstat_data_lookup(new, name);
438 	if (old && old->ks_data) {
439 		kstat_named_t *kold = kstat_data_lookup(old, name);
440 		return (knew->value.ui64 - kold->value.ui64);
441 	}
442 	return (knew->value.ui64);
443 }
444 
445 int
446 kstat_copy(const kstat_t *src, kstat_t *dst)
447 {
448 	*dst = *src;
449 
450 	if (src->ks_data != NULL) {
451 		if ((dst->ks_data = malloc(src->ks_data_size)) == NULL)
452 			return (-1);
453 		bcopy(src->ks_data, dst->ks_data, src->ks_data_size);
454 	} else {
455 		dst->ks_data = NULL;
456 		dst->ks_data_size = 0;
457 	}
458 	return (0);
459 }
460 
461 int
462 kstat_add(const kstat_t *src, kstat_t *dst)
463 {
464 	size_t i;
465 	kstat_named_t *from;
466 	kstat_named_t *to;
467 
468 	if (dst->ks_data == NULL)
469 		return (kstat_copy(src, dst));
470 
471 	from = src->ks_data;
472 	to = dst->ks_data;
473 
474 	for (i = 0; i < src->ks_ndata; i++) {
475 		/* "addition" makes little sense for strings */
476 		if (from->data_type != KSTAT_DATA_CHAR &&
477 		    from->data_type != KSTAT_DATA_STRING)
478 			(to)->value.ui64 += (from)->value.ui64;
479 		from++;
480 		to++;
481 	}
482 
483 	return (0);
484 }
485 
486 uint64_t
487 cpu_ticks_delta(kstat_t *old, kstat_t *new)
488 {
489 	uint64_t ticks = 0;
490 	size_t i;
491 	for (i = 0; i < ARRAY_SIZE(cpu_states); i++)
492 		ticks += kstat_delta(old, new, cpu_states[i]);
493 	return (ticks);
494 }
495 
496 int
497 nr_active_cpus(struct snapshot *ss)
498 {
499 	size_t i;
500 	int count = 0;
501 	for (i = 0; i < ss->s_nr_cpus; i++) {
502 		if (CPU_ACTIVE(&ss->s_cpus[i]))
503 			count++;
504 	}
505 
506 	return (count);
507 }
508 
509 /*
510  * Return the number of ticks delta between two hrtime_t
511  * values. Attempt to cater for various kinds of overflow
512  * in hrtime_t - no matter how improbable.
513  */
514 uint64_t
515 hrtime_delta(hrtime_t old, hrtime_t new)
516 {
517 	uint64_t del;
518 
519 	if ((new >= old) && (old >= 0L))
520 		return (new - old);
521 	else {
522 		/*
523 		 * We've overflowed the positive portion of an
524 		 * hrtime_t.
525 		 */
526 		if (new < 0L) {
527 			/*
528 			 * The new value is negative. Handle the
529 			 * case where the old value is positive or
530 			 * negative.
531 			 */
532 			uint64_t n1;
533 			uint64_t o1;
534 
535 			n1 = -new;
536 			if (old > 0L)
537 				return (n1 - old);
538 			else {
539 				o1 = -old;
540 				del = n1 - o1;
541 				return (del);
542 			}
543 		} else {
544 			/*
545 			 * Either we've just gone from being negative
546 			 * to positive *or* the last entry was positive
547 			 * and the new entry is also positive but *less*
548 			 * than the old entry. This implies we waited
549 			 * quite a few days on a very fast system between
550 			 * iostat displays.
551 			 */
552 			if (old < 0L) {
553 				uint64_t o2;
554 
555 				o2 = -old;
556 				del = UINT64_MAX - o2;
557 			} else {
558 				del = UINT64_MAX - old;
559 			}
560 			del += new;
561 			return (del);
562 		}
563 	}
564 }
565