xref: /titanic_41/usr/src/cmd/stat/common/acquire.c (revision b0fc0e77220f1fa4c933fd58a4e1dedcd650b0f1)
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 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include "statcommon.h"
29 #include "dsr.h"
30 
31 #include <stdlib.h>
32 #include <stdarg.h>
33 #include <unistd.h>
34 #include <strings.h>
35 #include <errno.h>
36 #include <limits.h>
37 #include <poll.h>
38 
39 #define	ARRAY_SIZE(a)	(sizeof (a) / sizeof (*a))
40 
41 /*
42  * The time we delay before retrying after an allocation
43  * failure, in milliseconds
44  */
45 #define	RETRY_DELAY 200
46 
47 static char *cpu_states[] = {
48 	"cpu_ticks_idle",
49 	"cpu_ticks_user",
50 	"cpu_ticks_kernel",
51 	"cpu_ticks_wait"
52 };
53 
54 extern char cmdname[];
55 
56 static kstat_t *
57 kstat_lookup_read(kstat_ctl_t *kc, char *module,
58 		int instance, char *name)
59 {
60 	kstat_t *ksp = kstat_lookup(kc, module, instance, name);
61 	if (ksp == NULL)
62 		return (NULL);
63 	if (kstat_read(kc, ksp, NULL) == -1)
64 		return (NULL);
65 	return (ksp);
66 }
67 
68 /*
69  * Note: the following helpers do not clean up on the failure case,
70  * because it is left to the free_snapshot() in the acquire_snapshot()
71  * failure path.
72  */
73 
74 static int
75 acquire_cpus(struct snapshot *ss, kstat_ctl_t *kc)
76 {
77 	size_t i;
78 
79 	ss->s_nr_cpus = sysconf(_SC_CPUID_MAX) + 1;
80 	ss->s_cpus = calloc(ss->s_nr_cpus, sizeof (struct cpu_snapshot));
81 	if (ss->s_cpus == NULL)
82 		goto out;
83 
84 	for (i = 0; i < ss->s_nr_cpus; i++) {
85 		kstat_t *ksp;
86 
87 		ss->s_cpus[i].cs_id = ID_NO_CPU;
88 		ss->s_cpus[i].cs_state = p_online(i, P_STATUS);
89 		/* If no valid CPU is present, move on to the next one */
90 		if (ss->s_cpus[i].cs_state == -1)
91 			continue;
92 		ss->s_cpus[i].cs_id = i;
93 
94 		if ((ksp = kstat_lookup_read(kc, "cpu_info", i, NULL)) == NULL)
95 			goto out;
96 
97 		(void) pset_assign(PS_QUERY, i, &ss->s_cpus[i].cs_pset_id);
98 		if (ss->s_cpus[i].cs_pset_id == PS_NONE)
99 			ss->s_cpus[i].cs_pset_id = ID_NO_PSET;
100 
101 		if (!CPU_ACTIVE(&ss->s_cpus[i]))
102 			continue;
103 
104 		if ((ksp = kstat_lookup_read(kc, "cpu", i, "vm")) == NULL)
105 			goto out;
106 
107 		if (kstat_copy(ksp, &ss->s_cpus[i].cs_vm))
108 			goto out;
109 
110 		if ((ksp = kstat_lookup_read(kc, "cpu", i, "sys")) == NULL)
111 			goto out;
112 
113 		if (kstat_copy(ksp, &ss->s_cpus[i].cs_sys))
114 			goto out;
115 	}
116 
117 	errno = 0;
118 out:
119 	return (errno);
120 }
121 
122 static int
123 acquire_psets(struct snapshot *ss)
124 {
125 	psetid_t *pids = NULL;
126 	struct pset_snapshot *ps;
127 	size_t pids_nr;
128 	size_t i, j;
129 
130 	/*
131 	 * Careful in this code. We have to use pset_list
132 	 * twice, but inbetween pids_nr can change at will.
133 	 * We delay the setting of s_nr_psets until we have
134 	 * the "final" value of pids_nr.
135 	 */
136 
137 	if (pset_list(NULL, &pids_nr) < 0)
138 		return (errno);
139 
140 	if ((pids = calloc(pids_nr, sizeof (psetid_t))) == NULL)
141 		goto out;
142 
143 	if (pset_list(pids, &pids_nr) < 0)
144 		goto out;
145 
146 	ss->s_psets = calloc(pids_nr + 1, sizeof (struct pset_snapshot));
147 	if (ss->s_psets == NULL)
148 		goto out;
149 	ss->s_nr_psets = pids_nr + 1;
150 
151 	/* CPUs not in any actual pset */
152 	ps = &ss->s_psets[0];
153 	ps->ps_id = 0;
154 	ps->ps_cpus = calloc(ss->s_nr_cpus, sizeof (struct cpu_snapshot *));
155 	if (ps->ps_cpus == NULL)
156 		goto out;
157 
158 	/* CPUs in a a pset */
159 	for (i = 1; i < ss->s_nr_psets; i++) {
160 		ps = &ss->s_psets[i];
161 
162 		ps->ps_id = pids[i - 1];
163 		ps->ps_cpus =
164 			calloc(ss->s_nr_cpus, sizeof (struct cpu_snapshot *));
165 		if (ps->ps_cpus == NULL)
166 			goto out;
167 	}
168 
169 	for (i = 0; i < ss->s_nr_psets; i++) {
170 		ps = &ss->s_psets[i];
171 
172 		for (j = 0; j < ss->s_nr_cpus; j++) {
173 			if (!CPU_ACTIVE(&ss->s_cpus[j]))
174 				continue;
175 			if (ss->s_cpus[j].cs_pset_id != ps->ps_id)
176 				continue;
177 
178 			ps->ps_cpus[ps->ps_nr_cpus++] = &ss->s_cpus[j];
179 		}
180 	}
181 
182 	errno = 0;
183 out:
184 	free(pids);
185 	return (errno);
186 }
187 
188 static int
189 acquire_intrs(struct snapshot *ss, kstat_ctl_t *kc)
190 {
191 	kstat_t *ksp;
192 	size_t i = 0;
193 	kstat_t *sys_misc;
194 	kstat_named_t *clock;
195 
196 	/* clock interrupt */
197 	ss->s_nr_intrs = 1;
198 
199 	for (ksp = kc->kc_chain; ksp; ksp = ksp->ks_next) {
200 		if (ksp->ks_type == KSTAT_TYPE_INTR)
201 			ss->s_nr_intrs++;
202 	}
203 
204 	ss->s_intrs = calloc(ss->s_nr_intrs, sizeof (struct intr_snapshot));
205 	if (ss->s_intrs == NULL)
206 		return (errno);
207 
208 	sys_misc = kstat_lookup_read(kc, "unix", 0, "system_misc");
209 	if (sys_misc == NULL)
210 		goto out;
211 
212 	clock = (kstat_named_t *)kstat_data_lookup(sys_misc, "clk_intr");
213 	if (clock == NULL)
214 		goto out;
215 
216 	(void) strlcpy(ss->s_intrs[0].is_name, "clock", KSTAT_STRLEN);
217 	ss->s_intrs[0].is_total = clock->value.ui32;
218 
219 	i = 1;
220 
221 	for (ksp = kc->kc_chain; ksp; ksp = ksp->ks_next) {
222 		kstat_intr_t *ki;
223 		int j;
224 
225 		if (ksp->ks_type != KSTAT_TYPE_INTR)
226 			continue;
227 		if (kstat_read(kc, ksp, NULL) == -1)
228 			goto out;
229 
230 		ki = KSTAT_INTR_PTR(ksp);
231 
232 		(void) strlcpy(ss->s_intrs[i].is_name, ksp->ks_name,
233 			KSTAT_STRLEN);
234 		ss->s_intrs[i].is_total = 0;
235 
236 		for (j = 0; j < KSTAT_NUM_INTRS; j++)
237 			ss->s_intrs[i].is_total += ki->intrs[j];
238 
239 		i++;
240 	}
241 
242 	errno = 0;
243 out:
244 	return (errno);
245 }
246 
247 int
248 acquire_sys(struct snapshot *ss, kstat_ctl_t *kc)
249 {
250 	size_t i;
251 	kstat_named_t *knp;
252 	kstat_t *ksp;
253 
254 	if ((ksp = kstat_lookup(kc, "unix", 0, "sysinfo")) == NULL)
255 		return (errno);
256 
257 	if (kstat_read(kc, ksp, &ss->s_sys.ss_sysinfo) == -1)
258 		return (errno);
259 
260 	if ((ksp = kstat_lookup(kc, "unix", 0, "vminfo")) == NULL)
261 		return (errno);
262 
263 	if (kstat_read(kc, ksp, &ss->s_sys.ss_vminfo) == -1)
264 		return (errno);
265 
266 	if ((ksp = kstat_lookup(kc, "unix", 0, "dnlcstats")) == NULL)
267 		return (errno);
268 
269 	if (kstat_read(kc, ksp, &ss->s_sys.ss_nc) == -1)
270 		return (errno);
271 
272 	if ((ksp = kstat_lookup(kc, "unix", 0, "system_misc")) == NULL)
273 		return (errno);
274 
275 	if (kstat_read(kc, ksp, NULL) == -1)
276 		return (errno);
277 
278 	knp = (kstat_named_t *)kstat_data_lookup(ksp, "clk_intr");
279 	if (knp == NULL)
280 		return (errno);
281 
282 	ss->s_sys.ss_ticks = knp->value.l;
283 
284 	knp = (kstat_named_t *)kstat_data_lookup(ksp, "deficit");
285 	if (knp == NULL)
286 		return (errno);
287 
288 	ss->s_sys.ss_deficit = knp->value.l;
289 
290 	for (i = 0; i < ss->s_nr_cpus; i++) {
291 		if (!CPU_ACTIVE(&ss->s_cpus[i]))
292 			continue;
293 
294 		if (kstat_add(&ss->s_cpus[i].cs_sys, &ss->s_sys.ss_agg_sys))
295 			return (errno);
296 		if (kstat_add(&ss->s_cpus[i].cs_vm, &ss->s_sys.ss_agg_vm))
297 			return (errno);
298 	}
299 
300 	return (0);
301 }
302 
303 struct snapshot *
304 acquire_snapshot(kstat_ctl_t *kc, int types, struct iodev_filter *iodev_filter)
305 {
306 	struct snapshot *ss = NULL;
307 	int err;
308 
309 retry:
310 	err = 0;
311 	/* ensure any partial resources are freed on a retry */
312 	free_snapshot(ss);
313 
314 	ss = safe_alloc(sizeof (struct snapshot));
315 
316 	(void) memset(ss, 0, sizeof (struct snapshot));
317 
318 	ss->s_types = types;
319 
320 	/* wait for a possibly up-to-date chain */
321 	while (kstat_chain_update(kc) == -1) {
322 		if (errno == EAGAIN)
323 			(void) poll(NULL, 0, RETRY_DELAY);
324 		else
325 			fail(1, "kstat_chain_update failed");
326 	}
327 
328 	if (types & SNAP_FLUSHES) {
329 		kstat_t *ksp;
330 		ksp = kstat_lookup(kc, "unix", 0, "flushmeter");
331 		if (ksp == NULL) {
332 			fail(0, "This machine does not have "
333 				"a virtual address cache");
334 		}
335 		if (kstat_read(kc, ksp, &ss->s_flushes) == -1)
336 			err = errno;
337 	}
338 
339 	if (!err && (types & SNAP_INTERRUPTS))
340 		err = acquire_intrs(ss, kc);
341 
342 	if (!err && (types & (SNAP_CPUS | SNAP_SYSTEM | SNAP_PSETS)))
343 		err = acquire_cpus(ss, kc);
344 
345 	if (!err && (types & SNAP_PSETS))
346 		err = acquire_psets(ss);
347 
348 	if (!err && (types & (SNAP_IODEVS | SNAP_CONTROLLERS |
349 	    SNAP_IOPATHS_LI | SNAP_IOPATHS_LTI)))
350 		err = acquire_iodevs(ss, kc, iodev_filter);
351 
352 	if (!err && (types & SNAP_SYSTEM))
353 		err = acquire_sys(ss, kc);
354 
355 	switch (err) {
356 		case 0:
357 			break;
358 		case EAGAIN:
359 			(void) poll(NULL, 0, RETRY_DELAY);
360 		/* a kstat disappeared from under us */
361 		/*FALLTHRU*/
362 		case ENXIO:
363 		case ENOENT:
364 			goto retry;
365 		default:
366 			fail(1, "acquiring snapshot failed");
367 	}
368 
369 	return (ss);
370 }
371 
372 void
373 free_snapshot(struct snapshot *ss)
374 {
375 	size_t i;
376 
377 	if (ss == NULL)
378 		return;
379 
380 	while (ss->s_iodevs) {
381 		struct iodev_snapshot *tmp = ss->s_iodevs;
382 		ss->s_iodevs = ss->s_iodevs->is_next;
383 		free_iodev(tmp);
384 	}
385 
386 	if (ss->s_cpus) {
387 		for (i = 0; i < ss->s_nr_cpus; i++) {
388 			free(ss->s_cpus[i].cs_vm.ks_data);
389 			free(ss->s_cpus[i].cs_sys.ks_data);
390 		}
391 		free(ss->s_cpus);
392 	}
393 
394 	if (ss->s_psets) {
395 		for (i = 0; i < ss->s_nr_psets; i++)
396 			free(ss->s_psets[i].ps_cpus);
397 		free(ss->s_psets);
398 	}
399 
400 	free(ss->s_sys.ss_agg_sys.ks_data);
401 	free(ss->s_sys.ss_agg_vm.ks_data);
402 	free(ss);
403 }
404 
405 kstat_ctl_t *
406 open_kstat(void)
407 {
408 	kstat_ctl_t *kc;
409 
410 	while ((kc = kstat_open()) == NULL) {
411 		if (errno == EAGAIN)
412 			(void) poll(NULL, 0, RETRY_DELAY);
413 		else
414 			fail(1, "kstat_open failed");
415 	}
416 
417 	return (kc);
418 }
419 
420 /*PRINTFLIKE2*/
421 void
422 fail(int do_perror, char *message, ...)
423 {
424 	va_list args;
425 	int save_errno = errno;
426 
427 	va_start(args, message);
428 	(void) fprintf(stderr, "%s: ", cmdname);
429 	(void) vfprintf(stderr, message, args);
430 	va_end(args);
431 	if (do_perror)
432 		(void) fprintf(stderr, ": %s", strerror(save_errno));
433 	(void) fprintf(stderr, "\n");
434 	exit(2);
435 }
436 
437 void *
438 safe_alloc(size_t size)
439 {
440 	void *ptr;
441 
442 	while ((ptr = malloc(size)) == NULL) {
443 		if (errno == EAGAIN)
444 			(void) poll(NULL, 0, RETRY_DELAY);
445 		else
446 			fail(1, "malloc failed");
447 	}
448 	return (ptr);
449 }
450 
451 char *
452 safe_strdup(char *str)
453 {
454 	char *ret;
455 
456 	if (str == NULL)
457 		return (NULL);
458 
459 	while ((ret = strdup(str)) == NULL) {
460 		if (errno == EAGAIN)
461 			(void) poll(NULL, 0, RETRY_DELAY);
462 		else
463 			fail(1, "malloc failed");
464 	}
465 	return (ret);
466 }
467 
468 uint64_t
469 kstat_delta(kstat_t *old, kstat_t *new, char *name)
470 {
471 	kstat_named_t *knew = kstat_data_lookup(new, name);
472 	if (old && old->ks_data) {
473 		kstat_named_t *kold = kstat_data_lookup(old, name);
474 		return (knew->value.ui64 - kold->value.ui64);
475 	}
476 	return (knew->value.ui64);
477 }
478 
479 int
480 kstat_copy(const kstat_t *src, kstat_t *dst)
481 {
482 	*dst = *src;
483 
484 	if (src->ks_data != NULL) {
485 		if ((dst->ks_data = malloc(src->ks_data_size)) == NULL)
486 			return (-1);
487 		bcopy(src->ks_data, dst->ks_data, src->ks_data_size);
488 	} else {
489 		dst->ks_data = NULL;
490 		dst->ks_data_size = 0;
491 	}
492 	return (0);
493 }
494 
495 int
496 kstat_add(const kstat_t *src, kstat_t *dst)
497 {
498 	size_t i;
499 	kstat_named_t *from;
500 	kstat_named_t *to;
501 
502 	if (dst->ks_data == NULL)
503 		return (kstat_copy(src, dst));
504 
505 	from = src->ks_data;
506 	to = dst->ks_data;
507 
508 	for (i = 0; i < src->ks_ndata; i++) {
509 		/* "addition" makes little sense for strings */
510 		if (from->data_type != KSTAT_DATA_CHAR &&
511 			from->data_type != KSTAT_DATA_STRING)
512 			(to)->value.ui64 += (from)->value.ui64;
513 		from++;
514 		to++;
515 	}
516 
517 	return (0);
518 }
519 
520 uint64_t
521 cpu_ticks_delta(kstat_t *old, kstat_t *new)
522 {
523 	uint64_t ticks = 0;
524 	size_t i;
525 	for (i = 0; i < ARRAY_SIZE(cpu_states); i++)
526 		ticks += kstat_delta(old, new, cpu_states[i]);
527 	return (ticks);
528 }
529 
530 int
531 nr_active_cpus(struct snapshot *ss)
532 {
533 	size_t i;
534 	int count = 0;
535 	for (i = 0; i < ss->s_nr_cpus; i++) {
536 		if (CPU_ACTIVE(&ss->s_cpus[i]))
537 			count++;
538 	}
539 
540 	return (count);
541 }
542 
543 /*
544  * Return the number of ticks delta between two hrtime_t
545  * values. Attempt to cater for various kinds of overflow
546  * in hrtime_t - no matter how improbable.
547  */
548 uint64_t
549 hrtime_delta(hrtime_t old, hrtime_t new)
550 {
551 	uint64_t del;
552 
553 	if ((new >= old) && (old >= 0L))
554 		return (new - old);
555 	else {
556 		/*
557 		 * We've overflowed the positive portion of an
558 		 * hrtime_t.
559 		 */
560 		if (new < 0L) {
561 			/*
562 			 * The new value is negative. Handle the
563 			 * case where the old value is positive or
564 			 * negative.
565 			 */
566 			uint64_t n1;
567 			uint64_t o1;
568 
569 			n1 = -new;
570 			if (old > 0L)
571 				return (n1 - old);
572 			else {
573 				o1 = -old;
574 				del = n1 - o1;
575 				return (del);
576 			}
577 		} else {
578 			/*
579 			 * Either we've just gone from being negative
580 			 * to positive *or* the last entry was positive
581 			 * and the new entry is also positive but *less*
582 			 * than the old entry. This implies we waited
583 			 * quite a few days on a very fast system between
584 			 * iostat displays.
585 			 */
586 			if (old < 0L) {
587 				uint64_t o2;
588 
589 				o2 = -old;
590 				del = UINT64_MAX - o2;
591 			} else {
592 				del = UINT64_MAX - old;
593 			}
594 			del += new;
595 			return (del);
596 		}
597 	}
598 }
599