1 /*-
2 * Copyright (c) 1997, 1998, 1999 Kenneth D. Merry.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. The name of the author may not be used to endorse or promote products
14 * derived from this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31
32 #include <sys/param.h>
33 #include <sys/kernel.h>
34 #include <sys/systm.h>
35 #include <sys/bio.h>
36 #include <sys/devicestat.h>
37 #include <sys/sdt.h>
38 #include <sys/sysctl.h>
39 #include <sys/malloc.h>
40 #include <sys/lock.h>
41 #include <sys/mutex.h>
42 #include <sys/conf.h>
43 #include <vm/vm.h>
44 #include <vm/pmap.h>
45
46 #include <machine/atomic.h>
47
48 SDT_PROVIDER_DEFINE(io);
49
50 SDT_PROBE_DEFINE2(io, , , start, "struct bio *", "struct devstat *");
51 SDT_PROBE_DEFINE2(io, , , done, "struct bio *", "struct devstat *");
52 SDT_PROBE_DEFINE2(io, , , wait__start, "struct bio *",
53 "struct devstat *");
54 SDT_PROBE_DEFINE2(io, , , wait__done, "struct bio *",
55 "struct devstat *");
56
57 #define DTRACE_DEVSTAT_START() SDT_PROBE2(io, , , start, NULL, ds)
58 #define DTRACE_DEVSTAT_BIO_START() SDT_PROBE2(io, , , start, bp, ds)
59 #define DTRACE_DEVSTAT_DONE() SDT_PROBE2(io, , , done, NULL, ds)
60 #define DTRACE_DEVSTAT_BIO_DONE() SDT_PROBE2(io, , , done, bp, ds)
61 #define DTRACE_DEVSTAT_WAIT_START() SDT_PROBE2(io, , , wait__start, NULL, ds)
62 #define DTRACE_DEVSTAT_WAIT_DONE() SDT_PROBE2(io, , , wait__done, NULL, ds)
63
64 static int devstat_num_devs;
65 static long devstat_generation = 1;
66 static int devstat_version = DEVSTAT_VERSION;
67 static int devstat_current_devnumber;
68 static struct mtx devstat_mutex;
69 MTX_SYSINIT(devstat_mutex, &devstat_mutex, "devstat", MTX_DEF);
70
71 static struct devstatlist device_statq = STAILQ_HEAD_INITIALIZER(device_statq);
72 static struct devstat *devstat_alloc(void);
73 static void devstat_free(struct devstat *);
74 static void devstat_add_entry(struct devstat *ds, const void *dev_name,
75 int unit_number, uint32_t block_size,
76 devstat_support_flags flags,
77 devstat_type_flags device_type,
78 devstat_priority priority);
79
80 /*
81 * Allocate a devstat and initialize it
82 */
83 struct devstat *
84 devstat_new_entry(const void *dev_name,
85 int unit_number, uint32_t block_size,
86 devstat_support_flags flags,
87 devstat_type_flags device_type,
88 devstat_priority priority)
89 {
90 struct devstat *ds;
91
92 mtx_assert(&devstat_mutex, MA_NOTOWNED);
93
94 ds = devstat_alloc();
95 mtx_lock(&devstat_mutex);
96 if (unit_number == -1) {
97 ds->unit_number = unit_number;
98 ds->id = dev_name;
99 binuptime(&ds->creation_time);
100 devstat_generation++;
101 } else {
102 devstat_add_entry(ds, dev_name, unit_number, block_size,
103 flags, device_type, priority);
104 }
105 mtx_unlock(&devstat_mutex);
106 return (ds);
107 }
108
109 /*
110 * Take a malloced and zeroed devstat structure given to us, fill it in
111 * and add it to the queue of devices.
112 */
113 static void
114 devstat_add_entry(struct devstat *ds, const void *dev_name,
115 int unit_number, uint32_t block_size,
116 devstat_support_flags flags,
117 devstat_type_flags device_type,
118 devstat_priority priority)
119 {
120 struct devstatlist *devstat_head;
121 struct devstat *ds_tmp;
122
123 mtx_assert(&devstat_mutex, MA_OWNED);
124 devstat_num_devs++;
125
126 devstat_head = &device_statq;
127
128 /*
129 * Priority sort. Each driver passes in its priority when it adds
130 * its devstat entry. Drivers are sorted first by priority, and
131 * then by probe order.
132 *
133 * For the first device, we just insert it, since the priority
134 * doesn't really matter yet. Subsequent devices are inserted into
135 * the list using the order outlined above.
136 */
137 if (devstat_num_devs == 1)
138 STAILQ_INSERT_TAIL(devstat_head, ds, dev_links);
139 else {
140 STAILQ_FOREACH(ds_tmp, devstat_head, dev_links) {
141 struct devstat *ds_next;
142
143 ds_next = STAILQ_NEXT(ds_tmp, dev_links);
144
145 /*
146 * If we find a break between higher and lower
147 * priority items, and if this item fits in the
148 * break, insert it. This also applies if the
149 * "lower priority item" is the end of the list.
150 */
151 if ((priority <= ds_tmp->priority)
152 && ((ds_next == NULL)
153 || (priority > ds_next->priority))) {
154 STAILQ_INSERT_AFTER(devstat_head, ds_tmp, ds,
155 dev_links);
156 break;
157 } else if (priority > ds_tmp->priority) {
158 /*
159 * If this is the case, we should be able
160 * to insert ourselves at the head of the
161 * list. If we can't, something is wrong.
162 */
163 if (ds_tmp == STAILQ_FIRST(devstat_head)) {
164 STAILQ_INSERT_HEAD(devstat_head,
165 ds, dev_links);
166 break;
167 } else {
168 STAILQ_INSERT_TAIL(devstat_head,
169 ds, dev_links);
170 printf("devstat_add_entry: HELP! "
171 "sorting problem detected "
172 "for name %p unit %d\n",
173 dev_name, unit_number);
174 break;
175 }
176 }
177 }
178 }
179
180 ds->device_number = devstat_current_devnumber++;
181 ds->unit_number = unit_number;
182 strlcpy(ds->device_name, dev_name, DEVSTAT_NAME_LEN);
183 ds->block_size = block_size;
184 ds->flags = flags;
185 ds->device_type = device_type;
186 ds->priority = priority;
187 binuptime(&ds->creation_time);
188 devstat_generation++;
189 }
190
191 /*
192 * Remove a devstat structure from the list of devices.
193 */
194 void
195 devstat_remove_entry(struct devstat *ds)
196 {
197 struct devstatlist *devstat_head;
198
199 mtx_assert(&devstat_mutex, MA_NOTOWNED);
200 if (ds == NULL)
201 return;
202
203 mtx_lock(&devstat_mutex);
204
205 devstat_head = &device_statq;
206
207 /* Remove this entry from the devstat queue */
208 atomic_add_acq_int(&ds->sequence1, 1);
209 if (ds->unit_number != -1) {
210 devstat_num_devs--;
211 STAILQ_REMOVE(devstat_head, ds, devstat, dev_links);
212 }
213 devstat_free(ds);
214 devstat_generation++;
215 mtx_unlock(&devstat_mutex);
216 }
217
218 /*
219 * Record a transaction start.
220 *
221 * See comments for devstat_end_transaction(). Ordering is very important
222 * here.
223 */
224 void
225 devstat_start_transaction(struct devstat *ds, struct bintime *now)
226 {
227
228 mtx_assert(&devstat_mutex, MA_NOTOWNED);
229
230 /* sanity check */
231 if (ds == NULL)
232 return;
233
234 atomic_add_acq_int(&ds->sequence1, 1);
235 /*
236 * We only want to set the start time when we are going from idle
237 * to busy. The start time is really the start of the latest busy
238 * period.
239 */
240 if (ds->start_count == ds->end_count) {
241 if (now != NULL)
242 ds->busy_from = *now;
243 else
244 binuptime(&ds->busy_from);
245 }
246 ds->start_count++;
247 atomic_add_rel_int(&ds->sequence0, 1);
248 DTRACE_DEVSTAT_START();
249 }
250
251 void
252 devstat_start_transaction_bio(struct devstat *ds, struct bio *bp)
253 {
254
255 mtx_assert(&devstat_mutex, MA_NOTOWNED);
256
257 /* sanity check */
258 if (ds == NULL)
259 return;
260
261 binuptime(&bp->bio_t0);
262 devstat_start_transaction(ds, &bp->bio_t0);
263 DTRACE_DEVSTAT_BIO_START();
264 }
265
266 /*
267 * Record the ending of a transaction, and incrment the various counters.
268 *
269 * Ordering in this function, and in devstat_start_transaction() is VERY
270 * important. The idea here is to run without locks, so we are very
271 * careful to only modify some fields on the way "down" (i.e. at
272 * transaction start) and some fields on the way "up" (i.e. at transaction
273 * completion). One exception is busy_from, which we only modify in
274 * devstat_start_transaction() when there are no outstanding transactions,
275 * and thus it can't be modified in devstat_end_transaction()
276 * simultaneously.
277 *
278 * The sequence0 and sequence1 fields are provided to enable an application
279 * spying on the structures with mmap(2) to tell when a structure is in a
280 * consistent state or not.
281 *
282 * For this to work 100% reliably, it is important that the two fields
283 * are at opposite ends of the structure and that they are incremented
284 * in the opposite order of how a memcpy(3) in userland would copy them.
285 * We assume that the copying happens front to back, but there is actually
286 * no way short of writing your own memcpy(3) replacement to guarantee
287 * this will be the case.
288 *
289 * In addition to this, being a kind of locks, they must be updated with
290 * atomic instructions using appropriate memory barriers.
291 */
292 void
293 devstat_end_transaction(struct devstat *ds, uint32_t bytes,
294 devstat_tag_type tag_type, devstat_trans_flags flags,
295 struct bintime *now, struct bintime *then)
296 {
297 struct bintime dt, lnow;
298
299 /* sanity check */
300 if (ds == NULL)
301 return;
302
303 if (now == NULL) {
304 now = &lnow;
305 binuptime(now);
306 }
307
308 atomic_add_acq_int(&ds->sequence1, 1);
309 /* Update byte and operations counts */
310 ds->bytes[flags] += bytes;
311 ds->operations[flags]++;
312
313 /*
314 * Keep a count of the various tag types sent.
315 */
316 if ((ds->flags & DEVSTAT_NO_ORDERED_TAGS) == 0 &&
317 tag_type != DEVSTAT_TAG_NONE)
318 ds->tag_types[tag_type]++;
319
320 if (then != NULL) {
321 /* Update duration of operations */
322 dt = *now;
323 bintime_sub(&dt, then);
324 bintime_add(&ds->duration[flags], &dt);
325 }
326
327 /* Accumulate busy time */
328 dt = *now;
329 bintime_sub(&dt, &ds->busy_from);
330 bintime_add(&ds->busy_time, &dt);
331 ds->busy_from = *now;
332
333 ds->end_count++;
334 atomic_add_rel_int(&ds->sequence0, 1);
335 DTRACE_DEVSTAT_DONE();
336 }
337
338 void
339 devstat_end_transaction_bio(struct devstat *ds, struct bio *bp)
340 {
341
342 devstat_end_transaction_bio_bt(ds, bp, NULL);
343 }
344
345 void
346 devstat_end_transaction_bio_bt(struct devstat *ds, struct bio *bp,
347 struct bintime *now)
348 {
349 devstat_trans_flags flg;
350
351 /* sanity check */
352 if (ds == NULL)
353 return;
354
355 if (bp->bio_cmd == BIO_DELETE)
356 flg = DEVSTAT_FREE;
357 else if ((bp->bio_cmd == BIO_READ)
358 || ((bp->bio_cmd == BIO_ZONE)
359 && (bp->bio_zone.zone_cmd == DISK_ZONE_REPORT_ZONES)))
360 flg = DEVSTAT_READ;
361 else if (bp->bio_cmd == BIO_WRITE)
362 flg = DEVSTAT_WRITE;
363 else
364 flg = DEVSTAT_NO_DATA;
365
366 devstat_end_transaction(ds, bp->bio_bcount - bp->bio_resid,
367 DEVSTAT_TAG_SIMPLE, flg, now, &bp->bio_t0);
368 DTRACE_DEVSTAT_BIO_DONE();
369 }
370
371 /*
372 * This is the sysctl handler for the devstat package. The data pushed out
373 * on the kern.devstat.all sysctl variable consists of the current devstat
374 * generation number, and then an array of devstat structures, one for each
375 * device in the system.
376 *
377 * This is more cryptic that obvious, but basically we neither can nor
378 * want to hold the devstat_mutex for any amount of time, so we grab it
379 * only when we need to and keep an eye on devstat_generation all the time.
380 */
381 static int
382 sysctl_devstat(SYSCTL_HANDLER_ARGS)
383 {
384 int error;
385 long mygen;
386 struct devstat *nds;
387
388 mtx_assert(&devstat_mutex, MA_NOTOWNED);
389
390 /*
391 * XXX devstat_generation should really be "volatile" but that
392 * XXX freaks out the sysctl macro below. The places where we
393 * XXX change it and inspect it are bracketed in the mutex which
394 * XXX guarantees us proper write barriers. I don't believe the
395 * XXX compiler is allowed to optimize mygen away across calls
396 * XXX to other functions, so the following is belived to be safe.
397 */
398 mygen = devstat_generation;
399
400 error = SYSCTL_OUT(req, &mygen, sizeof(mygen));
401
402 if (devstat_num_devs == 0)
403 return(0);
404
405 if (error != 0)
406 return (error);
407
408 mtx_lock(&devstat_mutex);
409 nds = STAILQ_FIRST(&device_statq);
410 if (mygen != devstat_generation)
411 error = EBUSY;
412 mtx_unlock(&devstat_mutex);
413
414 if (error != 0)
415 return (error);
416
417 for (;nds != NULL;) {
418 error = SYSCTL_OUT(req, nds, sizeof(struct devstat));
419 if (error != 0)
420 return (error);
421 mtx_lock(&devstat_mutex);
422 if (mygen != devstat_generation)
423 error = EBUSY;
424 else
425 nds = STAILQ_NEXT(nds, dev_links);
426 mtx_unlock(&devstat_mutex);
427 if (error != 0)
428 return (error);
429 }
430 return(error);
431 }
432
433 /*
434 * Sysctl entries for devstat. The first one is a node that all the rest
435 * hang off of.
436 */
437 static SYSCTL_NODE(_kern, OID_AUTO, devstat, CTLFLAG_RD, NULL,
438 "Device Statistics");
439
440 SYSCTL_PROC(_kern_devstat, OID_AUTO, all, CTLFLAG_RD|CTLTYPE_OPAQUE,
441 NULL, 0, sysctl_devstat, "S,devstat", "All devices in the devstat list");
442 /*
443 * Export the number of devices in the system so that userland utilities
444 * can determine how much memory to allocate to hold all the devices.
445 */
446 SYSCTL_INT(_kern_devstat, OID_AUTO, numdevs, CTLFLAG_RD,
447 &devstat_num_devs, 0, "Number of devices in the devstat list");
448 SYSCTL_LONG(_kern_devstat, OID_AUTO, generation, CTLFLAG_RD,
449 &devstat_generation, 0, "Devstat list generation");
450 SYSCTL_INT(_kern_devstat, OID_AUTO, version, CTLFLAG_RD,
451 &devstat_version, 0, "Devstat list version number");
452
453 /*
454 * Allocator for struct devstat structures. We sub-allocate these from pages
455 * which we get from malloc. These pages are exported for mmap(2)'ing through
456 * a miniature device driver
457 */
458
459 #define statsperpage (PAGE_SIZE / sizeof(struct devstat))
460
461 static d_mmap_t devstat_mmap;
462
463 static struct cdevsw devstat_cdevsw = {
464 .d_version = D_VERSION,
465 .d_mmap = devstat_mmap,
466 .d_name = "devstat",
467 };
468
469 struct statspage {
470 TAILQ_ENTRY(statspage) list;
471 struct devstat *stat;
472 u_int nfree;
473 };
474
475 static TAILQ_HEAD(, statspage) pagelist = TAILQ_HEAD_INITIALIZER(pagelist);
476 static MALLOC_DEFINE(M_DEVSTAT, "devstat", "Device statistics");
477
478 static int
479 devstat_mmap(struct cdev *dev, vm_ooffset_t offset, vm_paddr_t *paddr,
480 int nprot, vm_memattr_t *memattr)
481 {
482 struct statspage *spp;
483
484 if (nprot != VM_PROT_READ)
485 return (-1);
486 mtx_lock(&devstat_mutex);
487 TAILQ_FOREACH(spp, &pagelist, list) {
488 if (offset == 0) {
489 *paddr = vtophys(spp->stat);
490 mtx_unlock(&devstat_mutex);
491 return (0);
492 }
493 offset -= PAGE_SIZE;
494 }
495 mtx_unlock(&devstat_mutex);
496 return (-1);
497 }
498
499 static struct devstat *
500 devstat_alloc(void)
501 {
502 struct devstat *dsp;
503 struct statspage *spp, *spp2;
504 u_int u;
505 static int once;
506
507 mtx_assert(&devstat_mutex, MA_NOTOWNED);
508 if (!once) {
509 make_dev_credf(MAKEDEV_ETERNAL | MAKEDEV_CHECKNAME,
510 &devstat_cdevsw, 0, NULL, UID_ROOT, GID_WHEEL, 0444,
511 DEVSTAT_DEVICE_NAME);
512 once = 1;
513 }
514 spp2 = NULL;
515 mtx_lock(&devstat_mutex);
516 for (;;) {
517 TAILQ_FOREACH(spp, &pagelist, list) {
518 if (spp->nfree > 0)
519 break;
520 }
521 if (spp != NULL)
522 break;
523 mtx_unlock(&devstat_mutex);
524 spp2 = malloc(sizeof *spp, M_DEVSTAT, M_ZERO | M_WAITOK);
525 spp2->stat = malloc(PAGE_SIZE, M_DEVSTAT, M_ZERO | M_WAITOK);
526 spp2->nfree = statsperpage;
527
528 /*
529 * If free statspages were added while the lock was released
530 * just reuse them.
531 */
532 mtx_lock(&devstat_mutex);
533 TAILQ_FOREACH(spp, &pagelist, list)
534 if (spp->nfree > 0)
535 break;
536 if (spp == NULL) {
537 spp = spp2;
538
539 /*
540 * It would make more sense to add the new page at the
541 * head but the order on the list determine the
542 * sequence of the mapping so we can't do that.
543 */
544 TAILQ_INSERT_TAIL(&pagelist, spp, list);
545 } else
546 break;
547 }
548 dsp = spp->stat;
549 for (u = 0; u < statsperpage; u++) {
550 if (dsp->allocated == 0)
551 break;
552 dsp++;
553 }
554 spp->nfree--;
555 dsp->allocated = 1;
556 mtx_unlock(&devstat_mutex);
557 if (spp2 != NULL && spp2 != spp) {
558 free(spp2->stat, M_DEVSTAT);
559 free(spp2, M_DEVSTAT);
560 }
561 return (dsp);
562 }
563
564 static void
565 devstat_free(struct devstat *dsp)
566 {
567 struct statspage *spp;
568
569 mtx_assert(&devstat_mutex, MA_OWNED);
570 bzero(dsp, sizeof *dsp);
571 TAILQ_FOREACH(spp, &pagelist, list) {
572 if (dsp >= spp->stat && dsp < (spp->stat + statsperpage)) {
573 spp->nfree++;
574 return;
575 }
576 }
577 }
578
579 SYSCTL_INT(_debug_sizeof, OID_AUTO, devstat, CTLFLAG_RD,
580 SYSCTL_NULL_INT_PTR, sizeof(struct devstat), "sizeof(struct devstat)");
Cache object: 0f36361d0286118e608fafc4edc711f4
|