2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1997, 1998, 1999 Kenneth D. Merry.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. The name of the author may not be used to endorse or promote products
16 * derived from this software without specific prior written permission.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
36 #include <sys/kernel.h>
37 #include <sys/systm.h>
39 #include <sys/devicestat.h>
41 #include <sys/sysctl.h>
42 #include <sys/malloc.h>
44 #include <sys/mutex.h>
49 #include <machine/atomic.h>
51 SDT_PROVIDER_DEFINE(io);
53 SDT_PROBE_DEFINE2(io, , , start, "struct bio *", "struct devstat *");
54 SDT_PROBE_DEFINE2(io, , , done, "struct bio *", "struct devstat *");
55 SDT_PROBE_DEFINE2(io, , , wait__start, "struct bio *",
57 SDT_PROBE_DEFINE2(io, , , wait__done, "struct bio *",
60 #define DTRACE_DEVSTAT_START() SDT_PROBE2(io, , , start, NULL, ds)
61 #define DTRACE_DEVSTAT_BIO_START() SDT_PROBE2(io, , , start, bp, ds)
62 #define DTRACE_DEVSTAT_DONE() SDT_PROBE2(io, , , done, NULL, ds)
63 #define DTRACE_DEVSTAT_BIO_DONE() SDT_PROBE2(io, , , done, bp, ds)
64 #define DTRACE_DEVSTAT_WAIT_START() SDT_PROBE2(io, , , wait__start, NULL, ds)
65 #define DTRACE_DEVSTAT_WAIT_DONE() SDT_PROBE2(io, , , wait__done, NULL, ds)
67 static int devstat_num_devs;
68 static long devstat_generation = 1;
69 static int devstat_version = DEVSTAT_VERSION;
70 static int devstat_current_devnumber;
71 static struct mtx devstat_mutex;
72 MTX_SYSINIT(devstat_mutex, &devstat_mutex, "devstat", MTX_DEF);
74 static struct devstatlist device_statq = STAILQ_HEAD_INITIALIZER(device_statq);
75 static struct devstat *devstat_alloc(void);
76 static void devstat_free(struct devstat *);
77 static void devstat_add_entry(struct devstat *ds, const void *dev_name,
78 int unit_number, uint32_t block_size,
79 devstat_support_flags flags,
80 devstat_type_flags device_type,
81 devstat_priority priority);
84 * Allocate a devstat and initialize it
87 devstat_new_entry(const void *dev_name,
88 int unit_number, uint32_t block_size,
89 devstat_support_flags flags,
90 devstat_type_flags device_type,
91 devstat_priority priority)
95 mtx_assert(&devstat_mutex, MA_NOTOWNED);
98 mtx_lock(&devstat_mutex);
99 if (unit_number == -1) {
100 ds->unit_number = unit_number;
102 binuptime(&ds->creation_time);
103 devstat_generation++;
105 devstat_add_entry(ds, dev_name, unit_number, block_size,
106 flags, device_type, priority);
108 mtx_unlock(&devstat_mutex);
113 * Take a malloced and zeroed devstat structure given to us, fill it in
114 * and add it to the queue of devices.
117 devstat_add_entry(struct devstat *ds, const void *dev_name,
118 int unit_number, uint32_t block_size,
119 devstat_support_flags flags,
120 devstat_type_flags device_type,
121 devstat_priority priority)
123 struct devstatlist *devstat_head;
124 struct devstat *ds_tmp;
126 mtx_assert(&devstat_mutex, MA_OWNED);
129 devstat_head = &device_statq;
132 * Priority sort. Each driver passes in its priority when it adds
133 * its devstat entry. Drivers are sorted first by priority, and
134 * then by probe order.
136 * For the first device, we just insert it, since the priority
137 * doesn't really matter yet. Subsequent devices are inserted into
138 * the list using the order outlined above.
140 if (devstat_num_devs == 1)
141 STAILQ_INSERT_TAIL(devstat_head, ds, dev_links);
143 STAILQ_FOREACH(ds_tmp, devstat_head, dev_links) {
144 struct devstat *ds_next;
146 ds_next = STAILQ_NEXT(ds_tmp, dev_links);
149 * If we find a break between higher and lower
150 * priority items, and if this item fits in the
151 * break, insert it. This also applies if the
152 * "lower priority item" is the end of the list.
154 if ((priority <= ds_tmp->priority)
155 && ((ds_next == NULL)
156 || (priority > ds_next->priority))) {
157 STAILQ_INSERT_AFTER(devstat_head, ds_tmp, ds,
160 } else if (priority > ds_tmp->priority) {
162 * If this is the case, we should be able
163 * to insert ourselves at the head of the
164 * list. If we can't, something is wrong.
166 if (ds_tmp == STAILQ_FIRST(devstat_head)) {
167 STAILQ_INSERT_HEAD(devstat_head,
171 STAILQ_INSERT_TAIL(devstat_head,
173 printf("devstat_add_entry: HELP! "
174 "sorting problem detected "
175 "for name %p unit %d\n",
176 dev_name, unit_number);
183 ds->device_number = devstat_current_devnumber++;
184 ds->unit_number = unit_number;
185 strlcpy(ds->device_name, dev_name, DEVSTAT_NAME_LEN);
186 ds->block_size = block_size;
188 ds->device_type = device_type;
189 ds->priority = priority;
190 binuptime(&ds->creation_time);
191 devstat_generation++;
195 * Remove a devstat structure from the list of devices.
198 devstat_remove_entry(struct devstat *ds)
200 struct devstatlist *devstat_head;
202 mtx_assert(&devstat_mutex, MA_NOTOWNED);
206 mtx_lock(&devstat_mutex);
208 devstat_head = &device_statq;
210 /* Remove this entry from the devstat queue */
211 atomic_add_acq_int(&ds->sequence1, 1);
212 if (ds->unit_number != -1) {
214 STAILQ_REMOVE(devstat_head, ds, devstat, dev_links);
217 devstat_generation++;
218 mtx_unlock(&devstat_mutex);
222 * Record a transaction start.
224 * See comments for devstat_end_transaction(). Ordering is very important
228 devstat_start_transaction(struct devstat *ds, const struct bintime *now)
235 atomic_add_acq_int(&ds->sequence1, 1);
237 * We only want to set the start time when we are going from idle
238 * to busy. The start time is really the start of the latest busy
241 if (atomic_fetchadd_int(&ds->start_count, 1) == ds->end_count) {
243 ds->busy_from = *now;
245 binuptime(&ds->busy_from);
247 atomic_add_rel_int(&ds->sequence0, 1);
248 DTRACE_DEVSTAT_START();
252 devstat_start_transaction_bio(struct devstat *ds, struct bio *bp)
259 binuptime(&bp->bio_t0);
260 devstat_start_transaction_bio_t0(ds, bp);
264 devstat_start_transaction_bio_t0(struct devstat *ds, struct bio *bp)
271 devstat_start_transaction(ds, &bp->bio_t0);
272 DTRACE_DEVSTAT_BIO_START();
276 * Record the ending of a transaction, and incrment the various counters.
278 * Ordering in this function, and in devstat_start_transaction() is VERY
279 * important. The idea here is to run without locks, so we are very
280 * careful to only modify some fields on the way "down" (i.e. at
281 * transaction start) and some fields on the way "up" (i.e. at transaction
282 * completion). One exception is busy_from, which we only modify in
283 * devstat_start_transaction() when there are no outstanding transactions,
284 * and thus it can't be modified in devstat_end_transaction()
287 * The sequence0 and sequence1 fields are provided to enable an application
288 * spying on the structures with mmap(2) to tell when a structure is in a
289 * consistent state or not.
291 * For this to work 100% reliably, it is important that the two fields
292 * are at opposite ends of the structure and that they are incremented
293 * in the opposite order of how a memcpy(3) in userland would copy them.
294 * We assume that the copying happens front to back, but there is actually
295 * no way short of writing your own memcpy(3) replacement to guarantee
296 * this will be the case.
298 * In addition to this, being a kind of locks, they must be updated with
299 * atomic instructions using appropriate memory barriers.
302 devstat_end_transaction(struct devstat *ds, uint32_t bytes,
303 devstat_tag_type tag_type, devstat_trans_flags flags,
304 const struct bintime *now, const struct bintime *then)
306 struct bintime dt, lnow;
317 atomic_add_acq_int(&ds->sequence1, 1);
318 /* Update byte and operations counts */
319 ds->bytes[flags] += bytes;
320 ds->operations[flags]++;
323 * Keep a count of the various tag types sent.
325 if ((ds->flags & DEVSTAT_NO_ORDERED_TAGS) == 0 &&
326 tag_type != DEVSTAT_TAG_NONE)
327 ds->tag_types[tag_type]++;
330 /* Update duration of operations */
332 bintime_sub(&dt, then);
333 bintime_add(&ds->duration[flags], &dt);
336 /* Accumulate busy time */
338 bintime_sub(&dt, &ds->busy_from);
339 bintime_add(&ds->busy_time, &dt);
340 ds->busy_from = *now;
343 atomic_add_rel_int(&ds->sequence0, 1);
344 DTRACE_DEVSTAT_DONE();
348 devstat_end_transaction_bio(struct devstat *ds, const struct bio *bp)
351 devstat_end_transaction_bio_bt(ds, bp, NULL);
355 devstat_end_transaction_bio_bt(struct devstat *ds, const struct bio *bp,
356 const struct bintime *now)
358 devstat_trans_flags flg;
359 devstat_tag_type tag;
365 if (bp->bio_flags & BIO_ORDERED)
366 tag = DEVSTAT_TAG_ORDERED;
368 tag = DEVSTAT_TAG_SIMPLE;
369 if (bp->bio_cmd == BIO_DELETE)
371 else if ((bp->bio_cmd == BIO_READ)
372 || ((bp->bio_cmd == BIO_ZONE)
373 && (bp->bio_zone.zone_cmd == DISK_ZONE_REPORT_ZONES)))
375 else if (bp->bio_cmd == BIO_WRITE)
378 flg = DEVSTAT_NO_DATA;
380 devstat_end_transaction(ds, bp->bio_bcount - bp->bio_resid,
381 tag, flg, now, &bp->bio_t0);
382 DTRACE_DEVSTAT_BIO_DONE();
386 * This is the sysctl handler for the devstat package. The data pushed out
387 * on the kern.devstat.all sysctl variable consists of the current devstat
388 * generation number, and then an array of devstat structures, one for each
389 * device in the system.
391 * This is more cryptic that obvious, but basically we neither can nor
392 * want to hold the devstat_mutex for any amount of time, so we grab it
393 * only when we need to and keep an eye on devstat_generation all the time.
396 sysctl_devstat(SYSCTL_HANDLER_ARGS)
402 mtx_assert(&devstat_mutex, MA_NOTOWNED);
405 * XXX devstat_generation should really be "volatile" but that
406 * XXX freaks out the sysctl macro below. The places where we
407 * XXX change it and inspect it are bracketed in the mutex which
408 * XXX guarantees us proper write barriers. I don't believe the
409 * XXX compiler is allowed to optimize mygen away across calls
410 * XXX to other functions, so the following is belived to be safe.
412 mygen = devstat_generation;
414 error = SYSCTL_OUT(req, &mygen, sizeof(mygen));
416 if (devstat_num_devs == 0)
422 mtx_lock(&devstat_mutex);
423 nds = STAILQ_FIRST(&device_statq);
424 if (mygen != devstat_generation)
426 mtx_unlock(&devstat_mutex);
431 for (;nds != NULL;) {
432 error = SYSCTL_OUT(req, nds, sizeof(struct devstat));
435 mtx_lock(&devstat_mutex);
436 if (mygen != devstat_generation)
439 nds = STAILQ_NEXT(nds, dev_links);
440 mtx_unlock(&devstat_mutex);
448 * Sysctl entries for devstat. The first one is a node that all the rest
451 static SYSCTL_NODE(_kern, OID_AUTO, devstat, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
452 "Device Statistics");
454 SYSCTL_PROC(_kern_devstat, OID_AUTO, all,
455 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, NULL, 0,
456 sysctl_devstat, "S,devstat",
457 "All devices in the devstat list");
459 * Export the number of devices in the system so that userland utilities
460 * can determine how much memory to allocate to hold all the devices.
462 SYSCTL_INT(_kern_devstat, OID_AUTO, numdevs, CTLFLAG_RD,
463 &devstat_num_devs, 0, "Number of devices in the devstat list");
464 SYSCTL_LONG(_kern_devstat, OID_AUTO, generation, CTLFLAG_RD,
465 &devstat_generation, 0, "Devstat list generation");
466 SYSCTL_INT(_kern_devstat, OID_AUTO, version, CTLFLAG_RD,
467 &devstat_version, 0, "Devstat list version number");
470 * Allocator for struct devstat structures. We sub-allocate these from pages
471 * which we get from malloc. These pages are exported for mmap(2)'ing through
472 * a miniature device driver
475 #define statsperpage (PAGE_SIZE / sizeof(struct devstat))
477 static d_ioctl_t devstat_ioctl;
478 static d_mmap_t devstat_mmap;
480 static struct cdevsw devstat_cdevsw = {
481 .d_version = D_VERSION,
482 .d_ioctl = devstat_ioctl,
483 .d_mmap = devstat_mmap,
488 TAILQ_ENTRY(statspage) list;
489 struct devstat *stat;
493 static size_t pagelist_pages = 0;
494 static TAILQ_HEAD(, statspage) pagelist = TAILQ_HEAD_INITIALIZER(pagelist);
495 static MALLOC_DEFINE(M_DEVSTAT, "devstat", "Device statistics");
498 devstat_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag,
506 *(off_t *)data = pagelist_pages * PAGE_SIZE;
514 devstat_mmap(struct cdev *dev, vm_ooffset_t offset, vm_paddr_t *paddr,
515 int nprot, vm_memattr_t *memattr)
517 struct statspage *spp;
519 if (nprot != VM_PROT_READ)
521 mtx_lock(&devstat_mutex);
522 TAILQ_FOREACH(spp, &pagelist, list) {
524 *paddr = vtophys(spp->stat);
525 mtx_unlock(&devstat_mutex);
530 mtx_unlock(&devstat_mutex);
534 static struct devstat *
538 struct statspage *spp, *spp2;
542 mtx_assert(&devstat_mutex, MA_NOTOWNED);
544 make_dev_credf(MAKEDEV_ETERNAL | MAKEDEV_CHECKNAME,
545 &devstat_cdevsw, 0, NULL, UID_ROOT, GID_WHEEL, 0444,
546 DEVSTAT_DEVICE_NAME);
550 mtx_lock(&devstat_mutex);
552 TAILQ_FOREACH(spp, &pagelist, list) {
558 mtx_unlock(&devstat_mutex);
559 spp2 = malloc(sizeof *spp, M_DEVSTAT, M_ZERO | M_WAITOK);
560 spp2->stat = malloc(PAGE_SIZE, M_DEVSTAT, M_ZERO | M_WAITOK);
561 spp2->nfree = statsperpage;
564 * If free statspages were added while the lock was released
567 mtx_lock(&devstat_mutex);
568 TAILQ_FOREACH(spp, &pagelist, list)
575 * It would make more sense to add the new page at the
576 * head but the order on the list determine the
577 * sequence of the mapping so we can't do that.
580 TAILQ_INSERT_TAIL(&pagelist, spp, list);
585 for (u = 0; u < statsperpage; u++) {
586 if (dsp->allocated == 0)
592 mtx_unlock(&devstat_mutex);
593 if (spp2 != NULL && spp2 != spp) {
594 free(spp2->stat, M_DEVSTAT);
595 free(spp2, M_DEVSTAT);
601 devstat_free(struct devstat *dsp)
603 struct statspage *spp;
605 mtx_assert(&devstat_mutex, MA_OWNED);
606 bzero(dsp, sizeof *dsp);
607 TAILQ_FOREACH(spp, &pagelist, list) {
608 if (dsp >= spp->stat && dsp < (spp->stat + statsperpage)) {
615 SYSCTL_INT(_debug_sizeof, OID_AUTO, devstat, CTLFLAG_RD,
616 SYSCTL_NULL_INT_PTR, sizeof(struct devstat), "sizeof(struct devstat)");