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 *");
56 #define DTRACE_DEVSTAT_BIO_START() SDT_PROBE2(io, , , start, bp, ds)
57 #define DTRACE_DEVSTAT_BIO_DONE() SDT_PROBE2(io, , , done, bp, ds)
59 static int devstat_num_devs;
60 static long devstat_generation = 1;
61 static int devstat_version = DEVSTAT_VERSION;
62 static int devstat_current_devnumber;
63 static struct mtx devstat_mutex;
64 MTX_SYSINIT(devstat_mutex, &devstat_mutex, "devstat", MTX_DEF);
66 static struct devstatlist device_statq = STAILQ_HEAD_INITIALIZER(device_statq);
67 static struct devstat *devstat_alloc(void);
68 static void devstat_free(struct devstat *);
69 static void devstat_add_entry(struct devstat *ds, const void *dev_name,
70 int unit_number, uint32_t block_size,
71 devstat_support_flags flags,
72 devstat_type_flags device_type,
73 devstat_priority priority);
76 * Allocate a devstat and initialize it
79 devstat_new_entry(const void *dev_name,
80 int unit_number, uint32_t block_size,
81 devstat_support_flags flags,
82 devstat_type_flags device_type,
83 devstat_priority priority)
87 mtx_assert(&devstat_mutex, MA_NOTOWNED);
90 mtx_lock(&devstat_mutex);
91 if (unit_number == -1) {
92 ds->unit_number = unit_number;
94 binuptime(&ds->creation_time);
97 devstat_add_entry(ds, dev_name, unit_number, block_size,
98 flags, device_type, priority);
100 mtx_unlock(&devstat_mutex);
105 * Take a malloced and zeroed devstat structure given to us, fill it in
106 * and add it to the queue of devices.
109 devstat_add_entry(struct devstat *ds, const void *dev_name,
110 int unit_number, uint32_t block_size,
111 devstat_support_flags flags,
112 devstat_type_flags device_type,
113 devstat_priority priority)
115 struct devstatlist *devstat_head;
116 struct devstat *ds_tmp;
118 mtx_assert(&devstat_mutex, MA_OWNED);
121 devstat_head = &device_statq;
124 * Priority sort. Each driver passes in its priority when it adds
125 * its devstat entry. Drivers are sorted first by priority, and
126 * then by probe order.
128 * For the first device, we just insert it, since the priority
129 * doesn't really matter yet. Subsequent devices are inserted into
130 * the list using the order outlined above.
132 if (devstat_num_devs == 1)
133 STAILQ_INSERT_TAIL(devstat_head, ds, dev_links);
135 STAILQ_FOREACH(ds_tmp, devstat_head, dev_links) {
136 struct devstat *ds_next;
138 ds_next = STAILQ_NEXT(ds_tmp, dev_links);
141 * If we find a break between higher and lower
142 * priority items, and if this item fits in the
143 * break, insert it. This also applies if the
144 * "lower priority item" is the end of the list.
146 if ((priority <= ds_tmp->priority)
147 && ((ds_next == NULL)
148 || (priority > ds_next->priority))) {
149 STAILQ_INSERT_AFTER(devstat_head, ds_tmp, ds,
152 } else if (priority > ds_tmp->priority) {
154 * If this is the case, we should be able
155 * to insert ourselves at the head of the
156 * list. If we can't, something is wrong.
158 if (ds_tmp == STAILQ_FIRST(devstat_head)) {
159 STAILQ_INSERT_HEAD(devstat_head,
163 STAILQ_INSERT_TAIL(devstat_head,
165 printf("devstat_add_entry: HELP! "
166 "sorting problem detected "
167 "for name %p unit %d\n",
168 dev_name, unit_number);
175 ds->device_number = devstat_current_devnumber++;
176 ds->unit_number = unit_number;
177 strlcpy(ds->device_name, dev_name, DEVSTAT_NAME_LEN);
178 ds->block_size = block_size;
180 ds->device_type = device_type;
181 ds->priority = priority;
182 binuptime(&ds->creation_time);
183 devstat_generation++;
187 * Remove a devstat structure from the list of devices.
190 devstat_remove_entry(struct devstat *ds)
192 struct devstatlist *devstat_head;
194 mtx_assert(&devstat_mutex, MA_NOTOWNED);
198 mtx_lock(&devstat_mutex);
200 devstat_head = &device_statq;
202 /* Remove this entry from the devstat queue */
203 atomic_add_acq_int(&ds->sequence1, 1);
204 if (ds->unit_number != -1) {
206 STAILQ_REMOVE(devstat_head, ds, devstat, dev_links);
209 devstat_generation++;
210 mtx_unlock(&devstat_mutex);
214 * Record a transaction start.
216 * See comments for devstat_end_transaction(). Ordering is very important
220 devstat_start_transaction(struct devstat *ds, const struct bintime *now)
227 atomic_add_acq_int(&ds->sequence1, 1);
229 * We only want to set the start time when we are going from idle
230 * to busy. The start time is really the start of the latest busy
233 if (atomic_fetchadd_int(&ds->start_count, 1) == ds->end_count) {
235 ds->busy_from = *now;
237 binuptime(&ds->busy_from);
239 atomic_add_rel_int(&ds->sequence0, 1);
243 devstat_start_transaction_bio(struct devstat *ds, struct bio *bp)
250 binuptime(&bp->bio_t0);
251 devstat_start_transaction_bio_t0(ds, bp);
255 devstat_start_transaction_bio_t0(struct devstat *ds, struct bio *bp)
262 devstat_start_transaction(ds, &bp->bio_t0);
263 DTRACE_DEVSTAT_BIO_START();
267 * Record the ending of a transaction, and incrment the various counters.
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()
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.
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.
289 * In addition to this, being a kind of locks, they must be updated with
290 * atomic instructions using appropriate memory barriers.
293 devstat_end_transaction(struct devstat *ds, uint32_t bytes,
294 devstat_tag_type tag_type, devstat_trans_flags flags,
295 const struct bintime *now, const struct bintime *then)
297 struct bintime dt, lnow;
308 atomic_add_acq_int(&ds->sequence1, 1);
309 /* Update byte and operations counts */
310 ds->bytes[flags] += bytes;
311 ds->operations[flags]++;
314 * Keep a count of the various tag types sent.
316 if ((ds->flags & DEVSTAT_NO_ORDERED_TAGS) == 0 &&
317 tag_type != DEVSTAT_TAG_NONE)
318 ds->tag_types[tag_type]++;
321 /* Update duration of operations */
323 bintime_sub(&dt, then);
324 bintime_add(&ds->duration[flags], &dt);
327 /* Accumulate busy time */
329 bintime_sub(&dt, &ds->busy_from);
330 bintime_add(&ds->busy_time, &dt);
331 ds->busy_from = *now;
334 atomic_add_rel_int(&ds->sequence0, 1);
338 devstat_end_transaction_bio(struct devstat *ds, const struct bio *bp)
341 devstat_end_transaction_bio_bt(ds, bp, NULL);
345 devstat_end_transaction_bio_bt(struct devstat *ds, const struct bio *bp,
346 const struct bintime *now)
348 devstat_trans_flags flg;
349 devstat_tag_type tag;
355 if (bp->bio_flags & BIO_ORDERED)
356 tag = DEVSTAT_TAG_ORDERED;
358 tag = DEVSTAT_TAG_SIMPLE;
359 if (bp->bio_cmd == BIO_DELETE)
361 else if ((bp->bio_cmd == BIO_READ)
362 || ((bp->bio_cmd == BIO_ZONE)
363 && (bp->bio_zone.zone_cmd == DISK_ZONE_REPORT_ZONES)))
365 else if (bp->bio_cmd == BIO_WRITE)
368 flg = DEVSTAT_NO_DATA;
370 devstat_end_transaction(ds, bp->bio_bcount - bp->bio_resid,
371 tag, flg, now, &bp->bio_t0);
372 DTRACE_DEVSTAT_BIO_DONE();
376 * This is the sysctl handler for the devstat package. The data pushed out
377 * on the kern.devstat.all sysctl variable consists of the current devstat
378 * generation number, and then an array of devstat structures, one for each
379 * device in the system.
381 * This is more cryptic that obvious, but basically we neither can nor
382 * want to hold the devstat_mutex for any amount of time, so we grab it
383 * only when we need to and keep an eye on devstat_generation all the time.
386 sysctl_devstat(SYSCTL_HANDLER_ARGS)
392 mtx_assert(&devstat_mutex, MA_NOTOWNED);
395 * XXX devstat_generation should really be "volatile" but that
396 * XXX freaks out the sysctl macro below. The places where we
397 * XXX change it and inspect it are bracketed in the mutex which
398 * XXX guarantees us proper write barriers. I don't believe the
399 * XXX compiler is allowed to optimize mygen away across calls
400 * XXX to other functions, so the following is belived to be safe.
402 mygen = devstat_generation;
404 error = SYSCTL_OUT(req, &mygen, sizeof(mygen));
406 if (devstat_num_devs == 0)
412 mtx_lock(&devstat_mutex);
413 nds = STAILQ_FIRST(&device_statq);
414 if (mygen != devstat_generation)
416 mtx_unlock(&devstat_mutex);
421 for (;nds != NULL;) {
422 error = SYSCTL_OUT(req, nds, sizeof(struct devstat));
425 mtx_lock(&devstat_mutex);
426 if (mygen != devstat_generation)
429 nds = STAILQ_NEXT(nds, dev_links);
430 mtx_unlock(&devstat_mutex);
438 * Sysctl entries for devstat. The first one is a node that all the rest
441 static SYSCTL_NODE(_kern, OID_AUTO, devstat, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
442 "Device Statistics");
444 SYSCTL_PROC(_kern_devstat, OID_AUTO, all,
445 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, NULL, 0,
446 sysctl_devstat, "S,devstat",
447 "All devices in the devstat list");
449 * Export the number of devices in the system so that userland utilities
450 * can determine how much memory to allocate to hold all the devices.
452 SYSCTL_INT(_kern_devstat, OID_AUTO, numdevs, CTLFLAG_RD,
453 &devstat_num_devs, 0, "Number of devices in the devstat list");
454 SYSCTL_LONG(_kern_devstat, OID_AUTO, generation, CTLFLAG_RD,
455 &devstat_generation, 0, "Devstat list generation");
456 SYSCTL_INT(_kern_devstat, OID_AUTO, version, CTLFLAG_RD,
457 &devstat_version, 0, "Devstat list version number");
460 * Allocator for struct devstat structures. We sub-allocate these from pages
461 * which we get from malloc. These pages are exported for mmap(2)'ing through
462 * a miniature device driver
465 #define statsperpage (PAGE_SIZE / sizeof(struct devstat))
467 static d_ioctl_t devstat_ioctl;
468 static d_mmap_t devstat_mmap;
470 static struct cdevsw devstat_cdevsw = {
471 .d_version = D_VERSION,
472 .d_ioctl = devstat_ioctl,
473 .d_mmap = devstat_mmap,
478 TAILQ_ENTRY(statspage) list;
479 struct devstat *stat;
483 static size_t pagelist_pages = 0;
484 static TAILQ_HEAD(, statspage) pagelist = TAILQ_HEAD_INITIALIZER(pagelist);
485 static MALLOC_DEFINE(M_DEVSTAT, "devstat", "Device statistics");
488 devstat_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag,
496 *(off_t *)data = pagelist_pages * PAGE_SIZE;
504 devstat_mmap(struct cdev *dev, vm_ooffset_t offset, vm_paddr_t *paddr,
505 int nprot, vm_memattr_t *memattr)
507 struct statspage *spp;
509 if (nprot != VM_PROT_READ)
511 mtx_lock(&devstat_mutex);
512 TAILQ_FOREACH(spp, &pagelist, list) {
514 *paddr = vtophys(spp->stat);
515 mtx_unlock(&devstat_mutex);
520 mtx_unlock(&devstat_mutex);
524 static struct devstat *
528 struct statspage *spp, *spp2;
532 mtx_assert(&devstat_mutex, MA_NOTOWNED);
534 make_dev_credf(MAKEDEV_ETERNAL | MAKEDEV_CHECKNAME,
535 &devstat_cdevsw, 0, NULL, UID_ROOT, GID_WHEEL, 0444,
536 DEVSTAT_DEVICE_NAME);
540 mtx_lock(&devstat_mutex);
542 TAILQ_FOREACH(spp, &pagelist, list) {
548 mtx_unlock(&devstat_mutex);
549 spp2 = malloc(sizeof *spp, M_DEVSTAT, M_ZERO | M_WAITOK);
550 spp2->stat = malloc(PAGE_SIZE, M_DEVSTAT, M_ZERO | M_WAITOK);
551 spp2->nfree = statsperpage;
554 * If free statspages were added while the lock was released
557 mtx_lock(&devstat_mutex);
558 TAILQ_FOREACH(spp, &pagelist, list)
565 * It would make more sense to add the new page at the
566 * head but the order on the list determine the
567 * sequence of the mapping so we can't do that.
570 TAILQ_INSERT_TAIL(&pagelist, spp, list);
575 for (u = 0; u < statsperpage; u++) {
576 if (dsp->allocated == 0)
582 mtx_unlock(&devstat_mutex);
583 if (spp2 != NULL && spp2 != spp) {
584 free(spp2->stat, M_DEVSTAT);
585 free(spp2, M_DEVSTAT);
591 devstat_free(struct devstat *dsp)
593 struct statspage *spp;
595 mtx_assert(&devstat_mutex, MA_OWNED);
596 bzero(dsp, sizeof *dsp);
597 TAILQ_FOREACH(spp, &pagelist, list) {
598 if (dsp >= spp->stat && dsp < (spp->stat + statsperpage)) {
605 SYSCTL_INT(_debug_sizeof, OID_AUTO, devstat, CTLFLAG_RD,
606 SYSCTL_NULL_INT_PTR, sizeof(struct devstat), "sizeof(struct devstat)");