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 * $FreeBSD: releng/5.1/sys/kern/subr_devstat.c 113599 2003-04-17 15:06:28Z harti $
29 */
30
31 #include <sys/param.h>
32 #include <sys/kernel.h>
33 #include <sys/systm.h>
34 #include <sys/bio.h>
35 #include <sys/devicestat.h>
36 #include <sys/sysctl.h>
37 #include <sys/malloc.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/conf.h>
41 #include <vm/vm.h>
42 #include <vm/pmap.h>
43
44 #include <machine/atomic.h>
45
46 static int devstat_num_devs;
47 static long devstat_generation;
48 static int devstat_version = DEVSTAT_VERSION;
49 static int devstat_current_devnumber;
50 static struct mtx devstat_mutex;
51
52 static struct devstatlist device_statq;
53 static struct devstat *devstat_alloc(void);
54 static void devstat_free(struct devstat *);
55 static void devstat_add_entry(struct devstat *ds, const void *dev_name,
56 int unit_number, u_int32_t block_size,
57 devstat_support_flags flags,
58 devstat_type_flags device_type,
59 devstat_priority priority);
60
61 /*
62 * Allocate a devstat and initialize it
63 */
64 struct devstat *
65 devstat_new_entry(const void *dev_name,
66 int unit_number, u_int32_t block_size,
67 devstat_support_flags flags,
68 devstat_type_flags device_type,
69 devstat_priority priority)
70 {
71 struct devstat *ds;
72 static int once;
73
74 if (!once) {
75 STAILQ_INIT(&device_statq);
76 mtx_init(&devstat_mutex, "devstat", NULL, MTX_DEF);
77 once = 1;
78 }
79 mtx_assert(&devstat_mutex, MA_NOTOWNED);
80
81 ds = devstat_alloc();
82 mtx_lock(&devstat_mutex);
83 if (unit_number == -1) {
84 ds->id = dev_name;
85 binuptime(&ds->creation_time);
86 devstat_generation++;
87 } else {
88 devstat_add_entry(ds, dev_name, unit_number, block_size,
89 flags, device_type, priority);
90 }
91 mtx_unlock(&devstat_mutex);
92 return (ds);
93 }
94
95 /*
96 * Take a malloced and zeroed devstat structure given to us, fill it in
97 * and add it to the queue of devices.
98 */
99 static void
100 devstat_add_entry(struct devstat *ds, const void *dev_name,
101 int unit_number, u_int32_t block_size,
102 devstat_support_flags flags,
103 devstat_type_flags device_type,
104 devstat_priority priority)
105 {
106 struct devstatlist *devstat_head;
107 struct devstat *ds_tmp;
108
109 mtx_assert(&devstat_mutex, MA_OWNED);
110 devstat_num_devs++;
111
112 devstat_head = &device_statq;
113
114 /*
115 * Priority sort. Each driver passes in its priority when it adds
116 * its devstat entry. Drivers are sorted first by priority, and
117 * then by probe order.
118 *
119 * For the first device, we just insert it, since the priority
120 * doesn't really matter yet. Subsequent devices are inserted into
121 * the list using the order outlined above.
122 */
123 if (devstat_num_devs == 1)
124 STAILQ_INSERT_TAIL(devstat_head, ds, dev_links);
125 else {
126 STAILQ_FOREACH(ds_tmp, devstat_head, dev_links) {
127 struct devstat *ds_next;
128
129 ds_next = STAILQ_NEXT(ds_tmp, dev_links);
130
131 /*
132 * If we find a break between higher and lower
133 * priority items, and if this item fits in the
134 * break, insert it. This also applies if the
135 * "lower priority item" is the end of the list.
136 */
137 if ((priority <= ds_tmp->priority)
138 && ((ds_next == NULL)
139 || (priority > ds_next->priority))) {
140 STAILQ_INSERT_AFTER(devstat_head, ds_tmp, ds,
141 dev_links);
142 break;
143 } else if (priority > ds_tmp->priority) {
144 /*
145 * If this is the case, we should be able
146 * to insert ourselves at the head of the
147 * list. If we can't, something is wrong.
148 */
149 if (ds_tmp == STAILQ_FIRST(devstat_head)) {
150 STAILQ_INSERT_HEAD(devstat_head,
151 ds, dev_links);
152 break;
153 } else {
154 STAILQ_INSERT_TAIL(devstat_head,
155 ds, dev_links);
156 printf("devstat_add_entry: HELP! "
157 "sorting problem detected "
158 "for name %p unit %d\n",
159 dev_name, unit_number);
160 break;
161 }
162 }
163 }
164 }
165
166 ds->device_number = devstat_current_devnumber++;
167 ds->unit_number = unit_number;
168 strlcpy(ds->device_name, dev_name, DEVSTAT_NAME_LEN);
169 ds->block_size = block_size;
170 ds->flags = flags;
171 ds->device_type = device_type;
172 ds->priority = priority;
173 binuptime(&ds->creation_time);
174 devstat_generation++;
175 }
176
177 /*
178 * Remove a devstat structure from the list of devices.
179 */
180 void
181 devstat_remove_entry(struct devstat *ds)
182 {
183 struct devstatlist *devstat_head;
184
185 mtx_assert(&devstat_mutex, MA_NOTOWNED);
186 if (ds == NULL)
187 return;
188
189 mtx_lock(&devstat_mutex);
190
191 devstat_head = &device_statq;
192
193 /* Remove this entry from the devstat queue */
194 atomic_add_acq_int(&ds->sequence1, 1);
195 if (ds->id == NULL) {
196 devstat_num_devs--;
197 STAILQ_REMOVE(devstat_head, ds, devstat, dev_links);
198 }
199 devstat_free(ds);
200 devstat_generation++;
201 mtx_unlock(&devstat_mutex);
202 }
203
204 /*
205 * Record a transaction start.
206 *
207 * See comments for devstat_end_transaction(). Ordering is very important
208 * here.
209 */
210 void
211 devstat_start_transaction(struct devstat *ds, struct bintime *now)
212 {
213
214 mtx_assert(&devstat_mutex, MA_NOTOWNED);
215
216 /* sanity check */
217 if (ds == NULL)
218 return;
219
220 atomic_add_acq_int(&ds->sequence1, 1);
221 /*
222 * We only want to set the start time when we are going from idle
223 * to busy. The start time is really the start of the latest busy
224 * period.
225 */
226 if (ds->start_count == ds->end_count) {
227 if (now != NULL)
228 ds->busy_from = *now;
229 else
230 binuptime(&ds->busy_from);
231 }
232 ds->start_count++;
233 atomic_add_rel_int(&ds->sequence0, 1);
234 }
235
236 void
237 devstat_start_transaction_bio(struct devstat *ds, struct bio *bp)
238 {
239
240 mtx_assert(&devstat_mutex, MA_NOTOWNED);
241
242 /* sanity check */
243 if (ds == NULL)
244 return;
245
246 binuptime(&bp->bio_t0);
247 devstat_start_transaction(ds, &bp->bio_t0);
248 }
249
250 /*
251 * Record the ending of a transaction, and incrment the various counters.
252 *
253 * Ordering in this function, and in devstat_start_transaction() is VERY
254 * important. The idea here is to run without locks, so we are very
255 * careful to only modify some fields on the way "down" (i.e. at
256 * transaction start) and some fields on the way "up" (i.e. at transaction
257 * completion). One exception is busy_from, which we only modify in
258 * devstat_start_transaction() when there are no outstanding transactions,
259 * and thus it can't be modified in devstat_end_transaction()
260 * simultaneously.
261 *
262 * The sequence0 and sequence1 fields are provided to enable an application
263 * spying on the structures with mmap(2) to tell when a structure is in a
264 * consistent state or not.
265 *
266 * For this to work 100% reliably, it is important that the two fields
267 * are at opposite ends of the structure and that they are incremented
268 * in the opposite order of how a memcpy(3) in userland would copy them.
269 * We assume that the copying happens front to back, but there is actually
270 * no way short of writing your own memcpy(3) replacement to guarantee
271 * this will be the case.
272 *
273 * In addition to this, being a kind of locks, they must be updated with
274 * atomic instructions using appropriate memory barriers.
275 */
276 void
277 devstat_end_transaction(struct devstat *ds, u_int32_t bytes,
278 devstat_tag_type tag_type, devstat_trans_flags flags,
279 struct bintime *now, struct bintime *then)
280 {
281 struct bintime dt, lnow;
282
283 mtx_assert(&devstat_mutex, MA_NOTOWNED);
284
285 /* sanity check */
286 if (ds == NULL)
287 return;
288
289 if (now == NULL) {
290 now = &lnow;
291 binuptime(now);
292 }
293
294 atomic_add_acq_int(&ds->sequence1, 1);
295 /* Update byte and operations counts */
296 ds->bytes[flags] += bytes;
297 ds->operations[flags]++;
298
299 /*
300 * Keep a count of the various tag types sent.
301 */
302 if ((ds->flags & DEVSTAT_NO_ORDERED_TAGS) == 0 &&
303 tag_type != DEVSTAT_TAG_NONE)
304 ds->tag_types[tag_type]++;
305
306 if (then != NULL) {
307 /* Update duration of operations */
308 dt = *now;
309 bintime_sub(&dt, then);
310 bintime_add(&ds->duration[flags], &dt);
311 }
312
313 /* Accumulate busy time */
314 dt = *now;
315 bintime_sub(&dt, &ds->busy_from);
316 bintime_add(&ds->busy_time, &dt);
317 ds->busy_from = *now;
318
319 ds->end_count++;
320 atomic_add_rel_int(&ds->sequence0, 1);
321 }
322
323 void
324 devstat_end_transaction_bio(struct devstat *ds, struct bio *bp)
325 {
326 devstat_trans_flags flg;
327
328 mtx_assert(&devstat_mutex, MA_NOTOWNED);
329
330 /* sanity check */
331 if (ds == NULL)
332 return;
333
334 if (bp->bio_cmd == BIO_DELETE)
335 flg = DEVSTAT_FREE;
336 else if (bp->bio_cmd == BIO_READ)
337 flg = DEVSTAT_READ;
338 else if (bp->bio_cmd == BIO_WRITE)
339 flg = DEVSTAT_WRITE;
340 else
341 flg = DEVSTAT_NO_DATA;
342
343 devstat_end_transaction(ds, bp->bio_bcount - bp->bio_resid,
344 DEVSTAT_TAG_SIMPLE, flg, NULL, &bp->bio_t0);
345 }
346
347 /*
348 * This is the sysctl handler for the devstat package. The data pushed out
349 * on the kern.devstat.all sysctl variable consists of the current devstat
350 * generation number, and then an array of devstat structures, one for each
351 * device in the system.
352 *
353 * This is more cryptic that obvious, but basically we neither can nor
354 * want to hold the devstat_mutex for any amount of time, so we grab it
355 * only when we need to and keep an eye on devstat_generation all the time.
356 */
357 static int
358 sysctl_devstat(SYSCTL_HANDLER_ARGS)
359 {
360 int error;
361 long mygen;
362 struct devstat *nds;
363
364 mtx_assert(&devstat_mutex, MA_NOTOWNED);
365
366 if (devstat_num_devs == 0)
367 return(EINVAL);
368
369 /*
370 * XXX devstat_generation should really be "volatile" but that
371 * XXX freaks out the sysctl macro below. The places where we
372 * XXX change it and inspect it are bracketed in the mutex which
373 * XXX guarantees us proper write barriers. I don't belive the
374 * XXX compiler is allowed to optimize mygen away across calls
375 * XXX to other functions, so the following is belived to be safe.
376 */
377 mygen = devstat_generation;
378
379 error = SYSCTL_OUT(req, &mygen, sizeof(mygen));
380
381 if (error != 0)
382 return (error);
383
384 mtx_lock(&devstat_mutex);
385 nds = STAILQ_FIRST(&device_statq);
386 if (mygen != devstat_generation)
387 error = EBUSY;
388 mtx_unlock(&devstat_mutex);
389
390 if (error != 0)
391 return (error);
392
393 for (;nds != NULL;) {
394 error = SYSCTL_OUT(req, nds, sizeof(struct devstat));
395 if (error != 0)
396 return (error);
397 mtx_lock(&devstat_mutex);
398 if (mygen != devstat_generation)
399 error = EBUSY;
400 else
401 nds = STAILQ_NEXT(nds, dev_links);
402 mtx_unlock(&devstat_mutex);
403 if (error != 0)
404 return (error);
405 }
406 return(error);
407 }
408
409 /*
410 * Sysctl entries for devstat. The first one is a node that all the rest
411 * hang off of.
412 */
413 SYSCTL_NODE(_kern, OID_AUTO, devstat, CTLFLAG_RD, 0, "Device Statistics");
414
415 SYSCTL_PROC(_kern_devstat, OID_AUTO, all, CTLFLAG_RD|CTLTYPE_OPAQUE,
416 0, 0, sysctl_devstat, "S,devstat", "All devices in the devstat list");
417 /*
418 * Export the number of devices in the system so that userland utilities
419 * can determine how much memory to allocate to hold all the devices.
420 */
421 SYSCTL_INT(_kern_devstat, OID_AUTO, numdevs, CTLFLAG_RD,
422 &devstat_num_devs, 0, "Number of devices in the devstat list");
423 SYSCTL_LONG(_kern_devstat, OID_AUTO, generation, CTLFLAG_RD,
424 &devstat_generation, 0, "Devstat list generation");
425 SYSCTL_INT(_kern_devstat, OID_AUTO, version, CTLFLAG_RD,
426 &devstat_version, 0, "Devstat list version number");
427
428 /*
429 * Allocator for struct devstat structures. We sub-allocate these from pages
430 * which we get from malloc. These pages are exported for mmap(2)'ing through
431 * a miniature device driver
432 */
433
434 #define statsperpage (PAGE_SIZE / sizeof(struct devstat))
435
436 static d_mmap_t devstat_mmap;
437
438 static struct cdevsw devstat_cdevsw = {
439 .d_open = nullopen,
440 .d_close = nullclose,
441 .d_mmap = devstat_mmap,
442 .d_name = "devstat",
443 };
444
445 struct statspage {
446 TAILQ_ENTRY(statspage) list;
447 struct devstat *stat;
448 u_int nfree;
449 };
450
451 static TAILQ_HEAD(, statspage) pagelist = TAILQ_HEAD_INITIALIZER(pagelist);
452 static MALLOC_DEFINE(M_DEVSTAT, "devstat", "Device statistics");
453
454 static int
455 devstat_mmap(dev_t dev, vm_offset_t offset, vm_paddr_t *paddr, int nprot)
456 {
457 struct statspage *spp;
458
459 if (nprot != VM_PROT_READ)
460 return (-1);
461 TAILQ_FOREACH(spp, &pagelist, list) {
462 if (offset == 0) {
463 *paddr = vtophys(spp->stat);
464 return (0);
465 }
466 offset -= PAGE_SIZE;
467 }
468 return (-1);
469 }
470
471 static struct devstat *
472 devstat_alloc(void)
473 {
474 struct devstat *dsp;
475 struct statspage *spp;
476 u_int u;
477 static int once;
478
479 mtx_assert(&devstat_mutex, MA_NOTOWNED);
480 if (!once) {
481 make_dev(&devstat_cdevsw, 0,
482 UID_ROOT, GID_WHEEL, 0400, DEVSTAT_DEVICE_NAME);
483 once = 1;
484 }
485 mtx_lock(&devstat_mutex);
486 for (;;) {
487 TAILQ_FOREACH(spp, &pagelist, list) {
488 if (spp->nfree > 0)
489 break;
490 }
491 if (spp != NULL)
492 break;
493 /*
494 * We had no free slot in any of our pages, drop the mutex
495 * and get another page. In theory we could have more than
496 * one process doing this at the same time and consequently
497 * we may allocate more pages than we will need. That is
498 * Just Too Bad[tm], we can live with that.
499 */
500 mtx_unlock(&devstat_mutex);
501 spp = malloc(sizeof *spp, M_DEVSTAT, M_ZERO | M_WAITOK);
502 spp->stat = malloc(PAGE_SIZE, M_DEVSTAT, M_ZERO | M_WAITOK);
503 spp->nfree = statsperpage;
504 mtx_lock(&devstat_mutex);
505 /*
506 * It would make more sense to add the new page at the head
507 * but the order on the list determine the sequence of the
508 * mapping so we can't do that.
509 */
510 TAILQ_INSERT_TAIL(&pagelist, spp, list);
511 }
512 dsp = spp->stat;
513 for (u = 0; u < statsperpage; u++) {
514 if (dsp->allocated == 0)
515 break;
516 dsp++;
517 }
518 spp->nfree--;
519 dsp->allocated = 1;
520 mtx_unlock(&devstat_mutex);
521 return (dsp);
522 }
523
524 static void
525 devstat_free(struct devstat *dsp)
526 {
527 struct statspage *spp;
528
529 mtx_assert(&devstat_mutex, MA_OWNED);
530 bzero(dsp, sizeof *dsp);
531 TAILQ_FOREACH(spp, &pagelist, list) {
532 if (dsp >= spp->stat && dsp < (spp->stat + statsperpage)) {
533 spp->nfree++;
534 return;
535 }
536 }
537 }
538
539 SYSCTL_INT(_debug_sizeof, OID_AUTO, devstat, CTLFLAG_RD,
540 0, sizeof(struct devstat), "sizeof(struct devstat)");
Cache object: a6382586f98da64c29585a3bb9c51771
|