2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2005-2006 Robert N. M. Watson
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.
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
31 #include <sys/param.h>
32 #include <sys/cpuset.h>
33 #include <sys/sysctl.h>
36 #include <vm/uma_int.h>
49 #include "memstat_internal.h"
51 static struct nlist namelist[] = {
53 { .n_name = "_uma_kegs" },
55 { .n_name = "_mp_maxid" },
57 { .n_name = "_all_cpus" },
62 * Extract uma(9) statistics from the running kernel, and store all memory
63 * type information in the passed list. For each type, check the list for an
64 * existing entry with the right name/allocator -- if present, update that
65 * entry. Otherwise, add a new entry. On error, the entire list will be
66 * cleared, as entries will be in an inconsistent state.
68 * To reduce the level of work for a list that starts empty, we keep around a
69 * hint as to whether it was empty when we began, so we can avoid searching
70 * the list for entries to update. Updates are O(n^2) due to searching for
71 * each entry before adding it.
74 memstat_sysctl_uma(struct memory_type_list *list, int flags)
76 struct uma_stream_header *ushp;
77 struct uma_type_header *uthp;
78 struct uma_percpu_stat *upsp;
79 struct memory_type *mtp;
80 int count, hint_dontsearch, i, j, maxcpus, maxid;
84 hint_dontsearch = LIST_EMPTY(&list->mtl_list);
87 * Query the number of CPUs, number of malloc types so that we can
88 * guess an initial buffer size. We loop until we succeed or really
89 * fail. Note that the value of maxcpus we query using sysctl is not
90 * the version we use when processing the real data -- that is read
95 if (sysctlbyname("kern.smp.maxid", &maxid, &size, NULL, 0) < 0) {
96 if (errno == EACCES || errno == EPERM)
97 list->mtl_error = MEMSTAT_ERROR_PERMISSION;
99 list->mtl_error = MEMSTAT_ERROR_DATAERROR;
102 if (size != sizeof(maxid)) {
103 list->mtl_error = MEMSTAT_ERROR_DATAERROR;
107 size = sizeof(count);
108 if (sysctlbyname("vm.zone_count", &count, &size, NULL, 0) < 0) {
109 if (errno == EACCES || errno == EPERM)
110 list->mtl_error = MEMSTAT_ERROR_PERMISSION;
112 list->mtl_error = MEMSTAT_ERROR_VERSION;
115 if (size != sizeof(count)) {
116 list->mtl_error = MEMSTAT_ERROR_DATAERROR;
120 size = sizeof(*uthp) + count * (sizeof(*uthp) + sizeof(*upsp) *
123 buffer = malloc(size);
124 if (buffer == NULL) {
125 list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
129 if (sysctlbyname("vm.zone_stats", buffer, &size, NULL, 0) < 0) {
131 * XXXRW: ENOMEM is an ambiguous return, we should bound the
132 * number of loops, perhaps.
134 if (errno == ENOMEM) {
138 if (errno == EACCES || errno == EPERM)
139 list->mtl_error = MEMSTAT_ERROR_PERMISSION;
141 list->mtl_error = MEMSTAT_ERROR_VERSION;
151 if (size < sizeof(*ushp)) {
152 list->mtl_error = MEMSTAT_ERROR_VERSION;
157 ushp = (struct uma_stream_header *)p;
160 if (ushp->ush_version != UMA_STREAM_VERSION) {
161 list->mtl_error = MEMSTAT_ERROR_VERSION;
167 * For the remainder of this function, we are quite trusting about
168 * the layout of structures and sizes, since we've determined we have
169 * a matching version and acceptable CPU count.
171 maxcpus = ushp->ush_maxcpus;
172 count = ushp->ush_count;
173 for (i = 0; i < count; i++) {
174 uthp = (struct uma_type_header *)p;
177 if (hint_dontsearch == 0) {
178 mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
183 mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
184 uthp->uth_name, maxid + 1);
186 _memstat_mtl_empty(list);
188 list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
193 * Reset the statistics on a current node.
195 _memstat_mt_reset_stats(mtp, maxid + 1);
197 mtp->mt_numallocs = uthp->uth_allocs;
198 mtp->mt_numfrees = uthp->uth_frees;
199 mtp->mt_failures = uthp->uth_fails;
200 mtp->mt_sleeps = uthp->uth_sleeps;
202 for (j = 0; j < maxcpus; j++) {
203 upsp = (struct uma_percpu_stat *)p;
206 mtp->mt_percpu_cache[j].mtp_free =
207 upsp->ups_cache_free;
208 mtp->mt_free += upsp->ups_cache_free;
209 mtp->mt_numallocs += upsp->ups_allocs;
210 mtp->mt_numfrees += upsp->ups_frees;
213 mtp->mt_size = uthp->uth_size;
214 mtp->mt_rsize = uthp->uth_rsize;
215 mtp->mt_memalloced = mtp->mt_numallocs * uthp->uth_size;
216 mtp->mt_memfreed = mtp->mt_numfrees * uthp->uth_size;
217 mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
218 mtp->mt_countlimit = uthp->uth_limit;
219 mtp->mt_byteslimit = uthp->uth_limit * uthp->uth_size;
221 mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
222 mtp->mt_zonefree = uthp->uth_zone_free;
225 * UMA secondary zones share a keg with the primary zone. To
226 * avoid double-reporting of free items, report keg free
227 * items only in the primary zone.
229 if (!(uthp->uth_zone_flags & UTH_ZONE_SECONDARY)) {
230 mtp->mt_kegfree = uthp->uth_keg_free;
231 mtp->mt_free += mtp->mt_kegfree;
233 mtp->mt_free += mtp->mt_zonefree;
242 kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size,
247 ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address,
250 return (MEMSTAT_ERROR_KVM);
251 if ((size_t)ret != size)
252 return (MEMSTAT_ERROR_KVM_SHORTREAD);
257 kread_string(kvm_t *kvm, const void *kvm_pointer, char *buffer, int buflen)
262 for (i = 0; i < buflen; i++) {
263 ret = kvm_read(kvm, (unsigned long)kvm_pointer + i,
264 &(buffer[i]), sizeof(char));
266 return (MEMSTAT_ERROR_KVM);
267 if ((size_t)ret != sizeof(char))
268 return (MEMSTAT_ERROR_KVM_SHORTREAD);
269 if (buffer[i] == '\0')
278 kread_symbol(kvm_t *kvm, int index, void *address, size_t size,
283 ret = kvm_read(kvm, namelist[index].n_value + offset, address, size);
285 return (MEMSTAT_ERROR_KVM);
286 if ((size_t)ret != size)
287 return (MEMSTAT_ERROR_KVM_SHORTREAD);
292 * memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts
293 * UMA(9) statistics from a kernel core/memory file.
296 memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
298 LIST_HEAD(, uma_keg) uma_kegs;
299 struct memory_type *mtp;
300 struct uma_bucket *ubp, ub;
301 struct uma_cache *ucp, *ucp_array;
302 struct uma_zone *uzp, uz;
303 struct uma_keg *kzp, kz;
304 int hint_dontsearch, i, mp_maxid, ret;
305 char name[MEMTYPE_MAXNAME];
310 kvm = (kvm_t *)kvm_handle;
311 hint_dontsearch = LIST_EMPTY(&list->mtl_list);
312 if (kvm_nlist(kvm, namelist) != 0) {
313 list->mtl_error = MEMSTAT_ERROR_KVM;
316 if (namelist[X_UMA_KEGS].n_type == 0 ||
317 namelist[X_UMA_KEGS].n_value == 0) {
318 list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
321 ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
323 list->mtl_error = ret;
326 ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
328 list->mtl_error = ret;
331 cpusetsize = sysconf(_SC_CPUSET_SIZE);
332 if (cpusetsize == -1 || (u_long)cpusetsize > sizeof(cpuset_t)) {
333 list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
337 ret = kread_symbol(kvm, X_ALL_CPUS, &all_cpus, cpusetsize, 0);
339 list->mtl_error = ret;
342 ucp_array = malloc(sizeof(struct uma_cache) * (mp_maxid + 1));
343 if (ucp_array == NULL) {
344 list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
347 for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
348 LIST_NEXT(&kz, uk_link)) {
349 ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
352 _memstat_mtl_empty(list);
353 list->mtl_error = ret;
356 for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
357 LIST_NEXT(&uz, uz_link)) {
358 ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
361 _memstat_mtl_empty(list);
362 list->mtl_error = ret;
365 ret = kread(kvm, uzp, ucp_array,
366 sizeof(struct uma_cache) * (mp_maxid + 1),
367 offsetof(struct uma_zone, uz_cpu[0]));
370 _memstat_mtl_empty(list);
371 list->mtl_error = ret;
374 ret = kread_string(kvm, uz.uz_name, name,
378 _memstat_mtl_empty(list);
379 list->mtl_error = ret;
382 if (hint_dontsearch == 0) {
383 mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
388 mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
392 _memstat_mtl_empty(list);
393 list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
397 * Reset the statistics on a current node.
399 _memstat_mt_reset_stats(mtp, mp_maxid + 1);
400 mtp->mt_numallocs = uz.uz_allocs;
401 mtp->mt_numfrees = uz.uz_frees;
402 mtp->mt_failures = uz.uz_fails;
403 mtp->mt_sleeps = uz.uz_sleeps;
404 if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
406 for (i = 0; i < mp_maxid + 1; i++) {
407 if (!CPU_ISSET(i, &all_cpus))
410 mtp->mt_numallocs += ucp->uc_allocs;
411 mtp->mt_numfrees += ucp->uc_frees;
413 if (ucp->uc_allocbucket != NULL) {
414 ret = kread(kvm, ucp->uc_allocbucket,
418 _memstat_mtl_empty(list);
419 list->mtl_error = ret;
422 mtp->mt_free += ub.ub_cnt;
424 if (ucp->uc_freebucket != NULL) {
425 ret = kread(kvm, ucp->uc_freebucket,
429 _memstat_mtl_empty(list);
430 list->mtl_error = ret;
433 mtp->mt_free += ub.ub_cnt;
437 mtp->mt_size = kz.uk_size;
438 mtp->mt_rsize = kz.uk_rsize;
439 mtp->mt_memalloced = mtp->mt_numallocs * mtp->mt_size;
440 mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
441 mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
443 mtp->mt_countlimit = kz.uk_maxpages /
446 mtp->mt_countlimit = kz.uk_maxpages *
448 mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
449 mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
450 for (ubp = LIST_FIRST(&uz.uz_buckets); ubp !=
451 NULL; ubp = LIST_NEXT(&ub, ub_link)) {
452 ret = kread(kvm, ubp, &ub, sizeof(ub), 0);
453 mtp->mt_zonefree += ub.ub_cnt;
455 if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
456 LIST_FIRST(&kz.uk_zones) != uzp)) {
457 mtp->mt_kegfree = kz.uk_free;
458 mtp->mt_free += mtp->mt_kegfree;
460 mtp->mt_free += mtp->mt_zonefree;