2 * Copyright (c) 2005-2006 Robert N. M. Watson
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/param.h>
30 #include <sys/sysctl.h>
33 #include <vm/vm_page.h>
36 #include <vm/uma_int.h>
48 #include "memstat_internal.h"
50 static struct nlist namelist[] = {
52 { .n_name = "_uma_kegs" },
54 { .n_name = "_mp_maxid" },
56 { .n_name = "_all_cpus" },
61 * Extract uma(9) statistics from the running kernel, and store all memory
62 * type information in the passed list. For each type, check the list for an
63 * existing entry with the right name/allocator -- if present, update that
64 * entry. Otherwise, add a new entry. On error, the entire list will be
65 * cleared, as entries will be in an inconsistent state.
67 * To reduce the level of work for a list that starts empty, we keep around a
68 * hint as to whether it was empty when we began, so we can avoid searching
69 * the list for entries to update. Updates are O(n^2) due to searching for
70 * each entry before adding it.
73 memstat_sysctl_uma(struct memory_type_list *list, int flags)
75 struct uma_stream_header *ushp;
76 struct uma_type_header *uthp;
77 struct uma_percpu_stat *upsp;
78 struct memory_type *mtp;
79 int count, hint_dontsearch, i, j, maxcpus;
83 hint_dontsearch = LIST_EMPTY(&list->mtl_list);
86 * Query the number of CPUs, number of malloc types so that we can
87 * guess an initial buffer size. We loop until we succeed or really
88 * fail. Note that the value of maxcpus we query using sysctl is not
89 * the version we use when processing the real data -- that is read
93 size = sizeof(maxcpus);
94 if (sysctlbyname("kern.smp.maxcpus", &maxcpus, &size, NULL, 0) < 0) {
95 if (errno == EACCES || errno == EPERM)
96 list->mtl_error = MEMSTAT_ERROR_PERMISSION;
98 list->mtl_error = MEMSTAT_ERROR_DATAERROR;
101 if (size != sizeof(maxcpus)) {
102 list->mtl_error = MEMSTAT_ERROR_DATAERROR;
106 if (maxcpus > MEMSTAT_MAXCPU) {
107 list->mtl_error = MEMSTAT_ERROR_TOOMANYCPUS;
111 size = sizeof(count);
112 if (sysctlbyname("vm.zone_count", &count, &size, NULL, 0) < 0) {
113 if (errno == EACCES || errno == EPERM)
114 list->mtl_error = MEMSTAT_ERROR_PERMISSION;
116 list->mtl_error = MEMSTAT_ERROR_VERSION;
119 if (size != sizeof(count)) {
120 list->mtl_error = MEMSTAT_ERROR_DATAERROR;
124 size = sizeof(*uthp) + count * (sizeof(*uthp) + sizeof(*upsp) *
127 buffer = malloc(size);
128 if (buffer == NULL) {
129 list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
133 if (sysctlbyname("vm.zone_stats", buffer, &size, NULL, 0) < 0) {
135 * XXXRW: ENOMEM is an ambiguous return, we should bound the
136 * number of loops, perhaps.
138 if (errno == ENOMEM) {
142 if (errno == EACCES || errno == EPERM)
143 list->mtl_error = MEMSTAT_ERROR_PERMISSION;
145 list->mtl_error = MEMSTAT_ERROR_VERSION;
155 if (size < sizeof(*ushp)) {
156 list->mtl_error = MEMSTAT_ERROR_VERSION;
161 ushp = (struct uma_stream_header *)p;
164 if (ushp->ush_version != UMA_STREAM_VERSION) {
165 list->mtl_error = MEMSTAT_ERROR_VERSION;
170 if (ushp->ush_maxcpus > MEMSTAT_MAXCPU) {
171 list->mtl_error = MEMSTAT_ERROR_TOOMANYCPUS;
177 * For the remainder of this function, we are quite trusting about
178 * the layout of structures and sizes, since we've determined we have
179 * a matching version and acceptable CPU count.
181 maxcpus = ushp->ush_maxcpus;
182 count = ushp->ush_count;
183 for (i = 0; i < count; i++) {
184 uthp = (struct uma_type_header *)p;
187 if (hint_dontsearch == 0) {
188 mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
193 mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
196 _memstat_mtl_empty(list);
198 list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
203 * Reset the statistics on a current node.
205 _memstat_mt_reset_stats(mtp);
207 mtp->mt_numallocs = uthp->uth_allocs;
208 mtp->mt_numfrees = uthp->uth_frees;
209 mtp->mt_failures = uthp->uth_fails;
211 for (j = 0; j < maxcpus; j++) {
212 upsp = (struct uma_percpu_stat *)p;
215 mtp->mt_percpu_cache[j].mtp_free =
216 upsp->ups_cache_free;
217 mtp->mt_free += upsp->ups_cache_free;
218 mtp->mt_numallocs += upsp->ups_allocs;
219 mtp->mt_numfrees += upsp->ups_frees;
222 mtp->mt_size = uthp->uth_size;
223 mtp->mt_memalloced = mtp->mt_numallocs * uthp->uth_size;
224 mtp->mt_memfreed = mtp->mt_numfrees * uthp->uth_size;
225 mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
226 mtp->mt_countlimit = uthp->uth_limit;
227 mtp->mt_byteslimit = uthp->uth_limit * uthp->uth_size;
229 mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
230 mtp->mt_zonefree = uthp->uth_zone_free;
233 * UMA secondary zones share a keg with the primary zone. To
234 * avoid double-reporting of free items, report keg free
235 * items only in the primary zone.
237 if (!(uthp->uth_zone_flags & UTH_ZONE_SECONDARY)) {
238 mtp->mt_kegfree = uthp->uth_keg_free;
239 mtp->mt_free += mtp->mt_kegfree;
241 mtp->mt_free += mtp->mt_zonefree;
250 kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size,
255 ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address,
258 return (MEMSTAT_ERROR_KVM);
259 if ((size_t)ret != size)
260 return (MEMSTAT_ERROR_KVM_SHORTREAD);
265 kread_string(kvm_t *kvm, const void *kvm_pointer, char *buffer, int buflen)
270 for (i = 0; i < buflen; i++) {
271 ret = kvm_read(kvm, (unsigned long)kvm_pointer + i,
272 &(buffer[i]), sizeof(char));
274 return (MEMSTAT_ERROR_KVM);
275 if ((size_t)ret != sizeof(char))
276 return (MEMSTAT_ERROR_KVM_SHORTREAD);
277 if (buffer[i] == '\0')
286 kread_symbol(kvm_t *kvm, int index, void *address, size_t size,
291 ret = kvm_read(kvm, namelist[index].n_value + offset, address, size);
293 return (MEMSTAT_ERROR_KVM);
294 if ((size_t)ret != size)
295 return (MEMSTAT_ERROR_KVM_SHORTREAD);
300 * memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts
301 * UMA(9) statistics from a kernel core/memory file.
304 memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
306 LIST_HEAD(, uma_keg) uma_kegs;
307 struct memory_type *mtp;
308 struct uma_bucket *ubp, ub;
309 struct uma_cache *ucp, *ucp_array;
310 struct uma_zone *uzp, uz;
311 struct uma_keg *kzp, kz;
312 int hint_dontsearch, i, mp_maxid, ret;
313 char name[MEMTYPE_MAXNAME];
314 __cpumask_t all_cpus;
317 kvm = (kvm_t *)kvm_handle;
318 hint_dontsearch = LIST_EMPTY(&list->mtl_list);
319 if (kvm_nlist(kvm, namelist) != 0) {
320 list->mtl_error = MEMSTAT_ERROR_KVM;
323 if (namelist[X_UMA_KEGS].n_type == 0 ||
324 namelist[X_UMA_KEGS].n_value == 0) {
325 list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
328 ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
330 list->mtl_error = ret;
333 ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
335 list->mtl_error = ret;
338 ret = kread_symbol(kvm, X_ALL_CPUS, &all_cpus, sizeof(all_cpus), 0);
340 list->mtl_error = ret;
343 ucp_array = malloc(sizeof(struct uma_cache) * (mp_maxid + 1));
344 if (ucp_array == NULL) {
345 list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
348 for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
349 LIST_NEXT(&kz, uk_link)) {
350 ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
353 _memstat_mtl_empty(list);
354 list->mtl_error = ret;
357 for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
358 LIST_NEXT(&uz, uz_link)) {
359 ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
362 _memstat_mtl_empty(list);
363 list->mtl_error = ret;
366 ret = kread(kvm, uzp, ucp_array,
367 sizeof(struct uma_cache) * (mp_maxid + 1),
368 offsetof(struct uma_zone, uz_cpu[0]));
371 _memstat_mtl_empty(list);
372 list->mtl_error = ret;
375 ret = kread_string(kvm, uz.uz_name, name,
379 _memstat_mtl_empty(list);
380 list->mtl_error = ret;
383 if (hint_dontsearch == 0) {
384 mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
389 mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
393 _memstat_mtl_empty(list);
394 list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
398 * Reset the statistics on a current node.
400 _memstat_mt_reset_stats(mtp);
401 mtp->mt_numallocs = uz.uz_allocs;
402 mtp->mt_numfrees = uz.uz_frees;
403 mtp->mt_failures = uz.uz_fails;
404 if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
406 for (i = 0; i < mp_maxid + 1; i++) {
407 if ((all_cpus & (1 << i)) == 0)
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_memalloced = mtp->mt_numallocs * mtp->mt_size;
439 mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
440 mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
442 mtp->mt_countlimit = kz.uk_maxpages /
445 mtp->mt_countlimit = kz.uk_maxpages *
447 mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
448 mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
449 for (ubp = LIST_FIRST(&uz.uz_full_bucket); ubp !=
450 NULL; ubp = LIST_NEXT(&ub, ub_link)) {
451 ret = kread(kvm, ubp, &ub, sizeof(ub), 0);
452 mtp->mt_zonefree += ub.ub_cnt;
454 if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
455 LIST_FIRST(&kz.uk_zones) != uzp)) {
456 mtp->mt_kegfree = kz.uk_free;
457 mtp->mt_free += mtp->mt_kegfree;
459 mtp->mt_free += mtp->mt_zonefree;