vm_meter.c revision 228449
150472Speter/*- 233975Sjdp * Copyright (c) 1982, 1986, 1989, 1993 333975Sjdp * The Regents of the University of California. All rights reserved. 433975Sjdp * 544360Simp * Redistribution and use in source and binary forms, with or without 633975Sjdp * modification, are permitted provided that the following conditions 784902Sobrien * are met: 884902Sobrien * 1. Redistributions of source code must retain the above copyright 984902Sobrien * notice, this list of conditions and the following disclaimer. 1084902Sobrien * 2. Redistributions in binary form must reproduce the above copyright 1184902Sobrien * notice, this list of conditions and the following disclaimer in the 1284902Sobrien * documentation and/or other materials provided with the distribution. 1384902Sobrien * 4. Neither the name of the University nor the names of its contributors 1484902Sobrien * may be used to endorse or promote products derived from this software 1533975Sjdp * without specific prior written permission. 1633975Sjdp * 1734498Sjdp * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 1834498Sjdp * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1934495Sjdp * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2033975Sjdp * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 2152924Sjb * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2252924Sjb * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2352924Sjb * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2484902Sobrien * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2584902Sobrien * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2684902Sobrien * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2784902Sobrien * SUCH DAMAGE. 2852924Sjb * 2952924Sjb * @(#)vm_meter.c 8.4 (Berkeley) 1/4/94 3035711Sjb */ 3158953Ssteve 3284902Sobrien#include <sys/cdefs.h> 3358953Ssteve__FBSDID("$FreeBSD: head/sys/vm/vm_meter.c 228449 2011-12-13 00:38:50Z eadler $"); 3435711Sjb 3535711Sjb#include <sys/param.h> 3684902Sobrien#include <sys/systm.h> 3735711Sjb#include <sys/kernel.h> 3884902Sobrien#include <sys/lock.h> 3935711Sjb#include <sys/mutex.h> 4035711Sjb#include <sys/proc.h> 4158953Ssteve#include <sys/resource.h> 4284902Sobrien#include <sys/sx.h> 4335711Sjb#include <sys/vmmeter.h> 4435711Sjb#include <sys/smp.h> 4535711Sjb 4635711Sjb#include <vm/vm.h> 4784902Sobrien#include <vm/vm_page.h> 4835711Sjb#include <vm/vm_extern.h> 4984902Sobrien#include <vm/vm_param.h> 5035711Sjb#include <vm/pmap.h> 5135711Sjb#include <vm/vm_map.h> 5284902Sobrien#include <vm/vm_object.h> 5335711Sjb#include <sys/sysctl.h> 5433975Sjdp 5533975Sjdpstruct vmmeter cnt; 5633975Sjdp 5733975SjdpSYSCTL_UINT(_vm, VM_V_FREE_MIN, v_free_min, 58 CTLFLAG_RW, &cnt.v_free_min, 0, "Minimum low-free-pages threshold"); 59SYSCTL_UINT(_vm, VM_V_FREE_TARGET, v_free_target, 60 CTLFLAG_RW, &cnt.v_free_target, 0, "Desired free pages"); 61SYSCTL_UINT(_vm, VM_V_FREE_RESERVED, v_free_reserved, 62 CTLFLAG_RW, &cnt.v_free_reserved, 0, "Pages reserved for deadlock"); 63SYSCTL_UINT(_vm, VM_V_INACTIVE_TARGET, v_inactive_target, 64 CTLFLAG_RW, &cnt.v_inactive_target, 0, "Pages desired inactive"); 65SYSCTL_UINT(_vm, VM_V_CACHE_MIN, v_cache_min, 66 CTLFLAG_RW, &cnt.v_cache_min, 0, "Min pages on cache queue"); 67SYSCTL_UINT(_vm, VM_V_CACHE_MAX, v_cache_max, 68 CTLFLAG_RW, &cnt.v_cache_max, 0, "Max pages on cache queue"); 69SYSCTL_UINT(_vm, VM_V_PAGEOUT_FREE_MIN, v_pageout_free_min, 70 CTLFLAG_RW, &cnt.v_pageout_free_min, 0, "Min pages reserved for kernel"); 71SYSCTL_UINT(_vm, OID_AUTO, v_free_severe, 72 CTLFLAG_RW, &cnt.v_free_severe, 0, "Severe page depletion point"); 73 74static int 75sysctl_vm_loadavg(SYSCTL_HANDLER_ARGS) 76{ 77 78#ifdef SCTL_MASK32 79 u_int32_t la[4]; 80 81 if (req->flags & SCTL_MASK32) { 82 la[0] = averunnable.ldavg[0]; 83 la[1] = averunnable.ldavg[1]; 84 la[2] = averunnable.ldavg[2]; 85 la[3] = averunnable.fscale; 86 return SYSCTL_OUT(req, la, sizeof(la)); 87 } else 88#endif 89 return SYSCTL_OUT(req, &averunnable, sizeof(averunnable)); 90} 91SYSCTL_PROC(_vm, VM_LOADAVG, loadavg, CTLTYPE_STRUCT | CTLFLAG_RD | 92 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_loadavg, "S,loadavg", 93 "Machine loadaverage history"); 94 95static int 96vmtotal(SYSCTL_HANDLER_ARGS) 97{ 98 struct proc *p; 99 struct vmtotal total; 100 vm_map_entry_t entry; 101 vm_object_t object; 102 vm_map_t map; 103 int paging; 104 struct thread *td; 105 struct vmspace *vm; 106 107 bzero(&total, sizeof(total)); 108 /* 109 * Mark all objects as inactive. 110 */ 111 mtx_lock(&vm_object_list_mtx); 112 TAILQ_FOREACH(object, &vm_object_list, object_list) { 113 if (!VM_OBJECT_TRYLOCK(object)) { 114 /* 115 * Avoid a lock-order reversal. Consequently, 116 * the reported number of active pages may be 117 * greater than the actual number. 118 */ 119 continue; 120 } 121 vm_object_clear_flag(object, OBJ_ACTIVE); 122 VM_OBJECT_UNLOCK(object); 123 } 124 mtx_unlock(&vm_object_list_mtx); 125 /* 126 * Calculate process statistics. 127 */ 128 sx_slock(&allproc_lock); 129 FOREACH_PROC_IN_SYSTEM(p) { 130 if (p->p_flag & P_SYSTEM) 131 continue; 132 PROC_LOCK(p); 133 switch (p->p_state) { 134 case PRS_NEW: 135 PROC_UNLOCK(p); 136 continue; 137 break; 138 default: 139 FOREACH_THREAD_IN_PROC(p, td) { 140 thread_lock(td); 141 switch (td->td_state) { 142 case TDS_INHIBITED: 143 if (TD_IS_SWAPPED(td)) 144 total.t_sw++; 145 else if (TD_IS_SLEEPING(td) && 146 td->td_priority <= PZERO) 147 total.t_dw++; 148 else 149 total.t_sl++; 150 break; 151 152 case TDS_CAN_RUN: 153 total.t_sw++; 154 break; 155 case TDS_RUNQ: 156 case TDS_RUNNING: 157 total.t_rq++; 158 thread_unlock(td); 159 continue; 160 default: 161 break; 162 } 163 thread_unlock(td); 164 } 165 } 166 PROC_UNLOCK(p); 167 /* 168 * Note active objects. 169 */ 170 paging = 0; 171 vm = vmspace_acquire_ref(p); 172 if (vm == NULL) 173 continue; 174 map = &vm->vm_map; 175 vm_map_lock_read(map); 176 for (entry = map->header.next; 177 entry != &map->header; entry = entry->next) { 178 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) || 179 (object = entry->object.vm_object) == NULL) 180 continue; 181 VM_OBJECT_LOCK(object); 182 vm_object_set_flag(object, OBJ_ACTIVE); 183 paging |= object->paging_in_progress; 184 VM_OBJECT_UNLOCK(object); 185 } 186 vm_map_unlock_read(map); 187 vmspace_free(vm); 188 if (paging) 189 total.t_pw++; 190 } 191 sx_sunlock(&allproc_lock); 192 /* 193 * Calculate object memory usage statistics. 194 */ 195 mtx_lock(&vm_object_list_mtx); 196 TAILQ_FOREACH(object, &vm_object_list, object_list) { 197 /* 198 * Perform unsynchronized reads on the object to avoid 199 * a lock-order reversal. In this case, the lack of 200 * synchronization should not impair the accuracy of 201 * the reported statistics. 202 */ 203 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 204 /* 205 * Devices, like /dev/mem, will badly skew our totals. 206 */ 207 continue; 208 } 209 if (object->ref_count == 0) { 210 /* 211 * Also skip unreferenced objects, including 212 * vnodes representing mounted file systems. 213 */ 214 continue; 215 } 216 total.t_vm += object->size; 217 total.t_rm += object->resident_page_count; 218 if (object->flags & OBJ_ACTIVE) { 219 total.t_avm += object->size; 220 total.t_arm += object->resident_page_count; 221 } 222 if (object->shadow_count > 1) { 223 /* shared object */ 224 total.t_vmshr += object->size; 225 total.t_rmshr += object->resident_page_count; 226 if (object->flags & OBJ_ACTIVE) { 227 total.t_avmshr += object->size; 228 total.t_armshr += object->resident_page_count; 229 } 230 } 231 } 232 mtx_unlock(&vm_object_list_mtx); 233 total.t_free = cnt.v_free_count + cnt.v_cache_count; 234 return (sysctl_handle_opaque(oidp, &total, sizeof(total), req)); 235} 236 237/* 238 * vcnt() - accumulate statistics from all cpus and the global cnt 239 * structure. 240 * 241 * The vmmeter structure is now per-cpu as well as global. Those 242 * statistics which can be kept on a per-cpu basis (to avoid cache 243 * stalls between cpus) can be moved to the per-cpu vmmeter. Remaining 244 * statistics, such as v_free_reserved, are left in the global 245 * structure. 246 * 247 * (sysctl_oid *oidp, void *arg1, int arg2, struct sysctl_req *req) 248 */ 249static int 250vcnt(SYSCTL_HANDLER_ARGS) 251{ 252 int count = *(int *)arg1; 253 int offset = (char *)arg1 - (char *)&cnt; 254 int i; 255 256 CPU_FOREACH(i) { 257 struct pcpu *pcpu = pcpu_find(i); 258 count += *(int *)((char *)&pcpu->pc_cnt + offset); 259 } 260 return (SYSCTL_OUT(req, &count, sizeof(int))); 261} 262 263SYSCTL_PROC(_vm, VM_TOTAL, vmtotal, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 264 0, sizeof(struct vmtotal), vmtotal, "S,vmtotal", 265 "System virtual memory statistics"); 266SYSCTL_NODE(_vm, OID_AUTO, stats, CTLFLAG_RW, 0, "VM meter stats"); 267static SYSCTL_NODE(_vm_stats, OID_AUTO, sys, CTLFLAG_RW, 0, 268 "VM meter sys stats"); 269static SYSCTL_NODE(_vm_stats, OID_AUTO, vm, CTLFLAG_RW, 0, 270 "VM meter vm stats"); 271SYSCTL_NODE(_vm_stats, OID_AUTO, misc, CTLFLAG_RW, 0, "VM meter misc stats"); 272 273#define VM_STATS_SYS(var, descr) SYSCTL_PROC(_vm_stats_sys, OID_AUTO, var, \ 274 (CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE), &cnt.v_swtch, 0, vcnt, \ 275 "IU", descr); 276 277VM_STATS_SYS(v_swtch, "Context switches"); 278VM_STATS_SYS(v_trap, "Traps"); 279VM_STATS_SYS(v_syscall, "System calls"); 280VM_STATS_SYS(v_intr, "Device interrupts"); 281VM_STATS_SYS(v_soft, "Software interrupts"); 282VM_STATS_SYS(v_vm_faults, "Address memory faults"); 283VM_STATS_SYS(v_cow_faults, "Copy-on-write faults"); 284VM_STATS_SYS(v_cow_optim, "Optimized COW faults") 285VM_STATS_SYS(v_zfod, "Pages zero-filled on demand") 286VM_STATS_SYS(v_ozfod, "Optimized zero fill pages") 287VM_STATS_SYS(v_swapin, "Swap pager pageins") 288VM_STATS_SYS(v_swapout, "Swap pager pageouts") 289VM_STATS_SYS(v_swappgsin, "Swap pages swapped in") 290VM_STATS_SYS(v_swappgsout, "Swap pages swapped out") 291VM_STATS_SYS(v_vnodein, "Vnode pager pageins") 292VM_STATS_SYS(v_vnodeout, "Vnode pager pageouts") 293VM_STATS_SYS(v_vnodepgsin, "Vnode pages paged in") 294VM_STATS_SYS(v_vnodepgsout, "Vnode pages paged out") 295VM_STATS_SYS(v_intrans, "In transit page faults") 296VM_STATS_SYS(v_reactivated, "Pages reactivated from free list") 297VM_STATS_SYS(v_pdwakeups, "Pagedaemon wakeups") 298VM_STATS_SYS(v_pdpages, "Pages analyzed by pagedaemon") 299VM_STATS_SYS(v_tcached, "Total pages cached") 300VM_STATS_SYS(v_dfree, "Pages freed by pagedaemon") 301VM_STATS_SYS(v_pfree, "Pages freed by exiting processes") 302VM_STATS_SYS(v_tfree, "Total pages freed") 303VM_STATS_SYS(v_page_size, "Page size in bytes") 304VM_STATS_SYS(v_page_count, "Total number of pages in system") 305VM_STATS_SYS(v_free_reserved, "Pages reserved for deadlock") 306VM_STATS_SYS(v_free_target, "Pages desired free") 307VM_STATS_SYS(v_free_min, "Minimum low-free-pages threshold") 308VM_STATS_SYS(v_free_count, "Free pages") 309VM_STATS_SYS(v_wire_count, "Wired pages") 310VM_STATS_SYS(v_active_count, "Active pages") 311VM_STATS_SYS(v_inactive_target, "Desired inactive pages") 312VM_STATS_SYS(v_inactive_count, "Inactive pages") 313VM_STATS_SYS(v_cache_count, "Pages on cache queue") 314VM_STATS_SYS(v_cache_min, "Min pages on cache queue") 315VM_STATS_SYS(v_cache_max, "Max pages on cached queue") 316VM_STATS_SYS(v_pageout_free_min, "Min pages reserved for kernel") 317VM_STATS_SYS(v_interrupt_free_min, "Reserved pages for interrupt code") 318VM_STATS_SYS(v_forks, "Number of fork() calls") 319VM_STATS_SYS(v_vforks, "Number of vfork() calls") 320VM_STATS_SYS(v_rforks, "Number of rfork() calls") 321VM_STATS_SYS(v_kthreads, "Number of fork() calls by kernel") 322VM_STATS_SYS(v_forkpages, "VM pages affected by fork()") 323VM_STATS_SYS(v_vforkpages, "VM pages affected by vfork()") 324VM_STATS_SYS(v_rforkpages, "VM pages affected by rfork()") 325VM_STATS_SYS(v_kthreadpages, "VM pages affected by fork() by kernel") 326 327SYSCTL_INT(_vm_stats_misc, OID_AUTO, zero_page_count, CTLFLAG_RD, 328 &vm_page_zero_count, 0, "Number of zero-ed free pages"); 329