sys/vm/vm_meter.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1989, 1993
   3  *      The Regents of the University of California.  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. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)vm_meter.c  8.4 (Berkeley) 1/4/94
  30  */
  31
  32 #include <sys/cdefs.h>
  33 __FBSDID("$FreeBSD$");
  34
  35 #include "opt_compat.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/lock.h>
  41 #include <sys/malloc.h>
  42 #include <sys/mutex.h>
  43 #include <sys/proc.h>
  44 #include <sys/resource.h>
  45 #include <sys/rwlock.h>
  46 #include <sys/sx.h>
  47 #include <sys/vmmeter.h>
  48 #include <sys/smp.h>
  49
  50 #include <vm/vm.h>
  51 #include <vm/vm_page.h>
  52 #include <vm/vm_extern.h>
  53 #include <vm/vm_param.h>
  54 #include <vm/pmap.h>
  55 #include <vm/vm_map.h>
  56 #include <vm/vm_object.h>
  57 #include <sys/sysctl.h>
  58
  59 struct vmmeter vm_cnt = {
  60         .v_swtch = EARLY_COUNTER,
  61         .v_trap = EARLY_COUNTER,
  62         .v_syscall = EARLY_COUNTER,
  63         .v_intr = EARLY_COUNTER,
  64         .v_soft = EARLY_COUNTER,
  65         .v_vm_faults = EARLY_COUNTER,
  66         .v_io_faults = EARLY_COUNTER,
  67         .v_cow_faults = EARLY_COUNTER,
  68         .v_cow_optim = EARLY_COUNTER,
  69         .v_zfod = EARLY_COUNTER,
  70         .v_ozfod = EARLY_COUNTER,
  71         .v_swapin = EARLY_COUNTER,
  72         .v_swapout = EARLY_COUNTER,
  73         .v_swappgsin = EARLY_COUNTER,
  74         .v_swappgsout = EARLY_COUNTER,
  75         .v_vnodein = EARLY_COUNTER,
  76         .v_vnodeout = EARLY_COUNTER,
  77         .v_vnodepgsin = EARLY_COUNTER,
  78         .v_vnodepgsout = EARLY_COUNTER,
  79         .v_intrans = EARLY_COUNTER,
  80         .v_reactivated = EARLY_COUNTER,
  81         .v_pdwakeups = EARLY_COUNTER,
  82         .v_pdpages = EARLY_COUNTER,
  83         .v_pdshortfalls = EARLY_COUNTER,
  84         .v_dfree = EARLY_COUNTER,
  85         .v_pfree = EARLY_COUNTER,
  86         .v_tfree = EARLY_COUNTER,
  87         .v_forks = EARLY_COUNTER,
  88         .v_vforks = EARLY_COUNTER,
  89         .v_rforks = EARLY_COUNTER,
  90         .v_kthreads = EARLY_COUNTER,
  91         .v_forkpages = EARLY_COUNTER,
  92         .v_vforkpages = EARLY_COUNTER,
  93         .v_rforkpages = EARLY_COUNTER,
  94         .v_kthreadpages = EARLY_COUNTER,
  95 };
  96
  97 static void
  98 vmcounter_startup(void)
  99 {
 100         counter_u64_t *cnt = (counter_u64_t *)&vm_cnt;
 101
 102         COUNTER_ARRAY_ALLOC(cnt, VM_METER_NCOUNTERS, M_WAITOK);
 103 }
 104 SYSINIT(counter, SI_SUB_CPU, SI_ORDER_FOURTH + 1, vmcounter_startup, NULL);
 105
 106 SYSCTL_UINT(_vm, VM_V_FREE_MIN, v_free_min,
 107         CTLFLAG_RW, &vm_cnt.v_free_min, 0, "Minimum low-free-pages threshold");
 108 SYSCTL_UINT(_vm, VM_V_FREE_TARGET, v_free_target,
 109         CTLFLAG_RW, &vm_cnt.v_free_target, 0, "Desired free pages");
 110 SYSCTL_UINT(_vm, VM_V_FREE_RESERVED, v_free_reserved,
 111         CTLFLAG_RW, &vm_cnt.v_free_reserved, 0, "Pages reserved for deadlock");
 112 SYSCTL_UINT(_vm, VM_V_INACTIVE_TARGET, v_inactive_target,
 113         CTLFLAG_RW, &vm_cnt.v_inactive_target, 0, "Pages desired inactive");
 114 SYSCTL_UINT(_vm, VM_V_PAGEOUT_FREE_MIN, v_pageout_free_min,
 115         CTLFLAG_RW, &vm_cnt.v_pageout_free_min, 0, "Min pages reserved for kernel");
 116 SYSCTL_UINT(_vm, OID_AUTO, v_free_severe,
 117         CTLFLAG_RW, &vm_cnt.v_free_severe, 0, "Severe page depletion point");
 118
 119 static int
 120 sysctl_vm_loadavg(SYSCTL_HANDLER_ARGS)
 121 {
 122
 123 #ifdef SCTL_MASK32
 124         u_int32_t la[4];
 125
 126         if (req->flags & SCTL_MASK32) {
 127                 la[0] = averunnable.ldavg[0];
 128                 la[1] = averunnable.ldavg[1];
 129                 la[2] = averunnable.ldavg[2];
 130                 la[3] = averunnable.fscale;
 131                 return SYSCTL_OUT(req, la, sizeof(la));
 132         } else
 133 #endif
 134                 return SYSCTL_OUT(req, &averunnable, sizeof(averunnable));
 135 }
 136 SYSCTL_PROC(_vm, VM_LOADAVG, loadavg, CTLTYPE_STRUCT | CTLFLAG_RD |
 137     CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_loadavg, "S,loadavg",
 138     "Machine loadaverage history");
 139
 140 /*
 141  * This function aims to determine if the object is mapped,
 142  * specifically, if it is referenced by a vm_map_entry.  Because
 143  * objects occasionally acquire transient references that do not
 144  * represent a mapping, the method used here is inexact.  However, it
 145  * has very low overhead and is good enough for the advisory
 146  * vm.vmtotal sysctl.
 147  */
 148 static bool
 149 is_object_active(vm_object_t obj)
 150 {
 151
 152         return (obj->ref_count > obj->shadow_count);
 153 }
 154
 155 static int
 156 vmtotal(SYSCTL_HANDLER_ARGS)
 157 {
 158         struct vmtotal total;
 159         vm_object_t object;
 160         struct proc *p;
 161         struct thread *td;
 162
 163         bzero(&total, sizeof(total));
 164
 165         /*
 166          * Calculate process statistics.
 167          */
 168         sx_slock(&allproc_lock);
 169         FOREACH_PROC_IN_SYSTEM(p) {
 170                 if ((p->p_flag & P_SYSTEM) != 0)
 171                         continue;
 172                 PROC_LOCK(p);
 173                 if (p->p_state != PRS_NEW) {
 174                         FOREACH_THREAD_IN_PROC(p, td) {
 175                                 thread_lock(td);
 176                                 switch (td->td_state) {
 177                                 case TDS_INHIBITED:
 178                                         if (TD_IS_SWAPPED(td))
 179                                                 total.t_sw++;
 180                                         else if (TD_IS_SLEEPING(td)) {
 181                                                 if (td->td_priority <= PZERO)
 182                                                         total.t_dw++;
 183                                                 else
 184                                                         total.t_sl++;
 185                                                 if (td->td_wchan ==
 186                                                     &vm_cnt.v_free_count)
 187                                                         total.t_pw++;
 188                                         }
 189                                         break;
 190                                 case TDS_CAN_RUN:
 191                                         total.t_sw++;
 192                                         break;
 193                                 case TDS_RUNQ:
 194                                 case TDS_RUNNING:
 195                                         total.t_rq++;
 196                                         break;
 197                                 default:
 198                                         break;
 199                                 }
 200                                 thread_unlock(td);
 201                         }
 202                 }
 203                 PROC_UNLOCK(p);
 204         }
 205         sx_sunlock(&allproc_lock);
 206         /*
 207          * Calculate object memory usage statistics.
 208          */
 209         mtx_lock(&vm_object_list_mtx);
 210         TAILQ_FOREACH(object, &vm_object_list, object_list) {
 211                 /*
 212                  * Perform unsynchronized reads on the object.  In
 213                  * this case, the lack of synchronization should not
 214                  * impair the accuracy of the reported statistics.
 215                  */
 216                 if ((object->flags & OBJ_FICTITIOUS) != 0) {
 217                         /*
 218                          * Devices, like /dev/mem, will badly skew our totals.
 219                          */
 220                         continue;
 221                 }
 222                 if (object->ref_count == 0) {
 223                         /*
 224                          * Also skip unreferenced objects, including
 225                          * vnodes representing mounted file systems.
 226                          */
 227                         continue;
 228                 }
 229                 if (object->ref_count == 1 &&
 230                     (object->flags & OBJ_NOSPLIT) != 0) {
 231                         /*
 232                          * Also skip otherwise unreferenced swap
 233                          * objects backing tmpfs vnodes, and POSIX or
 234                          * SysV shared memory.
 235                          */
 236                         continue;
 237                 }
 238                 total.t_vm += object->size;
 239                 total.t_rm += object->resident_page_count;
 240                 if (is_object_active(object)) {
 241                         total.t_avm += object->size;
 242                         total.t_arm += object->resident_page_count;
 243                 }
 244                 if (object->shadow_count > 1) {
 245                         /* shared object */
 246                         total.t_vmshr += object->size;
 247                         total.t_rmshr += object->resident_page_count;
 248                         if (is_object_active(object)) {
 249                                 total.t_avmshr += object->size;
 250                                 total.t_armshr += object->resident_page_count;
 251                         }
 252                 }
 253         }
 254         mtx_unlock(&vm_object_list_mtx);
 255         total.t_free = vm_cnt.v_free_count;
 256         return (sysctl_handle_opaque(oidp, &total, sizeof(total), req));
 257 }
 258
 259 SYSCTL_PROC(_vm, VM_TOTAL, vmtotal, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
 260     0, sizeof(struct vmtotal), vmtotal, "S,vmtotal",
 261     "System virtual memory statistics");
 262 SYSCTL_NODE(_vm, OID_AUTO, stats, CTLFLAG_RW, 0, "VM meter stats");
 263 static SYSCTL_NODE(_vm_stats, OID_AUTO, sys, CTLFLAG_RW, 0,
 264         "VM meter sys stats");
 265 static SYSCTL_NODE(_vm_stats, OID_AUTO, vm, CTLFLAG_RW, 0,
 266         "VM meter vm stats");
 267 SYSCTL_NODE(_vm_stats, OID_AUTO, misc, CTLFLAG_RW, 0, "VM meter misc stats");
 268
 269 static int
 270 sysctl_handle_vmstat(SYSCTL_HANDLER_ARGS)
 271 {
 272         uint64_t val;
 273 #ifdef COMPAT_FREEBSD11
 274         uint32_t val32;
 275 #endif
 276
 277         val = counter_u64_fetch(*(counter_u64_t *)arg1);
 278 #ifdef COMPAT_FREEBSD11
 279         if (req->oldlen == sizeof(val32)) {
 280                 val32 = val;            /* truncate */
 281                 return (SYSCTL_OUT(req, &val32, sizeof(val32)));
 282         }
 283 #endif
 284         return (SYSCTL_OUT(req, &val, sizeof(val)));
 285 }
 286
 287 #define VM_STATS(parent, var, descr) \
 288     SYSCTL_OID(parent, OID_AUTO, var, CTLTYPE_U64 | CTLFLAG_MPSAFE | \
 289     CTLFLAG_RD, &vm_cnt.var, 0, sysctl_handle_vmstat, "QU", descr);
 290 #define VM_STATS_VM(var, descr)         VM_STATS(_vm_stats_vm, var, descr)
 291 #define VM_STATS_SYS(var, descr)        VM_STATS(_vm_stats_sys, var, descr)
 292
 293 VM_STATS_SYS(v_swtch, "Context switches");
 294 VM_STATS_SYS(v_trap, "Traps");
 295 VM_STATS_SYS(v_syscall, "System calls");
 296 VM_STATS_SYS(v_intr, "Device interrupts");
 297 VM_STATS_SYS(v_soft, "Software interrupts");
 298 VM_STATS_VM(v_vm_faults, "Address memory faults");
 299 VM_STATS_VM(v_io_faults, "Page faults requiring I/O");
 300 VM_STATS_VM(v_cow_faults, "Copy-on-write faults");
 301 VM_STATS_VM(v_cow_optim, "Optimized COW faults");
 302 VM_STATS_VM(v_zfod, "Pages zero-filled on demand");
 303 VM_STATS_VM(v_ozfod, "Optimized zero fill pages");
 304 VM_STATS_VM(v_swapin, "Swap pager pageins");
 305 VM_STATS_VM(v_swapout, "Swap pager pageouts");
 306 VM_STATS_VM(v_swappgsin, "Swap pages swapped in");
 307 VM_STATS_VM(v_swappgsout, "Swap pages swapped out");
 308 VM_STATS_VM(v_vnodein, "Vnode pager pageins");
 309 VM_STATS_VM(v_vnodeout, "Vnode pager pageouts");
 310 VM_STATS_VM(v_vnodepgsin, "Vnode pages paged in");
 311 VM_STATS_VM(v_vnodepgsout, "Vnode pages paged out");
 312 VM_STATS_VM(v_intrans, "In transit page faults");
 313 VM_STATS_VM(v_reactivated, "Pages reactivated by pagedaemon");
 314 VM_STATS_VM(v_pdwakeups, "Pagedaemon wakeups");
 315 VM_STATS_VM(v_pdpages, "Pages analyzed by pagedaemon");
 316 VM_STATS_VM(v_pdshortfalls, "Page reclamation shortfalls");
 317 VM_STATS_VM(v_dfree, "Pages freed by pagedaemon");
 318 VM_STATS_VM(v_pfree, "Pages freed by exiting processes");
 319 VM_STATS_VM(v_tfree, "Total pages freed");
 320 VM_STATS_VM(v_forks, "Number of fork() calls");
 321 VM_STATS_VM(v_vforks, "Number of vfork() calls");
 322 VM_STATS_VM(v_rforks, "Number of rfork() calls");
 323 VM_STATS_VM(v_kthreads, "Number of fork() calls by kernel");
 324 VM_STATS_VM(v_forkpages, "VM pages affected by fork()");
 325 VM_STATS_VM(v_vforkpages, "VM pages affected by vfork()");
 326 VM_STATS_VM(v_rforkpages, "VM pages affected by rfork()");
 327 VM_STATS_VM(v_kthreadpages, "VM pages affected by fork() by kernel");
 328
 329 #define VM_STATS_UINT(var, descr)       \
 330     SYSCTL_UINT(_vm_stats_vm, OID_AUTO, var, CTLFLAG_RD, &vm_cnt.var, 0, descr)
 331 VM_STATS_UINT(v_page_size, "Page size in bytes");
 332 VM_STATS_UINT(v_page_count, "Total number of pages in system");
 333 VM_STATS_UINT(v_free_reserved, "Pages reserved for deadlock");
 334 VM_STATS_UINT(v_free_target, "Pages desired free");
 335 VM_STATS_UINT(v_free_min, "Minimum low-free-pages threshold");
 336 VM_STATS_UINT(v_free_count, "Free pages");
 337 VM_STATS_UINT(v_wire_count, "Wired pages");
 338 VM_STATS_UINT(v_active_count, "Active pages");
 339 VM_STATS_UINT(v_inactive_target, "Desired inactive pages");
 340 VM_STATS_UINT(v_inactive_count, "Inactive pages");
 341 VM_STATS_UINT(v_laundry_count, "Pages eligible for laundering");
 342 VM_STATS_UINT(v_pageout_free_min, "Min pages reserved for kernel");
 343 VM_STATS_UINT(v_interrupt_free_min, "Reserved pages for interrupt code");
 344 VM_STATS_UINT(v_free_severe, "Severe page depletion point");
 345
 346 #ifdef COMPAT_FREEBSD11
 347 /*
 348  * Provide compatibility sysctls for the benefit of old utilities which exit
 349  * with an error if they cannot be found.
 350  */
 351 SYSCTL_UINT(_vm_stats_vm, OID_AUTO, v_cache_count, CTLFLAG_RD,
 352     SYSCTL_NULL_UINT_PTR, 0, "Dummy for compatibility");
 353 SYSCTL_UINT(_vm_stats_vm, OID_AUTO, v_tcached, CTLFLAG_RD,
 354     SYSCTL_NULL_UINT_PTR, 0, "Dummy for compatibility");
 355 #endif