sys/vm/vm_meter.c

   1 /*
   2  * (MPSAFE)
   3  *
   4  * Copyright (c) 1982, 1986, 1989, 1993
   5  *      The Regents of the University of California.  All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  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.
  15  * 3. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      @(#)vm_meter.c  8.4 (Berkeley) 1/4/94
  32  * $FreeBSD: src/sys/vm/vm_meter.c,v 1.34.2.7 2002/10/10 19:28:22 dillon Exp $
  33  * $DragonFly: src/sys/vm/vm_meter.c,v 1.15 2008/04/28 18:04:08 dillon Exp $
  34  */
  35
  36 #include <sys/param.h>
  37 #include <sys/proc.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/resource.h>
  41 #include <sys/vmmeter.h>
  42 #include <sys/kcollect.h>
  43
  44 #include <vm/vm.h>
  45 #include <vm/vm_page.h>
  46 #include <vm/vm_extern.h>
  47 #include <vm/vm_param.h>
  48 #include <sys/lock.h>
  49 #include <vm/pmap.h>
  50 #include <vm/vm_map.h>
  51 #include <vm/vm_object.h>
  52 #include <sys/sysctl.h>
  53
  54 /*
  55  * WARNING: vmstats represents the final say, but individual cpu's may
  56  *          accumulative adjustments in gd->gd_vmstats_adj.  These are
  57  *          synchronized to the global vmstats in hardclock.
  58  *
  59  *          In addition, most individual cpus check vmstats using a local
  60  *          copy of the global vmstats in gd->gd_vmstats.  Hardclock also
  61  *          sychronizes the copy.  The pageout code and vm_page_alloc will
  62  *          also synchronize their local copies as necessary.
  63  *
  64  *          Other consumers should not expect perfect values.
  65  */
  66 __cachealign struct vmstats vmstats;
  67
  68 static int maxslp = MAXSLP;
  69
  70 SYSCTL_UINT(_vm, VM_V_FREE_MIN, v_free_min,
  71         CTLFLAG_RW, &vmstats.v_free_min, 0,
  72         "Minimum number of pages desired free");
  73 SYSCTL_UINT(_vm, VM_V_FREE_TARGET, v_free_target,
  74         CTLFLAG_RW, &vmstats.v_free_target, 0,
  75         "Number of pages desired free");
  76 SYSCTL_UINT(_vm, VM_V_FREE_RESERVED, v_free_reserved,
  77         CTLFLAG_RW, &vmstats.v_free_reserved, 0,
  78         "Number of pages reserved for deadlock");
  79 SYSCTL_UINT(_vm, VM_V_INACTIVE_TARGET, v_inactive_target,
  80         CTLFLAG_RW, &vmstats.v_inactive_target, 0,
  81         "Number of pages desired inactive");
  82 SYSCTL_UINT(_vm, VM_V_CACHE_MIN, v_cache_min,
  83         CTLFLAG_RW, &vmstats.v_cache_min, 0,
  84         "Min number of pages desired on cache queue");
  85 SYSCTL_UINT(_vm, VM_V_CACHE_MAX, v_cache_max,
  86         CTLFLAG_RW, &vmstats.v_cache_max, 0,
  87         "Max number of pages in cached obj");
  88 SYSCTL_UINT(_vm, VM_V_PAGEOUT_FREE_MIN, v_pageout_free_min,
  89         CTLFLAG_RW, &vmstats.v_pageout_free_min, 0,
  90         "Min number pages reserved for kernel");
  91 SYSCTL_UINT(_vm, OID_AUTO, v_free_severe,
  92         CTLFLAG_RW, &vmstats.v_free_severe, 0, "");
  93
  94 SYSCTL_STRUCT(_vm, VM_LOADAVG, loadavg, CTLFLAG_RD,
  95     &averunnable, loadavg, "Machine loadaverage history");
  96
  97 static int do_vmtotal_callback(struct proc *p, void *data);
  98
  99 /*
 100  * No requirements.
 101  */
 102 static int
 103 do_vmtotal(SYSCTL_HANDLER_ARGS)
 104 {
 105         struct vmtotal total;
 106         globaldata_t gd;
 107         int n;
 108
 109         bzero(&total, sizeof(total));
 110         for (n = 0; n < ncpus; ++n) {
 111                 gd = globaldata_find(n);
 112
 113                 /* total.t_rq calculated separately */
 114                 /* total.t_dw calculated separately */
 115                 /* total.t_pw calculated separately */
 116                 /* total.t_sl calculated separately */
 117                 /* total.t_sw calculated separately */
 118                 total.t_vm += gd->gd_vmtotal.t_vm;
 119                 total.t_avm += gd->gd_vmtotal.t_avm;
 120                 total.t_rm += gd->gd_vmtotal.t_rm;
 121                 total.t_arm += gd->gd_vmtotal.t_arm;
 122                 total.t_vmshr += gd->gd_vmtotal.t_vmshr;
 123                 total.t_avmshr += gd->gd_vmtotal.t_avmshr;
 124                 total.t_rmshr += gd->gd_vmtotal.t_rmshr;
 125                 total.t_armshr += gd->gd_vmtotal.t_armshr;
 126                 /* total.t_free calculated separately */
 127         }
 128
 129         /*
 130          * Calculate process statistics.
 131          */
 132         allproc_scan(do_vmtotal_callback, &total, 0);
 133
 134         /*
 135          * Adjust for sysctl return.  Add real memory into virtual memory.
 136          * Set t_free.
 137          *
 138          * t_rm - Real memory
 139          * t_vm - Virtual memory (real + swap)
 140          */
 141         total.t_vm += total.t_rm;
 142         total.t_free = vmstats.v_free_count + vmstats.v_cache_count;
 143
 144         return (sysctl_handle_opaque(oidp, &total, sizeof(total), req));
 145 }
 146
 147 static int
 148 do_vmtotal_callback(struct proc *p, void *data)
 149 {
 150         struct vmtotal *totalp = data;
 151         struct lwp *lp;
 152
 153         if (p->p_flags & P_SYSTEM)
 154                 return(0);
 155
 156         lwkt_gettoken(&p->p_token);
 157
 158         FOREACH_LWP_IN_PROC(lp, p) {
 159                 switch (lp->lwp_stat) {
 160                 case LSSTOP:
 161                 case LSSLEEP:
 162                         if ((p->p_flags & P_SWAPPEDOUT) == 0) {
 163                                 if ((lp->lwp_flags & LWP_SINTR) == 0)
 164                                         totalp->t_dw++;
 165                                 else if (lp->lwp_slptime < maxslp)
 166                                         totalp->t_sl++;
 167                         } else if (lp->lwp_slptime < maxslp) {
 168                                 totalp->t_sw++;
 169                         }
 170                         if (lp->lwp_slptime >= maxslp)
 171                                 goto out;
 172                         break;
 173
 174                 case LSRUN:
 175                         if (p->p_flags & P_SWAPPEDOUT)
 176                                 totalp->t_sw++;
 177                         else
 178                                 totalp->t_rq++;
 179                         if (p->p_stat == SIDL)
 180                                 goto out;
 181                         break;
 182
 183                 default:
 184                         goto out;
 185                 }
 186
 187                 /*
 188                  * Set while in vm_fault()
 189                  */
 190                 if (lp->lwp_flags & LWP_PAGING)
 191                         totalp->t_pw++;
 192         }
 193 out:
 194         lwkt_reltoken(&p->p_token);
 195         return(0);
 196 }
 197
 198 /*
 199  * No requirements.
 200  */
 201 static int
 202 do_vmstats(SYSCTL_HANDLER_ARGS)
 203 {
 204         struct vmstats vms = vmstats;
 205         return (sysctl_handle_opaque(oidp, &vms, sizeof(vms), req));
 206 }
 207
 208 /*
 209  * No requirements.
 210  */
 211 static int
 212 do_vmmeter(SYSCTL_HANDLER_ARGS)
 213 {
 214         int boffset = offsetof(struct vmmeter, vmmeter_uint_begin);
 215         int eoffset = offsetof(struct vmmeter, vmmeter_uint_end);
 216         struct vmmeter vmm;
 217         int i;
 218
 219         bzero(&vmm, sizeof(vmm));
 220         for (i = 0; i < ncpus; ++i) {
 221                 int off;
 222                 struct globaldata *gd = globaldata_find(i);
 223
 224                 for (off = boffset; off <= eoffset; off += sizeof(u_int)) {
 225                         *(u_int *)((char *)&vmm + off) +=
 226                                 *(u_int *)((char *)&gd->gd_cnt + off);
 227                 }
 228
 229         }
 230         vmm.v_intr += vmm.v_ipi + vmm.v_timer;
 231         return (sysctl_handle_opaque(oidp, &vmm, sizeof(vmm), req));
 232 }
 233
 234 /*
 235  * vcnt() -     accumulate statistics from the cnt structure for each cpu
 236  *
 237  *      The vmmeter structure is now per-cpu as well as global.  Those
 238  *      statistics which can be kept on a per-cpu basis (to avoid cache
 239  *      stalls between cpus) can be moved to the per-cpu vmmeter.  Remaining
 240  *      statistics, such as v_free_reserved, are left in the global
 241  *      structure.
 242  *
 243  * (sysctl_oid *oidp, void *arg1, int arg2, struct sysctl_req *req)
 244  *
 245  * No requirements.
 246  */
 247 static int
 248 vcnt(SYSCTL_HANDLER_ARGS)
 249 {
 250         int i;
 251         int count = 0;
 252         int offset = arg2;
 253
 254         for (i = 0; i < ncpus; ++i) {
 255                 struct globaldata *gd = globaldata_find(i);
 256                 count += *(int *)((char *)&gd->gd_cnt + offset);
 257         }
 258         return(SYSCTL_OUT(req, &count, sizeof(int)));
 259 }
 260
 261 /*
 262  * No requirements.
 263  */
 264 static int
 265 vcnt_intr(SYSCTL_HANDLER_ARGS)
 266 {
 267         int i;
 268         int count = 0;
 269
 270         for (i = 0; i < ncpus; ++i) {
 271                 struct globaldata *gd = globaldata_find(i);
 272
 273                 count += gd->gd_cnt.v_intr + gd->gd_cnt.v_ipi +
 274                          gd->gd_cnt.v_timer;
 275         }
 276         return(SYSCTL_OUT(req, &count, sizeof(int)));
 277 }
 278
 279 #define VMMETEROFF(var) offsetof(struct vmmeter, var)
 280
 281 SYSCTL_PROC(_vm, OID_AUTO, vmtotal, CTLTYPE_OPAQUE|CTLFLAG_RD,
 282     0, sizeof(struct vmtotal), do_vmtotal, "S,vmtotal",
 283     "System virtual memory aggregate");
 284 SYSCTL_PROC(_vm, OID_AUTO, vmstats, CTLTYPE_OPAQUE|CTLFLAG_RD,
 285     0, sizeof(struct vmstats), do_vmstats, "S,vmstats",
 286     "System virtual memory statistics");
 287 SYSCTL_PROC(_vm, OID_AUTO, vmmeter, CTLTYPE_OPAQUE|CTLFLAG_RD,
 288     0, sizeof(struct vmmeter), do_vmmeter, "S,vmmeter",
 289     "System statistics");
 290 SYSCTL_NODE(_vm, OID_AUTO, stats, CTLFLAG_RW, 0, "VM meter stats");
 291 SYSCTL_NODE(_vm_stats, OID_AUTO, sys, CTLFLAG_RW, 0, "VM meter sys stats");
 292 SYSCTL_NODE(_vm_stats, OID_AUTO, vm, CTLFLAG_RW, 0, "VM meter vm stats");
 293 SYSCTL_NODE(_vm_stats, OID_AUTO, misc, CTLFLAG_RW, 0, "VM meter misc stats");
 294
 295 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_swtch, CTLTYPE_UINT|CTLFLAG_RD,
 296         0, VMMETEROFF(v_swtch), vcnt, "IU", "Context switches");
 297 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_intrans_coll, CTLTYPE_UINT|CTLFLAG_RD,
 298         0, VMMETEROFF(v_intrans_coll), vcnt, "IU", "Intransit map collisions (total)");
 299 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_intrans_wait, CTLTYPE_UINT|CTLFLAG_RD,
 300         0, VMMETEROFF(v_intrans_wait), vcnt, "IU", "Intransit map collisions which blocked");
 301 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_forwarded_ints, CTLTYPE_UINT|CTLFLAG_RD,
 302         0, VMMETEROFF(v_forwarded_ints), vcnt, "IU", "Forwarded interrupts due to MP lock");
 303 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_forwarded_hits, CTLTYPE_UINT|CTLFLAG_RD,
 304         0, VMMETEROFF(v_forwarded_hits), vcnt, "IU", "Forwarded hits due to MP lock");
 305 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_forwarded_misses, CTLTYPE_UINT|CTLFLAG_RD,
 306         0, VMMETEROFF(v_forwarded_misses), vcnt, "IU", "Forwarded misses due to MP lock");
 307 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_trap, CTLTYPE_UINT|CTLFLAG_RD,
 308         0, VMMETEROFF(v_trap), vcnt, "IU", "Traps");
 309 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_syscall, CTLTYPE_UINT|CTLFLAG_RD,
 310         0, VMMETEROFF(v_syscall), vcnt, "IU", "Syscalls");
 311 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_intr, CTLTYPE_UINT|CTLFLAG_RD,
 312         0, VMMETEROFF(v_intr), vcnt_intr, "IU", "Hardware interrupts");
 313 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_ipi, CTLTYPE_UINT|CTLFLAG_RD,
 314         0, VMMETEROFF(v_ipi), vcnt, "IU", "Inter-processor interrupts");
 315 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_timer, CTLTYPE_UINT|CTLFLAG_RD,
 316         0, VMMETEROFF(v_timer), vcnt, "IU", "LAPIC timer interrupts");
 317 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_soft, CTLTYPE_UINT|CTLFLAG_RD,
 318         0, VMMETEROFF(v_soft), vcnt, "IU", "Software interrupts");
 319 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vm_faults, CTLTYPE_UINT|CTLFLAG_RD,
 320         0, VMMETEROFF(v_vm_faults), vcnt, "IU", "VM faults");
 321 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_cow_faults, CTLTYPE_UINT|CTLFLAG_RD,
 322         0, VMMETEROFF(v_cow_faults), vcnt, "IU", "COW faults");
 323 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_cow_optim, CTLTYPE_UINT|CTLFLAG_RD,
 324         0, VMMETEROFF(v_cow_optim), vcnt, "IU", "Optimized COW faults");
 325 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_zfod, CTLTYPE_UINT|CTLFLAG_RD,
 326         0, VMMETEROFF(v_zfod), vcnt, "IU", "Zero fill");
 327 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_ozfod, CTLTYPE_UINT|CTLFLAG_RD,
 328         0, VMMETEROFF(v_ozfod), vcnt, "IU", "Optimized zero fill");
 329 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swapin, CTLTYPE_UINT|CTLFLAG_RD,
 330         0, VMMETEROFF(v_swapin), vcnt, "IU", "Swapin operations");
 331 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swapout, CTLTYPE_UINT|CTLFLAG_RD,
 332         0, VMMETEROFF(v_swapout), vcnt, "IU", "Swapout operations");
 333 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swappgsin, CTLTYPE_UINT|CTLFLAG_RD,
 334         0, VMMETEROFF(v_swappgsin), vcnt, "IU", "Swapin pages");
 335 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swappgsout, CTLTYPE_UINT|CTLFLAG_RD,
 336         0, VMMETEROFF(v_swappgsout), vcnt, "IU", "Swapout pages");
 337 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodein, CTLTYPE_UINT|CTLFLAG_RD,
 338         0, VMMETEROFF(v_vnodein), vcnt, "IU", "Vnodein operations");
 339 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodeout, CTLTYPE_UINT|CTLFLAG_RD,
 340         0, VMMETEROFF(v_vnodeout), vcnt, "IU", "Vnodeout operations");
 341 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodepgsin, CTLTYPE_UINT|CTLFLAG_RD,
 342         0, VMMETEROFF(v_vnodepgsin), vcnt, "IU", "Vnodein pages");
 343 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodepgsout, CTLTYPE_UINT|CTLFLAG_RD,
 344         0, VMMETEROFF(v_vnodepgsout), vcnt, "IU", "Vnodeout pages");
 345 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_intrans, CTLTYPE_UINT|CTLFLAG_RD,
 346         0, VMMETEROFF(v_intrans), vcnt, "IU", "In transit page blocking");
 347 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_reactivated, CTLTYPE_UINT|CTLFLAG_RD,
 348         0, VMMETEROFF(v_reactivated), vcnt, "IU", "Reactivated pages");
 349 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_pdwakeups, CTLTYPE_UINT|CTLFLAG_RD,
 350         0, VMMETEROFF(v_pdwakeups), vcnt, "IU", "Pagedaemon wakeups");
 351 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_ppwakeups, CTLTYPE_UINT|CTLFLAG_RD,
 352         0, VMMETEROFF(v_ppwakeups), vcnt, "IU", "vm_wait wakeups");
 353 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_pdpages, CTLTYPE_UINT|CTLFLAG_RD,
 354         0, VMMETEROFF(v_pdpages), vcnt, "IU", "Pagedaemon page scans");
 355 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_dfree, CTLTYPE_UINT|CTLFLAG_RD,
 356         0, VMMETEROFF(v_dfree), vcnt, "IU", "Pages freed by daemon");
 357 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_pfree, CTLTYPE_UINT|CTLFLAG_RD,
 358         0, VMMETEROFF(v_pfree), vcnt, "IU", "Pages freed by exiting processes");
 359 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_tfree, CTLTYPE_UINT|CTLFLAG_RD,
 360         0, VMMETEROFF(v_tfree), vcnt, "IU", "Total pages freed");
 361 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_forks, CTLTYPE_UINT|CTLFLAG_RD,
 362         0, VMMETEROFF(v_forks), vcnt, "IU", "Number of fork() calls");
 363 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vforks, CTLTYPE_UINT|CTLFLAG_RD,
 364         0, VMMETEROFF(v_vforks), vcnt, "IU", "Number of vfork() calls");
 365 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_rforks, CTLTYPE_UINT|CTLFLAG_RD,
 366         0, VMMETEROFF(v_rforks), vcnt, "IU", "Number of rfork() calls");
 367 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_kthreads, CTLTYPE_UINT|CTLFLAG_RD,
 368         0, VMMETEROFF(v_kthreads), vcnt, "IU", "Number of fork() calls by kernel");
 369 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_forkpages, CTLTYPE_UINT|CTLFLAG_RD,
 370         0, VMMETEROFF(v_forkpages), vcnt, "IU", "VM pages affected by fork()");
 371 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vforkpages, CTLTYPE_UINT|CTLFLAG_RD,
 372         0, VMMETEROFF(v_vforkpages), vcnt, "IU", "VM pages affected by vfork()");
 373 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_rforkpages, CTLTYPE_UINT|CTLFLAG_RD,
 374         0, VMMETEROFF(v_rforkpages), vcnt, "IU", "VM pages affected by rfork()");
 375 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_kthreadpages, CTLTYPE_UINT|CTLFLAG_RD,
 376         0, VMMETEROFF(v_kthreadpages), vcnt, "IU", "VM pages affected by fork() by kernel");
 377
 378 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 379         v_page_size, CTLFLAG_RD, &vmstats.v_page_size, 0,
 380         "Page size in bytes");
 381 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 382         v_page_count, CTLFLAG_RD, &vmstats.v_page_count, 0,
 383         "Total number of pages in system");
 384 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 385         v_free_reserved, CTLFLAG_RD, &vmstats.v_free_reserved, 0,
 386         "Number of pages reserved for deadlock");
 387 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 388         v_free_target, CTLFLAG_RD, &vmstats.v_free_target, 0,
 389         "Number of pages desired free");
 390 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 391         v_free_min, CTLFLAG_RD, &vmstats.v_free_min, 0,
 392         "Minimum number of pages desired free");
 393 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 394         v_free_count, CTLFLAG_RD, &vmstats.v_free_count, 0,
 395         "Number of pages free");
 396 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 397         v_wire_count, CTLFLAG_RD, &vmstats.v_wire_count, 0,
 398         "Number of pages wired down");
 399 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 400         v_active_count, CTLFLAG_RD, &vmstats.v_active_count, 0,
 401         "Number of pages active");
 402 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 403         v_inactive_target, CTLFLAG_RD, &vmstats.v_inactive_target, 0,
 404         "Number of pages desired inactive");
 405 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 406         v_inactive_count, CTLFLAG_RD, &vmstats.v_inactive_count, 0,
 407         "Number of pages inactive");
 408 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 409         v_cache_count, CTLFLAG_RD, &vmstats.v_cache_count, 0,
 410         "Number of pages on buffer cache queue");
 411 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 412         v_cache_min, CTLFLAG_RD, &vmstats.v_cache_min, 0,
 413         "Min number of pages desired on cache queue");
 414 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 415         v_cache_max, CTLFLAG_RD, &vmstats.v_cache_max, 0,
 416         "Max number of pages in cached obj");
 417 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 418         v_pageout_free_min, CTLFLAG_RD, &vmstats.v_pageout_free_min, 0,
 419         "Min number pages reserved for kernel");
 420 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
 421         v_interrupt_free_min, CTLFLAG_RD, &vmstats.v_interrupt_free_min, 0,
 422         "Reserved number of pages for int code");
 423
 424 /*
 425  * No requirements.
 426  */
 427 static int
 428 do_vmmeter_pcpu(SYSCTL_HANDLER_ARGS)
 429 {
 430         int boffset = offsetof(struct vmmeter, vmmeter_uint_begin);
 431         int eoffset = offsetof(struct vmmeter, vmmeter_uint_end);
 432         struct globaldata *gd = arg1;
 433         struct vmmeter vmm;
 434         int off;
 435
 436         bzero(&vmm, sizeof(vmm));
 437         for (off = boffset; off <= eoffset; off += sizeof(u_int)) {
 438                 *(u_int *)((char *)&vmm + off) +=
 439                         *(u_int *)((char *)&gd->gd_cnt + off);
 440         }
 441         vmm.v_intr += vmm.v_ipi + vmm.v_timer;
 442         return (sysctl_handle_opaque(oidp, &vmm, sizeof(vmm), req));
 443 }
 444
 445 /*
 446  * Callback for long-term slow data collection on 10-second interval.
 447  *
 448  * Return faults, set data for other entries.
 449  */
 450 #define PTOB(value)     ((uint64_t)(value) << PAGE_SHIFT)
 451
 452 static uint64_t
 453 collect_vmstats_callback(int n)
 454 {
 455         static struct vmmeter last_vmm;
 456         struct vmmeter cur_vmm;
 457         const int boffset = offsetof(struct vmmeter, vmmeter_uint_begin);
 458         const int eoffset = offsetof(struct vmmeter, vmmeter_uint_end);
 459         uint64_t total;
 460
 461         /*
 462          * The hardclock already rolls up vmstats for us.
 463          */
 464         kcollect_setvalue(KCOLLECT_MEMFRE, PTOB(vmstats.v_free_count));
 465         kcollect_setvalue(KCOLLECT_MEMCAC, PTOB(vmstats.v_cache_count));
 466         kcollect_setvalue(KCOLLECT_MEMINA, PTOB(vmstats.v_inactive_count));
 467         kcollect_setvalue(KCOLLECT_MEMACT, PTOB(vmstats.v_active_count));
 468         kcollect_setvalue(KCOLLECT_MEMWIR, PTOB(vmstats.v_wire_count));
 469
 470         /*
 471          * Collect pcpu statistics for things like faults.
 472          */
 473         bzero(&cur_vmm, sizeof(cur_vmm));
 474         for (n = 0; n < ncpus; ++n) {
 475                 struct globaldata *gd = globaldata_find(n);
 476                 int off;
 477
 478                 for (off = boffset; off <= eoffset; off += sizeof(u_int)) {
 479                         *(u_int *)((char *)&cur_vmm + off) +=
 480                                 *(u_int *)((char *)&gd->gd_cnt + off);
 481                 }
 482
 483         }
 484
 485         total = cur_vmm.v_cow_faults - last_vmm.v_cow_faults;
 486         last_vmm.v_cow_faults = cur_vmm.v_cow_faults;
 487         kcollect_setvalue(KCOLLECT_COWFAULT, total);
 488
 489         total = cur_vmm.v_zfod - last_vmm.v_zfod;
 490         last_vmm.v_zfod = cur_vmm.v_zfod;
 491         kcollect_setvalue(KCOLLECT_ZFILL, total);
 492
 493         total = cur_vmm.v_syscall - last_vmm.v_syscall;
 494         last_vmm.v_syscall = cur_vmm.v_syscall;
 495         kcollect_setvalue(KCOLLECT_SYSCALLS, total);
 496
 497         total = cur_vmm.v_intr - last_vmm.v_intr;
 498         last_vmm.v_intr = cur_vmm.v_intr;
 499         kcollect_setvalue(KCOLLECT_INTR, total);
 500
 501         total = cur_vmm.v_ipi - last_vmm.v_ipi;
 502         last_vmm.v_ipi = cur_vmm.v_ipi;
 503         kcollect_setvalue(KCOLLECT_IPI, total);
 504
 505         total = cur_vmm.v_timer - last_vmm.v_timer;
 506         last_vmm.v_timer = cur_vmm.v_timer;
 507         kcollect_setvalue(KCOLLECT_TIMER, total);
 508
 509         total = cur_vmm.v_vm_faults - last_vmm.v_vm_faults;
 510         last_vmm.v_vm_faults = cur_vmm.v_vm_faults;
 511
 512         return total;
 513 }
 514
 515 /*
 516  * Called from the low level boot code only.
 517  */
 518 static void
 519 vmmeter_init(void *dummy __unused)
 520 {
 521         int i;
 522
 523         for (i = 0; i < ncpus; ++i) {
 524                 struct sysctl_ctx_list *ctx;
 525                 struct sysctl_oid *oid;
 526                 struct globaldata *gd;
 527                 char name[32];
 528
 529                 ksnprintf(name, sizeof(name), "cpu%d", i);
 530
 531                 ctx = kmalloc(sizeof(*ctx), M_TEMP, M_WAITOK);
 532                 sysctl_ctx_init(ctx);
 533                 oid = SYSCTL_ADD_NODE(ctx, SYSCTL_STATIC_CHILDREN(_vm),
 534                                       OID_AUTO, name, CTLFLAG_RD, 0, "");
 535
 536                 gd = globaldata_find(i);
 537                 SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
 538                                 "vmmeter", CTLTYPE_OPAQUE|CTLFLAG_RD,
 539                                 gd, sizeof(struct vmmeter), do_vmmeter_pcpu,
 540                                 "S,vmmeter", "System per-cpu statistics");
 541         }
 542         kcollect_register(KCOLLECT_VMFAULT, "fault", collect_vmstats_callback,
 543                           KCOLLECT_SCALE(KCOLLECT_VMFAULT_FORMAT, 0));
 544         kcollect_register(KCOLLECT_COWFAULT, "cow", NULL,
 545                           KCOLLECT_SCALE(KCOLLECT_COWFAULT_FORMAT, 0));
 546         kcollect_register(KCOLLECT_ZFILL, "zfill", NULL,
 547                           KCOLLECT_SCALE(KCOLLECT_ZFILL_FORMAT, 0));
 548
 549         kcollect_register(KCOLLECT_MEMFRE, "free", NULL,
 550                           KCOLLECT_SCALE(KCOLLECT_MEMFRE_FORMAT,
 551                                          PTOB(vmstats.v_page_count)));
 552         kcollect_register(KCOLLECT_MEMCAC, "cache", NULL,
 553                           KCOLLECT_SCALE(KCOLLECT_MEMCAC_FORMAT,
 554                                          PTOB(vmstats.v_page_count)));
 555         kcollect_register(KCOLLECT_MEMINA, "inact", NULL,
 556                           KCOLLECT_SCALE(KCOLLECT_MEMINA_FORMAT,
 557                                          PTOB(vmstats.v_page_count)));
 558         kcollect_register(KCOLLECT_MEMACT, "act", NULL,
 559                           KCOLLECT_SCALE(KCOLLECT_MEMACT_FORMAT,
 560                                          PTOB(vmstats.v_page_count)));
 561         kcollect_register(KCOLLECT_MEMWIR, "wired", NULL,
 562                           KCOLLECT_SCALE(KCOLLECT_MEMWIR_FORMAT,
 563                                          PTOB(vmstats.v_page_count)));
 564
 565         kcollect_register(KCOLLECT_SYSCALLS, "syscalls", NULL,
 566                           KCOLLECT_SCALE(KCOLLECT_SYSCALLS_FORMAT, 0));
 567
 568         kcollect_register(KCOLLECT_INTR, "intr", NULL,
 569                           KCOLLECT_SCALE(KCOLLECT_INTR_FORMAT, 0));
 570         kcollect_register(KCOLLECT_IPI, "ipi", NULL,
 571                           KCOLLECT_SCALE(KCOLLECT_IPI_FORMAT, 0));
 572         kcollect_register(KCOLLECT_TIMER, "timer", NULL,
 573                           KCOLLECT_SCALE(KCOLLECT_TIMER_FORMAT, 0));
 574 }
 575 SYSINIT(vmmeter, SI_SUB_PSEUDO, SI_ORDER_ANY, vmmeter_init, 0);
 576
 577 /*
 578  * Rolls up accumulated pcpu adjustments to vmstats counts into the global
 579  * structure, copy the global structure into our pcpu structure.  Critical
 580  * path checks will use our pcpu structure.
 581  *
 582  * This is somewhat expensive and only called when needed, and by the
 583  * hardclock.
 584  */
 585 void
 586 vmstats_rollup(void)
 587 {
 588         int cpu;
 589
 590         for (cpu = 0; cpu < ncpus; ++cpu) {
 591                 vmstats_rollup_cpu(globaldata_find(cpu));
 592         }
 593         mycpu->gd_vmstats = vmstats;
 594 }
 595
 596 void
 597 vmstats_rollup_cpu(globaldata_t gd)
 598 {
 599         int value;
 600
 601         if (gd->gd_vmstats_adj.v_free_count) {
 602                 value = atomic_swap_int(&gd->gd_vmstats_adj.v_free_count, 0);
 603                 atomic_add_int(&vmstats.v_free_count, value);
 604         }
 605         if (gd->gd_vmstats_adj.v_cache_count) {
 606                 value = atomic_swap_int(&gd->gd_vmstats_adj.v_cache_count, 0);
 607                 atomic_add_int(&vmstats.v_cache_count, value);
 608         }
 609         if (gd->gd_vmstats_adj.v_inactive_count) {
 610                 value =atomic_swap_int(&gd->gd_vmstats_adj.v_inactive_count, 0);
 611                 atomic_add_int(&vmstats.v_inactive_count, value);
 612         }
 613         if (gd->gd_vmstats_adj.v_active_count) {
 614                 value = atomic_swap_int(&gd->gd_vmstats_adj.v_active_count, 0);
 615                 atomic_add_int(&vmstats.v_active_count, value);
 616         }
 617         if (gd->gd_vmstats_adj.v_wire_count) {
 618                 value = atomic_swap_int(&gd->gd_vmstats_adj.v_wire_count, 0);
 619                 atomic_add_int(&vmstats.v_wire_count, value);
 620         }
 621 }