xref: /dragonfly/sys/vm/vm_meter.c (revision c8860c9a)
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  */
34 
35 #include <sys/param.h>
36 #include <sys/proc.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/malloc.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  *	    accumualte 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 __exclusive_cache_line struct vmstats vmstats;
67 
68 static int maxslp = MAXSLP;
69 
70 SYSCTL_ULONG(_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_ULONG(_vm, VM_V_PAGING_WAIT, v_paging_wait,
74 	CTLFLAG_RW, &vmstats.v_paging_wait, 0,
75 	"Userland slows down allocations");
76 SYSCTL_ULONG(_vm, VM_V_PAGING_START, v_paging_start,
77 	CTLFLAG_RW, &vmstats.v_paging_start, 0,
78 	"Pageout daemon begins running");
79 SYSCTL_ULONG(_vm, VM_V_PAGING_TARGET1, v_paging_target1,
80 	CTLFLAG_RW, &vmstats.v_paging_target1, 0,
81 	"Mid pageout daemon target");
82 SYSCTL_ULONG(_vm, VM_V_PAGING_TARGET2, v_paging_target2,
83 	CTLFLAG_RW, &vmstats.v_paging_target2, 0,
84 	"Final pageout daemon target");
85 SYSCTL_ULONG(_vm, VM_V_FREE_RESERVED, v_free_reserved,
86 	CTLFLAG_RW, &vmstats.v_free_reserved, 0,
87 	"Number of pages reserved for deadlock");
88 SYSCTL_ULONG(_vm, VM_V_INACTIVE_TARGET, v_inactive_target,
89 	CTLFLAG_RW, &vmstats.v_inactive_target, 0,
90 	"Maximum inactive pages during pageout");
91 SYSCTL_ULONG(_vm, VM_V_PAGEOUT_FREE_MIN, v_pageout_free_min,
92 	CTLFLAG_RW, &vmstats.v_pageout_free_min, 0,
93 	"Min number pages reserved for kernel");
94 SYSCTL_ULONG(_vm, OID_AUTO, v_free_severe,
95 	CTLFLAG_RW, &vmstats.v_free_severe, 0, "");
96 
97 SYSCTL_STRUCT(_vm, VM_LOADAVG, loadavg,
98 	CTLFLAG_RD | CTLFLAG_NOLOCK,
99 	&averunnable, loadavg, "Machine loadaverage history");
100 
101 static int do_vmtotal_callback(struct proc *p, void *data);
102 
103 /*
104  * No requirements.
105  */
106 static int
107 do_vmtotal(SYSCTL_HANDLER_ARGS)
108 {
109 	struct vmtotal total;
110 	globaldata_t gd;
111 	int n;
112 
113 	bzero(&total, sizeof(total));
114 	for (n = 0; n < ncpus; ++n) {
115 		gd = globaldata_find(n);
116 
117 		/* total.t_rq calculated separately */
118 		/* total.t_dw calculated separately */
119 		/* total.t_pw calculated separately */
120 		/* total.t_sl calculated separately */
121 		/* total.t_sw calculated separately */
122 		total.t_vm += gd->gd_vmtotal.t_vm;
123 		total.t_avm += gd->gd_vmtotal.t_avm;
124 		total.t_rm += gd->gd_vmtotal.t_rm;
125 		total.t_arm += gd->gd_vmtotal.t_arm;
126 		total.t_vmshr += gd->gd_vmtotal.t_vmshr;
127 		total.t_avmshr += gd->gd_vmtotal.t_avmshr;
128 		total.t_rmshr += gd->gd_vmtotal.t_rmshr;
129 		total.t_armshr += gd->gd_vmtotal.t_armshr;
130 		/* total.t_free calculated separately */
131 	}
132 
133 	/*
134 	 * Calculate process statistics.
135 	 */
136 	allproc_scan(do_vmtotal_callback, &total, 0);
137 
138 	/*
139 	 * Adjust for sysctl return.  Add real memory into virtual memory.
140 	 * Set t_free.
141 	 *
142 	 * t_rm - Real memory
143 	 * t_vm - Virtual memory (real + swap)
144 	 */
145 	total.t_vm += total.t_rm;
146 	total.t_free = vmstats.v_free_count + vmstats.v_cache_count;
147 
148 	return (sysctl_handle_opaque(oidp, &total, sizeof(total), req));
149 }
150 
151 static int
152 do_vmtotal_callback(struct proc *p, void *data)
153 {
154 	struct vmtotal *totalp = data;
155 	struct lwp *lp;
156 
157 	if (p->p_flags & P_SYSTEM)
158 		return(0);
159 
160 	lwkt_gettoken_shared(&p->p_token);
161 
162 	FOREACH_LWP_IN_PROC(lp, p) {
163 		switch (lp->lwp_stat) {
164 		case LSSTOP:
165 		case LSSLEEP:
166 			if ((lp->lwp_flags & LWP_SINTR) == 0)
167 				totalp->t_dw++;
168 			else if (lp->lwp_slptime < maxslp)
169 				totalp->t_sl++;
170 			if (lp->lwp_slptime >= maxslp)
171 				goto out;
172 			break;
173 		case LSRUN:
174 			totalp->t_rq++;
175 			if (p->p_stat == SIDL)
176 				goto out;
177 			break;
178 		default:
179 			goto out;
180 		}
181 
182 		/*
183 		 * Set while in vm_fault()
184 		 */
185 		if (lp->lwp_flags & LWP_PAGING)
186 			totalp->t_pw++;
187 	}
188 out:
189 	lwkt_reltoken(&p->p_token);
190 	return(0);
191 }
192 
193 /*
194  * No requirements.
195  */
196 static int
197 do_vmstats(SYSCTL_HANDLER_ARGS)
198 {
199 	struct vmstats vms = vmstats;
200 	return (sysctl_handle_opaque(oidp, &vms, sizeof(vms), req));
201 }
202 
203 /*
204  * No requirements.
205  */
206 static int
207 do_vmmeter(SYSCTL_HANDLER_ARGS)
208 {
209 	int boffset = offsetof(struct vmmeter, vmmeter_uint_begin);
210 	int eoffset = offsetof(struct vmmeter, vmmeter_uint_end);
211 	struct vmmeter vmm;
212 	int i;
213 
214 	bzero(&vmm, sizeof(vmm));
215 	for (i = 0; i < ncpus; ++i) {
216 		int off;
217 		struct globaldata *gd = globaldata_find(i);
218 
219 		for (off = boffset; off <= eoffset; off += sizeof(u_int)) {
220 			*(u_int *)((char *)&vmm + off) +=
221 				*(u_int *)((char *)&gd->gd_cnt + off);
222 		}
223 
224 	}
225 	vmm.v_intr += vmm.v_ipi + vmm.v_timer;
226 	return (sysctl_handle_opaque(oidp, &vmm, sizeof(vmm), req));
227 }
228 
229 /*
230  * vcnt() -	accumulate statistics from the cnt structure for each cpu
231  *
232  *	The vmmeter structure is now per-cpu as well as global.  Those
233  *	statistics which can be kept on a per-cpu basis (to avoid cache
234  *	stalls between cpus) can be moved to the per-cpu vmmeter.  Remaining
235  *	statistics, such as v_free_reserved, are left in the global
236  *	structure.
237  *
238  * (sysctl_oid *oidp, void *arg1, int arg2, struct sysctl_req *req)
239  *
240  * No requirements.
241  */
242 static int
243 vcnt(SYSCTL_HANDLER_ARGS)
244 {
245 	int i;
246 	int count = 0;
247 	int offset = arg2;
248 
249 	for (i = 0; i < ncpus; ++i) {
250 		struct globaldata *gd = globaldata_find(i);
251 		count += *(int *)((char *)&gd->gd_cnt + offset);
252 	}
253 	return(SYSCTL_OUT(req, &count, sizeof(int)));
254 }
255 
256 /*
257  * No requirements.
258  */
259 static int
260 vcnt_intr(SYSCTL_HANDLER_ARGS)
261 {
262 	int i;
263 	int count = 0;
264 
265 	for (i = 0; i < ncpus; ++i) {
266 		struct globaldata *gd = globaldata_find(i);
267 
268 		count += gd->gd_cnt.v_intr + gd->gd_cnt.v_ipi +
269 			 gd->gd_cnt.v_timer;
270 	}
271 	return(SYSCTL_OUT(req, &count, sizeof(int)));
272 }
273 
274 #define VMMETEROFF(var)	offsetof(struct vmmeter, var)
275 
276 SYSCTL_PROC(_vm, OID_AUTO, vmtotal, CTLTYPE_OPAQUE|CTLFLAG_RD,
277     0, sizeof(struct vmtotal), do_vmtotal, "S,vmtotal",
278     "System virtual memory aggregate");
279 SYSCTL_PROC(_vm, OID_AUTO, vmstats, CTLTYPE_OPAQUE|CTLFLAG_RD,
280     0, sizeof(struct vmstats), do_vmstats, "S,vmstats",
281     "System virtual memory statistics");
282 SYSCTL_PROC(_vm, OID_AUTO, vmmeter, CTLTYPE_OPAQUE|CTLFLAG_RD,
283     0, sizeof(struct vmmeter), do_vmmeter, "S,vmmeter",
284     "System statistics");
285 SYSCTL_NODE(_vm, OID_AUTO, stats, CTLFLAG_RW, 0, "VM meter stats");
286 SYSCTL_NODE(_vm_stats, OID_AUTO, sys, CTLFLAG_RW, 0, "VM meter sys stats");
287 SYSCTL_NODE(_vm_stats, OID_AUTO, vm, CTLFLAG_RW, 0, "VM meter vm stats");
288 SYSCTL_NODE(_vm_stats, OID_AUTO, misc, CTLFLAG_RW, 0, "VM meter misc stats");
289 
290 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_swtch, CTLTYPE_UINT|CTLFLAG_RD,
291 	0, VMMETEROFF(v_swtch), vcnt, "IU", "Context switches");
292 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_intrans_coll, CTLTYPE_UINT|CTLFLAG_RD,
293 	0, VMMETEROFF(v_intrans_coll), vcnt, "IU", "Intransit map collisions (total)");
294 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_intrans_wait, CTLTYPE_UINT|CTLFLAG_RD,
295 	0, VMMETEROFF(v_intrans_wait), vcnt, "IU", "Intransit map collisions which blocked");
296 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_forwarded_ints, CTLTYPE_UINT|CTLFLAG_RD,
297 	0, VMMETEROFF(v_forwarded_ints), vcnt, "IU", "Forwarded interrupts due to MP lock");
298 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_forwarded_hits, CTLTYPE_UINT|CTLFLAG_RD,
299 	0, VMMETEROFF(v_forwarded_hits), vcnt, "IU", "Forwarded hits due to MP lock");
300 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_forwarded_misses, CTLTYPE_UINT|CTLFLAG_RD,
301 	0, VMMETEROFF(v_forwarded_misses), vcnt, "IU", "Forwarded misses due to MP lock");
302 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_trap, CTLTYPE_UINT|CTLFLAG_RD,
303 	0, VMMETEROFF(v_trap), vcnt, "IU", "Traps");
304 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_syscall, CTLTYPE_UINT|CTLFLAG_RD,
305 	0, VMMETEROFF(v_syscall), vcnt, "IU", "Syscalls");
306 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_intr, CTLTYPE_UINT|CTLFLAG_RD,
307 	0, VMMETEROFF(v_intr), vcnt_intr, "IU", "Hardware interrupts");
308 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_ipi, CTLTYPE_UINT|CTLFLAG_RD,
309 	0, VMMETEROFF(v_ipi), vcnt, "IU", "Inter-processor interrupts");
310 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_timer, CTLTYPE_UINT|CTLFLAG_RD,
311 	0, VMMETEROFF(v_timer), vcnt, "IU", "LAPIC timer interrupts");
312 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_soft, CTLTYPE_UINT|CTLFLAG_RD,
313 	0, VMMETEROFF(v_soft), vcnt, "IU", "Software interrupts");
314 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vm_faults, CTLTYPE_UINT|CTLFLAG_RD,
315 	0, VMMETEROFF(v_vm_faults), vcnt, "IU", "VM faults");
316 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_cow_faults, CTLTYPE_UINT|CTLFLAG_RD,
317 	0, VMMETEROFF(v_cow_faults), vcnt, "IU", "COW faults");
318 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_cow_optim, CTLTYPE_UINT|CTLFLAG_RD,
319 	0, VMMETEROFF(v_cow_optim), vcnt, "IU", "Optimized COW faults");
320 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_zfod, CTLTYPE_UINT|CTLFLAG_RD,
321 	0, VMMETEROFF(v_zfod), vcnt, "IU", "Zero fill");
322 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_ozfod, CTLTYPE_UINT|CTLFLAG_RD,
323 	0, VMMETEROFF(v_ozfod), vcnt, "IU", "Optimized zero fill");
324 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swapin, CTLTYPE_UINT|CTLFLAG_RD,
325 	0, VMMETEROFF(v_swapin), vcnt, "IU", "Swapin operations");
326 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swapout, CTLTYPE_UINT|CTLFLAG_RD,
327 	0, VMMETEROFF(v_swapout), vcnt, "IU", "Swapout operations");
328 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swappgsin, CTLTYPE_UINT|CTLFLAG_RD,
329 	0, VMMETEROFF(v_swappgsin), vcnt, "IU", "Swapin pages");
330 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swappgsout, CTLTYPE_UINT|CTLFLAG_RD,
331 	0, VMMETEROFF(v_swappgsout), vcnt, "IU", "Swapout pages");
332 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodein, CTLTYPE_UINT|CTLFLAG_RD,
333 	0, VMMETEROFF(v_vnodein), vcnt, "IU", "Vnodein operations");
334 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodeout, CTLTYPE_UINT|CTLFLAG_RD,
335 	0, VMMETEROFF(v_vnodeout), vcnt, "IU", "Vnodeout operations");
336 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodepgsin, CTLTYPE_UINT|CTLFLAG_RD,
337 	0, VMMETEROFF(v_vnodepgsin), vcnt, "IU", "Vnodein pages");
338 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodepgsout, CTLTYPE_UINT|CTLFLAG_RD,
339 	0, VMMETEROFF(v_vnodepgsout), vcnt, "IU", "Vnodeout pages");
340 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_intrans, CTLTYPE_UINT|CTLFLAG_RD,
341 	0, VMMETEROFF(v_intrans), vcnt, "IU", "In transit page blocking");
342 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_reactivated, CTLTYPE_UINT|CTLFLAG_RD,
343 	0, VMMETEROFF(v_reactivated), vcnt, "IU", "Reactivated pages");
344 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_pdwakeups, CTLTYPE_UINT|CTLFLAG_RD,
345 	0, VMMETEROFF(v_pdwakeups), vcnt, "IU", "Pagedaemon wakeups");
346 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_ppwakeups, CTLTYPE_UINT|CTLFLAG_RD,
347 	0, VMMETEROFF(v_ppwakeups), vcnt, "IU", "vm_wait wakeups");
348 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_pdpages, CTLTYPE_UINT|CTLFLAG_RD,
349 	0, VMMETEROFF(v_pdpages), vcnt, "IU", "Pagedaemon page scans");
350 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_dfree, CTLTYPE_UINT|CTLFLAG_RD,
351 	0, VMMETEROFF(v_dfree), vcnt, "IU", "Pages freed by daemon");
352 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_pfree, CTLTYPE_UINT|CTLFLAG_RD,
353 	0, VMMETEROFF(v_pfree), vcnt, "IU", "Pages freed by exiting processes");
354 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_tfree, CTLTYPE_UINT|CTLFLAG_RD,
355 	0, VMMETEROFF(v_tfree), vcnt, "IU", "Total pages freed");
356 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_forks, CTLTYPE_UINT|CTLFLAG_RD,
357 	0, VMMETEROFF(v_forks), vcnt, "IU", "Number of fork() calls");
358 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vforks, CTLTYPE_UINT|CTLFLAG_RD,
359 	0, VMMETEROFF(v_vforks), vcnt, "IU", "Number of vfork() calls");
360 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_rforks, CTLTYPE_UINT|CTLFLAG_RD,
361 	0, VMMETEROFF(v_rforks), vcnt, "IU", "Number of rfork() calls");
362 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_kthreads, CTLTYPE_UINT|CTLFLAG_RD,
363 	0, VMMETEROFF(v_kthreads), vcnt, "IU", "Number of fork() calls by kernel");
364 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_forkpages, CTLTYPE_UINT|CTLFLAG_RD,
365 	0, VMMETEROFF(v_forkpages), vcnt, "IU", "VM pages affected by fork()");
366 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vforkpages, CTLTYPE_UINT|CTLFLAG_RD,
367 	0, VMMETEROFF(v_vforkpages), vcnt, "IU", "VM pages affected by vfork()");
368 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_rforkpages, CTLTYPE_UINT|CTLFLAG_RD,
369 	0, VMMETEROFF(v_rforkpages), vcnt, "IU", "VM pages affected by rfork()");
370 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_kthreadpages, CTLTYPE_UINT|CTLFLAG_RD,
371 	0, VMMETEROFF(v_kthreadpages), vcnt, "IU", "VM pages affected by fork() by kernel");
372 
373 SYSCTL_UINT(_vm_stats_vm, OID_AUTO,
374 	v_page_size, CTLFLAG_RD, &vmstats.v_page_size, 0,
375 	"Page size in bytes");
376 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
377 	v_page_count, CTLFLAG_RD, &vmstats.v_page_count, 0,
378 	"Total number of pages in system");
379 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
380 	v_free_reserved, CTLFLAG_RD, &vmstats.v_free_reserved, 0,
381 	"Number of pages reserved for deadlock");
382 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
383 	v_free_min, CTLFLAG_RD, &vmstats.v_free_min, 0,
384 	"Minimum number of pages desired free");
385 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
386 	v_paging_wait, CTLFLAG_RW, &vmstats.v_paging_wait, 0,
387 	"Userland slows down allocations");
388 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
389 	v_paging_start, CTLFLAG_RW, &vmstats.v_paging_start, 0,
390 	"Pageout daemon begins running");
391 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
392 	v_paging_target1, CTLFLAG_RW, &vmstats.v_paging_target1, 0,
393 	"Mid pageout daemon target");
394 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
395 	v_paging_target2, CTLFLAG_RW, &vmstats.v_paging_target2, 0,
396 	"Final pageout daemon target");
397 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
398 	v_free_count, CTLFLAG_RD, &vmstats.v_free_count, 0,
399 	"Number of pages free");
400 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
401 	v_wire_count, CTLFLAG_RD, &vmstats.v_wire_count, 0,
402 	"Number of pages wired down");
403 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
404 	v_active_count, CTLFLAG_RD, &vmstats.v_active_count, 0,
405 	"Number of pages active");
406 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
407 	v_inactive_target, CTLFLAG_RD, &vmstats.v_inactive_target, 0,
408 	"Maximum inactive pages during pageout");
409 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
410 	v_inactive_count, CTLFLAG_RD, &vmstats.v_inactive_count, 0,
411 	"Number of pages inactive");
412 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
413 	v_cache_count, CTLFLAG_RD, &vmstats.v_cache_count, 0,
414 	"Number of pages on buffer cache queue");
415 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
416 	v_pageout_free_min, CTLFLAG_RD, &vmstats.v_pageout_free_min, 0,
417 	"Min number pages reserved for kernel");
418 SYSCTL_ULONG(_vm_stats_vm, OID_AUTO,
419 	v_interrupt_free_min, CTLFLAG_RD, &vmstats.v_interrupt_free_min, 0,
420 	"Reserved number of pages for int code");
421 
422 /*
423  * No requirements.
424  */
425 static int
426 do_vmmeter_pcpu(SYSCTL_HANDLER_ARGS)
427 {
428 	int boffset = offsetof(struct vmmeter, vmmeter_uint_begin);
429 	int eoffset = offsetof(struct vmmeter, vmmeter_uint_end);
430 	struct globaldata *gd = arg1;
431 	struct vmmeter vmm;
432 	int off;
433 
434 	bzero(&vmm, sizeof(vmm));
435 	for (off = boffset; off <= eoffset; off += sizeof(u_int)) {
436 		*(u_int *)((char *)&vmm + off) +=
437 			*(u_int *)((char *)&gd->gd_cnt + off);
438 	}
439 	vmm.v_intr += vmm.v_ipi + vmm.v_timer;
440 	return (sysctl_handle_opaque(oidp, &vmm, sizeof(vmm), req));
441 }
442 
443 /*
444  * Callback for long-term slow data collection on 10-second interval.
445  *
446  * Return faults, set data for other entries.
447  */
448 #define PTOB(value)	((uint64_t)(value) << PAGE_SHIFT)
449 
450 static uint64_t
451 collect_vmstats_callback(int n)
452 {
453 	static struct vmmeter last_vmm;
454 	struct vmmeter cur_vmm;
455 	const int boffset = offsetof(struct vmmeter, vmmeter_uint_begin);
456 	const int eoffset = offsetof(struct vmmeter, vmmeter_uint_end);
457 	uint64_t total;
458 
459 	/*
460 	 * The hardclock already rolls up vmstats for us.
461 	 */
462 	kcollect_setvalue(KCOLLECT_MEMFRE, PTOB(vmstats.v_free_count));
463 	kcollect_setvalue(KCOLLECT_MEMCAC, PTOB(vmstats.v_cache_count));
464 	kcollect_setvalue(KCOLLECT_MEMINA, PTOB(vmstats.v_inactive_count));
465 	kcollect_setvalue(KCOLLECT_MEMACT, PTOB(vmstats.v_active_count));
466 	kcollect_setvalue(KCOLLECT_MEMWIR, PTOB(vmstats.v_wire_count));
467 
468 	/*
469 	 * Collect pcpu statistics for things like faults.
470 	 */
471 	bzero(&cur_vmm, sizeof(cur_vmm));
472         for (n = 0; n < ncpus; ++n) {
473                 struct globaldata *gd = globaldata_find(n);
474                 int off;
475 
476                 for (off = boffset; off <= eoffset; off += sizeof(u_int)) {
477                         *(u_int *)((char *)&cur_vmm + off) +=
478                                 *(u_int *)((char *)&gd->gd_cnt + off);
479                 }
480 
481         }
482 
483 	total = cur_vmm.v_cow_faults - last_vmm.v_cow_faults;
484 	last_vmm.v_cow_faults = cur_vmm.v_cow_faults;
485 	kcollect_setvalue(KCOLLECT_COWFAULT, total);
486 
487 	total = cur_vmm.v_zfod - last_vmm.v_zfod;
488 	last_vmm.v_zfod = cur_vmm.v_zfod;
489 	kcollect_setvalue(KCOLLECT_ZFILL, total);
490 
491 	total = cur_vmm.v_syscall - last_vmm.v_syscall;
492 	last_vmm.v_syscall = cur_vmm.v_syscall;
493 	kcollect_setvalue(KCOLLECT_SYSCALLS, total);
494 
495 	total = cur_vmm.v_intr - last_vmm.v_intr;
496 	last_vmm.v_intr = cur_vmm.v_intr;
497 	kcollect_setvalue(KCOLLECT_INTR, total);
498 
499 	total = cur_vmm.v_ipi - last_vmm.v_ipi;
500 	last_vmm.v_ipi = cur_vmm.v_ipi;
501 	kcollect_setvalue(KCOLLECT_IPI, total);
502 
503 	total = cur_vmm.v_timer - last_vmm.v_timer;
504 	last_vmm.v_timer = cur_vmm.v_timer;
505 	kcollect_setvalue(KCOLLECT_TIMER, total);
506 
507 	total = cur_vmm.v_vm_faults - last_vmm.v_vm_faults;
508 	last_vmm.v_vm_faults = cur_vmm.v_vm_faults;
509 
510 	return total;
511 }
512 
513 /*
514  * Called from the low level boot code only.
515  */
516 static void
517 vmmeter_init(void *dummy __unused)
518 {
519 	int i;
520 
521 	for (i = 0; i < ncpus; ++i) {
522 		struct sysctl_ctx_list *ctx;
523 		struct sysctl_oid *oid;
524 		struct globaldata *gd;
525 		char name[32];
526 
527 		ksnprintf(name, sizeof(name), "cpu%d", i);
528 
529 		ctx = kmalloc(sizeof(*ctx), M_TEMP, M_WAITOK);
530 		sysctl_ctx_init(ctx);
531 		oid = SYSCTL_ADD_NODE(ctx, SYSCTL_STATIC_CHILDREN(_vm),
532 				      OID_AUTO, name, CTLFLAG_RD, 0, "");
533 
534 		gd = globaldata_find(i);
535 		SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
536 				"vmmeter", CTLTYPE_OPAQUE|CTLFLAG_RD,
537 				gd, sizeof(struct vmmeter), do_vmmeter_pcpu,
538 				"S,vmmeter", "System per-cpu statistics");
539 	}
540 	kcollect_register(KCOLLECT_VMFAULT, "fault", collect_vmstats_callback,
541 			  KCOLLECT_SCALE(KCOLLECT_VMFAULT_FORMAT, 0));
542 	kcollect_register(KCOLLECT_COWFAULT, "cow", NULL,
543 			  KCOLLECT_SCALE(KCOLLECT_COWFAULT_FORMAT, 0));
544 	kcollect_register(KCOLLECT_ZFILL, "zfill", NULL,
545 			  KCOLLECT_SCALE(KCOLLECT_ZFILL_FORMAT, 0));
546 
547 	kcollect_register(KCOLLECT_MEMFRE, "free", NULL,
548 			  KCOLLECT_SCALE(KCOLLECT_MEMFRE_FORMAT,
549 					 PTOB(vmstats.v_page_count)));
550 	kcollect_register(KCOLLECT_MEMCAC, "cache", NULL,
551 			  KCOLLECT_SCALE(KCOLLECT_MEMCAC_FORMAT,
552 					 PTOB(vmstats.v_page_count)));
553 	kcollect_register(KCOLLECT_MEMINA, "inact", NULL,
554 			  KCOLLECT_SCALE(KCOLLECT_MEMINA_FORMAT,
555 					 PTOB(vmstats.v_page_count)));
556 	kcollect_register(KCOLLECT_MEMACT, "act", NULL,
557 			  KCOLLECT_SCALE(KCOLLECT_MEMACT_FORMAT,
558 					 PTOB(vmstats.v_page_count)));
559 	kcollect_register(KCOLLECT_MEMWIR, "wired", NULL,
560 			  KCOLLECT_SCALE(KCOLLECT_MEMWIR_FORMAT,
561 					 PTOB(vmstats.v_page_count)));
562 
563 	kcollect_register(KCOLLECT_SYSCALLS, "syscalls", NULL,
564 			  KCOLLECT_SCALE(KCOLLECT_SYSCALLS_FORMAT, 0));
565 
566 	kcollect_register(KCOLLECT_INTR, "intr", NULL,
567 			  KCOLLECT_SCALE(KCOLLECT_INTR_FORMAT, 0));
568 	kcollect_register(KCOLLECT_IPI, "ipi", NULL,
569 			  KCOLLECT_SCALE(KCOLLECT_IPI_FORMAT, 0));
570 	kcollect_register(KCOLLECT_TIMER, "timer", NULL,
571 			  KCOLLECT_SCALE(KCOLLECT_TIMER_FORMAT, 0));
572 }
573 SYSINIT(vmmeter, SI_SUB_PSEUDO, SI_ORDER_ANY, vmmeter_init, 0);
574 
575 /*
576  * Rolls up accumulated pcpu adjustments to vmstats counts into the global
577  * structure, copy the global structure into our pcpu structure.  Critical
578  * path checks will use our pcpu structure.
579  *
580  * This is somewhat expensive and only called when needed, and by the
581  * hardclock.
582  */
583 void
584 vmstats_rollup(void)
585 {
586 	int cpu;
587 
588 	for (cpu = 0; cpu < ncpus; ++cpu) {
589 		vmstats_rollup_cpu(globaldata_find(cpu));
590 	}
591 	mycpu->gd_vmstats = vmstats;
592 }
593 
594 void
595 vmstats_rollup_cpu(globaldata_t gd)
596 {
597 	long value;
598 
599 	if (gd->gd_vmstats_adj.v_free_count) {
600 		value = atomic_swap_long(&gd->gd_vmstats_adj.v_free_count, 0);
601 		atomic_add_long(&vmstats.v_free_count, value);
602 	}
603 	if (gd->gd_vmstats_adj.v_cache_count) {
604 		value = atomic_swap_long(&gd->gd_vmstats_adj.v_cache_count, 0);
605 		atomic_add_long(&vmstats.v_cache_count, value);
606 	}
607 	if (gd->gd_vmstats_adj.v_inactive_count) {
608 		value=atomic_swap_long(&gd->gd_vmstats_adj.v_inactive_count, 0);
609 		atomic_add_long(&vmstats.v_inactive_count, value);
610 	}
611 	if (gd->gd_vmstats_adj.v_active_count) {
612 		value = atomic_swap_long(&gd->gd_vmstats_adj.v_active_count, 0);
613 		atomic_add_long(&vmstats.v_active_count, value);
614 	}
615 	if (gd->gd_vmstats_adj.v_wire_count) {
616 		value = atomic_swap_long(&gd->gd_vmstats_adj.v_wire_count, 0);
617 		atomic_add_long(&vmstats.v_wire_count, value);
618 	}
619 }
620