xref: /freebsd/sys/vm/vm_meter.c (revision b2ea244070ff84eab79e04befb7aa30c982fc84d)
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 __exclusive_cache_line 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 #if defined(COMPAT_FREEBSD11)
156 struct vmtotal11 {
157 	int16_t	t_rq;
158 	int16_t	t_dw;
159 	int16_t	t_pw;
160 	int16_t	t_sl;
161 	int16_t	t_sw;
162 	int32_t	t_vm;
163 	int32_t	t_avm;
164 	int32_t	t_rm;
165 	int32_t	t_arm;
166 	int32_t	t_vmshr;
167 	int32_t	t_avmshr;
168 	int32_t	t_rmshr;
169 	int32_t	t_armshr;
170 	int32_t	t_free;
171 };
172 #endif
173 
174 static int
175 vmtotal(SYSCTL_HANDLER_ARGS)
176 {
177 	struct vmtotal total;
178 #if defined(COMPAT_FREEBSD11)
179 	struct vmtotal11 total11;
180 #endif
181 	vm_object_t object;
182 	struct proc *p;
183 	struct thread *td;
184 
185 	if (req->oldptr == NULL) {
186 #if defined(COMPAT_FREEBSD11)
187 		if (curproc->p_osrel < P_OSREL_VMTOTAL64)
188 			return (SYSCTL_OUT(req, NULL, sizeof(total11)));
189 #endif
190 		return (SYSCTL_OUT(req, NULL, sizeof(total)));
191 	}
192 	bzero(&total, sizeof(total));
193 
194 	/*
195 	 * Calculate process statistics.
196 	 */
197 	sx_slock(&allproc_lock);
198 	FOREACH_PROC_IN_SYSTEM(p) {
199 		if ((p->p_flag & P_SYSTEM) != 0)
200 			continue;
201 		PROC_LOCK(p);
202 		if (p->p_state != PRS_NEW) {
203 			FOREACH_THREAD_IN_PROC(p, td) {
204 				thread_lock(td);
205 				switch (td->td_state) {
206 				case TDS_INHIBITED:
207 					if (TD_IS_SWAPPED(td))
208 						total.t_sw++;
209 					else if (TD_IS_SLEEPING(td)) {
210 						if (td->td_priority <= PZERO)
211 							total.t_dw++;
212 						else
213 							total.t_sl++;
214 						if (td->td_wchan ==
215 						    &vm_cnt.v_free_count)
216 							total.t_pw++;
217 					}
218 					break;
219 				case TDS_CAN_RUN:
220 					total.t_sw++;
221 					break;
222 				case TDS_RUNQ:
223 				case TDS_RUNNING:
224 					total.t_rq++;
225 					break;
226 				default:
227 					break;
228 				}
229 				thread_unlock(td);
230 			}
231 		}
232 		PROC_UNLOCK(p);
233 	}
234 	sx_sunlock(&allproc_lock);
235 	/*
236 	 * Calculate object memory usage statistics.
237 	 */
238 	mtx_lock(&vm_object_list_mtx);
239 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
240 		/*
241 		 * Perform unsynchronized reads on the object.  In
242 		 * this case, the lack of synchronization should not
243 		 * impair the accuracy of the reported statistics.
244 		 */
245 		if ((object->flags & OBJ_FICTITIOUS) != 0) {
246 			/*
247 			 * Devices, like /dev/mem, will badly skew our totals.
248 			 */
249 			continue;
250 		}
251 		if (object->ref_count == 0) {
252 			/*
253 			 * Also skip unreferenced objects, including
254 			 * vnodes representing mounted file systems.
255 			 */
256 			continue;
257 		}
258 		if (object->ref_count == 1 &&
259 		    (object->flags & OBJ_NOSPLIT) != 0) {
260 			/*
261 			 * Also skip otherwise unreferenced swap
262 			 * objects backing tmpfs vnodes, and POSIX or
263 			 * SysV shared memory.
264 			 */
265 			continue;
266 		}
267 		total.t_vm += object->size;
268 		total.t_rm += object->resident_page_count;
269 		if (is_object_active(object)) {
270 			total.t_avm += object->size;
271 			total.t_arm += object->resident_page_count;
272 		}
273 		if (object->shadow_count > 1) {
274 			/* shared object */
275 			total.t_vmshr += object->size;
276 			total.t_rmshr += object->resident_page_count;
277 			if (is_object_active(object)) {
278 				total.t_avmshr += object->size;
279 				total.t_armshr += object->resident_page_count;
280 			}
281 		}
282 	}
283 	mtx_unlock(&vm_object_list_mtx);
284 	total.t_free = vm_cnt.v_free_count;
285 #if defined(COMPAT_FREEBSD11)
286 	/* sysctl(8) allocates twice as much memory as reported by sysctl(3) */
287 	if (curproc->p_osrel < P_OSREL_VMTOTAL64 && (req->oldlen ==
288 	    sizeof(total11) || req->oldlen == 2 * sizeof(total11))) {
289 		bzero(&total11, sizeof(total11));
290 		total11.t_rq = total.t_rq;
291 		total11.t_dw = total.t_dw;
292 		total11.t_pw = total.t_pw;
293 		total11.t_sl = total.t_sl;
294 		total11.t_sw = total.t_sw;
295 		total11.t_vm = total.t_vm;	/* truncate */
296 		total11.t_avm = total.t_avm;	/* truncate */
297 		total11.t_rm = total.t_rm;	/* truncate */
298 		total11.t_arm = total.t_arm;	/* truncate */
299 		total11.t_vmshr = total.t_vmshr;	/* truncate */
300 		total11.t_avmshr = total.t_avmshr;	/* truncate */
301 		total11.t_rmshr = total.t_rmshr;	/* truncate */
302 		total11.t_armshr = total.t_armshr;	/* truncate */
303 		total11.t_free = total.t_free;		/* truncate */
304 		return (SYSCTL_OUT(req, &total11, sizeof(total11)));
305 	}
306 #endif
307 	return (SYSCTL_OUT(req, &total, sizeof(total)));
308 }
309 
310 SYSCTL_PROC(_vm, VM_TOTAL, vmtotal, CTLTYPE_OPAQUE | CTLFLAG_RD |
311     CTLFLAG_MPSAFE, NULL, 0, vmtotal, "S,vmtotal",
312     "System virtual memory statistics");
313 SYSCTL_NODE(_vm, OID_AUTO, stats, CTLFLAG_RW, 0, "VM meter stats");
314 static SYSCTL_NODE(_vm_stats, OID_AUTO, sys, CTLFLAG_RW, 0,
315 	"VM meter sys stats");
316 static SYSCTL_NODE(_vm_stats, OID_AUTO, vm, CTLFLAG_RW, 0,
317 	"VM meter vm stats");
318 SYSCTL_NODE(_vm_stats, OID_AUTO, misc, CTLFLAG_RW, 0, "VM meter misc stats");
319 
320 static int
321 sysctl_handle_vmstat(SYSCTL_HANDLER_ARGS)
322 {
323 	uint64_t val;
324 #ifdef COMPAT_FREEBSD11
325 	uint32_t val32;
326 #endif
327 
328 	val = counter_u64_fetch(*(counter_u64_t *)arg1);
329 #ifdef COMPAT_FREEBSD11
330 	if (req->oldlen == sizeof(val32)) {
331 		val32 = val;		/* truncate */
332 		return (SYSCTL_OUT(req, &val32, sizeof(val32)));
333 	}
334 #endif
335 	return (SYSCTL_OUT(req, &val, sizeof(val)));
336 }
337 
338 #define	VM_STATS(parent, var, descr) \
339     SYSCTL_OID(parent, OID_AUTO, var, CTLTYPE_U64 | CTLFLAG_MPSAFE | \
340     CTLFLAG_RD, &vm_cnt.var, 0, sysctl_handle_vmstat, "QU", descr);
341 #define	VM_STATS_VM(var, descr)		VM_STATS(_vm_stats_vm, var, descr)
342 #define	VM_STATS_SYS(var, descr)	VM_STATS(_vm_stats_sys, var, descr)
343 
344 VM_STATS_SYS(v_swtch, "Context switches");
345 VM_STATS_SYS(v_trap, "Traps");
346 VM_STATS_SYS(v_syscall, "System calls");
347 VM_STATS_SYS(v_intr, "Device interrupts");
348 VM_STATS_SYS(v_soft, "Software interrupts");
349 VM_STATS_VM(v_vm_faults, "Address memory faults");
350 VM_STATS_VM(v_io_faults, "Page faults requiring I/O");
351 VM_STATS_VM(v_cow_faults, "Copy-on-write faults");
352 VM_STATS_VM(v_cow_optim, "Optimized COW faults");
353 VM_STATS_VM(v_zfod, "Pages zero-filled on demand");
354 VM_STATS_VM(v_ozfod, "Optimized zero fill pages");
355 VM_STATS_VM(v_swapin, "Swap pager pageins");
356 VM_STATS_VM(v_swapout, "Swap pager pageouts");
357 VM_STATS_VM(v_swappgsin, "Swap pages swapped in");
358 VM_STATS_VM(v_swappgsout, "Swap pages swapped out");
359 VM_STATS_VM(v_vnodein, "Vnode pager pageins");
360 VM_STATS_VM(v_vnodeout, "Vnode pager pageouts");
361 VM_STATS_VM(v_vnodepgsin, "Vnode pages paged in");
362 VM_STATS_VM(v_vnodepgsout, "Vnode pages paged out");
363 VM_STATS_VM(v_intrans, "In transit page faults");
364 VM_STATS_VM(v_reactivated, "Pages reactivated by pagedaemon");
365 VM_STATS_VM(v_pdwakeups, "Pagedaemon wakeups");
366 VM_STATS_VM(v_pdpages, "Pages analyzed by pagedaemon");
367 VM_STATS_VM(v_pdshortfalls, "Page reclamation shortfalls");
368 VM_STATS_VM(v_dfree, "Pages freed by pagedaemon");
369 VM_STATS_VM(v_pfree, "Pages freed by exiting processes");
370 VM_STATS_VM(v_tfree, "Total pages freed");
371 VM_STATS_VM(v_forks, "Number of fork() calls");
372 VM_STATS_VM(v_vforks, "Number of vfork() calls");
373 VM_STATS_VM(v_rforks, "Number of rfork() calls");
374 VM_STATS_VM(v_kthreads, "Number of fork() calls by kernel");
375 VM_STATS_VM(v_forkpages, "VM pages affected by fork()");
376 VM_STATS_VM(v_vforkpages, "VM pages affected by vfork()");
377 VM_STATS_VM(v_rforkpages, "VM pages affected by rfork()");
378 VM_STATS_VM(v_kthreadpages, "VM pages affected by fork() by kernel");
379 
380 #define	VM_STATS_UINT(var, descr)	\
381     SYSCTL_UINT(_vm_stats_vm, OID_AUTO, var, CTLFLAG_RD, &vm_cnt.var, 0, descr)
382 VM_STATS_UINT(v_page_size, "Page size in bytes");
383 VM_STATS_UINT(v_page_count, "Total number of pages in system");
384 VM_STATS_UINT(v_free_reserved, "Pages reserved for deadlock");
385 VM_STATS_UINT(v_free_target, "Pages desired free");
386 VM_STATS_UINT(v_free_min, "Minimum low-free-pages threshold");
387 VM_STATS_UINT(v_free_count, "Free pages");
388 VM_STATS_UINT(v_wire_count, "Wired pages");
389 VM_STATS_UINT(v_active_count, "Active pages");
390 VM_STATS_UINT(v_inactive_target, "Desired inactive pages");
391 VM_STATS_UINT(v_inactive_count, "Inactive pages");
392 VM_STATS_UINT(v_laundry_count, "Pages eligible for laundering");
393 VM_STATS_UINT(v_pageout_free_min, "Min pages reserved for kernel");
394 VM_STATS_UINT(v_interrupt_free_min, "Reserved pages for interrupt code");
395 VM_STATS_UINT(v_free_severe, "Severe page depletion point");
396 
397 #ifdef COMPAT_FREEBSD11
398 /*
399  * Provide compatibility sysctls for the benefit of old utilities which exit
400  * with an error if they cannot be found.
401  */
402 SYSCTL_UINT(_vm_stats_vm, OID_AUTO, v_cache_count, CTLFLAG_RD,
403     SYSCTL_NULL_UINT_PTR, 0, "Dummy for compatibility");
404 SYSCTL_UINT(_vm_stats_vm, OID_AUTO, v_tcached, CTLFLAG_RD,
405     SYSCTL_NULL_UINT_PTR, 0, "Dummy for compatibility");
406 #endif
407