xref: /freebsd/sys/vm/vm_meter.c (revision 168fce73b59d6023cab45d063a452551a1f2103e)
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  * 4. 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 <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/proc.h>
41 #include <sys/resource.h>
42 #include <sys/rwlock.h>
43 #include <sys/sx.h>
44 #include <sys/vmmeter.h>
45 #include <sys/smp.h>
46 
47 #include <vm/vm.h>
48 #include <vm/vm_page.h>
49 #include <vm/vm_extern.h>
50 #include <vm/vm_param.h>
51 #include <vm/pmap.h>
52 #include <vm/vm_map.h>
53 #include <vm/vm_object.h>
54 #include <sys/sysctl.h>
55 
56 struct vmmeter vm_cnt;
57 
58 SYSCTL_UINT(_vm, VM_V_FREE_MIN, v_free_min,
59 	CTLFLAG_RW, &vm_cnt.v_free_min, 0, "Minimum low-free-pages threshold");
60 SYSCTL_UINT(_vm, VM_V_FREE_TARGET, v_free_target,
61 	CTLFLAG_RW, &vm_cnt.v_free_target, 0, "Desired free pages");
62 SYSCTL_UINT(_vm, VM_V_FREE_RESERVED, v_free_reserved,
63 	CTLFLAG_RW, &vm_cnt.v_free_reserved, 0, "Pages reserved for deadlock");
64 SYSCTL_UINT(_vm, VM_V_INACTIVE_TARGET, v_inactive_target,
65 	CTLFLAG_RW, &vm_cnt.v_inactive_target, 0, "Pages desired inactive");
66 SYSCTL_UINT(_vm, VM_V_PAGEOUT_FREE_MIN, v_pageout_free_min,
67 	CTLFLAG_RW, &vm_cnt.v_pageout_free_min, 0, "Min pages reserved for kernel");
68 SYSCTL_UINT(_vm, OID_AUTO, v_free_severe,
69 	CTLFLAG_RW, &vm_cnt.v_free_severe, 0, "Severe page depletion point");
70 
71 static int
72 sysctl_vm_loadavg(SYSCTL_HANDLER_ARGS)
73 {
74 
75 #ifdef SCTL_MASK32
76 	u_int32_t la[4];
77 
78 	if (req->flags & SCTL_MASK32) {
79 		la[0] = averunnable.ldavg[0];
80 		la[1] = averunnable.ldavg[1];
81 		la[2] = averunnable.ldavg[2];
82 		la[3] = averunnable.fscale;
83 		return SYSCTL_OUT(req, la, sizeof(la));
84 	} else
85 #endif
86 		return SYSCTL_OUT(req, &averunnable, sizeof(averunnable));
87 }
88 SYSCTL_PROC(_vm, VM_LOADAVG, loadavg, CTLTYPE_STRUCT | CTLFLAG_RD |
89     CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_loadavg, "S,loadavg",
90     "Machine loadaverage history");
91 
92 /*
93  * This function aims to determine if the object is mapped,
94  * specifically, if it is referenced by a vm_map_entry.  Because
95  * objects occasionally acquire transient references that do not
96  * represent a mapping, the method used here is inexact.  However, it
97  * has very low overhead and is good enough for the advisory
98  * vm.vmtotal sysctl.
99  */
100 static bool
101 is_object_active(vm_object_t obj)
102 {
103 
104 	return (obj->ref_count > obj->shadow_count);
105 }
106 
107 static int
108 vmtotal(SYSCTL_HANDLER_ARGS)
109 {
110 	struct vmtotal total;
111 	vm_object_t object;
112 	struct proc *p;
113 	struct thread *td;
114 
115 	bzero(&total, sizeof(total));
116 
117 	/*
118 	 * Calculate process statistics.
119 	 */
120 	sx_slock(&allproc_lock);
121 	FOREACH_PROC_IN_SYSTEM(p) {
122 		if (p->p_flag & P_SYSTEM)
123 			continue;
124 		PROC_LOCK(p);
125 		switch (p->p_state) {
126 		case PRS_NEW:
127 			PROC_UNLOCK(p);
128 			continue;
129 			break;
130 		default:
131 			FOREACH_THREAD_IN_PROC(p, td) {
132 				thread_lock(td);
133 				switch (td->td_state) {
134 				case TDS_INHIBITED:
135 					if (TD_IS_SWAPPED(td))
136 						total.t_sw++;
137 					else if (TD_IS_SLEEPING(td)) {
138 						if (td->td_priority <= PZERO)
139 							total.t_dw++;
140 						else
141 							total.t_sl++;
142 						if (td->td_wchan ==
143 						    &vm_cnt.v_free_count)
144 							total.t_pw++;
145 					}
146 					break;
147 
148 				case TDS_CAN_RUN:
149 					total.t_sw++;
150 					break;
151 				case TDS_RUNQ:
152 				case TDS_RUNNING:
153 					total.t_rq++;
154 					thread_unlock(td);
155 					continue;
156 				default:
157 					break;
158 				}
159 				thread_unlock(td);
160 			}
161 		}
162 		PROC_UNLOCK(p);
163 	}
164 	sx_sunlock(&allproc_lock);
165 	/*
166 	 * Calculate object memory usage statistics.
167 	 */
168 	mtx_lock(&vm_object_list_mtx);
169 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
170 		/*
171 		 * Perform unsynchronized reads on the object.  In
172 		 * this case, the lack of synchronization should not
173 		 * impair the accuracy of the reported statistics.
174 		 */
175 		if ((object->flags & OBJ_FICTITIOUS) != 0) {
176 			/*
177 			 * Devices, like /dev/mem, will badly skew our totals.
178 			 */
179 			continue;
180 		}
181 		if (object->ref_count == 0) {
182 			/*
183 			 * Also skip unreferenced objects, including
184 			 * vnodes representing mounted file systems.
185 			 */
186 			continue;
187 		}
188 		if (object->ref_count == 1 &&
189 		    (object->flags & OBJ_NOSPLIT) != 0) {
190 			/*
191 			 * Also skip otherwise unreferenced swap
192 			 * objects backing tmpfs vnodes, and POSIX or
193 			 * SysV shared memory.
194 			 */
195 			continue;
196 		}
197 		total.t_vm += object->size;
198 		total.t_rm += object->resident_page_count;
199 		if (is_object_active(object)) {
200 			total.t_avm += object->size;
201 			total.t_arm += object->resident_page_count;
202 		}
203 		if (object->shadow_count > 1) {
204 			/* shared object */
205 			total.t_vmshr += object->size;
206 			total.t_rmshr += object->resident_page_count;
207 			if (is_object_active(object)) {
208 				total.t_avmshr += object->size;
209 				total.t_armshr += object->resident_page_count;
210 			}
211 		}
212 	}
213 	mtx_unlock(&vm_object_list_mtx);
214 	total.t_free = vm_cnt.v_free_count + vm_cnt.v_cache_count;
215 	return (sysctl_handle_opaque(oidp, &total, sizeof(total), req));
216 }
217 
218 /*
219  * vm_meter_cnt() -	accumulate statistics from all cpus and the global cnt
220  *			structure.
221  *
222  *	The vmmeter structure is now per-cpu as well as global.  Those
223  *	statistics which can be kept on a per-cpu basis (to avoid cache
224  *	stalls between cpus) can be moved to the per-cpu vmmeter.  Remaining
225  *	statistics, such as v_free_reserved, are left in the global
226  *	structure.
227  */
228 u_int
229 vm_meter_cnt(size_t offset)
230 {
231 	struct pcpu *pcpu;
232 	u_int count;
233 	int i;
234 
235 	count = *(u_int *)((char *)&vm_cnt + offset);
236 	CPU_FOREACH(i) {
237 		pcpu = pcpu_find(i);
238 		count += *(u_int *)((char *)&pcpu->pc_cnt + offset);
239 	}
240 	return (count);
241 }
242 
243 static int
244 cnt_sysctl(SYSCTL_HANDLER_ARGS)
245 {
246 	u_int count;
247 
248 	count = vm_meter_cnt((char *)arg1 - (char *)&vm_cnt);
249 	return (SYSCTL_OUT(req, &count, sizeof(count)));
250 }
251 
252 SYSCTL_PROC(_vm, VM_TOTAL, vmtotal, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
253     0, sizeof(struct vmtotal), vmtotal, "S,vmtotal",
254     "System virtual memory statistics");
255 SYSCTL_NODE(_vm, OID_AUTO, stats, CTLFLAG_RW, 0, "VM meter stats");
256 static SYSCTL_NODE(_vm_stats, OID_AUTO, sys, CTLFLAG_RW, 0,
257 	"VM meter sys stats");
258 static SYSCTL_NODE(_vm_stats, OID_AUTO, vm, CTLFLAG_RW, 0,
259 	"VM meter vm stats");
260 SYSCTL_NODE(_vm_stats, OID_AUTO, misc, CTLFLAG_RW, 0, "VM meter misc stats");
261 
262 #define	VM_STATS(parent, var, descr) \
263 	SYSCTL_PROC(parent, OID_AUTO, var, \
264 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, &vm_cnt.var, 0,	\
265 	    cnt_sysctl, "IU", descr)
266 #define	VM_STATS_VM(var, descr)		VM_STATS(_vm_stats_vm, var, descr)
267 #define	VM_STATS_SYS(var, descr)	VM_STATS(_vm_stats_sys, var, descr)
268 
269 VM_STATS_SYS(v_swtch, "Context switches");
270 VM_STATS_SYS(v_trap, "Traps");
271 VM_STATS_SYS(v_syscall, "System calls");
272 VM_STATS_SYS(v_intr, "Device interrupts");
273 VM_STATS_SYS(v_soft, "Software interrupts");
274 VM_STATS_VM(v_vm_faults, "Address memory faults");
275 VM_STATS_VM(v_io_faults, "Page faults requiring I/O");
276 VM_STATS_VM(v_cow_faults, "Copy-on-write faults");
277 VM_STATS_VM(v_cow_optim, "Optimized COW faults");
278 VM_STATS_VM(v_zfod, "Pages zero-filled on demand");
279 VM_STATS_VM(v_ozfod, "Optimized zero fill pages");
280 VM_STATS_VM(v_swapin, "Swap pager pageins");
281 VM_STATS_VM(v_swapout, "Swap pager pageouts");
282 VM_STATS_VM(v_swappgsin, "Swap pages swapped in");
283 VM_STATS_VM(v_swappgsout, "Swap pages swapped out");
284 VM_STATS_VM(v_vnodein, "Vnode pager pageins");
285 VM_STATS_VM(v_vnodeout, "Vnode pager pageouts");
286 VM_STATS_VM(v_vnodepgsin, "Vnode pages paged in");
287 VM_STATS_VM(v_vnodepgsout, "Vnode pages paged out");
288 VM_STATS_VM(v_intrans, "In transit page faults");
289 VM_STATS_VM(v_reactivated, "Pages reactivated by pagedaemon");
290 VM_STATS_VM(v_pdwakeups, "Pagedaemon wakeups");
291 VM_STATS_VM(v_pdpages, "Pages analyzed by pagedaemon");
292 VM_STATS_VM(v_pdshortfalls, "Page reclamation shortfalls");
293 VM_STATS_VM(v_tcached, "Total pages cached");
294 VM_STATS_VM(v_dfree, "Pages freed by pagedaemon");
295 VM_STATS_VM(v_pfree, "Pages freed by exiting processes");
296 VM_STATS_VM(v_tfree, "Total pages freed");
297 VM_STATS_VM(v_page_size, "Page size in bytes");
298 VM_STATS_VM(v_page_count, "Total number of pages in system");
299 VM_STATS_VM(v_free_reserved, "Pages reserved for deadlock");
300 VM_STATS_VM(v_free_target, "Pages desired free");
301 VM_STATS_VM(v_free_min, "Minimum low-free-pages threshold");
302 VM_STATS_VM(v_free_count, "Free pages");
303 VM_STATS_VM(v_wire_count, "Wired pages");
304 VM_STATS_VM(v_active_count, "Active pages");
305 VM_STATS_VM(v_inactive_target, "Desired inactive pages");
306 VM_STATS_VM(v_inactive_count, "Inactive pages");
307 VM_STATS_VM(v_laundry_count, "Pages eligible for laundering");
308 VM_STATS_VM(v_cache_count, "Pages on cache queue");
309 VM_STATS_VM(v_pageout_free_min, "Min pages reserved for kernel");
310 VM_STATS_VM(v_interrupt_free_min, "Reserved pages for interrupt code");
311 VM_STATS_VM(v_forks, "Number of fork() calls");
312 VM_STATS_VM(v_vforks, "Number of vfork() calls");
313 VM_STATS_VM(v_rforks, "Number of rfork() calls");
314 VM_STATS_VM(v_kthreads, "Number of fork() calls by kernel");
315 VM_STATS_VM(v_forkpages, "VM pages affected by fork()");
316 VM_STATS_VM(v_vforkpages, "VM pages affected by vfork()");
317 VM_STATS_VM(v_rforkpages, "VM pages affected by rfork()");
318 VM_STATS_VM(v_kthreadpages, "VM pages affected by fork() by kernel");
319