xref: /netbsd-src/sys/uvm/uvm_meter.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: uvm_meter.c,v 1.47 2007/02/26 09:20:54 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
5  * Copyright (c) 1982, 1986, 1989, 1993
6  *      The Regents of the University of California.
7  *
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *      This product includes software developed by Charles D. Cranor,
21  *      Washington University, and the University of California, Berkeley
22  *      and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  *      @(#)vm_meter.c  8.4 (Berkeley) 1/4/94
40  * from: Id: uvm_meter.c,v 1.1.2.1 1997/08/14 19:10:35 chuck Exp
41  */
42 
43 #include <sys/cdefs.h>
44 __KERNEL_RCSID(0, "$NetBSD: uvm_meter.c,v 1.47 2007/02/26 09:20:54 yamt Exp $");
45 
46 #include <sys/param.h>
47 #include <sys/proc.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 
52 #include <uvm/uvm_extern.h>
53 #include <uvm/uvm_pdpolicy.h>
54 
55 /*
56  * maxslp: ???? XXXCDC
57  */
58 
59 int maxslp = MAXSLP;	/* patchable ... */
60 struct loadavg averunnable;
61 
62 /*
63  * constants for averages over 1, 5, and 15 minutes when sampling at
64  * 5 second intervals.
65  */
66 
67 static const fixpt_t cexp[3] = {
68 	0.9200444146293232 * FSCALE,	/* exp(-1/12) */
69 	0.9834714538216174 * FSCALE,	/* exp(-1/60) */
70 	0.9944598480048967 * FSCALE,	/* exp(-1/180) */
71 };
72 
73 /*
74  * prototypes
75  */
76 
77 static void uvm_loadav(struct loadavg *);
78 static void uvm_total(struct vmtotal *);
79 
80 /*
81  * uvm_meter: calculate load average and wake up the swapper (if needed)
82  */
83 void
84 uvm_meter(void)
85 {
86 	static int count;
87 
88 	if (++count >= 5) {
89 		count = 0;
90 		uvm_loadav(&averunnable);
91 	}
92 	if (lwp0.l_slptime > (maxslp / 2))
93 		uvm_kick_scheduler();
94 }
95 
96 /*
97  * uvm_loadav: compute a tenex style load average of a quantity on
98  * 1, 5, and 15 minute intervals.
99  */
100 static void
101 uvm_loadav(struct loadavg *avg)
102 {
103 	int i, nrun;
104 	struct lwp *l;
105 
106 	nrun = 0;
107 
108 	mutex_enter(&proclist_mutex);
109 	LIST_FOREACH(l, &alllwp, l_list) {
110 		if ((l->l_flag & (LW_SINTR | LW_SYSTEM)) != 0)
111 			continue;
112 		switch (l->l_stat) {
113 		case LSSLEEP:
114 			if (l->l_slptime > 1)
115 				continue;
116 		/* fall through */
117 		case LSRUN:
118 		case LSONPROC:
119 		case LSIDL:
120 			nrun++;
121 		}
122 	}
123 	mutex_exit(&proclist_mutex);
124 
125 	for (i = 0; i < 3; i++)
126 		avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
127 		    nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
128 }
129 
130 /*
131  * sysctl helper routine for the vm.vmmeter node.
132  */
133 static int
134 sysctl_vm_meter(SYSCTLFN_ARGS)
135 {
136 	struct sysctlnode node;
137 	struct vmtotal vmtotals;
138 
139 	node = *rnode;
140 	node.sysctl_data = &vmtotals;
141 	uvm_total(&vmtotals);
142 
143 	return (sysctl_lookup(SYSCTLFN_CALL(&node)));
144 }
145 
146 /*
147  * sysctl helper routine for the vm.uvmexp node.
148  */
149 static int
150 sysctl_vm_uvmexp(SYSCTLFN_ARGS)
151 {
152 	struct sysctlnode node;
153 
154 	node = *rnode;
155 	if (oldp)
156 		node.sysctl_size = min(*oldlenp, node.sysctl_size);
157 
158 	return (sysctl_lookup(SYSCTLFN_CALL(&node)));
159 }
160 
161 static int
162 sysctl_vm_uvmexp2(SYSCTLFN_ARGS)
163 {
164 	struct sysctlnode node;
165 	struct uvmexp_sysctl u;
166 	int active, inactive;
167 
168 	uvm_estimatepageable(&active, &inactive);
169 
170 	memset(&u, 0, sizeof(u));
171 
172 	/* Entries here are in order of uvmexp_sysctl, not uvmexp */
173 	u.pagesize = uvmexp.pagesize;
174 	u.pagemask = uvmexp.pagemask;
175 	u.pageshift = uvmexp.pageshift;
176 	u.npages = uvmexp.npages;
177 	u.free = uvmexp.free;
178 	u.active = active;
179 	u.inactive = inactive;
180 	u.paging = uvmexp.paging;
181 	u.wired = uvmexp.wired;
182 	u.zeropages = uvmexp.zeropages;
183 	u.reserve_pagedaemon = uvmexp.reserve_pagedaemon;
184 	u.reserve_kernel = uvmexp.reserve_kernel;
185 	u.freemin = uvmexp.freemin;
186 	u.freetarg = uvmexp.freetarg;
187 	u.inactarg = 0; /* unused */
188 	u.wiredmax = uvmexp.wiredmax;
189 	u.nswapdev = uvmexp.nswapdev;
190 	u.swpages = uvmexp.swpages;
191 	u.swpginuse = uvmexp.swpginuse;
192 	u.swpgonly = uvmexp.swpgonly;
193 	u.nswget = uvmexp.nswget;
194 	u.faults = uvmexp.faults;
195 	u.traps = uvmexp.traps;
196 	u.intrs = uvmexp.intrs;
197 	u.swtch = uvmexp.swtch;
198 	u.softs = uvmexp.softs;
199 	u.syscalls = uvmexp.syscalls;
200 	u.pageins = uvmexp.pageins;
201 	u.swapins = uvmexp.swapins;
202 	u.swapouts = uvmexp.swapouts;
203 	u.pgswapin = uvmexp.pgswapin;
204 	u.pgswapout = uvmexp.pgswapout;
205 	u.forks = uvmexp.forks;
206 	u.forks_ppwait = uvmexp.forks_ppwait;
207 	u.forks_sharevm = uvmexp.forks_sharevm;
208 	u.pga_zerohit = uvmexp.pga_zerohit;
209 	u.pga_zeromiss = uvmexp.pga_zeromiss;
210 	u.zeroaborts = uvmexp.zeroaborts;
211 	u.fltnoram = uvmexp.fltnoram;
212 	u.fltnoanon = uvmexp.fltnoanon;
213 	u.fltpgwait = uvmexp.fltpgwait;
214 	u.fltpgrele = uvmexp.fltpgrele;
215 	u.fltrelck = uvmexp.fltrelck;
216 	u.fltrelckok = uvmexp.fltrelckok;
217 	u.fltanget = uvmexp.fltanget;
218 	u.fltanretry = uvmexp.fltanretry;
219 	u.fltamcopy = uvmexp.fltamcopy;
220 	u.fltnamap = uvmexp.fltnamap;
221 	u.fltnomap = uvmexp.fltnomap;
222 	u.fltlget = uvmexp.fltlget;
223 	u.fltget = uvmexp.fltget;
224 	u.flt_anon = uvmexp.flt_anon;
225 	u.flt_acow = uvmexp.flt_acow;
226 	u.flt_obj = uvmexp.flt_obj;
227 	u.flt_prcopy = uvmexp.flt_prcopy;
228 	u.flt_przero = uvmexp.flt_przero;
229 	u.pdwoke = uvmexp.pdwoke;
230 	u.pdrevs = uvmexp.pdrevs;
231 	u.pdswout = uvmexp.pdswout;
232 	u.pdfreed = uvmexp.pdfreed;
233 	u.pdscans = uvmexp.pdscans;
234 	u.pdanscan = uvmexp.pdanscan;
235 	u.pdobscan = uvmexp.pdobscan;
236 	u.pdreact = uvmexp.pdreact;
237 	u.pdbusy = uvmexp.pdbusy;
238 	u.pdpageouts = uvmexp.pdpageouts;
239 	u.pdpending = uvmexp.pdpending;
240 	u.pddeact = uvmexp.pddeact;
241 	u.anonpages = uvmexp.anonpages;
242 	u.filepages = uvmexp.filepages;
243 	u.execpages = uvmexp.execpages;
244 	u.colorhit = uvmexp.colorhit;
245 	u.colormiss = uvmexp.colormiss;
246 
247 	node = *rnode;
248 	node.sysctl_data = &u;
249 	node.sysctl_size = sizeof(u);
250 	return (sysctl_lookup(SYSCTLFN_CALL(&node)));
251 }
252 
253 /*
254  * sysctl helper routine for uvm_pctparam.
255  */
256 static int
257 uvm_sysctlpctparam(SYSCTLFN_ARGS)
258 {
259 	int t, error;
260 	struct sysctlnode node;
261 	struct uvm_pctparam *pct;
262 
263 	pct = rnode->sysctl_data;
264 	t = pct->pct_pct;
265 
266 	node = *rnode;
267 	node.sysctl_data = &t;
268 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
269 	if (error || newp == NULL)
270 		return error;
271 
272 	if (t < 0 || t > 100)
273 		return EINVAL;
274 
275 	error = uvm_pctparam_check(pct, t);
276 	if (error) {
277 		return error;
278 	}
279 	uvm_pctparam_set(pct, t);
280 
281 	return (0);
282 }
283 
284 /*
285  * uvm_sysctl: sysctl hook into UVM system.
286  */
287 SYSCTL_SETUP(sysctl_vm_setup, "sysctl vm subtree setup")
288 {
289 
290 	sysctl_createv(clog, 0, NULL, NULL,
291 		       CTLFLAG_PERMANENT,
292 		       CTLTYPE_NODE, "vm", NULL,
293 		       NULL, 0, NULL, 0,
294 		       CTL_VM, CTL_EOL);
295 	sysctl_createv(clog, 0, NULL, NULL,
296 		       CTLFLAG_PERMANENT,
297 		       CTLTYPE_STRUCT, "vmmeter",
298 		       SYSCTL_DESCR("Simple system-wide virtual memory "
299 				    "statistics"),
300 		       sysctl_vm_meter, 0, NULL, sizeof(struct vmtotal),
301 		       CTL_VM, VM_METER, CTL_EOL);
302 	sysctl_createv(clog, 0, NULL, NULL,
303 		       CTLFLAG_PERMANENT,
304 		       CTLTYPE_STRUCT, "loadavg",
305 		       SYSCTL_DESCR("System load average history"),
306 		       NULL, 0, &averunnable, sizeof(averunnable),
307 		       CTL_VM, VM_LOADAVG, CTL_EOL);
308 	sysctl_createv(clog, 0, NULL, NULL,
309 		       CTLFLAG_PERMANENT,
310 		       CTLTYPE_STRUCT, "uvmexp",
311 		       SYSCTL_DESCR("Detailed system-wide virtual memory "
312 				    "statistics"),
313 		       sysctl_vm_uvmexp, 0, &uvmexp, sizeof(uvmexp),
314 		       CTL_VM, VM_UVMEXP, CTL_EOL);
315 	sysctl_createv(clog, 0, NULL, NULL,
316 		       CTLFLAG_PERMANENT,
317 		       CTLTYPE_INT, "nkmempages",
318 		       SYSCTL_DESCR("Default number of pages in kmem_map"),
319 		       NULL, 0, &nkmempages, 0,
320 		       CTL_VM, VM_NKMEMPAGES, CTL_EOL);
321 	sysctl_createv(clog, 0, NULL, NULL,
322 		       CTLFLAG_PERMANENT,
323 		       CTLTYPE_STRUCT, "uvmexp2",
324 		       SYSCTL_DESCR("Detailed system-wide virtual memory "
325 				    "statistics (MI)"),
326 		       sysctl_vm_uvmexp2, 0, NULL, 0,
327 		       CTL_VM, VM_UVMEXP2, CTL_EOL);
328 	sysctl_createv(clog, 0, NULL, NULL,
329 		       CTLFLAG_PERMANENT, CTLTYPE_INT, "maxslp",
330 		       SYSCTL_DESCR("Maximum process sleep time before being "
331 				    "swapped"),
332 		       NULL, 0, &maxslp, 0,
333 		       CTL_VM, VM_MAXSLP, CTL_EOL);
334 	sysctl_createv(clog, 0, NULL, NULL,
335 		       CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
336 		       CTLTYPE_INT, "uspace",
337 		       SYSCTL_DESCR("Number of bytes allocated for a kernel "
338 				    "stack"),
339 		       NULL, USPACE, NULL, 0,
340 		       CTL_VM, VM_USPACE, CTL_EOL);
341 	sysctl_createv(clog, 0, NULL, NULL,
342 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
343 		       CTLTYPE_INT, "idlezero",
344 		       SYSCTL_DESCR("Whether try to zero pages in idle loop"),
345 		       NULL, 0, &vm_page_zero_enable, 0,
346 		       CTL_VM, CTL_CREATE, CTL_EOL);
347 
348 	uvmpdpol_sysctlsetup();
349 }
350 
351 /*
352  * uvm_total: calculate the current state of the system.
353  */
354 static void
355 uvm_total(struct vmtotal *totalp)
356 {
357 	struct lwp *l;
358 #if 0
359 	struct vm_map_entry *	entry;
360 	struct vm_map *map;
361 	int paging;
362 #endif
363 	int active;
364 
365 	memset(totalp, 0, sizeof *totalp);
366 
367 	/*
368 	 * calculate process statistics
369 	 */
370 	mutex_enter(&proclist_mutex);
371 	LIST_FOREACH(l, &alllwp, l_list) {
372 		if (l->l_proc->p_flag & PK_SYSTEM)
373 			continue;
374 		switch (l->l_stat) {
375 		case 0:
376 			continue;
377 
378 		case LSSLEEP:
379 		case LSSTOP:
380 			if (l->l_flag & LW_INMEM) {
381 				if (lwp_eprio(l) <= PZERO)
382 					totalp->t_dw++;
383 				else if (l->l_slptime < maxslp)
384 					totalp->t_sl++;
385 			} else if (l->l_slptime < maxslp)
386 				totalp->t_sw++;
387 			if (l->l_slptime >= maxslp)
388 				continue;
389 			break;
390 
391 		case LSRUN:
392 		case LSONPROC:
393 		case LSIDL:
394 			if (l->l_flag & LW_INMEM)
395 				totalp->t_rq++;
396 			else
397 				totalp->t_sw++;
398 			if (l->l_stat == LSIDL)
399 				continue;
400 			break;
401 		}
402 		/*
403 		 * note active objects
404 		 */
405 #if 0
406 		/*
407 		 * XXXCDC: BOGUS!  rethink this.  in the mean time
408 		 * don't do it.
409 		 */
410 		paging = 0;
411 		vm_map_lock(map);
412 		for (map = &p->p_vmspace->vm_map, entry = map->header.next;
413 		    entry != &map->header; entry = entry->next) {
414 			if (entry->is_a_map || entry->is_sub_map ||
415 			    entry->object.uvm_obj == NULL)
416 				continue;
417 			/* XXX how to do this with uvm */
418 		}
419 		vm_map_unlock(map);
420 		if (paging)
421 			totalp->t_pw++;
422 #endif
423 	}
424 	mutex_exit(&proclist_mutex);
425 
426 	/*
427 	 * Calculate object memory usage statistics.
428 	 */
429 	uvm_estimatepageable(&active, NULL);
430 	totalp->t_free = uvmexp.free;
431 	totalp->t_vm = uvmexp.npages - uvmexp.free + uvmexp.swpginuse;
432 	totalp->t_avm = active + uvmexp.swpginuse;	/* XXX */
433 	totalp->t_rm = uvmexp.npages - uvmexp.free;
434 	totalp->t_arm = active;
435 	totalp->t_vmshr = 0;		/* XXX */
436 	totalp->t_avmshr = 0;		/* XXX */
437 	totalp->t_rmshr = 0;		/* XXX */
438 	totalp->t_armshr = 0;		/* XXX */
439 }
440 
441 void
442 uvm_pctparam_set(struct uvm_pctparam *pct, int val)
443 {
444 
445 	pct->pct_pct = val;
446 	pct->pct_scaled = val * UVM_PCTPARAM_SCALE / 100;
447 }
448 
449 int
450 uvm_pctparam_get(struct uvm_pctparam *pct)
451 {
452 
453 	return pct->pct_pct;
454 }
455 
456 int
457 uvm_pctparam_check(struct uvm_pctparam *pct, int val)
458 {
459 
460 	if (pct->pct_check == NULL) {
461 		return 0;
462 	}
463 	return (*pct->pct_check)(pct, val);
464 }
465 
466 void
467 uvm_pctparam_init(struct uvm_pctparam *pct, int val,
468     int (*fn)(struct uvm_pctparam *, int))
469 {
470 
471 	pct->pct_check = fn;
472 	uvm_pctparam_set(pct, val);
473 }
474 
475 int
476 uvm_pctparam_createsysctlnode(struct uvm_pctparam *pct, const char *name,
477     const char *desc)
478 {
479 
480 	return sysctl_createv(NULL, 0, NULL, NULL,
481 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
482 	    CTLTYPE_INT, name, SYSCTL_DESCR(desc),
483 	    uvm_sysctlpctparam, 0, pct, 0, CTL_VM, CTL_CREATE, CTL_EOL);
484 }
485