kernel - Major signal path adjustments to fix races, tsleep race fixes, +more
[dragonfly.git] / sys / vm / vm_glue.c
CommitLineData
984263bc 1/*
99ad9bc4
MD
2 * (MPSAFE)
3 *
984263bc
MD
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
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 the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94
39 *
40 *
41 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
42 * All rights reserved.
43 *
44 * Permission to use, copy, modify and distribute this software and
45 * its documentation is hereby granted, provided that both the copyright
46 * notice and this permission notice appear in all copies of the
47 * software, derivative works or modified versions, and any portions
48 * thereof, and that both notices appear in supporting documentation.
49 *
50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53 *
54 * Carnegie Mellon requests users of this software to return to
55 *
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
60 *
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
63 *
64 * $FreeBSD: src/sys/vm/vm_glue.c,v 1.94.2.4 2003/01/13 22:51:17 dillon Exp $
4ecf7cc9 65 * $DragonFly: src/sys/vm/vm_glue.c,v 1.56 2008/07/01 02:02:56 dillon Exp $
984263bc
MD
66 */
67
68#include "opt_vm.h"
69
70#include <sys/param.h>
71#include <sys/systm.h>
72#include <sys/proc.h>
73#include <sys/resourcevar.h>
74#include <sys/buf.h>
75#include <sys/shm.h>
76#include <sys/vmmeter.h>
77#include <sys/sysctl.h>
78
79#include <sys/kernel.h>
80#include <sys/unistd.h>
81
82#include <machine/limits.h>
83
84#include <vm/vm.h>
85#include <vm/vm_param.h>
86#include <sys/lock.h>
87#include <vm/pmap.h>
88#include <vm/vm_map.h>
89#include <vm/vm_page.h>
90#include <vm/vm_pageout.h>
91#include <vm/vm_kern.h>
92#include <vm/vm_extern.h>
93
94#include <sys/user.h>
12e4aaff 95#include <vm/vm_page2.h>
cdd46d2e 96#include <sys/thread2.h>
e3161323 97#include <sys/sysref2.h>
984263bc
MD
98
99/*
984263bc
MD
100 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
101 *
102 * Note: run scheduling should be divorced from the vm system.
103 */
1388df65 104static void scheduler (void *);
984263bc
MD
105SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
106
344ad853
MD
107#ifdef INVARIANTS
108
109static int swap_debug = 0;
110SYSCTL_INT(_vm, OID_AUTO, swap_debug,
111 CTLFLAG_RW, &swap_debug, 0, "");
112
113#endif
114
115static int scheduler_notify;
984263bc 116
1388df65 117static void swapout (struct proc *);
984263bc 118
99ad9bc4
MD
119/*
120 * No requirements.
121 */
984263bc 122int
6cebf0fc 123kernacc(c_caddr_t addr, int len, int rw)
984263bc
MD
124{
125 boolean_t rv;
126 vm_offset_t saddr, eaddr;
127 vm_prot_t prot;
128
129 KASSERT((rw & (~VM_PROT_ALL)) == 0,
130 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
ed468fd2 131
03debc4a
MD
132 /*
133 * The globaldata space is not part of the kernel_map proper,
134 * check access separately.
135 */
136 if (is_globaldata_space((vm_offset_t)addr, (vm_offset_t)(addr + len)))
137 return (TRUE);
138
139 /*
140 * Nominal kernel memory access - check access via kernel_map.
141 */
e4846942 142 if ((vm_offset_t)addr + len > kernel_map.max_offset ||
ed468fd2
MD
143 (vm_offset_t)addr + len < (vm_offset_t)addr) {
144 return (FALSE);
145 }
984263bc
MD
146 prot = rw;
147 saddr = trunc_page((vm_offset_t)addr);
148 eaddr = round_page((vm_offset_t)addr + len);
46754a20
MD
149 rv = vm_map_check_protection(&kernel_map, saddr, eaddr, prot, FALSE);
150
984263bc
MD
151 return (rv == TRUE);
152}
153
99ad9bc4
MD
154/*
155 * No requirements.
156 */
984263bc 157int
6cebf0fc 158useracc(c_caddr_t addr, int len, int rw)
984263bc
MD
159{
160 boolean_t rv;
161 vm_prot_t prot;
162 vm_map_t map;
163 vm_map_entry_t save_hint;
b0c15cdf 164 vm_offset_t wrap;
984263bc
MD
165
166 KASSERT((rw & (~VM_PROT_ALL)) == 0,
167 ("illegal ``rw'' argument to useracc (%x)\n", rw));
168 prot = rw;
169 /*
170 * XXX - check separately to disallow access to user area and user
171 * page tables - they are in the map.
984263bc 172 */
b0c15cdf
MD
173 wrap = (vm_offset_t)addr + len;
174 if (wrap > VM_MAX_USER_ADDRESS || wrap < (vm_offset_t)addr) {
984263bc
MD
175 return (FALSE);
176 }
177 map = &curproc->p_vmspace->vm_map;
178 vm_map_lock_read(map);
179 /*
180 * We save the map hint, and restore it. Useracc appears to distort
181 * the map hint unnecessarily.
182 */
183 save_hint = map->hint;
46754a20 184 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr),
b0c15cdf 185 round_page(wrap), prot, TRUE);
984263bc
MD
186 map->hint = save_hint;
187 vm_map_unlock_read(map);
188
189 return (rv == TRUE);
190}
191
99ad9bc4
MD
192/*
193 * No requirements.
194 */
984263bc 195void
57e43348 196vslock(caddr_t addr, u_int len)
984263bc 197{
fc00aa2b
MD
198 if (len) {
199 vm_map_wire(&curproc->p_vmspace->vm_map,
200 trunc_page((vm_offset_t)addr),
201 round_page((vm_offset_t)addr + len), 0);
202 }
984263bc
MD
203}
204
99ad9bc4
MD
205/*
206 * No requirements.
207 */
984263bc 208void
57e43348 209vsunlock(caddr_t addr, u_int len)
984263bc 210{
fc00aa2b
MD
211 if (len) {
212 vm_map_wire(&curproc->p_vmspace->vm_map,
213 trunc_page((vm_offset_t)addr),
214 round_page((vm_offset_t)addr + len),
215 KM_PAGEABLE);
216 }
984263bc
MD
217}
218
219/*
220 * Implement fork's actions on an address space.
221 * Here we arrange for the address space to be copied or referenced,
222 * allocate a user struct (pcb and kernel stack), then call the
223 * machine-dependent layer to fill those in and make the new process
224 * ready to run. The new process is set up so that it returns directly
225 * to user mode to avoid stack copying and relocation problems.
99ad9bc4
MD
226 *
227 * No requirements.
984263bc
MD
228 */
229void
13d13d89 230vm_fork(struct proc *p1, struct proc *p2, int flags)
984263bc 231{
984263bc
MD
232 if ((flags & RFPROC) == 0) {
233 /*
234 * Divorce the memory, if it is shared, essentially
235 * this changes shared memory amongst threads, into
236 * COW locally.
237 */
238 if ((flags & RFMEM) == 0) {
e3161323 239 if (p1->p_vmspace->vm_sysref.refcnt > 1) {
984263bc
MD
240 vmspace_unshare(p1);
241 }
242 }
13d13d89 243 cpu_fork(ONLY_LWP_IN_PROC(p1), NULL, flags);
984263bc
MD
244 return;
245 }
246
247 if (flags & RFMEM) {
248 p2->p_vmspace = p1->p_vmspace;
e3161323 249 sysref_get(&p1->p_vmspace->vm_sysref);
984263bc
MD
250 }
251
252 while (vm_page_count_severe()) {
4ecf7cc9 253 vm_wait(0);
984263bc
MD
254 }
255
256 if ((flags & RFMEM) == 0) {
257 p2->p_vmspace = vmspace_fork(p1->p_vmspace);
258
259 pmap_pinit2(vmspace_pmap(p2->p_vmspace));
260
261 if (p1->p_vmspace->vm_shm)
262 shmfork(p1, p2);
263 }
264
13d13d89 265 pmap_init_proc(p2);
984263bc
MD
266}
267
268/*
ba39e2e0 269 * Set default limits for VM system. Call during proc0's initialization.
99ad9bc4
MD
270 *
271 * Called from the low level boot code only.
984263bc 272 */
ba39e2e0
MD
273void
274vm_init_limits(struct proc *p)
984263bc 275{
984263bc
MD
276 int rss_limit;
277
278 /*
279 * Set up the initial limits on process VM. Set the maximum resident
280 * set size to be half of (reasonably) available memory. Since this
281 * is a soft limit, it comes into effect only when the system is out
282 * of memory - half of main memory helps to favor smaller processes,
283 * and reduces thrashing of the object cache.
284 */
285 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
286 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
287 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
288 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
289 /* limit the limit to no less than 2MB */
12e4aaff 290 rss_limit = max(vmstats.v_free_count, 512);
984263bc
MD
291 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
292 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
293}
294
344ad853
MD
295/*
296 * Faultin the specified process. Note that the process can be in any
297 * state. Just clear P_SWAPPEDOUT and call wakeup in case the process is
298 * sleeping.
99ad9bc4
MD
299 *
300 * No requirements.
344ad853 301 */
984263bc 302void
57e43348 303faultin(struct proc *p)
984263bc 304{
4643740a 305 if (p->p_flags & P_SWAPPEDOUT) {
8ec60c3f 306 /*
344ad853
MD
307 * The process is waiting in the kernel to return to user
308 * mode but cannot until P_SWAPPEDOUT gets cleared.
8ec60c3f 309 */
a5fc46c9 310 lwkt_gettoken(&p->p_token);
4643740a 311 p->p_flags &= ~(P_SWAPPEDOUT | P_SWAPWAIT);
344ad853
MD
312#ifdef INVARIANTS
313 if (swap_debug)
086c1d7e 314 kprintf("swapping in %d (%s)\n", p->p_pid, p->p_comm);
344ad853
MD
315#endif
316 wakeup(p);
a5fc46c9 317 lwkt_reltoken(&p->p_token);
984263bc
MD
318 }
319}
320
321/*
f8c3996b
MD
322 * Kernel initialization eventually falls through to this function,
323 * which is process 0.
324 *
984263bc
MD
325 * This swapin algorithm attempts to swap-in processes only if there
326 * is enough space for them. Of course, if a process waits for a long
327 * time, it will be swapped in anyway.
328 */
8fa76237
MD
329struct scheduler_info {
330 struct proc *pp;
331 int ppri;
332};
333
334static int scheduler_callback(struct proc *p, void *data);
335
984263bc 336static void
57e43348 337scheduler(void *dummy)
984263bc 338{
8fa76237 339 struct scheduler_info info;
5f910b2f 340 struct proc *p;
984263bc 341
f8c3996b 342 KKASSERT(!IN_CRITICAL_SECT(curthread));
984263bc 343loop:
344ad853
MD
344 scheduler_notify = 0;
345 /*
346 * Don't try to swap anything in if we are low on memory.
347 */
20479584 348 if (vm_page_count_severe()) {
4ecf7cc9 349 vm_wait(0);
984263bc
MD
350 goto loop;
351 }
352
344ad853
MD
353 /*
354 * Look for a good candidate to wake up
355 */
8fa76237
MD
356 info.pp = NULL;
357 info.ppri = INT_MIN;
358 allproc_scan(scheduler_callback, &info);
984263bc
MD
359
360 /*
344ad853
MD
361 * Nothing to do, back to sleep for at least 1/10 of a second. If
362 * we are woken up, immediately process the next request. If
363 * multiple requests have built up the first is processed
364 * immediately and the rest are staggered.
984263bc 365 */
8fa76237 366 if ((p = info.pp) == NULL) {
344ad853
MD
367 tsleep(&proc0, 0, "nowork", hz / 10);
368 if (scheduler_notify == 0)
369 tsleep(&scheduler_notify, 0, "nowork", 0);
984263bc
MD
370 goto loop;
371 }
984263bc
MD
372
373 /*
344ad853
MD
374 * Fault the selected process in, then wait for a short period of
375 * time and loop up.
376 *
377 * XXX we need a heuristic to get a measure of system stress and
378 * then adjust our stagger wakeup delay accordingly.
984263bc 379 */
99ad9bc4 380 lwkt_gettoken(&proc_token);
984263bc
MD
381 faultin(p);
382 p->p_swtime = 0;
8fa76237 383 PRELE(p);
99ad9bc4 384 lwkt_reltoken(&proc_token);
344ad853 385 tsleep(&proc0, 0, "swapin", hz / 10);
984263bc
MD
386 goto loop;
387}
388
99ad9bc4
MD
389/*
390 * The caller must hold proc_token.
391 */
8fa76237
MD
392static int
393scheduler_callback(struct proc *p, void *data)
394{
395 struct scheduler_info *info = data;
08f2f1bb 396 struct lwp *lp;
8fa76237
MD
397 segsz_t pgs;
398 int pri;
399
4643740a 400 if (p->p_flags & P_SWAPWAIT) {
c7e98b2f
SS
401 pri = 0;
402 FOREACH_LWP_IN_PROC(lp, p) {
403 /* XXX lwp might need a different metric */
404 pri += lp->lwp_slptime;
405 }
406 pri += p->p_swtime - p->p_nice * 8;
8fa76237
MD
407
408 /*
409 * The more pages paged out while we were swapped,
410 * the more work we have to do to get up and running
411 * again and the lower our wakeup priority.
412 *
413 * Each second of sleep time is worth ~1MB
414 */
b12defdc 415 lwkt_gettoken(&p->p_vmspace->vm_map.token);
8fa76237
MD
416 pgs = vmspace_resident_count(p->p_vmspace);
417 if (pgs < p->p_vmspace->vm_swrss) {
418 pri -= (p->p_vmspace->vm_swrss - pgs) /
419 (1024 * 1024 / PAGE_SIZE);
420 }
b12defdc 421 lwkt_reltoken(&p->p_vmspace->vm_map.token);
8fa76237
MD
422
423 /*
424 * If this process is higher priority and there is
425 * enough space, then select this process instead of
426 * the previous selection.
427 */
428 if (pri > info->ppri) {
429 if (info->pp)
430 PRELE(info->pp);
431 PHOLD(p);
432 info->pp = p;
433 info->ppri = pri;
434 }
435 }
436 return(0);
437}
438
99ad9bc4
MD
439/*
440 * SMP races ok.
441 * No requirements.
442 */
344ad853
MD
443void
444swapin_request(void)
445{
446 if (scheduler_notify == 0) {
447 scheduler_notify = 1;
448 wakeup(&scheduler_notify);
449 }
450}
451
984263bc
MD
452#ifndef NO_SWAPPING
453
454#define swappable(p) \
455 (((p)->p_lock == 0) && \
4643740a 456 ((p)->p_flags & (P_TRACED|P_SYSTEM|P_SWAPPEDOUT|P_WEXIT)) == 0)
984263bc
MD
457
458
459/*
460 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
461 */
46311ac2 462static int swap_idle_threshold1 = 15;
984263bc 463SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1,
9733f757 464 CTLFLAG_RW, &swap_idle_threshold1, 0, "Guaranteed process resident time (sec)");
984263bc
MD
465
466/*
467 * Swap_idle_threshold2 is the time that a process can be idle before
344ad853
MD
468 * it will be swapped out, if idle swapping is enabled. Default is
469 * one minute.
984263bc 470 */
344ad853 471static int swap_idle_threshold2 = 60;
984263bc 472SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2,
9733f757 473 CTLFLAG_RW, &swap_idle_threshold2, 0, "Time (sec) a process can idle before being swapped");
984263bc
MD
474
475/*
476 * Swapout is driven by the pageout daemon. Very simple, we find eligible
344ad853
MD
477 * procs and mark them as being swapped out. This will cause the kernel
478 * to prefer to pageout those proc's pages first and the procs in question
479 * will not return to user mode until the swapper tells them they can.
480 *
984263bc
MD
481 * If any procs have been sleeping/stopped for at least maxslp seconds,
482 * they are swapped. Else, we swap the longest-sleeping or stopped process,
483 * if any, otherwise the longest-resident process.
484 */
8fa76237
MD
485
486static int swapout_procs_callback(struct proc *p, void *data);
487
99ad9bc4
MD
488/*
489 * No requirements.
490 */
984263bc 491void
57e43348 492swapout_procs(int action)
984263bc 493{
8fa76237
MD
494 allproc_scan(swapout_procs_callback, &action);
495}
984263bc 496
99ad9bc4 497/*
b12defdc 498 * The caller must hold proc_token
99ad9bc4 499 */
8fa76237
MD
500static int
501swapout_procs_callback(struct proc *p, void *data)
502{
503 struct vmspace *vm;
08f2f1bb 504 struct lwp *lp;
8fa76237 505 int action = *(int *)data;
c7e98b2f 506 int minslp = -1;
8fa76237
MD
507
508 if (!swappable(p))
509 return(0);
510
a5fc46c9 511 lwkt_gettoken(&p->p_token);
8fa76237
MD
512 vm = p->p_vmspace;
513
c7e98b2f
SS
514 /*
515 * We only consider active processes.
516 */
a5fc46c9
MD
517 if (p->p_stat != SACTIVE && p->p_stat != SSTOP) {
518 lwkt_reltoken(&p->p_token);
c7e98b2f 519 return(0);
a5fc46c9 520 }
164b8401 521
c7e98b2f 522 FOREACH_LWP_IN_PROC(lp, p) {
8fa76237
MD
523 /*
524 * do not swap out a realtime process
525 */
a5fc46c9
MD
526 if (RTP_PRIO_IS_REALTIME(lp->lwp_rtprio.type)) {
527 lwkt_reltoken(&p->p_token);
8fa76237 528 return(0);
a5fc46c9 529 }
984263bc 530
8fa76237
MD
531 /*
532 * Guarentee swap_idle_threshold time in memory
533 */
a5fc46c9
MD
534 if (lp->lwp_slptime < swap_idle_threshold1) {
535 lwkt_reltoken(&p->p_token);
8fa76237 536 return(0);
a5fc46c9 537 }
8fa76237
MD
538
539 /*
540 * If the system is under memory stress, or if we
541 * are swapping idle processes >= swap_idle_threshold2,
542 * then swap the process out.
543 */
544 if (((action & VM_SWAP_NORMAL) == 0) &&
545 (((action & VM_SWAP_IDLE) == 0) ||
08f2f1bb 546 (lp->lwp_slptime < swap_idle_threshold2))) {
a5fc46c9 547 lwkt_reltoken(&p->p_token);
8fa76237 548 return(0);
984263bc 549 }
8fa76237 550
c7e98b2f
SS
551 if (minslp == -1 || lp->lwp_slptime < minslp)
552 minslp = lp->lwp_slptime;
553 }
8fa76237 554
c7e98b2f
SS
555 /*
556 * If the process has been asleep for awhile, swap
557 * it out.
558 */
559 if ((action & VM_SWAP_NORMAL) ||
560 ((action & VM_SWAP_IDLE) &&
561 (minslp > swap_idle_threshold2))) {
562 swapout(p);
984263bc 563 }
c7e98b2f
SS
564
565 /*
566 * cleanup our reference
567 */
a5fc46c9 568 lwkt_reltoken(&p->p_token);
c7e98b2f 569
8fa76237 570 return(0);
984263bc
MD
571}
572
99ad9bc4 573/*
b12defdc 574 * The caller must hold proc_token and p->p_token
99ad9bc4 575 */
984263bc 576static void
57e43348 577swapout(struct proc *p)
984263bc 578{
344ad853
MD
579#ifdef INVARIANTS
580 if (swap_debug)
086c1d7e 581 kprintf("swapping out %d (%s)\n", p->p_pid, p->p_comm);
984263bc 582#endif
fde7ac71 583 ++p->p_ru.ru_nswap;
b0c15cdf 584
984263bc
MD
585 /*
586 * remember the process resident count
587 */
588 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
4643740a 589 p->p_flags |= P_SWAPPEDOUT;
984263bc
MD
590 p->p_swtime = 0;
591}
344ad853 592
984263bc 593#endif /* !NO_SWAPPING */
344ad853 594