1 /* 2 * Copyright (c) 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * The Mach Operating System project at Carnegie-Mellon University. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 37 * 38 * 39 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 40 * All rights reserved. 41 * 42 * Permission to use, copy, modify and distribute this software and 43 * its documentation is hereby granted, provided that both the copyright 44 * notice and this permission notice appear in all copies of the 45 * software, derivative works or modified versions, and any portions 46 * thereof, and that both notices appear in supporting documentation. 47 * 48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 51 * 52 * Carnegie Mellon requests users of this software to return to 53 * 54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55 * School of Computer Science 56 * Carnegie Mellon University 57 * Pittsburgh PA 15213-3890 58 * 59 * any improvements or extensions that they make and grant Carnegie the 60 * rights to redistribute these changes. 61 * 62 * $FreeBSD: src/sys/vm/vm_glue.c,v 1.94.2.4 2003/01/13 22:51:17 dillon Exp $ 63 * $DragonFly: src/sys/vm/vm_glue.c,v 1.39 2006/03/15 07:58:37 dillon Exp $ 64 */ 65 66 #include "opt_vm.h" 67 68 #include <sys/param.h> 69 #include <sys/systm.h> 70 #include <sys/proc.h> 71 #include <sys/resourcevar.h> 72 #include <sys/buf.h> 73 #include <sys/shm.h> 74 #include <sys/vmmeter.h> 75 #include <sys/sysctl.h> 76 77 #include <sys/kernel.h> 78 #include <sys/unistd.h> 79 80 #include <machine/limits.h> 81 82 #include <vm/vm.h> 83 #include <vm/vm_param.h> 84 #include <sys/lock.h> 85 #include <vm/pmap.h> 86 #include <vm/vm_map.h> 87 #include <vm/vm_page.h> 88 #include <vm/vm_pageout.h> 89 #include <vm/vm_kern.h> 90 #include <vm/vm_extern.h> 91 92 #include <sys/user.h> 93 #include <vm/vm_page2.h> 94 #include <sys/thread2.h> 95 96 /* 97 * System initialization 98 * 99 * Note: proc0 from proc.h 100 */ 101 102 static void vm_init_limits (void *); 103 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 104 105 /* 106 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 107 * 108 * Note: run scheduling should be divorced from the vm system. 109 */ 110 static void scheduler (void *); 111 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 112 113 #ifdef INVARIANTS 114 115 static int swap_debug = 0; 116 SYSCTL_INT(_vm, OID_AUTO, swap_debug, 117 CTLFLAG_RW, &swap_debug, 0, ""); 118 119 #endif 120 121 static int scheduler_notify; 122 123 static void swapout (struct proc *); 124 125 int 126 kernacc(c_caddr_t addr, int len, int rw) 127 { 128 boolean_t rv; 129 vm_offset_t saddr, eaddr; 130 vm_prot_t prot; 131 132 KASSERT((rw & (~VM_PROT_ALL)) == 0, 133 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 134 prot = rw; 135 saddr = trunc_page((vm_offset_t)addr); 136 eaddr = round_page((vm_offset_t)addr + len); 137 vm_map_lock_read(kernel_map); 138 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 139 vm_map_unlock_read(kernel_map); 140 if (rv == FALSE && is_globaldata_space(saddr, eaddr)) 141 rv = TRUE; 142 return (rv == TRUE); 143 } 144 145 int 146 useracc(c_caddr_t addr, int len, int rw) 147 { 148 boolean_t rv; 149 vm_prot_t prot; 150 vm_map_t map; 151 vm_map_entry_t save_hint; 152 153 KASSERT((rw & (~VM_PROT_ALL)) == 0, 154 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 155 prot = rw; 156 /* 157 * XXX - check separately to disallow access to user area and user 158 * page tables - they are in the map. 159 * 160 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 161 * only used (as an end address) in trap.c. Use it as an end address 162 * here too. This bogusness has spread. I just fixed where it was 163 * used as a max in vm_mmap.c. 164 */ 165 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 166 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 167 return (FALSE); 168 } 169 map = &curproc->p_vmspace->vm_map; 170 vm_map_lock_read(map); 171 /* 172 * We save the map hint, and restore it. Useracc appears to distort 173 * the map hint unnecessarily. 174 */ 175 save_hint = map->hint; 176 rv = vm_map_check_protection(map, 177 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot); 178 map->hint = save_hint; 179 vm_map_unlock_read(map); 180 181 return (rv == TRUE); 182 } 183 184 void 185 vslock(caddr_t addr, u_int len) 186 { 187 vm_map_wire(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 188 round_page((vm_offset_t)addr + len), 0); 189 } 190 191 void 192 vsunlock(caddr_t addr, u_int len) 193 { 194 vm_map_wire(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 195 round_page((vm_offset_t)addr + len), KM_PAGEABLE); 196 } 197 198 /* 199 * Implement fork's actions on an address space. 200 * Here we arrange for the address space to be copied or referenced, 201 * allocate a user struct (pcb and kernel stack), then call the 202 * machine-dependent layer to fill those in and make the new process 203 * ready to run. The new process is set up so that it returns directly 204 * to user mode to avoid stack copying and relocation problems. 205 */ 206 void 207 vm_fork(struct proc *p1, struct proc *p2, int flags) 208 { 209 struct user *up; 210 struct thread *td2; 211 212 if ((flags & RFPROC) == 0) { 213 /* 214 * Divorce the memory, if it is shared, essentially 215 * this changes shared memory amongst threads, into 216 * COW locally. 217 */ 218 if ((flags & RFMEM) == 0) { 219 if (p1->p_vmspace->vm_refcnt > 1) { 220 vmspace_unshare(p1); 221 } 222 } 223 cpu_fork(p1, p2, flags); 224 return; 225 } 226 227 if (flags & RFMEM) { 228 p2->p_vmspace = p1->p_vmspace; 229 p1->p_vmspace->vm_refcnt++; 230 } 231 232 while (vm_page_count_severe()) { 233 vm_wait(); 234 } 235 236 if ((flags & RFMEM) == 0) { 237 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 238 239 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 240 241 if (p1->p_vmspace->vm_shm) 242 shmfork(p1, p2); 243 } 244 245 td2 = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0); 246 pmap_init_proc(p2, td2); 247 lwkt_setpri(td2, TDPRI_KERN_USER); 248 lwkt_set_comm(td2, "%s", p1->p_comm); 249 250 up = p2->p_addr; 251 252 /* 253 * p_stats currently points at fields in the user struct 254 * but not at &u, instead at p_addr. Copy parts of 255 * p_stats; zero the rest of p_stats (statistics). 256 * 257 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 258 * to share sigacts, so we use the up->u_sigacts. 259 */ 260 p2->p_stats = &up->u_stats; 261 if (p2->p_sigacts == NULL) { 262 if (p2->p_procsig->ps_refcnt != 1) 263 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 264 p2->p_sigacts = &up->u_sigacts; 265 up->u_sigacts = *p1->p_sigacts; 266 } 267 268 bzero(&up->u_stats, sizeof(struct pstats)); 269 270 /* 271 * cpu_fork will copy and update the pcb, set up the kernel stack, 272 * and make the child ready to run. 273 */ 274 cpu_fork(p1, p2, flags); 275 } 276 277 /* 278 * Called after process has been wait(2)'ed apon and is being reaped. 279 * The idea is to reclaim resources that we could not reclaim while 280 * the process was still executing. 281 */ 282 void 283 vm_waitproc(struct proc *p) 284 { 285 p->p_stats = NULL; 286 cpu_proc_wait(p); 287 vmspace_exitfree(p); /* and clean-out the vmspace */ 288 } 289 290 /* 291 * Set default limits for VM system. 292 * Called for proc 0, and then inherited by all others. 293 * 294 * XXX should probably act directly on proc0. 295 */ 296 static void 297 vm_init_limits(void *udata) 298 { 299 struct proc *p = udata; 300 int rss_limit; 301 302 /* 303 * Set up the initial limits on process VM. Set the maximum resident 304 * set size to be half of (reasonably) available memory. Since this 305 * is a soft limit, it comes into effect only when the system is out 306 * of memory - half of main memory helps to favor smaller processes, 307 * and reduces thrashing of the object cache. 308 */ 309 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 310 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 311 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 312 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 313 /* limit the limit to no less than 2MB */ 314 rss_limit = max(vmstats.v_free_count, 512); 315 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 316 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 317 } 318 319 /* 320 * Faultin the specified process. Note that the process can be in any 321 * state. Just clear P_SWAPPEDOUT and call wakeup in case the process is 322 * sleeping. 323 */ 324 void 325 faultin(struct proc *p) 326 { 327 if (p->p_flag & P_SWAPPEDOUT) { 328 /* 329 * The process is waiting in the kernel to return to user 330 * mode but cannot until P_SWAPPEDOUT gets cleared. 331 */ 332 crit_enter(); 333 p->p_flag &= ~(P_SWAPPEDOUT | P_SWAPWAIT); 334 #ifdef INVARIANTS 335 if (swap_debug) 336 printf("swapping in %d (%s)\n", p->p_pid, p->p_comm); 337 #endif 338 wakeup(p); 339 340 crit_exit(); 341 } 342 } 343 344 /* 345 * Kernel initialization eventually falls through to this function, 346 * which is process 0. 347 * 348 * This swapin algorithm attempts to swap-in processes only if there 349 * is enough space for them. Of course, if a process waits for a long 350 * time, it will be swapped in anyway. 351 */ 352 353 /* ARGSUSED*/ 354 static void 355 scheduler(void *dummy) 356 { 357 struct proc *p; 358 struct proc *pp; 359 int pri; 360 int ppri; 361 segsz_t pgs; 362 363 KKASSERT(!IN_CRITICAL_SECT(curthread)); 364 loop: 365 scheduler_notify = 0; 366 /* 367 * Don't try to swap anything in if we are low on memory. 368 */ 369 if (vm_page_count_min()) { 370 vm_wait(); 371 goto loop; 372 } 373 374 /* 375 * Look for a good candidate to wake up 376 */ 377 pp = NULL; 378 ppri = INT_MIN; 379 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 380 if (p->p_flag & P_SWAPWAIT) { 381 pri = p->p_swtime + p->p_slptime - p->p_nice * 8; 382 383 /* 384 * The more pages paged out while we were swapped, 385 * the more work we have to do to get up and running 386 * again and the lower our wakeup priority. 387 * 388 * Each second of sleep time is worth ~1MB 389 */ 390 pgs = vmspace_resident_count(p->p_vmspace); 391 if (pgs < p->p_vmspace->vm_swrss) { 392 pri -= (p->p_vmspace->vm_swrss - pgs) / 393 (1024 * 1024 / PAGE_SIZE); 394 } 395 396 /* 397 * if this process is higher priority and there is 398 * enough space, then select this process instead of 399 * the previous selection. 400 */ 401 if (pri > ppri) { 402 pp = p; 403 ppri = pri; 404 } 405 } 406 } 407 408 /* 409 * Nothing to do, back to sleep for at least 1/10 of a second. If 410 * we are woken up, immediately process the next request. If 411 * multiple requests have built up the first is processed 412 * immediately and the rest are staggered. 413 */ 414 if ((p = pp) == NULL) { 415 tsleep(&proc0, 0, "nowork", hz / 10); 416 if (scheduler_notify == 0) 417 tsleep(&scheduler_notify, 0, "nowork", 0); 418 goto loop; 419 } 420 421 /* 422 * Fault the selected process in, then wait for a short period of 423 * time and loop up. 424 * 425 * XXX we need a heuristic to get a measure of system stress and 426 * then adjust our stagger wakeup delay accordingly. 427 */ 428 faultin(p); 429 p->p_swtime = 0; 430 tsleep(&proc0, 0, "swapin", hz / 10); 431 goto loop; 432 } 433 434 void 435 swapin_request(void) 436 { 437 if (scheduler_notify == 0) { 438 scheduler_notify = 1; 439 wakeup(&scheduler_notify); 440 } 441 } 442 443 #ifndef NO_SWAPPING 444 445 #define swappable(p) \ 446 (((p)->p_lock == 0) && \ 447 ((p)->p_flag & (P_TRACED|P_SYSTEM|P_SWAPPEDOUT|P_WEXIT)) == 0) 448 449 450 /* 451 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 452 */ 453 static int swap_idle_threshold1 = 15; 454 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 455 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 456 457 /* 458 * Swap_idle_threshold2 is the time that a process can be idle before 459 * it will be swapped out, if idle swapping is enabled. Default is 460 * one minute. 461 */ 462 static int swap_idle_threshold2 = 60; 463 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 464 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 465 466 /* 467 * Swapout is driven by the pageout daemon. Very simple, we find eligible 468 * procs and mark them as being swapped out. This will cause the kernel 469 * to prefer to pageout those proc's pages first and the procs in question 470 * will not return to user mode until the swapper tells them they can. 471 * 472 * If any procs have been sleeping/stopped for at least maxslp seconds, 473 * they are swapped. Else, we swap the longest-sleeping or stopped process, 474 * if any, otherwise the longest-resident process. 475 */ 476 void 477 swapout_procs(int action) 478 { 479 struct proc *p; 480 struct proc *outp, *outp2; 481 int outpri, outpri2; 482 483 outp = outp2 = NULL; 484 outpri = outpri2 = INT_MIN; 485 retry: 486 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 487 struct vmspace *vm; 488 if (!swappable(p)) 489 continue; 490 491 vm = p->p_vmspace; 492 493 if (p->p_stat == SSLEEP || p->p_stat == SRUN) { 494 /* 495 * do not swap out a realtime process 496 */ 497 if (RTP_PRIO_IS_REALTIME(p->p_lwp.lwp_rtprio.type)) 498 continue; 499 500 /* 501 * Guarentee swap_idle_threshold time in memory 502 */ 503 if (p->p_slptime < swap_idle_threshold1) 504 continue; 505 506 /* 507 * If the system is under memory stress, or if we 508 * are swapping idle processes >= swap_idle_threshold2, 509 * then swap the process out. 510 */ 511 if (((action & VM_SWAP_NORMAL) == 0) && 512 (((action & VM_SWAP_IDLE) == 0) || 513 (p->p_slptime < swap_idle_threshold2))) { 514 continue; 515 } 516 517 ++vm->vm_refcnt; 518 519 /* 520 * If the process has been asleep for awhile, swap 521 * it out. 522 */ 523 if ((action & VM_SWAP_NORMAL) || 524 ((action & VM_SWAP_IDLE) && 525 (p->p_slptime > swap_idle_threshold2))) { 526 swapout(p); 527 vmspace_free(vm); 528 goto retry; 529 } 530 531 /* 532 * cleanup our reference 533 */ 534 vmspace_free(vm); 535 } 536 } 537 } 538 539 static void 540 swapout(struct proc *p) 541 { 542 #ifdef INVARIANTS 543 if (swap_debug) 544 printf("swapping out %d (%s)\n", p->p_pid, p->p_comm); 545 #endif 546 ++p->p_stats->p_ru.ru_nswap; 547 /* 548 * remember the process resident count 549 */ 550 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 551 p->p_flag |= P_SWAPPEDOUT; 552 p->p_swtime = 0; 553 } 554 555 #endif /* !NO_SWAPPING */ 556 557