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 1.28 2004/09/05 21:25:53 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 95 /* 96 * System initialization 97 * 98 * Note: proc0 from proc.h 99 */ 100 101 static void vm_init_limits (void *); 102 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 103 104 /* 105 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 106 * 107 * Note: run scheduling should be divorced from the vm system. 108 */ 109 static void scheduler (void *); 110 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 111 112 113 static void swapout (struct proc *); 114 115 int 116 kernacc(c_caddr_t addr, int len, int rw) 117 { 118 boolean_t rv; 119 vm_offset_t saddr, eaddr; 120 vm_prot_t prot; 121 122 KASSERT((rw & (~VM_PROT_ALL)) == 0, 123 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 124 prot = rw; 125 saddr = trunc_page((vm_offset_t)addr); 126 eaddr = round_page((vm_offset_t)addr + len); 127 vm_map_lock_read(kernel_map); 128 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 129 vm_map_unlock_read(kernel_map); 130 return (rv == TRUE); 131 } 132 133 int 134 useracc(c_caddr_t addr, int len, int rw) 135 { 136 boolean_t rv; 137 vm_prot_t prot; 138 vm_map_t map; 139 vm_map_entry_t save_hint; 140 141 KASSERT((rw & (~VM_PROT_ALL)) == 0, 142 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 143 prot = rw; 144 /* 145 * XXX - check separately to disallow access to user area and user 146 * page tables - they are in the map. 147 * 148 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 149 * only used (as an end address) in trap.c. Use it as an end address 150 * here too. This bogusness has spread. I just fixed where it was 151 * used as a max in vm_mmap.c. 152 */ 153 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 154 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 155 return (FALSE); 156 } 157 map = &curproc->p_vmspace->vm_map; 158 vm_map_lock_read(map); 159 /* 160 * We save the map hint, and restore it. Useracc appears to distort 161 * the map hint unnecessarily. 162 */ 163 save_hint = map->hint; 164 rv = vm_map_check_protection(map, 165 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot); 166 map->hint = save_hint; 167 vm_map_unlock_read(map); 168 169 return (rv == TRUE); 170 } 171 172 void 173 vslock(caddr_t addr, u_int len) 174 { 175 vm_map_wire(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 176 round_page((vm_offset_t)addr + len), 0); 177 } 178 179 void 180 vsunlock(caddr_t addr, u_int len) 181 { 182 vm_map_wire(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 183 round_page((vm_offset_t)addr + len), KM_PAGEABLE); 184 } 185 186 /* 187 * Implement fork's actions on an address space. 188 * Here we arrange for the address space to be copied or referenced, 189 * allocate a user struct (pcb and kernel stack), then call the 190 * machine-dependent layer to fill those in and make the new process 191 * ready to run. The new process is set up so that it returns directly 192 * to user mode to avoid stack copying and relocation problems. 193 */ 194 void 195 vm_fork(struct proc *p1, struct proc *p2, int flags) 196 { 197 struct user *up; 198 struct thread *td2; 199 200 if ((flags & RFPROC) == 0) { 201 /* 202 * Divorce the memory, if it is shared, essentially 203 * this changes shared memory amongst threads, into 204 * COW locally. 205 */ 206 if ((flags & RFMEM) == 0) { 207 if (p1->p_vmspace->vm_refcnt > 1) { 208 vmspace_unshare(p1); 209 } 210 } 211 cpu_fork(p1, p2, flags); 212 return; 213 } 214 215 if (flags & RFMEM) { 216 p2->p_vmspace = p1->p_vmspace; 217 p1->p_vmspace->vm_refcnt++; 218 } 219 220 while (vm_page_count_severe()) { 221 vm_wait(); 222 } 223 224 if ((flags & RFMEM) == 0) { 225 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 226 227 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 228 229 if (p1->p_vmspace->vm_shm) 230 shmfork(p1, p2); 231 } 232 233 td2 = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1); 234 pmap_init_proc(p2, td2); 235 lwkt_setpri(td2, TDPRI_KERN_USER); 236 lwkt_set_comm(td2, "%s", p1->p_comm); 237 238 up = p2->p_addr; 239 240 /* 241 * p_stats currently points at fields in the user struct 242 * but not at &u, instead at p_addr. Copy parts of 243 * p_stats; zero the rest of p_stats (statistics). 244 * 245 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 246 * to share sigacts, so we use the up->u_sigacts. 247 */ 248 p2->p_stats = &up->u_stats; 249 if (p2->p_sigacts == NULL) { 250 if (p2->p_procsig->ps_refcnt != 1) 251 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 252 p2->p_sigacts = &up->u_sigacts; 253 up->u_sigacts = *p1->p_sigacts; 254 } 255 256 bzero(&up->u_stats, sizeof(struct pstats)); 257 bcopy(&p1->p_stats->p_prof, &up->u_stats.p_prof, 258 sizeof(struct uprof)); 259 bcopy(&p1->p_thread->td_start, &p2->p_thread->td_start, 260 sizeof(struct timeval)); 261 262 263 /* 264 * cpu_fork will copy and update the pcb, set up the kernel stack, 265 * and make the child ready to run. 266 */ 267 cpu_fork(p1, p2, flags); 268 } 269 270 /* 271 * Called after process has been wait(2)'ed apon and is being reaped. 272 * The idea is to reclaim resources that we could not reclaim while 273 * the process was still executing. 274 */ 275 void 276 vm_waitproc(struct proc *p) 277 { 278 p->p_stats = NULL; 279 cpu_proc_wait(p); 280 vmspace_exitfree(p); /* and clean-out the vmspace */ 281 } 282 283 /* 284 * Set default limits for VM system. 285 * Called for proc 0, and then inherited by all others. 286 * 287 * XXX should probably act directly on proc0. 288 */ 289 static void 290 vm_init_limits(void *udata) 291 { 292 struct proc *p = udata; 293 int rss_limit; 294 295 /* 296 * Set up the initial limits on process VM. Set the maximum resident 297 * set size to be half of (reasonably) available memory. Since this 298 * is a soft limit, it comes into effect only when the system is out 299 * of memory - half of main memory helps to favor smaller processes, 300 * and reduces thrashing of the object cache. 301 */ 302 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 303 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 304 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 305 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 306 /* limit the limit to no less than 2MB */ 307 rss_limit = max(vmstats.v_free_count, 512); 308 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 309 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 310 } 311 312 void 313 faultin(struct proc *p) 314 { 315 int s; 316 317 if ((p->p_flag & P_INMEM) == 0) { 318 319 ++p->p_lock; 320 321 pmap_swapin_proc(p); 322 323 s = splhigh(); 324 325 /* 326 * The process is in the kernel and controlled by LWKT, 327 * so we just schedule it rather then call setrunqueue(). 328 */ 329 if (p->p_stat == SRUN) 330 lwkt_schedule(p->p_thread); 331 332 p->p_flag |= P_INMEM; 333 334 /* undo the effect of setting SLOCK above */ 335 --p->p_lock; 336 splx(s); 337 338 } 339 } 340 341 /* 342 * Kernel initialization eventually falls through to this function, 343 * which is process 0. 344 * 345 * This swapin algorithm attempts to swap-in processes only if there 346 * is enough space for them. Of course, if a process waits for a long 347 * time, it will be swapped in anyway. 348 */ 349 /* ARGSUSED*/ 350 static void 351 scheduler(void *dummy) 352 { 353 struct proc *p; 354 int pri; 355 struct proc *pp; 356 int ppri; 357 358 KKASSERT(!IN_CRITICAL_SECT(curthread)); 359 loop: 360 if (vm_page_count_min()) { 361 vm_wait(); 362 goto loop; 363 } 364 365 pp = NULL; 366 ppri = INT_MIN; 367 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 368 if (p->p_stat == SRUN && 369 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 370 371 pri = p->p_swtime + p->p_slptime; 372 if ((p->p_flag & P_SWAPINREQ) == 0) { 373 pri -= p->p_nice * 8; 374 } 375 376 /* 377 * if this process is higher priority and there is 378 * enough space, then select this process instead of 379 * the previous selection. 380 */ 381 if (pri > ppri) { 382 pp = p; 383 ppri = pri; 384 } 385 } 386 } 387 388 /* 389 * Nothing to do, back to sleep. 390 */ 391 if ((p = pp) == NULL) { 392 tsleep(&proc0, 0, "sched", 0); 393 goto loop; 394 } 395 p->p_flag &= ~P_SWAPINREQ; 396 397 /* 398 * We would like to bring someone in. (only if there is space). 399 */ 400 faultin(p); 401 p->p_swtime = 0; 402 goto loop; 403 } 404 405 #ifndef NO_SWAPPING 406 407 #define swappable(p) \ 408 (((p)->p_lock == 0) && \ 409 ((p)->p_flag & (P_TRACED|P_SYSTEM|P_INMEM|P_WEXIT|P_SWAPPING)) == P_INMEM) 410 411 412 /* 413 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 414 */ 415 static int swap_idle_threshold1 = 2; 416 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 417 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 418 419 /* 420 * Swap_idle_threshold2 is the time that a process can be idle before 421 * it will be swapped out, if idle swapping is enabled. 422 */ 423 static int swap_idle_threshold2 = 10; 424 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 425 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 426 427 /* 428 * Swapout is driven by the pageout daemon. Very simple, we find eligible 429 * procs and unwire their u-areas. We try to always "swap" at least one 430 * process in case we need the room for a swapin. 431 * If any procs have been sleeping/stopped for at least maxslp seconds, 432 * they are swapped. Else, we swap the longest-sleeping or stopped process, 433 * if any, otherwise the longest-resident process. 434 */ 435 void 436 swapout_procs(int action) 437 { 438 struct proc *p; 439 struct proc *outp, *outp2; 440 int outpri, outpri2; 441 int didswap = 0; 442 443 outp = outp2 = NULL; 444 outpri = outpri2 = INT_MIN; 445 retry: 446 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 447 struct vmspace *vm; 448 if (!swappable(p)) 449 continue; 450 451 vm = p->p_vmspace; 452 453 switch (p->p_stat) { 454 default: 455 continue; 456 457 case SSLEEP: 458 case SSTOP: 459 /* 460 * do not swapout a realtime process 461 */ 462 if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type)) 463 continue; 464 465 /* 466 * YYY do not swapout a proc waiting on a critical 467 * event. 468 * 469 * Guarentee swap_idle_threshold time in memory 470 */ 471 if (p->p_slptime < swap_idle_threshold1) 472 continue; 473 474 /* 475 * If the system is under memory stress, or if we 476 * are swapping idle processes >= swap_idle_threshold2, 477 * then swap the process out. 478 */ 479 if (((action & VM_SWAP_NORMAL) == 0) && 480 (((action & VM_SWAP_IDLE) == 0) || 481 (p->p_slptime < swap_idle_threshold2))) 482 continue; 483 484 ++vm->vm_refcnt; 485 /* 486 * do not swapout a process that is waiting for VM 487 * data structures there is a possible deadlock. 488 */ 489 if (lockmgr(&vm->vm_map.lock, 490 LK_EXCLUSIVE | LK_NOWAIT, 491 NULL, curthread)) { 492 vmspace_free(vm); 493 continue; 494 } 495 vm_map_unlock(&vm->vm_map); 496 /* 497 * If the process has been asleep for awhile and had 498 * most of its pages taken away already, swap it out. 499 */ 500 if ((action & VM_SWAP_NORMAL) || 501 ((action & VM_SWAP_IDLE) && 502 (p->p_slptime > swap_idle_threshold2))) { 503 swapout(p); 504 vmspace_free(vm); 505 didswap++; 506 goto retry; 507 } 508 509 /* 510 * cleanup our reference 511 */ 512 vmspace_free(vm); 513 } 514 } 515 /* 516 * If we swapped something out, and another process needed memory, 517 * then wakeup the sched process. 518 */ 519 if (didswap) 520 wakeup(&proc0); 521 } 522 523 static void 524 swapout(struct proc *p) 525 { 526 527 #if defined(SWAP_DEBUG) 528 printf("swapping out %d\n", p->p_pid); 529 #endif 530 ++p->p_stats->p_ru.ru_nswap; 531 /* 532 * remember the process resident count 533 */ 534 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 535 536 (void) splhigh(); 537 p->p_flag &= ~P_INMEM; 538 p->p_flag |= P_SWAPPING; 539 if (p->p_flag & P_ONRUNQ) 540 remrunqueue(p); 541 (void) spl0(); 542 543 pmap_swapout_proc(p); 544 545 p->p_flag &= ~P_SWAPPING; 546 p->p_swtime = 0; 547 } 548 #endif /* !NO_SWAPPING */ 549