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 1.29 2005/02/07 20:39:01 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 if (rv == FALSE && is_globaldata_space(saddr, eaddr)) 131 rv = TRUE; 132 return (rv == TRUE); 133 } 134 135 int 136 useracc(c_caddr_t addr, int len, int rw) 137 { 138 boolean_t rv; 139 vm_prot_t prot; 140 vm_map_t map; 141 vm_map_entry_t save_hint; 142 143 KASSERT((rw & (~VM_PROT_ALL)) == 0, 144 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 145 prot = rw; 146 /* 147 * XXX - check separately to disallow access to user area and user 148 * page tables - they are in the map. 149 * 150 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 151 * only used (as an end address) in trap.c. Use it as an end address 152 * here too. This bogusness has spread. I just fixed where it was 153 * used as a max in vm_mmap.c. 154 */ 155 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 156 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 157 return (FALSE); 158 } 159 map = &curproc->p_vmspace->vm_map; 160 vm_map_lock_read(map); 161 /* 162 * We save the map hint, and restore it. Useracc appears to distort 163 * the map hint unnecessarily. 164 */ 165 save_hint = map->hint; 166 rv = vm_map_check_protection(map, 167 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot); 168 map->hint = save_hint; 169 vm_map_unlock_read(map); 170 171 return (rv == TRUE); 172 } 173 174 void 175 vslock(caddr_t addr, u_int len) 176 { 177 vm_map_wire(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 178 round_page((vm_offset_t)addr + len), 0); 179 } 180 181 void 182 vsunlock(caddr_t addr, u_int len) 183 { 184 vm_map_wire(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 185 round_page((vm_offset_t)addr + len), KM_PAGEABLE); 186 } 187 188 /* 189 * Implement fork's actions on an address space. 190 * Here we arrange for the address space to be copied or referenced, 191 * allocate a user struct (pcb and kernel stack), then call the 192 * machine-dependent layer to fill those in and make the new process 193 * ready to run. The new process is set up so that it returns directly 194 * to user mode to avoid stack copying and relocation problems. 195 */ 196 void 197 vm_fork(struct proc *p1, struct proc *p2, int flags) 198 { 199 struct user *up; 200 struct thread *td2; 201 202 if ((flags & RFPROC) == 0) { 203 /* 204 * Divorce the memory, if it is shared, essentially 205 * this changes shared memory amongst threads, into 206 * COW locally. 207 */ 208 if ((flags & RFMEM) == 0) { 209 if (p1->p_vmspace->vm_refcnt > 1) { 210 vmspace_unshare(p1); 211 } 212 } 213 cpu_fork(p1, p2, flags); 214 return; 215 } 216 217 if (flags & RFMEM) { 218 p2->p_vmspace = p1->p_vmspace; 219 p1->p_vmspace->vm_refcnt++; 220 } 221 222 while (vm_page_count_severe()) { 223 vm_wait(); 224 } 225 226 if ((flags & RFMEM) == 0) { 227 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 228 229 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 230 231 if (p1->p_vmspace->vm_shm) 232 shmfork(p1, p2); 233 } 234 235 td2 = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1); 236 pmap_init_proc(p2, td2); 237 lwkt_setpri(td2, TDPRI_KERN_USER); 238 lwkt_set_comm(td2, "%s", p1->p_comm); 239 240 up = p2->p_addr; 241 242 /* 243 * p_stats currently points at fields in the user struct 244 * but not at &u, instead at p_addr. Copy parts of 245 * p_stats; zero the rest of p_stats (statistics). 246 * 247 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 248 * to share sigacts, so we use the up->u_sigacts. 249 */ 250 p2->p_stats = &up->u_stats; 251 if (p2->p_sigacts == NULL) { 252 if (p2->p_procsig->ps_refcnt != 1) 253 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 254 p2->p_sigacts = &up->u_sigacts; 255 up->u_sigacts = *p1->p_sigacts; 256 } 257 258 bzero(&up->u_stats, sizeof(struct pstats)); 259 bcopy(&p1->p_stats->p_prof, &up->u_stats.p_prof, 260 sizeof(struct uprof)); 261 bcopy(&p1->p_thread->td_start, &p2->p_thread->td_start, 262 sizeof(struct timeval)); 263 264 265 /* 266 * cpu_fork will copy and update the pcb, set up the kernel stack, 267 * and make the child ready to run. 268 */ 269 cpu_fork(p1, p2, flags); 270 } 271 272 /* 273 * Called after process has been wait(2)'ed apon and is being reaped. 274 * The idea is to reclaim resources that we could not reclaim while 275 * the process was still executing. 276 */ 277 void 278 vm_waitproc(struct proc *p) 279 { 280 p->p_stats = NULL; 281 cpu_proc_wait(p); 282 vmspace_exitfree(p); /* and clean-out the vmspace */ 283 } 284 285 /* 286 * Set default limits for VM system. 287 * Called for proc 0, and then inherited by all others. 288 * 289 * XXX should probably act directly on proc0. 290 */ 291 static void 292 vm_init_limits(void *udata) 293 { 294 struct proc *p = udata; 295 int rss_limit; 296 297 /* 298 * Set up the initial limits on process VM. Set the maximum resident 299 * set size to be half of (reasonably) available memory. Since this 300 * is a soft limit, it comes into effect only when the system is out 301 * of memory - half of main memory helps to favor smaller processes, 302 * and reduces thrashing of the object cache. 303 */ 304 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 305 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 306 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 307 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 308 /* limit the limit to no less than 2MB */ 309 rss_limit = max(vmstats.v_free_count, 512); 310 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 311 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 312 } 313 314 void 315 faultin(struct proc *p) 316 { 317 int s; 318 319 if ((p->p_flag & P_INMEM) == 0) { 320 321 ++p->p_lock; 322 323 pmap_swapin_proc(p); 324 325 s = splhigh(); 326 327 /* 328 * The process is in the kernel and controlled by LWKT, 329 * so we just schedule it rather then call setrunqueue(). 330 */ 331 if (p->p_stat == SRUN) 332 lwkt_schedule(p->p_thread); 333 334 p->p_flag |= P_INMEM; 335 336 /* undo the effect of setting SLOCK above */ 337 --p->p_lock; 338 splx(s); 339 340 } 341 } 342 343 /* 344 * Kernel initialization eventually falls through to this function, 345 * which is process 0. 346 * 347 * This swapin algorithm attempts to swap-in processes only if there 348 * is enough space for them. Of course, if a process waits for a long 349 * time, it will be swapped in anyway. 350 */ 351 /* ARGSUSED*/ 352 static void 353 scheduler(void *dummy) 354 { 355 struct proc *p; 356 int pri; 357 struct proc *pp; 358 int ppri; 359 360 KKASSERT(!IN_CRITICAL_SECT(curthread)); 361 loop: 362 if (vm_page_count_min()) { 363 vm_wait(); 364 goto loop; 365 } 366 367 pp = NULL; 368 ppri = INT_MIN; 369 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 370 if (p->p_stat == SRUN && 371 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 372 373 pri = p->p_swtime + p->p_slptime; 374 if ((p->p_flag & P_SWAPINREQ) == 0) { 375 pri -= p->p_nice * 8; 376 } 377 378 /* 379 * if this process is higher priority and there is 380 * enough space, then select this process instead of 381 * the previous selection. 382 */ 383 if (pri > ppri) { 384 pp = p; 385 ppri = pri; 386 } 387 } 388 } 389 390 /* 391 * Nothing to do, back to sleep. 392 */ 393 if ((p = pp) == NULL) { 394 tsleep(&proc0, 0, "sched", 0); 395 goto loop; 396 } 397 p->p_flag &= ~P_SWAPINREQ; 398 399 /* 400 * We would like to bring someone in. (only if there is space). 401 */ 402 faultin(p); 403 p->p_swtime = 0; 404 goto loop; 405 } 406 407 #ifndef NO_SWAPPING 408 409 #define swappable(p) \ 410 (((p)->p_lock == 0) && \ 411 ((p)->p_flag & (P_TRACED|P_SYSTEM|P_INMEM|P_WEXIT|P_SWAPPING)) == P_INMEM) 412 413 414 /* 415 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 416 */ 417 static int swap_idle_threshold1 = 2; 418 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 419 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 420 421 /* 422 * Swap_idle_threshold2 is the time that a process can be idle before 423 * it will be swapped out, if idle swapping is enabled. 424 */ 425 static int swap_idle_threshold2 = 10; 426 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 427 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 428 429 /* 430 * Swapout is driven by the pageout daemon. Very simple, we find eligible 431 * procs and unwire their u-areas. We try to always "swap" at least one 432 * process in case we need the room for a swapin. 433 * If any procs have been sleeping/stopped for at least maxslp seconds, 434 * they are swapped. Else, we swap the longest-sleeping or stopped process, 435 * if any, otherwise the longest-resident process. 436 */ 437 void 438 swapout_procs(int action) 439 { 440 struct proc *p; 441 struct proc *outp, *outp2; 442 int outpri, outpri2; 443 int didswap = 0; 444 445 outp = outp2 = NULL; 446 outpri = outpri2 = INT_MIN; 447 retry: 448 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 449 struct vmspace *vm; 450 if (!swappable(p)) 451 continue; 452 453 vm = p->p_vmspace; 454 455 switch (p->p_stat) { 456 default: 457 continue; 458 459 case SSLEEP: 460 case SSTOP: 461 /* 462 * do not swapout a realtime process 463 */ 464 if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type)) 465 continue; 466 467 /* 468 * YYY do not swapout a proc waiting on a critical 469 * event. 470 * 471 * Guarentee swap_idle_threshold time in memory 472 */ 473 if (p->p_slptime < swap_idle_threshold1) 474 continue; 475 476 /* 477 * If the system is under memory stress, or if we 478 * are swapping idle processes >= swap_idle_threshold2, 479 * then swap the process out. 480 */ 481 if (((action & VM_SWAP_NORMAL) == 0) && 482 (((action & VM_SWAP_IDLE) == 0) || 483 (p->p_slptime < swap_idle_threshold2))) 484 continue; 485 486 ++vm->vm_refcnt; 487 /* 488 * do not swapout a process that is waiting for VM 489 * data structures there is a possible deadlock. 490 */ 491 if (lockmgr(&vm->vm_map.lock, 492 LK_EXCLUSIVE | LK_NOWAIT, 493 NULL, curthread)) { 494 vmspace_free(vm); 495 continue; 496 } 497 vm_map_unlock(&vm->vm_map); 498 /* 499 * If the process has been asleep for awhile and had 500 * most of its pages taken away already, swap it out. 501 */ 502 if ((action & VM_SWAP_NORMAL) || 503 ((action & VM_SWAP_IDLE) && 504 (p->p_slptime > swap_idle_threshold2))) { 505 swapout(p); 506 vmspace_free(vm); 507 didswap++; 508 goto retry; 509 } 510 511 /* 512 * cleanup our reference 513 */ 514 vmspace_free(vm); 515 } 516 } 517 /* 518 * If we swapped something out, and another process needed memory, 519 * then wakeup the sched process. 520 */ 521 if (didswap) 522 wakeup(&proc0); 523 } 524 525 static void 526 swapout(struct proc *p) 527 { 528 529 #if defined(SWAP_DEBUG) 530 printf("swapping out %d\n", p->p_pid); 531 #endif 532 ++p->p_stats->p_ru.ru_nswap; 533 /* 534 * remember the process resident count 535 */ 536 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 537 538 (void) splhigh(); 539 p->p_flag &= ~P_INMEM; 540 p->p_flag |= P_SWAPPING; 541 if (p->p_flag & P_ONRUNQ) 542 remrunqueue(p); 543 (void) spl0(); 544 545 pmap_swapout_proc(p); 546 547 p->p_flag &= ~P_SWAPPING; 548 p->p_swtime = 0; 549 } 550 #endif /* !NO_SWAPPING */ 551