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 * $FreeBSD$ 63 */ 64 65 #include "opt_vm.h" 66 67 #include <sys/param.h> 68 #include <sys/systm.h> 69 #include <sys/lock.h> 70 #include <sys/mutex.h> 71 #include <sys/proc.h> 72 #include <sys/resourcevar.h> 73 #include <sys/shm.h> 74 #include <sys/vmmeter.h> 75 #include <sys/sx.h> 76 #include <sys/sysctl.h> 77 78 #include <sys/kernel.h> 79 #include <sys/ktr.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 <vm/pmap.h> 87 #include <vm/vm_map.h> 88 #include <vm/vm_page.h> 89 #include <vm/vm_pageout.h> 90 #include <vm/vm_kern.h> 91 #include <vm/vm_extern.h> 92 93 #include <sys/user.h> 94 95 extern int maxslp; 96 97 /* 98 * System initialization 99 * 100 * Note: proc0 from proc.h 101 */ 102 103 static void vm_init_limits __P((void *)); 104 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 105 106 /* 107 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 108 * 109 * Note: run scheduling should be divorced from the vm system. 110 */ 111 static void scheduler __P((void *)); 112 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 113 114 #ifndef NO_SWAPPING 115 static void swapout __P((struct proc *)); 116 #endif 117 118 int 119 kernacc(addr, len, rw) 120 caddr_t addr; 121 int len, rw; 122 { 123 boolean_t rv; 124 vm_offset_t saddr, eaddr; 125 vm_prot_t prot; 126 127 KASSERT((rw & ~VM_PROT_ALL) == 0, 128 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 129 prot = rw; 130 saddr = trunc_page((vm_offset_t)addr); 131 eaddr = round_page((vm_offset_t)addr + len); 132 vm_map_lock_read(kernel_map); 133 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 134 vm_map_unlock_read(kernel_map); 135 return (rv == TRUE); 136 } 137 138 int 139 useracc(addr, len, rw) 140 caddr_t addr; 141 int len, rw; 142 { 143 boolean_t rv; 144 vm_prot_t prot; 145 vm_map_t map; 146 vm_map_entry_t save_hint; 147 148 GIANT_REQUIRED; 149 150 KASSERT((rw & ~VM_PROT_ALL) == 0, 151 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 152 prot = rw; 153 /* 154 * XXX - check separately to disallow access to user area and user 155 * page tables - they are in the map. 156 * 157 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 158 * only used (as an end address) in trap.c. Use it as an end address 159 * here too. This bogusness has spread. I just fixed where it was 160 * used as a max in vm_mmap.c. 161 */ 162 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 163 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 164 return (FALSE); 165 } 166 map = &curproc->p_vmspace->vm_map; 167 vm_map_lock_read(map); 168 /* 169 * We save the map hint, and restore it. Useracc appears to distort 170 * the map hint unnecessarily. 171 */ 172 save_hint = map->hint; 173 rv = vm_map_check_protection(map, 174 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot); 175 map->hint = save_hint; 176 vm_map_unlock_read(map); 177 178 return (rv == TRUE); 179 } 180 181 void 182 vslock(addr, len) 183 caddr_t addr; 184 u_int len; 185 { 186 GIANT_REQUIRED; 187 vm_map_pageable(&curproc->p_vmspace->vm_map, 188 trunc_page((vm_offset_t)addr), 189 round_page((vm_offset_t)addr + len), FALSE); 190 } 191 192 void 193 vsunlock(addr, len) 194 caddr_t addr; 195 u_int len; 196 { 197 GIANT_REQUIRED; 198 vm_map_pageable(&curproc->p_vmspace->vm_map, 199 trunc_page((vm_offset_t)addr), 200 round_page((vm_offset_t)addr + len), TRUE); 201 } 202 203 /* 204 * Implement fork's actions on an address space. 205 * Here we arrange for the address space to be copied or referenced, 206 * allocate a user struct (pcb and kernel stack), then call the 207 * machine-dependent layer to fill those in and make the new process 208 * ready to run. The new process is set up so that it returns directly 209 * to user mode to avoid stack copying and relocation problems. 210 */ 211 void 212 vm_forkproc(td, p2, flags) 213 struct thread *td; 214 struct proc *p2; 215 int flags; 216 { 217 struct proc *p1 = td->td_proc; 218 struct user *up; 219 220 GIANT_REQUIRED; 221 222 if ((flags & RFPROC) == 0) { 223 /* 224 * Divorce the memory, if it is shared, essentially 225 * this changes shared memory amongst threads, into 226 * COW locally. 227 */ 228 if ((flags & RFMEM) == 0) { 229 if (p1->p_vmspace->vm_refcnt > 1) { 230 vmspace_unshare(p1); 231 } 232 } 233 cpu_fork(td, p2, flags); 234 return; 235 } 236 237 if (flags & RFMEM) { 238 p2->p_vmspace = p1->p_vmspace; 239 p1->p_vmspace->vm_refcnt++; 240 } 241 242 while (vm_page_count_severe()) { 243 VM_WAIT; 244 } 245 246 if ((flags & RFMEM) == 0) { 247 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 248 249 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 250 251 if (p1->p_vmspace->vm_shm) 252 shmfork(p1, p2); 253 } 254 255 pmap_new_proc(p2); 256 pmap_new_thread(&p2->p_thread); /* Initial thread */ 257 258 /* XXXKSE this is unsatisfactory but should be adequate */ 259 up = p2->p_uarea; 260 261 /* 262 * p_stats currently points at fields in the user struct 263 * but not at &u, instead at p_addr. Copy parts of 264 * p_stats; zero the rest of p_stats (statistics). 265 * 266 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 267 * to share sigacts, so we use the up->u_sigacts. 268 */ 269 p2->p_stats = &up->u_stats; 270 if (p2->p_sigacts == NULL) { 271 if (p2->p_procsig->ps_refcnt != 1) 272 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 273 p2->p_sigacts = &up->u_sigacts; 274 up->u_sigacts = *p1->p_sigacts; 275 } 276 277 bzero(&up->u_stats.pstat_startzero, 278 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 279 (caddr_t) &up->u_stats.pstat_startzero)); 280 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 281 ((caddr_t) &up->u_stats.pstat_endcopy - 282 (caddr_t) &up->u_stats.pstat_startcopy)); 283 284 285 /* 286 * cpu_fork will copy and update the pcb, set up the kernel stack, 287 * and make the child ready to run. 288 */ 289 cpu_fork(td, p2, flags); 290 } 291 292 /* 293 * Called after process has been wait(2)'ed apon and is being reaped. 294 * The idea is to reclaim resources that we could not reclaim while 295 * the process was still executing. 296 */ 297 void 298 vm_waitproc(p) 299 struct proc *p; 300 { 301 struct thread *td; 302 303 GIANT_REQUIRED; 304 cpu_wait(p); 305 pmap_dispose_proc(p); /* drop per-process resources */ 306 FOREACH_THREAD_IN_PROC(p, td) 307 pmap_dispose_thread(td); 308 vmspace_free(p->p_vmspace); /* and clean-out the vmspace */ 309 } 310 311 /* 312 * Set default limits for VM system. 313 * Called for proc 0, and then inherited by all others. 314 * 315 * XXX should probably act directly on proc0. 316 */ 317 static void 318 vm_init_limits(udata) 319 void *udata; 320 { 321 struct proc *p = udata; 322 int rss_limit; 323 324 /* 325 * Set up the initial limits on process VM. Set the maximum resident 326 * set size to be half of (reasonably) available memory. Since this 327 * is a soft limit, it comes into effect only when the system is out 328 * of memory - half of main memory helps to favor smaller processes, 329 * and reduces thrashing of the object cache. 330 */ 331 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 332 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 333 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 334 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 335 /* limit the limit to no less than 2MB */ 336 rss_limit = max(cnt.v_free_count, 512); 337 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 338 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 339 } 340 341 void 342 faultin(p) 343 struct proc *p; 344 { 345 struct thread *td; 346 GIANT_REQUIRED; 347 348 PROC_LOCK_ASSERT(p, MA_OWNED); 349 mtx_lock_spin(&sched_lock); 350 if ((p->p_sflag & PS_INMEM) == 0) { 351 ++p->p_lock; 352 mtx_unlock_spin(&sched_lock); 353 PROC_UNLOCK(p); 354 355 pmap_swapin_proc(p); 356 FOREACH_THREAD_IN_PROC (p, td) 357 pmap_swapin_thread(td); 358 359 PROC_LOCK(p); 360 mtx_lock_spin(&sched_lock); 361 FOREACH_THREAD_IN_PROC (p, td) 362 if (td->td_proc->p_stat == SRUN) /* XXXKSE */ 363 setrunqueue(td); 364 365 p->p_sflag |= PS_INMEM; 366 367 /* undo the effect of setting SLOCK above */ 368 --p->p_lock; 369 } 370 mtx_unlock_spin(&sched_lock); 371 } 372 373 /* 374 * This swapin algorithm attempts to swap-in processes only if there 375 * is enough space for them. Of course, if a process waits for a long 376 * time, it will be swapped in anyway. 377 * 378 * XXXKSE - KSEGRP with highest priority counts.. 379 * 380 * Giant is still held at this point, to be released in tsleep. 381 */ 382 /* ARGSUSED*/ 383 static void 384 scheduler(dummy) 385 void *dummy; 386 { 387 struct proc *p; 388 int pri; 389 struct proc *pp; 390 int ppri; 391 392 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED); 393 /* GIANT_REQUIRED */ 394 395 loop: 396 if (vm_page_count_min()) { 397 VM_WAIT; 398 goto loop; 399 } 400 401 pp = NULL; 402 ppri = INT_MIN; 403 sx_slock(&allproc_lock); 404 FOREACH_PROC_IN_SYSTEM(p) { 405 struct ksegrp *kg; 406 mtx_lock_spin(&sched_lock); 407 if (p->p_stat == SRUN 408 && (p->p_sflag & (PS_INMEM | PS_SWAPPING)) == 0) { 409 /* Find the minimum sleeptime for the process */ 410 FOREACH_KSEGRP_IN_PROC(p, kg) { 411 pri = p->p_swtime + kg->kg_slptime; 412 if ((p->p_sflag & PS_SWAPINREQ) == 0) { 413 pri -= kg->kg_nice * 8; 414 } 415 416 417 /* 418 * if this ksegrp is higher priority 419 * and there is enough space, then select 420 * this process instead of the previous 421 * selection. 422 */ 423 if (pri > ppri) { 424 pp = p; 425 ppri = pri; 426 } 427 } 428 } 429 mtx_unlock_spin(&sched_lock); 430 } 431 sx_sunlock(&allproc_lock); 432 433 /* 434 * Nothing to do, back to sleep. 435 */ 436 if ((p = pp) == NULL) { 437 tsleep(&proc0, PVM, "sched", maxslp * hz / 2); 438 goto loop; 439 } 440 mtx_lock_spin(&sched_lock); 441 p->p_sflag &= ~PS_SWAPINREQ; 442 mtx_unlock_spin(&sched_lock); 443 444 /* 445 * We would like to bring someone in. (only if there is space). 446 */ 447 PROC_LOCK(p); 448 faultin(p); 449 PROC_UNLOCK(p); 450 mtx_lock_spin(&sched_lock); 451 p->p_swtime = 0; 452 mtx_unlock_spin(&sched_lock); 453 goto loop; 454 } 455 456 #ifndef NO_SWAPPING 457 458 /* 459 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 460 */ 461 static int swap_idle_threshold1 = 2; 462 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 463 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 464 465 /* 466 * Swap_idle_threshold2 is the time that a process can be idle before 467 * it will be swapped out, if idle swapping is enabled. 468 */ 469 static int swap_idle_threshold2 = 10; 470 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 471 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 472 473 /* 474 * Swapout is driven by the pageout daemon. Very simple, we find eligible 475 * procs and unwire their u-areas. We try to always "swap" at least one 476 * process in case we need the room for a swapin. 477 * If any procs have been sleeping/stopped for at least maxslp seconds, 478 * they are swapped. Else, we swap the longest-sleeping or stopped process, 479 * if any, otherwise the longest-resident process. 480 */ 481 void 482 swapout_procs(action) 483 int action; 484 { 485 struct proc *p; 486 struct ksegrp *kg; 487 struct proc *outp, *outp2; 488 int outpri, outpri2; 489 int didswap = 0; 490 491 GIANT_REQUIRED; 492 493 outp = outp2 = NULL; 494 outpri = outpri2 = INT_MIN; 495 retry: 496 sx_slock(&allproc_lock); 497 LIST_FOREACH(p, &allproc, p_list) { 498 struct vmspace *vm; 499 int minslptime = 100000; 500 501 PROC_LOCK(p); 502 if (p->p_lock != 0 || 503 (p->p_flag & (P_TRACED|P_SYSTEM|P_WEXIT)) != 0) { 504 PROC_UNLOCK(p); 505 continue; 506 } 507 /* 508 * only aiod changes vmspace, however it will be 509 * skipped because of the if statement above checking 510 * for P_SYSTEM 511 */ 512 vm = p->p_vmspace; 513 mtx_lock_spin(&sched_lock); 514 if ((p->p_sflag & (PS_INMEM|PS_SWAPPING)) != PS_INMEM) { 515 mtx_unlock_spin(&sched_lock); 516 PROC_UNLOCK(p); 517 continue; 518 } 519 520 switch (p->p_stat) { 521 default: 522 mtx_unlock_spin(&sched_lock); 523 PROC_UNLOCK(p); 524 continue; 525 526 case SSLEEP: 527 case SSTOP: 528 /* 529 * do not swapout a realtime process 530 * Check all the thread groups.. 531 */ 532 FOREACH_KSEGRP_IN_PROC(p, kg) { 533 if (PRI_IS_REALTIME(kg->kg_pri.pri_class)) { 534 mtx_unlock_spin(&sched_lock); 535 PROC_UNLOCK(p); 536 goto nextproc; 537 } 538 539 /* 540 * Do not swapout a process waiting 541 * on a critical event of some kind. 542 * Also guarantee swap_idle_threshold1 543 * time in memory. 544 */ 545 if (((kg->kg_pri.pri_level) < PSOCK) || 546 (kg->kg_slptime < swap_idle_threshold1)) { 547 mtx_unlock_spin(&sched_lock); 548 PROC_UNLOCK(p); 549 goto nextproc; 550 } 551 552 /* 553 * If the system is under memory stress, 554 * or if we are swapping 555 * idle processes >= swap_idle_threshold2, 556 * then swap the process out. 557 */ 558 if (((action & VM_SWAP_NORMAL) == 0) && 559 (((action & VM_SWAP_IDLE) == 0) || 560 (kg->kg_slptime < swap_idle_threshold2))) { 561 mtx_unlock_spin(&sched_lock); 562 PROC_UNLOCK(p); 563 goto nextproc; 564 } 565 if (minslptime > kg->kg_slptime) 566 minslptime = kg->kg_slptime; 567 } 568 569 mtx_unlock_spin(&sched_lock); 570 ++vm->vm_refcnt; 571 /* 572 * do not swapout a process that 573 * is waiting for VM 574 * data structures there is a 575 * possible deadlock. 576 */ 577 if (lockmgr(&vm->vm_map.lock, 578 LK_EXCLUSIVE | LK_NOWAIT, 579 NULL, curthread)) { 580 vmspace_free(vm); 581 PROC_UNLOCK(p); 582 goto nextproc; 583 } 584 vm_map_unlock(&vm->vm_map); 585 /* 586 * If the process has been asleep for awhile and had 587 * most of its pages taken away already, swap it out. 588 */ 589 if ((action & VM_SWAP_NORMAL) || 590 ((action & VM_SWAP_IDLE) && 591 (minslptime > swap_idle_threshold2))) { 592 sx_sunlock(&allproc_lock); 593 swapout(p); 594 vmspace_free(vm); 595 didswap++; 596 goto retry; 597 } 598 PROC_UNLOCK(p); 599 vmspace_free(vm); 600 } 601 nextproc: 602 } 603 sx_sunlock(&allproc_lock); 604 /* 605 * If we swapped something out, and another process needed memory, 606 * then wakeup the sched process. 607 */ 608 if (didswap) 609 wakeup(&proc0); 610 } 611 612 static void 613 swapout(p) 614 struct proc *p; 615 { 616 struct thread *td; 617 618 PROC_LOCK_ASSERT(p, MA_OWNED); 619 #if defined(SWAP_DEBUG) 620 printf("swapping out %d\n", p->p_pid); 621 #endif 622 ++p->p_stats->p_ru.ru_nswap; 623 /* 624 * remember the process resident count 625 */ 626 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 627 628 mtx_lock_spin(&sched_lock); 629 p->p_sflag &= ~PS_INMEM; 630 p->p_sflag |= PS_SWAPPING; 631 PROC_UNLOCK(p); 632 FOREACH_THREAD_IN_PROC (p, td) 633 if (td->td_proc->p_stat == SRUN) /* XXXKSE */ 634 remrunqueue(td); /* XXXKSE */ 635 mtx_unlock_spin(&sched_lock); 636 637 pmap_swapout_proc(p); 638 FOREACH_THREAD_IN_PROC(p, td) 639 pmap_swapout_thread(td); 640 641 mtx_lock_spin(&sched_lock); 642 p->p_sflag &= ~PS_SWAPPING; 643 p->p_swtime = 0; 644 mtx_unlock_spin(&sched_lock); 645 } 646 #endif /* !NO_SWAPPING */ 647