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) 113 114 115 static void swapout __P((struct proc *)); 116 117 int 118 kernacc(addr, len, rw) 119 caddr_t addr; 120 int len, rw; 121 { 122 boolean_t rv; 123 vm_offset_t saddr, eaddr; 124 vm_prot_t prot; 125 126 KASSERT((rw & (~VM_PROT_ALL)) == 0, 127 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 128 prot = rw; 129 saddr = trunc_page((vm_offset_t)addr); 130 eaddr = round_page((vm_offset_t)addr + len); 131 vm_map_lock_read(kernel_map); 132 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 133 vm_map_unlock_read(kernel_map); 134 return (rv == TRUE); 135 } 136 137 int 138 useracc(addr, len, rw) 139 caddr_t addr; 140 int len, rw; 141 { 142 boolean_t rv; 143 vm_prot_t prot; 144 vm_map_t map; 145 vm_map_entry_t save_hint; 146 147 GIANT_REQUIRED; 148 149 KASSERT((rw & (~VM_PROT_ALL)) == 0, 150 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 151 prot = rw; 152 /* 153 * XXX - check separately to disallow access to user area and user 154 * page tables - they are in the map. 155 * 156 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 157 * only used (as an end address) in trap.c. Use it as an end address 158 * here too. This bogusness has spread. I just fixed where it was 159 * used as a max in vm_mmap.c. 160 */ 161 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 162 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 163 return (FALSE); 164 } 165 map = &curproc->p_vmspace->vm_map; 166 vm_map_lock_read(map); 167 /* 168 * We save the map hint, and restore it. Useracc appears to distort 169 * the map hint unnecessarily. 170 */ 171 save_hint = map->hint; 172 rv = vm_map_check_protection(map, 173 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot); 174 map->hint = save_hint; 175 vm_map_unlock_read(map); 176 177 return (rv == TRUE); 178 } 179 180 void 181 vslock(addr, len) 182 caddr_t addr; 183 u_int len; 184 { 185 GIANT_REQUIRED; 186 vm_map_pageable(&curproc->p_vmspace->vm_map, 187 trunc_page((vm_offset_t)addr), 188 round_page((vm_offset_t)addr + len), FALSE); 189 } 190 191 void 192 vsunlock(addr, len) 193 caddr_t addr; 194 u_int len; 195 { 196 GIANT_REQUIRED; 197 vm_map_pageable(&curproc->p_vmspace->vm_map, 198 trunc_page((vm_offset_t)addr), 199 round_page((vm_offset_t)addr + len), TRUE); 200 } 201 202 /* 203 * Implement fork's actions on an address space. 204 * Here we arrange for the address space to be copied or referenced, 205 * allocate a user struct (pcb and kernel stack), then call the 206 * machine-dependent layer to fill those in and make the new process 207 * ready to run. The new process is set up so that it returns directly 208 * to user mode to avoid stack copying and relocation problems. 209 */ 210 void 211 vm_forkproc(td, p2, flags) 212 struct thread *td; 213 struct proc *p2; 214 int flags; 215 { 216 struct proc *p1 = td->td_proc; 217 struct user *up; 218 219 GIANT_REQUIRED; 220 221 if ((flags & RFPROC) == 0) { 222 /* 223 * Divorce the memory, if it is shared, essentially 224 * this changes shared memory amongst threads, into 225 * COW locally. 226 */ 227 if ((flags & RFMEM) == 0) { 228 if (p1->p_vmspace->vm_refcnt > 1) { 229 vmspace_unshare(p1); 230 } 231 } 232 cpu_fork(td, p2, flags); 233 return; 234 } 235 236 if (flags & RFMEM) { 237 p2->p_vmspace = p1->p_vmspace; 238 p1->p_vmspace->vm_refcnt++; 239 } 240 241 while (vm_page_count_severe()) { 242 VM_WAIT; 243 } 244 245 if ((flags & RFMEM) == 0) { 246 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 247 248 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 249 250 if (p1->p_vmspace->vm_shm) 251 shmfork(p1, p2); 252 } 253 254 pmap_new_proc(p2); 255 pmap_new_thread(&p2->p_thread); /* Initial thread */ 256 257 /* XXXKSE this is unsatisfactory but should be adequate */ 258 up = p2->p_uarea; 259 260 /* 261 * p_stats currently points at fields in the user struct 262 * but not at &u, instead at p_addr. Copy parts of 263 * p_stats; zero the rest of p_stats (statistics). 264 * 265 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 266 * to share sigacts, so we use the up->u_sigacts. 267 */ 268 p2->p_stats = &up->u_stats; 269 if (p2->p_sigacts == NULL) { 270 if (p2->p_procsig->ps_refcnt != 1) 271 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 272 p2->p_sigacts = &up->u_sigacts; 273 up->u_sigacts = *p1->p_sigacts; 274 } 275 276 bzero(&up->u_stats.pstat_startzero, 277 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 278 (caddr_t) &up->u_stats.pstat_startzero)); 279 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 280 ((caddr_t) &up->u_stats.pstat_endcopy - 281 (caddr_t) &up->u_stats.pstat_startcopy)); 282 283 284 /* 285 * cpu_fork will copy and update the pcb, set up the kernel stack, 286 * and make the child ready to run. 287 */ 288 cpu_fork(td, p2, flags); 289 } 290 291 /* 292 * Called after process has been wait(2)'ed apon and is being reaped. 293 * The idea is to reclaim resources that we could not reclaim while 294 * the process was still executing. 295 */ 296 void 297 vm_waitproc(p) 298 struct proc *p; 299 { 300 struct thread *td; 301 302 GIANT_REQUIRED; 303 cpu_wait(p); 304 pmap_dispose_proc(p); /* drop per-process resources */ 305 FOREACH_THREAD_IN_PROC(p, td) 306 pmap_dispose_thread(td); 307 vmspace_free(p->p_vmspace); /* and clean-out the vmspace */ 308 } 309 310 /* 311 * Set default limits for VM system. 312 * Called for proc 0, and then inherited by all others. 313 * 314 * XXX should probably act directly on proc0. 315 */ 316 static void 317 vm_init_limits(udata) 318 void *udata; 319 { 320 struct proc *p = udata; 321 int rss_limit; 322 323 /* 324 * Set up the initial limits on process VM. Set the maximum resident 325 * set size to be half of (reasonably) available memory. Since this 326 * is a soft limit, it comes into effect only when the system is out 327 * of memory - half of main memory helps to favor smaller processes, 328 * and reduces thrashing of the object cache. 329 */ 330 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 331 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 332 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 333 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 334 /* limit the limit to no less than 2MB */ 335 rss_limit = max(cnt.v_free_count, 512); 336 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 337 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 338 } 339 340 /* 341 * Must be called with the proc struc mutex held. 342 */ 343 void 344 faultin(p) 345 struct proc *p; 346 { 347 struct thread *td; 348 GIANT_REQUIRED; 349 350 PROC_LOCK_ASSERT(p, MA_OWNED); 351 mtx_lock_spin(&sched_lock); 352 if ((p->p_sflag & PS_INMEM) == 0) { 353 ++p->p_lock; 354 mtx_unlock_spin(&sched_lock); 355 PROC_UNLOCK(p); 356 357 pmap_swapin_proc(p); 358 FOREACH_THREAD_IN_PROC (p, td) 359 pmap_swapin_thread(td); 360 361 PROC_LOCK(p); 362 mtx_lock_spin(&sched_lock); 363 FOREACH_THREAD_IN_PROC (p, td) 364 if (td->td_proc->p_stat == SRUN) /* XXXKSE */ 365 setrunqueue(td); 366 367 p->p_sflag |= PS_INMEM; 368 369 /* undo the effect of setting SLOCK above */ 370 --p->p_lock; 371 } 372 mtx_unlock_spin(&sched_lock); 373 } 374 375 /* 376 * This swapin algorithm attempts to swap-in processes only if there 377 * is enough space for them. Of course, if a process waits for a long 378 * time, it will be swapped in anyway. 379 * 380 * XXXKSE - KSEGRP with highest priority counts.. 381 * 382 * Giant is still held at this point, to be released in tsleep. 383 */ 384 /* ARGSUSED*/ 385 static void 386 scheduler(dummy) 387 void *dummy; 388 { 389 struct proc *p; 390 int pri; 391 struct proc *pp; 392 int ppri; 393 394 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED); 395 /* GIANT_REQUIRED */ 396 397 loop: 398 if (vm_page_count_min()) { 399 VM_WAIT; 400 goto loop; 401 } 402 403 pp = NULL; 404 ppri = INT_MIN; 405 sx_slock(&allproc_lock); 406 FOREACH_PROC_IN_SYSTEM(p) { 407 struct ksegrp *kg; 408 mtx_lock_spin(&sched_lock); 409 if (p->p_stat == SRUN 410 && (p->p_sflag & (PS_INMEM | PS_SWAPPING)) == 0) { 411 /* Find the minimum sleeptime for the process */ 412 FOREACH_KSEGRP_IN_PROC(p, kg) { 413 pri = p->p_swtime + kg->kg_slptime; 414 if ((p->p_sflag & PS_SWAPINREQ) == 0) { 415 pri -= kg->kg_nice * 8; 416 } 417 418 419 /* 420 * if this ksegrp is higher priority 421 * and there is enough space, then select 422 * this process instead of the previous 423 * selection. 424 */ 425 if (pri > ppri) { 426 pp = p; 427 ppri = pri; 428 } 429 } 430 } 431 mtx_unlock_spin(&sched_lock); 432 } 433 sx_sunlock(&allproc_lock); 434 435 /* 436 * Nothing to do, back to sleep. 437 */ 438 if ((p = pp) == NULL) { 439 tsleep(&proc0, PVM, "sched", maxslp * hz / 2); 440 goto loop; 441 } 442 mtx_lock_spin(&sched_lock); 443 p->p_sflag &= ~PS_SWAPINREQ; 444 mtx_unlock_spin(&sched_lock); 445 446 /* 447 * We would like to bring someone in. (only if there is space). 448 */ 449 PROC_LOCK(p); 450 faultin(p); 451 PROC_UNLOCK(p); 452 mtx_lock_spin(&sched_lock); 453 p->p_swtime = 0; 454 mtx_unlock_spin(&sched_lock); 455 goto loop; 456 } 457 458 #ifndef NO_SWAPPING 459 460 /* 461 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 462 */ 463 static int swap_idle_threshold1 = 2; 464 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 465 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 466 467 /* 468 * Swap_idle_threshold2 is the time that a process can be idle before 469 * it will be swapped out, if idle swapping is enabled. 470 */ 471 static int swap_idle_threshold2 = 10; 472 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 473 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 474 475 /* 476 * Swapout is driven by the pageout daemon. Very simple, we find eligible 477 * procs and unwire their u-areas. We try to always "swap" at least one 478 * process in case we need the room for a swapin. 479 * If any procs have been sleeping/stopped for at least maxslp seconds, 480 * they are swapped. Else, we swap the longest-sleeping or stopped process, 481 * if any, otherwise the longest-resident process. 482 */ 483 void 484 swapout_procs(action) 485 int action; 486 { 487 struct proc *p; 488 struct ksegrp *kg; 489 struct proc *outp, *outp2; 490 int outpri, outpri2; 491 int didswap = 0; 492 493 GIANT_REQUIRED; 494 495 outp = outp2 = NULL; 496 outpri = outpri2 = INT_MIN; 497 retry: 498 sx_slock(&allproc_lock); 499 LIST_FOREACH(p, &allproc, p_list) { 500 struct vmspace *vm; 501 int minslptime = 100000; 502 503 PROC_LOCK(p); 504 if (p->p_lock != 0 || 505 (p->p_flag & (P_TRACED|P_SYSTEM|P_WEXIT)) != 0) { 506 PROC_UNLOCK(p); 507 continue; 508 } 509 /* 510 * only aiod changes vmspace, however it will be 511 * skipped because of the if statement above checking 512 * for P_SYSTEM 513 */ 514 vm = p->p_vmspace; 515 mtx_lock_spin(&sched_lock); 516 if ((p->p_sflag & (PS_INMEM|PS_SWAPPING)) != PS_INMEM) { 517 mtx_unlock_spin(&sched_lock); 518 PROC_UNLOCK(p); 519 continue; 520 } 521 522 switch (p->p_stat) { 523 default: 524 mtx_unlock_spin(&sched_lock); 525 PROC_UNLOCK(p); 526 continue; 527 528 case SSLEEP: 529 case SSTOP: 530 /* 531 * do not swapout a realtime process 532 * Check all the thread groups.. 533 */ 534 FOREACH_KSEGRP_IN_PROC(p, kg) { 535 if (PRI_IS_REALTIME(kg->kg_pri.pri_class)) { 536 mtx_unlock_spin(&sched_lock); 537 PROC_UNLOCK(p); 538 goto nextproc; 539 } 540 541 /* 542 * Do not swapout a process waiting 543 * on a critical event of some kind. 544 * Also guarantee swap_idle_threshold1 545 * time in memory. 546 */ 547 if (((kg->kg_pri.pri_level) < PSOCK) || 548 (kg->kg_slptime < swap_idle_threshold1)) { 549 mtx_unlock_spin(&sched_lock); 550 PROC_UNLOCK(p); 551 goto nextproc; 552 } 553 554 /* 555 * If the system is under memory stress, 556 * or if we are swapping 557 * idle processes >= swap_idle_threshold2, 558 * then swap the process out. 559 */ 560 if (((action & VM_SWAP_NORMAL) == 0) && 561 (((action & VM_SWAP_IDLE) == 0) || 562 (kg->kg_slptime < swap_idle_threshold2))) { 563 mtx_unlock_spin(&sched_lock); 564 PROC_UNLOCK(p); 565 goto nextproc; 566 } 567 if (minslptime > kg->kg_slptime) 568 minslptime = kg->kg_slptime; 569 } 570 571 mtx_unlock_spin(&sched_lock); 572 ++vm->vm_refcnt; 573 /* 574 * do not swapout a process that 575 * is waiting for VM 576 * data structures there is a 577 * possible deadlock. 578 */ 579 if (lockmgr(&vm->vm_map.lock, 580 LK_EXCLUSIVE | LK_NOWAIT, 581 NULL, curthread)) { 582 vmspace_free(vm); 583 PROC_UNLOCK(p); 584 goto nextproc; 585 } 586 vm_map_unlock(&vm->vm_map); 587 /* 588 * If the process has been asleep for awhile and had 589 * most of its pages taken away already, swap it out. 590 */ 591 if ((action & VM_SWAP_NORMAL) || 592 ((action & VM_SWAP_IDLE) && 593 (minslptime > swap_idle_threshold2))) { 594 sx_sunlock(&allproc_lock); 595 swapout(p); 596 vmspace_free(vm); 597 didswap++; 598 goto retry; 599 } 600 PROC_UNLOCK(p); 601 vmspace_free(vm); 602 } 603 nextproc: 604 } 605 sx_sunlock(&allproc_lock); 606 /* 607 * If we swapped something out, and another process needed memory, 608 * then wakeup the sched process. 609 */ 610 if (didswap) 611 wakeup(&proc0); 612 } 613 614 static void 615 swapout(p) 616 struct proc *p; 617 { 618 struct thread *td; 619 620 PROC_LOCK_ASSERT(p, MA_OWNED); 621 #if defined(SWAP_DEBUG) 622 printf("swapping out %d\n", p->p_pid); 623 #endif 624 ++p->p_stats->p_ru.ru_nswap; 625 /* 626 * remember the process resident count 627 */ 628 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 629 630 mtx_lock_spin(&sched_lock); 631 p->p_sflag &= ~PS_INMEM; 632 p->p_sflag |= PS_SWAPPING; 633 PROC_UNLOCK_NOSWITCH(p); 634 FOREACH_THREAD_IN_PROC (p, td) 635 if (td->td_proc->p_stat == SRUN) /* XXXKSE */ 636 remrunqueue(td); /* XXXKSE */ 637 mtx_unlock_spin(&sched_lock); 638 639 pmap_swapout_proc(p); 640 FOREACH_THREAD_IN_PROC(p, td) 641 pmap_swapout_thread(td); 642 643 mtx_lock_spin(&sched_lock); 644 p->p_sflag &= ~PS_SWAPPING; 645 p->p_swtime = 0; 646 mtx_unlock_spin(&sched_lock); 647 } 648 #endif /* !NO_SWAPPING */ 649