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 * $Id: vm_glue.c,v 1.56 1996/12/22 23:17:09 joerg Exp $ 63 */ 64 65 #include "opt_rlimit.h" 66 67 #include <sys/param.h> 68 #include <sys/systm.h> 69 #include <sys/proc.h> 70 #include <sys/resourcevar.h> 71 #include <sys/buf.h> 72 #include <sys/shm.h> 73 #include <sys/vmmeter.h> 74 75 #include <sys/kernel.h> 76 #include <sys/dkstat.h> 77 78 #include <vm/vm.h> 79 #include <vm/vm_param.h> 80 #include <vm/vm_inherit.h> 81 #include <vm/vm_prot.h> 82 #include <vm/lock.h> 83 #include <vm/pmap.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_pageout.h> 87 #include <vm/vm_kern.h> 88 #include <vm/vm_extern.h> 89 #include <vm/vm_object.h> 90 #include <vm/vm_pager.h> 91 92 #include <sys/user.h> 93 94 /* 95 * System initialization 96 * 97 * Note: proc0 from proc.h 98 */ 99 100 static void vm_init_limits __P((void *)); 101 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 102 103 /* 104 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 105 * 106 * Note: run scheduling should be divorced from the vm system. 107 */ 108 static void scheduler __P((void *)); 109 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 110 111 112 static void swapout __P((struct proc *)); 113 114 extern char kstack[]; 115 116 /* vm_map_t upages_map; */ 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 = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 126 127 saddr = trunc_page(addr); 128 eaddr = round_page(addr + len); 129 vm_map_lock_read(kernel_map); 130 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 131 vm_map_unlock_read(kernel_map); 132 return (rv == TRUE); 133 } 134 135 int 136 useracc(addr, len, rw) 137 caddr_t addr; 138 int len, rw; 139 { 140 boolean_t rv; 141 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 142 vm_map_t map; 143 vm_map_entry_t save_hint; 144 145 /* 146 * XXX - check separately to disallow access to user area and user 147 * page tables - they are in the map. 148 * 149 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 150 * only used (as an end address) in trap.c. Use it as an end address 151 * here too. This bogusness has spread. I just fixed where it was 152 * used as a max in vm_mmap.c. 153 */ 154 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 155 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 156 return (FALSE); 157 } 158 map = &curproc->p_vmspace->vm_map; 159 vm_map_lock_read(map); 160 /* 161 * We save the map hint, and restore it. Useracc appears to distort 162 * the map hint unnecessarily. 163 */ 164 save_hint = map->hint; 165 rv = vm_map_check_protection(map, 166 trunc_page(addr), round_page(addr + len), prot); 167 map->hint = save_hint; 168 vm_map_unlock_read(map); 169 170 return (rv == TRUE); 171 } 172 173 void 174 vslock(addr, len) 175 caddr_t addr; 176 u_int len; 177 { 178 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 179 round_page(addr + len), FALSE); 180 } 181 182 void 183 vsunlock(addr, len, dirtied) 184 caddr_t addr; 185 u_int len; 186 int dirtied; 187 { 188 #ifdef lint 189 dirtied++; 190 #endif /* lint */ 191 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 192 round_page(addr + len), TRUE); 193 } 194 195 /* 196 * Implement fork's actions on an address space. 197 * Here we arrange for the address space to be copied or referenced, 198 * allocate a user struct (pcb and kernel stack), then call the 199 * machine-dependent layer to fill those in and make the new process 200 * ready to run. 201 * NOTE: the kernel stack may be at a different location in the child 202 * process, and thus addresses of automatic variables may be invalid 203 * after cpu_fork returns in the child process. We do nothing here 204 * after cpu_fork returns. 205 */ 206 int 207 vm_fork(p1, p2) 208 register struct proc *p1, *p2; 209 { 210 register struct user *up; 211 int i; 212 pmap_t pvp; 213 vm_object_t upobj; 214 215 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 216 VM_WAIT; 217 } 218 219 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 220 221 if (p1->p_vmspace->vm_shm) 222 shmfork(p1, p2); 223 224 pmap_new_proc(p2); 225 226 up = p2->p_addr; 227 228 /* 229 * p_stats and p_sigacts currently point at fields in the user struct 230 * but not at &u, instead at p_addr. Copy p_sigacts and parts of 231 * p_stats; zero the rest of p_stats (statistics). 232 */ 233 p2->p_stats = &up->u_stats; 234 p2->p_sigacts = &up->u_sigacts; 235 up->u_sigacts = *p1->p_sigacts; 236 bzero(&up->u_stats.pstat_startzero, 237 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 238 (caddr_t) &up->u_stats.pstat_startzero)); 239 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 240 ((caddr_t) &up->u_stats.pstat_endcopy - 241 (caddr_t) &up->u_stats.pstat_startcopy)); 242 243 244 /* 245 * cpu_fork will copy and update the kernel stack and pcb, and make 246 * the child ready to run. It marks the child so that it can return 247 * differently than the parent. It returns twice, once in the parent 248 * process and once in the child. 249 */ 250 return (cpu_fork(p1, p2)); 251 } 252 253 /* 254 * Set default limits for VM system. 255 * Called for proc 0, and then inherited by all others. 256 * 257 * XXX should probably act directly on proc0. 258 */ 259 static void 260 vm_init_limits(udata) 261 void *udata; 262 { 263 register struct proc *p = udata; 264 int rss_limit; 265 266 /* 267 * Set up the initial limits on process VM. Set the maximum resident 268 * set size to be half of (reasonably) available memory. Since this 269 * is a soft limit, it comes into effect only when the system is out 270 * of memory - half of main memory helps to favor smaller processes, 271 * and reduces thrashing of the object cache. 272 */ 273 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 274 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 275 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 276 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 277 /* limit the limit to no less than 2MB */ 278 rss_limit = max(cnt.v_free_count, 512); 279 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 280 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 281 } 282 283 void 284 faultin(p) 285 struct proc *p; 286 { 287 vm_offset_t i; 288 int s; 289 290 if ((p->p_flag & P_INMEM) == 0) { 291 292 ++p->p_lock; 293 294 pmap_swapin_proc(p); 295 296 s = splhigh(); 297 298 if (p->p_stat == SRUN) 299 setrunqueue(p); 300 301 p->p_flag |= P_INMEM; 302 303 /* undo the effect of setting SLOCK above */ 304 --p->p_lock; 305 splx(s); 306 307 } 308 } 309 310 /* 311 * This swapin algorithm attempts to swap-in processes only if there 312 * is enough space for them. Of course, if a process waits for a long 313 * time, it will be swapped in anyway. 314 */ 315 /* ARGSUSED*/ 316 static void 317 scheduler(dummy) 318 void *dummy; 319 { 320 register struct proc *p; 321 register int pri; 322 struct proc *pp; 323 int ppri; 324 325 spl0(); 326 loop: 327 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 328 VM_WAIT; 329 } 330 331 pp = NULL; 332 ppri = INT_MIN; 333 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 334 if (p->p_stat == SRUN && 335 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 336 int mempri; 337 338 pri = p->p_swtime + p->p_slptime; 339 if ((p->p_flag & P_SWAPINREQ) == 0) { 340 pri -= p->p_nice * 8; 341 } 342 mempri = pri > 0 ? pri : 0; 343 /* 344 * if this process is higher priority and there is 345 * enough space, then select this process instead of 346 * the previous selection. 347 */ 348 if (pri > ppri) { 349 pp = p; 350 ppri = pri; 351 } 352 } 353 } 354 355 /* 356 * Nothing to do, back to sleep. 357 */ 358 if ((p = pp) == NULL) { 359 tsleep(&proc0, PVM, "sched", 0); 360 goto loop; 361 } 362 p->p_flag &= ~P_SWAPINREQ; 363 364 /* 365 * We would like to bring someone in. (only if there is space). 366 */ 367 faultin(p); 368 p->p_swtime = 0; 369 goto loop; 370 } 371 372 #ifndef NO_SWAPPING 373 374 #define swappable(p) \ 375 (((p)->p_lock == 0) && \ 376 ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM) 377 378 /* 379 * Swapout is driven by the pageout daemon. Very simple, we find eligible 380 * procs and unwire their u-areas. We try to always "swap" at least one 381 * process in case we need the room for a swapin. 382 * If any procs have been sleeping/stopped for at least maxslp seconds, 383 * they are swapped. Else, we swap the longest-sleeping or stopped process, 384 * if any, otherwise the longest-resident process. 385 */ 386 void 387 swapout_procs() 388 { 389 register struct proc *p; 390 struct proc *outp, *outp2; 391 int outpri, outpri2; 392 int didswap = 0; 393 394 outp = outp2 = NULL; 395 outpri = outpri2 = INT_MIN; 396 retry: 397 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 398 struct vmspace *vm; 399 if (!swappable(p)) 400 continue; 401 402 vm = p->p_vmspace; 403 404 switch (p->p_stat) { 405 default: 406 continue; 407 408 case SSLEEP: 409 case SSTOP: 410 /* 411 * do not swapout a realtime process 412 */ 413 if (p->p_rtprio.type == RTP_PRIO_REALTIME) 414 continue; 415 416 /* 417 * do not swapout a process waiting on a critical 418 * event of some kind 419 */ 420 if (((p->p_priority & 0x7f) < PSOCK) || 421 (p->p_slptime <= 10)) 422 continue; 423 424 ++vm->vm_refcnt; 425 vm_map_reference(&vm->vm_map); 426 /* 427 * do not swapout a process that is waiting for VM 428 * datastructures there is a possible deadlock. 429 */ 430 if (!lock_try_write(&vm->vm_map.lock)) { 431 vm_map_deallocate(&vm->vm_map); 432 vmspace_free(vm); 433 continue; 434 } 435 vm_map_unlock(&vm->vm_map); 436 /* 437 * If the process has been asleep for awhile and had 438 * most of its pages taken away already, swap it out. 439 */ 440 swapout(p); 441 vm_map_deallocate(&vm->vm_map); 442 vmspace_free(vm); 443 didswap++; 444 goto retry; 445 } 446 } 447 /* 448 * If we swapped something out, and another process needed memory, 449 * then wakeup the sched process. 450 */ 451 if (didswap) 452 wakeup(&proc0); 453 } 454 455 static void 456 swapout(p) 457 register struct proc *p; 458 { 459 pmap_t pmap = &p->p_vmspace->vm_pmap; 460 int i; 461 462 #if defined(SWAP_DEBUG) 463 printf("swapping out %d\n", p->p_pid); 464 #endif 465 ++p->p_stats->p_ru.ru_nswap; 466 /* 467 * remember the process resident count 468 */ 469 p->p_vmspace->vm_swrss = 470 p->p_vmspace->vm_pmap.pm_stats.resident_count; 471 472 (void) splhigh(); 473 p->p_flag &= ~P_INMEM; 474 p->p_flag |= P_SWAPPING; 475 if (p->p_stat == SRUN) 476 remrq(p); 477 (void) spl0(); 478 479 pmap_swapout_proc(p); 480 481 p->p_flag &= ~P_SWAPPING; 482 p->p_swtime = 0; 483 } 484 #endif /* !NO_SWAPPING */ 485