Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_glue.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 33 * 34 * 35 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 36 * All rights reserved. 37 * 38 * Permission to use, copy, modify and distribute this software and 39 * its documentation is hereby granted, provided that both the copyright 40 * notice and this permission notice appear in all copies of the 41 * software, derivative works or modified versions, and any portions 42 * thereof, and that both notices appear in supporting documentation. 43 * 44 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 45 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 46 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 47 * 48 * Carnegie Mellon requests users of this software to return to 49 * 50 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 51 * School of Computer Science 52 * Carnegie Mellon University 53 * Pittsburgh PA 15213-3890 54 * 55 * any improvements or extensions that they make and grant Carnegie the 56 * rights to redistribute these changes. 57 */ 58 59 #include <sys/cdefs.h> 60 __FBSDID("$FreeBSD$"); 61 62 #include "opt_vm.h" 63 #include "opt_kstack_pages.h" 64 #include "opt_kstack_max_pages.h" 65 66 #include <sys/param.h> 67 #include <sys/systm.h> 68 #include <sys/limits.h> 69 #include <sys/lock.h> 70 #include <sys/mutex.h> 71 #include <sys/proc.h> 72 #include <sys/resourcevar.h> 73 #include <sys/sched.h> 74 #include <sys/sf_buf.h> 75 #include <sys/shm.h> 76 #include <sys/vmmeter.h> 77 #include <sys/sx.h> 78 #include <sys/sysctl.h> 79 80 #include <sys/eventhandler.h> 81 #include <sys/kernel.h> 82 #include <sys/ktr.h> 83 #include <sys/unistd.h> 84 85 #include <vm/vm.h> 86 #include <vm/vm_param.h> 87 #include <vm/pmap.h> 88 #include <vm/vm_map.h> 89 #include <vm/vm_page.h> 90 #include <vm/vm_pageout.h> 91 #include <vm/vm_object.h> 92 #include <vm/vm_kern.h> 93 #include <vm/vm_extern.h> 94 #include <vm/vm_pager.h> 95 #include <vm/swap_pager.h> 96 97 extern int maxslp; 98 99 /* 100 * System initialization 101 * 102 * Note: proc0 from proc.h 103 */ 104 static void vm_init_limits(void *); 105 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0); 106 107 #ifndef NO_SWAPPING 108 static int swapout(struct proc *); 109 static void swapclear(struct proc *); 110 #endif 111 112 /* 113 * MPSAFE 114 * 115 * WARNING! This code calls vm_map_check_protection() which only checks 116 * the associated vm_map_entry range. It does not determine whether the 117 * contents of the memory is actually readable or writable. In most cases 118 * just checking the vm_map_entry is sufficient within the kernel's address 119 * space. 120 */ 121 int 122 kernacc(addr, len, rw) 123 void *addr; 124 int len, rw; 125 { 126 boolean_t rv; 127 vm_offset_t saddr, eaddr; 128 vm_prot_t prot; 129 130 KASSERT((rw & ~VM_PROT_ALL) == 0, 131 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 132 133 if ((vm_offset_t)addr + len > kernel_map->max_offset || 134 (vm_offset_t)addr + len < (vm_offset_t)addr) 135 return (FALSE); 136 137 prot = rw; 138 saddr = trunc_page((vm_offset_t)addr); 139 eaddr = round_page((vm_offset_t)addr + len); 140 vm_map_lock_read(kernel_map); 141 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 142 vm_map_unlock_read(kernel_map); 143 return (rv == TRUE); 144 } 145 146 /* 147 * MPSAFE 148 * 149 * WARNING! This code calls vm_map_check_protection() which only checks 150 * the associated vm_map_entry range. It does not determine whether the 151 * contents of the memory is actually readable or writable. vmapbuf(), 152 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be 153 * used in conjuction with this call. 154 */ 155 int 156 useracc(addr, len, rw) 157 void *addr; 158 int len, rw; 159 { 160 boolean_t rv; 161 vm_prot_t prot; 162 vm_map_t map; 163 164 KASSERT((rw & ~VM_PROT_ALL) == 0, 165 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 166 prot = rw; 167 map = &curproc->p_vmspace->vm_map; 168 if ((vm_offset_t)addr + len > vm_map_max(map) || 169 (vm_offset_t)addr + len < (vm_offset_t)addr) { 170 return (FALSE); 171 } 172 vm_map_lock_read(map); 173 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 174 round_page((vm_offset_t)addr + len), prot); 175 vm_map_unlock_read(map); 176 return (rv == TRUE); 177 } 178 179 int 180 vslock(void *addr, size_t len) 181 { 182 vm_offset_t end, last, start; 183 vm_size_t npages; 184 int error; 185 186 last = (vm_offset_t)addr + len; 187 start = trunc_page((vm_offset_t)addr); 188 end = round_page(last); 189 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr) 190 return (EINVAL); 191 npages = atop(end - start); 192 if (npages > vm_page_max_wired) 193 return (ENOMEM); 194 #if 0 195 /* 196 * XXX - not yet 197 * 198 * The limit for transient usage of wired pages should be 199 * larger than for "permanent" wired pages (mlock()). 200 * 201 * Also, the sysctl code, which is the only present user 202 * of vslock(), does a hard loop on EAGAIN. 203 */ 204 if (npages + cnt.v_wire_count > vm_page_max_wired) 205 return (EAGAIN); 206 #endif 207 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end, 208 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 209 /* 210 * Return EFAULT on error to match copy{in,out}() behaviour 211 * rather than returning ENOMEM like mlock() would. 212 */ 213 return (error == KERN_SUCCESS ? 0 : EFAULT); 214 } 215 216 void 217 vsunlock(void *addr, size_t len) 218 { 219 220 /* Rely on the parameter sanity checks performed by vslock(). */ 221 (void)vm_map_unwire(&curproc->p_vmspace->vm_map, 222 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), 223 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 224 } 225 226 /* 227 * Pin the page contained within the given object at the given offset. If the 228 * page is not resident, allocate and load it using the given object's pager. 229 * Return the pinned page if successful; otherwise, return NULL. 230 */ 231 static vm_page_t 232 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset) 233 { 234 vm_page_t m, ma[1]; 235 vm_pindex_t pindex; 236 int rv; 237 238 VM_OBJECT_LOCK(object); 239 pindex = OFF_TO_IDX(offset); 240 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 241 if (m->valid != VM_PAGE_BITS_ALL) { 242 ma[0] = m; 243 rv = vm_pager_get_pages(object, ma, 1, 0); 244 m = vm_page_lookup(object, pindex); 245 if (m == NULL) 246 goto out; 247 if (rv != VM_PAGER_OK) { 248 vm_page_lock_queues(); 249 vm_page_free(m); 250 vm_page_unlock_queues(); 251 m = NULL; 252 goto out; 253 } 254 } 255 vm_page_lock_queues(); 256 vm_page_hold(m); 257 vm_page_unlock_queues(); 258 vm_page_wakeup(m); 259 out: 260 VM_OBJECT_UNLOCK(object); 261 return (m); 262 } 263 264 /* 265 * Return a CPU private mapping to the page at the given offset within the 266 * given object. The page is pinned before it is mapped. 267 */ 268 struct sf_buf * 269 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset) 270 { 271 vm_page_t m; 272 273 m = vm_imgact_hold_page(object, offset); 274 if (m == NULL) 275 return (NULL); 276 sched_pin(); 277 return (sf_buf_alloc(m, SFB_CPUPRIVATE)); 278 } 279 280 /* 281 * Destroy the given CPU private mapping and unpin the page that it mapped. 282 */ 283 void 284 vm_imgact_unmap_page(struct sf_buf *sf) 285 { 286 vm_page_t m; 287 288 m = sf_buf_page(sf); 289 sf_buf_free(sf); 290 sched_unpin(); 291 vm_page_lock_queues(); 292 vm_page_unhold(m); 293 vm_page_unlock_queues(); 294 } 295 296 void 297 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz) 298 { 299 300 pmap_sync_icache(map->pmap, va, sz); 301 } 302 303 struct kstack_cache_entry { 304 vm_object_t ksobj; 305 struct kstack_cache_entry *next_ks_entry; 306 }; 307 308 static struct kstack_cache_entry *kstack_cache; 309 static int kstack_cache_size = 128; 310 static int kstacks; 311 static struct mtx kstack_cache_mtx; 312 SYSCTL_INT(_vm, OID_AUTO, kstack_cache_size, CTLFLAG_RW, &kstack_cache_size, 0, 313 ""); 314 SYSCTL_INT(_vm, OID_AUTO, kstacks, CTLFLAG_RD, &kstacks, 0, 315 ""); 316 317 #ifndef KSTACK_MAX_PAGES 318 #define KSTACK_MAX_PAGES 32 319 #endif 320 321 /* 322 * Create the kernel stack (including pcb for i386) for a new thread. 323 * This routine directly affects the fork perf for a process and 324 * create performance for a thread. 325 */ 326 int 327 vm_thread_new(struct thread *td, int pages) 328 { 329 vm_object_t ksobj; 330 vm_offset_t ks; 331 vm_page_t m, ma[KSTACK_MAX_PAGES]; 332 struct kstack_cache_entry *ks_ce; 333 int i; 334 335 /* Bounds check */ 336 if (pages <= 1) 337 pages = KSTACK_PAGES; 338 else if (pages > KSTACK_MAX_PAGES) 339 pages = KSTACK_MAX_PAGES; 340 341 if (pages == KSTACK_PAGES) { 342 mtx_lock(&kstack_cache_mtx); 343 if (kstack_cache != NULL) { 344 ks_ce = kstack_cache; 345 kstack_cache = ks_ce->next_ks_entry; 346 mtx_unlock(&kstack_cache_mtx); 347 348 td->td_kstack_obj = ks_ce->ksobj; 349 td->td_kstack = (vm_offset_t)ks_ce; 350 td->td_kstack_pages = KSTACK_PAGES; 351 return (1); 352 } 353 mtx_unlock(&kstack_cache_mtx); 354 } 355 356 /* 357 * Allocate an object for the kstack. 358 */ 359 ksobj = vm_object_allocate(OBJT_DEFAULT, pages); 360 361 /* 362 * Get a kernel virtual address for this thread's kstack. 363 */ 364 #if defined(__mips__) 365 /* 366 * We need to align the kstack's mapped address to fit within 367 * a single TLB entry. 368 */ 369 ks = kmem_alloc_nofault_space(kernel_map, 370 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE, VMFS_TLB_ALIGNED_SPACE); 371 #else 372 ks = kmem_alloc_nofault(kernel_map, 373 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 374 #endif 375 if (ks == 0) { 376 printf("vm_thread_new: kstack allocation failed\n"); 377 vm_object_deallocate(ksobj); 378 return (0); 379 } 380 381 atomic_add_int(&kstacks, 1); 382 if (KSTACK_GUARD_PAGES != 0) { 383 pmap_qremove(ks, KSTACK_GUARD_PAGES); 384 ks += KSTACK_GUARD_PAGES * PAGE_SIZE; 385 } 386 td->td_kstack_obj = ksobj; 387 td->td_kstack = ks; 388 /* 389 * Knowing the number of pages allocated is useful when you 390 * want to deallocate them. 391 */ 392 td->td_kstack_pages = pages; 393 /* 394 * For the length of the stack, link in a real page of ram for each 395 * page of stack. 396 */ 397 VM_OBJECT_LOCK(ksobj); 398 for (i = 0; i < pages; i++) { 399 /* 400 * Get a kernel stack page. 401 */ 402 m = vm_page_grab(ksobj, i, VM_ALLOC_NOBUSY | 403 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_WIRED); 404 ma[i] = m; 405 m->valid = VM_PAGE_BITS_ALL; 406 } 407 VM_OBJECT_UNLOCK(ksobj); 408 pmap_qenter(ks, ma, pages); 409 return (1); 410 } 411 412 static void 413 vm_thread_stack_dispose(vm_object_t ksobj, vm_offset_t ks, int pages) 414 { 415 vm_page_t m; 416 int i; 417 418 atomic_add_int(&kstacks, -1); 419 pmap_qremove(ks, pages); 420 VM_OBJECT_LOCK(ksobj); 421 for (i = 0; i < pages; i++) { 422 m = vm_page_lookup(ksobj, i); 423 if (m == NULL) 424 panic("vm_thread_dispose: kstack already missing?"); 425 vm_page_lock_queues(); 426 vm_page_unwire(m, 0); 427 vm_page_free(m); 428 vm_page_unlock_queues(); 429 } 430 VM_OBJECT_UNLOCK(ksobj); 431 vm_object_deallocate(ksobj); 432 kmem_free(kernel_map, ks - (KSTACK_GUARD_PAGES * PAGE_SIZE), 433 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 434 } 435 436 /* 437 * Dispose of a thread's kernel stack. 438 */ 439 void 440 vm_thread_dispose(struct thread *td) 441 { 442 vm_object_t ksobj; 443 vm_offset_t ks; 444 struct kstack_cache_entry *ks_ce; 445 int pages; 446 447 pages = td->td_kstack_pages; 448 ksobj = td->td_kstack_obj; 449 ks = td->td_kstack; 450 td->td_kstack = 0; 451 td->td_kstack_pages = 0; 452 if (pages == KSTACK_PAGES && kstacks <= kstack_cache_size) { 453 ks_ce = (struct kstack_cache_entry *)ks; 454 ks_ce->ksobj = ksobj; 455 mtx_lock(&kstack_cache_mtx); 456 ks_ce->next_ks_entry = kstack_cache; 457 kstack_cache = ks_ce; 458 mtx_unlock(&kstack_cache_mtx); 459 return; 460 } 461 vm_thread_stack_dispose(ksobj, ks, pages); 462 } 463 464 static void 465 vm_thread_stack_lowmem(void *nulll) 466 { 467 struct kstack_cache_entry *ks_ce, *ks_ce1; 468 469 mtx_lock(&kstack_cache_mtx); 470 ks_ce = kstack_cache; 471 kstack_cache = NULL; 472 mtx_unlock(&kstack_cache_mtx); 473 474 while (ks_ce != NULL) { 475 ks_ce1 = ks_ce; 476 ks_ce = ks_ce->next_ks_entry; 477 478 vm_thread_stack_dispose(ks_ce1->ksobj, (vm_offset_t)ks_ce1, 479 KSTACK_PAGES); 480 } 481 } 482 483 static void 484 kstack_cache_init(void *nulll) 485 { 486 487 EVENTHANDLER_REGISTER(vm_lowmem, vm_thread_stack_lowmem, NULL, 488 EVENTHANDLER_PRI_ANY); 489 } 490 491 MTX_SYSINIT(kstack_cache, &kstack_cache_mtx, "kstkch", MTX_DEF); 492 SYSINIT(vm_kstacks, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY, kstack_cache_init, NULL); 493 494 /* 495 * Allow a thread's kernel stack to be paged out. 496 */ 497 void 498 vm_thread_swapout(struct thread *td) 499 { 500 vm_object_t ksobj; 501 vm_page_t m; 502 int i, pages; 503 504 cpu_thread_swapout(td); 505 pages = td->td_kstack_pages; 506 ksobj = td->td_kstack_obj; 507 pmap_qremove(td->td_kstack, pages); 508 VM_OBJECT_LOCK(ksobj); 509 for (i = 0; i < pages; i++) { 510 m = vm_page_lookup(ksobj, i); 511 if (m == NULL) 512 panic("vm_thread_swapout: kstack already missing?"); 513 vm_page_lock_queues(); 514 vm_page_dirty(m); 515 vm_page_unwire(m, 0); 516 vm_page_unlock_queues(); 517 } 518 VM_OBJECT_UNLOCK(ksobj); 519 } 520 521 /* 522 * Bring the kernel stack for a specified thread back in. 523 */ 524 void 525 vm_thread_swapin(struct thread *td) 526 { 527 vm_object_t ksobj; 528 vm_page_t ma[KSTACK_MAX_PAGES]; 529 int i, j, k, pages, rv; 530 531 pages = td->td_kstack_pages; 532 ksobj = td->td_kstack_obj; 533 VM_OBJECT_LOCK(ksobj); 534 for (i = 0; i < pages; i++) 535 ma[i] = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY | 536 VM_ALLOC_WIRED); 537 for (i = 0; i < pages; i++) { 538 if (ma[i]->valid != VM_PAGE_BITS_ALL) { 539 KASSERT(ma[i]->oflags & VPO_BUSY, 540 ("lost busy 1")); 541 vm_object_pip_add(ksobj, 1); 542 for (j = i + 1; j < pages; j++) { 543 KASSERT(ma[j]->valid == VM_PAGE_BITS_ALL || 544 (ma[j]->oflags & VPO_BUSY), 545 ("lost busy 2")); 546 if (ma[j]->valid == VM_PAGE_BITS_ALL) 547 break; 548 } 549 rv = vm_pager_get_pages(ksobj, ma + i, j - i, 0); 550 if (rv != VM_PAGER_OK) 551 panic("vm_thread_swapin: cannot get kstack for proc: %d", 552 td->td_proc->p_pid); 553 vm_object_pip_wakeup(ksobj); 554 for (k = i; k < j; k++) 555 ma[k] = vm_page_lookup(ksobj, k); 556 vm_page_wakeup(ma[i]); 557 } else if (ma[i]->oflags & VPO_BUSY) 558 vm_page_wakeup(ma[i]); 559 } 560 VM_OBJECT_UNLOCK(ksobj); 561 pmap_qenter(td->td_kstack, ma, pages); 562 cpu_thread_swapin(td); 563 } 564 565 /* 566 * Implement fork's actions on an address space. 567 * Here we arrange for the address space to be copied or referenced, 568 * allocate a user struct (pcb and kernel stack), then call the 569 * machine-dependent layer to fill those in and make the new process 570 * ready to run. The new process is set up so that it returns directly 571 * to user mode to avoid stack copying and relocation problems. 572 */ 573 int 574 vm_forkproc(td, p2, td2, vm2, flags) 575 struct thread *td; 576 struct proc *p2; 577 struct thread *td2; 578 struct vmspace *vm2; 579 int flags; 580 { 581 struct proc *p1 = td->td_proc; 582 int error; 583 584 if ((flags & RFPROC) == 0) { 585 /* 586 * Divorce the memory, if it is shared, essentially 587 * this changes shared memory amongst threads, into 588 * COW locally. 589 */ 590 if ((flags & RFMEM) == 0) { 591 if (p1->p_vmspace->vm_refcnt > 1) { 592 error = vmspace_unshare(p1); 593 if (error) 594 return (error); 595 } 596 } 597 cpu_fork(td, p2, td2, flags); 598 return (0); 599 } 600 601 if (flags & RFMEM) { 602 p2->p_vmspace = p1->p_vmspace; 603 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1); 604 } 605 606 while (vm_page_count_severe()) { 607 VM_WAIT; 608 } 609 610 if ((flags & RFMEM) == 0) { 611 p2->p_vmspace = vm2; 612 if (p1->p_vmspace->vm_shm) 613 shmfork(p1, p2); 614 } 615 616 /* 617 * cpu_fork will copy and update the pcb, set up the kernel stack, 618 * and make the child ready to run. 619 */ 620 cpu_fork(td, p2, td2, flags); 621 return (0); 622 } 623 624 /* 625 * Called after process has been wait(2)'ed apon and is being reaped. 626 * The idea is to reclaim resources that we could not reclaim while 627 * the process was still executing. 628 */ 629 void 630 vm_waitproc(p) 631 struct proc *p; 632 { 633 634 vmspace_exitfree(p); /* and clean-out the vmspace */ 635 } 636 637 /* 638 * Set default limits for VM system. 639 * Called for proc 0, and then inherited by all others. 640 * 641 * XXX should probably act directly on proc0. 642 */ 643 static void 644 vm_init_limits(udata) 645 void *udata; 646 { 647 struct proc *p = udata; 648 struct plimit *limp; 649 int rss_limit; 650 651 /* 652 * Set up the initial limits on process VM. Set the maximum resident 653 * set size to be half of (reasonably) available memory. Since this 654 * is a soft limit, it comes into effect only when the system is out 655 * of memory - half of main memory helps to favor smaller processes, 656 * and reduces thrashing of the object cache. 657 */ 658 limp = p->p_limit; 659 limp->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 660 limp->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 661 limp->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 662 limp->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 663 /* limit the limit to no less than 2MB */ 664 rss_limit = max(cnt.v_free_count, 512); 665 limp->pl_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 666 limp->pl_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 667 } 668 669 void 670 faultin(p) 671 struct proc *p; 672 { 673 #ifdef NO_SWAPPING 674 675 PROC_LOCK_ASSERT(p, MA_OWNED); 676 if ((p->p_flag & P_INMEM) == 0) 677 panic("faultin: proc swapped out with NO_SWAPPING!"); 678 #else /* !NO_SWAPPING */ 679 struct thread *td; 680 681 PROC_LOCK_ASSERT(p, MA_OWNED); 682 /* 683 * If another process is swapping in this process, 684 * just wait until it finishes. 685 */ 686 if (p->p_flag & P_SWAPPINGIN) { 687 while (p->p_flag & P_SWAPPINGIN) 688 msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0); 689 return; 690 } 691 if ((p->p_flag & P_INMEM) == 0) { 692 /* 693 * Don't let another thread swap process p out while we are 694 * busy swapping it in. 695 */ 696 ++p->p_lock; 697 p->p_flag |= P_SWAPPINGIN; 698 PROC_UNLOCK(p); 699 700 /* 701 * We hold no lock here because the list of threads 702 * can not change while all threads in the process are 703 * swapped out. 704 */ 705 FOREACH_THREAD_IN_PROC(p, td) 706 vm_thread_swapin(td); 707 PROC_LOCK(p); 708 swapclear(p); 709 p->p_swtick = ticks; 710 711 wakeup(&p->p_flag); 712 713 /* Allow other threads to swap p out now. */ 714 --p->p_lock; 715 } 716 #endif /* NO_SWAPPING */ 717 } 718 719 /* 720 * This swapin algorithm attempts to swap-in processes only if there 721 * is enough space for them. Of course, if a process waits for a long 722 * time, it will be swapped in anyway. 723 * 724 * Giant is held on entry. 725 */ 726 void 727 swapper(void) 728 { 729 struct proc *p; 730 struct thread *td; 731 struct proc *pp; 732 int slptime; 733 int swtime; 734 int ppri; 735 int pri; 736 737 loop: 738 if (vm_page_count_min()) { 739 VM_WAIT; 740 goto loop; 741 } 742 743 pp = NULL; 744 ppri = INT_MIN; 745 sx_slock(&allproc_lock); 746 FOREACH_PROC_IN_SYSTEM(p) { 747 PROC_LOCK(p); 748 if (p->p_state == PRS_NEW || 749 p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) { 750 PROC_UNLOCK(p); 751 continue; 752 } 753 swtime = (ticks - p->p_swtick) / hz; 754 FOREACH_THREAD_IN_PROC(p, td) { 755 /* 756 * An otherwise runnable thread of a process 757 * swapped out has only the TDI_SWAPPED bit set. 758 * 759 */ 760 thread_lock(td); 761 if (td->td_inhibitors == TDI_SWAPPED) { 762 slptime = (ticks - td->td_slptick) / hz; 763 pri = swtime + slptime; 764 if ((td->td_flags & TDF_SWAPINREQ) == 0) 765 pri -= p->p_nice * 8; 766 /* 767 * if this thread is higher priority 768 * and there is enough space, then select 769 * this process instead of the previous 770 * selection. 771 */ 772 if (pri > ppri) { 773 pp = p; 774 ppri = pri; 775 } 776 } 777 thread_unlock(td); 778 } 779 PROC_UNLOCK(p); 780 } 781 sx_sunlock(&allproc_lock); 782 783 /* 784 * Nothing to do, back to sleep. 785 */ 786 if ((p = pp) == NULL) { 787 tsleep(&proc0, PVM, "swapin", maxslp * hz / 2); 788 goto loop; 789 } 790 PROC_LOCK(p); 791 792 /* 793 * Another process may be bringing or may have already 794 * brought this process in while we traverse all threads. 795 * Or, this process may even be being swapped out again. 796 */ 797 if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) { 798 PROC_UNLOCK(p); 799 goto loop; 800 } 801 802 /* 803 * We would like to bring someone in. (only if there is space). 804 * [What checks the space? ] 805 */ 806 faultin(p); 807 PROC_UNLOCK(p); 808 goto loop; 809 } 810 811 void 812 kick_proc0(void) 813 { 814 815 wakeup(&proc0); 816 } 817 818 #ifndef NO_SWAPPING 819 820 /* 821 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 822 */ 823 static int swap_idle_threshold1 = 2; 824 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW, 825 &swap_idle_threshold1, 0, "Guaranteed swapped in time for a process"); 826 827 /* 828 * Swap_idle_threshold2 is the time that a process can be idle before 829 * it will be swapped out, if idle swapping is enabled. 830 */ 831 static int swap_idle_threshold2 = 10; 832 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW, 833 &swap_idle_threshold2, 0, "Time before a process will be swapped out"); 834 835 /* 836 * Swapout is driven by the pageout daemon. Very simple, we find eligible 837 * procs and swap out their stacks. We try to always "swap" at least one 838 * process in case we need the room for a swapin. 839 * If any procs have been sleeping/stopped for at least maxslp seconds, 840 * they are swapped. Else, we swap the longest-sleeping or stopped process, 841 * if any, otherwise the longest-resident process. 842 */ 843 void 844 swapout_procs(action) 845 int action; 846 { 847 struct proc *p; 848 struct thread *td; 849 int didswap = 0; 850 851 retry: 852 sx_slock(&allproc_lock); 853 FOREACH_PROC_IN_SYSTEM(p) { 854 struct vmspace *vm; 855 int minslptime = 100000; 856 int slptime; 857 858 /* 859 * Watch out for a process in 860 * creation. It may have no 861 * address space or lock yet. 862 */ 863 if (p->p_state == PRS_NEW) 864 continue; 865 /* 866 * An aio daemon switches its 867 * address space while running. 868 * Perform a quick check whether 869 * a process has P_SYSTEM. 870 */ 871 if ((p->p_flag & P_SYSTEM) != 0) 872 continue; 873 /* 874 * Do not swapout a process that 875 * is waiting for VM data 876 * structures as there is a possible 877 * deadlock. Test this first as 878 * this may block. 879 * 880 * Lock the map until swapout 881 * finishes, or a thread of this 882 * process may attempt to alter 883 * the map. 884 */ 885 vm = vmspace_acquire_ref(p); 886 if (vm == NULL) 887 continue; 888 if (!vm_map_trylock(&vm->vm_map)) 889 goto nextproc1; 890 891 PROC_LOCK(p); 892 if (p->p_lock != 0 || 893 (p->p_flag & (P_STOPPED_SINGLE|P_TRACED|P_SYSTEM|P_WEXIT) 894 ) != 0) { 895 goto nextproc; 896 } 897 /* 898 * only aiod changes vmspace, however it will be 899 * skipped because of the if statement above checking 900 * for P_SYSTEM 901 */ 902 if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM) 903 goto nextproc; 904 905 switch (p->p_state) { 906 default: 907 /* Don't swap out processes in any sort 908 * of 'special' state. */ 909 break; 910 911 case PRS_NORMAL: 912 /* 913 * do not swapout a realtime process 914 * Check all the thread groups.. 915 */ 916 FOREACH_THREAD_IN_PROC(p, td) { 917 thread_lock(td); 918 if (PRI_IS_REALTIME(td->td_pri_class)) { 919 thread_unlock(td); 920 goto nextproc; 921 } 922 slptime = (ticks - td->td_slptick) / hz; 923 /* 924 * Guarantee swap_idle_threshold1 925 * time in memory. 926 */ 927 if (slptime < swap_idle_threshold1) { 928 thread_unlock(td); 929 goto nextproc; 930 } 931 932 /* 933 * Do not swapout a process if it is 934 * waiting on a critical event of some 935 * kind or there is a thread whose 936 * pageable memory may be accessed. 937 * 938 * This could be refined to support 939 * swapping out a thread. 940 */ 941 if (!thread_safetoswapout(td)) { 942 thread_unlock(td); 943 goto nextproc; 944 } 945 /* 946 * If the system is under memory stress, 947 * or if we are swapping 948 * idle processes >= swap_idle_threshold2, 949 * then swap the process out. 950 */ 951 if (((action & VM_SWAP_NORMAL) == 0) && 952 (((action & VM_SWAP_IDLE) == 0) || 953 (slptime < swap_idle_threshold2))) { 954 thread_unlock(td); 955 goto nextproc; 956 } 957 958 if (minslptime > slptime) 959 minslptime = slptime; 960 thread_unlock(td); 961 } 962 963 /* 964 * If the pageout daemon didn't free enough pages, 965 * or if this process is idle and the system is 966 * configured to swap proactively, swap it out. 967 */ 968 if ((action & VM_SWAP_NORMAL) || 969 ((action & VM_SWAP_IDLE) && 970 (minslptime > swap_idle_threshold2))) { 971 if (swapout(p) == 0) 972 didswap++; 973 PROC_UNLOCK(p); 974 vm_map_unlock(&vm->vm_map); 975 vmspace_free(vm); 976 sx_sunlock(&allproc_lock); 977 goto retry; 978 } 979 } 980 nextproc: 981 PROC_UNLOCK(p); 982 vm_map_unlock(&vm->vm_map); 983 nextproc1: 984 vmspace_free(vm); 985 continue; 986 } 987 sx_sunlock(&allproc_lock); 988 /* 989 * If we swapped something out, and another process needed memory, 990 * then wakeup the sched process. 991 */ 992 if (didswap) 993 wakeup(&proc0); 994 } 995 996 static void 997 swapclear(p) 998 struct proc *p; 999 { 1000 struct thread *td; 1001 1002 PROC_LOCK_ASSERT(p, MA_OWNED); 1003 1004 FOREACH_THREAD_IN_PROC(p, td) { 1005 thread_lock(td); 1006 td->td_flags |= TDF_INMEM; 1007 td->td_flags &= ~TDF_SWAPINREQ; 1008 TD_CLR_SWAPPED(td); 1009 if (TD_CAN_RUN(td)) 1010 if (setrunnable(td)) { 1011 #ifdef INVARIANTS 1012 /* 1013 * XXX: We just cleared TDI_SWAPPED 1014 * above and set TDF_INMEM, so this 1015 * should never happen. 1016 */ 1017 panic("not waking up swapper"); 1018 #endif 1019 } 1020 thread_unlock(td); 1021 } 1022 p->p_flag &= ~(P_SWAPPINGIN|P_SWAPPINGOUT); 1023 p->p_flag |= P_INMEM; 1024 } 1025 1026 static int 1027 swapout(p) 1028 struct proc *p; 1029 { 1030 struct thread *td; 1031 1032 PROC_LOCK_ASSERT(p, MA_OWNED); 1033 #if defined(SWAP_DEBUG) 1034 printf("swapping out %d\n", p->p_pid); 1035 #endif 1036 1037 /* 1038 * The states of this process and its threads may have changed 1039 * by now. Assuming that there is only one pageout daemon thread, 1040 * this process should still be in memory. 1041 */ 1042 KASSERT((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) == P_INMEM, 1043 ("swapout: lost a swapout race?")); 1044 1045 /* 1046 * remember the process resident count 1047 */ 1048 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 1049 /* 1050 * Check and mark all threads before we proceed. 1051 */ 1052 p->p_flag &= ~P_INMEM; 1053 p->p_flag |= P_SWAPPINGOUT; 1054 FOREACH_THREAD_IN_PROC(p, td) { 1055 thread_lock(td); 1056 if (!thread_safetoswapout(td)) { 1057 thread_unlock(td); 1058 swapclear(p); 1059 return (EBUSY); 1060 } 1061 td->td_flags &= ~TDF_INMEM; 1062 TD_SET_SWAPPED(td); 1063 thread_unlock(td); 1064 } 1065 td = FIRST_THREAD_IN_PROC(p); 1066 ++td->td_ru.ru_nswap; 1067 PROC_UNLOCK(p); 1068 1069 /* 1070 * This list is stable because all threads are now prevented from 1071 * running. The list is only modified in the context of a running 1072 * thread in this process. 1073 */ 1074 FOREACH_THREAD_IN_PROC(p, td) 1075 vm_thread_swapout(td); 1076 1077 PROC_LOCK(p); 1078 p->p_flag &= ~P_SWAPPINGOUT; 1079 p->p_swtick = ticks; 1080 return (0); 1081 } 1082 #endif /* !NO_SWAPPING */
Cache object: f34acbc23f640b4a0d7eec6644b5cfe1
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]