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: releng/8.3/sys/vm/vm_glue.c 222868 2011-06-08 19:58:20Z jhb $"); 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 /* 108 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 109 * 110 * Note: run scheduling should be divorced from the vm system. 111 */ 112 static void scheduler(void *); 113 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_ANY, scheduler, NULL); 114 115 #ifndef NO_SWAPPING 116 static int swapout(struct proc *); 117 static void swapclear(struct proc *); 118 #endif 119 120 /* 121 * MPSAFE 122 * 123 * WARNING! This code calls vm_map_check_protection() which only checks 124 * the associated vm_map_entry range. It does not determine whether the 125 * contents of the memory is actually readable or writable. In most cases 126 * just checking the vm_map_entry is sufficient within the kernel's address 127 * space. 128 */ 129 int 130 kernacc(addr, len, rw) 131 void *addr; 132 int len, rw; 133 { 134 boolean_t rv; 135 vm_offset_t saddr, eaddr; 136 vm_prot_t prot; 137 138 KASSERT((rw & ~VM_PROT_ALL) == 0, 139 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 140 141 if ((vm_offset_t)addr + len > kernel_map->max_offset || 142 (vm_offset_t)addr + len < (vm_offset_t)addr) 143 return (FALSE); 144 145 prot = rw; 146 saddr = trunc_page((vm_offset_t)addr); 147 eaddr = round_page((vm_offset_t)addr + len); 148 vm_map_lock_read(kernel_map); 149 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 150 vm_map_unlock_read(kernel_map); 151 return (rv == TRUE); 152 } 153 154 /* 155 * MPSAFE 156 * 157 * WARNING! This code calls vm_map_check_protection() which only checks 158 * the associated vm_map_entry range. It does not determine whether the 159 * contents of the memory is actually readable or writable. vmapbuf(), 160 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be 161 * used in conjuction with this call. 162 */ 163 int 164 useracc(addr, len, rw) 165 void *addr; 166 int len, rw; 167 { 168 boolean_t rv; 169 vm_prot_t prot; 170 vm_map_t map; 171 172 KASSERT((rw & ~VM_PROT_ALL) == 0, 173 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 174 prot = rw; 175 map = &curproc->p_vmspace->vm_map; 176 if ((vm_offset_t)addr + len > vm_map_max(map) || 177 (vm_offset_t)addr + len < (vm_offset_t)addr) { 178 return (FALSE); 179 } 180 vm_map_lock_read(map); 181 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 182 round_page((vm_offset_t)addr + len), prot); 183 vm_map_unlock_read(map); 184 return (rv == TRUE); 185 } 186 187 int 188 vslock(void *addr, size_t len) 189 { 190 vm_offset_t end, last, start; 191 vm_size_t npages; 192 int error; 193 194 last = (vm_offset_t)addr + len; 195 start = trunc_page((vm_offset_t)addr); 196 end = round_page(last); 197 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr) 198 return (EINVAL); 199 npages = atop(end - start); 200 if (npages > vm_page_max_wired) 201 return (ENOMEM); 202 PROC_LOCK(curproc); 203 if (ptoa(npages + 204 pmap_wired_count(vm_map_pmap(&curproc->p_vmspace->vm_map))) > 205 lim_cur(curproc, RLIMIT_MEMLOCK)) { 206 PROC_UNLOCK(curproc); 207 return (ENOMEM); 208 } 209 PROC_UNLOCK(curproc); 210 #if 0 211 /* 212 * XXX - not yet 213 * 214 * The limit for transient usage of wired pages should be 215 * larger than for "permanent" wired pages (mlock()). 216 * 217 * Also, the sysctl code, which is the only present user 218 * of vslock(), does a hard loop on EAGAIN. 219 */ 220 if (npages + cnt.v_wire_count > vm_page_max_wired) 221 return (EAGAIN); 222 #endif 223 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end, 224 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 225 /* 226 * Return EFAULT on error to match copy{in,out}() behaviour 227 * rather than returning ENOMEM like mlock() would. 228 */ 229 return (error == KERN_SUCCESS ? 0 : EFAULT); 230 } 231 232 void 233 vsunlock(void *addr, size_t len) 234 { 235 236 /* Rely on the parameter sanity checks performed by vslock(). */ 237 (void)vm_map_unwire(&curproc->p_vmspace->vm_map, 238 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), 239 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 240 } 241 242 /* 243 * Pin the page contained within the given object at the given offset. If the 244 * page is not resident, allocate and load it using the given object's pager. 245 * Return the pinned page if successful; otherwise, return NULL. 246 */ 247 static vm_page_t 248 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset) 249 { 250 vm_page_t m, ma[1]; 251 vm_pindex_t pindex; 252 int rv; 253 254 VM_OBJECT_LOCK(object); 255 pindex = OFF_TO_IDX(offset); 256 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 257 if (m->valid != VM_PAGE_BITS_ALL) { 258 ma[0] = m; 259 rv = vm_pager_get_pages(object, ma, 1, 0); 260 m = vm_page_lookup(object, pindex); 261 if (m == NULL) 262 goto out; 263 if (rv != VM_PAGER_OK) { 264 vm_page_lock_queues(); 265 vm_page_free(m); 266 vm_page_unlock_queues(); 267 m = NULL; 268 goto out; 269 } 270 } 271 vm_page_lock_queues(); 272 vm_page_hold(m); 273 vm_page_unlock_queues(); 274 vm_page_wakeup(m); 275 out: 276 VM_OBJECT_UNLOCK(object); 277 return (m); 278 } 279 280 /* 281 * Return a CPU private mapping to the page at the given offset within the 282 * given object. The page is pinned before it is mapped. 283 */ 284 struct sf_buf * 285 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset) 286 { 287 vm_page_t m; 288 289 m = vm_imgact_hold_page(object, offset); 290 if (m == NULL) 291 return (NULL); 292 sched_pin(); 293 return (sf_buf_alloc(m, SFB_CPUPRIVATE)); 294 } 295 296 /* 297 * Destroy the given CPU private mapping and unpin the page that it mapped. 298 */ 299 void 300 vm_imgact_unmap_page(struct sf_buf *sf) 301 { 302 vm_page_t m; 303 304 m = sf_buf_page(sf); 305 sf_buf_free(sf); 306 sched_unpin(); 307 vm_page_lock_queues(); 308 vm_page_unhold(m); 309 vm_page_unlock_queues(); 310 } 311 312 void 313 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz) 314 { 315 316 pmap_sync_icache(map->pmap, va, sz); 317 } 318 319 struct kstack_cache_entry { 320 vm_object_t ksobj; 321 struct kstack_cache_entry *next_ks_entry; 322 }; 323 324 static struct kstack_cache_entry *kstack_cache; 325 static int kstack_cache_size = 128; 326 static int kstacks; 327 static struct mtx kstack_cache_mtx; 328 SYSCTL_INT(_vm, OID_AUTO, kstack_cache_size, CTLFLAG_RW, &kstack_cache_size, 0, 329 ""); 330 SYSCTL_INT(_vm, OID_AUTO, kstacks, CTLFLAG_RD, &kstacks, 0, 331 ""); 332 333 #ifndef KSTACK_MAX_PAGES 334 #define KSTACK_MAX_PAGES 32 335 #endif 336 337 /* 338 * Create the kernel stack (including pcb for i386) for a new thread. 339 * This routine directly affects the fork perf for a process and 340 * create performance for a thread. 341 */ 342 int 343 vm_thread_new(struct thread *td, int pages) 344 { 345 vm_object_t ksobj; 346 vm_offset_t ks; 347 vm_page_t m, ma[KSTACK_MAX_PAGES]; 348 struct kstack_cache_entry *ks_ce; 349 int i; 350 351 /* Bounds check */ 352 if (pages <= 1) 353 pages = KSTACK_PAGES; 354 else if (pages > KSTACK_MAX_PAGES) 355 pages = KSTACK_MAX_PAGES; 356 357 if (pages == KSTACK_PAGES) { 358 mtx_lock(&kstack_cache_mtx); 359 if (kstack_cache != NULL) { 360 ks_ce = kstack_cache; 361 kstack_cache = ks_ce->next_ks_entry; 362 mtx_unlock(&kstack_cache_mtx); 363 364 td->td_kstack_obj = ks_ce->ksobj; 365 td->td_kstack = (vm_offset_t)ks_ce; 366 td->td_kstack_pages = KSTACK_PAGES; 367 return (1); 368 } 369 mtx_unlock(&kstack_cache_mtx); 370 } 371 372 /* 373 * Allocate an object for the kstack. 374 */ 375 ksobj = vm_object_allocate(OBJT_DEFAULT, pages); 376 377 /* 378 * Get a kernel virtual address for this thread's kstack. 379 */ 380 #if defined(__mips__) 381 /* 382 * We need to align the kstack's mapped address to fit within 383 * a single TLB entry. 384 */ 385 ks = kmem_alloc_nofault_space(kernel_map, 386 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE, VMFS_TLB_ALIGNED_SPACE); 387 #else 388 ks = kmem_alloc_nofault(kernel_map, 389 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 390 #endif 391 if (ks == 0) { 392 printf("vm_thread_new: kstack allocation failed\n"); 393 vm_object_deallocate(ksobj); 394 return (0); 395 } 396 397 atomic_add_int(&kstacks, 1); 398 if (KSTACK_GUARD_PAGES != 0) { 399 pmap_qremove(ks, KSTACK_GUARD_PAGES); 400 ks += KSTACK_GUARD_PAGES * PAGE_SIZE; 401 } 402 td->td_kstack_obj = ksobj; 403 td->td_kstack = ks; 404 /* 405 * Knowing the number of pages allocated is useful when you 406 * want to deallocate them. 407 */ 408 td->td_kstack_pages = pages; 409 /* 410 * For the length of the stack, link in a real page of ram for each 411 * page of stack. 412 */ 413 VM_OBJECT_LOCK(ksobj); 414 for (i = 0; i < pages; i++) { 415 /* 416 * Get a kernel stack page. 417 */ 418 m = vm_page_grab(ksobj, i, VM_ALLOC_NOBUSY | 419 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_WIRED); 420 ma[i] = m; 421 m->valid = VM_PAGE_BITS_ALL; 422 } 423 VM_OBJECT_UNLOCK(ksobj); 424 pmap_qenter(ks, ma, pages); 425 return (1); 426 } 427 428 static void 429 vm_thread_stack_dispose(vm_object_t ksobj, vm_offset_t ks, int pages) 430 { 431 vm_page_t m; 432 int i; 433 434 atomic_add_int(&kstacks, -1); 435 pmap_qremove(ks, pages); 436 VM_OBJECT_LOCK(ksobj); 437 for (i = 0; i < pages; i++) { 438 m = vm_page_lookup(ksobj, i); 439 if (m == NULL) 440 panic("vm_thread_dispose: kstack already missing?"); 441 vm_page_lock_queues(); 442 vm_page_unwire(m, 0); 443 vm_page_free(m); 444 vm_page_unlock_queues(); 445 } 446 VM_OBJECT_UNLOCK(ksobj); 447 vm_object_deallocate(ksobj); 448 kmem_free(kernel_map, ks - (KSTACK_GUARD_PAGES * PAGE_SIZE), 449 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 450 } 451 452 /* 453 * Dispose of a thread's kernel stack. 454 */ 455 void 456 vm_thread_dispose(struct thread *td) 457 { 458 vm_object_t ksobj; 459 vm_offset_t ks; 460 struct kstack_cache_entry *ks_ce; 461 int pages; 462 463 pages = td->td_kstack_pages; 464 ksobj = td->td_kstack_obj; 465 ks = td->td_kstack; 466 td->td_kstack = 0; 467 td->td_kstack_pages = 0; 468 if (pages == KSTACK_PAGES && kstacks <= kstack_cache_size) { 469 ks_ce = (struct kstack_cache_entry *)ks; 470 ks_ce->ksobj = ksobj; 471 mtx_lock(&kstack_cache_mtx); 472 ks_ce->next_ks_entry = kstack_cache; 473 kstack_cache = ks_ce; 474 mtx_unlock(&kstack_cache_mtx); 475 return; 476 } 477 vm_thread_stack_dispose(ksobj, ks, pages); 478 } 479 480 static void 481 vm_thread_stack_lowmem(void *nulll) 482 { 483 struct kstack_cache_entry *ks_ce, *ks_ce1; 484 485 mtx_lock(&kstack_cache_mtx); 486 ks_ce = kstack_cache; 487 kstack_cache = NULL; 488 mtx_unlock(&kstack_cache_mtx); 489 490 while (ks_ce != NULL) { 491 ks_ce1 = ks_ce; 492 ks_ce = ks_ce->next_ks_entry; 493 494 vm_thread_stack_dispose(ks_ce1->ksobj, (vm_offset_t)ks_ce1, 495 KSTACK_PAGES); 496 } 497 } 498 499 static void 500 kstack_cache_init(void *nulll) 501 { 502 503 EVENTHANDLER_REGISTER(vm_lowmem, vm_thread_stack_lowmem, NULL, 504 EVENTHANDLER_PRI_ANY); 505 } 506 507 MTX_SYSINIT(kstack_cache, &kstack_cache_mtx, "kstkch", MTX_DEF); 508 SYSINIT(vm_kstacks, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY, kstack_cache_init, NULL); 509 510 /* 511 * Allow a thread's kernel stack to be paged out. 512 */ 513 void 514 vm_thread_swapout(struct thread *td) 515 { 516 vm_object_t ksobj; 517 vm_page_t m; 518 int i, pages; 519 520 cpu_thread_swapout(td); 521 pages = td->td_kstack_pages; 522 ksobj = td->td_kstack_obj; 523 pmap_qremove(td->td_kstack, pages); 524 VM_OBJECT_LOCK(ksobj); 525 for (i = 0; i < pages; i++) { 526 m = vm_page_lookup(ksobj, i); 527 if (m == NULL) 528 panic("vm_thread_swapout: kstack already missing?"); 529 vm_page_lock_queues(); 530 vm_page_dirty(m); 531 vm_page_unwire(m, 0); 532 vm_page_unlock_queues(); 533 } 534 VM_OBJECT_UNLOCK(ksobj); 535 } 536 537 /* 538 * Bring the kernel stack for a specified thread back in. 539 */ 540 void 541 vm_thread_swapin(struct thread *td) 542 { 543 vm_object_t ksobj; 544 vm_page_t ma[KSTACK_MAX_PAGES]; 545 int i, j, k, pages, rv; 546 547 pages = td->td_kstack_pages; 548 ksobj = td->td_kstack_obj; 549 VM_OBJECT_LOCK(ksobj); 550 for (i = 0; i < pages; i++) 551 ma[i] = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY | 552 VM_ALLOC_WIRED); 553 for (i = 0; i < pages; i++) { 554 if (ma[i]->valid != VM_PAGE_BITS_ALL) { 555 KASSERT(ma[i]->oflags & VPO_BUSY, 556 ("lost busy 1")); 557 vm_object_pip_add(ksobj, 1); 558 for (j = i + 1; j < pages; j++) { 559 KASSERT(ma[j]->valid == VM_PAGE_BITS_ALL || 560 (ma[j]->oflags & VPO_BUSY), 561 ("lost busy 2")); 562 if (ma[j]->valid == VM_PAGE_BITS_ALL) 563 break; 564 } 565 rv = vm_pager_get_pages(ksobj, ma + i, j - i, 0); 566 if (rv != VM_PAGER_OK) 567 panic("vm_thread_swapin: cannot get kstack for proc: %d", 568 td->td_proc->p_pid); 569 vm_object_pip_wakeup(ksobj); 570 for (k = i; k < j; k++) 571 ma[k] = vm_page_lookup(ksobj, k); 572 vm_page_wakeup(ma[i]); 573 } else if (ma[i]->oflags & VPO_BUSY) 574 vm_page_wakeup(ma[i]); 575 } 576 VM_OBJECT_UNLOCK(ksobj); 577 pmap_qenter(td->td_kstack, ma, pages); 578 cpu_thread_swapin(td); 579 } 580 581 /* 582 * Implement fork's actions on an address space. 583 * Here we arrange for the address space to be copied or referenced, 584 * allocate a user struct (pcb and kernel stack), then call the 585 * machine-dependent layer to fill those in and make the new process 586 * ready to run. The new process is set up so that it returns directly 587 * to user mode to avoid stack copying and relocation problems. 588 */ 589 int 590 vm_forkproc(td, p2, td2, vm2, flags) 591 struct thread *td; 592 struct proc *p2; 593 struct thread *td2; 594 struct vmspace *vm2; 595 int flags; 596 { 597 struct proc *p1 = td->td_proc; 598 int error; 599 600 if ((flags & RFPROC) == 0) { 601 /* 602 * Divorce the memory, if it is shared, essentially 603 * this changes shared memory amongst threads, into 604 * COW locally. 605 */ 606 if ((flags & RFMEM) == 0) { 607 if (p1->p_vmspace->vm_refcnt > 1) { 608 error = vmspace_unshare(p1); 609 if (error) 610 return (error); 611 } 612 } 613 cpu_fork(td, p2, td2, flags); 614 return (0); 615 } 616 617 if (flags & RFMEM) { 618 p2->p_vmspace = p1->p_vmspace; 619 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1); 620 } 621 622 while (vm_page_count_severe()) { 623 VM_WAIT; 624 } 625 626 if ((flags & RFMEM) == 0) { 627 p2->p_vmspace = vm2; 628 if (p1->p_vmspace->vm_shm) 629 shmfork(p1, p2); 630 } 631 632 /* 633 * cpu_fork will copy and update the pcb, set up the kernel stack, 634 * and make the child ready to run. 635 */ 636 cpu_fork(td, p2, td2, flags); 637 return (0); 638 } 639 640 /* 641 * Called after process has been wait(2)'ed apon and is being reaped. 642 * The idea is to reclaim resources that we could not reclaim while 643 * the process was still executing. 644 */ 645 void 646 vm_waitproc(p) 647 struct proc *p; 648 { 649 650 vmspace_exitfree(p); /* and clean-out the vmspace */ 651 } 652 653 /* 654 * Set default limits for VM system. 655 * Called for proc 0, and then inherited by all others. 656 * 657 * XXX should probably act directly on proc0. 658 */ 659 static void 660 vm_init_limits(udata) 661 void *udata; 662 { 663 struct proc *p = udata; 664 struct plimit *limp; 665 int rss_limit; 666 667 /* 668 * Set up the initial limits on process VM. Set the maximum resident 669 * set size to be half of (reasonably) available memory. Since this 670 * is a soft limit, it comes into effect only when the system is out 671 * of memory - half of main memory helps to favor smaller processes, 672 * and reduces thrashing of the object cache. 673 */ 674 limp = p->p_limit; 675 limp->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 676 limp->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 677 limp->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 678 limp->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 679 /* limit the limit to no less than 2MB */ 680 rss_limit = max(cnt.v_free_count, 512); 681 limp->pl_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 682 limp->pl_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 683 } 684 685 void 686 faultin(p) 687 struct proc *p; 688 { 689 #ifdef NO_SWAPPING 690 691 PROC_LOCK_ASSERT(p, MA_OWNED); 692 if ((p->p_flag & P_INMEM) == 0) 693 panic("faultin: proc swapped out with NO_SWAPPING!"); 694 #else /* !NO_SWAPPING */ 695 struct thread *td; 696 697 PROC_LOCK_ASSERT(p, MA_OWNED); 698 /* 699 * If another process is swapping in this process, 700 * just wait until it finishes. 701 */ 702 if (p->p_flag & P_SWAPPINGIN) { 703 while (p->p_flag & P_SWAPPINGIN) 704 msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0); 705 return; 706 } 707 if ((p->p_flag & P_INMEM) == 0) { 708 /* 709 * Don't let another thread swap process p out while we are 710 * busy swapping it in. 711 */ 712 ++p->p_lock; 713 p->p_flag |= P_SWAPPINGIN; 714 PROC_UNLOCK(p); 715 716 /* 717 * We hold no lock here because the list of threads 718 * can not change while all threads in the process are 719 * swapped out. 720 */ 721 FOREACH_THREAD_IN_PROC(p, td) 722 vm_thread_swapin(td); 723 PROC_LOCK(p); 724 swapclear(p); 725 p->p_swtick = ticks; 726 727 wakeup(&p->p_flag); 728 729 /* Allow other threads to swap p out now. */ 730 --p->p_lock; 731 } 732 #endif /* NO_SWAPPING */ 733 } 734 735 /* 736 * This swapin algorithm attempts to swap-in processes only if there 737 * is enough space for them. Of course, if a process waits for a long 738 * time, it will be swapped in anyway. 739 * 740 * Giant is held on entry. 741 */ 742 /* ARGSUSED*/ 743 static void 744 scheduler(dummy) 745 void *dummy; 746 { 747 struct proc *p; 748 struct thread *td; 749 struct proc *pp; 750 int slptime; 751 int swtime; 752 int ppri; 753 int pri; 754 755 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED); 756 mtx_unlock(&Giant); 757 758 loop: 759 if (vm_page_count_min()) { 760 VM_WAIT; 761 goto loop; 762 } 763 764 pp = NULL; 765 ppri = INT_MIN; 766 sx_slock(&allproc_lock); 767 FOREACH_PROC_IN_SYSTEM(p) { 768 PROC_LOCK(p); 769 if (p->p_state == PRS_NEW || 770 p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) { 771 PROC_UNLOCK(p); 772 continue; 773 } 774 swtime = (ticks - p->p_swtick) / hz; 775 FOREACH_THREAD_IN_PROC(p, td) { 776 /* 777 * An otherwise runnable thread of a process 778 * swapped out has only the TDI_SWAPPED bit set. 779 * 780 */ 781 thread_lock(td); 782 if (td->td_inhibitors == TDI_SWAPPED) { 783 slptime = (ticks - td->td_slptick) / hz; 784 pri = swtime + slptime; 785 if ((td->td_flags & TDF_SWAPINREQ) == 0) 786 pri -= p->p_nice * 8; 787 /* 788 * if this thread is higher priority 789 * and there is enough space, then select 790 * this process instead of the previous 791 * selection. 792 */ 793 if (pri > ppri) { 794 pp = p; 795 ppri = pri; 796 } 797 } 798 thread_unlock(td); 799 } 800 PROC_UNLOCK(p); 801 } 802 sx_sunlock(&allproc_lock); 803 804 /* 805 * Nothing to do, back to sleep. 806 */ 807 if ((p = pp) == NULL) { 808 tsleep(&proc0, PVM, "sched", maxslp * hz / 2); 809 goto loop; 810 } 811 PROC_LOCK(p); 812 813 /* 814 * Another process may be bringing or may have already 815 * brought this process in while we traverse all threads. 816 * Or, this process may even be being swapped out again. 817 */ 818 if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) { 819 PROC_UNLOCK(p); 820 goto loop; 821 } 822 823 /* 824 * We would like to bring someone in. (only if there is space). 825 * [What checks the space? ] 826 */ 827 faultin(p); 828 PROC_UNLOCK(p); 829 goto loop; 830 } 831 832 void 833 kick_proc0(void) 834 { 835 836 wakeup(&proc0); 837 } 838 839 #ifndef NO_SWAPPING 840 841 /* 842 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 843 */ 844 static int swap_idle_threshold1 = 2; 845 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW, 846 &swap_idle_threshold1, 0, "Guaranteed swapped in time for a process"); 847 848 /* 849 * Swap_idle_threshold2 is the time that a process can be idle before 850 * it will be swapped out, if idle swapping is enabled. 851 */ 852 static int swap_idle_threshold2 = 10; 853 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW, 854 &swap_idle_threshold2, 0, "Time before a process will be swapped out"); 855 856 /* 857 * Swapout is driven by the pageout daemon. Very simple, we find eligible 858 * procs and swap out their stacks. We try to always "swap" at least one 859 * process in case we need the room for a swapin. 860 * If any procs have been sleeping/stopped for at least maxslp seconds, 861 * they are swapped. Else, we swap the longest-sleeping or stopped process, 862 * if any, otherwise the longest-resident process. 863 */ 864 void 865 swapout_procs(action) 866 int action; 867 { 868 struct proc *p; 869 struct thread *td; 870 int didswap = 0; 871 872 retry: 873 sx_slock(&allproc_lock); 874 FOREACH_PROC_IN_SYSTEM(p) { 875 struct vmspace *vm; 876 int minslptime = 100000; 877 int slptime; 878 879 /* 880 * Watch out for a process in 881 * creation. It may have no 882 * address space or lock yet. 883 */ 884 if (p->p_state == PRS_NEW) 885 continue; 886 /* 887 * An aio daemon switches its 888 * address space while running. 889 * Perform a quick check whether 890 * a process has P_SYSTEM. 891 */ 892 if ((p->p_flag & P_SYSTEM) != 0) 893 continue; 894 /* 895 * Do not swapout a process that 896 * is waiting for VM data 897 * structures as there is a possible 898 * deadlock. Test this first as 899 * this may block. 900 * 901 * Lock the map until swapout 902 * finishes, or a thread of this 903 * process may attempt to alter 904 * the map. 905 */ 906 vm = vmspace_acquire_ref(p); 907 if (vm == NULL) 908 continue; 909 if (!vm_map_trylock(&vm->vm_map)) 910 goto nextproc1; 911 912 PROC_LOCK(p); 913 if (p->p_lock != 0 || 914 (p->p_flag & (P_STOPPED_SINGLE|P_TRACED|P_SYSTEM|P_WEXIT) 915 ) != 0) { 916 goto nextproc; 917 } 918 /* 919 * only aiod changes vmspace, however it will be 920 * skipped because of the if statement above checking 921 * for P_SYSTEM 922 */ 923 if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM) 924 goto nextproc; 925 926 switch (p->p_state) { 927 default: 928 /* Don't swap out processes in any sort 929 * of 'special' state. */ 930 break; 931 932 case PRS_NORMAL: 933 /* 934 * do not swapout a realtime process 935 * Check all the thread groups.. 936 */ 937 FOREACH_THREAD_IN_PROC(p, td) { 938 thread_lock(td); 939 if (PRI_IS_REALTIME(td->td_pri_class)) { 940 thread_unlock(td); 941 goto nextproc; 942 } 943 slptime = (ticks - td->td_slptick) / hz; 944 /* 945 * Guarantee swap_idle_threshold1 946 * time in memory. 947 */ 948 if (slptime < swap_idle_threshold1) { 949 thread_unlock(td); 950 goto nextproc; 951 } 952 953 /* 954 * Do not swapout a process if it is 955 * waiting on a critical event of some 956 * kind or there is a thread whose 957 * pageable memory may be accessed. 958 * 959 * This could be refined to support 960 * swapping out a thread. 961 */ 962 if (!thread_safetoswapout(td)) { 963 thread_unlock(td); 964 goto nextproc; 965 } 966 /* 967 * If the system is under memory stress, 968 * or if we are swapping 969 * idle processes >= swap_idle_threshold2, 970 * then swap the process out. 971 */ 972 if (((action & VM_SWAP_NORMAL) == 0) && 973 (((action & VM_SWAP_IDLE) == 0) || 974 (slptime < swap_idle_threshold2))) { 975 thread_unlock(td); 976 goto nextproc; 977 } 978 979 if (minslptime > slptime) 980 minslptime = slptime; 981 thread_unlock(td); 982 } 983 984 /* 985 * If the pageout daemon didn't free enough pages, 986 * or if this process is idle and the system is 987 * configured to swap proactively, swap it out. 988 */ 989 if ((action & VM_SWAP_NORMAL) || 990 ((action & VM_SWAP_IDLE) && 991 (minslptime > swap_idle_threshold2))) { 992 if (swapout(p) == 0) 993 didswap++; 994 PROC_UNLOCK(p); 995 vm_map_unlock(&vm->vm_map); 996 vmspace_free(vm); 997 sx_sunlock(&allproc_lock); 998 goto retry; 999 } 1000 } 1001 nextproc: 1002 PROC_UNLOCK(p); 1003 vm_map_unlock(&vm->vm_map); 1004 nextproc1: 1005 vmspace_free(vm); 1006 continue; 1007 } 1008 sx_sunlock(&allproc_lock); 1009 /* 1010 * If we swapped something out, and another process needed memory, 1011 * then wakeup the sched process. 1012 */ 1013 if (didswap) 1014 wakeup(&proc0); 1015 } 1016 1017 static void 1018 swapclear(p) 1019 struct proc *p; 1020 { 1021 struct thread *td; 1022 1023 PROC_LOCK_ASSERT(p, MA_OWNED); 1024 1025 FOREACH_THREAD_IN_PROC(p, td) { 1026 thread_lock(td); 1027 td->td_flags |= TDF_INMEM; 1028 td->td_flags &= ~TDF_SWAPINREQ; 1029 TD_CLR_SWAPPED(td); 1030 if (TD_CAN_RUN(td)) 1031 if (setrunnable(td)) { 1032 #ifdef INVARIANTS 1033 /* 1034 * XXX: We just cleared TDI_SWAPPED 1035 * above and set TDF_INMEM, so this 1036 * should never happen. 1037 */ 1038 panic("not waking up swapper"); 1039 #endif 1040 } 1041 thread_unlock(td); 1042 } 1043 p->p_flag &= ~(P_SWAPPINGIN|P_SWAPPINGOUT); 1044 p->p_flag |= P_INMEM; 1045 } 1046 1047 static int 1048 swapout(p) 1049 struct proc *p; 1050 { 1051 struct thread *td; 1052 1053 PROC_LOCK_ASSERT(p, MA_OWNED); 1054 #if defined(SWAP_DEBUG) 1055 printf("swapping out %d\n", p->p_pid); 1056 #endif 1057 1058 /* 1059 * The states of this process and its threads may have changed 1060 * by now. Assuming that there is only one pageout daemon thread, 1061 * this process should still be in memory. 1062 */ 1063 KASSERT((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) == P_INMEM, 1064 ("swapout: lost a swapout race?")); 1065 1066 /* 1067 * remember the process resident count 1068 */ 1069 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 1070 /* 1071 * Check and mark all threads before we proceed. 1072 */ 1073 p->p_flag &= ~P_INMEM; 1074 p->p_flag |= P_SWAPPINGOUT; 1075 FOREACH_THREAD_IN_PROC(p, td) { 1076 thread_lock(td); 1077 if (!thread_safetoswapout(td)) { 1078 thread_unlock(td); 1079 swapclear(p); 1080 return (EBUSY); 1081 } 1082 td->td_flags &= ~TDF_INMEM; 1083 TD_SET_SWAPPED(td); 1084 thread_unlock(td); 1085 } 1086 td = FIRST_THREAD_IN_PROC(p); 1087 ++td->td_ru.ru_nswap; 1088 PROC_UNLOCK(p); 1089 1090 /* 1091 * This list is stable because all threads are now prevented from 1092 * running. The list is only modified in the context of a running 1093 * thread in this process. 1094 */ 1095 FOREACH_THREAD_IN_PROC(p, td) 1096 vm_thread_swapout(td); 1097 1098 PROC_LOCK(p); 1099 p->p_flag &= ~P_SWAPPINGOUT; 1100 p->p_swtick = ticks; 1101 return (0); 1102 } 1103 #endif /* !NO_SWAPPING */
Cache object: 556249f74f05520afc4a29c178e94afe
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]