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 63 #include <sys/cdefs.h> 64 __FBSDID("$FreeBSD$"); 65 66 #include "opt_vm.h" 67 #include "opt_kstack_pages.h" 68 #include "opt_kstack_max_pages.h" 69 70 #include <sys/param.h> 71 #include <sys/systm.h> 72 #include <sys/limits.h> 73 #include <sys/lock.h> 74 #include <sys/mutex.h> 75 #include <sys/proc.h> 76 #include <sys/resourcevar.h> 77 #include <sys/shm.h> 78 #include <sys/vmmeter.h> 79 #include <sys/sx.h> 80 #include <sys/sysctl.h> 81 82 #include <sys/kernel.h> 83 #include <sys/ktr.h> 84 #include <sys/unistd.h> 85 86 #include <vm/vm.h> 87 #include <vm/vm_param.h> 88 #include <vm/pmap.h> 89 #include <vm/vm_map.h> 90 #include <vm/vm_page.h> 91 #include <vm/vm_pageout.h> 92 #include <vm/vm_object.h> 93 #include <vm/vm_kern.h> 94 #include <vm/vm_extern.h> 95 #include <vm/vm_pager.h> 96 #include <vm/swap_pager.h> 97 98 #include <sys/user.h> 99 100 extern int maxslp; 101 102 /* 103 * System initialization 104 * 105 * Note: proc0 from proc.h 106 */ 107 static void vm_init_limits(void *); 108 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 109 110 /* 111 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 112 * 113 * Note: run scheduling should be divorced from the vm system. 114 */ 115 static void scheduler(void *); 116 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_ANY, scheduler, NULL) 117 118 #ifndef NO_SWAPPING 119 static void swapout(struct proc *); 120 static void vm_proc_swapin(struct proc *p); 121 static void vm_proc_swapout(struct proc *p); 122 #endif 123 124 /* 125 * MPSAFE 126 * 127 * WARNING! This code calls vm_map_check_protection() which only checks 128 * the associated vm_map_entry range. It does not determine whether the 129 * contents of the memory is actually readable or writable. In most cases 130 * just checking the vm_map_entry is sufficient within the kernel's address 131 * space. 132 */ 133 int 134 kernacc(addr, len, rw) 135 void *addr; 136 int len, rw; 137 { 138 boolean_t rv; 139 vm_offset_t saddr, eaddr; 140 vm_prot_t prot; 141 142 KASSERT((rw & ~VM_PROT_ALL) == 0, 143 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 144 prot = rw; 145 saddr = trunc_page((vm_offset_t)addr); 146 eaddr = round_page((vm_offset_t)addr + len); 147 vm_map_lock_read(kernel_map); 148 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 149 vm_map_unlock_read(kernel_map); 150 return (rv == TRUE); 151 } 152 153 /* 154 * MPSAFE 155 * 156 * WARNING! This code calls vm_map_check_protection() which only checks 157 * the associated vm_map_entry range. It does not determine whether the 158 * contents of the memory is actually readable or writable. vmapbuf(), 159 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be 160 * used in conjuction with this call. 161 */ 162 int 163 useracc(addr, len, rw) 164 void *addr; 165 int len, rw; 166 { 167 boolean_t rv; 168 vm_prot_t prot; 169 vm_map_t map; 170 171 KASSERT((rw & ~VM_PROT_ALL) == 0, 172 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 173 prot = rw; 174 map = &curproc->p_vmspace->vm_map; 175 if ((vm_offset_t)addr + len > vm_map_max(map) || 176 (vm_offset_t)addr + len < (vm_offset_t)addr) { 177 return (FALSE); 178 } 179 vm_map_lock_read(map); 180 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 181 round_page((vm_offset_t)addr + len), prot); 182 vm_map_unlock_read(map); 183 return (rv == TRUE); 184 } 185 186 int 187 vslock(void *addr, size_t len) 188 { 189 vm_offset_t end, last, start; 190 vm_size_t npages; 191 int error; 192 193 last = (vm_offset_t)addr + len; 194 start = trunc_page((vm_offset_t)addr); 195 end = round_page(last); 196 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr) 197 return (EINVAL); 198 npages = atop(end - start); 199 if (npages > vm_page_max_wired) 200 return (ENOMEM); 201 PROC_LOCK(curproc); 202 if (ptoa(npages + 203 pmap_wired_count(vm_map_pmap(&curproc->p_vmspace->vm_map))) > 204 lim_cur(curproc, RLIMIT_MEMLOCK)) { 205 PROC_UNLOCK(curproc); 206 return (ENOMEM); 207 } 208 PROC_UNLOCK(curproc); 209 #if 0 210 /* 211 * XXX - not yet 212 * 213 * The limit for transient usage of wired pages should be 214 * larger than for "permanent" wired pages (mlock()). 215 * 216 * Also, the sysctl code, which is the only present user 217 * of vslock(), does a hard loop on EAGAIN. 218 */ 219 if (npages + cnt.v_wire_count > vm_page_max_wired) 220 return (EAGAIN); 221 #endif 222 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end, 223 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 224 /* 225 * Return EFAULT on error to match copy{in,out}() behaviour 226 * rather than returning ENOMEM like mlock() would. 227 */ 228 return (error == KERN_SUCCESS ? 0 : EFAULT); 229 } 230 231 void 232 vsunlock(void *addr, size_t len) 233 { 234 235 /* Rely on the parameter sanity checks performed by vslock(). */ 236 (void)vm_map_unwire(&curproc->p_vmspace->vm_map, 237 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), 238 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 239 } 240 241 /* 242 * Create the U area for a new process. 243 * This routine directly affects the fork perf for a process. 244 */ 245 void 246 vm_proc_new(struct proc *p) 247 { 248 vm_page_t ma[UAREA_PAGES]; 249 vm_object_t upobj; 250 vm_offset_t up; 251 vm_page_t m; 252 u_int i; 253 254 /* 255 * Get a kernel virtual address for the U area for this process. 256 */ 257 up = kmem_alloc_nofault(kernel_map, UAREA_PAGES * PAGE_SIZE); 258 if (up == 0) 259 panic("vm_proc_new: upage allocation failed"); 260 p->p_uarea = (struct user *)up; 261 262 /* 263 * Allocate object and page(s) for the U area. 264 */ 265 upobj = vm_object_allocate(OBJT_DEFAULT, UAREA_PAGES); 266 p->p_upages_obj = upobj; 267 VM_OBJECT_LOCK(upobj); 268 for (i = 0; i < UAREA_PAGES; i++) { 269 m = vm_page_grab(upobj, i, 270 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_WIRED); 271 ma[i] = m; 272 273 vm_page_lock_queues(); 274 vm_page_wakeup(m); 275 m->valid = VM_PAGE_BITS_ALL; 276 vm_page_unlock_queues(); 277 } 278 VM_OBJECT_UNLOCK(upobj); 279 280 /* 281 * Enter the pages into the kernel address space. 282 */ 283 pmap_qenter(up, ma, UAREA_PAGES); 284 } 285 286 /* 287 * Dispose the U area for a process that has exited. 288 * This routine directly impacts the exit perf of a process. 289 * XXX proc_zone is marked UMA_ZONE_NOFREE, so this should never be called. 290 */ 291 void 292 vm_proc_dispose(struct proc *p) 293 { 294 vm_object_t upobj; 295 vm_offset_t up; 296 vm_page_t m; 297 298 upobj = p->p_upages_obj; 299 VM_OBJECT_LOCK(upobj); 300 if (upobj->resident_page_count != UAREA_PAGES) 301 panic("vm_proc_dispose: incorrect number of pages in upobj"); 302 vm_page_lock_queues(); 303 while ((m = TAILQ_FIRST(&upobj->memq)) != NULL) { 304 vm_page_busy(m); 305 vm_page_unwire(m, 0); 306 vm_page_free(m); 307 } 308 vm_page_unlock_queues(); 309 VM_OBJECT_UNLOCK(upobj); 310 up = (vm_offset_t)p->p_uarea; 311 pmap_qremove(up, UAREA_PAGES); 312 kmem_free(kernel_map, up, UAREA_PAGES * PAGE_SIZE); 313 vm_object_deallocate(upobj); 314 } 315 316 #ifndef NO_SWAPPING 317 /* 318 * Allow the U area for a process to be prejudicially paged out. 319 */ 320 static void 321 vm_proc_swapout(struct proc *p) 322 { 323 vm_object_t upobj; 324 vm_offset_t up; 325 vm_page_t m; 326 327 upobj = p->p_upages_obj; 328 VM_OBJECT_LOCK(upobj); 329 if (upobj->resident_page_count != UAREA_PAGES) 330 panic("vm_proc_dispose: incorrect number of pages in upobj"); 331 vm_page_lock_queues(); 332 TAILQ_FOREACH(m, &upobj->memq, listq) { 333 vm_page_dirty(m); 334 vm_page_unwire(m, 0); 335 } 336 vm_page_unlock_queues(); 337 VM_OBJECT_UNLOCK(upobj); 338 up = (vm_offset_t)p->p_uarea; 339 pmap_qremove(up, UAREA_PAGES); 340 } 341 342 /* 343 * Bring the U area for a specified process back in. 344 */ 345 static void 346 vm_proc_swapin(struct proc *p) 347 { 348 vm_page_t ma[UAREA_PAGES]; 349 vm_object_t upobj; 350 vm_offset_t up; 351 vm_page_t m; 352 int rv; 353 int i; 354 355 upobj = p->p_upages_obj; 356 VM_OBJECT_LOCK(upobj); 357 for (i = 0; i < UAREA_PAGES; i++) { 358 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 359 if (m->valid != VM_PAGE_BITS_ALL) { 360 rv = vm_pager_get_pages(upobj, &m, 1, 0); 361 if (rv != VM_PAGER_OK) 362 panic("vm_proc_swapin: cannot get upage"); 363 } 364 ma[i] = m; 365 } 366 if (upobj->resident_page_count != UAREA_PAGES) 367 panic("vm_proc_swapin: lost pages from upobj"); 368 vm_page_lock_queues(); 369 TAILQ_FOREACH(m, &upobj->memq, listq) { 370 m->valid = VM_PAGE_BITS_ALL; 371 vm_page_wire(m); 372 vm_page_wakeup(m); 373 } 374 vm_page_unlock_queues(); 375 VM_OBJECT_UNLOCK(upobj); 376 up = (vm_offset_t)p->p_uarea; 377 pmap_qenter(up, ma, UAREA_PAGES); 378 } 379 380 /* 381 * Swap in the UAREAs of all processes swapped out to the given device. 382 * The pages in the UAREA are marked dirty and their swap metadata is freed. 383 */ 384 void 385 vm_proc_swapin_all(struct swdevt *devidx) 386 { 387 struct proc *p; 388 vm_object_t object; 389 vm_page_t m; 390 391 retry: 392 sx_slock(&allproc_lock); 393 FOREACH_PROC_IN_SYSTEM(p) { 394 PROC_LOCK(p); 395 object = p->p_upages_obj; 396 if (object != NULL) { 397 VM_OBJECT_LOCK(object); 398 if (swap_pager_isswapped(object, devidx)) { 399 VM_OBJECT_UNLOCK(object); 400 sx_sunlock(&allproc_lock); 401 faultin(p); 402 PROC_UNLOCK(p); 403 VM_OBJECT_LOCK(object); 404 vm_page_lock_queues(); 405 TAILQ_FOREACH(m, &object->memq, listq) 406 vm_page_dirty(m); 407 vm_page_unlock_queues(); 408 swap_pager_freespace(object, 0, 409 object->un_pager.swp.swp_bcount); 410 VM_OBJECT_UNLOCK(object); 411 goto retry; 412 } 413 VM_OBJECT_UNLOCK(object); 414 } 415 PROC_UNLOCK(p); 416 } 417 sx_sunlock(&allproc_lock); 418 } 419 #endif 420 421 #ifndef KSTACK_MAX_PAGES 422 #define KSTACK_MAX_PAGES 32 423 #endif 424 425 /* 426 * Create the kernel stack (including pcb for i386) for a new thread. 427 * This routine directly affects the fork perf for a process and 428 * create performance for a thread. 429 */ 430 void 431 vm_thread_new(struct thread *td, int pages) 432 { 433 vm_object_t ksobj; 434 vm_offset_t ks; 435 vm_page_t m, ma[KSTACK_MAX_PAGES]; 436 int i; 437 438 /* Bounds check */ 439 if (pages <= 1) 440 pages = KSTACK_PAGES; 441 else if (pages > KSTACK_MAX_PAGES) 442 pages = KSTACK_MAX_PAGES; 443 /* 444 * Allocate an object for the kstack. 445 */ 446 ksobj = vm_object_allocate(OBJT_DEFAULT, pages); 447 td->td_kstack_obj = ksobj; 448 /* 449 * Get a kernel virtual address for this thread's kstack. 450 */ 451 ks = kmem_alloc_nofault(kernel_map, 452 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 453 if (ks == 0) 454 panic("vm_thread_new: kstack allocation failed"); 455 if (KSTACK_GUARD_PAGES != 0) { 456 pmap_qremove(ks, KSTACK_GUARD_PAGES); 457 ks += KSTACK_GUARD_PAGES * PAGE_SIZE; 458 } 459 td->td_kstack = ks; 460 /* 461 * Knowing the number of pages allocated is useful when you 462 * want to deallocate them. 463 */ 464 td->td_kstack_pages = pages; 465 /* 466 * For the length of the stack, link in a real page of ram for each 467 * page of stack. 468 */ 469 VM_OBJECT_LOCK(ksobj); 470 for (i = 0; i < pages; i++) { 471 /* 472 * Get a kernel stack page. 473 */ 474 m = vm_page_grab(ksobj, i, 475 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_WIRED); 476 ma[i] = m; 477 vm_page_lock_queues(); 478 vm_page_wakeup(m); 479 m->valid = VM_PAGE_BITS_ALL; 480 vm_page_unlock_queues(); 481 } 482 VM_OBJECT_UNLOCK(ksobj); 483 pmap_qenter(ks, ma, pages); 484 } 485 486 /* 487 * Dispose of a thread's kernel stack. 488 */ 489 void 490 vm_thread_dispose(struct thread *td) 491 { 492 vm_object_t ksobj; 493 vm_offset_t ks; 494 vm_page_t m; 495 int i, pages; 496 497 pages = td->td_kstack_pages; 498 ksobj = td->td_kstack_obj; 499 ks = td->td_kstack; 500 pmap_qremove(ks, pages); 501 VM_OBJECT_LOCK(ksobj); 502 for (i = 0; i < pages; i++) { 503 m = vm_page_lookup(ksobj, i); 504 if (m == NULL) 505 panic("vm_thread_dispose: kstack already missing?"); 506 vm_page_lock_queues(); 507 vm_page_busy(m); 508 vm_page_unwire(m, 0); 509 vm_page_free(m); 510 vm_page_unlock_queues(); 511 } 512 VM_OBJECT_UNLOCK(ksobj); 513 vm_object_deallocate(ksobj); 514 kmem_free(kernel_map, ks - (KSTACK_GUARD_PAGES * PAGE_SIZE), 515 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 516 } 517 518 /* 519 * Allow a thread's kernel stack to be paged out. 520 */ 521 void 522 vm_thread_swapout(struct thread *td) 523 { 524 vm_object_t ksobj; 525 vm_page_t m; 526 int i, pages; 527 528 cpu_thread_swapout(td); 529 pages = td->td_kstack_pages; 530 ksobj = td->td_kstack_obj; 531 pmap_qremove(td->td_kstack, pages); 532 VM_OBJECT_LOCK(ksobj); 533 for (i = 0; i < pages; i++) { 534 m = vm_page_lookup(ksobj, i); 535 if (m == NULL) 536 panic("vm_thread_swapout: kstack already missing?"); 537 vm_page_lock_queues(); 538 vm_page_dirty(m); 539 vm_page_unwire(m, 0); 540 vm_page_unlock_queues(); 541 } 542 VM_OBJECT_UNLOCK(ksobj); 543 } 544 545 /* 546 * Bring the kernel stack for a specified thread back in. 547 */ 548 void 549 vm_thread_swapin(struct thread *td) 550 { 551 vm_object_t ksobj; 552 vm_page_t m, ma[KSTACK_MAX_PAGES]; 553 int i, pages, rv; 554 555 pages = td->td_kstack_pages; 556 ksobj = td->td_kstack_obj; 557 VM_OBJECT_LOCK(ksobj); 558 for (i = 0; i < pages; i++) { 559 m = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 560 if (m->valid != VM_PAGE_BITS_ALL) { 561 rv = vm_pager_get_pages(ksobj, &m, 1, 0); 562 if (rv != VM_PAGER_OK) 563 panic("vm_thread_swapin: cannot get kstack for proc: %d", td->td_proc->p_pid); 564 m = vm_page_lookup(ksobj, i); 565 m->valid = VM_PAGE_BITS_ALL; 566 } 567 ma[i] = m; 568 vm_page_lock_queues(); 569 vm_page_wire(m); 570 vm_page_wakeup(m); 571 vm_page_unlock_queues(); 572 } 573 VM_OBJECT_UNLOCK(ksobj); 574 pmap_qenter(td->td_kstack, ma, pages); 575 cpu_thread_swapin(td); 576 } 577 578 /* 579 * Set up a variable-sized alternate kstack. 580 */ 581 void 582 vm_thread_new_altkstack(struct thread *td, int pages) 583 { 584 585 td->td_altkstack = td->td_kstack; 586 td->td_altkstack_obj = td->td_kstack_obj; 587 td->td_altkstack_pages = td->td_kstack_pages; 588 589 vm_thread_new(td, pages); 590 } 591 592 /* 593 * Restore the original kstack. 594 */ 595 void 596 vm_thread_dispose_altkstack(struct thread *td) 597 { 598 599 vm_thread_dispose(td); 600 601 td->td_kstack = td->td_altkstack; 602 td->td_kstack_obj = td->td_altkstack_obj; 603 td->td_kstack_pages = td->td_altkstack_pages; 604 td->td_altkstack = 0; 605 td->td_altkstack_obj = NULL; 606 td->td_altkstack_pages = 0; 607 } 608 609 /* 610 * Implement fork's actions on an address space. 611 * Here we arrange for the address space to be copied or referenced, 612 * allocate a user struct (pcb and kernel stack), then call the 613 * machine-dependent layer to fill those in and make the new process 614 * ready to run. The new process is set up so that it returns directly 615 * to user mode to avoid stack copying and relocation problems. 616 */ 617 void 618 vm_forkproc(td, p2, td2, flags) 619 struct thread *td; 620 struct proc *p2; 621 struct thread *td2; 622 int flags; 623 { 624 struct proc *p1 = td->td_proc; 625 struct user *up; 626 627 GIANT_REQUIRED; 628 629 if ((flags & RFPROC) == 0) { 630 /* 631 * Divorce the memory, if it is shared, essentially 632 * this changes shared memory amongst threads, into 633 * COW locally. 634 */ 635 if ((flags & RFMEM) == 0) { 636 if (p1->p_vmspace->vm_refcnt > 1) { 637 vmspace_unshare(p1); 638 } 639 } 640 cpu_fork(td, p2, td2, flags); 641 return; 642 } 643 644 if (flags & RFMEM) { 645 p2->p_vmspace = p1->p_vmspace; 646 p1->p_vmspace->vm_refcnt++; 647 } 648 649 while (vm_page_count_severe()) { 650 VM_WAIT; 651 } 652 653 if ((flags & RFMEM) == 0) { 654 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 655 if (p1->p_vmspace->vm_shm) 656 shmfork(p1, p2); 657 } 658 659 /* XXXKSE this is unsatisfactory but should be adequate */ 660 up = p2->p_uarea; 661 MPASS(p2->p_sigacts != NULL); 662 663 /* 664 * p_stats currently points at fields in the user struct 665 * but not at &u, instead at p_addr. Copy parts of 666 * p_stats; zero the rest of p_stats (statistics). 667 */ 668 p2->p_stats = &up->u_stats; 669 bzero(&up->u_stats.pstat_startzero, 670 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 671 (caddr_t) &up->u_stats.pstat_startzero)); 672 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 673 ((caddr_t) &up->u_stats.pstat_endcopy - 674 (caddr_t) &up->u_stats.pstat_startcopy)); 675 676 /* 677 * cpu_fork will copy and update the pcb, set up the kernel stack, 678 * and make the child ready to run. 679 */ 680 cpu_fork(td, p2, td2, flags); 681 } 682 683 /* 684 * Called after process has been wait(2)'ed apon and is being reaped. 685 * The idea is to reclaim resources that we could not reclaim while 686 * the process was still executing. 687 */ 688 void 689 vm_waitproc(p) 690 struct proc *p; 691 { 692 693 GIANT_REQUIRED; 694 vmspace_exitfree(p); /* and clean-out the vmspace */ 695 } 696 697 /* 698 * Set default limits for VM system. 699 * Called for proc 0, and then inherited by all others. 700 * 701 * XXX should probably act directly on proc0. 702 */ 703 static void 704 vm_init_limits(udata) 705 void *udata; 706 { 707 struct proc *p = udata; 708 struct plimit *limp; 709 int rss_limit; 710 711 /* 712 * Set up the initial limits on process VM. Set the maximum resident 713 * set size to be half of (reasonably) available memory. Since this 714 * is a soft limit, it comes into effect only when the system is out 715 * of memory - half of main memory helps to favor smaller processes, 716 * and reduces thrashing of the object cache. 717 */ 718 limp = p->p_limit; 719 limp->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 720 limp->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 721 limp->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 722 limp->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 723 /* limit the limit to no less than 2MB */ 724 rss_limit = max(cnt.v_free_count, 512); 725 limp->pl_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 726 limp->pl_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 727 } 728 729 void 730 faultin(p) 731 struct proc *p; 732 { 733 #ifdef NO_SWAPPING 734 735 PROC_LOCK_ASSERT(p, MA_OWNED); 736 if ((p->p_sflag & PS_INMEM) == 0) 737 panic("faultin: proc swapped out with NO_SWAPPING!"); 738 #else /* !NO_SWAPPING */ 739 struct thread *td; 740 741 GIANT_REQUIRED; 742 PROC_LOCK_ASSERT(p, MA_OWNED); 743 /* 744 * If another process is swapping in this process, 745 * just wait until it finishes. 746 */ 747 if (p->p_sflag & PS_SWAPPINGIN) 748 msleep(&p->p_sflag, &p->p_mtx, PVM, "faultin", 0); 749 else if ((p->p_sflag & PS_INMEM) == 0) { 750 /* 751 * Don't let another thread swap process p out while we are 752 * busy swapping it in. 753 */ 754 ++p->p_lock; 755 mtx_lock_spin(&sched_lock); 756 p->p_sflag |= PS_SWAPPINGIN; 757 mtx_unlock_spin(&sched_lock); 758 PROC_UNLOCK(p); 759 760 vm_proc_swapin(p); 761 FOREACH_THREAD_IN_PROC(p, td) 762 vm_thread_swapin(td); 763 764 PROC_LOCK(p); 765 mtx_lock_spin(&sched_lock); 766 p->p_sflag &= ~PS_SWAPPINGIN; 767 p->p_sflag |= PS_INMEM; 768 FOREACH_THREAD_IN_PROC(p, td) { 769 TD_CLR_SWAPPED(td); 770 if (TD_CAN_RUN(td)) 771 setrunnable(td); 772 } 773 mtx_unlock_spin(&sched_lock); 774 775 wakeup(&p->p_sflag); 776 777 /* Allow other threads to swap p out now. */ 778 --p->p_lock; 779 } 780 #endif /* NO_SWAPPING */ 781 } 782 783 /* 784 * This swapin algorithm attempts to swap-in processes only if there 785 * is enough space for them. Of course, if a process waits for a long 786 * time, it will be swapped in anyway. 787 * 788 * XXXKSE - process with the thread with highest priority counts.. 789 * 790 * Giant is still held at this point, to be released in tsleep. 791 */ 792 /* ARGSUSED*/ 793 static void 794 scheduler(dummy) 795 void *dummy; 796 { 797 struct proc *p; 798 struct thread *td; 799 int pri; 800 struct proc *pp; 801 int ppri; 802 803 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED); 804 /* GIANT_REQUIRED */ 805 806 loop: 807 if (vm_page_count_min()) { 808 VM_WAIT; 809 goto loop; 810 } 811 812 pp = NULL; 813 ppri = INT_MIN; 814 sx_slock(&allproc_lock); 815 FOREACH_PROC_IN_SYSTEM(p) { 816 struct ksegrp *kg; 817 if (p->p_sflag & (PS_INMEM | PS_SWAPPINGOUT | PS_SWAPPINGIN)) { 818 continue; 819 } 820 mtx_lock_spin(&sched_lock); 821 FOREACH_THREAD_IN_PROC(p, td) { 822 /* 823 * An otherwise runnable thread of a process 824 * swapped out has only the TDI_SWAPPED bit set. 825 * 826 */ 827 if (td->td_inhibitors == TDI_SWAPPED) { 828 kg = td->td_ksegrp; 829 pri = p->p_swtime + kg->kg_slptime; 830 if ((p->p_sflag & PS_SWAPINREQ) == 0) { 831 pri -= kg->kg_nice * 8; 832 } 833 834 /* 835 * if this ksegrp is higher priority 836 * and there is enough space, then select 837 * this process instead of the previous 838 * selection. 839 */ 840 if (pri > ppri) { 841 pp = p; 842 ppri = pri; 843 } 844 } 845 } 846 mtx_unlock_spin(&sched_lock); 847 } 848 sx_sunlock(&allproc_lock); 849 850 /* 851 * Nothing to do, back to sleep. 852 */ 853 if ((p = pp) == NULL) { 854 tsleep(&proc0, PVM, "sched", maxslp * hz / 2); 855 goto loop; 856 } 857 PROC_LOCK(p); 858 859 /* 860 * Another process may be bringing or may have already 861 * brought this process in while we traverse all threads. 862 * Or, this process may even be being swapped out again. 863 */ 864 if (p->p_sflag & (PS_INMEM | PS_SWAPPINGOUT | PS_SWAPPINGIN)) { 865 PROC_UNLOCK(p); 866 goto loop; 867 } 868 869 mtx_lock_spin(&sched_lock); 870 p->p_sflag &= ~PS_SWAPINREQ; 871 mtx_unlock_spin(&sched_lock); 872 873 /* 874 * We would like to bring someone in. (only if there is space). 875 * [What checks the space? ] 876 */ 877 faultin(p); 878 PROC_UNLOCK(p); 879 mtx_lock_spin(&sched_lock); 880 p->p_swtime = 0; 881 mtx_unlock_spin(&sched_lock); 882 goto loop; 883 } 884 885 #ifndef NO_SWAPPING 886 887 /* 888 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 889 */ 890 static int swap_idle_threshold1 = 2; 891 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW, 892 &swap_idle_threshold1, 0, "Guaranteed swapped in time for a process"); 893 894 /* 895 * Swap_idle_threshold2 is the time that a process can be idle before 896 * it will be swapped out, if idle swapping is enabled. 897 */ 898 static int swap_idle_threshold2 = 10; 899 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW, 900 &swap_idle_threshold2, 0, "Time before a process will be swapped out"); 901 902 /* 903 * Swapout is driven by the pageout daemon. Very simple, we find eligible 904 * procs and unwire their u-areas. We try to always "swap" at least one 905 * process in case we need the room for a swapin. 906 * If any procs have been sleeping/stopped for at least maxslp seconds, 907 * they are swapped. Else, we swap the longest-sleeping or stopped process, 908 * if any, otherwise the longest-resident process. 909 */ 910 void 911 swapout_procs(action) 912 int action; 913 { 914 struct proc *p; 915 struct thread *td; 916 struct ksegrp *kg; 917 int didswap = 0; 918 919 GIANT_REQUIRED; 920 921 retry: 922 sx_slock(&allproc_lock); 923 FOREACH_PROC_IN_SYSTEM(p) { 924 struct vmspace *vm; 925 int minslptime = 100000; 926 927 /* 928 * Watch out for a process in 929 * creation. It may have no 930 * address space or lock yet. 931 */ 932 mtx_lock_spin(&sched_lock); 933 if (p->p_state == PRS_NEW) { 934 mtx_unlock_spin(&sched_lock); 935 continue; 936 } 937 mtx_unlock_spin(&sched_lock); 938 939 /* 940 * An aio daemon switches its 941 * address space while running. 942 * Perform a quick check whether 943 * a process has P_SYSTEM. 944 */ 945 if ((p->p_flag & P_SYSTEM) != 0) 946 continue; 947 948 /* 949 * Do not swapout a process that 950 * is waiting for VM data 951 * structures as there is a possible 952 * deadlock. Test this first as 953 * this may block. 954 * 955 * Lock the map until swapout 956 * finishes, or a thread of this 957 * process may attempt to alter 958 * the map. 959 */ 960 PROC_LOCK(p); 961 vm = p->p_vmspace; 962 KASSERT(vm != NULL, 963 ("swapout_procs: a process has no address space")); 964 ++vm->vm_refcnt; 965 PROC_UNLOCK(p); 966 if (!vm_map_trylock(&vm->vm_map)) 967 goto nextproc1; 968 969 PROC_LOCK(p); 970 if (p->p_lock != 0 || 971 (p->p_flag & (P_STOPPED_SINGLE|P_TRACED|P_SYSTEM|P_WEXIT) 972 ) != 0) { 973 goto nextproc2; 974 } 975 /* 976 * only aiod changes vmspace, however it will be 977 * skipped because of the if statement above checking 978 * for P_SYSTEM 979 */ 980 if ((p->p_sflag & (PS_INMEM|PS_SWAPPINGOUT|PS_SWAPPINGIN)) != PS_INMEM) 981 goto nextproc2; 982 983 switch (p->p_state) { 984 default: 985 /* Don't swap out processes in any sort 986 * of 'special' state. */ 987 break; 988 989 case PRS_NORMAL: 990 mtx_lock_spin(&sched_lock); 991 /* 992 * do not swapout a realtime process 993 * Check all the thread groups.. 994 */ 995 FOREACH_KSEGRP_IN_PROC(p, kg) { 996 if (PRI_IS_REALTIME(kg->kg_pri_class)) 997 goto nextproc; 998 999 /* 1000 * Guarantee swap_idle_threshold1 1001 * time in memory. 1002 */ 1003 if (kg->kg_slptime < swap_idle_threshold1) 1004 goto nextproc; 1005 1006 /* 1007 * Do not swapout a process if it is 1008 * waiting on a critical event of some 1009 * kind or there is a thread whose 1010 * pageable memory may be accessed. 1011 * 1012 * This could be refined to support 1013 * swapping out a thread. 1014 */ 1015 FOREACH_THREAD_IN_GROUP(kg, td) { 1016 if ((td->td_priority) < PSOCK || 1017 !thread_safetoswapout(td)) 1018 goto nextproc; 1019 } 1020 /* 1021 * If the system is under memory stress, 1022 * or if we are swapping 1023 * idle processes >= swap_idle_threshold2, 1024 * then swap the process out. 1025 */ 1026 if (((action & VM_SWAP_NORMAL) == 0) && 1027 (((action & VM_SWAP_IDLE) == 0) || 1028 (kg->kg_slptime < swap_idle_threshold2))) 1029 goto nextproc; 1030 1031 if (minslptime > kg->kg_slptime) 1032 minslptime = kg->kg_slptime; 1033 } 1034 1035 /* 1036 * If the process has been asleep for awhile and had 1037 * most of its pages taken away already, swap it out. 1038 */ 1039 if ((action & VM_SWAP_NORMAL) || 1040 ((action & VM_SWAP_IDLE) && 1041 (minslptime > swap_idle_threshold2))) { 1042 swapout(p); 1043 didswap++; 1044 mtx_unlock_spin(&sched_lock); 1045 PROC_UNLOCK(p); 1046 vm_map_unlock(&vm->vm_map); 1047 vmspace_free(vm); 1048 sx_sunlock(&allproc_lock); 1049 goto retry; 1050 } 1051 nextproc: 1052 mtx_unlock_spin(&sched_lock); 1053 } 1054 nextproc2: 1055 PROC_UNLOCK(p); 1056 vm_map_unlock(&vm->vm_map); 1057 nextproc1: 1058 vmspace_free(vm); 1059 continue; 1060 } 1061 sx_sunlock(&allproc_lock); 1062 /* 1063 * If we swapped something out, and another process needed memory, 1064 * then wakeup the sched process. 1065 */ 1066 if (didswap) 1067 wakeup(&proc0); 1068 } 1069 1070 static void 1071 swapout(p) 1072 struct proc *p; 1073 { 1074 struct thread *td; 1075 1076 PROC_LOCK_ASSERT(p, MA_OWNED); 1077 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 1078 #if defined(SWAP_DEBUG) 1079 printf("swapping out %d\n", p->p_pid); 1080 #endif 1081 1082 /* 1083 * The states of this process and its threads may have changed 1084 * by now. Assuming that there is only one pageout daemon thread, 1085 * this process should still be in memory. 1086 */ 1087 KASSERT((p->p_sflag & (PS_INMEM|PS_SWAPPINGOUT|PS_SWAPPINGIN)) == PS_INMEM, 1088 ("swapout: lost a swapout race?")); 1089 1090 #if defined(INVARIANTS) 1091 /* 1092 * Make sure that all threads are safe to be swapped out. 1093 * 1094 * Alternatively, we could swap out only safe threads. 1095 */ 1096 FOREACH_THREAD_IN_PROC(p, td) { 1097 KASSERT(thread_safetoswapout(td), 1098 ("swapout: there is a thread not safe for swapout")); 1099 } 1100 #endif /* INVARIANTS */ 1101 1102 ++p->p_stats->p_ru.ru_nswap; 1103 /* 1104 * remember the process resident count 1105 */ 1106 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 1107 1108 p->p_sflag &= ~PS_INMEM; 1109 p->p_sflag |= PS_SWAPPINGOUT; 1110 PROC_UNLOCK(p); 1111 FOREACH_THREAD_IN_PROC(p, td) 1112 TD_SET_SWAPPED(td); 1113 mtx_unlock_spin(&sched_lock); 1114 1115 vm_proc_swapout(p); 1116 FOREACH_THREAD_IN_PROC(p, td) 1117 vm_thread_swapout(td); 1118 1119 PROC_LOCK(p); 1120 mtx_lock_spin(&sched_lock); 1121 p->p_sflag &= ~PS_SWAPPINGOUT; 1122 p->p_swtime = 0; 1123 } 1124 #endif /* !NO_SWAPPING */ 1125