1 /*- 2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) 3 * 4 * Copyright (c) 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * The Mach Operating System project at Carnegie-Mellon University. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 35 * 36 * 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 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 /* 64 * Kernel memory management. 65 */ 66 67 #include <sys/cdefs.h> 68 __FBSDID("$FreeBSD$"); 69 70 #include "opt_vm.h" 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/kernel.h> /* for ticks and hz */ 75 #include <sys/domainset.h> 76 #include <sys/eventhandler.h> 77 #include <sys/lock.h> 78 #include <sys/proc.h> 79 #include <sys/malloc.h> 80 #include <sys/rwlock.h> 81 #include <sys/sysctl.h> 82 #include <sys/vmem.h> 83 #include <sys/vmmeter.h> 84 85 #include <vm/vm.h> 86 #include <vm/vm_param.h> 87 #include <vm/vm_domainset.h> 88 #include <vm/vm_kern.h> 89 #include <vm/pmap.h> 90 #include <vm/vm_map.h> 91 #include <vm/vm_object.h> 92 #include <vm/vm_page.h> 93 #include <vm/vm_pageout.h> 94 #include <vm/vm_phys.h> 95 #include <vm/vm_pagequeue.h> 96 #include <vm/vm_radix.h> 97 #include <vm/vm_extern.h> 98 #include <vm/uma.h> 99 100 vm_map_t kernel_map; 101 vm_map_t exec_map; 102 vm_map_t pipe_map; 103 104 const void *zero_region; 105 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0); 106 107 /* NB: Used by kernel debuggers. */ 108 const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS; 109 110 u_int exec_map_entry_size; 111 u_int exec_map_entries; 112 113 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD, 114 SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address"); 115 116 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD, 117 #if defined(__arm__) 118 &vm_max_kernel_address, 0, 119 #else 120 SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS, 121 #endif 122 "Max kernel address"); 123 124 #if VM_NRESERVLEVEL > 0 125 #define KVA_QUANTUM_SHIFT (VM_LEVEL_0_ORDER + PAGE_SHIFT) 126 #else 127 /* On non-superpage architectures we want large import sizes. */ 128 #define KVA_QUANTUM_SHIFT (8 + PAGE_SHIFT) 129 #endif 130 #define KVA_QUANTUM (1 << KVA_QUANTUM_SHIFT) 131 132 extern void uma_startup2(void); 133 134 /* 135 * kva_alloc: 136 * 137 * Allocate a virtual address range with no underlying object and 138 * no initial mapping to physical memory. Any mapping from this 139 * range to physical memory must be explicitly created prior to 140 * its use, typically with pmap_qenter(). Any attempt to create 141 * a mapping on demand through vm_fault() will result in a panic. 142 */ 143 vm_offset_t 144 kva_alloc(vm_size_t size) 145 { 146 vm_offset_t addr; 147 148 size = round_page(size); 149 if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr)) 150 return (0); 151 152 return (addr); 153 } 154 155 /* 156 * kva_free: 157 * 158 * Release a region of kernel virtual memory allocated 159 * with kva_alloc, and return the physical pages 160 * associated with that region. 161 * 162 * This routine may not block on kernel maps. 163 */ 164 void 165 kva_free(vm_offset_t addr, vm_size_t size) 166 { 167 168 size = round_page(size); 169 vmem_free(kernel_arena, addr, size); 170 } 171 172 static vm_page_t 173 kmem_alloc_contig_pages(vm_object_t object, vm_pindex_t pindex, int domain, 174 int pflags, u_long npages, vm_paddr_t low, vm_paddr_t high, 175 u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr) 176 { 177 vm_page_t m; 178 int tries; 179 bool wait; 180 181 VM_OBJECT_ASSERT_WLOCKED(object); 182 183 wait = (pflags & VM_ALLOC_WAITOK) != 0; 184 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL); 185 pflags |= VM_ALLOC_NOWAIT; 186 for (tries = wait ? 3 : 1;; tries--) { 187 m = vm_page_alloc_contig_domain(object, pindex, domain, pflags, 188 npages, low, high, alignment, boundary, memattr); 189 if (m != NULL || tries == 0) 190 break; 191 192 VM_OBJECT_WUNLOCK(object); 193 if (!vm_page_reclaim_contig_domain(domain, pflags, npages, 194 low, high, alignment, boundary) && wait) 195 vm_wait_domain(domain); 196 VM_OBJECT_WLOCK(object); 197 } 198 return (m); 199 } 200 201 /* 202 * Allocates a region from the kernel address map and physical pages 203 * within the specified address range to the kernel object. Creates a 204 * wired mapping from this region to these pages, and returns the 205 * region's starting virtual address. The allocated pages are not 206 * necessarily physically contiguous. If M_ZERO is specified through the 207 * given flags, then the pages are zeroed before they are mapped. 208 */ 209 static vm_offset_t 210 kmem_alloc_attr_domain(int domain, vm_size_t size, int flags, vm_paddr_t low, 211 vm_paddr_t high, vm_memattr_t memattr) 212 { 213 vmem_t *vmem; 214 vm_object_t object; 215 vm_offset_t addr, i, offset; 216 vm_page_t m; 217 int pflags; 218 vm_prot_t prot; 219 220 object = kernel_object; 221 size = round_page(size); 222 vmem = vm_dom[domain].vmd_kernel_arena; 223 if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr)) 224 return (0); 225 offset = addr - VM_MIN_KERNEL_ADDRESS; 226 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED; 227 prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW; 228 VM_OBJECT_WLOCK(object); 229 for (i = 0; i < size; i += PAGE_SIZE) { 230 m = kmem_alloc_contig_pages(object, atop(offset + i), 231 domain, pflags, 1, low, high, PAGE_SIZE, 0, memattr); 232 if (m == NULL) { 233 VM_OBJECT_WUNLOCK(object); 234 kmem_unback(object, addr, i); 235 vmem_free(vmem, addr, size); 236 return (0); 237 } 238 KASSERT(vm_phys_domain(m) == domain, 239 ("kmem_alloc_attr_domain: Domain mismatch %d != %d", 240 vm_phys_domain(m), domain)); 241 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0) 242 pmap_zero_page(m); 243 vm_page_valid(m); 244 pmap_enter(kernel_pmap, addr + i, m, prot, 245 prot | PMAP_ENTER_WIRED, 0); 246 } 247 VM_OBJECT_WUNLOCK(object); 248 return (addr); 249 } 250 251 vm_offset_t 252 kmem_alloc_attr(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high, 253 vm_memattr_t memattr) 254 { 255 256 return (kmem_alloc_attr_domainset(DOMAINSET_RR(), size, flags, low, 257 high, memattr)); 258 } 259 260 vm_offset_t 261 kmem_alloc_attr_domainset(struct domainset *ds, vm_size_t size, int flags, 262 vm_paddr_t low, vm_paddr_t high, vm_memattr_t memattr) 263 { 264 struct vm_domainset_iter di; 265 vm_offset_t addr; 266 int domain; 267 268 vm_domainset_iter_policy_init(&di, ds, &domain, &flags); 269 do { 270 addr = kmem_alloc_attr_domain(domain, size, flags, low, high, 271 memattr); 272 if (addr != 0) 273 break; 274 } while (vm_domainset_iter_policy(&di, &domain) == 0); 275 276 return (addr); 277 } 278 279 /* 280 * Allocates a region from the kernel address map and physically 281 * contiguous pages within the specified address range to the kernel 282 * object. Creates a wired mapping from this region to these pages, and 283 * returns the region's starting virtual address. If M_ZERO is specified 284 * through the given flags, then the pages are zeroed before they are 285 * mapped. 286 */ 287 static vm_offset_t 288 kmem_alloc_contig_domain(int domain, vm_size_t size, int flags, vm_paddr_t low, 289 vm_paddr_t high, u_long alignment, vm_paddr_t boundary, 290 vm_memattr_t memattr) 291 { 292 vmem_t *vmem; 293 vm_object_t object; 294 vm_offset_t addr, offset, tmp; 295 vm_page_t end_m, m; 296 u_long npages; 297 int pflags; 298 299 object = kernel_object; 300 size = round_page(size); 301 vmem = vm_dom[domain].vmd_kernel_arena; 302 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr)) 303 return (0); 304 offset = addr - VM_MIN_KERNEL_ADDRESS; 305 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED; 306 npages = atop(size); 307 VM_OBJECT_WLOCK(object); 308 m = kmem_alloc_contig_pages(object, atop(offset), domain, 309 pflags, npages, low, high, alignment, boundary, memattr); 310 if (m == NULL) { 311 VM_OBJECT_WUNLOCK(object); 312 vmem_free(vmem, addr, size); 313 return (0); 314 } 315 KASSERT(vm_phys_domain(m) == domain, 316 ("kmem_alloc_contig_domain: Domain mismatch %d != %d", 317 vm_phys_domain(m), domain)); 318 end_m = m + npages; 319 tmp = addr; 320 for (; m < end_m; m++) { 321 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0) 322 pmap_zero_page(m); 323 vm_page_valid(m); 324 pmap_enter(kernel_pmap, tmp, m, VM_PROT_RW, 325 VM_PROT_RW | PMAP_ENTER_WIRED, 0); 326 tmp += PAGE_SIZE; 327 } 328 VM_OBJECT_WUNLOCK(object); 329 return (addr); 330 } 331 332 vm_offset_t 333 kmem_alloc_contig(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high, 334 u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr) 335 { 336 337 return (kmem_alloc_contig_domainset(DOMAINSET_RR(), size, flags, low, 338 high, alignment, boundary, memattr)); 339 } 340 341 vm_offset_t 342 kmem_alloc_contig_domainset(struct domainset *ds, vm_size_t size, int flags, 343 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, 344 vm_memattr_t memattr) 345 { 346 struct vm_domainset_iter di; 347 vm_offset_t addr; 348 int domain; 349 350 vm_domainset_iter_policy_init(&di, ds, &domain, &flags); 351 do { 352 addr = kmem_alloc_contig_domain(domain, size, flags, low, high, 353 alignment, boundary, memattr); 354 if (addr != 0) 355 break; 356 } while (vm_domainset_iter_policy(&di, &domain) == 0); 357 358 return (addr); 359 } 360 361 /* 362 * kmem_suballoc: 363 * 364 * Allocates a map to manage a subrange 365 * of the kernel virtual address space. 366 * 367 * Arguments are as follows: 368 * 369 * parent Map to take range from 370 * min, max Returned endpoints of map 371 * size Size of range to find 372 * superpage_align Request that min is superpage aligned 373 */ 374 vm_map_t 375 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max, 376 vm_size_t size, boolean_t superpage_align) 377 { 378 int ret; 379 vm_map_t result; 380 381 size = round_page(size); 382 383 *min = vm_map_min(parent); 384 ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ? 385 VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL, 386 MAP_ACC_NO_CHARGE); 387 if (ret != KERN_SUCCESS) 388 panic("kmem_suballoc: bad status return of %d", ret); 389 *max = *min + size; 390 result = vm_map_create(vm_map_pmap(parent), *min, *max); 391 if (result == NULL) 392 panic("kmem_suballoc: cannot create submap"); 393 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS) 394 panic("kmem_suballoc: unable to change range to submap"); 395 return (result); 396 } 397 398 /* 399 * kmem_malloc_domain: 400 * 401 * Allocate wired-down pages in the kernel's address space. 402 */ 403 static vm_offset_t 404 kmem_malloc_domain(int domain, vm_size_t size, int flags) 405 { 406 vmem_t *arena; 407 vm_offset_t addr; 408 int rv; 409 410 if (__predict_true((flags & M_EXEC) == 0)) 411 arena = vm_dom[domain].vmd_kernel_arena; 412 else 413 arena = vm_dom[domain].vmd_kernel_rwx_arena; 414 size = round_page(size); 415 if (vmem_alloc(arena, size, flags | M_BESTFIT, &addr)) 416 return (0); 417 418 rv = kmem_back_domain(domain, kernel_object, addr, size, flags); 419 if (rv != KERN_SUCCESS) { 420 vmem_free(arena, addr, size); 421 return (0); 422 } 423 return (addr); 424 } 425 426 vm_offset_t 427 kmem_malloc(vm_size_t size, int flags) 428 { 429 430 return (kmem_malloc_domainset(DOMAINSET_RR(), size, flags)); 431 } 432 433 vm_offset_t 434 kmem_malloc_domainset(struct domainset *ds, vm_size_t size, int flags) 435 { 436 struct vm_domainset_iter di; 437 vm_offset_t addr; 438 int domain; 439 440 vm_domainset_iter_policy_init(&di, ds, &domain, &flags); 441 do { 442 addr = kmem_malloc_domain(domain, size, flags); 443 if (addr != 0) 444 break; 445 } while (vm_domainset_iter_policy(&di, &domain) == 0); 446 447 return (addr); 448 } 449 450 /* 451 * kmem_back_domain: 452 * 453 * Allocate physical pages from the specified domain for the specified 454 * virtual address range. 455 */ 456 int 457 kmem_back_domain(int domain, vm_object_t object, vm_offset_t addr, 458 vm_size_t size, int flags) 459 { 460 vm_offset_t offset, i; 461 vm_page_t m, mpred; 462 vm_prot_t prot; 463 int pflags; 464 465 KASSERT(object == kernel_object, 466 ("kmem_back_domain: only supports kernel object.")); 467 468 offset = addr - VM_MIN_KERNEL_ADDRESS; 469 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED; 470 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL); 471 if (flags & M_WAITOK) 472 pflags |= VM_ALLOC_WAITFAIL; 473 prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW; 474 475 i = 0; 476 VM_OBJECT_WLOCK(object); 477 retry: 478 mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i)); 479 for (; i < size; i += PAGE_SIZE, mpred = m) { 480 m = vm_page_alloc_domain_after(object, atop(offset + i), 481 domain, pflags, mpred); 482 483 /* 484 * Ran out of space, free everything up and return. Don't need 485 * to lock page queues here as we know that the pages we got 486 * aren't on any queues. 487 */ 488 if (m == NULL) { 489 if ((flags & M_NOWAIT) == 0) 490 goto retry; 491 VM_OBJECT_WUNLOCK(object); 492 kmem_unback(object, addr, i); 493 return (KERN_NO_SPACE); 494 } 495 KASSERT(vm_phys_domain(m) == domain, 496 ("kmem_back_domain: Domain mismatch %d != %d", 497 vm_phys_domain(m), domain)); 498 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0) 499 pmap_zero_page(m); 500 KASSERT((m->oflags & VPO_UNMANAGED) != 0, 501 ("kmem_malloc: page %p is managed", m)); 502 vm_page_valid(m); 503 pmap_enter(kernel_pmap, addr + i, m, prot, 504 prot | PMAP_ENTER_WIRED, 0); 505 if (__predict_false((prot & VM_PROT_EXECUTE) != 0)) 506 m->oflags |= VPO_KMEM_EXEC; 507 } 508 VM_OBJECT_WUNLOCK(object); 509 510 return (KERN_SUCCESS); 511 } 512 513 /* 514 * kmem_back: 515 * 516 * Allocate physical pages for the specified virtual address range. 517 */ 518 int 519 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags) 520 { 521 vm_offset_t end, next, start; 522 int domain, rv; 523 524 KASSERT(object == kernel_object, 525 ("kmem_back: only supports kernel object.")); 526 527 for (start = addr, end = addr + size; addr < end; addr = next) { 528 /* 529 * We must ensure that pages backing a given large virtual page 530 * all come from the same physical domain. 531 */ 532 if (vm_ndomains > 1) { 533 domain = (addr >> KVA_QUANTUM_SHIFT) % vm_ndomains; 534 while (VM_DOMAIN_EMPTY(domain)) 535 domain++; 536 next = roundup2(addr + 1, KVA_QUANTUM); 537 if (next > end || next < start) 538 next = end; 539 } else { 540 domain = 0; 541 next = end; 542 } 543 rv = kmem_back_domain(domain, object, addr, next - addr, flags); 544 if (rv != KERN_SUCCESS) { 545 kmem_unback(object, start, addr - start); 546 break; 547 } 548 } 549 return (rv); 550 } 551 552 /* 553 * kmem_unback: 554 * 555 * Unmap and free the physical pages underlying the specified virtual 556 * address range. 557 * 558 * A physical page must exist within the specified object at each index 559 * that is being unmapped. 560 */ 561 static struct vmem * 562 _kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size) 563 { 564 struct vmem *arena; 565 vm_page_t m, next; 566 vm_offset_t end, offset; 567 int domain; 568 569 KASSERT(object == kernel_object, 570 ("kmem_unback: only supports kernel object.")); 571 572 if (size == 0) 573 return (NULL); 574 pmap_remove(kernel_pmap, addr, addr + size); 575 offset = addr - VM_MIN_KERNEL_ADDRESS; 576 end = offset + size; 577 VM_OBJECT_WLOCK(object); 578 m = vm_page_lookup(object, atop(offset)); 579 domain = vm_phys_domain(m); 580 if (__predict_true((m->oflags & VPO_KMEM_EXEC) == 0)) 581 arena = vm_dom[domain].vmd_kernel_arena; 582 else 583 arena = vm_dom[domain].vmd_kernel_rwx_arena; 584 for (; offset < end; offset += PAGE_SIZE, m = next) { 585 next = vm_page_next(m); 586 vm_page_xbusy_claim(m); 587 vm_page_unwire_noq(m); 588 vm_page_free(m); 589 } 590 VM_OBJECT_WUNLOCK(object); 591 592 return (arena); 593 } 594 595 void 596 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size) 597 { 598 599 (void)_kmem_unback(object, addr, size); 600 } 601 602 /* 603 * kmem_free: 604 * 605 * Free memory allocated with kmem_malloc. The size must match the 606 * original allocation. 607 */ 608 void 609 kmem_free(vm_offset_t addr, vm_size_t size) 610 { 611 struct vmem *arena; 612 613 size = round_page(size); 614 arena = _kmem_unback(kernel_object, addr, size); 615 if (arena != NULL) 616 vmem_free(arena, addr, size); 617 } 618 619 /* 620 * kmap_alloc_wait: 621 * 622 * Allocates pageable memory from a sub-map of the kernel. If the submap 623 * has no room, the caller sleeps waiting for more memory in the submap. 624 * 625 * This routine may block. 626 */ 627 vm_offset_t 628 kmap_alloc_wait(vm_map_t map, vm_size_t size) 629 { 630 vm_offset_t addr; 631 632 size = round_page(size); 633 if (!swap_reserve(size)) 634 return (0); 635 636 for (;;) { 637 /* 638 * To make this work for more than one map, use the map's lock 639 * to lock out sleepers/wakers. 640 */ 641 vm_map_lock(map); 642 addr = vm_map_findspace(map, vm_map_min(map), size); 643 if (addr + size <= vm_map_max(map)) 644 break; 645 /* no space now; see if we can ever get space */ 646 if (vm_map_max(map) - vm_map_min(map) < size) { 647 vm_map_unlock(map); 648 swap_release(size); 649 return (0); 650 } 651 map->needs_wakeup = TRUE; 652 vm_map_unlock_and_wait(map, 0); 653 } 654 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_RW, VM_PROT_RW, 655 MAP_ACC_CHARGED); 656 vm_map_unlock(map); 657 return (addr); 658 } 659 660 /* 661 * kmap_free_wakeup: 662 * 663 * Returns memory to a submap of the kernel, and wakes up any processes 664 * waiting for memory in that map. 665 */ 666 void 667 kmap_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size) 668 { 669 670 vm_map_lock(map); 671 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 672 if (map->needs_wakeup) { 673 map->needs_wakeup = FALSE; 674 vm_map_wakeup(map); 675 } 676 vm_map_unlock(map); 677 } 678 679 void 680 kmem_init_zero_region(void) 681 { 682 vm_offset_t addr, i; 683 vm_page_t m; 684 685 /* 686 * Map a single physical page of zeros to a larger virtual range. 687 * This requires less looping in places that want large amounts of 688 * zeros, while not using much more physical resources. 689 */ 690 addr = kva_alloc(ZERO_REGION_SIZE); 691 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL | 692 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO); 693 if ((m->flags & PG_ZERO) == 0) 694 pmap_zero_page(m); 695 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE) 696 pmap_qenter(addr + i, &m, 1); 697 pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ); 698 699 zero_region = (const void *)addr; 700 } 701 702 /* 703 * Import KVA from the kernel map into the kernel arena. 704 */ 705 static int 706 kva_import(void *unused, vmem_size_t size, int flags, vmem_addr_t *addrp) 707 { 708 vm_offset_t addr; 709 int result; 710 711 KASSERT((size % KVA_QUANTUM) == 0, 712 ("kva_import: Size %jd is not a multiple of %d", 713 (intmax_t)size, (int)KVA_QUANTUM)); 714 addr = vm_map_min(kernel_map); 715 result = vm_map_find(kernel_map, NULL, 0, &addr, size, 0, 716 VMFS_SUPER_SPACE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT); 717 if (result != KERN_SUCCESS) 718 return (ENOMEM); 719 720 *addrp = addr; 721 722 return (0); 723 } 724 725 /* 726 * Import KVA from a parent arena into a per-domain arena. Imports must be 727 * KVA_QUANTUM-aligned and a multiple of KVA_QUANTUM in size. 728 */ 729 static int 730 kva_import_domain(void *arena, vmem_size_t size, int flags, vmem_addr_t *addrp) 731 { 732 733 KASSERT((size % KVA_QUANTUM) == 0, 734 ("kva_import_domain: Size %jd is not a multiple of %d", 735 (intmax_t)size, (int)KVA_QUANTUM)); 736 return (vmem_xalloc(arena, size, KVA_QUANTUM, 0, 0, VMEM_ADDR_MIN, 737 VMEM_ADDR_MAX, flags, addrp)); 738 } 739 740 /* 741 * kmem_init: 742 * 743 * Create the kernel map; insert a mapping covering kernel text, 744 * data, bss, and all space allocated thus far (`boostrap' data). The 745 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and 746 * `start' as allocated, and the range between `start' and `end' as free. 747 * Create the kernel vmem arena and its per-domain children. 748 */ 749 void 750 kmem_init(vm_offset_t start, vm_offset_t end) 751 { 752 vm_map_t m; 753 int domain; 754 755 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end); 756 m->system_map = 1; 757 vm_map_lock(m); 758 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 759 kernel_map = m; 760 (void)vm_map_insert(m, NULL, 0, 761 #ifdef __amd64__ 762 KERNBASE, 763 #else 764 VM_MIN_KERNEL_ADDRESS, 765 #endif 766 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT); 767 /* ... and ending with the completion of the above `insert' */ 768 769 #ifdef __amd64__ 770 /* 771 * Mark KVA used for the page array as allocated. Other platforms 772 * that handle vm_page_array allocation can simply adjust virtual_avail 773 * instead. 774 */ 775 (void)vm_map_insert(m, NULL, 0, (vm_offset_t)vm_page_array, 776 (vm_offset_t)vm_page_array + round_2mpage(vm_page_array_size * 777 sizeof(struct vm_page)), 778 VM_PROT_RW, VM_PROT_RW, MAP_NOFAULT); 779 #endif 780 vm_map_unlock(m); 781 782 /* 783 * Initialize the kernel_arena. This can grow on demand. 784 */ 785 vmem_init(kernel_arena, "kernel arena", 0, 0, PAGE_SIZE, 0, 0); 786 vmem_set_import(kernel_arena, kva_import, NULL, NULL, KVA_QUANTUM); 787 788 for (domain = 0; domain < vm_ndomains; domain++) { 789 /* 790 * Initialize the per-domain arenas. These are used to color 791 * the KVA space in a way that ensures that virtual large pages 792 * are backed by memory from the same physical domain, 793 * maximizing the potential for superpage promotion. 794 */ 795 vm_dom[domain].vmd_kernel_arena = vmem_create( 796 "kernel arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK); 797 vmem_set_import(vm_dom[domain].vmd_kernel_arena, 798 kva_import_domain, NULL, kernel_arena, KVA_QUANTUM); 799 800 /* 801 * In architectures with superpages, maintain separate arenas 802 * for allocations with permissions that differ from the 803 * "standard" read/write permissions used for kernel memory, 804 * so as not to inhibit superpage promotion. 805 */ 806 #if VM_NRESERVLEVEL > 0 807 vm_dom[domain].vmd_kernel_rwx_arena = vmem_create( 808 "kernel rwx arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK); 809 vmem_set_import(vm_dom[domain].vmd_kernel_rwx_arena, 810 kva_import_domain, (vmem_release_t *)vmem_xfree, 811 kernel_arena, KVA_QUANTUM); 812 #else 813 vm_dom[domain].vmd_kernel_rwx_arena = 814 vm_dom[domain].vmd_kernel_arena; 815 #endif 816 } 817 818 /* 819 * This must be the very first call so that the virtual address 820 * space used for early allocations is properly marked used in 821 * the map. 822 */ 823 uma_startup2(); 824 } 825 826 /* 827 * kmem_bootstrap_free: 828 * 829 * Free pages backing preloaded data (e.g., kernel modules) to the 830 * system. Currently only supported on platforms that create a 831 * vm_phys segment for preloaded data. 832 */ 833 void 834 kmem_bootstrap_free(vm_offset_t start, vm_size_t size) 835 { 836 #if defined(__i386__) || defined(__amd64__) 837 struct vm_domain *vmd; 838 vm_offset_t end, va; 839 vm_paddr_t pa; 840 vm_page_t m; 841 842 end = trunc_page(start + size); 843 start = round_page(start); 844 845 #ifdef __amd64__ 846 /* 847 * Preloaded files do not have execute permissions by default on amd64. 848 * Restore the default permissions to ensure that the direct map alias 849 * is updated. 850 */ 851 pmap_change_prot(start, end - start, VM_PROT_RW); 852 #endif 853 for (va = start; va < end; va += PAGE_SIZE) { 854 pa = pmap_kextract(va); 855 m = PHYS_TO_VM_PAGE(pa); 856 857 vmd = vm_pagequeue_domain(m); 858 vm_domain_free_lock(vmd); 859 vm_phys_free_pages(m, 0); 860 vm_domain_free_unlock(vmd); 861 862 vm_domain_freecnt_inc(vmd, 1); 863 vm_cnt.v_page_count++; 864 } 865 pmap_remove(kernel_pmap, start, end); 866 (void)vmem_add(kernel_arena, start, end - start, M_WAITOK); 867 #endif 868 } 869 870 /* 871 * Allow userspace to directly trigger the VM drain routine for testing 872 * purposes. 873 */ 874 static int 875 debug_vm_lowmem(SYSCTL_HANDLER_ARGS) 876 { 877 int error, i; 878 879 i = 0; 880 error = sysctl_handle_int(oidp, &i, 0, req); 881 if (error) 882 return (error); 883 if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0) 884 return (EINVAL); 885 if (i != 0) 886 EVENTHANDLER_INVOKE(vm_lowmem, i); 887 return (0); 888 } 889 890 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0, 891 debug_vm_lowmem, "I", "set to trigger vm_lowmem event with given flags"); 892