uvm_km.c revision 1.48
1/* $NetBSD: uvm_km.c,v 1.48 2001/05/26 16:32:47 chs Exp $ */ 2 3/* 4 * Copyright (c) 1997 Charles D. Cranor and Washington University. 5 * Copyright (c) 1991, 1993, The Regents of the University of California. 6 * 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to Berkeley by 10 * The Mach Operating System project at Carnegie-Mellon University. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. All advertising materials mentioning features or use of this software 21 * must display the following acknowledgement: 22 * This product includes software developed by Charles D. Cranor, 23 * Washington University, the University of California, Berkeley and 24 * its contributors. 25 * 4. Neither the name of the University nor the names of its contributors 26 * may be used to endorse or promote products derived from this software 27 * without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39 * SUCH DAMAGE. 40 * 41 * @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 42 * from: Id: uvm_km.c,v 1.1.2.14 1998/02/06 05:19:27 chs Exp 43 * 44 * 45 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 46 * All rights reserved. 47 * 48 * Permission to use, copy, modify and distribute this software and 49 * its documentation is hereby granted, provided that both the copyright 50 * notice and this permission notice appear in all copies of the 51 * software, derivative works or modified versions, and any portions 52 * thereof, and that both notices appear in supporting documentation. 53 * 54 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 55 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 56 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 57 * 58 * Carnegie Mellon requests users of this software to return to 59 * 60 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 61 * School of Computer Science 62 * Carnegie Mellon University 63 * Pittsburgh PA 15213-3890 64 * 65 * any improvements or extensions that they make and grant Carnegie the 66 * rights to redistribute these changes. 67 */ 68 69#include "opt_uvmhist.h" 70 71/* 72 * uvm_km.c: handle kernel memory allocation and management 73 */ 74 75/* 76 * overview of kernel memory management: 77 * 78 * the kernel virtual address space is mapped by "kernel_map." kernel_map 79 * starts at VM_MIN_KERNEL_ADDRESS and goes to VM_MAX_KERNEL_ADDRESS. 80 * note that VM_MIN_KERNEL_ADDRESS is equal to vm_map_min(kernel_map). 81 * 82 * the kernel_map has several "submaps." submaps can only appear in 83 * the kernel_map (user processes can't use them). submaps "take over" 84 * the management of a sub-range of the kernel's address space. submaps 85 * are typically allocated at boot time and are never released. kernel 86 * virtual address space that is mapped by a submap is locked by the 87 * submap's lock -- not the kernel_map's lock. 88 * 89 * thus, the useful feature of submaps is that they allow us to break 90 * up the locking and protection of the kernel address space into smaller 91 * chunks. 92 * 93 * the vm system has several standard kernel submaps, including: 94 * kmem_map => contains only wired kernel memory for the kernel 95 * malloc. *** access to kmem_map must be protected 96 * by splvm() because we are allowed to call malloc() 97 * at interrupt time *** 98 * mb_map => memory for large mbufs, *** protected by splvm *** 99 * pager_map => used to map "buf" structures into kernel space 100 * exec_map => used during exec to handle exec args 101 * etc... 102 * 103 * the kernel allocates its private memory out of special uvm_objects whose 104 * reference count is set to UVM_OBJ_KERN (thus indicating that the objects 105 * are "special" and never die). all kernel objects should be thought of 106 * as large, fixed-sized, sparsely populated uvm_objects. each kernel 107 * object is equal to the size of kernel virtual address space (i.e. the 108 * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS"). 109 * 110 * most kernel private memory lives in kernel_object. the only exception 111 * to this is for memory that belongs to submaps that must be protected 112 * by splvm(). each of these submaps has their own private kernel 113 * object (e.g. kmem_object, mb_object). 114 * 115 * note that just because a kernel object spans the entire kernel virutal 116 * address space doesn't mean that it has to be mapped into the entire space. 117 * large chunks of a kernel object's space go unused either because 118 * that area of kernel VM is unmapped, or there is some other type of 119 * object mapped into that range (e.g. a vnode). for submap's kernel 120 * objects, the only part of the object that can ever be populated is the 121 * offsets that are managed by the submap. 122 * 123 * note that the "offset" in a kernel object is always the kernel virtual 124 * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)). 125 * example: 126 * suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a 127 * uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the 128 * kernel map]. if uvm_km_alloc returns virtual address 0xf8235000, 129 * then that means that the page at offset 0x235000 in kernel_object is 130 * mapped at 0xf8235000. 131 * 132 * note that the offsets in kmem_object and mb_object also follow this 133 * rule. this means that the offsets for kmem_object must fall in the 134 * range of [vm_map_min(kmem_object) - vm_map_min(kernel_map)] to 135 * [vm_map_max(kmem_object) - vm_map_min(kernel_map)], so the offsets 136 * in those objects will typically not start at zero. 137 * 138 * kernel object have one other special property: when the kernel virtual 139 * memory mapping them is unmapped, the backing memory in the object is 140 * freed right away. this is done with the uvm_km_pgremove() function. 141 * this has to be done because there is no backing store for kernel pages 142 * and no need to save them after they are no longer referenced. 143 */ 144 145#include <sys/param.h> 146#include <sys/systm.h> 147#include <sys/proc.h> 148 149#include <uvm/uvm.h> 150 151/* 152 * global data structures 153 */ 154 155vm_map_t kernel_map = NULL; 156 157struct vmi_list vmi_list; 158struct simplelock vmi_list_slock; 159 160/* 161 * local data structues 162 */ 163 164static struct vm_map kernel_map_store; 165static struct uvm_object kmem_object_store; 166static struct uvm_object mb_object_store; 167 168/* 169 * All pager operations here are NULL, but the object must have 170 * a pager ops vector associated with it; various places assume 171 * it to be so. 172 */ 173static struct uvm_pagerops km_pager; 174 175/* 176 * uvm_km_init: init kernel maps and objects to reflect reality (i.e. 177 * KVM already allocated for text, data, bss, and static data structures). 178 * 179 * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS. 180 * we assume that [min -> start] has already been allocated and that 181 * "end" is the end. 182 */ 183 184void 185uvm_km_init(start, end) 186 vaddr_t start, end; 187{ 188 vaddr_t base = VM_MIN_KERNEL_ADDRESS; 189 190 /* 191 * first, initialize the interrupt-safe map list. 192 */ 193 LIST_INIT(&vmi_list); 194 simple_lock_init(&vmi_list_slock); 195 196 /* 197 * next, init kernel memory objects. 198 */ 199 200 /* kernel_object: for pageable anonymous kernel memory */ 201 uao_init(); 202 uvm.kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS - 203 VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ); 204 205 /* 206 * kmem_object: for use by the kernel malloc(). Memory is always 207 * wired, and this object (and the kmem_map) can be accessed at 208 * interrupt time. 209 */ 210 simple_lock_init(&kmem_object_store.vmobjlock); 211 kmem_object_store.pgops = &km_pager; 212 TAILQ_INIT(&kmem_object_store.memq); 213 kmem_object_store.uo_npages = 0; 214 /* we are special. we never die */ 215 kmem_object_store.uo_refs = UVM_OBJ_KERN_INTRSAFE; 216 uvmexp.kmem_object = &kmem_object_store; 217 218 /* 219 * mb_object: for mbuf cluster pages on platforms which use the 220 * mb_map. Memory is always wired, and this object (and the mb_map) 221 * can be accessed at interrupt time. 222 */ 223 simple_lock_init(&mb_object_store.vmobjlock); 224 mb_object_store.pgops = &km_pager; 225 TAILQ_INIT(&mb_object_store.memq); 226 mb_object_store.uo_npages = 0; 227 /* we are special. we never die */ 228 mb_object_store.uo_refs = UVM_OBJ_KERN_INTRSAFE; 229 uvmexp.mb_object = &mb_object_store; 230 231 /* 232 * init the map and reserve allready allocated kernel space 233 * before installing. 234 */ 235 236 uvm_map_setup(&kernel_map_store, base, end, VM_MAP_PAGEABLE); 237 kernel_map_store.pmap = pmap_kernel(); 238 if (uvm_map(&kernel_map_store, &base, start - base, NULL, 239 UVM_UNKNOWN_OFFSET, 0, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, 240 UVM_INH_NONE, UVM_ADV_RANDOM,UVM_FLAG_FIXED)) != 0) 241 panic("uvm_km_init: could not reserve space for kernel"); 242 243 /* 244 * install! 245 */ 246 247 kernel_map = &kernel_map_store; 248} 249 250/* 251 * uvm_km_suballoc: allocate a submap in the kernel map. once a submap 252 * is allocated all references to that area of VM must go through it. this 253 * allows the locking of VAs in kernel_map to be broken up into regions. 254 * 255 * => if `fixed' is true, *min specifies where the region described 256 * by the submap must start 257 * => if submap is non NULL we use that as the submap, otherwise we 258 * alloc a new map 259 */ 260struct vm_map * 261uvm_km_suballoc(map, min, max, size, flags, fixed, submap) 262 struct vm_map *map; 263 vaddr_t *min, *max; /* OUT, OUT */ 264 vsize_t size; 265 int flags; 266 boolean_t fixed; 267 struct vm_map *submap; 268{ 269 int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0); 270 271 size = round_page(size); /* round up to pagesize */ 272 273 /* 274 * first allocate a blank spot in the parent map 275 */ 276 277 if (uvm_map(map, min, size, NULL, UVM_UNKNOWN_OFFSET, 0, 278 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 279 UVM_ADV_RANDOM, mapflags)) != 0) { 280 panic("uvm_km_suballoc: unable to allocate space in parent map"); 281 } 282 283 /* 284 * set VM bounds (min is filled in by uvm_map) 285 */ 286 287 *max = *min + size; 288 289 /* 290 * add references to pmap and create or init the submap 291 */ 292 293 pmap_reference(vm_map_pmap(map)); 294 if (submap == NULL) { 295 submap = uvm_map_create(vm_map_pmap(map), *min, *max, flags); 296 if (submap == NULL) 297 panic("uvm_km_suballoc: unable to create submap"); 298 } else { 299 uvm_map_setup(submap, *min, *max, flags); 300 submap->pmap = vm_map_pmap(map); 301 } 302 303 /* 304 * now let uvm_map_submap plug in it... 305 */ 306 307 if (uvm_map_submap(map, *min, *max, submap) != 0) 308 panic("uvm_km_suballoc: submap allocation failed"); 309 310 return(submap); 311} 312 313/* 314 * uvm_km_pgremove: remove pages from a kernel uvm_object. 315 * 316 * => when you unmap a part of anonymous kernel memory you want to toss 317 * the pages right away. (this gets called from uvm_unmap_...). 318 */ 319 320#define UKM_HASH_PENALTY 4 /* a guess */ 321 322void 323uvm_km_pgremove(uobj, start, end) 324 struct uvm_object *uobj; 325 vaddr_t start, end; 326{ 327 boolean_t by_list; 328 struct vm_page *pp, *ppnext; 329 vaddr_t curoff; 330 UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist); 331 332 KASSERT(uobj->pgops == &aobj_pager); 333 simple_lock(&uobj->vmobjlock); 334 335 /* choose cheapest traversal */ 336 by_list = (uobj->uo_npages <= 337 ((end - start) >> PAGE_SHIFT) * UKM_HASH_PENALTY); 338 339 if (by_list) 340 goto loop_by_list; 341 342 /* by hash */ 343 344 for (curoff = start ; curoff < end ; curoff += PAGE_SIZE) { 345 pp = uvm_pagelookup(uobj, curoff); 346 if (pp == NULL) 347 continue; 348 349 UVMHIST_LOG(maphist," page 0x%x, busy=%d", pp, 350 pp->flags & PG_BUSY, 0, 0); 351 352 /* now do the actual work */ 353 if (pp->flags & PG_BUSY) { 354 /* owner must check for this when done */ 355 pp->flags |= PG_RELEASED; 356 } else { 357 /* free the swap slot... */ 358 uao_dropswap(uobj, curoff >> PAGE_SHIFT); 359 360 /* 361 * ...and free the page; note it may be on the 362 * active or inactive queues. 363 */ 364 uvm_lock_pageq(); 365 uvm_pagefree(pp); 366 uvm_unlock_pageq(); 367 } 368 } 369 simple_unlock(&uobj->vmobjlock); 370 return; 371 372loop_by_list: 373 374 for (pp = TAILQ_FIRST(&uobj->memq); pp != NULL; pp = ppnext) { 375 ppnext = TAILQ_NEXT(pp, listq); 376 if (pp->offset < start || pp->offset >= end) { 377 continue; 378 } 379 380 UVMHIST_LOG(maphist," page 0x%x, busy=%d", pp, 381 pp->flags & PG_BUSY, 0, 0); 382 383 if (pp->flags & PG_BUSY) { 384 /* owner must check for this when done */ 385 pp->flags |= PG_RELEASED; 386 } else { 387 /* free the swap slot... */ 388 uao_dropswap(uobj, pp->offset >> PAGE_SHIFT); 389 390 /* 391 * ...and free the page; note it may be on the 392 * active or inactive queues. 393 */ 394 uvm_lock_pageq(); 395 uvm_pagefree(pp); 396 uvm_unlock_pageq(); 397 } 398 } 399 simple_unlock(&uobj->vmobjlock); 400} 401 402 403/* 404 * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for "intrsafe" 405 * objects 406 * 407 * => when you unmap a part of anonymous kernel memory you want to toss 408 * the pages right away. (this gets called from uvm_unmap_...). 409 * => none of the pages will ever be busy, and none of them will ever 410 * be on the active or inactive queues (because these objects are 411 * never allowed to "page"). 412 */ 413 414void 415uvm_km_pgremove_intrsafe(uobj, start, end) 416 struct uvm_object *uobj; 417 vaddr_t start, end; 418{ 419 boolean_t by_list; 420 struct vm_page *pp, *ppnext; 421 vaddr_t curoff; 422 UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist); 423 424 KASSERT(UVM_OBJ_IS_INTRSAFE_OBJECT(uobj)); 425 simple_lock(&uobj->vmobjlock); /* lock object */ 426 427 /* choose cheapest traversal */ 428 by_list = (uobj->uo_npages <= 429 ((end - start) >> PAGE_SHIFT) * UKM_HASH_PENALTY); 430 431 if (by_list) 432 goto loop_by_list; 433 434 /* by hash */ 435 436 for (curoff = start ; curoff < end ; curoff += PAGE_SIZE) { 437 pp = uvm_pagelookup(uobj, curoff); 438 if (pp == NULL) { 439 continue; 440 } 441 442 UVMHIST_LOG(maphist," page 0x%x, busy=%d", pp, 443 pp->flags & PG_BUSY, 0, 0); 444 KASSERT((pp->flags & PG_BUSY) == 0); 445 KASSERT((pp->pqflags & PQ_ACTIVE) == 0); 446 KASSERT((pp->pqflags & PQ_INACTIVE) == 0); 447 uvm_pagefree(pp); 448 } 449 simple_unlock(&uobj->vmobjlock); 450 return; 451 452loop_by_list: 453 454 for (pp = TAILQ_FIRST(&uobj->memq); pp != NULL; pp = ppnext) { 455 ppnext = TAILQ_NEXT(pp, listq); 456 if (pp->offset < start || pp->offset >= end) { 457 continue; 458 } 459 460 UVMHIST_LOG(maphist," page 0x%x, busy=%d", pp, 461 pp->flags & PG_BUSY, 0, 0); 462 KASSERT((pp->flags & PG_BUSY) == 0); 463 KASSERT((pp->pqflags & PQ_ACTIVE) == 0); 464 KASSERT((pp->pqflags & PQ_INACTIVE) == 0); 465 uvm_pagefree(pp); 466 } 467 simple_unlock(&uobj->vmobjlock); 468} 469 470 471/* 472 * uvm_km_kmemalloc: lower level kernel memory allocator for malloc() 473 * 474 * => we map wired memory into the specified map using the obj passed in 475 * => NOTE: we can return NULL even if we can wait if there is not enough 476 * free VM space in the map... caller should be prepared to handle 477 * this case. 478 * => we return KVA of memory allocated 479 * => flags: NOWAIT, VALLOC - just allocate VA, TRYLOCK - fail if we can't 480 * lock the map 481 */ 482 483vaddr_t 484uvm_km_kmemalloc(map, obj, size, flags) 485 vm_map_t map; 486 struct uvm_object *obj; 487 vsize_t size; 488 int flags; 489{ 490 vaddr_t kva, loopva; 491 vaddr_t offset; 492 vsize_t loopsize; 493 struct vm_page *pg; 494 UVMHIST_FUNC("uvm_km_kmemalloc"); UVMHIST_CALLED(maphist); 495 496 UVMHIST_LOG(maphist," (map=0x%x, obj=0x%x, size=0x%x, flags=%d)", 497 map, obj, size, flags); 498 KASSERT(vm_map_pmap(map) == pmap_kernel()); 499 500 /* 501 * setup for call 502 */ 503 504 size = round_page(size); 505 kva = vm_map_min(map); /* hint */ 506 507 /* 508 * allocate some virtual space 509 */ 510 511 if (__predict_false(uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET, 512 0, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 513 UVM_ADV_RANDOM, (flags & UVM_KMF_TRYLOCK))) 514 != 0)) { 515 UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0); 516 return(0); 517 } 518 519 /* 520 * if all we wanted was VA, return now 521 */ 522 523 if (flags & UVM_KMF_VALLOC) { 524 UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0); 525 return(kva); 526 } 527 528 /* 529 * recover object offset from virtual address 530 */ 531 532 offset = kva - vm_map_min(kernel_map); 533 UVMHIST_LOG(maphist, " kva=0x%x, offset=0x%x", kva, offset,0,0); 534 535 /* 536 * now allocate and map in the memory... note that we are the only ones 537 * whom should ever get a handle on this area of VM. 538 */ 539 540 loopva = kva; 541 loopsize = size; 542 while (loopsize) { 543 simple_lock(&obj->vmobjlock); 544 pg = uvm_pagealloc(obj, offset, NULL, 0); 545 if (__predict_true(pg != NULL)) { 546 pg->flags &= ~PG_BUSY; /* new page */ 547 UVM_PAGE_OWN(pg, NULL); 548 } 549 simple_unlock(&obj->vmobjlock); 550 551 /* 552 * out of memory? 553 */ 554 555 if (__predict_false(pg == NULL)) { 556 if (flags & UVM_KMF_NOWAIT) { 557 /* free everything! */ 558 uvm_unmap(map, kva, kva + size); 559 return(0); 560 } else { 561 uvm_wait("km_getwait2"); /* sleep here */ 562 continue; 563 } 564 } 565 566 /* 567 * map it in: note that we call pmap_enter with the map and 568 * object unlocked in case we are kmem_map/kmem_object 569 * (because if pmap_enter wants to allocate out of kmem_object 570 * it will need to lock it itself!) 571 */ 572 573 if (UVM_OBJ_IS_INTRSAFE_OBJECT(obj)) { 574 pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg), 575 VM_PROT_ALL); 576 } else { 577 pmap_enter(map->pmap, loopva, VM_PAGE_TO_PHYS(pg), 578 UVM_PROT_ALL, 579 PMAP_WIRED | VM_PROT_READ | VM_PROT_WRITE); 580 } 581 loopva += PAGE_SIZE; 582 offset += PAGE_SIZE; 583 loopsize -= PAGE_SIZE; 584 } 585 pmap_update(); 586 UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0); 587 return(kva); 588} 589 590/* 591 * uvm_km_free: free an area of kernel memory 592 */ 593 594void 595uvm_km_free(map, addr, size) 596 vm_map_t map; 597 vaddr_t addr; 598 vsize_t size; 599{ 600 uvm_unmap(map, trunc_page(addr), round_page(addr+size)); 601} 602 603/* 604 * uvm_km_free_wakeup: free an area of kernel memory and wake up 605 * anyone waiting for vm space. 606 * 607 * => XXX: "wanted" bit + unlock&wait on other end? 608 */ 609 610void 611uvm_km_free_wakeup(map, addr, size) 612 vm_map_t map; 613 vaddr_t addr; 614 vsize_t size; 615{ 616 vm_map_entry_t dead_entries; 617 618 vm_map_lock(map); 619 uvm_unmap_remove(map, trunc_page(addr), round_page(addr + size), 620 &dead_entries); 621 wakeup(map); 622 vm_map_unlock(map); 623 if (dead_entries != NULL) 624 uvm_unmap_detach(dead_entries, 0); 625} 626 627/* 628 * uvm_km_alloc1: allocate wired down memory in the kernel map. 629 * 630 * => we can sleep if needed 631 */ 632 633vaddr_t 634uvm_km_alloc1(map, size, zeroit) 635 vm_map_t map; 636 vsize_t size; 637 boolean_t zeroit; 638{ 639 vaddr_t kva, loopva, offset; 640 struct vm_page *pg; 641 UVMHIST_FUNC("uvm_km_alloc1"); UVMHIST_CALLED(maphist); 642 643 UVMHIST_LOG(maphist,"(map=0x%x, size=0x%x)", map, size,0,0); 644 KASSERT(vm_map_pmap(map) == pmap_kernel()); 645 646 size = round_page(size); 647 kva = vm_map_min(map); /* hint */ 648 649 /* 650 * allocate some virtual space 651 */ 652 653 if (__predict_false(uvm_map(map, &kva, size, uvm.kernel_object, 654 UVM_UNKNOWN_OFFSET, 0, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, 655 UVM_INH_NONE, UVM_ADV_RANDOM, 656 0)) != 0)) { 657 UVMHIST_LOG(maphist,"<- done (no VM)",0,0,0,0); 658 return(0); 659 } 660 661 /* 662 * recover object offset from virtual address 663 */ 664 665 offset = kva - vm_map_min(kernel_map); 666 UVMHIST_LOG(maphist," kva=0x%x, offset=0x%x", kva, offset,0,0); 667 668 /* 669 * now allocate the memory. we must be careful about released pages. 670 */ 671 672 loopva = kva; 673 while (size) { 674 simple_lock(&uvm.kernel_object->vmobjlock); 675 pg = uvm_pagelookup(uvm.kernel_object, offset); 676 677 /* 678 * if we found a page in an unallocated region, it must be 679 * released 680 */ 681 if (pg) { 682 if ((pg->flags & PG_RELEASED) == 0) 683 panic("uvm_km_alloc1: non-released page"); 684 pg->flags |= PG_WANTED; 685 UVM_UNLOCK_AND_WAIT(pg, &uvm.kernel_object->vmobjlock, 686 FALSE, "km_alloc", 0); 687 continue; /* retry */ 688 } 689 690 /* allocate ram */ 691 pg = uvm_pagealloc(uvm.kernel_object, offset, NULL, 0); 692 if (pg) { 693 pg->flags &= ~PG_BUSY; /* new page */ 694 UVM_PAGE_OWN(pg, NULL); 695 } 696 simple_unlock(&uvm.kernel_object->vmobjlock); 697 if (__predict_false(pg == NULL)) { 698 uvm_wait("km_alloc1w"); /* wait for memory */ 699 continue; 700 } 701 702 /* 703 * map it in; note we're never called with an intrsafe 704 * object, so we always use regular old pmap_enter(). 705 */ 706 pmap_enter(map->pmap, loopva, VM_PAGE_TO_PHYS(pg), 707 UVM_PROT_ALL, PMAP_WIRED | VM_PROT_READ | VM_PROT_WRITE); 708 709 loopva += PAGE_SIZE; 710 offset += PAGE_SIZE; 711 size -= PAGE_SIZE; 712 } 713 714 pmap_update(); 715 716 /* 717 * zero on request (note that "size" is now zero due to the above loop 718 * so we need to subtract kva from loopva to reconstruct the size). 719 */ 720 721 if (zeroit) 722 memset((caddr_t)kva, 0, loopva - kva); 723 724 UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0); 725 return(kva); 726} 727 728/* 729 * uvm_km_valloc: allocate zero-fill memory in the kernel's address space 730 * 731 * => memory is not allocated until fault time 732 */ 733 734vaddr_t 735uvm_km_valloc(map, size) 736 vm_map_t map; 737 vsize_t size; 738{ 739 return(uvm_km_valloc_align(map, size, 0)); 740} 741 742vaddr_t 743uvm_km_valloc_align(map, size, align) 744 vm_map_t map; 745 vsize_t size; 746 vsize_t align; 747{ 748 vaddr_t kva; 749 UVMHIST_FUNC("uvm_km_valloc"); UVMHIST_CALLED(maphist); 750 751 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x)", map, size, 0,0); 752 KASSERT(vm_map_pmap(map) == pmap_kernel()); 753 754 size = round_page(size); 755 kva = vm_map_min(map); /* hint */ 756 757 /* 758 * allocate some virtual space. will be demand filled by kernel_object. 759 */ 760 761 if (__predict_false(uvm_map(map, &kva, size, uvm.kernel_object, 762 UVM_UNKNOWN_OFFSET, align, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, 763 UVM_INH_NONE, UVM_ADV_RANDOM, 764 0)) != 0)) { 765 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 766 return(0); 767 } 768 769 UVMHIST_LOG(maphist, "<- done (kva=0x%x)", kva,0,0,0); 770 return(kva); 771} 772 773/* 774 * uvm_km_valloc_wait: allocate zero-fill memory in the kernel's address space 775 * 776 * => memory is not allocated until fault time 777 * => if no room in map, wait for space to free, unless requested size 778 * is larger than map (in which case we return 0) 779 */ 780 781vaddr_t 782uvm_km_valloc_prefer_wait(map, size, prefer) 783 vm_map_t map; 784 vsize_t size; 785 voff_t prefer; 786{ 787 vaddr_t kva; 788 UVMHIST_FUNC("uvm_km_valloc_prefer_wait"); UVMHIST_CALLED(maphist); 789 790 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x)", map, size, 0,0); 791 KASSERT(vm_map_pmap(map) == pmap_kernel()); 792 793 size = round_page(size); 794 if (size > vm_map_max(map) - vm_map_min(map)) 795 return(0); 796 797 while (1) { 798 kva = vm_map_min(map); /* hint */ 799 800 /* 801 * allocate some virtual space. will be demand filled 802 * by kernel_object. 803 */ 804 805 if (__predict_true(uvm_map(map, &kva, size, uvm.kernel_object, 806 prefer, 0, UVM_MAPFLAG(UVM_PROT_ALL, 807 UVM_PROT_ALL, UVM_INH_NONE, UVM_ADV_RANDOM, 0)) 808 == 0)) { 809 UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0); 810 return(kva); 811 } 812 813 /* 814 * failed. sleep for a while (on map) 815 */ 816 817 UVMHIST_LOG(maphist,"<<<sleeping>>>",0,0,0,0); 818 tsleep((caddr_t)map, PVM, "vallocwait", 0); 819 } 820 /*NOTREACHED*/ 821} 822 823vaddr_t 824uvm_km_valloc_wait(map, size) 825 vm_map_t map; 826 vsize_t size; 827{ 828 return uvm_km_valloc_prefer_wait(map, size, UVM_UNKNOWN_OFFSET); 829} 830 831/* Sanity; must specify both or none. */ 832#if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \ 833 (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE)) 834#error Must specify MAP and UNMAP together. 835#endif 836 837/* 838 * uvm_km_alloc_poolpage: allocate a page for the pool allocator 839 * 840 * => if the pmap specifies an alternate mapping method, we use it. 841 */ 842 843/* ARGSUSED */ 844vaddr_t 845uvm_km_alloc_poolpage1(map, obj, waitok) 846 vm_map_t map; 847 struct uvm_object *obj; 848 boolean_t waitok; 849{ 850#if defined(PMAP_MAP_POOLPAGE) 851 struct vm_page *pg; 852 vaddr_t va; 853 854 again: 855 pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE); 856 if (__predict_false(pg == NULL)) { 857 if (waitok) { 858 uvm_wait("plpg"); 859 goto again; 860 } else 861 return (0); 862 } 863 va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg)); 864 if (__predict_false(va == 0)) 865 uvm_pagefree(pg); 866 return (va); 867#else 868 vaddr_t va; 869 int s; 870 871 /* 872 * NOTE: We may be called with a map that doens't require splvm 873 * protection (e.g. kernel_map). However, it does not hurt to 874 * go to splvm in this case (since unprocted maps will never be 875 * accessed in interrupt context). 876 * 877 * XXX We may want to consider changing the interface to this 878 * XXX function. 879 */ 880 881 s = splvm(); 882 va = uvm_km_kmemalloc(map, obj, PAGE_SIZE, waitok ? 0 : UVM_KMF_NOWAIT); 883 splx(s); 884 return (va); 885#endif /* PMAP_MAP_POOLPAGE */ 886} 887 888/* 889 * uvm_km_free_poolpage: free a previously allocated pool page 890 * 891 * => if the pmap specifies an alternate unmapping method, we use it. 892 */ 893 894/* ARGSUSED */ 895void 896uvm_km_free_poolpage1(map, addr) 897 vm_map_t map; 898 vaddr_t addr; 899{ 900#if defined(PMAP_UNMAP_POOLPAGE) 901 paddr_t pa; 902 903 pa = PMAP_UNMAP_POOLPAGE(addr); 904 uvm_pagefree(PHYS_TO_VM_PAGE(pa)); 905#else 906 int s; 907 908 /* 909 * NOTE: We may be called with a map that doens't require splvm 910 * protection (e.g. kernel_map). However, it does not hurt to 911 * go to splvm in this case (since unprocted maps will never be 912 * accessed in interrupt context). 913 * 914 * XXX We may want to consider changing the interface to this 915 * XXX function. 916 */ 917 918 s = splvm(); 919 uvm_km_free(map, addr, PAGE_SIZE); 920 splx(s); 921#endif /* PMAP_UNMAP_POOLPAGE */ 922} 923