1 /* $NetBSD: uvm_km.c,v 1.76 2005/01/13 11:50:32 yamt 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 /* 70 * uvm_km.c: handle kernel memory allocation and management 71 */ 72 73 /* 74 * overview of kernel memory management: 75 * 76 * the kernel virtual address space is mapped by "kernel_map." kernel_map 77 * starts at VM_MIN_KERNEL_ADDRESS and goes to VM_MAX_KERNEL_ADDRESS. 78 * note that VM_MIN_KERNEL_ADDRESS is equal to vm_map_min(kernel_map). 79 * 80 * the kernel_map has several "submaps." submaps can only appear in 81 * the kernel_map (user processes can't use them). submaps "take over" 82 * the management of a sub-range of the kernel's address space. submaps 83 * are typically allocated at boot time and are never released. kernel 84 * virtual address space that is mapped by a submap is locked by the 85 * submap's lock -- not the kernel_map's lock. 86 * 87 * thus, the useful feature of submaps is that they allow us to break 88 * up the locking and protection of the kernel address space into smaller 89 * chunks. 90 * 91 * the vm system has several standard kernel submaps, including: 92 * kmem_map => contains only wired kernel memory for the kernel 93 * malloc. *** access to kmem_map must be protected 94 * by splvm() because we are allowed to call malloc() 95 * at interrupt time *** 96 * mb_map => memory for large mbufs, *** protected by splvm *** 97 * pager_map => used to map "buf" structures into kernel space 98 * exec_map => used during exec to handle exec args 99 * etc... 100 * 101 * the kernel allocates its private memory out of special uvm_objects whose 102 * reference count is set to UVM_OBJ_KERN (thus indicating that the objects 103 * are "special" and never die). all kernel objects should be thought of 104 * as large, fixed-sized, sparsely populated uvm_objects. each kernel 105 * object is equal to the size of kernel virtual address space (i.e. the 106 * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS"). 107 * 108 * most kernel private memory lives in kernel_object. the only exception 109 * to this is for memory that belongs to submaps that must be protected 110 * by splvm(). pages in these submaps are not assigned to an object. 111 * 112 * note that just because a kernel object spans the entire kernel virutal 113 * address space doesn't mean that it has to be mapped into the entire space. 114 * large chunks of a kernel object's space go unused either because 115 * that area of kernel VM is unmapped, or there is some other type of 116 * object mapped into that range (e.g. a vnode). for submap's kernel 117 * objects, the only part of the object that can ever be populated is the 118 * offsets that are managed by the submap. 119 * 120 * note that the "offset" in a kernel object is always the kernel virtual 121 * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)). 122 * example: 123 * suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a 124 * uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the 125 * kernel map]. if uvm_km_alloc returns virtual address 0xf8235000, 126 * then that means that the page at offset 0x235000 in kernel_object is 127 * mapped at 0xf8235000. 128 * 129 * kernel object have one other special property: when the kernel virtual 130 * memory mapping them is unmapped, the backing memory in the object is 131 * freed right away. this is done with the uvm_km_pgremove() function. 132 * this has to be done because there is no backing store for kernel pages 133 * and no need to save them after they are no longer referenced. 134 */ 135 136 #include <sys/cdefs.h> 137 __KERNEL_RCSID(0, "$NetBSD: uvm_km.c,v 1.76 2005/01/13 11:50:32 yamt Exp $"); 138 139 #include "opt_uvmhist.h" 140 141 #include <sys/param.h> 142 #include <sys/malloc.h> 143 #include <sys/systm.h> 144 #include <sys/proc.h> 145 #include <sys/pool.h> 146 147 #include <uvm/uvm.h> 148 149 /* 150 * global data structures 151 */ 152 153 struct vm_map *kernel_map = NULL; 154 155 /* 156 * local data structues 157 */ 158 159 static struct vm_map_kernel kernel_map_store; 160 static struct vm_map_entry kernel_first_mapent_store; 161 162 #if !defined(PMAP_MAP_POOLPAGE) 163 164 /* 165 * kva cache 166 * 167 * XXX maybe it's better to do this at the uvm_map layer. 168 */ 169 170 #define KM_VACACHE_SIZE (32 * PAGE_SIZE) /* XXX tune */ 171 172 static void *km_vacache_alloc(struct pool *, int); 173 static void km_vacache_free(struct pool *, void *); 174 static void km_vacache_init(struct vm_map *, const char *, size_t); 175 176 /* XXX */ 177 #define KM_VACACHE_POOL_TO_MAP(pp) \ 178 ((struct vm_map *)((char *)(pp) - \ 179 offsetof(struct vm_map_kernel, vmk_vacache))) 180 181 static void * 182 km_vacache_alloc(struct pool *pp, int flags) 183 { 184 vaddr_t va; 185 size_t size; 186 struct vm_map *map; 187 #if defined(DEBUG) 188 vaddr_t loopva; 189 #endif 190 size = pp->pr_alloc->pa_pagesz; 191 192 map = KM_VACACHE_POOL_TO_MAP(pp); 193 194 va = vm_map_min(map); /* hint */ 195 if (uvm_map(map, &va, size, NULL, UVM_UNKNOWN_OFFSET, size, 196 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 197 UVM_ADV_RANDOM, UVM_FLAG_QUANTUM | 198 ((flags & PR_WAITOK) ? 0 : UVM_FLAG_TRYLOCK | UVM_FLAG_NOWAIT)))) 199 return NULL; 200 201 #if defined(DEBUG) 202 for (loopva = va; loopva < va + size; loopva += PAGE_SIZE) { 203 if (pmap_extract(pmap_kernel(), loopva, NULL)) 204 panic("km_vacache_free: has mapping"); 205 } 206 #endif 207 208 return (void *)va; 209 } 210 211 static void 212 km_vacache_free(struct pool *pp, void *v) 213 { 214 vaddr_t va = (vaddr_t)v; 215 size_t size = pp->pr_alloc->pa_pagesz; 216 struct vm_map *map; 217 #if defined(DEBUG) 218 vaddr_t loopva; 219 220 for (loopva = va; loopva < va + size; loopva += PAGE_SIZE) { 221 if (pmap_extract(pmap_kernel(), loopva, NULL)) 222 panic("km_vacache_free: has mapping"); 223 } 224 #endif 225 map = KM_VACACHE_POOL_TO_MAP(pp); 226 uvm_unmap1(map, va, va + size, UVM_FLAG_QUANTUM); 227 } 228 229 /* 230 * km_vacache_init: initialize kva cache. 231 */ 232 233 static void 234 km_vacache_init(struct vm_map *map, const char *name, size_t size) 235 { 236 struct vm_map_kernel *vmk; 237 struct pool *pp; 238 struct pool_allocator *pa; 239 240 KASSERT(VM_MAP_IS_KERNEL(map)); 241 KASSERT(size < (vm_map_max(map) - vm_map_min(map)) / 2); /* sanity */ 242 243 vmk = vm_map_to_kernel(map); 244 pp = &vmk->vmk_vacache; 245 pa = &vmk->vmk_vacache_allocator; 246 memset(pa, 0, sizeof(*pa)); 247 pa->pa_alloc = km_vacache_alloc; 248 pa->pa_free = km_vacache_free; 249 pa->pa_pagesz = (unsigned int)size; 250 pool_init(pp, PAGE_SIZE, 0, 0, PR_NOTOUCH | PR_RECURSIVE, name, pa); 251 252 /* XXX for now.. */ 253 pool_sethiwat(pp, 0); 254 } 255 256 void 257 uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size) 258 { 259 260 map->flags |= VM_MAP_VACACHE; 261 if (size == 0) 262 size = KM_VACACHE_SIZE; 263 km_vacache_init(map, name, size); 264 } 265 266 #else /* !defined(PMAP_MAP_POOLPAGE) */ 267 268 void 269 uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size) 270 { 271 272 /* nothing */ 273 } 274 275 #endif /* !defined(PMAP_MAP_POOLPAGE) */ 276 277 /* 278 * uvm_km_init: init kernel maps and objects to reflect reality (i.e. 279 * KVM already allocated for text, data, bss, and static data structures). 280 * 281 * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS. 282 * we assume that [min -> start] has already been allocated and that 283 * "end" is the end. 284 */ 285 286 void 287 uvm_km_init(start, end) 288 vaddr_t start, end; 289 { 290 vaddr_t base = VM_MIN_KERNEL_ADDRESS; 291 292 /* 293 * next, init kernel memory objects. 294 */ 295 296 /* kernel_object: for pageable anonymous kernel memory */ 297 uao_init(); 298 uvm.kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS - 299 VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ); 300 301 /* 302 * init the map and reserve any space that might already 303 * have been allocated kernel space before installing. 304 */ 305 306 uvm_map_setup_kernel(&kernel_map_store, base, end, VM_MAP_PAGEABLE); 307 kernel_map_store.vmk_map.pmap = pmap_kernel(); 308 if (start != base) { 309 int error; 310 struct uvm_map_args args; 311 312 error = uvm_map_prepare(&kernel_map_store.vmk_map, 313 base, start - base, 314 NULL, UVM_UNKNOWN_OFFSET, 0, 315 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 316 UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args); 317 if (!error) { 318 kernel_first_mapent_store.flags = 319 UVM_MAP_KERNEL | UVM_MAP_FIRST; 320 error = uvm_map_enter(&kernel_map_store.vmk_map, &args, 321 &kernel_first_mapent_store); 322 } 323 324 if (error) 325 panic( 326 "uvm_km_init: could not reserve space for kernel"); 327 } 328 329 /* 330 * install! 331 */ 332 333 kernel_map = &kernel_map_store.vmk_map; 334 uvm_km_vacache_init(kernel_map, "kvakernel", 0); 335 } 336 337 /* 338 * uvm_km_suballoc: allocate a submap in the kernel map. once a submap 339 * is allocated all references to that area of VM must go through it. this 340 * allows the locking of VAs in kernel_map to be broken up into regions. 341 * 342 * => if `fixed' is true, *min specifies where the region described 343 * by the submap must start 344 * => if submap is non NULL we use that as the submap, otherwise we 345 * alloc a new map 346 */ 347 struct vm_map * 348 uvm_km_suballoc(map, min, max, size, flags, fixed, submap) 349 struct vm_map *map; 350 vaddr_t *min, *max; /* IN/OUT, OUT */ 351 vsize_t size; 352 int flags; 353 boolean_t fixed; 354 struct vm_map_kernel *submap; 355 { 356 int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0); 357 358 KASSERT(vm_map_pmap(map) == pmap_kernel()); 359 360 size = round_page(size); /* round up to pagesize */ 361 362 /* 363 * first allocate a blank spot in the parent map 364 */ 365 366 if (uvm_map(map, min, size, NULL, UVM_UNKNOWN_OFFSET, 0, 367 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 368 UVM_ADV_RANDOM, mapflags)) != 0) { 369 panic("uvm_km_suballoc: unable to allocate space in parent map"); 370 } 371 372 /* 373 * set VM bounds (min is filled in by uvm_map) 374 */ 375 376 *max = *min + size; 377 378 /* 379 * add references to pmap and create or init the submap 380 */ 381 382 pmap_reference(vm_map_pmap(map)); 383 if (submap == NULL) { 384 submap = malloc(sizeof(*submap), M_VMMAP, M_WAITOK); 385 if (submap == NULL) 386 panic("uvm_km_suballoc: unable to create submap"); 387 } 388 uvm_map_setup_kernel(submap, *min, *max, flags); 389 submap->vmk_map.pmap = vm_map_pmap(map); 390 391 /* 392 * now let uvm_map_submap plug in it... 393 */ 394 395 if (uvm_map_submap(map, *min, *max, &submap->vmk_map) != 0) 396 panic("uvm_km_suballoc: submap allocation failed"); 397 398 return(&submap->vmk_map); 399 } 400 401 /* 402 * uvm_km_pgremove: remove pages from a kernel uvm_object. 403 * 404 * => when you unmap a part of anonymous kernel memory you want to toss 405 * the pages right away. (this gets called from uvm_unmap_...). 406 */ 407 408 void 409 uvm_km_pgremove(uobj, start, end) 410 struct uvm_object *uobj; 411 vaddr_t start, end; 412 { 413 struct vm_page *pg; 414 voff_t curoff, nextoff; 415 int swpgonlydelta = 0; 416 UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist); 417 418 KASSERT(uobj->pgops == &aobj_pager); 419 simple_lock(&uobj->vmobjlock); 420 421 for (curoff = start; curoff < end; curoff = nextoff) { 422 nextoff = curoff + PAGE_SIZE; 423 pg = uvm_pagelookup(uobj, curoff); 424 if (pg != NULL && pg->flags & PG_BUSY) { 425 pg->flags |= PG_WANTED; 426 UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0, 427 "km_pgrm", 0); 428 simple_lock(&uobj->vmobjlock); 429 nextoff = curoff; 430 continue; 431 } 432 433 /* 434 * free the swap slot, then the page. 435 */ 436 437 if (pg == NULL && 438 uao_find_swslot(uobj, curoff >> PAGE_SHIFT) > 0) { 439 swpgonlydelta++; 440 } 441 uao_dropswap(uobj, curoff >> PAGE_SHIFT); 442 if (pg != NULL) { 443 uvm_lock_pageq(); 444 uvm_pagefree(pg); 445 uvm_unlock_pageq(); 446 } 447 } 448 simple_unlock(&uobj->vmobjlock); 449 450 if (swpgonlydelta > 0) { 451 simple_lock(&uvm.swap_data_lock); 452 KASSERT(uvmexp.swpgonly >= swpgonlydelta); 453 uvmexp.swpgonly -= swpgonlydelta; 454 simple_unlock(&uvm.swap_data_lock); 455 } 456 } 457 458 459 /* 460 * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for "intrsafe" 461 * maps 462 * 463 * => when you unmap a part of anonymous kernel memory you want to toss 464 * the pages right away. (this is called from uvm_unmap_...). 465 * => none of the pages will ever be busy, and none of them will ever 466 * be on the active or inactive queues (because they have no object). 467 */ 468 469 void 470 uvm_km_pgremove_intrsafe(start, end) 471 vaddr_t start, end; 472 { 473 struct vm_page *pg; 474 paddr_t pa; 475 UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist); 476 477 for (; start < end; start += PAGE_SIZE) { 478 if (!pmap_extract(pmap_kernel(), start, &pa)) { 479 continue; 480 } 481 pg = PHYS_TO_VM_PAGE(pa); 482 KASSERT(pg); 483 KASSERT(pg->uobject == NULL && pg->uanon == NULL); 484 uvm_pagefree(pg); 485 } 486 } 487 488 489 /* 490 * uvm_km_kmemalloc: lower level kernel memory allocator for malloc() 491 * 492 * => we map wired memory into the specified map using the obj passed in 493 * => NOTE: we can return NULL even if we can wait if there is not enough 494 * free VM space in the map... caller should be prepared to handle 495 * this case. 496 * => we return KVA of memory allocated 497 * => align,prefer - passed on to uvm_map() 498 * => flags: NOWAIT, VALLOC - just allocate VA, TRYLOCK - fail if we can't 499 * lock the map 500 */ 501 502 vaddr_t 503 uvm_km_kmemalloc1(map, obj, size, align, prefer, flags) 504 struct vm_map *map; 505 struct uvm_object *obj; 506 vsize_t size; 507 vsize_t align; 508 voff_t prefer; 509 int flags; 510 { 511 vaddr_t kva, loopva; 512 vaddr_t offset; 513 vsize_t loopsize; 514 struct vm_page *pg; 515 UVMHIST_FUNC("uvm_km_kmemalloc"); UVMHIST_CALLED(maphist); 516 517 UVMHIST_LOG(maphist," (map=0x%x, obj=0x%x, size=0x%x, flags=%d)", 518 map, obj, size, flags); 519 KASSERT(vm_map_pmap(map) == pmap_kernel()); 520 521 /* 522 * setup for call 523 */ 524 525 size = round_page(size); 526 kva = vm_map_min(map); /* hint */ 527 528 /* 529 * allocate some virtual space 530 */ 531 532 if (__predict_false(uvm_map(map, &kva, size, obj, prefer, align, 533 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 534 UVM_ADV_RANDOM, 535 (flags & (UVM_KMF_TRYLOCK | UVM_KMF_NOWAIT)) 536 | UVM_FLAG_QUANTUM)) 537 != 0)) { 538 UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0); 539 return(0); 540 } 541 542 /* 543 * if all we wanted was VA, return now 544 */ 545 546 if (flags & UVM_KMF_VALLOC) { 547 UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0); 548 return(kva); 549 } 550 551 /* 552 * recover object offset from virtual address 553 */ 554 555 offset = kva - vm_map_min(kernel_map); 556 UVMHIST_LOG(maphist, " kva=0x%x, offset=0x%x", kva, offset,0,0); 557 558 /* 559 * now allocate and map in the memory... note that we are the only ones 560 * whom should ever get a handle on this area of VM. 561 */ 562 563 loopva = kva; 564 loopsize = size; 565 while (loopsize) { 566 if (obj) { 567 simple_lock(&obj->vmobjlock); 568 } 569 pg = uvm_pagealloc(obj, offset, NULL, UVM_PGA_USERESERVE); 570 if (__predict_true(pg != NULL)) { 571 pg->flags &= ~PG_BUSY; /* new page */ 572 UVM_PAGE_OWN(pg, NULL); 573 } 574 if (obj) { 575 simple_unlock(&obj->vmobjlock); 576 } 577 578 /* 579 * out of memory? 580 */ 581 582 if (__predict_false(pg == NULL)) { 583 if ((flags & UVM_KMF_NOWAIT) || 584 ((flags & UVM_KMF_CANFAIL) && uvm_swapisfull())) { 585 /* free everything! */ 586 uvm_unmap1(map, kva, kva + size, 587 UVM_FLAG_QUANTUM); 588 return (0); 589 } else { 590 uvm_wait("km_getwait2"); /* sleep here */ 591 continue; 592 } 593 } 594 595 /* 596 * map it in 597 */ 598 599 if (obj == NULL) { 600 pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg), 601 VM_PROT_READ | VM_PROT_WRITE); 602 } else { 603 pmap_enter(map->pmap, loopva, VM_PAGE_TO_PHYS(pg), 604 UVM_PROT_ALL, 605 PMAP_WIRED | VM_PROT_READ | VM_PROT_WRITE); 606 } 607 loopva += PAGE_SIZE; 608 offset += PAGE_SIZE; 609 loopsize -= PAGE_SIZE; 610 } 611 612 pmap_update(pmap_kernel()); 613 614 UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0); 615 return(kva); 616 } 617 618 /* 619 * uvm_km_free: free an area of kernel memory 620 */ 621 622 void 623 uvm_km_free(map, addr, size) 624 struct vm_map *map; 625 vaddr_t addr; 626 vsize_t size; 627 { 628 uvm_unmap1(map, trunc_page(addr), round_page(addr+size), 629 UVM_FLAG_QUANTUM); 630 } 631 632 /* 633 * uvm_km_alloc1: allocate wired down memory in the kernel map. 634 * 635 * => we can sleep if needed 636 */ 637 638 vaddr_t 639 uvm_km_alloc1(map, size, zeroit) 640 struct vm_map *map; 641 vsize_t size; 642 boolean_t zeroit; 643 { 644 vaddr_t kva, loopva, offset; 645 struct vm_page *pg; 646 UVMHIST_FUNC("uvm_km_alloc1"); UVMHIST_CALLED(maphist); 647 648 UVMHIST_LOG(maphist,"(map=0x%x, size=0x%x)", map, size,0,0); 649 KASSERT(vm_map_pmap(map) == pmap_kernel()); 650 651 size = round_page(size); 652 kva = vm_map_min(map); /* hint */ 653 654 /* 655 * allocate some virtual space 656 */ 657 658 if (__predict_false(uvm_map(map, &kva, size, uvm.kernel_object, 659 UVM_UNKNOWN_OFFSET, 0, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, 660 UVM_INH_NONE, UVM_ADV_RANDOM, 661 UVM_FLAG_QUANTUM)) != 0)) { 662 UVMHIST_LOG(maphist,"<- done (no VM)",0,0,0,0); 663 return(0); 664 } 665 666 /* 667 * recover object offset from virtual address 668 */ 669 670 offset = kva - vm_map_min(kernel_map); 671 UVMHIST_LOG(maphist," kva=0x%x, offset=0x%x", kva, offset,0,0); 672 673 /* 674 * now allocate the memory. 675 */ 676 677 loopva = kva; 678 while (size) { 679 simple_lock(&uvm.kernel_object->vmobjlock); 680 KASSERT(uvm_pagelookup(uvm.kernel_object, offset) == NULL); 681 pg = uvm_pagealloc(uvm.kernel_object, offset, NULL, 0); 682 if (pg) { 683 pg->flags &= ~PG_BUSY; 684 UVM_PAGE_OWN(pg, NULL); 685 } 686 simple_unlock(&uvm.kernel_object->vmobjlock); 687 if (pg == NULL) { 688 uvm_wait("km_alloc1w"); 689 continue; 690 } 691 pmap_enter(map->pmap, loopva, VM_PAGE_TO_PHYS(pg), 692 UVM_PROT_ALL, PMAP_WIRED | VM_PROT_READ | VM_PROT_WRITE); 693 loopva += PAGE_SIZE; 694 offset += PAGE_SIZE; 695 size -= PAGE_SIZE; 696 } 697 pmap_update(map->pmap); 698 699 /* 700 * zero on request (note that "size" is now zero due to the above loop 701 * so we need to subtract kva from loopva to reconstruct the size). 702 */ 703 704 if (zeroit) 705 memset((caddr_t)kva, 0, loopva - kva); 706 UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0); 707 return(kva); 708 } 709 710 /* 711 * uvm_km_valloc1: allocate zero-fill memory in the kernel's address space 712 * 713 * => memory is not allocated until fault time 714 * => the align, prefer and flags parameters are passed on to uvm_map(). 715 * 716 * Note: this function is also the backend for these macros: 717 * uvm_km_valloc 718 * uvm_km_valloc_wait 719 * uvm_km_valloc_prefer 720 * uvm_km_valloc_prefer_wait 721 * uvm_km_valloc_align 722 */ 723 724 vaddr_t 725 uvm_km_valloc1(map, size, align, prefer, flags) 726 struct vm_map *map; 727 vsize_t size; 728 vsize_t align; 729 voff_t prefer; 730 uvm_flag_t flags; 731 { 732 vaddr_t kva; 733 int error; 734 UVMHIST_FUNC("uvm_km_valloc1"); UVMHIST_CALLED(maphist); 735 736 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, align=0x%x, prefer=0x%x)", 737 map, size, align, prefer); 738 739 KASSERT(vm_map_pmap(map) == pmap_kernel()); 740 741 size = round_page(size); 742 /* 743 * Check if requested size is larger than the map, in which 744 * case we can't succeed. 745 */ 746 if (size > vm_map_max(map) - vm_map_min(map)) 747 return (0); 748 749 flags |= UVM_FLAG_QUANTUM; 750 if ((flags & UVM_KMF_NOWAIT) == 0) /* XXX */ 751 flags |= UVM_FLAG_WAITVA; /* XXX */ 752 753 kva = vm_map_min(map); /* hint */ 754 755 /* 756 * allocate some virtual space. will be demand filled 757 * by kernel_object. 758 */ 759 760 error = uvm_map(map, &kva, size, uvm.kernel_object, 761 prefer, align, UVM_MAPFLAG(UVM_PROT_ALL, 762 UVM_PROT_ALL, UVM_INH_NONE, UVM_ADV_RANDOM, flags)); 763 764 KASSERT(error == 0 || (flags & UVM_KMF_NOWAIT) != 0); 765 766 UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0); 767 768 return (kva); 769 } 770 771 /* Function definitions for binary compatibility */ 772 vaddr_t 773 uvm_km_kmemalloc(struct vm_map *map, struct uvm_object *obj, 774 vsize_t sz, int flags) 775 { 776 return uvm_km_kmemalloc1(map, obj, sz, 0, UVM_UNKNOWN_OFFSET, flags); 777 } 778 779 vaddr_t uvm_km_valloc(struct vm_map *map, vsize_t sz) 780 { 781 return uvm_km_valloc1(map, sz, 0, UVM_UNKNOWN_OFFSET, UVM_KMF_NOWAIT); 782 } 783 784 vaddr_t uvm_km_valloc_align(struct vm_map *map, vsize_t sz, vsize_t align) 785 { 786 return uvm_km_valloc1(map, sz, align, UVM_UNKNOWN_OFFSET, UVM_KMF_NOWAIT); 787 } 788 789 vaddr_t uvm_km_valloc_prefer_wait(struct vm_map *map, vsize_t sz, voff_t prefer) 790 { 791 return uvm_km_valloc1(map, sz, 0, prefer, 0); 792 } 793 794 vaddr_t uvm_km_valloc_wait(struct vm_map *map, vsize_t sz) 795 { 796 return uvm_km_valloc1(map, sz, 0, UVM_UNKNOWN_OFFSET, 0); 797 } 798 799 /* Sanity; must specify both or none. */ 800 #if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \ 801 (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE)) 802 #error Must specify MAP and UNMAP together. 803 #endif 804 805 /* 806 * uvm_km_alloc_poolpage: allocate a page for the pool allocator 807 * 808 * => if the pmap specifies an alternate mapping method, we use it. 809 */ 810 811 /* ARGSUSED */ 812 vaddr_t 813 uvm_km_alloc_poolpage_cache(map, obj, waitok) 814 struct vm_map *map; 815 struct uvm_object *obj; 816 boolean_t waitok; 817 { 818 #if defined(PMAP_MAP_POOLPAGE) 819 return uvm_km_alloc_poolpage1(map, obj, waitok); 820 #else 821 struct vm_page *pg; 822 struct pool *pp = &vm_map_to_kernel(map)->vmk_vacache; 823 vaddr_t va; 824 int s = 0xdeadbeaf; /* XXX: gcc */ 825 const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0; 826 827 if ((map->flags & VM_MAP_VACACHE) == 0) 828 return uvm_km_alloc_poolpage1(map, obj, waitok); 829 830 if (intrsafe) 831 s = splvm(); 832 va = (vaddr_t)pool_get(pp, waitok ? PR_WAITOK : PR_NOWAIT); 833 if (intrsafe) 834 splx(s); 835 if (va == 0) 836 return 0; 837 KASSERT(!pmap_extract(pmap_kernel(), va, NULL)); 838 again: 839 pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE); 840 if (__predict_false(pg == NULL)) { 841 if (waitok) { 842 uvm_wait("plpg"); 843 goto again; 844 } else { 845 if (intrsafe) 846 s = splvm(); 847 pool_put(pp, (void *)va); 848 if (intrsafe) 849 splx(s); 850 return 0; 851 } 852 } 853 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), 854 VM_PROT_READ|VM_PROT_WRITE); 855 pmap_update(pmap_kernel()); 856 857 return va; 858 #endif /* PMAP_MAP_POOLPAGE */ 859 } 860 861 vaddr_t 862 uvm_km_alloc_poolpage1(map, obj, waitok) 863 struct vm_map *map; 864 struct uvm_object *obj; 865 boolean_t waitok; 866 { 867 #if defined(PMAP_MAP_POOLPAGE) 868 struct vm_page *pg; 869 vaddr_t va; 870 871 again: 872 pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE); 873 if (__predict_false(pg == NULL)) { 874 if (waitok) { 875 uvm_wait("plpg"); 876 goto again; 877 } else 878 return (0); 879 } 880 va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg)); 881 if (__predict_false(va == 0)) 882 uvm_pagefree(pg); 883 return (va); 884 #else 885 vaddr_t va; 886 int s = 0xdeadbeaf; /* XXX: gcc */ 887 const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0; 888 889 if (intrsafe) 890 s = splvm(); 891 va = uvm_km_kmemalloc(map, obj, PAGE_SIZE, 892 waitok ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK); 893 if (intrsafe) 894 splx(s); 895 return (va); 896 #endif /* PMAP_MAP_POOLPAGE */ 897 } 898 899 /* 900 * uvm_km_free_poolpage: free a previously allocated pool page 901 * 902 * => if the pmap specifies an alternate unmapping method, we use it. 903 */ 904 905 /* ARGSUSED */ 906 void 907 uvm_km_free_poolpage_cache(map, addr) 908 struct vm_map *map; 909 vaddr_t addr; 910 { 911 #if defined(PMAP_UNMAP_POOLPAGE) 912 uvm_km_free_poolpage1(map, addr); 913 #else 914 struct pool *pp; 915 int s = 0xdeadbeaf; /* XXX: gcc */ 916 const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0; 917 918 if ((map->flags & VM_MAP_VACACHE) == 0) { 919 uvm_km_free_poolpage1(map, addr); 920 return; 921 } 922 923 KASSERT(pmap_extract(pmap_kernel(), addr, NULL)); 924 uvm_km_pgremove_intrsafe(addr, addr + PAGE_SIZE); 925 pmap_kremove(addr, PAGE_SIZE); 926 #if defined(DEBUG) 927 pmap_update(pmap_kernel()); 928 #endif 929 KASSERT(!pmap_extract(pmap_kernel(), addr, NULL)); 930 pp = &vm_map_to_kernel(map)->vmk_vacache; 931 if (intrsafe) 932 s = splvm(); 933 pool_put(pp, (void *)addr); 934 if (intrsafe) 935 splx(s); 936 #endif 937 } 938 939 /* ARGSUSED */ 940 void 941 uvm_km_free_poolpage1(map, addr) 942 struct vm_map *map; 943 vaddr_t addr; 944 { 945 #if defined(PMAP_UNMAP_POOLPAGE) 946 paddr_t pa; 947 948 pa = PMAP_UNMAP_POOLPAGE(addr); 949 uvm_pagefree(PHYS_TO_VM_PAGE(pa)); 950 #else 951 int s = 0xdeadbeaf; /* XXX: gcc */ 952 const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0; 953 954 if (intrsafe) 955 s = splvm(); 956 uvm_km_free(map, addr, PAGE_SIZE); 957 if (intrsafe) 958 splx(s); 959 #endif /* PMAP_UNMAP_POOLPAGE */ 960 } 961