uvm_km.c revision 1.126
1/* $NetBSD: uvm_km.c,v 1.126 2012/06/02 08:42:37 para 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. 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 * @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 37 * from: Id: uvm_km.c,v 1.1.2.14 1998/02/06 05:19:27 chs Exp 38 * 39 * 40 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 41 * All rights reserved. 42 * 43 * Permission to use, copy, modify and distribute this software and 44 * its documentation is hereby granted, provided that both the copyright 45 * notice and this permission notice appear in all copies of the 46 * software, derivative works or modified versions, and any portions 47 * thereof, and that both notices appear in supporting documentation. 48 * 49 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 50 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 51 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 52 * 53 * Carnegie Mellon requests users of this software to return to 54 * 55 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 56 * School of Computer Science 57 * Carnegie Mellon University 58 * Pittsburgh PA 15213-3890 59 * 60 * any improvements or extensions that they make and grant Carnegie the 61 * rights to redistribute these changes. 62 */ 63 64/* 65 * uvm_km.c: handle kernel memory allocation and management 66 */ 67 68/* 69 * overview of kernel memory management: 70 * 71 * the kernel virtual address space is mapped by "kernel_map." kernel_map 72 * starts at VM_MIN_KERNEL_ADDRESS and goes to VM_MAX_KERNEL_ADDRESS. 73 * note that VM_MIN_KERNEL_ADDRESS is equal to vm_map_min(kernel_map). 74 * 75 * the kernel_map has several "submaps." submaps can only appear in 76 * the kernel_map (user processes can't use them). submaps "take over" 77 * the management of a sub-range of the kernel's address space. submaps 78 * are typically allocated at boot time and are never released. kernel 79 * virtual address space that is mapped by a submap is locked by the 80 * submap's lock -- not the kernel_map's lock. 81 * 82 * thus, the useful feature of submaps is that they allow us to break 83 * up the locking and protection of the kernel address space into smaller 84 * chunks. 85 * 86 * the vm system has several standard kernel submaps/arenas, including: 87 * kmem_arena => used for kmem/pool (memoryallocators(9)) 88 * pager_map => used to map "buf" structures into kernel space 89 * exec_map => used during exec to handle exec args 90 * etc... 91 * 92 * the kmem_arena is a "special submap", as it lives a fixed map entry 93 * within the kernel_map and controlled by vmem(9). 94 * 95 * the kernel allocates its private memory out of special uvm_objects whose 96 * reference count is set to UVM_OBJ_KERN (thus indicating that the objects 97 * are "special" and never die). all kernel objects should be thought of 98 * as large, fixed-sized, sparsely populated uvm_objects. each kernel 99 * object is equal to the size of kernel virtual address space (i.e. the 100 * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS"). 101 * 102 * note that just because a kernel object spans the entire kernel virtual 103 * address space doesn't mean that it has to be mapped into the entire space. 104 * large chunks of a kernel object's space go unused either because 105 * that area of kernel VM is unmapped, or there is some other type of 106 * object mapped into that range (e.g. a vnode). for submap's kernel 107 * objects, the only part of the object that can ever be populated is the 108 * offsets that are managed by the submap. 109 * 110 * note that the "offset" in a kernel object is always the kernel virtual 111 * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)). 112 * example: 113 * suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a 114 * uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the 115 * kernel map]. if uvm_km_alloc returns virtual address 0xf8235000, 116 * then that means that the page at offset 0x235000 in kernel_object is 117 * mapped at 0xf8235000. 118 * 119 * kernel object have one other special property: when the kernel virtual 120 * memory mapping them is unmapped, the backing memory in the object is 121 * freed right away. this is done with the uvm_km_pgremove() function. 122 * this has to be done because there is no backing store for kernel pages 123 * and no need to save them after they are no longer referenced. 124 * 125 * kmem_arena: main arena controlling the kernel kva used by other arenas. 126 * kmem_va_arena: it utilizes quantum caching for fast allocations and to 127 * lower fragmentation. the pool and kmem allocate from this arena 128 * except for some pool meta-data. 129 * 130 * arenas for metadata allocations used by vmem(9) and pool(9) 131 * note: these arenas can't use quantum caching, the kmem_va_meta_arena 132 * compensates for this by importing larger chunks from kmem_arena. 133 * 134 * kmem_va_meta_arena: space for metadata is allocated from this arena. 135 * kmem_meta_arena: imports from kmem_va_meta_arena. 136 * allocations from this arena are backed with vm_pages. 137 * 138 * arena stacking: 139 * kmem_arena 140 * kmem_va_arena 141 * kmem_va_meta_arena 142 * kmem_meta_arena 143 * 144 */ 145 146#include <sys/cdefs.h> 147__KERNEL_RCSID(0, "$NetBSD: uvm_km.c,v 1.126 2012/06/02 08:42:37 para Exp $"); 148 149#include "opt_uvmhist.h" 150 151#include "opt_kmempages.h" 152 153#ifndef NKMEMPAGES 154#define NKMEMPAGES 0 155#endif 156 157/* 158 * Defaults for lower and upper-bounds for the kmem_arena page count. 159 * Can be overridden by kernel config options. 160 */ 161#ifndef NKMEMPAGES_MIN 162#define NKMEMPAGES_MIN NKMEMPAGES_MIN_DEFAULT 163#endif 164 165#ifndef NKMEMPAGES_MAX 166#define NKMEMPAGES_MAX NKMEMPAGES_MAX_DEFAULT 167#endif 168 169 170#include <sys/param.h> 171#include <sys/systm.h> 172#include <sys/proc.h> 173#include <sys/pool.h> 174#include <sys/vmem.h> 175#include <sys/kmem.h> 176 177#include <uvm/uvm.h> 178 179/* 180 * global data structures 181 */ 182 183struct vm_map *kernel_map = NULL; 184 185/* 186 * local data structues 187 */ 188 189static struct vm_map kernel_map_store; 190static struct vm_map_entry kernel_image_mapent_store; 191static struct vm_map_entry kernel_kmem_mapent_store; 192 193int nkmempages = 0; 194vaddr_t kmembase; 195vsize_t kmemsize; 196 197vmem_t *kmem_arena; 198vmem_t *kmem_va_arena; 199 200/* 201 * kmeminit_nkmempages: calculate the size of kmem_arena. 202 */ 203void 204kmeminit_nkmempages(void) 205{ 206 int npages; 207 208 if (nkmempages != 0) { 209 /* 210 * It's already been set (by us being here before) 211 * bail out now; 212 */ 213 return; 214 } 215 216#if defined(PMAP_MAP_POOLPAGE) 217 npages = (physmem / 4); 218#else 219 npages = (physmem / 3) * 2; 220#endif /* defined(PMAP_MAP_POOLPAGE) */ 221 222#ifndef NKMEMPAGES_MAX_UNLIMITED 223 if (npages > NKMEMPAGES_MAX) 224 npages = NKMEMPAGES_MAX; 225#endif 226 227 if (npages < NKMEMPAGES_MIN) 228 npages = NKMEMPAGES_MIN; 229 230 nkmempages = npages; 231} 232 233/* 234 * uvm_km_bootstrap: init kernel maps and objects to reflect reality (i.e. 235 * KVM already allocated for text, data, bss, and static data structures). 236 * 237 * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS. 238 * we assume that [vmin -> start] has already been allocated and that 239 * "end" is the end. 240 */ 241 242void 243uvm_km_bootstrap(vaddr_t start, vaddr_t end) 244{ 245 bool kmem_arena_small; 246 vaddr_t base = VM_MIN_KERNEL_ADDRESS; 247 struct uvm_map_args args; 248 int error; 249 250 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 251 UVMHIST_LOG(maphist, "start=%"PRIxVADDR" end=%#"PRIxVADDR, 252 start, end, 0,0); 253 254 kmeminit_nkmempages(); 255 kmemsize = (vsize_t)nkmempages * PAGE_SIZE; 256 kmem_arena_small = kmemsize < 64 * 1024 * 1024; 257 258 UVMHIST_LOG(maphist, "kmemsize=%#"PRIxVSIZE, kmemsize, 0,0,0); 259 260 /* 261 * next, init kernel memory objects. 262 */ 263 264 /* kernel_object: for pageable anonymous kernel memory */ 265 uvm_kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS - 266 VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ); 267 268 /* 269 * init the map and reserve any space that might already 270 * have been allocated kernel space before installing. 271 */ 272 273 uvm_map_setup(&kernel_map_store, base, end, VM_MAP_PAGEABLE); 274 kernel_map_store.pmap = pmap_kernel(); 275 if (start != base) { 276 error = uvm_map_prepare(&kernel_map_store, 277 base, start - base, 278 NULL, UVM_UNKNOWN_OFFSET, 0, 279 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 280 UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args); 281 if (!error) { 282 kernel_image_mapent_store.flags = 283 UVM_MAP_KERNEL | UVM_MAP_STATIC | UVM_MAP_NOMERGE; 284 error = uvm_map_enter(&kernel_map_store, &args, 285 &kernel_image_mapent_store); 286 } 287 288 if (error) 289 panic( 290 "uvm_km_bootstrap: could not reserve space for kernel"); 291 292 kmembase = args.uma_start + args.uma_size; 293 } else { 294 kmembase = base; 295 } 296 297 error = uvm_map_prepare(&kernel_map_store, 298 kmembase, kmemsize, 299 NULL, UVM_UNKNOWN_OFFSET, 0, 300 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 301 UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args); 302 if (!error) { 303 kernel_kmem_mapent_store.flags = 304 UVM_MAP_KERNEL | UVM_MAP_STATIC | UVM_MAP_NOMERGE; 305 error = uvm_map_enter(&kernel_map_store, &args, 306 &kernel_kmem_mapent_store); 307 } 308 309 if (error) 310 panic("uvm_km_bootstrap: could not reserve kernel kmem"); 311 312 /* 313 * install! 314 */ 315 316 kernel_map = &kernel_map_store; 317 318 pool_subsystem_init(); 319 vmem_bootstrap(); 320 321 kmem_arena = vmem_create("kmem", kmembase, kmemsize, PAGE_SIZE, 322 NULL, NULL, NULL, 323 0, VM_NOSLEEP | VM_BOOTSTRAP, IPL_VM); 324 325 vmem_init(kmem_arena); 326 327 UVMHIST_LOG(maphist, "kmem vmem created (base=%#"PRIxVADDR 328 ", size=%#"PRIxVSIZE, kmembase, kmemsize, 0,0); 329 330 kmem_va_arena = vmem_create("kva", 0, 0, PAGE_SIZE, 331 vmem_alloc, vmem_free, kmem_arena, 332 (kmem_arena_small ? 4 : 8) * PAGE_SIZE, 333 VM_NOSLEEP | VM_BOOTSTRAP, IPL_VM); 334 335 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 336} 337 338/* 339 * uvm_km_init: init the kernel maps virtual memory caches 340 * and start the pool/kmem allocator. 341 */ 342void 343uvm_km_init(void) 344{ 345 346 kmem_init(); 347 348 kmeminit(); // killme 349} 350 351/* 352 * uvm_km_suballoc: allocate a submap in the kernel map. once a submap 353 * is allocated all references to that area of VM must go through it. this 354 * allows the locking of VAs in kernel_map to be broken up into regions. 355 * 356 * => if `fixed' is true, *vmin specifies where the region described 357 * pager_map => used to map "buf" structures into kernel space 358 * by the submap must start 359 * => if submap is non NULL we use that as the submap, otherwise we 360 * alloc a new map 361 */ 362 363struct vm_map * 364uvm_km_suballoc(struct vm_map *map, vaddr_t *vmin /* IN/OUT */, 365 vaddr_t *vmax /* OUT */, vsize_t size, int flags, bool fixed, 366 struct vm_map *submap) 367{ 368 int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0); 369 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 370 371 KASSERT(vm_map_pmap(map) == pmap_kernel()); 372 373 size = round_page(size); /* round up to pagesize */ 374 375 /* 376 * first allocate a blank spot in the parent map 377 */ 378 379 if (uvm_map(map, vmin, size, NULL, UVM_UNKNOWN_OFFSET, 0, 380 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 381 UVM_ADV_RANDOM, mapflags)) != 0) { 382 panic("%s: unable to allocate space in parent map", __func__); 383 } 384 385 /* 386 * set VM bounds (vmin is filled in by uvm_map) 387 */ 388 389 *vmax = *vmin + size; 390 391 /* 392 * add references to pmap and create or init the submap 393 */ 394 395 pmap_reference(vm_map_pmap(map)); 396 if (submap == NULL) { 397 submap = kmem_alloc(sizeof(*submap), KM_SLEEP); 398 if (submap == NULL) 399 panic("uvm_km_suballoc: unable to create submap"); 400 } 401 uvm_map_setup(submap, *vmin, *vmax, flags); 402 submap->pmap = vm_map_pmap(map); 403 404 /* 405 * now let uvm_map_submap plug in it... 406 */ 407 408 if (uvm_map_submap(map, *vmin, *vmax, submap) != 0) 409 panic("uvm_km_suballoc: submap allocation failed"); 410 411 return(submap); 412} 413 414/* 415 * uvm_km_pgremove: remove pages from a kernel uvm_object and KVA. 416 */ 417 418void 419uvm_km_pgremove(vaddr_t startva, vaddr_t endva) 420{ 421 struct uvm_object * const uobj = uvm_kernel_object; 422 const voff_t start = startva - vm_map_min(kernel_map); 423 const voff_t end = endva - vm_map_min(kernel_map); 424 struct vm_page *pg; 425 voff_t curoff, nextoff; 426 int swpgonlydelta = 0; 427 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 428 429 KASSERT(VM_MIN_KERNEL_ADDRESS <= startva); 430 KASSERT(startva < endva); 431 KASSERT(endva <= VM_MAX_KERNEL_ADDRESS); 432 433 mutex_enter(uobj->vmobjlock); 434 pmap_remove(pmap_kernel(), startva, endva); 435 for (curoff = start; curoff < end; curoff = nextoff) { 436 nextoff = curoff + PAGE_SIZE; 437 pg = uvm_pagelookup(uobj, curoff); 438 if (pg != NULL && pg->flags & PG_BUSY) { 439 pg->flags |= PG_WANTED; 440 UVM_UNLOCK_AND_WAIT(pg, uobj->vmobjlock, 0, 441 "km_pgrm", 0); 442 mutex_enter(uobj->vmobjlock); 443 nextoff = curoff; 444 continue; 445 } 446 447 /* 448 * free the swap slot, then the page. 449 */ 450 451 if (pg == NULL && 452 uao_find_swslot(uobj, curoff >> PAGE_SHIFT) > 0) { 453 swpgonlydelta++; 454 } 455 uao_dropswap(uobj, curoff >> PAGE_SHIFT); 456 if (pg != NULL) { 457 mutex_enter(&uvm_pageqlock); 458 uvm_pagefree(pg); 459 mutex_exit(&uvm_pageqlock); 460 } 461 } 462 mutex_exit(uobj->vmobjlock); 463 464 if (swpgonlydelta > 0) { 465 mutex_enter(&uvm_swap_data_lock); 466 KASSERT(uvmexp.swpgonly >= swpgonlydelta); 467 uvmexp.swpgonly -= swpgonlydelta; 468 mutex_exit(&uvm_swap_data_lock); 469 } 470} 471 472 473/* 474 * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for non object backed 475 * regions. 476 * 477 * => when you unmap a part of anonymous kernel memory you want to toss 478 * the pages right away. (this is called from uvm_unmap_...). 479 * => none of the pages will ever be busy, and none of them will ever 480 * be on the active or inactive queues (because they have no object). 481 */ 482 483void 484uvm_km_pgremove_intrsafe(struct vm_map *map, vaddr_t start, vaddr_t end) 485{ 486#define __PGRM_BATCH 16 487 struct vm_page *pg; 488 paddr_t pa[__PGRM_BATCH]; 489 int npgrm, i; 490 vaddr_t va, batch_vastart; 491 492 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 493 494 KASSERT(VM_MAP_IS_KERNEL(map)); 495 KASSERT(vm_map_min(map) <= start); 496 KASSERT(start < end); 497 KASSERT(end <= vm_map_max(map)); 498 499 for (va = start; va < end;) { 500 batch_vastart = va; 501 /* create a batch of at most __PGRM_BATCH pages to free */ 502 for (i = 0; 503 i < __PGRM_BATCH && va < end; 504 va += PAGE_SIZE) { 505 if (!pmap_extract(pmap_kernel(), va, &pa[i])) { 506 continue; 507 } 508 i++; 509 } 510 npgrm = i; 511 /* now remove the mappings */ 512 pmap_kremove(batch_vastart, va - batch_vastart); 513 /* and free the pages */ 514 for (i = 0; i < npgrm; i++) { 515 pg = PHYS_TO_VM_PAGE(pa[i]); 516 KASSERT(pg); 517 KASSERT(pg->uobject == NULL && pg->uanon == NULL); 518 KASSERT((pg->flags & PG_BUSY) == 0); 519 uvm_pagefree(pg); 520 } 521 } 522#undef __PGRM_BATCH 523} 524 525#if defined(DEBUG) 526void 527uvm_km_check_empty(struct vm_map *map, vaddr_t start, vaddr_t end) 528{ 529 struct vm_page *pg; 530 vaddr_t va; 531 paddr_t pa; 532 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 533 534 KDASSERT(VM_MAP_IS_KERNEL(map)); 535 KDASSERT(vm_map_min(map) <= start); 536 KDASSERT(start < end); 537 KDASSERT(end <= vm_map_max(map)); 538 539 for (va = start; va < end; va += PAGE_SIZE) { 540 if (pmap_extract(pmap_kernel(), va, &pa)) { 541 panic("uvm_km_check_empty: va %p has pa 0x%llx", 542 (void *)va, (long long)pa); 543 } 544 mutex_enter(uvm_kernel_object->vmobjlock); 545 pg = uvm_pagelookup(uvm_kernel_object, 546 va - vm_map_min(kernel_map)); 547 mutex_exit(uvm_kernel_object->vmobjlock); 548 if (pg) { 549 panic("uvm_km_check_empty: " 550 "has page hashed at %p", (const void *)va); 551 } 552 } 553} 554#endif /* defined(DEBUG) */ 555 556/* 557 * uvm_km_alloc: allocate an area of kernel memory. 558 * 559 * => NOTE: we can return 0 even if we can wait if there is not enough 560 * free VM space in the map... caller should be prepared to handle 561 * this case. 562 * => we return KVA of memory allocated 563 */ 564 565vaddr_t 566uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags) 567{ 568 vaddr_t kva, loopva; 569 vaddr_t offset; 570 vsize_t loopsize; 571 struct vm_page *pg; 572 struct uvm_object *obj; 573 int pgaflags; 574 vm_prot_t prot; 575 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 576 577 KASSERT(vm_map_pmap(map) == pmap_kernel()); 578 KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED || 579 (flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE || 580 (flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY); 581 KASSERT((flags & UVM_KMF_VAONLY) != 0 || (flags & UVM_KMF_COLORMATCH) == 0); 582 KASSERT((flags & UVM_KMF_COLORMATCH) == 0 || (flags & UVM_KMF_VAONLY) != 0); 583 584 /* 585 * setup for call 586 */ 587 588 kva = vm_map_min(map); /* hint */ 589 size = round_page(size); 590 obj = (flags & UVM_KMF_PAGEABLE) ? uvm_kernel_object : NULL; 591 UVMHIST_LOG(maphist," (map=0x%x, obj=0x%x, size=0x%x, flags=%d)", 592 map, obj, size, flags); 593 594 /* 595 * allocate some virtual space 596 */ 597 598 if (__predict_false(uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET, 599 align, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 600 UVM_ADV_RANDOM, 601 (flags & (UVM_KMF_TRYLOCK | UVM_KMF_NOWAIT | UVM_KMF_WAITVA 602 | UVM_KMF_COLORMATCH)))) != 0)) { 603 UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0); 604 return(0); 605 } 606 607 /* 608 * if all we wanted was VA, return now 609 */ 610 611 if (flags & (UVM_KMF_VAONLY | UVM_KMF_PAGEABLE)) { 612 UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0); 613 return(kva); 614 } 615 616 /* 617 * recover object offset from virtual address 618 */ 619 620 offset = kva - vm_map_min(kernel_map); 621 UVMHIST_LOG(maphist, " kva=0x%x, offset=0x%x", kva, offset,0,0); 622 623 /* 624 * now allocate and map in the memory... note that we are the only ones 625 * whom should ever get a handle on this area of VM. 626 */ 627 628 loopva = kva; 629 loopsize = size; 630 631 pgaflags = UVM_FLAG_COLORMATCH; 632 if (flags & UVM_KMF_NOWAIT) 633 pgaflags |= UVM_PGA_USERESERVE; 634 if (flags & UVM_KMF_ZERO) 635 pgaflags |= UVM_PGA_ZERO; 636 prot = VM_PROT_READ | VM_PROT_WRITE; 637 if (flags & UVM_KMF_EXEC) 638 prot |= VM_PROT_EXECUTE; 639 while (loopsize) { 640 KASSERTMSG(!pmap_extract(pmap_kernel(), loopva, NULL), 641 "loopva=%#"PRIxVADDR, loopva); 642 643 pg = uvm_pagealloc_strat(NULL, offset, NULL, pgaflags, 644#ifdef UVM_KM_VMFREELIST 645 UVM_PGA_STRAT_ONLY, UVM_KM_VMFREELIST 646#else 647 UVM_PGA_STRAT_NORMAL, 0 648#endif 649 ); 650 651 /* 652 * out of memory? 653 */ 654 655 if (__predict_false(pg == NULL)) { 656 if ((flags & UVM_KMF_NOWAIT) || 657 ((flags & UVM_KMF_CANFAIL) && !uvm_reclaimable())) { 658 /* free everything! */ 659 uvm_km_free(map, kva, size, 660 flags & UVM_KMF_TYPEMASK); 661 return (0); 662 } else { 663 uvm_wait("km_getwait2"); /* sleep here */ 664 continue; 665 } 666 } 667 668 pg->flags &= ~PG_BUSY; /* new page */ 669 UVM_PAGE_OWN(pg, NULL); 670 671 /* 672 * map it in 673 */ 674 675 pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg), 676 prot, PMAP_KMPAGE); 677 loopva += PAGE_SIZE; 678 offset += PAGE_SIZE; 679 loopsize -= PAGE_SIZE; 680 } 681 682 pmap_update(pmap_kernel()); 683 684 UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0); 685 return(kva); 686} 687 688/* 689 * uvm_km_free: free an area of kernel memory 690 */ 691 692void 693uvm_km_free(struct vm_map *map, vaddr_t addr, vsize_t size, uvm_flag_t flags) 694{ 695 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 696 697 KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED || 698 (flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE || 699 (flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY); 700 KASSERT((addr & PAGE_MASK) == 0); 701 KASSERT(vm_map_pmap(map) == pmap_kernel()); 702 703 size = round_page(size); 704 705 if (flags & UVM_KMF_PAGEABLE) { 706 uvm_km_pgremove(addr, addr + size); 707 } else if (flags & UVM_KMF_WIRED) { 708 /* 709 * Note: uvm_km_pgremove_intrsafe() extracts mapping, thus 710 * remove it after. See comment below about KVA visibility. 711 */ 712 uvm_km_pgremove_intrsafe(map, addr, addr + size); 713 } 714 715 /* 716 * Note: uvm_unmap_remove() calls pmap_update() for us, before 717 * KVA becomes globally available. 718 */ 719 720 uvm_unmap1(map, addr, addr + size, UVM_FLAG_VAONLY); 721} 722 723/* Sanity; must specify both or none. */ 724#if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \ 725 (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE)) 726#error Must specify MAP and UNMAP together. 727#endif 728 729int 730uvm_km_kmem_alloc(vmem_t *vm, vmem_size_t size, vm_flag_t flags, 731 vmem_addr_t *addr) 732{ 733 struct vm_page *pg; 734 vmem_addr_t va; 735 int rc; 736 vaddr_t loopva; 737 vsize_t loopsize; 738 739 size = round_page(size); 740 741#if defined(PMAP_MAP_POOLPAGE) 742 if (size == PAGE_SIZE) { 743again: 744#ifdef PMAP_ALLOC_POOLPAGE 745 pg = PMAP_ALLOC_POOLPAGE((flags & VM_SLEEP) ? 746 0 : UVM_PGA_USERESERVE); 747#else 748 pg = uvm_pagealloc(NULL, 0, NULL, 749 (flags & VM_SLEEP) ? 0 : UVM_PGA_USERESERVE); 750#endif /* PMAP_ALLOC_POOLPAGE */ 751 if (__predict_false(pg == NULL)) { 752 if (flags & VM_SLEEP) { 753 uvm_wait("plpg"); 754 goto again; 755 } 756 return ENOMEM; 757 } 758 va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg)); 759 if (__predict_false(va == 0)) { 760 uvm_pagefree(pg); 761 return ENOMEM; 762 } 763 *addr = va; 764 return 0; 765 } 766#endif /* PMAP_MAP_POOLPAGE */ 767 768 rc = vmem_alloc(vm, size, flags, &va); 769 if (rc != 0) 770 return rc; 771 772 loopva = va; 773 loopsize = size; 774 775 while (loopsize) { 776 KASSERTMSG(!pmap_extract(pmap_kernel(), loopva, NULL), 777 "loopva=%#"PRIxVADDR" loopsize=%#"PRIxVSIZE" vmem=%p", 778 loopva, loopsize, vm); 779 780 pg = uvm_pagealloc(NULL, loopva, NULL, 781 UVM_FLAG_COLORMATCH 782 | ((flags & VM_SLEEP) ? 0 : UVM_PGA_USERESERVE)); 783 if (__predict_false(pg == NULL)) { 784 if (flags & VM_SLEEP) { 785 uvm_wait("plpg"); 786 continue; 787 } else { 788 uvm_km_pgremove_intrsafe(kernel_map, va, 789 va + size); 790 vmem_free(vm, va, size); 791 return ENOMEM; 792 } 793 } 794 795 pg->flags &= ~PG_BUSY; /* new page */ 796 UVM_PAGE_OWN(pg, NULL); 797 pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg), 798 VM_PROT_READ|VM_PROT_WRITE, PMAP_KMPAGE); 799 800 loopva += PAGE_SIZE; 801 loopsize -= PAGE_SIZE; 802 } 803 pmap_update(pmap_kernel()); 804 805 *addr = va; 806 807 return 0; 808} 809 810void 811uvm_km_kmem_free(vmem_t *vm, vmem_addr_t addr, size_t size) 812{ 813 814 size = round_page(size); 815#if defined(PMAP_UNMAP_POOLPAGE) 816 if (size == PAGE_SIZE) { 817 paddr_t pa; 818 819 pa = PMAP_UNMAP_POOLPAGE(addr); 820 uvm_pagefree(PHYS_TO_VM_PAGE(pa)); 821 return; 822 } 823#endif /* PMAP_UNMAP_POOLPAGE */ 824 uvm_km_pgremove_intrsafe(kernel_map, addr, addr + size); 825 pmap_update(pmap_kernel()); 826 827 vmem_free(vm, addr, size); 828} 829 830bool 831uvm_km_va_starved_p(void) 832{ 833 vmem_size_t total; 834 vmem_size_t free; 835 836 total = vmem_size(kmem_arena, VMEM_ALLOC|VMEM_FREE); 837 free = vmem_size(kmem_arena, VMEM_FREE); 838 839 return (free < (total / 10)); 840} 841