1 /* 2 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Hiten Pandya <hmp@backplane.com>. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 */ 35 /* 36 * Copyright (c) 1991 Regents of the University of California. 37 * All rights reserved. 38 * 39 * This code is derived from software contributed to Berkeley by 40 * The Mach Operating System project at Carnegie-Mellon University. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91 67 * $DragonFly: src/sys/vm/vm_contig.c,v 1.9 2004/08/17 18:57:36 dillon Exp $ 68 */ 69 70 /* 71 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 72 * All rights reserved. 73 * 74 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 75 * 76 * Permission to use, copy, modify and distribute this software and 77 * its documentation is hereby granted, provided that both the copyright 78 * notice and this permission notice appear in all copies of the 79 * software, derivative works or modified versions, and any portions 80 * thereof, and that both notices appear in supporting documentation. 81 * 82 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 83 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 84 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 85 * 86 * Carnegie Mellon requests users of this software to return to 87 * 88 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 89 * School of Computer Science 90 * Carnegie Mellon University 91 * Pittsburgh PA 15213-3890 92 * 93 * any improvements or extensions that they make and grant Carnegie the 94 * rights to redistribute these changes. 95 */ 96 97 /* 98 * Contiguous memory allocation API. 99 */ 100 101 #include <sys/param.h> 102 #include <sys/systm.h> 103 #include <sys/malloc.h> 104 #include <sys/proc.h> 105 #include <sys/lock.h> 106 #include <sys/vmmeter.h> 107 #include <sys/vnode.h> 108 109 #include <vm/vm.h> 110 #include <vm/vm_param.h> 111 #include <vm/vm_kern.h> 112 #include <vm/pmap.h> 113 #include <vm/vm_map.h> 114 #include <vm/vm_object.h> 115 #include <vm/vm_page.h> 116 #include <vm/vm_pageout.h> 117 #include <vm/vm_pager.h> 118 #include <vm/vm_extern.h> 119 #include <vm/vm_page2.h> 120 121 /* 122 * vm_contig_pg_clean: 123 * 124 * Do a thorough cleanup of the specified 'queue', which can be either 125 * PQ_ACTIVE or PQ_INACTIVE by doing a walkthrough. If the page is not 126 * marked dirty, it is shoved into the page cache, provided no one has 127 * currently aqcuired it, otherwise localized action per object type 128 * is taken for cleanup: 129 * 130 * In the OBJT_VNODE case, the whole page range is cleaned up 131 * using the vm_object_page_clean() routine, by specyfing a 132 * start and end of '0'. 133 * 134 * Otherwise if the object is of any other type, the generic 135 * pageout (daemon) flush routine is invoked. 136 */ 137 static int 138 vm_contig_pg_clean(int queue) 139 { 140 vm_object_t object; 141 vm_page_t m, m_tmp, next; 142 143 for (m = TAILQ_FIRST(&vm_page_queues[queue].pl); m != NULL; m = next) { 144 KASSERT(m->queue == queue, 145 ("vm_contig_clean: page %p's queue is not %d", m, queue)); 146 147 next = TAILQ_NEXT(m, pageq); 148 149 if (vm_page_sleep_busy(m, TRUE, "vpctw0")) 150 return (TRUE); 151 152 vm_page_test_dirty(m); 153 if (m->dirty) { 154 object = m->object; 155 if (object->type == OBJT_VNODE) { 156 vn_lock(object->handle, NULL, 157 LK_EXCLUSIVE | LK_RETRY, curthread); 158 vm_object_page_clean(object, 0, 0, OBJPC_SYNC); 159 VOP_UNLOCK(((struct vnode *)object->handle), 160 NULL, 0, curthread); 161 return (TRUE); 162 } else if (object->type == OBJT_SWAP || 163 object->type == OBJT_DEFAULT) { 164 m_tmp = m; 165 vm_pageout_flush(&m_tmp, 1, 0); 166 return (TRUE); 167 } 168 } 169 170 if ((m->dirty == 0) && (m->busy == 0) && (m->hold_count == 0)) 171 vm_page_cache(m); 172 } 173 174 return (FALSE); 175 } 176 177 /* 178 * vm_contig_pg_alloc: 179 * 180 * Allocate contiguous pages from the VM. This function does not 181 * map the allocated pages into the kernel map, otherwise it is 182 * impossible to make large allocations (i.e. >2G). 183 * 184 * Malloc()'s data structures have been used for collection of 185 * statistics and for allocations of less than a page. 186 * 187 */ 188 int 189 vm_contig_pg_alloc( 190 unsigned long size, 191 vm_paddr_t low, 192 vm_paddr_t high, 193 unsigned long alignment, 194 unsigned long boundary) 195 { 196 int i, s, start, pass; 197 vm_offset_t phys; 198 vm_page_t pga = vm_page_array; 199 200 size = round_page(size); 201 if (size == 0) 202 panic("vm_contig_pg_alloc: size must not be 0"); 203 if ((alignment & (alignment - 1)) != 0) 204 panic("vm_contig_pg_alloc: alignment must be a power of 2"); 205 if ((boundary & (boundary - 1)) != 0) 206 panic("vm_contig_pg_alloc: boundary must be a power of 2"); 207 208 start = 0; 209 for (pass = 0; pass <= 1; pass++) { 210 s = splvm(); 211 again: 212 /* 213 * Find first page in array that is free, within range, aligned, and 214 * such that the boundary won't be crossed. 215 */ 216 for (i = start; i < vmstats.v_page_count; i++) { 217 int pqtype; 218 phys = VM_PAGE_TO_PHYS(&pga[i]); 219 pqtype = pga[i].queue - pga[i].pc; 220 if (((pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) && 221 (phys >= low) && (phys < high) && 222 ((phys & (alignment - 1)) == 0) && 223 (((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0)) 224 break; 225 } 226 227 /* 228 * If we cannot find the page in the given range, or we have 229 * crossed the boundary, call the vm_contig_pg_clean() function 230 * for flushing out the queues, and returning it back to 231 * normal state. 232 */ 233 if ((i == vmstats.v_page_count) || 234 ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) { 235 236 again1: 237 if (vm_contig_pg_clean(PQ_INACTIVE)) 238 goto again1; 239 if (vm_contig_pg_clean(PQ_ACTIVE)) 240 goto again1; 241 242 splx(s); 243 continue; /* next pass */ 244 } 245 start = i; 246 247 /* 248 * Check successive pages for contiguous and free. 249 */ 250 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) { 251 int pqtype; 252 pqtype = pga[i].queue - pga[i].pc; 253 if ((VM_PAGE_TO_PHYS(&pga[i]) != 254 (VM_PAGE_TO_PHYS(&pga[i - 1]) + PAGE_SIZE)) || 255 ((pqtype != PQ_FREE) && (pqtype != PQ_CACHE))) { 256 start++; 257 goto again; 258 } 259 } 260 261 for (i = start; i < (start + size / PAGE_SIZE); i++) { 262 int pqtype; 263 vm_page_t m = &pga[i]; 264 265 pqtype = m->queue - m->pc; 266 if (pqtype == PQ_CACHE) { 267 vm_page_busy(m); 268 vm_page_free(m); 269 } 270 vm_page_unqueue_nowakeup(m); 271 m->valid = VM_PAGE_BITS_ALL; 272 if (m->flags & PG_ZERO) 273 vm_page_zero_count--; 274 /* Don't clear the PG_ZERO flag, we'll need it later. */ 275 m->flags &= PG_ZERO; 276 KASSERT(m->dirty == 0, 277 ("vm_contig_pg_alloc: page %p was dirty", m)); 278 m->wire_count = 0; 279 m->busy = 0; 280 m->object = NULL; 281 } 282 283 /* 284 * Our job is done, return the index page of vm_page_array. 285 */ 286 287 splx(s); 288 return (start); /* aka &pga[start] */ 289 } 290 291 /* 292 * Failed. 293 */ 294 splx(s); 295 return (-1); 296 } 297 298 /* 299 * vm_contig_pg_free: 300 * 301 * Remove pages previously allocated by vm_contig_pg_alloc, and 302 * assume all references to the pages have been removed, and that 303 * it is OK to add them back to the free list. 304 */ 305 void 306 vm_contig_pg_free(int start, u_long size) 307 { 308 vm_page_t pga = vm_page_array; 309 int i; 310 311 size = round_page(size); 312 if (size == 0) 313 panic("vm_contig_pg_free: size must not be 0"); 314 315 for (i = start; i < (start + size / PAGE_SIZE); i++) { 316 vm_page_free(&pga[i]); 317 } 318 } 319 320 /* 321 * vm_contig_pg_kmap: 322 * 323 * Map previously allocated (vm_contig_pg_alloc) range of pages from 324 * vm_page_array[] into the KVA. Once mapped, the pages are part of 325 * the Kernel, and are to free'ed with kmem_free(kernel_map, addr, size). 326 */ 327 vm_offset_t 328 vm_contig_pg_kmap(int start, u_long size, vm_map_t map, int flags) 329 { 330 vm_offset_t addr, tmp_addr; 331 vm_page_t pga = vm_page_array; 332 int i, s, count; 333 334 size = round_page(size); 335 if (size == 0) 336 panic("vm_contig_pg_kmap: size must not be 0"); 337 338 s = splvm(); /* XXX: is this really needed? */ 339 340 /* 341 * We've found a contiguous chunk that meets our requirements. 342 * Allocate KVM, and assign phys pages and return a kernel VM 343 * pointer. 344 */ 345 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 346 vm_map_lock(map); 347 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr) != 348 KERN_SUCCESS) { 349 /* 350 * XXX We almost never run out of kernel virtual 351 * space, so we don't make the allocated memory 352 * above available. 353 */ 354 vm_map_unlock(map); 355 vm_map_entry_release(count); 356 splx(s); 357 return (0); 358 } 359 vm_object_reference(kernel_object); 360 vm_map_insert(map, &count, 361 kernel_object, addr - VM_MIN_KERNEL_ADDRESS, 362 addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0); 363 vm_map_unlock(map); 364 vm_map_entry_release(count); 365 366 tmp_addr = addr; 367 for (i = start; i < (start + size / PAGE_SIZE); i++) { 368 vm_page_t m = &pga[i]; 369 vm_page_insert(m, kernel_object, 370 OFF_TO_IDX(tmp_addr - VM_MIN_KERNEL_ADDRESS)); 371 if ((flags & M_ZERO) && !(m->flags & PG_ZERO)) 372 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 373 m->flags = 0; 374 tmp_addr += PAGE_SIZE; 375 } 376 vm_map_wire(map, addr, addr + size, 0); 377 378 splx(s); 379 return (addr); 380 } 381 382 void * 383 contigmalloc( 384 unsigned long size, /* should be size_t here and for malloc() */ 385 struct malloc_type *type, 386 int flags, 387 vm_paddr_t low, 388 vm_paddr_t high, 389 unsigned long alignment, 390 unsigned long boundary) 391 { 392 return contigmalloc_map(size, type, flags, low, high, alignment, 393 boundary, kernel_map); 394 } 395 396 void * 397 contigmalloc_map( 398 unsigned long size, /* should be size_t here and for malloc() */ 399 struct malloc_type *type, 400 int flags, 401 vm_paddr_t low, 402 vm_paddr_t high, 403 unsigned long alignment, 404 unsigned long boundary, 405 vm_map_t map) 406 { 407 int index; 408 void *rv; 409 410 index = vm_contig_pg_alloc(size, low, high, alignment, boundary); 411 if (index < 0) { 412 printf("contigmalloc_map: failed in index < 0 case!"); 413 return NULL; 414 } 415 416 rv = (void *) vm_contig_pg_kmap(index, size, map, flags); 417 if (!rv) 418 vm_contig_pg_free(index, size); 419 420 return rv; 421 } 422 423 void 424 contigfree(void *addr, unsigned long size, struct malloc_type *type) 425 { 426 kmem_free(kernel_map, (vm_offset_t)addr, size); 427 } 428 429 vm_offset_t 430 vm_page_alloc_contig( 431 vm_offset_t size, 432 vm_paddr_t low, 433 vm_paddr_t high, 434 vm_offset_t alignment) 435 { 436 return ((vm_offset_t)contigmalloc_map(size, M_DEVBUF, M_NOWAIT, low, 437 high, alignment, 0ul, kernel_map)); 438 } 439