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 1.12 2004/11/10 20:19:51 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> 119 120 #include <sys/thread2.h> 121 #include <vm/vm_page2.h> 122 123 /* 124 * vm_contig_pg_clean: 125 * 126 * Do a thorough cleanup of the specified 'queue', which can be either 127 * PQ_ACTIVE or PQ_INACTIVE by doing a walkthrough. If the page is not 128 * marked dirty, it is shoved into the page cache, provided no one has 129 * currently aqcuired it, otherwise localized action per object type 130 * is taken for cleanup: 131 * 132 * In the OBJT_VNODE case, the whole page range is cleaned up 133 * using the vm_object_page_clean() routine, by specyfing a 134 * start and end of '0'. 135 * 136 * Otherwise if the object is of any other type, the generic 137 * pageout (daemon) flush routine is invoked. 138 * 139 * We must be in a critical section. 140 */ 141 static int 142 vm_contig_pg_clean(int queue) 143 { 144 vm_object_t object; 145 vm_page_t m, m_tmp, next; 146 147 for (m = TAILQ_FIRST(&vm_page_queues[queue].pl); m != NULL; m = next) { 148 KASSERT(m->queue == queue, 149 ("vm_contig_clean: page %p's queue is not %d", m, queue)); 150 151 next = TAILQ_NEXT(m, pageq); 152 153 if (vm_page_sleep_busy(m, TRUE, "vpctw0")) 154 return (TRUE); 155 156 vm_page_test_dirty(m); 157 if (m->dirty) { 158 object = m->object; 159 if (object->type == OBJT_VNODE) { 160 vn_lock(object->handle, 161 LK_EXCLUSIVE | LK_RETRY, curthread); 162 vm_object_page_clean(object, 0, 0, OBJPC_SYNC); 163 VOP_UNLOCK(((struct vnode *)object->handle), 164 0, curthread); 165 return (TRUE); 166 } else if (object->type == OBJT_SWAP || 167 object->type == OBJT_DEFAULT) { 168 m_tmp = m; 169 vm_pageout_flush(&m_tmp, 1, 0); 170 return (TRUE); 171 } 172 } 173 174 if ((m->dirty == 0) && (m->busy == 0) && (m->hold_count == 0)) 175 vm_page_cache(m); 176 } 177 178 return (FALSE); 179 } 180 181 /* 182 * vm_contig_pg_alloc: 183 * 184 * Allocate contiguous pages from the VM. This function does not 185 * map the allocated pages into the kernel map, otherwise it is 186 * impossible to make large allocations (i.e. >2G). 187 * 188 * Malloc()'s data structures have been used for collection of 189 * statistics and for allocations of less than a page. 190 * 191 */ 192 int 193 vm_contig_pg_alloc( 194 unsigned long size, 195 vm_paddr_t low, 196 vm_paddr_t high, 197 unsigned long alignment, 198 unsigned long boundary) 199 { 200 int i, start, pass; 201 vm_offset_t phys; 202 vm_page_t pga = vm_page_array; 203 vm_page_t m; 204 int pqtype; 205 206 size = round_page(size); 207 if (size == 0) 208 panic("vm_contig_pg_alloc: size must not be 0"); 209 if ((alignment & (alignment - 1)) != 0) 210 panic("vm_contig_pg_alloc: alignment must be a power of 2"); 211 if ((boundary & (boundary - 1)) != 0) 212 panic("vm_contig_pg_alloc: boundary must be a power of 2"); 213 214 start = 0; 215 for (pass = 0; pass <= 1; pass++) { 216 crit_enter(); 217 again: 218 /* 219 * Find first page in array that is free, within range, aligned, and 220 * such that the boundary won't be crossed. 221 */ 222 for (i = start; i < vmstats.v_page_count; i++) { 223 m = &pga[i]; 224 phys = VM_PAGE_TO_PHYS(m); 225 pqtype = m->queue - m->pc; 226 if (((pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) && 227 (phys >= low) && (phys < high) && 228 ((phys & (alignment - 1)) == 0) && 229 (((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0) && 230 m->busy == 0 && m->wire_count == 0 && 231 m->hold_count == 0 && (m->flags & PG_BUSY) == 0 232 233 ) { 234 break; 235 } 236 } 237 238 /* 239 * If we cannot find the page in the given range, or we have 240 * crossed the boundary, call the vm_contig_pg_clean() function 241 * for flushing out the queues, and returning it back to 242 * normal state. 243 */ 244 if ((i == vmstats.v_page_count) || 245 ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) { 246 247 again1: 248 if (vm_contig_pg_clean(PQ_INACTIVE)) 249 goto again1; 250 if (vm_contig_pg_clean(PQ_ACTIVE)) 251 goto again1; 252 253 crit_exit(); 254 continue; /* next pass */ 255 } 256 start = i; 257 258 /* 259 * Check successive pages for contiguous and free. 260 * 261 * (still in critical section) 262 */ 263 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) { 264 m = &pga[i]; 265 pqtype = m->queue - m->pc; 266 if ((VM_PAGE_TO_PHYS(&m[0]) != 267 (VM_PAGE_TO_PHYS(&m[-1]) + PAGE_SIZE)) || 268 ((pqtype != PQ_FREE) && (pqtype != PQ_CACHE)) || 269 m->busy || m->wire_count || 270 m->hold_count || (m->flags & PG_BUSY) 271 ) { 272 start++; 273 goto again; 274 } 275 } 276 277 /* 278 * (still in critical section) 279 */ 280 for (i = start; i < (start + size / PAGE_SIZE); i++) { 281 m = &pga[i]; 282 pqtype = m->queue - m->pc; 283 if (pqtype == PQ_CACHE) { 284 vm_page_busy(m); 285 vm_page_free(m); 286 } 287 KKASSERT(m->object == NULL); 288 vm_page_unqueue_nowakeup(m); 289 m->valid = VM_PAGE_BITS_ALL; 290 if (m->flags & PG_ZERO) 291 vm_page_zero_count--; 292 /* Don't clear the PG_ZERO flag, we'll need it later. */ 293 m->flags &= PG_ZERO; 294 KASSERT(m->dirty == 0, 295 ("vm_contig_pg_alloc: page %p was dirty", m)); 296 m->wire_count = 0; 297 m->busy = 0; 298 } 299 300 /* 301 * Our job is done, return the index page of vm_page_array. 302 */ 303 crit_exit(); 304 return (start); /* aka &pga[start] */ 305 } 306 307 /* 308 * Failed. 309 */ 310 crit_exit(); 311 return (-1); 312 } 313 314 /* 315 * vm_contig_pg_free: 316 * 317 * Remove pages previously allocated by vm_contig_pg_alloc, and 318 * assume all references to the pages have been removed, and that 319 * it is OK to add them back to the free list. 320 */ 321 void 322 vm_contig_pg_free(int start, u_long size) 323 { 324 vm_page_t pga = vm_page_array; 325 int i; 326 327 size = round_page(size); 328 if (size == 0) 329 panic("vm_contig_pg_free: size must not be 0"); 330 331 for (i = start; i < (start + size / PAGE_SIZE); i++) { 332 vm_page_free(&pga[i]); 333 } 334 } 335 336 /* 337 * vm_contig_pg_kmap: 338 * 339 * Map previously allocated (vm_contig_pg_alloc) range of pages from 340 * vm_page_array[] into the KVA. Once mapped, the pages are part of 341 * the Kernel, and are to free'ed with kmem_free(kernel_map, addr, size). 342 */ 343 vm_offset_t 344 vm_contig_pg_kmap(int start, u_long size, vm_map_t map, int flags) 345 { 346 vm_offset_t addr, tmp_addr; 347 vm_page_t pga = vm_page_array; 348 int i, count; 349 350 size = round_page(size); 351 if (size == 0) 352 panic("vm_contig_pg_kmap: size must not be 0"); 353 354 crit_enter(); 355 356 /* 357 * We've found a contiguous chunk that meets our requirements. 358 * Allocate KVM, and assign phys pages and return a kernel VM 359 * pointer. 360 */ 361 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 362 vm_map_lock(map); 363 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr) != 364 KERN_SUCCESS) { 365 /* 366 * XXX We almost never run out of kernel virtual 367 * space, so we don't make the allocated memory 368 * above available. 369 */ 370 vm_map_unlock(map); 371 vm_map_entry_release(count); 372 crit_exit(); 373 return (0); 374 } 375 vm_object_reference(kernel_object); 376 vm_map_insert(map, &count, 377 kernel_object, addr - VM_MIN_KERNEL_ADDRESS, 378 addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0); 379 vm_map_unlock(map); 380 vm_map_entry_release(count); 381 382 tmp_addr = addr; 383 for (i = start; i < (start + size / PAGE_SIZE); i++) { 384 vm_page_t m = &pga[i]; 385 vm_page_insert(m, kernel_object, 386 OFF_TO_IDX(tmp_addr - VM_MIN_KERNEL_ADDRESS)); 387 if ((flags & M_ZERO) && !(m->flags & PG_ZERO)) 388 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 389 m->flags = 0; 390 tmp_addr += PAGE_SIZE; 391 } 392 vm_map_wire(map, addr, addr + size, 0); 393 394 crit_exit(); 395 return (addr); 396 } 397 398 void * 399 contigmalloc( 400 unsigned long size, /* should be size_t here and for malloc() */ 401 struct malloc_type *type, 402 int flags, 403 vm_paddr_t low, 404 vm_paddr_t high, 405 unsigned long alignment, 406 unsigned long boundary) 407 { 408 return contigmalloc_map(size, type, flags, low, high, alignment, 409 boundary, kernel_map); 410 } 411 412 void * 413 contigmalloc_map( 414 unsigned long size, /* should be size_t here and for malloc() */ 415 struct malloc_type *type, 416 int flags, 417 vm_paddr_t low, 418 vm_paddr_t high, 419 unsigned long alignment, 420 unsigned long boundary, 421 vm_map_t map) 422 { 423 int index; 424 void *rv; 425 426 index = vm_contig_pg_alloc(size, low, high, alignment, boundary); 427 if (index < 0) { 428 printf("contigmalloc_map: failed in index < 0 case!"); 429 return NULL; 430 } 431 432 rv = (void *) vm_contig_pg_kmap(index, size, map, flags); 433 if (!rv) 434 vm_contig_pg_free(index, size); 435 436 return rv; 437 } 438 439 void 440 contigfree(void *addr, unsigned long size, struct malloc_type *type) 441 { 442 kmem_free(kernel_map, (vm_offset_t)addr, size); 443 } 444 445 vm_offset_t 446 vm_page_alloc_contig( 447 vm_offset_t size, 448 vm_paddr_t low, 449 vm_paddr_t high, 450 vm_offset_t alignment) 451 { 452 return ((vm_offset_t)contigmalloc_map(size, M_DEVBUF, M_NOWAIT, low, 453 high, alignment, 0ul, kernel_map)); 454 } 455