1 /*- 2 * Copyright (c) 2002-2006 Rice University 3 * Copyright (c) 2007 Alan L. Cox <alc@cs.rice.edu> 4 * All rights reserved. 5 * 6 * This software was developed for the FreeBSD Project by Alan L. Cox, 7 * Olivier Crameri, Peter Druschel, Sitaram Iyer, and Juan Navarro. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 21 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 22 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 24 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS 25 * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 26 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY 28 * WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Physical memory system implementation 34 * 35 * Any external functions defined by this module are only to be used by the 36 * virtual memory system. 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_ddb.h" 43 #include "opt_vm.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/lock.h> 48 #include <sys/kernel.h> 49 #include <sys/malloc.h> 50 #include <sys/mutex.h> 51 #if MAXMEMDOM > 1 52 #include <sys/proc.h> 53 #endif 54 #include <sys/queue.h> 55 #include <sys/rwlock.h> 56 #include <sys/sbuf.h> 57 #include <sys/sysctl.h> 58 #include <sys/tree.h> 59 #include <sys/vmmeter.h> 60 61 #include <ddb/ddb.h> 62 63 #include <vm/vm.h> 64 #include <vm/vm_param.h> 65 #include <vm/vm_kern.h> 66 #include <vm/vm_object.h> 67 #include <vm/vm_page.h> 68 #include <vm/vm_phys.h> 69 70 _Static_assert(sizeof(long) * NBBY >= VM_PHYSSEG_MAX, 71 "Too many physsegs."); 72 73 struct mem_affinity *mem_affinity; 74 75 int vm_ndomains = 1; 76 77 struct vm_phys_seg vm_phys_segs[VM_PHYSSEG_MAX]; 78 int vm_phys_nsegs; 79 80 struct vm_phys_fictitious_seg; 81 static int vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *, 82 struct vm_phys_fictitious_seg *); 83 84 RB_HEAD(fict_tree, vm_phys_fictitious_seg) vm_phys_fictitious_tree = 85 RB_INITIALIZER(_vm_phys_fictitious_tree); 86 87 struct vm_phys_fictitious_seg { 88 RB_ENTRY(vm_phys_fictitious_seg) node; 89 /* Memory region data */ 90 vm_paddr_t start; 91 vm_paddr_t end; 92 vm_page_t first_page; 93 }; 94 95 RB_GENERATE_STATIC(fict_tree, vm_phys_fictitious_seg, node, 96 vm_phys_fictitious_cmp); 97 98 static struct rwlock vm_phys_fictitious_reg_lock; 99 MALLOC_DEFINE(M_FICT_PAGES, "vm_fictitious", "Fictitious VM pages"); 100 101 static struct vm_freelist 102 vm_phys_free_queues[MAXMEMDOM][VM_NFREELIST][VM_NFREEPOOL][VM_NFREEORDER]; 103 104 static int vm_nfreelists; 105 106 /* 107 * Provides the mapping from VM_FREELIST_* to free list indices (flind). 108 */ 109 static int vm_freelist_to_flind[VM_NFREELIST]; 110 111 CTASSERT(VM_FREELIST_DEFAULT == 0); 112 113 #ifdef VM_FREELIST_ISADMA 114 #define VM_ISADMA_BOUNDARY 16777216 115 #endif 116 #ifdef VM_FREELIST_DMA32 117 #define VM_DMA32_BOUNDARY ((vm_paddr_t)1 << 32) 118 #endif 119 120 /* 121 * Enforce the assumptions made by vm_phys_add_seg() and vm_phys_init() about 122 * the ordering of the free list boundaries. 123 */ 124 #if defined(VM_ISADMA_BOUNDARY) && defined(VM_LOWMEM_BOUNDARY) 125 CTASSERT(VM_ISADMA_BOUNDARY < VM_LOWMEM_BOUNDARY); 126 #endif 127 #if defined(VM_LOWMEM_BOUNDARY) && defined(VM_DMA32_BOUNDARY) 128 CTASSERT(VM_LOWMEM_BOUNDARY < VM_DMA32_BOUNDARY); 129 #endif 130 131 static int cnt_prezero; 132 SYSCTL_INT(_vm_stats_misc, OID_AUTO, cnt_prezero, CTLFLAG_RD, 133 &cnt_prezero, 0, "The number of physical pages prezeroed at idle time"); 134 135 static int sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS); 136 SYSCTL_OID(_vm, OID_AUTO, phys_free, CTLTYPE_STRING | CTLFLAG_RD, 137 NULL, 0, sysctl_vm_phys_free, "A", "Phys Free Info"); 138 139 static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS); 140 SYSCTL_OID(_vm, OID_AUTO, phys_segs, CTLTYPE_STRING | CTLFLAG_RD, 141 NULL, 0, sysctl_vm_phys_segs, "A", "Phys Seg Info"); 142 143 SYSCTL_INT(_vm, OID_AUTO, ndomains, CTLFLAG_RD, 144 &vm_ndomains, 0, "Number of physical memory domains available."); 145 146 static vm_page_t vm_phys_alloc_domain_pages(int domain, int flind, int pool, 147 int order); 148 static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain); 149 static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end); 150 static int vm_phys_paddr_to_segind(vm_paddr_t pa); 151 static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, 152 int order); 153 154 /* 155 * Red-black tree helpers for vm fictitious range management. 156 */ 157 static inline int 158 vm_phys_fictitious_in_range(struct vm_phys_fictitious_seg *p, 159 struct vm_phys_fictitious_seg *range) 160 { 161 162 KASSERT(range->start != 0 && range->end != 0, 163 ("Invalid range passed on search for vm_fictitious page")); 164 if (p->start >= range->end) 165 return (1); 166 if (p->start < range->start) 167 return (-1); 168 169 return (0); 170 } 171 172 static int 173 vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *p1, 174 struct vm_phys_fictitious_seg *p2) 175 { 176 177 /* Check if this is a search for a page */ 178 if (p1->end == 0) 179 return (vm_phys_fictitious_in_range(p1, p2)); 180 181 KASSERT(p2->end != 0, 182 ("Invalid range passed as second parameter to vm fictitious comparison")); 183 184 /* Searching to add a new range */ 185 if (p1->end <= p2->start) 186 return (-1); 187 if (p1->start >= p2->end) 188 return (1); 189 190 panic("Trying to add overlapping vm fictitious ranges:\n" 191 "[%#jx:%#jx] and [%#jx:%#jx]", (uintmax_t)p1->start, 192 (uintmax_t)p1->end, (uintmax_t)p2->start, (uintmax_t)p2->end); 193 } 194 195 static __inline int 196 vm_rr_selectdomain(void) 197 { 198 #if MAXMEMDOM > 1 199 struct thread *td; 200 201 td = curthread; 202 203 td->td_dom_rr_idx++; 204 td->td_dom_rr_idx %= vm_ndomains; 205 return (td->td_dom_rr_idx); 206 #else 207 return (0); 208 #endif 209 } 210 211 boolean_t 212 vm_phys_domain_intersects(long mask, vm_paddr_t low, vm_paddr_t high) 213 { 214 struct vm_phys_seg *s; 215 int idx; 216 217 while ((idx = ffsl(mask)) != 0) { 218 idx--; /* ffsl counts from 1 */ 219 mask &= ~(1UL << idx); 220 s = &vm_phys_segs[idx]; 221 if (low < s->end && high > s->start) 222 return (TRUE); 223 } 224 return (FALSE); 225 } 226 227 /* 228 * Outputs the state of the physical memory allocator, specifically, 229 * the amount of physical memory in each free list. 230 */ 231 static int 232 sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS) 233 { 234 struct sbuf sbuf; 235 struct vm_freelist *fl; 236 int dom, error, flind, oind, pind; 237 238 error = sysctl_wire_old_buffer(req, 0); 239 if (error != 0) 240 return (error); 241 sbuf_new_for_sysctl(&sbuf, NULL, 128 * vm_ndomains, req); 242 for (dom = 0; dom < vm_ndomains; dom++) { 243 sbuf_printf(&sbuf,"\nDOMAIN %d:\n", dom); 244 for (flind = 0; flind < vm_nfreelists; flind++) { 245 sbuf_printf(&sbuf, "\nFREE LIST %d:\n" 246 "\n ORDER (SIZE) | NUMBER" 247 "\n ", flind); 248 for (pind = 0; pind < VM_NFREEPOOL; pind++) 249 sbuf_printf(&sbuf, " | POOL %d", pind); 250 sbuf_printf(&sbuf, "\n-- "); 251 for (pind = 0; pind < VM_NFREEPOOL; pind++) 252 sbuf_printf(&sbuf, "-- -- "); 253 sbuf_printf(&sbuf, "--\n"); 254 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) { 255 sbuf_printf(&sbuf, " %2d (%6dK)", oind, 256 1 << (PAGE_SHIFT - 10 + oind)); 257 for (pind = 0; pind < VM_NFREEPOOL; pind++) { 258 fl = vm_phys_free_queues[dom][flind][pind]; 259 sbuf_printf(&sbuf, " | %6d", 260 fl[oind].lcnt); 261 } 262 sbuf_printf(&sbuf, "\n"); 263 } 264 } 265 } 266 error = sbuf_finish(&sbuf); 267 sbuf_delete(&sbuf); 268 return (error); 269 } 270 271 /* 272 * Outputs the set of physical memory segments. 273 */ 274 static int 275 sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS) 276 { 277 struct sbuf sbuf; 278 struct vm_phys_seg *seg; 279 int error, segind; 280 281 error = sysctl_wire_old_buffer(req, 0); 282 if (error != 0) 283 return (error); 284 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 285 for (segind = 0; segind < vm_phys_nsegs; segind++) { 286 sbuf_printf(&sbuf, "\nSEGMENT %d:\n\n", segind); 287 seg = &vm_phys_segs[segind]; 288 sbuf_printf(&sbuf, "start: %#jx\n", 289 (uintmax_t)seg->start); 290 sbuf_printf(&sbuf, "end: %#jx\n", 291 (uintmax_t)seg->end); 292 sbuf_printf(&sbuf, "domain: %d\n", seg->domain); 293 sbuf_printf(&sbuf, "free list: %p\n", seg->free_queues); 294 } 295 error = sbuf_finish(&sbuf); 296 sbuf_delete(&sbuf); 297 return (error); 298 } 299 300 static void 301 vm_freelist_add(struct vm_freelist *fl, vm_page_t m, int order, int tail) 302 { 303 304 m->order = order; 305 if (tail) 306 TAILQ_INSERT_TAIL(&fl[order].pl, m, plinks.q); 307 else 308 TAILQ_INSERT_HEAD(&fl[order].pl, m, plinks.q); 309 fl[order].lcnt++; 310 } 311 312 static void 313 vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order) 314 { 315 316 TAILQ_REMOVE(&fl[order].pl, m, plinks.q); 317 fl[order].lcnt--; 318 m->order = VM_NFREEORDER; 319 } 320 321 /* 322 * Create a physical memory segment. 323 */ 324 static void 325 _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain) 326 { 327 struct vm_phys_seg *seg; 328 329 KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX, 330 ("vm_phys_create_seg: increase VM_PHYSSEG_MAX")); 331 KASSERT(domain < vm_ndomains, 332 ("vm_phys_create_seg: invalid domain provided")); 333 seg = &vm_phys_segs[vm_phys_nsegs++]; 334 while (seg > vm_phys_segs && (seg - 1)->start >= end) { 335 *seg = *(seg - 1); 336 seg--; 337 } 338 seg->start = start; 339 seg->end = end; 340 seg->domain = domain; 341 } 342 343 static void 344 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end) 345 { 346 int i; 347 348 if (mem_affinity == NULL) { 349 _vm_phys_create_seg(start, end, 0); 350 return; 351 } 352 353 for (i = 0;; i++) { 354 if (mem_affinity[i].end == 0) 355 panic("Reached end of affinity info"); 356 if (mem_affinity[i].end <= start) 357 continue; 358 if (mem_affinity[i].start > start) 359 panic("No affinity info for start %jx", 360 (uintmax_t)start); 361 if (mem_affinity[i].end >= end) { 362 _vm_phys_create_seg(start, end, 363 mem_affinity[i].domain); 364 break; 365 } 366 _vm_phys_create_seg(start, mem_affinity[i].end, 367 mem_affinity[i].domain); 368 start = mem_affinity[i].end; 369 } 370 } 371 372 /* 373 * Add a physical memory segment. 374 */ 375 void 376 vm_phys_add_seg(vm_paddr_t start, vm_paddr_t end) 377 { 378 vm_paddr_t paddr; 379 380 KASSERT((start & PAGE_MASK) == 0, 381 ("vm_phys_define_seg: start is not page aligned")); 382 KASSERT((end & PAGE_MASK) == 0, 383 ("vm_phys_define_seg: end is not page aligned")); 384 385 /* 386 * Split the physical memory segment if it spans two or more free 387 * list boundaries. 388 */ 389 paddr = start; 390 #ifdef VM_FREELIST_ISADMA 391 if (paddr < VM_ISADMA_BOUNDARY && end > VM_ISADMA_BOUNDARY) { 392 vm_phys_create_seg(paddr, VM_ISADMA_BOUNDARY); 393 paddr = VM_ISADMA_BOUNDARY; 394 } 395 #endif 396 #ifdef VM_FREELIST_LOWMEM 397 if (paddr < VM_LOWMEM_BOUNDARY && end > VM_LOWMEM_BOUNDARY) { 398 vm_phys_create_seg(paddr, VM_LOWMEM_BOUNDARY); 399 paddr = VM_LOWMEM_BOUNDARY; 400 } 401 #endif 402 #ifdef VM_FREELIST_DMA32 403 if (paddr < VM_DMA32_BOUNDARY && end > VM_DMA32_BOUNDARY) { 404 vm_phys_create_seg(paddr, VM_DMA32_BOUNDARY); 405 paddr = VM_DMA32_BOUNDARY; 406 } 407 #endif 408 vm_phys_create_seg(paddr, end); 409 } 410 411 /* 412 * Initialize the physical memory allocator. 413 * 414 * Requires that vm_page_array is initialized! 415 */ 416 void 417 vm_phys_init(void) 418 { 419 struct vm_freelist *fl; 420 struct vm_phys_seg *seg; 421 u_long npages; 422 int dom, flind, freelist, oind, pind, segind; 423 424 /* 425 * Compute the number of free lists, and generate the mapping from the 426 * manifest constants VM_FREELIST_* to the free list indices. 427 * 428 * Initially, the entries of vm_freelist_to_flind[] are set to either 429 * 0 or 1 to indicate which free lists should be created. 430 */ 431 npages = 0; 432 for (segind = vm_phys_nsegs - 1; segind >= 0; segind--) { 433 seg = &vm_phys_segs[segind]; 434 #ifdef VM_FREELIST_ISADMA 435 if (seg->end <= VM_ISADMA_BOUNDARY) 436 vm_freelist_to_flind[VM_FREELIST_ISADMA] = 1; 437 else 438 #endif 439 #ifdef VM_FREELIST_LOWMEM 440 if (seg->end <= VM_LOWMEM_BOUNDARY) 441 vm_freelist_to_flind[VM_FREELIST_LOWMEM] = 1; 442 else 443 #endif 444 #ifdef VM_FREELIST_DMA32 445 if ( 446 #ifdef VM_DMA32_NPAGES_THRESHOLD 447 /* 448 * Create the DMA32 free list only if the amount of 449 * physical memory above physical address 4G exceeds the 450 * given threshold. 451 */ 452 npages > VM_DMA32_NPAGES_THRESHOLD && 453 #endif 454 seg->end <= VM_DMA32_BOUNDARY) 455 vm_freelist_to_flind[VM_FREELIST_DMA32] = 1; 456 else 457 #endif 458 { 459 npages += atop(seg->end - seg->start); 460 vm_freelist_to_flind[VM_FREELIST_DEFAULT] = 1; 461 } 462 } 463 /* Change each entry into a running total of the free lists. */ 464 for (freelist = 1; freelist < VM_NFREELIST; freelist++) { 465 vm_freelist_to_flind[freelist] += 466 vm_freelist_to_flind[freelist - 1]; 467 } 468 vm_nfreelists = vm_freelist_to_flind[VM_NFREELIST - 1]; 469 KASSERT(vm_nfreelists > 0, ("vm_phys_init: no free lists")); 470 /* Change each entry into a free list index. */ 471 for (freelist = 0; freelist < VM_NFREELIST; freelist++) 472 vm_freelist_to_flind[freelist]--; 473 474 /* 475 * Initialize the first_page and free_queues fields of each physical 476 * memory segment. 477 */ 478 #ifdef VM_PHYSSEG_SPARSE 479 npages = 0; 480 #endif 481 for (segind = 0; segind < vm_phys_nsegs; segind++) { 482 seg = &vm_phys_segs[segind]; 483 #ifdef VM_PHYSSEG_SPARSE 484 seg->first_page = &vm_page_array[npages]; 485 npages += atop(seg->end - seg->start); 486 #else 487 seg->first_page = PHYS_TO_VM_PAGE(seg->start); 488 #endif 489 #ifdef VM_FREELIST_ISADMA 490 if (seg->end <= VM_ISADMA_BOUNDARY) { 491 flind = vm_freelist_to_flind[VM_FREELIST_ISADMA]; 492 KASSERT(flind >= 0, 493 ("vm_phys_init: ISADMA flind < 0")); 494 } else 495 #endif 496 #ifdef VM_FREELIST_LOWMEM 497 if (seg->end <= VM_LOWMEM_BOUNDARY) { 498 flind = vm_freelist_to_flind[VM_FREELIST_LOWMEM]; 499 KASSERT(flind >= 0, 500 ("vm_phys_init: LOWMEM flind < 0")); 501 } else 502 #endif 503 #ifdef VM_FREELIST_DMA32 504 if (seg->end <= VM_DMA32_BOUNDARY) { 505 flind = vm_freelist_to_flind[VM_FREELIST_DMA32]; 506 KASSERT(flind >= 0, 507 ("vm_phys_init: DMA32 flind < 0")); 508 } else 509 #endif 510 { 511 flind = vm_freelist_to_flind[VM_FREELIST_DEFAULT]; 512 KASSERT(flind >= 0, 513 ("vm_phys_init: DEFAULT flind < 0")); 514 } 515 seg->free_queues = &vm_phys_free_queues[seg->domain][flind]; 516 } 517 518 /* 519 * Initialize the free queues. 520 */ 521 for (dom = 0; dom < vm_ndomains; dom++) { 522 for (flind = 0; flind < vm_nfreelists; flind++) { 523 for (pind = 0; pind < VM_NFREEPOOL; pind++) { 524 fl = vm_phys_free_queues[dom][flind][pind]; 525 for (oind = 0; oind < VM_NFREEORDER; oind++) 526 TAILQ_INIT(&fl[oind].pl); 527 } 528 } 529 } 530 531 rw_init(&vm_phys_fictitious_reg_lock, "vmfctr"); 532 } 533 534 /* 535 * Split a contiguous, power of two-sized set of physical pages. 536 */ 537 static __inline void 538 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order) 539 { 540 vm_page_t m_buddy; 541 542 while (oind > order) { 543 oind--; 544 m_buddy = &m[1 << oind]; 545 KASSERT(m_buddy->order == VM_NFREEORDER, 546 ("vm_phys_split_pages: page %p has unexpected order %d", 547 m_buddy, m_buddy->order)); 548 vm_freelist_add(fl, m_buddy, oind, 0); 549 } 550 } 551 552 /* 553 * Initialize a physical page and add it to the free lists. 554 */ 555 void 556 vm_phys_add_page(vm_paddr_t pa) 557 { 558 vm_page_t m; 559 struct vm_domain *vmd; 560 561 vm_cnt.v_page_count++; 562 m = vm_phys_paddr_to_vm_page(pa); 563 m->phys_addr = pa; 564 m->queue = PQ_NONE; 565 m->segind = vm_phys_paddr_to_segind(pa); 566 vmd = vm_phys_domain(m); 567 vmd->vmd_page_count++; 568 vmd->vmd_segs |= 1UL << m->segind; 569 KASSERT(m->order == VM_NFREEORDER, 570 ("vm_phys_add_page: page %p has unexpected order %d", 571 m, m->order)); 572 m->pool = VM_FREEPOOL_DEFAULT; 573 pmap_page_init(m); 574 mtx_lock(&vm_page_queue_free_mtx); 575 vm_phys_freecnt_adj(m, 1); 576 vm_phys_free_pages(m, 0); 577 mtx_unlock(&vm_page_queue_free_mtx); 578 } 579 580 /* 581 * Allocate a contiguous, power of two-sized set of physical pages 582 * from the free lists. 583 * 584 * The free page queues must be locked. 585 */ 586 vm_page_t 587 vm_phys_alloc_pages(int pool, int order) 588 { 589 vm_page_t m; 590 int dom, domain, flind; 591 592 KASSERT(pool < VM_NFREEPOOL, 593 ("vm_phys_alloc_pages: pool %d is out of range", pool)); 594 KASSERT(order < VM_NFREEORDER, 595 ("vm_phys_alloc_pages: order %d is out of range", order)); 596 597 for (dom = 0; dom < vm_ndomains; dom++) { 598 domain = vm_rr_selectdomain(); 599 for (flind = 0; flind < vm_nfreelists; flind++) { 600 m = vm_phys_alloc_domain_pages(domain, flind, pool, 601 order); 602 if (m != NULL) 603 return (m); 604 } 605 } 606 return (NULL); 607 } 608 609 /* 610 * Allocate a contiguous, power of two-sized set of physical pages from the 611 * specified free list. The free list must be specified using one of the 612 * manifest constants VM_FREELIST_*. 613 * 614 * The free page queues must be locked. 615 */ 616 vm_page_t 617 vm_phys_alloc_freelist_pages(int freelist, int pool, int order) 618 { 619 vm_page_t m; 620 int dom, domain; 621 622 KASSERT(freelist < VM_NFREELIST, 623 ("vm_phys_alloc_freelist_pages: freelist %d is out of range", 624 freelist)); 625 KASSERT(pool < VM_NFREEPOOL, 626 ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool)); 627 KASSERT(order < VM_NFREEORDER, 628 ("vm_phys_alloc_freelist_pages: order %d is out of range", order)); 629 for (dom = 0; dom < vm_ndomains; dom++) { 630 domain = vm_rr_selectdomain(); 631 m = vm_phys_alloc_domain_pages(domain, 632 vm_freelist_to_flind[freelist], pool, order); 633 if (m != NULL) 634 return (m); 635 } 636 return (NULL); 637 } 638 639 static vm_page_t 640 vm_phys_alloc_domain_pages(int domain, int flind, int pool, int order) 641 { 642 struct vm_freelist *fl; 643 struct vm_freelist *alt; 644 int oind, pind; 645 vm_page_t m; 646 647 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 648 fl = &vm_phys_free_queues[domain][flind][pool][0]; 649 for (oind = order; oind < VM_NFREEORDER; oind++) { 650 m = TAILQ_FIRST(&fl[oind].pl); 651 if (m != NULL) { 652 vm_freelist_rem(fl, m, oind); 653 vm_phys_split_pages(m, oind, fl, order); 654 return (m); 655 } 656 } 657 658 /* 659 * The given pool was empty. Find the largest 660 * contiguous, power-of-two-sized set of pages in any 661 * pool. Transfer these pages to the given pool, and 662 * use them to satisfy the allocation. 663 */ 664 for (oind = VM_NFREEORDER - 1; oind >= order; oind--) { 665 for (pind = 0; pind < VM_NFREEPOOL; pind++) { 666 alt = &vm_phys_free_queues[domain][flind][pind][0]; 667 m = TAILQ_FIRST(&alt[oind].pl); 668 if (m != NULL) { 669 vm_freelist_rem(alt, m, oind); 670 vm_phys_set_pool(pool, m, oind); 671 vm_phys_split_pages(m, oind, fl, order); 672 return (m); 673 } 674 } 675 } 676 return (NULL); 677 } 678 679 /* 680 * Find the vm_page corresponding to the given physical address. 681 */ 682 vm_page_t 683 vm_phys_paddr_to_vm_page(vm_paddr_t pa) 684 { 685 struct vm_phys_seg *seg; 686 int segind; 687 688 for (segind = 0; segind < vm_phys_nsegs; segind++) { 689 seg = &vm_phys_segs[segind]; 690 if (pa >= seg->start && pa < seg->end) 691 return (&seg->first_page[atop(pa - seg->start)]); 692 } 693 return (NULL); 694 } 695 696 vm_page_t 697 vm_phys_fictitious_to_vm_page(vm_paddr_t pa) 698 { 699 struct vm_phys_fictitious_seg tmp, *seg; 700 vm_page_t m; 701 702 m = NULL; 703 tmp.start = pa; 704 tmp.end = 0; 705 706 rw_rlock(&vm_phys_fictitious_reg_lock); 707 seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp); 708 rw_runlock(&vm_phys_fictitious_reg_lock); 709 if (seg == NULL) 710 return (NULL); 711 712 m = &seg->first_page[atop(pa - seg->start)]; 713 KASSERT((m->flags & PG_FICTITIOUS) != 0, ("%p not fictitious", m)); 714 715 return (m); 716 } 717 718 static inline void 719 vm_phys_fictitious_init_range(vm_page_t range, vm_paddr_t start, 720 long page_count, vm_memattr_t memattr) 721 { 722 long i; 723 724 for (i = 0; i < page_count; i++) { 725 vm_page_initfake(&range[i], start + PAGE_SIZE * i, memattr); 726 range[i].oflags &= ~VPO_UNMANAGED; 727 range[i].busy_lock = VPB_UNBUSIED; 728 } 729 } 730 731 int 732 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end, 733 vm_memattr_t memattr) 734 { 735 struct vm_phys_fictitious_seg *seg; 736 vm_page_t fp; 737 long page_count; 738 #ifdef VM_PHYSSEG_DENSE 739 long pi, pe; 740 long dpage_count; 741 #endif 742 743 KASSERT(start < end, 744 ("Start of segment isn't less than end (start: %jx end: %jx)", 745 (uintmax_t)start, (uintmax_t)end)); 746 747 page_count = (end - start) / PAGE_SIZE; 748 749 #ifdef VM_PHYSSEG_DENSE 750 pi = atop(start); 751 pe = atop(end); 752 if (pi >= first_page && (pi - first_page) < vm_page_array_size) { 753 fp = &vm_page_array[pi - first_page]; 754 if ((pe - first_page) > vm_page_array_size) { 755 /* 756 * We have a segment that starts inside 757 * of vm_page_array, but ends outside of it. 758 * 759 * Use vm_page_array pages for those that are 760 * inside of the vm_page_array range, and 761 * allocate the remaining ones. 762 */ 763 dpage_count = vm_page_array_size - (pi - first_page); 764 vm_phys_fictitious_init_range(fp, start, dpage_count, 765 memattr); 766 page_count -= dpage_count; 767 start += ptoa(dpage_count); 768 goto alloc; 769 } 770 /* 771 * We can allocate the full range from vm_page_array, 772 * so there's no need to register the range in the tree. 773 */ 774 vm_phys_fictitious_init_range(fp, start, page_count, memattr); 775 return (0); 776 } else if (pe > first_page && (pe - first_page) < vm_page_array_size) { 777 /* 778 * We have a segment that ends inside of vm_page_array, 779 * but starts outside of it. 780 */ 781 fp = &vm_page_array[0]; 782 dpage_count = pe - first_page; 783 vm_phys_fictitious_init_range(fp, ptoa(first_page), dpage_count, 784 memattr); 785 end -= ptoa(dpage_count); 786 page_count -= dpage_count; 787 goto alloc; 788 } else if (pi < first_page && pe > (first_page + vm_page_array_size)) { 789 /* 790 * Trying to register a fictitious range that expands before 791 * and after vm_page_array. 792 */ 793 return (EINVAL); 794 } else { 795 alloc: 796 #endif 797 fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES, 798 M_WAITOK | M_ZERO); 799 #ifdef VM_PHYSSEG_DENSE 800 } 801 #endif 802 vm_phys_fictitious_init_range(fp, start, page_count, memattr); 803 804 seg = malloc(sizeof(*seg), M_FICT_PAGES, M_WAITOK | M_ZERO); 805 seg->start = start; 806 seg->end = end; 807 seg->first_page = fp; 808 809 rw_wlock(&vm_phys_fictitious_reg_lock); 810 RB_INSERT(fict_tree, &vm_phys_fictitious_tree, seg); 811 rw_wunlock(&vm_phys_fictitious_reg_lock); 812 813 return (0); 814 } 815 816 void 817 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end) 818 { 819 struct vm_phys_fictitious_seg *seg, tmp; 820 #ifdef VM_PHYSSEG_DENSE 821 long pi, pe; 822 #endif 823 824 KASSERT(start < end, 825 ("Start of segment isn't less than end (start: %jx end: %jx)", 826 (uintmax_t)start, (uintmax_t)end)); 827 828 #ifdef VM_PHYSSEG_DENSE 829 pi = atop(start); 830 pe = atop(end); 831 if (pi >= first_page && (pi - first_page) < vm_page_array_size) { 832 if ((pe - first_page) <= vm_page_array_size) { 833 /* 834 * This segment was allocated using vm_page_array 835 * only, there's nothing to do since those pages 836 * were never added to the tree. 837 */ 838 return; 839 } 840 /* 841 * We have a segment that starts inside 842 * of vm_page_array, but ends outside of it. 843 * 844 * Calculate how many pages were added to the 845 * tree and free them. 846 */ 847 start = ptoa(first_page + vm_page_array_size); 848 } else if (pe > first_page && (pe - first_page) < vm_page_array_size) { 849 /* 850 * We have a segment that ends inside of vm_page_array, 851 * but starts outside of it. 852 */ 853 end = ptoa(first_page); 854 } else if (pi < first_page && pe > (first_page + vm_page_array_size)) { 855 /* Since it's not possible to register such a range, panic. */ 856 panic( 857 "Unregistering not registered fictitious range [%#jx:%#jx]", 858 (uintmax_t)start, (uintmax_t)end); 859 } 860 #endif 861 tmp.start = start; 862 tmp.end = 0; 863 864 rw_wlock(&vm_phys_fictitious_reg_lock); 865 seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp); 866 if (seg->start != start || seg->end != end) { 867 rw_wunlock(&vm_phys_fictitious_reg_lock); 868 panic( 869 "Unregistering not registered fictitious range [%#jx:%#jx]", 870 (uintmax_t)start, (uintmax_t)end); 871 } 872 RB_REMOVE(fict_tree, &vm_phys_fictitious_tree, seg); 873 rw_wunlock(&vm_phys_fictitious_reg_lock); 874 free(seg->first_page, M_FICT_PAGES); 875 free(seg, M_FICT_PAGES); 876 } 877 878 /* 879 * Find the segment containing the given physical address. 880 */ 881 static int 882 vm_phys_paddr_to_segind(vm_paddr_t pa) 883 { 884 struct vm_phys_seg *seg; 885 int segind; 886 887 for (segind = 0; segind < vm_phys_nsegs; segind++) { 888 seg = &vm_phys_segs[segind]; 889 if (pa >= seg->start && pa < seg->end) 890 return (segind); 891 } 892 panic("vm_phys_paddr_to_segind: paddr %#jx is not in any segment" , 893 (uintmax_t)pa); 894 } 895 896 /* 897 * Free a contiguous, power of two-sized set of physical pages. 898 * 899 * The free page queues must be locked. 900 */ 901 void 902 vm_phys_free_pages(vm_page_t m, int order) 903 { 904 struct vm_freelist *fl; 905 struct vm_phys_seg *seg; 906 vm_paddr_t pa; 907 vm_page_t m_buddy; 908 909 KASSERT(m->order == VM_NFREEORDER, 910 ("vm_phys_free_pages: page %p has unexpected order %d", 911 m, m->order)); 912 KASSERT(m->pool < VM_NFREEPOOL, 913 ("vm_phys_free_pages: page %p has unexpected pool %d", 914 m, m->pool)); 915 KASSERT(order < VM_NFREEORDER, 916 ("vm_phys_free_pages: order %d is out of range", order)); 917 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 918 seg = &vm_phys_segs[m->segind]; 919 if (order < VM_NFREEORDER - 1) { 920 pa = VM_PAGE_TO_PHYS(m); 921 do { 922 pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order)); 923 if (pa < seg->start || pa >= seg->end) 924 break; 925 m_buddy = &seg->first_page[atop(pa - seg->start)]; 926 if (m_buddy->order != order) 927 break; 928 fl = (*seg->free_queues)[m_buddy->pool]; 929 vm_freelist_rem(fl, m_buddy, order); 930 if (m_buddy->pool != m->pool) 931 vm_phys_set_pool(m->pool, m_buddy, order); 932 order++; 933 pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1); 934 m = &seg->first_page[atop(pa - seg->start)]; 935 } while (order < VM_NFREEORDER - 1); 936 } 937 fl = (*seg->free_queues)[m->pool]; 938 vm_freelist_add(fl, m, order, 1); 939 } 940 941 /* 942 * Free a contiguous, arbitrarily sized set of physical pages. 943 * 944 * The free page queues must be locked. 945 */ 946 void 947 vm_phys_free_contig(vm_page_t m, u_long npages) 948 { 949 u_int n; 950 int order; 951 952 /* 953 * Avoid unnecessary coalescing by freeing the pages in the largest 954 * possible power-of-two-sized subsets. 955 */ 956 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 957 for (;; npages -= n) { 958 /* 959 * Unsigned "min" is used here so that "order" is assigned 960 * "VM_NFREEORDER - 1" when "m"'s physical address is zero 961 * or the low-order bits of its physical address are zero 962 * because the size of a physical address exceeds the size of 963 * a long. 964 */ 965 order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1, 966 VM_NFREEORDER - 1); 967 n = 1 << order; 968 if (npages < n) 969 break; 970 vm_phys_free_pages(m, order); 971 m += n; 972 } 973 /* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */ 974 for (; npages > 0; npages -= n) { 975 order = flsl(npages) - 1; 976 n = 1 << order; 977 vm_phys_free_pages(m, order); 978 m += n; 979 } 980 } 981 982 /* 983 * Set the pool for a contiguous, power of two-sized set of physical pages. 984 */ 985 void 986 vm_phys_set_pool(int pool, vm_page_t m, int order) 987 { 988 vm_page_t m_tmp; 989 990 for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++) 991 m_tmp->pool = pool; 992 } 993 994 /* 995 * Search for the given physical page "m" in the free lists. If the search 996 * succeeds, remove "m" from the free lists and return TRUE. Otherwise, return 997 * FALSE, indicating that "m" is not in the free lists. 998 * 999 * The free page queues must be locked. 1000 */ 1001 boolean_t 1002 vm_phys_unfree_page(vm_page_t m) 1003 { 1004 struct vm_freelist *fl; 1005 struct vm_phys_seg *seg; 1006 vm_paddr_t pa, pa_half; 1007 vm_page_t m_set, m_tmp; 1008 int order; 1009 1010 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 1011 1012 /* 1013 * First, find the contiguous, power of two-sized set of free 1014 * physical pages containing the given physical page "m" and 1015 * assign it to "m_set". 1016 */ 1017 seg = &vm_phys_segs[m->segind]; 1018 for (m_set = m, order = 0; m_set->order == VM_NFREEORDER && 1019 order < VM_NFREEORDER - 1; ) { 1020 order++; 1021 pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order)); 1022 if (pa >= seg->start) 1023 m_set = &seg->first_page[atop(pa - seg->start)]; 1024 else 1025 return (FALSE); 1026 } 1027 if (m_set->order < order) 1028 return (FALSE); 1029 if (m_set->order == VM_NFREEORDER) 1030 return (FALSE); 1031 KASSERT(m_set->order < VM_NFREEORDER, 1032 ("vm_phys_unfree_page: page %p has unexpected order %d", 1033 m_set, m_set->order)); 1034 1035 /* 1036 * Next, remove "m_set" from the free lists. Finally, extract 1037 * "m" from "m_set" using an iterative algorithm: While "m_set" 1038 * is larger than a page, shrink "m_set" by returning the half 1039 * of "m_set" that does not contain "m" to the free lists. 1040 */ 1041 fl = (*seg->free_queues)[m_set->pool]; 1042 order = m_set->order; 1043 vm_freelist_rem(fl, m_set, order); 1044 while (order > 0) { 1045 order--; 1046 pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order)); 1047 if (m->phys_addr < pa_half) 1048 m_tmp = &seg->first_page[atop(pa_half - seg->start)]; 1049 else { 1050 m_tmp = m_set; 1051 m_set = &seg->first_page[atop(pa_half - seg->start)]; 1052 } 1053 vm_freelist_add(fl, m_tmp, order, 0); 1054 } 1055 KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency")); 1056 return (TRUE); 1057 } 1058 1059 /* 1060 * Try to zero one physical page. Used by an idle priority thread. 1061 */ 1062 boolean_t 1063 vm_phys_zero_pages_idle(void) 1064 { 1065 static struct vm_freelist *fl; 1066 static int flind, oind, pind; 1067 vm_page_t m, m_tmp; 1068 int domain; 1069 1070 domain = vm_rr_selectdomain(); 1071 fl = vm_phys_free_queues[domain][0][0]; 1072 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 1073 for (;;) { 1074 TAILQ_FOREACH_REVERSE(m, &fl[oind].pl, pglist, plinks.q) { 1075 for (m_tmp = m; m_tmp < &m[1 << oind]; m_tmp++) { 1076 if ((m_tmp->flags & (PG_CACHED | PG_ZERO)) == 0) { 1077 vm_phys_unfree_page(m_tmp); 1078 vm_phys_freecnt_adj(m, -1); 1079 mtx_unlock(&vm_page_queue_free_mtx); 1080 pmap_zero_page_idle(m_tmp); 1081 m_tmp->flags |= PG_ZERO; 1082 mtx_lock(&vm_page_queue_free_mtx); 1083 vm_phys_freecnt_adj(m, 1); 1084 vm_phys_free_pages(m_tmp, 0); 1085 vm_page_zero_count++; 1086 cnt_prezero++; 1087 return (TRUE); 1088 } 1089 } 1090 } 1091 oind++; 1092 if (oind == VM_NFREEORDER) { 1093 oind = 0; 1094 pind++; 1095 if (pind == VM_NFREEPOOL) { 1096 pind = 0; 1097 flind++; 1098 if (flind == vm_nfreelists) 1099 flind = 0; 1100 } 1101 fl = vm_phys_free_queues[domain][flind][pind]; 1102 } 1103 } 1104 } 1105 1106 /* 1107 * Allocate a contiguous set of physical pages of the given size 1108 * "npages" from the free lists. All of the physical pages must be at 1109 * or above the given physical address "low" and below the given 1110 * physical address "high". The given value "alignment" determines the 1111 * alignment of the first physical page in the set. If the given value 1112 * "boundary" is non-zero, then the set of physical pages cannot cross 1113 * any physical address boundary that is a multiple of that value. Both 1114 * "alignment" and "boundary" must be a power of two. 1115 */ 1116 vm_page_t 1117 vm_phys_alloc_contig(u_long npages, vm_paddr_t low, vm_paddr_t high, 1118 u_long alignment, vm_paddr_t boundary) 1119 { 1120 struct vm_freelist *fl; 1121 struct vm_phys_seg *seg; 1122 vm_paddr_t pa, pa_last, size; 1123 vm_page_t m, m_ret; 1124 u_long npages_end; 1125 int dom, domain, flind, oind, order, pind; 1126 1127 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 1128 size = npages << PAGE_SHIFT; 1129 KASSERT(size != 0, 1130 ("vm_phys_alloc_contig: size must not be 0")); 1131 KASSERT((alignment & (alignment - 1)) == 0, 1132 ("vm_phys_alloc_contig: alignment must be a power of 2")); 1133 KASSERT((boundary & (boundary - 1)) == 0, 1134 ("vm_phys_alloc_contig: boundary must be a power of 2")); 1135 /* Compute the queue that is the best fit for npages. */ 1136 for (order = 0; (1 << order) < npages; order++); 1137 dom = 0; 1138 restartdom: 1139 domain = vm_rr_selectdomain(); 1140 for (flind = 0; flind < vm_nfreelists; flind++) { 1141 for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER; oind++) { 1142 for (pind = 0; pind < VM_NFREEPOOL; pind++) { 1143 fl = &vm_phys_free_queues[domain][flind][pind][0]; 1144 TAILQ_FOREACH(m_ret, &fl[oind].pl, plinks.q) { 1145 /* 1146 * A free list may contain physical pages 1147 * from one or more segments. 1148 */ 1149 seg = &vm_phys_segs[m_ret->segind]; 1150 if (seg->start > high || 1151 low >= seg->end) 1152 continue; 1153 1154 /* 1155 * Is the size of this allocation request 1156 * larger than the largest block size? 1157 */ 1158 if (order >= VM_NFREEORDER) { 1159 /* 1160 * Determine if a sufficient number 1161 * of subsequent blocks to satisfy 1162 * the allocation request are free. 1163 */ 1164 pa = VM_PAGE_TO_PHYS(m_ret); 1165 pa_last = pa + size; 1166 for (;;) { 1167 pa += 1 << (PAGE_SHIFT + VM_NFREEORDER - 1); 1168 if (pa >= pa_last) 1169 break; 1170 if (pa < seg->start || 1171 pa >= seg->end) 1172 break; 1173 m = &seg->first_page[atop(pa - seg->start)]; 1174 if (m->order != VM_NFREEORDER - 1) 1175 break; 1176 } 1177 /* If not, continue to the next block. */ 1178 if (pa < pa_last) 1179 continue; 1180 } 1181 1182 /* 1183 * Determine if the blocks are within the given range, 1184 * satisfy the given alignment, and do not cross the 1185 * given boundary. 1186 */ 1187 pa = VM_PAGE_TO_PHYS(m_ret); 1188 if (pa >= low && 1189 pa + size <= high && 1190 (pa & (alignment - 1)) == 0 && 1191 ((pa ^ (pa + size - 1)) & ~(boundary - 1)) == 0) 1192 goto done; 1193 } 1194 } 1195 } 1196 } 1197 if (++dom < vm_ndomains) 1198 goto restartdom; 1199 return (NULL); 1200 done: 1201 for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) { 1202 fl = (*seg->free_queues)[m->pool]; 1203 vm_freelist_rem(fl, m, m->order); 1204 } 1205 if (m_ret->pool != VM_FREEPOOL_DEFAULT) 1206 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m_ret, oind); 1207 fl = (*seg->free_queues)[m_ret->pool]; 1208 vm_phys_split_pages(m_ret, oind, fl, order); 1209 /* Return excess pages to the free lists. */ 1210 npages_end = roundup2(npages, 1 << imin(oind, order)); 1211 if (npages < npages_end) 1212 vm_phys_free_contig(&m_ret[npages], npages_end - npages); 1213 return (m_ret); 1214 } 1215 1216 #ifdef DDB 1217 /* 1218 * Show the number of physical pages in each of the free lists. 1219 */ 1220 DB_SHOW_COMMAND(freepages, db_show_freepages) 1221 { 1222 struct vm_freelist *fl; 1223 int flind, oind, pind, dom; 1224 1225 for (dom = 0; dom < vm_ndomains; dom++) { 1226 db_printf("DOMAIN: %d\n", dom); 1227 for (flind = 0; flind < vm_nfreelists; flind++) { 1228 db_printf("FREE LIST %d:\n" 1229 "\n ORDER (SIZE) | NUMBER" 1230 "\n ", flind); 1231 for (pind = 0; pind < VM_NFREEPOOL; pind++) 1232 db_printf(" | POOL %d", pind); 1233 db_printf("\n-- "); 1234 for (pind = 0; pind < VM_NFREEPOOL; pind++) 1235 db_printf("-- -- "); 1236 db_printf("--\n"); 1237 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) { 1238 db_printf(" %2.2d (%6.6dK)", oind, 1239 1 << (PAGE_SHIFT - 10 + oind)); 1240 for (pind = 0; pind < VM_NFREEPOOL; pind++) { 1241 fl = vm_phys_free_queues[dom][flind][pind]; 1242 db_printf(" | %6.6d", fl[oind].lcnt); 1243 } 1244 db_printf("\n"); 1245 } 1246 db_printf("\n"); 1247 } 1248 db_printf("\n"); 1249 } 1250 } 1251 #endif 1252