sys/vm/vm_phys.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2002-2006 Rice University
   5  * Copyright (c) 2007 Alan L. Cox <alc@cs.rice.edu>
   6  * All rights reserved.
   7  *
   8  * This software was developed for the FreeBSD Project by Alan L. Cox,
   9  * Olivier Crameri, Peter Druschel, Sitaram Iyer, and Juan Navarro.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  23  * A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT
  24  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
  27  * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY
  30  * WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  31  * POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 /*
  35  *      Physical memory system implementation
  36  *
  37  * Any external functions defined by this module are only to be used by the
  38  * virtual memory system.
  39  */
  40
  41 #include <sys/cdefs.h>
  42 __FBSDID("$FreeBSD$");
  43
  44 #include "opt_ddb.h"
  45 #include "opt_vm.h"
  46
  47 #include <sys/param.h>
  48 #include <sys/systm.h>
  49 #include <sys/lock.h>
  50 #include <sys/kernel.h>
  51 #include <sys/malloc.h>
  52 #include <sys/mutex.h>
  53 #include <sys/proc.h>
  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 #include <sys/seq.h>
  61
  62 #include <ddb/ddb.h>
  63
  64 #include <vm/vm.h>
  65 #include <vm/vm_param.h>
  66 #include <vm/vm_kern.h>
  67 #include <vm/vm_object.h>
  68 #include <vm/vm_page.h>
  69 #include <vm/vm_phys.h>
  70 #include <vm/vm_pagequeue.h>
  71
  72 _Static_assert(sizeof(long) * NBBY >= VM_PHYSSEG_MAX,
  73     "Too many physsegs.");
  74
  75 #ifdef NUMA
  76 struct mem_affinity __read_mostly *mem_affinity;
  77 int __read_mostly *mem_locality;
  78 #endif
  79
  80 int __read_mostly vm_ndomains = 1;
  81
  82 struct vm_phys_seg __read_mostly vm_phys_segs[VM_PHYSSEG_MAX];
  83 int __read_mostly vm_phys_nsegs;
  84
  85 struct vm_phys_fictitious_seg;
  86 static int vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *,
  87     struct vm_phys_fictitious_seg *);
  88
  89 RB_HEAD(fict_tree, vm_phys_fictitious_seg) vm_phys_fictitious_tree =
  90     RB_INITIALIZER(_vm_phys_fictitious_tree);
  91
  92 struct vm_phys_fictitious_seg {
  93         RB_ENTRY(vm_phys_fictitious_seg) node;
  94         /* Memory region data */
  95         vm_paddr_t      start;
  96         vm_paddr_t      end;
  97         vm_page_t       first_page;
  98 };
  99
 100 RB_GENERATE_STATIC(fict_tree, vm_phys_fictitious_seg, node,
 101     vm_phys_fictitious_cmp);
 102
 103 static struct rwlock_padalign vm_phys_fictitious_reg_lock;
 104 MALLOC_DEFINE(M_FICT_PAGES, "vm_fictitious", "Fictitious VM pages");
 105
 106 static struct vm_freelist __aligned(CACHE_LINE_SIZE)
 107     vm_phys_free_queues[MAXMEMDOM][VM_NFREELIST][VM_NFREEPOOL][VM_NFREEORDER];
 108
 109 static int __read_mostly vm_nfreelists;
 110
 111 /*
 112  * Provides the mapping from VM_FREELIST_* to free list indices (flind).
 113  */
 114 static int __read_mostly vm_freelist_to_flind[VM_NFREELIST];
 115
 116 CTASSERT(VM_FREELIST_DEFAULT == 0);
 117
 118 #ifdef VM_FREELIST_ISADMA
 119 #define VM_ISADMA_BOUNDARY      16777216
 120 #endif
 121 #ifdef VM_FREELIST_DMA32
 122 #define VM_DMA32_BOUNDARY       ((vm_paddr_t)1 << 32)
 123 #endif
 124
 125 /*
 126  * Enforce the assumptions made by vm_phys_add_seg() and vm_phys_init() about
 127  * the ordering of the free list boundaries.
 128  */
 129 #if defined(VM_ISADMA_BOUNDARY) && defined(VM_LOWMEM_BOUNDARY)
 130 CTASSERT(VM_ISADMA_BOUNDARY < VM_LOWMEM_BOUNDARY);
 131 #endif
 132 #if defined(VM_LOWMEM_BOUNDARY) && defined(VM_DMA32_BOUNDARY)
 133 CTASSERT(VM_LOWMEM_BOUNDARY < VM_DMA32_BOUNDARY);
 134 #endif
 135
 136 static int sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS);
 137 SYSCTL_OID(_vm, OID_AUTO, phys_free, CTLTYPE_STRING | CTLFLAG_RD,
 138     NULL, 0, sysctl_vm_phys_free, "A", "Phys Free Info");
 139
 140 static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS);
 141 SYSCTL_OID(_vm, OID_AUTO, phys_segs, CTLTYPE_STRING | CTLFLAG_RD,
 142     NULL, 0, sysctl_vm_phys_segs, "A", "Phys Seg Info");
 143
 144 #ifdef NUMA
 145 static int sysctl_vm_phys_locality(SYSCTL_HANDLER_ARGS);
 146 SYSCTL_OID(_vm, OID_AUTO, phys_locality, CTLTYPE_STRING | CTLFLAG_RD,
 147     NULL, 0, sysctl_vm_phys_locality, "A", "Phys Locality Info");
 148 #endif
 149
 150 SYSCTL_INT(_vm, OID_AUTO, ndomains, CTLFLAG_RD,
 151     &vm_ndomains, 0, "Number of physical memory domains available.");
 152
 153 static vm_page_t vm_phys_alloc_seg_contig(struct vm_phys_seg *seg,
 154     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
 155     vm_paddr_t boundary);
 156 static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain);
 157 static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end);
 158 static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl,
 159     int order);
 160
 161 /*
 162  * Red-black tree helpers for vm fictitious range management.
 163  */
 164 static inline int
 165 vm_phys_fictitious_in_range(struct vm_phys_fictitious_seg *p,
 166     struct vm_phys_fictitious_seg *range)
 167 {
 168
 169         KASSERT(range->start != 0 && range->end != 0,
 170             ("Invalid range passed on search for vm_fictitious page"));
 171         if (p->start >= range->end)
 172                 return (1);
 173         if (p->start < range->start)
 174                 return (-1);
 175
 176         return (0);
 177 }
 178
 179 static int
 180 vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *p1,
 181     struct vm_phys_fictitious_seg *p2)
 182 {
 183
 184         /* Check if this is a search for a page */
 185         if (p1->end == 0)
 186                 return (vm_phys_fictitious_in_range(p1, p2));
 187
 188         KASSERT(p2->end != 0,
 189     ("Invalid range passed as second parameter to vm fictitious comparison"));
 190
 191         /* Searching to add a new range */
 192         if (p1->end <= p2->start)
 193                 return (-1);
 194         if (p1->start >= p2->end)
 195                 return (1);
 196
 197         panic("Trying to add overlapping vm fictitious ranges:\n"
 198             "[%#jx:%#jx] and [%#jx:%#jx]", (uintmax_t)p1->start,
 199             (uintmax_t)p1->end, (uintmax_t)p2->start, (uintmax_t)p2->end);
 200 }
 201
 202 int
 203 vm_phys_domain_match(int prefer, vm_paddr_t low, vm_paddr_t high)
 204 {
 205 #ifdef NUMA
 206         domainset_t mask;
 207         int i;
 208
 209         if (vm_ndomains == 1 || mem_affinity == NULL)
 210                 return (0);
 211
 212         DOMAINSET_ZERO(&mask);
 213         /*
 214          * Check for any memory that overlaps low, high.
 215          */
 216         for (i = 0; mem_affinity[i].end != 0; i++)
 217                 if (mem_affinity[i].start <= high &&
 218                     mem_affinity[i].end >= low)
 219                         DOMAINSET_SET(mem_affinity[i].domain, &mask);
 220         if (prefer != -1 && DOMAINSET_ISSET(prefer, &mask))
 221                 return (prefer);
 222         if (DOMAINSET_EMPTY(&mask))
 223                 panic("vm_phys_domain_match:  Impossible constraint");
 224         return (DOMAINSET_FFS(&mask) - 1);
 225 #else
 226         return (0);
 227 #endif
 228 }
 229
 230 /*
 231  * Outputs the state of the physical memory allocator, specifically,
 232  * the amount of physical memory in each free list.
 233  */
 234 static int
 235 sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS)
 236 {
 237         struct sbuf sbuf;
 238         struct vm_freelist *fl;
 239         int dom, error, flind, oind, pind;
 240
 241         error = sysctl_wire_old_buffer(req, 0);
 242         if (error != 0)
 243                 return (error);
 244         sbuf_new_for_sysctl(&sbuf, NULL, 128 * vm_ndomains, req);
 245         for (dom = 0; dom < vm_ndomains; dom++) {
 246                 sbuf_printf(&sbuf,"\nDOMAIN %d:\n", dom);
 247                 for (flind = 0; flind < vm_nfreelists; flind++) {
 248                         sbuf_printf(&sbuf, "\nFREE LIST %d:\n"
 249                             "\n  ORDER (SIZE)  |  NUMBER"
 250                             "\n              ", flind);
 251                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
 252                                 sbuf_printf(&sbuf, "  |  POOL %d", pind);
 253                         sbuf_printf(&sbuf, "\n--            ");
 254                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
 255                                 sbuf_printf(&sbuf, "-- --      ");
 256                         sbuf_printf(&sbuf, "--\n");
 257                         for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
 258                                 sbuf_printf(&sbuf, "  %2d (%6dK)", oind,
 259                                     1 << (PAGE_SHIFT - 10 + oind));
 260                                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 261                                 fl = vm_phys_free_queues[dom][flind][pind];
 262                                         sbuf_printf(&sbuf, "  |  %6d",
 263                                             fl[oind].lcnt);
 264                                 }
 265                                 sbuf_printf(&sbuf, "\n");
 266                         }
 267                 }
 268         }
 269         error = sbuf_finish(&sbuf);
 270         sbuf_delete(&sbuf);
 271         return (error);
 272 }
 273
 274 /*
 275  * Outputs the set of physical memory segments.
 276  */
 277 static int
 278 sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS)
 279 {
 280         struct sbuf sbuf;
 281         struct vm_phys_seg *seg;
 282         int error, segind;
 283
 284         error = sysctl_wire_old_buffer(req, 0);
 285         if (error != 0)
 286                 return (error);
 287         sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
 288         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 289                 sbuf_printf(&sbuf, "\nSEGMENT %d:\n\n", segind);
 290                 seg = &vm_phys_segs[segind];
 291                 sbuf_printf(&sbuf, "start:     %#jx\n",
 292                     (uintmax_t)seg->start);
 293                 sbuf_printf(&sbuf, "end:       %#jx\n",
 294                     (uintmax_t)seg->end);
 295                 sbuf_printf(&sbuf, "domain:    %d\n", seg->domain);
 296                 sbuf_printf(&sbuf, "free list: %p\n", seg->free_queues);
 297         }
 298         error = sbuf_finish(&sbuf);
 299         sbuf_delete(&sbuf);
 300         return (error);
 301 }
 302
 303 /*
 304  * Return affinity, or -1 if there's no affinity information.
 305  */
 306 int
 307 vm_phys_mem_affinity(int f, int t)
 308 {
 309
 310 #ifdef NUMA
 311         if (mem_locality == NULL)
 312                 return (-1);
 313         if (f >= vm_ndomains || t >= vm_ndomains)
 314                 return (-1);
 315         return (mem_locality[f * vm_ndomains + t]);
 316 #else
 317         return (-1);
 318 #endif
 319 }
 320
 321 #ifdef NUMA
 322 /*
 323  * Outputs the VM locality table.
 324  */
 325 static int
 326 sysctl_vm_phys_locality(SYSCTL_HANDLER_ARGS)
 327 {
 328         struct sbuf sbuf;
 329         int error, i, j;
 330
 331         error = sysctl_wire_old_buffer(req, 0);
 332         if (error != 0)
 333                 return (error);
 334         sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
 335
 336         sbuf_printf(&sbuf, "\n");
 337
 338         for (i = 0; i < vm_ndomains; i++) {
 339                 sbuf_printf(&sbuf, "%d: ", i);
 340                 for (j = 0; j < vm_ndomains; j++) {
 341                         sbuf_printf(&sbuf, "%d ", vm_phys_mem_affinity(i, j));
 342                 }
 343                 sbuf_printf(&sbuf, "\n");
 344         }
 345         error = sbuf_finish(&sbuf);
 346         sbuf_delete(&sbuf);
 347         return (error);
 348 }
 349 #endif
 350
 351 static void
 352 vm_freelist_add(struct vm_freelist *fl, vm_page_t m, int order, int tail)
 353 {
 354
 355         m->order = order;
 356         if (tail)
 357                 TAILQ_INSERT_TAIL(&fl[order].pl, m, listq);
 358         else
 359                 TAILQ_INSERT_HEAD(&fl[order].pl, m, listq);
 360         fl[order].lcnt++;
 361 }
 362
 363 static void
 364 vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order)
 365 {
 366
 367         TAILQ_REMOVE(&fl[order].pl, m, listq);
 368         fl[order].lcnt--;
 369         m->order = VM_NFREEORDER;
 370 }
 371
 372 /*
 373  * Create a physical memory segment.
 374  */
 375 static void
 376 _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain)
 377 {
 378         struct vm_phys_seg *seg;
 379
 380         KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX,
 381             ("vm_phys_create_seg: increase VM_PHYSSEG_MAX"));
 382         KASSERT(domain >= 0 && domain < vm_ndomains,
 383             ("vm_phys_create_seg: invalid domain provided"));
 384         seg = &vm_phys_segs[vm_phys_nsegs++];
 385         while (seg > vm_phys_segs && (seg - 1)->start >= end) {
 386                 *seg = *(seg - 1);
 387                 seg--;
 388         }
 389         seg->start = start;
 390         seg->end = end;
 391         seg->domain = domain;
 392 }
 393
 394 static void
 395 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end)
 396 {
 397 #ifdef NUMA
 398         int i;
 399
 400         if (mem_affinity == NULL) {
 401                 _vm_phys_create_seg(start, end, 0);
 402                 return;
 403         }
 404
 405         for (i = 0;; i++) {
 406                 if (mem_affinity[i].end == 0)
 407                         panic("Reached end of affinity info");
 408                 if (mem_affinity[i].end <= start)
 409                         continue;
 410                 if (mem_affinity[i].start > start)
 411                         panic("No affinity info for start %jx",
 412                             (uintmax_t)start);
 413                 if (mem_affinity[i].end >= end) {
 414                         _vm_phys_create_seg(start, end,
 415                             mem_affinity[i].domain);
 416                         break;
 417                 }
 418                 _vm_phys_create_seg(start, mem_affinity[i].end,
 419                     mem_affinity[i].domain);
 420                 start = mem_affinity[i].end;
 421         }
 422 #else
 423         _vm_phys_create_seg(start, end, 0);
 424 #endif
 425 }
 426
 427 /*
 428  * Add a physical memory segment.
 429  */
 430 void
 431 vm_phys_add_seg(vm_paddr_t start, vm_paddr_t end)
 432 {
 433         vm_paddr_t paddr;
 434
 435         KASSERT((start & PAGE_MASK) == 0,
 436             ("vm_phys_define_seg: start is not page aligned"));
 437         KASSERT((end & PAGE_MASK) == 0,
 438             ("vm_phys_define_seg: end is not page aligned"));
 439
 440         /*
 441          * Split the physical memory segment if it spans two or more free
 442          * list boundaries.
 443          */
 444         paddr = start;
 445 #ifdef  VM_FREELIST_ISADMA
 446         if (paddr < VM_ISADMA_BOUNDARY && end > VM_ISADMA_BOUNDARY) {
 447                 vm_phys_create_seg(paddr, VM_ISADMA_BOUNDARY);
 448                 paddr = VM_ISADMA_BOUNDARY;
 449         }
 450 #endif
 451 #ifdef  VM_FREELIST_LOWMEM
 452         if (paddr < VM_LOWMEM_BOUNDARY && end > VM_LOWMEM_BOUNDARY) {
 453                 vm_phys_create_seg(paddr, VM_LOWMEM_BOUNDARY);
 454                 paddr = VM_LOWMEM_BOUNDARY;
 455         }
 456 #endif
 457 #ifdef  VM_FREELIST_DMA32
 458         if (paddr < VM_DMA32_BOUNDARY && end > VM_DMA32_BOUNDARY) {
 459                 vm_phys_create_seg(paddr, VM_DMA32_BOUNDARY);
 460                 paddr = VM_DMA32_BOUNDARY;
 461         }
 462 #endif
 463         vm_phys_create_seg(paddr, end);
 464 }
 465
 466 /*
 467  * Initialize the physical memory allocator.
 468  *
 469  * Requires that vm_page_array is initialized!
 470  */
 471 void
 472 vm_phys_init(void)
 473 {
 474         struct vm_freelist *fl;
 475         struct vm_phys_seg *seg;
 476         u_long npages;
 477         int dom, flind, freelist, oind, pind, segind;
 478
 479         /*
 480          * Compute the number of free lists, and generate the mapping from the
 481          * manifest constants VM_FREELIST_* to the free list indices.
 482          *
 483          * Initially, the entries of vm_freelist_to_flind[] are set to either
 484          * 0 or 1 to indicate which free lists should be created.
 485          */
 486         npages = 0;
 487         for (segind = vm_phys_nsegs - 1; segind >= 0; segind--) {
 488                 seg = &vm_phys_segs[segind];
 489 #ifdef  VM_FREELIST_ISADMA
 490                 if (seg->end <= VM_ISADMA_BOUNDARY)
 491                         vm_freelist_to_flind[VM_FREELIST_ISADMA] = 1;
 492                 else
 493 #endif
 494 #ifdef  VM_FREELIST_LOWMEM
 495                 if (seg->end <= VM_LOWMEM_BOUNDARY)
 496                         vm_freelist_to_flind[VM_FREELIST_LOWMEM] = 1;
 497                 else
 498 #endif
 499 #ifdef  VM_FREELIST_DMA32
 500                 if (
 501 #ifdef  VM_DMA32_NPAGES_THRESHOLD
 502                     /*
 503                      * Create the DMA32 free list only if the amount of
 504                      * physical memory above physical address 4G exceeds the
 505                      * given threshold.
 506                      */
 507                     npages > VM_DMA32_NPAGES_THRESHOLD &&
 508 #endif
 509                     seg->end <= VM_DMA32_BOUNDARY)
 510                         vm_freelist_to_flind[VM_FREELIST_DMA32] = 1;
 511                 else
 512 #endif
 513                 {
 514                         npages += atop(seg->end - seg->start);
 515                         vm_freelist_to_flind[VM_FREELIST_DEFAULT] = 1;
 516                 }
 517         }
 518         /* Change each entry into a running total of the free lists. */
 519         for (freelist = 1; freelist < VM_NFREELIST; freelist++) {
 520                 vm_freelist_to_flind[freelist] +=
 521                     vm_freelist_to_flind[freelist - 1];
 522         }
 523         vm_nfreelists = vm_freelist_to_flind[VM_NFREELIST - 1];
 524         KASSERT(vm_nfreelists > 0, ("vm_phys_init: no free lists"));
 525         /* Change each entry into a free list index. */
 526         for (freelist = 0; freelist < VM_NFREELIST; freelist++)
 527                 vm_freelist_to_flind[freelist]--;
 528
 529         /*
 530          * Initialize the first_page and free_queues fields of each physical
 531          * memory segment.
 532          */
 533 #ifdef VM_PHYSSEG_SPARSE
 534         npages = 0;
 535 #endif
 536         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 537                 seg = &vm_phys_segs[segind];
 538 #ifdef VM_PHYSSEG_SPARSE
 539                 seg->first_page = &vm_page_array[npages];
 540                 npages += atop(seg->end - seg->start);
 541 #else
 542                 seg->first_page = PHYS_TO_VM_PAGE(seg->start);
 543 #endif
 544 #ifdef  VM_FREELIST_ISADMA
 545                 if (seg->end <= VM_ISADMA_BOUNDARY) {
 546                         flind = vm_freelist_to_flind[VM_FREELIST_ISADMA];
 547                         KASSERT(flind >= 0,
 548                             ("vm_phys_init: ISADMA flind < 0"));
 549                 } else
 550 #endif
 551 #ifdef  VM_FREELIST_LOWMEM
 552                 if (seg->end <= VM_LOWMEM_BOUNDARY) {
 553                         flind = vm_freelist_to_flind[VM_FREELIST_LOWMEM];
 554                         KASSERT(flind >= 0,
 555                             ("vm_phys_init: LOWMEM flind < 0"));
 556                 } else
 557 #endif
 558 #ifdef  VM_FREELIST_DMA32
 559                 if (seg->end <= VM_DMA32_BOUNDARY) {
 560                         flind = vm_freelist_to_flind[VM_FREELIST_DMA32];
 561                         KASSERT(flind >= 0,
 562                             ("vm_phys_init: DMA32 flind < 0"));
 563                 } else
 564 #endif
 565                 {
 566                         flind = vm_freelist_to_flind[VM_FREELIST_DEFAULT];
 567                         KASSERT(flind >= 0,
 568                             ("vm_phys_init: DEFAULT flind < 0"));
 569                 }
 570                 seg->free_queues = &vm_phys_free_queues[seg->domain][flind];
 571         }
 572
 573         /*
 574          * Initialize the free queues.
 575          */
 576         for (dom = 0; dom < vm_ndomains; dom++) {
 577                 for (flind = 0; flind < vm_nfreelists; flind++) {
 578                         for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 579                                 fl = vm_phys_free_queues[dom][flind][pind];
 580                                 for (oind = 0; oind < VM_NFREEORDER; oind++)
 581                                         TAILQ_INIT(&fl[oind].pl);
 582                         }
 583                 }
 584         }
 585
 586         rw_init(&vm_phys_fictitious_reg_lock, "vmfctr");
 587 }
 588
 589 /*
 590  * Split a contiguous, power of two-sized set of physical pages.
 591  */
 592 static __inline void
 593 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order)
 594 {
 595         vm_page_t m_buddy;
 596
 597         while (oind > order) {
 598                 oind--;
 599                 m_buddy = &m[1 << oind];
 600                 KASSERT(m_buddy->order == VM_NFREEORDER,
 601                     ("vm_phys_split_pages: page %p has unexpected order %d",
 602                     m_buddy, m_buddy->order));
 603                 vm_freelist_add(fl, m_buddy, oind, 0);
 604         }
 605 }
 606
 607 /*
 608  * Allocate a contiguous, power of two-sized set of physical pages
 609  * from the free lists.
 610  *
 611  * The free page queues must be locked.
 612  */
 613 vm_page_t
 614 vm_phys_alloc_pages(int domain, int pool, int order)
 615 {
 616         vm_page_t m;
 617         int freelist;
 618
 619         for (freelist = 0; freelist < VM_NFREELIST; freelist++) {
 620                 m = vm_phys_alloc_freelist_pages(domain, freelist, pool, order);
 621                 if (m != NULL)
 622                         return (m);
 623         }
 624         return (NULL);
 625 }
 626
 627 int
 628 vm_phys_alloc_npages(int domain, int pool, vm_page_t *mp, int cnt)
 629 {
 630         vm_page_t m;
 631         int order, freelist;
 632
 633         for (freelist = 0; freelist < VM_NFREELIST; freelist++) {
 634                 for (order = fls(cnt) -1; order >= 0; order--) {
 635                         m = vm_phys_alloc_freelist_pages(domain, freelist,
 636                             pool, order);
 637                         if (m != NULL) {
 638                                 *mp = m;
 639                                 return (1 << order);
 640                         }
 641                 }
 642         }
 643         *mp = NULL;
 644         return (0);
 645 }
 646
 647 /*
 648  * Allocate a contiguous, power of two-sized set of physical pages from the
 649  * specified free list.  The free list must be specified using one of the
 650  * manifest constants VM_FREELIST_*.
 651  *
 652  * The free page queues must be locked.
 653  */
 654 vm_page_t
 655 vm_phys_alloc_freelist_pages(int domain, int freelist, int pool, int order)
 656 {
 657         struct vm_freelist *alt, *fl;
 658         vm_page_t m;
 659         int oind, pind, flind;
 660
 661         KASSERT(domain >= 0 && domain < vm_ndomains,
 662             ("vm_phys_alloc_freelist_pages: domain %d is out of range",
 663             domain));
 664         KASSERT(freelist < VM_NFREELIST,
 665             ("vm_phys_alloc_freelist_pages: freelist %d is out of range",
 666             freelist));
 667         KASSERT(pool < VM_NFREEPOOL,
 668             ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
 669         KASSERT(order < VM_NFREEORDER,
 670             ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
 671
 672         flind = vm_freelist_to_flind[freelist];
 673         /* Check if freelist is present */
 674         if (flind < 0)
 675                 return (NULL);
 676
 677         vm_domain_free_assert_locked(VM_DOMAIN(domain));
 678         fl = &vm_phys_free_queues[domain][flind][pool][0];
 679         for (oind = order; oind < VM_NFREEORDER; oind++) {
 680                 m = TAILQ_FIRST(&fl[oind].pl);
 681                 if (m != NULL) {
 682                         vm_freelist_rem(fl, m, oind);
 683                         vm_phys_split_pages(m, oind, fl, order);
 684                         return (m);
 685                 }
 686         }
 687
 688         /*
 689          * The given pool was empty.  Find the largest
 690          * contiguous, power-of-two-sized set of pages in any
 691          * pool.  Transfer these pages to the given pool, and
 692          * use them to satisfy the allocation.
 693          */
 694         for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
 695                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 696                         alt = &vm_phys_free_queues[domain][flind][pind][0];
 697                         m = TAILQ_FIRST(&alt[oind].pl);
 698                         if (m != NULL) {
 699                                 vm_freelist_rem(alt, m, oind);
 700                                 vm_phys_set_pool(pool, m, oind);
 701                                 vm_phys_split_pages(m, oind, fl, order);
 702                                 return (m);
 703                         }
 704                 }
 705         }
 706         return (NULL);
 707 }
 708
 709 /*
 710  * Find the vm_page corresponding to the given physical address.
 711  */
 712 vm_page_t
 713 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
 714 {
 715         struct vm_phys_seg *seg;
 716         int segind;
 717
 718         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 719                 seg = &vm_phys_segs[segind];
 720                 if (pa >= seg->start && pa < seg->end)
 721                         return (&seg->first_page[atop(pa - seg->start)]);
 722         }
 723         return (NULL);
 724 }
 725
 726 vm_page_t
 727 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
 728 {
 729         struct vm_phys_fictitious_seg tmp, *seg;
 730         vm_page_t m;
 731
 732         m = NULL;
 733         tmp.start = pa;
 734         tmp.end = 0;
 735
 736         rw_rlock(&vm_phys_fictitious_reg_lock);
 737         seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
 738         rw_runlock(&vm_phys_fictitious_reg_lock);
 739         if (seg == NULL)
 740                 return (NULL);
 741
 742         m = &seg->first_page[atop(pa - seg->start)];
 743         KASSERT((m->flags & PG_FICTITIOUS) != 0, ("%p not fictitious", m));
 744
 745         return (m);
 746 }
 747
 748 static inline void
 749 vm_phys_fictitious_init_range(vm_page_t range, vm_paddr_t start,
 750     long page_count, vm_memattr_t memattr)
 751 {
 752         long i;
 753
 754         bzero(range, page_count * sizeof(*range));
 755         for (i = 0; i < page_count; i++) {
 756                 vm_page_initfake(&range[i], start + PAGE_SIZE * i, memattr);
 757                 range[i].oflags &= ~VPO_UNMANAGED;
 758                 range[i].busy_lock = VPB_UNBUSIED;
 759         }
 760 }
 761
 762 int
 763 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
 764     vm_memattr_t memattr)
 765 {
 766         struct vm_phys_fictitious_seg *seg;
 767         vm_page_t fp;
 768         long page_count;
 769 #ifdef VM_PHYSSEG_DENSE
 770         long pi, pe;
 771         long dpage_count;
 772 #endif
 773
 774         KASSERT(start < end,
 775             ("Start of segment isn't less than end (start: %jx end: %jx)",
 776             (uintmax_t)start, (uintmax_t)end));
 777
 778         page_count = (end - start) / PAGE_SIZE;
 779
 780 #ifdef VM_PHYSSEG_DENSE
 781         pi = atop(start);
 782         pe = atop(end);
 783         if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
 784                 fp = &vm_page_array[pi - first_page];
 785                 if ((pe - first_page) > vm_page_array_size) {
 786                         /*
 787                          * We have a segment that starts inside
 788                          * of vm_page_array, but ends outside of it.
 789                          *
 790                          * Use vm_page_array pages for those that are
 791                          * inside of the vm_page_array range, and
 792                          * allocate the remaining ones.
 793                          */
 794                         dpage_count = vm_page_array_size - (pi - first_page);
 795                         vm_phys_fictitious_init_range(fp, start, dpage_count,
 796                             memattr);
 797                         page_count -= dpage_count;
 798                         start += ptoa(dpage_count);
 799                         goto alloc;
 800                 }
 801                 /*
 802                  * We can allocate the full range from vm_page_array,
 803                  * so there's no need to register the range in the tree.
 804                  */
 805                 vm_phys_fictitious_init_range(fp, start, page_count, memattr);
 806                 return (0);
 807         } else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
 808                 /*
 809                  * We have a segment that ends inside of vm_page_array,
 810                  * but starts outside of it.
 811                  */
 812                 fp = &vm_page_array[0];
 813                 dpage_count = pe - first_page;
 814                 vm_phys_fictitious_init_range(fp, ptoa(first_page), dpage_count,
 815                     memattr);
 816                 end -= ptoa(dpage_count);
 817                 page_count -= dpage_count;
 818                 goto alloc;
 819         } else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
 820                 /*
 821                  * Trying to register a fictitious range that expands before
 822                  * and after vm_page_array.
 823                  */
 824                 return (EINVAL);
 825         } else {
 826 alloc:
 827 #endif
 828                 fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
 829                     M_WAITOK);
 830 #ifdef VM_PHYSSEG_DENSE
 831         }
 832 #endif
 833         vm_phys_fictitious_init_range(fp, start, page_count, memattr);
 834
 835         seg = malloc(sizeof(*seg), M_FICT_PAGES, M_WAITOK | M_ZERO);
 836         seg->start = start;
 837         seg->end = end;
 838         seg->first_page = fp;
 839
 840         rw_wlock(&vm_phys_fictitious_reg_lock);
 841         RB_INSERT(fict_tree, &vm_phys_fictitious_tree, seg);
 842         rw_wunlock(&vm_phys_fictitious_reg_lock);
 843
 844         return (0);
 845 }
 846
 847 void
 848 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
 849 {
 850         struct vm_phys_fictitious_seg *seg, tmp;
 851 #ifdef VM_PHYSSEG_DENSE
 852         long pi, pe;
 853 #endif
 854
 855         KASSERT(start < end,
 856             ("Start of segment isn't less than end (start: %jx end: %jx)",
 857             (uintmax_t)start, (uintmax_t)end));
 858
 859 #ifdef VM_PHYSSEG_DENSE
 860         pi = atop(start);
 861         pe = atop(end);
 862         if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
 863                 if ((pe - first_page) <= vm_page_array_size) {
 864                         /*
 865                          * This segment was allocated using vm_page_array
 866                          * only, there's nothing to do since those pages
 867                          * were never added to the tree.
 868                          */
 869                         return;
 870                 }
 871                 /*
 872                  * We have a segment that starts inside
 873                  * of vm_page_array, but ends outside of it.
 874                  *
 875                  * Calculate how many pages were added to the
 876                  * tree and free them.
 877                  */
 878                 start = ptoa(first_page + vm_page_array_size);
 879         } else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
 880                 /*
 881                  * We have a segment that ends inside of vm_page_array,
 882                  * but starts outside of it.
 883                  */
 884                 end = ptoa(first_page);
 885         } else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
 886                 /* Since it's not possible to register such a range, panic. */
 887                 panic(
 888                     "Unregistering not registered fictitious range [%#jx:%#jx]",
 889                     (uintmax_t)start, (uintmax_t)end);
 890         }
 891 #endif
 892         tmp.start = start;
 893         tmp.end = 0;
 894
 895         rw_wlock(&vm_phys_fictitious_reg_lock);
 896         seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
 897         if (seg->start != start || seg->end != end) {
 898                 rw_wunlock(&vm_phys_fictitious_reg_lock);
 899                 panic(
 900                     "Unregistering not registered fictitious range [%#jx:%#jx]",
 901                     (uintmax_t)start, (uintmax_t)end);
 902         }
 903         RB_REMOVE(fict_tree, &vm_phys_fictitious_tree, seg);
 904         rw_wunlock(&vm_phys_fictitious_reg_lock);
 905         free(seg->first_page, M_FICT_PAGES);
 906         free(seg, M_FICT_PAGES);
 907 }
 908
 909 /*
 910  * Free a contiguous, power of two-sized set of physical pages.
 911  *
 912  * The free page queues must be locked.
 913  */
 914 void
 915 vm_phys_free_pages(vm_page_t m, int order)
 916 {
 917         struct vm_freelist *fl;
 918         struct vm_phys_seg *seg;
 919         vm_paddr_t pa;
 920         vm_page_t m_buddy;
 921
 922         KASSERT(m->order == VM_NFREEORDER,
 923             ("vm_phys_free_pages: page %p has unexpected order %d",
 924             m, m->order));
 925         KASSERT(m->pool < VM_NFREEPOOL,
 926             ("vm_phys_free_pages: page %p has unexpected pool %d",
 927             m, m->pool));
 928         KASSERT(order < VM_NFREEORDER,
 929             ("vm_phys_free_pages: order %d is out of range", order));
 930         seg = &vm_phys_segs[m->segind];
 931         vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
 932         if (order < VM_NFREEORDER - 1) {
 933                 pa = VM_PAGE_TO_PHYS(m);
 934                 do {
 935                         pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
 936                         if (pa < seg->start || pa >= seg->end)
 937                                 break;
 938                         m_buddy = &seg->first_page[atop(pa - seg->start)];
 939                         if (m_buddy->order != order)
 940                                 break;
 941                         fl = (*seg->free_queues)[m_buddy->pool];
 942                         vm_freelist_rem(fl, m_buddy, order);
 943                         if (m_buddy->pool != m->pool)
 944                                 vm_phys_set_pool(m->pool, m_buddy, order);
 945                         order++;
 946                         pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
 947                         m = &seg->first_page[atop(pa - seg->start)];
 948                 } while (order < VM_NFREEORDER - 1);
 949         }
 950         fl = (*seg->free_queues)[m->pool];
 951         vm_freelist_add(fl, m, order, 1);
 952 }
 953
 954 /*
 955  * Free a contiguous, arbitrarily sized set of physical pages.
 956  *
 957  * The free page queues must be locked.
 958  */
 959 void
 960 vm_phys_free_contig(vm_page_t m, u_long npages)
 961 {
 962         u_int n;
 963         int order;
 964
 965         /*
 966          * Avoid unnecessary coalescing by freeing the pages in the largest
 967          * possible power-of-two-sized subsets.
 968          */
 969         vm_domain_free_assert_locked(vm_pagequeue_domain(m));
 970         for (;; npages -= n) {
 971                 /*
 972                  * Unsigned "min" is used here so that "order" is assigned
 973                  * "VM_NFREEORDER - 1" when "m"'s physical address is zero
 974                  * or the low-order bits of its physical address are zero
 975                  * because the size of a physical address exceeds the size of
 976                  * a long.
 977                  */
 978                 order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
 979                     VM_NFREEORDER - 1);
 980                 n = 1 << order;
 981                 if (npages < n)
 982                         break;
 983                 vm_phys_free_pages(m, order);
 984                 m += n;
 985         }
 986         /* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */
 987         for (; npages > 0; npages -= n) {
 988                 order = flsl(npages) - 1;
 989                 n = 1 << order;
 990                 vm_phys_free_pages(m, order);
 991                 m += n;
 992         }
 993 }
 994
 995 /*
 996  * Scan physical memory between the specified addresses "low" and "high" for a
 997  * run of contiguous physical pages that satisfy the specified conditions, and
 998  * return the lowest page in the run.  The specified "alignment" determines
 999  * the alignment of the lowest physical page in the run.  If the specified
1000  * "boundary" is non-zero, then the run of physical pages cannot span a
1001  * physical address that is a multiple of "boundary".
1002  *
1003  * "npages" must be greater than zero.  Both "alignment" and "boundary" must
1004  * be a power of two.
1005  */
1006 vm_page_t
1007 vm_phys_scan_contig(int domain, u_long npages, vm_paddr_t low, vm_paddr_t high,
1008     u_long alignment, vm_paddr_t boundary, int options)
1009 {
1010         vm_paddr_t pa_end;
1011         vm_page_t m_end, m_run, m_start;
1012         struct vm_phys_seg *seg;
1013         int segind;
1014
1015         KASSERT(npages > 0, ("npages is 0"));
1016         KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1017         KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1018         if (low >= high)
1019                 return (NULL);
1020         for (segind = 0; segind < vm_phys_nsegs; segind++) {
1021                 seg = &vm_phys_segs[segind];
1022                 if (seg->domain != domain)
1023                         continue;
1024                 if (seg->start >= high)
1025                         break;
1026                 if (low >= seg->end)
1027                         continue;
1028                 if (low <= seg->start)
1029                         m_start = seg->first_page;
1030                 else
1031                         m_start = &seg->first_page[atop(low - seg->start)];
1032                 if (high < seg->end)
1033                         pa_end = high;
1034                 else
1035                         pa_end = seg->end;
1036                 if (pa_end - VM_PAGE_TO_PHYS(m_start) < ptoa(npages))
1037                         continue;
1038                 m_end = &seg->first_page[atop(pa_end - seg->start)];
1039                 m_run = vm_page_scan_contig(npages, m_start, m_end,
1040                     alignment, boundary, options);
1041                 if (m_run != NULL)
1042                         return (m_run);
1043         }
1044         return (NULL);
1045 }
1046
1047 /*
1048  * Set the pool for a contiguous, power of two-sized set of physical pages.
1049  */
1050 void
1051 vm_phys_set_pool(int pool, vm_page_t m, int order)
1052 {
1053         vm_page_t m_tmp;
1054
1055         for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
1056                 m_tmp->pool = pool;
1057 }
1058
1059 /*
1060  * Search for the given physical page "m" in the free lists.  If the search
1061  * succeeds, remove "m" from the free lists and return TRUE.  Otherwise, return
1062  * FALSE, indicating that "m" is not in the free lists.
1063  *
1064  * The free page queues must be locked.
1065  */
1066 boolean_t
1067 vm_phys_unfree_page(vm_page_t m)
1068 {
1069         struct vm_freelist *fl;
1070         struct vm_phys_seg *seg;
1071         vm_paddr_t pa, pa_half;
1072         vm_page_t m_set, m_tmp;
1073         int order;
1074
1075         /*
1076          * First, find the contiguous, power of two-sized set of free
1077          * physical pages containing the given physical page "m" and
1078          * assign it to "m_set".
1079          */
1080         seg = &vm_phys_segs[m->segind];
1081         vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
1082         for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
1083             order < VM_NFREEORDER - 1; ) {
1084                 order++;
1085                 pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
1086                 if (pa >= seg->start)
1087                         m_set = &seg->first_page[atop(pa - seg->start)];
1088                 else
1089                         return (FALSE);
1090         }
1091         if (m_set->order < order)
1092                 return (FALSE);
1093         if (m_set->order == VM_NFREEORDER)
1094                 return (FALSE);
1095         KASSERT(m_set->order < VM_NFREEORDER,
1096             ("vm_phys_unfree_page: page %p has unexpected order %d",
1097             m_set, m_set->order));
1098
1099         /*
1100          * Next, remove "m_set" from the free lists.  Finally, extract
1101          * "m" from "m_set" using an iterative algorithm: While "m_set"
1102          * is larger than a page, shrink "m_set" by returning the half
1103          * of "m_set" that does not contain "m" to the free lists.
1104          */
1105         fl = (*seg->free_queues)[m_set->pool];
1106         order = m_set->order;
1107         vm_freelist_rem(fl, m_set, order);
1108         while (order > 0) {
1109                 order--;
1110                 pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
1111                 if (m->phys_addr < pa_half)
1112                         m_tmp = &seg->first_page[atop(pa_half - seg->start)];
1113                 else {
1114                         m_tmp = m_set;
1115                         m_set = &seg->first_page[atop(pa_half - seg->start)];
1116                 }
1117                 vm_freelist_add(fl, m_tmp, order, 0);
1118         }
1119         KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
1120         return (TRUE);
1121 }
1122
1123 /*
1124  * Allocate a contiguous set of physical pages of the given size
1125  * "npages" from the free lists.  All of the physical pages must be at
1126  * or above the given physical address "low" and below the given
1127  * physical address "high".  The given value "alignment" determines the
1128  * alignment of the first physical page in the set.  If the given value
1129  * "boundary" is non-zero, then the set of physical pages cannot cross
1130  * any physical address boundary that is a multiple of that value.  Both
1131  * "alignment" and "boundary" must be a power of two.
1132  */
1133 vm_page_t
1134 vm_phys_alloc_contig(int domain, u_long npages, vm_paddr_t low, vm_paddr_t high,
1135     u_long alignment, vm_paddr_t boundary)
1136 {
1137         vm_paddr_t pa_end, pa_start;
1138         vm_page_t m_run;
1139         struct vm_phys_seg *seg;
1140         int segind;
1141
1142         KASSERT(npages > 0, ("npages is 0"));
1143         KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1144         KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1145         vm_domain_free_assert_locked(VM_DOMAIN(domain));
1146         if (low >= high)
1147                 return (NULL);
1148         m_run = NULL;
1149         for (segind = vm_phys_nsegs - 1; segind >= 0; segind--) {
1150                 seg = &vm_phys_segs[segind];
1151                 if (seg->start >= high || seg->domain != domain)
1152                         continue;
1153                 if (low >= seg->end)
1154                         break;
1155                 if (low <= seg->start)
1156                         pa_start = seg->start;
1157                 else
1158                         pa_start = low;
1159                 if (high < seg->end)
1160                         pa_end = high;
1161                 else
1162                         pa_end = seg->end;
1163                 if (pa_end - pa_start < ptoa(npages))
1164                         continue;
1165                 m_run = vm_phys_alloc_seg_contig(seg, npages, low, high,
1166                     alignment, boundary);
1167                 if (m_run != NULL)
1168                         break;
1169         }
1170         return (m_run);
1171 }
1172
1173 /*
1174  * Allocate a run of contiguous physical pages from the free list for the
1175  * specified segment.
1176  */
1177 static vm_page_t
1178 vm_phys_alloc_seg_contig(struct vm_phys_seg *seg, u_long npages,
1179     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
1180 {
1181         struct vm_freelist *fl;
1182         vm_paddr_t pa, pa_end, size;
1183         vm_page_t m, m_ret;
1184         u_long npages_end;
1185         int oind, order, pind;
1186
1187         KASSERT(npages > 0, ("npages is 0"));
1188         KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1189         KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1190         vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
1191         /* Compute the queue that is the best fit for npages. */
1192         for (order = 0; (1 << order) < npages; order++);
1193         /* Search for a run satisfying the specified conditions. */
1194         size = npages << PAGE_SHIFT;
1195         for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER;
1196             oind++) {
1197                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1198                         fl = (*seg->free_queues)[pind];
1199                         TAILQ_FOREACH(m_ret, &fl[oind].pl, listq) {
1200                                 /*
1201                                  * Is the size of this allocation request
1202                                  * larger than the largest block size?
1203                                  */
1204                                 if (order >= VM_NFREEORDER) {
1205                                         /*
1206                                          * Determine if a sufficient number of
1207                                          * subsequent blocks to satisfy the
1208                                          * allocation request are free.
1209                                          */
1210                                         pa = VM_PAGE_TO_PHYS(m_ret);
1211                                         pa_end = pa + size;
1212                                         if (pa_end < pa)
1213                                                 continue;
1214                                         for (;;) {
1215                                                 pa += 1 << (PAGE_SHIFT +
1216                                                     VM_NFREEORDER - 1);
1217                                                 if (pa >= pa_end ||
1218                                                     pa < seg->start ||
1219                                                     pa >= seg->end)
1220                                                         break;
1221                                                 m = &seg->first_page[atop(pa -
1222                                                     seg->start)];
1223                                                 if (m->order != VM_NFREEORDER -
1224                                                     1)
1225                                                         break;
1226                                         }
1227                                         /* If not, go to the next block. */
1228                                         if (pa < pa_end)
1229                                                 continue;
1230                                 }
1231
1232                                 /*
1233                                  * Determine if the blocks are within the
1234                                  * given range, satisfy the given alignment,
1235                                  * and do not cross the given boundary.
1236                                  */
1237                                 pa = VM_PAGE_TO_PHYS(m_ret);
1238                                 pa_end = pa + size;
1239                                 if (pa >= low && pa_end <= high &&
1240                                     (pa & (alignment - 1)) == 0 &&
1241                                     rounddown2(pa ^ (pa_end - 1), boundary) == 0)
1242                                         goto done;
1243                         }
1244                 }
1245         }
1246         return (NULL);
1247 done:
1248         for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
1249                 fl = (*seg->free_queues)[m->pool];
1250                 vm_freelist_rem(fl, m, m->order);
1251         }
1252         if (m_ret->pool != VM_FREEPOOL_DEFAULT)
1253                 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m_ret, oind);
1254         fl = (*seg->free_queues)[m_ret->pool];
1255         vm_phys_split_pages(m_ret, oind, fl, order);
1256         /* Return excess pages to the free lists. */
1257         npages_end = roundup2(npages, 1 << imin(oind, order));
1258         if (npages < npages_end)
1259                 vm_phys_free_contig(&m_ret[npages], npages_end - npages);
1260         return (m_ret);
1261 }
1262
1263 #ifdef DDB
1264 /*
1265  * Show the number of physical pages in each of the free lists.
1266  */
1267 DB_SHOW_COMMAND(freepages, db_show_freepages)
1268 {
1269         struct vm_freelist *fl;
1270         int flind, oind, pind, dom;
1271
1272         for (dom = 0; dom < vm_ndomains; dom++) {
1273                 db_printf("DOMAIN: %d\n", dom);
1274                 for (flind = 0; flind < vm_nfreelists; flind++) {
1275                         db_printf("FREE LIST %d:\n"
1276                             "\n  ORDER (SIZE)  |  NUMBER"
1277                             "\n              ", flind);
1278                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
1279                                 db_printf("  |  POOL %d", pind);
1280                         db_printf("\n--            ");
1281                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
1282                                 db_printf("-- --      ");
1283                         db_printf("--\n");
1284                         for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
1285                                 db_printf("  %2.2d (%6.6dK)", oind,
1286                                     1 << (PAGE_SHIFT - 10 + oind));
1287                                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1288                                 fl = vm_phys_free_queues[dom][flind][pind];
1289                                         db_printf("  |  %6.6d", fl[oind].lcnt);
1290                                 }
1291                                 db_printf("\n");
1292                         }
1293                         db_printf("\n");
1294                 }
1295                 db_printf("\n");
1296         }
1297 }
1298 #endif