sys/vm/vm_phys.c

   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/sbuf.h>
  56 #include <sys/sysctl.h>
  57 #include <sys/vmmeter.h>
  58
  59 #include <ddb/ddb.h>
  60
  61 #include <vm/vm.h>
  62 #include <vm/vm_param.h>
  63 #include <vm/vm_kern.h>
  64 #include <vm/vm_object.h>
  65 #include <vm/vm_page.h>
  66 #include <vm/vm_phys.h>
  67
  68 _Static_assert(sizeof(long) * NBBY >= VM_PHYSSEG_MAX,
  69     "Too many physsegs.");
  70
  71 struct mem_affinity *mem_affinity;
  72
  73 int vm_ndomains = 1;
  74
  75 struct vm_phys_seg vm_phys_segs[VM_PHYSSEG_MAX];
  76 int vm_phys_nsegs;
  77
  78 #define VM_PHYS_FICTITIOUS_NSEGS        8
  79 static struct vm_phys_fictitious_seg {
  80         vm_paddr_t      start;
  81         vm_paddr_t      end;
  82         vm_page_t       first_page;
  83 } vm_phys_fictitious_segs[VM_PHYS_FICTITIOUS_NSEGS];
  84 static struct mtx vm_phys_fictitious_reg_mtx;
  85 MALLOC_DEFINE(M_FICT_PAGES, "vm_fictitious", "Fictitious VM pages");
  86
  87 static struct vm_freelist
  88     vm_phys_free_queues[MAXMEMDOM][VM_NFREELIST][VM_NFREEPOOL][VM_NFREEORDER];
  89
  90 static int vm_nfreelists = VM_FREELIST_DEFAULT + 1;
  91
  92 static int cnt_prezero;
  93 SYSCTL_INT(_vm_stats_misc, OID_AUTO, cnt_prezero, CTLFLAG_RD,
  94     &cnt_prezero, 0, "The number of physical pages prezeroed at idle time");
  95
  96 static int sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS);
  97 SYSCTL_OID(_vm, OID_AUTO, phys_free, CTLTYPE_STRING | CTLFLAG_RD,
  98     NULL, 0, sysctl_vm_phys_free, "A", "Phys Free Info");
  99
 100 static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS);
 101 SYSCTL_OID(_vm, OID_AUTO, phys_segs, CTLTYPE_STRING | CTLFLAG_RD,
 102     NULL, 0, sysctl_vm_phys_segs, "A", "Phys Seg Info");
 103
 104 SYSCTL_INT(_vm, OID_AUTO, ndomains, CTLFLAG_RD,
 105     &vm_ndomains, 0, "Number of physical memory domains available.");
 106
 107 static vm_page_t vm_phys_alloc_domain_pages(int domain, int flind, int pool,
 108     int order);
 109 static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind,
 110     int domain);
 111 static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind);
 112 static int vm_phys_paddr_to_segind(vm_paddr_t pa);
 113 static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl,
 114     int order);
 115
 116 static __inline int
 117 vm_rr_selectdomain(void)
 118 {
 119 #if MAXMEMDOM > 1
 120         struct thread *td;
 121
 122         td = curthread;
 123
 124         td->td_dom_rr_idx++;
 125         td->td_dom_rr_idx %= vm_ndomains;
 126         return (td->td_dom_rr_idx);
 127 #else
 128         return (0);
 129 #endif
 130 }
 131
 132 boolean_t
 133 vm_phys_domain_intersects(long mask, vm_paddr_t low, vm_paddr_t high)
 134 {
 135         struct vm_phys_seg *s;
 136         int idx;
 137
 138         while ((idx = ffsl(mask)) != 0) {
 139                 idx--;  /* ffsl counts from 1 */
 140                 mask &= ~(1UL << idx);
 141                 s = &vm_phys_segs[idx];
 142                 if (low < s->end && high > s->start)
 143                         return (TRUE);
 144         }
 145         return (FALSE);
 146 }
 147
 148 /*
 149  * Outputs the state of the physical memory allocator, specifically,
 150  * the amount of physical memory in each free list.
 151  */
 152 static int
 153 sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS)
 154 {
 155         struct sbuf sbuf;
 156         struct vm_freelist *fl;
 157         int dom, error, flind, oind, pind;
 158
 159         error = sysctl_wire_old_buffer(req, 0);
 160         if (error != 0)
 161                 return (error);
 162         sbuf_new_for_sysctl(&sbuf, NULL, 128 * vm_ndomains, req);
 163         for (dom = 0; dom < vm_ndomains; dom++) {
 164                 sbuf_printf(&sbuf,"\nDOMAIN %d:\n", dom);
 165                 for (flind = 0; flind < vm_nfreelists; flind++) {
 166                         sbuf_printf(&sbuf, "\nFREE LIST %d:\n"
 167                             "\n  ORDER (SIZE)  |  NUMBER"
 168                             "\n              ", flind);
 169                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
 170                                 sbuf_printf(&sbuf, "  |  POOL %d", pind);
 171                         sbuf_printf(&sbuf, "\n--            ");
 172                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
 173                                 sbuf_printf(&sbuf, "-- --      ");
 174                         sbuf_printf(&sbuf, "--\n");
 175                         for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
 176                                 sbuf_printf(&sbuf, "  %2d (%6dK)", oind,
 177                                     1 << (PAGE_SHIFT - 10 + oind));
 178                                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 179                                 fl = vm_phys_free_queues[dom][flind][pind];
 180                                         sbuf_printf(&sbuf, "  |  %6d",
 181                                             fl[oind].lcnt);
 182                                 }
 183                                 sbuf_printf(&sbuf, "\n");
 184                         }
 185                 }
 186         }
 187         error = sbuf_finish(&sbuf);
 188         sbuf_delete(&sbuf);
 189         return (error);
 190 }
 191
 192 /*
 193  * Outputs the set of physical memory segments.
 194  */
 195 static int
 196 sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS)
 197 {
 198         struct sbuf sbuf;
 199         struct vm_phys_seg *seg;
 200         int error, segind;
 201
 202         error = sysctl_wire_old_buffer(req, 0);
 203         if (error != 0)
 204                 return (error);
 205         sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
 206         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 207                 sbuf_printf(&sbuf, "\nSEGMENT %d:\n\n", segind);
 208                 seg = &vm_phys_segs[segind];
 209                 sbuf_printf(&sbuf, "start:     %#jx\n",
 210                     (uintmax_t)seg->start);
 211                 sbuf_printf(&sbuf, "end:       %#jx\n",
 212                     (uintmax_t)seg->end);
 213                 sbuf_printf(&sbuf, "domain:    %d\n", seg->domain);
 214                 sbuf_printf(&sbuf, "free list: %p\n", seg->free_queues);
 215         }
 216         error = sbuf_finish(&sbuf);
 217         sbuf_delete(&sbuf);
 218         return (error);
 219 }
 220
 221 static void
 222 vm_freelist_add(struct vm_freelist *fl, vm_page_t m, int order, int tail)
 223 {
 224
 225         m->order = order;
 226         if (tail)
 227                 TAILQ_INSERT_TAIL(&fl[order].pl, m, plinks.q);
 228         else
 229                 TAILQ_INSERT_HEAD(&fl[order].pl, m, plinks.q);
 230         fl[order].lcnt++;
 231 }
 232
 233 static void
 234 vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order)
 235 {
 236
 237         TAILQ_REMOVE(&fl[order].pl, m, plinks.q);
 238         fl[order].lcnt--;
 239         m->order = VM_NFREEORDER;
 240 }
 241
 242 /*
 243  * Create a physical memory segment.
 244  */
 245 static void
 246 _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind, int domain)
 247 {
 248         struct vm_phys_seg *seg;
 249 #ifdef VM_PHYSSEG_SPARSE
 250         long pages;
 251         int segind;
 252
 253         pages = 0;
 254         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 255                 seg = &vm_phys_segs[segind];
 256                 pages += atop(seg->end - seg->start);
 257         }
 258 #endif
 259         KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX,
 260             ("vm_phys_create_seg: increase VM_PHYSSEG_MAX"));
 261         KASSERT(domain < vm_ndomains,
 262             ("vm_phys_create_seg: invalid domain provided"));
 263         seg = &vm_phys_segs[vm_phys_nsegs++];
 264         seg->start = start;
 265         seg->end = end;
 266         seg->domain = domain;
 267 #ifdef VM_PHYSSEG_SPARSE
 268         seg->first_page = &vm_page_array[pages];
 269 #else
 270         seg->first_page = PHYS_TO_VM_PAGE(start);
 271 #endif
 272         seg->free_queues = &vm_phys_free_queues[domain][flind];
 273 }
 274
 275 static void
 276 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind)
 277 {
 278         int i;
 279
 280         if (mem_affinity == NULL) {
 281                 _vm_phys_create_seg(start, end, flind, 0);
 282                 return;
 283         }
 284
 285         for (i = 0;; i++) {
 286                 if (mem_affinity[i].end == 0)
 287                         panic("Reached end of affinity info");
 288                 if (mem_affinity[i].end <= start)
 289                         continue;
 290                 if (mem_affinity[i].start > start)
 291                         panic("No affinity info for start %jx",
 292                             (uintmax_t)start);
 293                 if (mem_affinity[i].end >= end) {
 294                         _vm_phys_create_seg(start, end, flind,
 295                             mem_affinity[i].domain);
 296                         break;
 297                 }
 298                 _vm_phys_create_seg(start, mem_affinity[i].end, flind,
 299                     mem_affinity[i].domain);
 300                 start = mem_affinity[i].end;
 301         }
 302 }
 303
 304 /*
 305  * Initialize the physical memory allocator.
 306  */
 307 void
 308 vm_phys_init(void)
 309 {
 310         struct vm_freelist *fl;
 311         int dom, flind, i, oind, pind;
 312
 313         for (i = 0; phys_avail[i + 1] != 0; i += 2) {
 314 #ifdef  VM_FREELIST_ISADMA
 315                 if (phys_avail[i] < 16777216) {
 316                         if (phys_avail[i + 1] > 16777216) {
 317                                 vm_phys_create_seg(phys_avail[i], 16777216,
 318                                     VM_FREELIST_ISADMA);
 319                                 vm_phys_create_seg(16777216, phys_avail[i + 1],
 320                                     VM_FREELIST_DEFAULT);
 321                         } else {
 322                                 vm_phys_create_seg(phys_avail[i],
 323                                     phys_avail[i + 1], VM_FREELIST_ISADMA);
 324                         }
 325                         if (VM_FREELIST_ISADMA >= vm_nfreelists)
 326                                 vm_nfreelists = VM_FREELIST_ISADMA + 1;
 327                 } else
 328 #endif
 329 #ifdef  VM_FREELIST_HIGHMEM
 330                 if (phys_avail[i + 1] > VM_HIGHMEM_ADDRESS) {
 331                         if (phys_avail[i] < VM_HIGHMEM_ADDRESS) {
 332                                 vm_phys_create_seg(phys_avail[i],
 333                                     VM_HIGHMEM_ADDRESS, VM_FREELIST_DEFAULT);
 334                                 vm_phys_create_seg(VM_HIGHMEM_ADDRESS,
 335                                     phys_avail[i + 1], VM_FREELIST_HIGHMEM);
 336                         } else {
 337                                 vm_phys_create_seg(phys_avail[i],
 338                                     phys_avail[i + 1], VM_FREELIST_HIGHMEM);
 339                         }
 340                         if (VM_FREELIST_HIGHMEM >= vm_nfreelists)
 341                                 vm_nfreelists = VM_FREELIST_HIGHMEM + 1;
 342                 } else
 343 #endif
 344                 vm_phys_create_seg(phys_avail[i], phys_avail[i + 1],
 345                     VM_FREELIST_DEFAULT);
 346         }
 347         for (dom = 0; dom < vm_ndomains; dom++) {
 348                 for (flind = 0; flind < vm_nfreelists; flind++) {
 349                         for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 350                                 fl = vm_phys_free_queues[dom][flind][pind];
 351                                 for (oind = 0; oind < VM_NFREEORDER; oind++)
 352                                         TAILQ_INIT(&fl[oind].pl);
 353                         }
 354                 }
 355         }
 356         mtx_init(&vm_phys_fictitious_reg_mtx, "vmfctr", NULL, MTX_DEF);
 357 }
 358
 359 /*
 360  * Split a contiguous, power of two-sized set of physical pages.
 361  */
 362 static __inline void
 363 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order)
 364 {
 365         vm_page_t m_buddy;
 366
 367         while (oind > order) {
 368                 oind--;
 369                 m_buddy = &m[1 << oind];
 370                 KASSERT(m_buddy->order == VM_NFREEORDER,
 371                     ("vm_phys_split_pages: page %p has unexpected order %d",
 372                     m_buddy, m_buddy->order));
 373                 vm_freelist_add(fl, m_buddy, oind, 0);
 374         }
 375 }
 376
 377 /*
 378  * Initialize a physical page and add it to the free lists.
 379  */
 380 void
 381 vm_phys_add_page(vm_paddr_t pa)
 382 {
 383         vm_page_t m;
 384         struct vm_domain *vmd;
 385
 386         vm_cnt.v_page_count++;
 387         m = vm_phys_paddr_to_vm_page(pa);
 388         m->phys_addr = pa;
 389         m->queue = PQ_NONE;
 390         m->segind = vm_phys_paddr_to_segind(pa);
 391         vmd = vm_phys_domain(m);
 392         vmd->vmd_page_count++;
 393         vmd->vmd_segs |= 1UL << m->segind;
 394         KASSERT(m->order == VM_NFREEORDER,
 395             ("vm_phys_add_page: page %p has unexpected order %d",
 396             m, m->order));
 397         m->pool = VM_FREEPOOL_DEFAULT;
 398         pmap_page_init(m);
 399         mtx_lock(&vm_page_queue_free_mtx);
 400         vm_phys_freecnt_adj(m, 1);
 401         vm_phys_free_pages(m, 0);
 402         mtx_unlock(&vm_page_queue_free_mtx);
 403 }
 404
 405 /*
 406  * Allocate a contiguous, power of two-sized set of physical pages
 407  * from the free lists.
 408  *
 409  * The free page queues must be locked.
 410  */
 411 vm_page_t
 412 vm_phys_alloc_pages(int pool, int order)
 413 {
 414         vm_page_t m;
 415         int dom, domain, flind;
 416
 417         KASSERT(pool < VM_NFREEPOOL,
 418             ("vm_phys_alloc_pages: pool %d is out of range", pool));
 419         KASSERT(order < VM_NFREEORDER,
 420             ("vm_phys_alloc_pages: order %d is out of range", order));
 421
 422         for (dom = 0; dom < vm_ndomains; dom++) {
 423                 domain = vm_rr_selectdomain();
 424                 for (flind = 0; flind < vm_nfreelists; flind++) {
 425                         m = vm_phys_alloc_domain_pages(domain, flind, pool,
 426                             order);
 427                         if (m != NULL)
 428                                 return (m);
 429                 }
 430         }
 431         return (NULL);
 432 }
 433
 434 /*
 435  * Find and dequeue a free page on the given free list, with the
 436  * specified pool and order
 437  */
 438 vm_page_t
 439 vm_phys_alloc_freelist_pages(int flind, int pool, int order)
 440 {
 441         vm_page_t m;
 442         int dom, domain;
 443
 444         KASSERT(flind < VM_NFREELIST,
 445             ("vm_phys_alloc_freelist_pages: freelist %d is out of range", flind));
 446         KASSERT(pool < VM_NFREEPOOL,
 447             ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
 448         KASSERT(order < VM_NFREEORDER,
 449             ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
 450
 451         for (dom = 0; dom < vm_ndomains; dom++) {
 452                 domain = vm_rr_selectdomain();
 453                 m = vm_phys_alloc_domain_pages(domain, flind, pool, order);
 454                 if (m != NULL)
 455                         return (m);
 456         }
 457         return (NULL);
 458 }
 459
 460 static vm_page_t
 461 vm_phys_alloc_domain_pages(int domain, int flind, int pool, int order)
 462 {
 463         struct vm_freelist *fl;
 464         struct vm_freelist *alt;
 465         int oind, pind;
 466         vm_page_t m;
 467
 468         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 469         fl = &vm_phys_free_queues[domain][flind][pool][0];
 470         for (oind = order; oind < VM_NFREEORDER; oind++) {
 471                 m = TAILQ_FIRST(&fl[oind].pl);
 472                 if (m != NULL) {
 473                         vm_freelist_rem(fl, m, oind);
 474                         vm_phys_split_pages(m, oind, fl, order);
 475                         return (m);
 476                 }
 477         }
 478
 479         /*
 480          * The given pool was empty.  Find the largest
 481          * contiguous, power-of-two-sized set of pages in any
 482          * pool.  Transfer these pages to the given pool, and
 483          * use them to satisfy the allocation.
 484          */
 485         for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
 486                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 487                         alt = &vm_phys_free_queues[domain][flind][pind][0];
 488                         m = TAILQ_FIRST(&alt[oind].pl);
 489                         if (m != NULL) {
 490                                 vm_freelist_rem(alt, m, oind);
 491                                 vm_phys_set_pool(pool, m, oind);
 492                                 vm_phys_split_pages(m, oind, fl, order);
 493                                 return (m);
 494                         }
 495                 }
 496         }
 497         return (NULL);
 498 }
 499
 500 /*
 501  * Find the vm_page corresponding to the given physical address.
 502  */
 503 vm_page_t
 504 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
 505 {
 506         struct vm_phys_seg *seg;
 507         int segind;
 508
 509         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 510                 seg = &vm_phys_segs[segind];
 511                 if (pa >= seg->start && pa < seg->end)
 512                         return (&seg->first_page[atop(pa - seg->start)]);
 513         }
 514         return (NULL);
 515 }
 516
 517 vm_page_t
 518 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
 519 {
 520         struct vm_phys_fictitious_seg *seg;
 521         vm_page_t m;
 522         int segind;
 523
 524         m = NULL;
 525         for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
 526                 seg = &vm_phys_fictitious_segs[segind];
 527                 if (pa >= seg->start && pa < seg->end) {
 528                         m = &seg->first_page[atop(pa - seg->start)];
 529                         KASSERT((m->flags & PG_FICTITIOUS) != 0,
 530                             ("%p not fictitious", m));
 531                         break;
 532                 }
 533         }
 534         return (m);
 535 }
 536
 537 int
 538 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
 539     vm_memattr_t memattr)
 540 {
 541         struct vm_phys_fictitious_seg *seg;
 542         vm_page_t fp;
 543         long i, page_count;
 544         int segind;
 545 #ifdef VM_PHYSSEG_DENSE
 546         long pi;
 547         boolean_t malloced;
 548 #endif
 549
 550         page_count = (end - start) / PAGE_SIZE;
 551
 552 #ifdef VM_PHYSSEG_DENSE
 553         pi = atop(start);
 554         if (pi >= first_page && atop(end) < vm_page_array_size) {
 555                 fp = &vm_page_array[pi - first_page];
 556                 malloced = FALSE;
 557         } else
 558 #endif
 559         {
 560                 fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
 561                     M_WAITOK | M_ZERO);
 562 #ifdef VM_PHYSSEG_DENSE
 563                 malloced = TRUE;
 564 #endif
 565         }
 566         for (i = 0; i < page_count; i++) {
 567                 vm_page_initfake(&fp[i], start + PAGE_SIZE * i, memattr);
 568                 fp[i].oflags &= ~VPO_UNMANAGED;
 569                 fp[i].busy_lock = VPB_UNBUSIED;
 570         }
 571         mtx_lock(&vm_phys_fictitious_reg_mtx);
 572         for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
 573                 seg = &vm_phys_fictitious_segs[segind];
 574                 if (seg->start == 0 && seg->end == 0) {
 575                         seg->start = start;
 576                         seg->end = end;
 577                         seg->first_page = fp;
 578                         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 579                         return (0);
 580                 }
 581         }
 582         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 583 #ifdef VM_PHYSSEG_DENSE
 584         if (malloced)
 585 #endif
 586                 free(fp, M_FICT_PAGES);
 587         return (EBUSY);
 588 }
 589
 590 void
 591 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
 592 {
 593         struct vm_phys_fictitious_seg *seg;
 594         vm_page_t fp;
 595         int segind;
 596 #ifdef VM_PHYSSEG_DENSE
 597         long pi;
 598 #endif
 599
 600 #ifdef VM_PHYSSEG_DENSE
 601         pi = atop(start);
 602 #endif
 603
 604         mtx_lock(&vm_phys_fictitious_reg_mtx);
 605         for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
 606                 seg = &vm_phys_fictitious_segs[segind];
 607                 if (seg->start == start && seg->end == end) {
 608                         seg->start = seg->end = 0;
 609                         fp = seg->first_page;
 610                         seg->first_page = NULL;
 611                         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 612 #ifdef VM_PHYSSEG_DENSE
 613                         if (pi < first_page || atop(end) >= vm_page_array_size)
 614 #endif
 615                                 free(fp, M_FICT_PAGES);
 616                         return;
 617                 }
 618         }
 619         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 620         KASSERT(0, ("Unregistering not registered fictitious range"));
 621 }
 622
 623 /*
 624  * Find the segment containing the given physical address.
 625  */
 626 static int
 627 vm_phys_paddr_to_segind(vm_paddr_t pa)
 628 {
 629         struct vm_phys_seg *seg;
 630         int segind;
 631
 632         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 633                 seg = &vm_phys_segs[segind];
 634                 if (pa >= seg->start && pa < seg->end)
 635                         return (segind);
 636         }
 637         panic("vm_phys_paddr_to_segind: paddr %#jx is not in any segment" ,
 638             (uintmax_t)pa);
 639 }
 640
 641 /*
 642  * Free a contiguous, power of two-sized set of physical pages.
 643  *
 644  * The free page queues must be locked.
 645  */
 646 void
 647 vm_phys_free_pages(vm_page_t m, int order)
 648 {
 649         struct vm_freelist *fl;
 650         struct vm_phys_seg *seg;
 651         vm_paddr_t pa;
 652         vm_page_t m_buddy;
 653
 654         KASSERT(m->order == VM_NFREEORDER,
 655             ("vm_phys_free_pages: page %p has unexpected order %d",
 656             m, m->order));
 657         KASSERT(m->pool < VM_NFREEPOOL,
 658             ("vm_phys_free_pages: page %p has unexpected pool %d",
 659             m, m->pool));
 660         KASSERT(order < VM_NFREEORDER,
 661             ("vm_phys_free_pages: order %d is out of range", order));
 662         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 663         seg = &vm_phys_segs[m->segind];
 664         if (order < VM_NFREEORDER - 1) {
 665                 pa = VM_PAGE_TO_PHYS(m);
 666                 do {
 667                         pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
 668                         if (pa < seg->start || pa >= seg->end)
 669                                 break;
 670                         m_buddy = &seg->first_page[atop(pa - seg->start)];
 671                         if (m_buddy->order != order)
 672                                 break;
 673                         fl = (*seg->free_queues)[m_buddy->pool];
 674                         vm_freelist_rem(fl, m_buddy, order);
 675                         if (m_buddy->pool != m->pool)
 676                                 vm_phys_set_pool(m->pool, m_buddy, order);
 677                         order++;
 678                         pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
 679                         m = &seg->first_page[atop(pa - seg->start)];
 680                 } while (order < VM_NFREEORDER - 1);
 681         }
 682         fl = (*seg->free_queues)[m->pool];
 683         vm_freelist_add(fl, m, order, 1);
 684 }
 685
 686 /*
 687  * Free a contiguous, arbitrarily sized set of physical pages.
 688  *
 689  * The free page queues must be locked.
 690  */
 691 void
 692 vm_phys_free_contig(vm_page_t m, u_long npages)
 693 {
 694         u_int n;
 695         int order;
 696
 697         /*
 698          * Avoid unnecessary coalescing by freeing the pages in the largest
 699          * possible power-of-two-sized subsets.
 700          */
 701         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 702         for (;; npages -= n) {
 703                 /*
 704                  * Unsigned "min" is used here so that "order" is assigned
 705                  * "VM_NFREEORDER - 1" when "m"'s physical address is zero
 706                  * or the low-order bits of its physical address are zero
 707                  * because the size of a physical address exceeds the size of
 708                  * a long.
 709                  */
 710                 order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
 711                     VM_NFREEORDER - 1);
 712                 n = 1 << order;
 713                 if (npages < n)
 714                         break;
 715                 vm_phys_free_pages(m, order);
 716                 m += n;
 717         }
 718         /* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */
 719         for (; npages > 0; npages -= n) {
 720                 order = flsl(npages) - 1;
 721                 n = 1 << order;
 722                 vm_phys_free_pages(m, order);
 723                 m += n;
 724         }
 725 }
 726
 727 /*
 728  * Set the pool for a contiguous, power of two-sized set of physical pages.
 729  */
 730 void
 731 vm_phys_set_pool(int pool, vm_page_t m, int order)
 732 {
 733         vm_page_t m_tmp;
 734
 735         for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
 736                 m_tmp->pool = pool;
 737 }
 738
 739 /*
 740  * Search for the given physical page "m" in the free lists.  If the search
 741  * succeeds, remove "m" from the free lists and return TRUE.  Otherwise, return
 742  * FALSE, indicating that "m" is not in the free lists.
 743  *
 744  * The free page queues must be locked.
 745  */
 746 boolean_t
 747 vm_phys_unfree_page(vm_page_t m)
 748 {
 749         struct vm_freelist *fl;
 750         struct vm_phys_seg *seg;
 751         vm_paddr_t pa, pa_half;
 752         vm_page_t m_set, m_tmp;
 753         int order;
 754
 755         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 756
 757         /*
 758          * First, find the contiguous, power of two-sized set of free
 759          * physical pages containing the given physical page "m" and
 760          * assign it to "m_set".
 761          */
 762         seg = &vm_phys_segs[m->segind];
 763         for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
 764             order < VM_NFREEORDER - 1; ) {
 765                 order++;
 766                 pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
 767                 if (pa >= seg->start)
 768                         m_set = &seg->first_page[atop(pa - seg->start)];
 769                 else
 770                         return (FALSE);
 771         }
 772         if (m_set->order < order)
 773                 return (FALSE);
 774         if (m_set->order == VM_NFREEORDER)
 775                 return (FALSE);
 776         KASSERT(m_set->order < VM_NFREEORDER,
 777             ("vm_phys_unfree_page: page %p has unexpected order %d",
 778             m_set, m_set->order));
 779
 780         /*
 781          * Next, remove "m_set" from the free lists.  Finally, extract
 782          * "m" from "m_set" using an iterative algorithm: While "m_set"
 783          * is larger than a page, shrink "m_set" by returning the half
 784          * of "m_set" that does not contain "m" to the free lists.
 785          */
 786         fl = (*seg->free_queues)[m_set->pool];
 787         order = m_set->order;
 788         vm_freelist_rem(fl, m_set, order);
 789         while (order > 0) {
 790                 order--;
 791                 pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
 792                 if (m->phys_addr < pa_half)
 793                         m_tmp = &seg->first_page[atop(pa_half - seg->start)];
 794                 else {
 795                         m_tmp = m_set;
 796                         m_set = &seg->first_page[atop(pa_half - seg->start)];
 797                 }
 798                 vm_freelist_add(fl, m_tmp, order, 0);
 799         }
 800         KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
 801         return (TRUE);
 802 }
 803
 804 /*
 805  * Try to zero one physical page.  Used by an idle priority thread.
 806  */
 807 boolean_t
 808 vm_phys_zero_pages_idle(void)
 809 {
 810         static struct vm_freelist *fl;
 811         static int flind, oind, pind;
 812         vm_page_t m, m_tmp;
 813         int domain;
 814
 815         domain = vm_rr_selectdomain();
 816         fl = vm_phys_free_queues[domain][0][0];
 817         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 818         for (;;) {
 819                 TAILQ_FOREACH_REVERSE(m, &fl[oind].pl, pglist, plinks.q) {
 820                         for (m_tmp = m; m_tmp < &m[1 << oind]; m_tmp++) {
 821                                 if ((m_tmp->flags & (PG_CACHED | PG_ZERO)) == 0) {
 822                                         vm_phys_unfree_page(m_tmp);
 823                                         vm_phys_freecnt_adj(m, -1);
 824                                         mtx_unlock(&vm_page_queue_free_mtx);
 825                                         pmap_zero_page_idle(m_tmp);
 826                                         m_tmp->flags |= PG_ZERO;
 827                                         mtx_lock(&vm_page_queue_free_mtx);
 828                                         vm_phys_freecnt_adj(m, 1);
 829                                         vm_phys_free_pages(m_tmp, 0);
 830                                         vm_page_zero_count++;
 831                                         cnt_prezero++;
 832                                         return (TRUE);
 833                                 }
 834                         }
 835                 }
 836                 oind++;
 837                 if (oind == VM_NFREEORDER) {
 838                         oind = 0;
 839                         pind++;
 840                         if (pind == VM_NFREEPOOL) {
 841                                 pind = 0;
 842                                 flind++;
 843                                 if (flind == vm_nfreelists)
 844                                         flind = 0;
 845                         }
 846                         fl = vm_phys_free_queues[domain][flind][pind];
 847                 }
 848         }
 849 }
 850
 851 /*
 852  * Allocate a contiguous set of physical pages of the given size
 853  * "npages" from the free lists.  All of the physical pages must be at
 854  * or above the given physical address "low" and below the given
 855  * physical address "high".  The given value "alignment" determines the
 856  * alignment of the first physical page in the set.  If the given value
 857  * "boundary" is non-zero, then the set of physical pages cannot cross
 858  * any physical address boundary that is a multiple of that value.  Both
 859  * "alignment" and "boundary" must be a power of two.
 860  */
 861 vm_page_t
 862 vm_phys_alloc_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
 863     u_long alignment, vm_paddr_t boundary)
 864 {
 865         struct vm_freelist *fl;
 866         struct vm_phys_seg *seg;
 867         vm_paddr_t pa, pa_last, size;
 868         vm_page_t m, m_ret;
 869         u_long npages_end;
 870         int dom, domain, flind, oind, order, pind;
 871
 872         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 873         size = npages << PAGE_SHIFT;
 874         KASSERT(size != 0,
 875             ("vm_phys_alloc_contig: size must not be 0"));
 876         KASSERT((alignment & (alignment - 1)) == 0,
 877             ("vm_phys_alloc_contig: alignment must be a power of 2"));
 878         KASSERT((boundary & (boundary - 1)) == 0,
 879             ("vm_phys_alloc_contig: boundary must be a power of 2"));
 880         /* Compute the queue that is the best fit for npages. */
 881         for (order = 0; (1 << order) < npages; order++);
 882         dom = 0;
 883 restartdom:
 884         domain = vm_rr_selectdomain();
 885         for (flind = 0; flind < vm_nfreelists; flind++) {
 886                 for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER; oind++) {
 887                         for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 888                                 fl = &vm_phys_free_queues[domain][flind][pind][0];
 889                                 TAILQ_FOREACH(m_ret, &fl[oind].pl, plinks.q) {
 890                                         /*
 891                                          * A free list may contain physical pages
 892                                          * from one or more segments.
 893                                          */
 894                                         seg = &vm_phys_segs[m_ret->segind];
 895                                         if (seg->start > high ||
 896                                             low >= seg->end)
 897                                                 continue;
 898
 899                                         /*
 900                                          * Is the size of this allocation request
 901                                          * larger than the largest block size?
 902                                          */
 903                                         if (order >= VM_NFREEORDER) {
 904                                                 /*
 905                                                  * Determine if a sufficient number
 906                                                  * of subsequent blocks to satisfy
 907                                                  * the allocation request are free.
 908                                                  */
 909                                                 pa = VM_PAGE_TO_PHYS(m_ret);
 910                                                 pa_last = pa + size;
 911                                                 for (;;) {
 912                                                         pa += 1 << (PAGE_SHIFT + VM_NFREEORDER - 1);
 913                                                         if (pa >= pa_last)
 914                                                                 break;
 915                                                         if (pa < seg->start ||
 916                                                             pa >= seg->end)
 917                                                                 break;
 918                                                         m = &seg->first_page[atop(pa - seg->start)];
 919                                                         if (m->order != VM_NFREEORDER - 1)
 920                                                                 break;
 921                                                 }
 922                                                 /* If not, continue to the next block. */
 923                                                 if (pa < pa_last)
 924                                                         continue;
 925                                         }
 926
 927                                         /*
 928                                          * Determine if the blocks are within the given range,
 929                                          * satisfy the given alignment, and do not cross the
 930                                          * given boundary.
 931                                          */
 932                                         pa = VM_PAGE_TO_PHYS(m_ret);
 933                                         if (pa >= low &&
 934                                             pa + size <= high &&
 935                                             (pa & (alignment - 1)) == 0 &&
 936                                             ((pa ^ (pa + size - 1)) & ~(boundary - 1)) == 0)
 937                                                 goto done;
 938                                 }
 939                         }
 940                 }
 941         }
 942         if (++dom < vm_ndomains)
 943                 goto restartdom;
 944         return (NULL);
 945 done:
 946         for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
 947                 fl = (*seg->free_queues)[m->pool];
 948                 vm_freelist_rem(fl, m, m->order);
 949         }
 950         if (m_ret->pool != VM_FREEPOOL_DEFAULT)
 951                 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m_ret, oind);
 952         fl = (*seg->free_queues)[m_ret->pool];
 953         vm_phys_split_pages(m_ret, oind, fl, order);
 954         /* Return excess pages to the free lists. */
 955         npages_end = roundup2(npages, 1 << imin(oind, order));
 956         if (npages < npages_end)
 957                 vm_phys_free_contig(&m_ret[npages], npages_end - npages);
 958         return (m_ret);
 959 }
 960
 961 #ifdef DDB
 962 /*
 963  * Show the number of physical pages in each of the free lists.
 964  */
 965 DB_SHOW_COMMAND(freepages, db_show_freepages)
 966 {
 967         struct vm_freelist *fl;
 968         int flind, oind, pind, dom;
 969
 970         for (dom = 0; dom < vm_ndomains; dom++) {
 971                 db_printf("DOMAIN: %d\n", dom);
 972                 for (flind = 0; flind < vm_nfreelists; flind++) {
 973                         db_printf("FREE LIST %d:\n"
 974                             "\n  ORDER (SIZE)  |  NUMBER"
 975                             "\n              ", flind);
 976                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
 977                                 db_printf("  |  POOL %d", pind);
 978                         db_printf("\n--            ");
 979                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
 980                                 db_printf("-- --      ");
 981                         db_printf("--\n");
 982                         for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
 983                                 db_printf("  %2.2d (%6.6dK)", oind,
 984                                     1 << (PAGE_SHIFT - 10 + oind));
 985                                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 986                                 fl = vm_phys_free_queues[dom][flind][pind];
 987                                         db_printf("  |  %6.6d", fl[oind].lcnt);
 988                                 }
 989                                 db_printf("\n");
 990                         }
 991                         db_printf("\n");
 992                 }
 993                 db_printf("\n");
 994         }
 995 }
 996 #endif