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         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         m->flags = PG_FREE;
 395         KASSERT(m->order == VM_NFREEORDER,
 396             ("vm_phys_add_page: page %p has unexpected order %d",
 397             m, m->order));
 398         m->pool = VM_FREEPOOL_DEFAULT;
 399         pmap_page_init(m);
 400         mtx_lock(&vm_page_queue_free_mtx);
 401         vm_phys_freecnt_adj(m, 1);
 402         vm_phys_free_pages(m, 0);
 403         mtx_unlock(&vm_page_queue_free_mtx);
 404 }
 405
 406 /*
 407  * Allocate a contiguous, power of two-sized set of physical pages
 408  * from the free lists.
 409  *
 410  * The free page queues must be locked.
 411  */
 412 vm_page_t
 413 vm_phys_alloc_pages(int pool, int order)
 414 {
 415         vm_page_t m;
 416         int dom, domain, flind;
 417
 418         KASSERT(pool < VM_NFREEPOOL,
 419             ("vm_phys_alloc_pages: pool %d is out of range", pool));
 420         KASSERT(order < VM_NFREEORDER,
 421             ("vm_phys_alloc_pages: order %d is out of range", order));
 422
 423         for (dom = 0; dom < vm_ndomains; dom++) {
 424                 domain = vm_rr_selectdomain();
 425                 for (flind = 0; flind < vm_nfreelists; flind++) {
 426                         m = vm_phys_alloc_domain_pages(domain, flind, pool,
 427                             order);
 428                         if (m != NULL)
 429                                 return (m);
 430                 }
 431         }
 432         return (NULL);
 433 }
 434
 435 /*
 436  * Find and dequeue a free page on the given free list, with the
 437  * specified pool and order
 438  */
 439 vm_page_t
 440 vm_phys_alloc_freelist_pages(int flind, int pool, int order)
 441 {
 442         vm_page_t m;
 443         int dom, domain;
 444
 445         KASSERT(flind < VM_NFREELIST,
 446             ("vm_phys_alloc_freelist_pages: freelist %d is out of range", flind));
 447         KASSERT(pool < VM_NFREEPOOL,
 448             ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
 449         KASSERT(order < VM_NFREEORDER,
 450             ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
 451
 452         for (dom = 0; dom < vm_ndomains; dom++) {
 453                 domain = vm_rr_selectdomain();
 454                 m = vm_phys_alloc_domain_pages(domain, flind, pool, order);
 455                 if (m != NULL)
 456                         return (m);
 457         }
 458         return (NULL);
 459 }
 460
 461 static vm_page_t
 462 vm_phys_alloc_domain_pages(int domain, int flind, int pool, int order)
 463 {
 464         struct vm_freelist *fl;
 465         struct vm_freelist *alt;
 466         int oind, pind;
 467         vm_page_t m;
 468
 469         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 470         fl = &vm_phys_free_queues[domain][flind][pool][0];
 471         for (oind = order; oind < VM_NFREEORDER; oind++) {
 472                 m = TAILQ_FIRST(&fl[oind].pl);
 473                 if (m != NULL) {
 474                         vm_freelist_rem(fl, m, oind);
 475                         vm_phys_split_pages(m, oind, fl, order);
 476                         return (m);
 477                 }
 478         }
 479
 480         /*
 481          * The given pool was empty.  Find the largest
 482          * contiguous, power-of-two-sized set of pages in any
 483          * pool.  Transfer these pages to the given pool, and
 484          * use them to satisfy the allocation.
 485          */
 486         for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
 487                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 488                         alt = &vm_phys_free_queues[domain][flind][pind][0];
 489                         m = TAILQ_FIRST(&alt[oind].pl);
 490                         if (m != NULL) {
 491                                 vm_freelist_rem(alt, m, oind);
 492                                 vm_phys_set_pool(pool, m, oind);
 493                                 vm_phys_split_pages(m, oind, fl, order);
 494                                 return (m);
 495                         }
 496                 }
 497         }
 498         return (NULL);
 499 }
 500
 501 /*
 502  * Find the vm_page corresponding to the given physical address.
 503  */
 504 vm_page_t
 505 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
 506 {
 507         struct vm_phys_seg *seg;
 508         int segind;
 509
 510         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 511                 seg = &vm_phys_segs[segind];
 512                 if (pa >= seg->start && pa < seg->end)
 513                         return (&seg->first_page[atop(pa - seg->start)]);
 514         }
 515         return (NULL);
 516 }
 517
 518 vm_page_t
 519 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
 520 {
 521         struct vm_phys_fictitious_seg *seg;
 522         vm_page_t m;
 523         int segind;
 524
 525         m = NULL;
 526         for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
 527                 seg = &vm_phys_fictitious_segs[segind];
 528                 if (pa >= seg->start && pa < seg->end) {
 529                         m = &seg->first_page[atop(pa - seg->start)];
 530                         KASSERT((m->flags & PG_FICTITIOUS) != 0,
 531                             ("%p not fictitious", m));
 532                         break;
 533                 }
 534         }
 535         return (m);
 536 }
 537
 538 int
 539 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
 540     vm_memattr_t memattr)
 541 {
 542         struct vm_phys_fictitious_seg *seg;
 543         vm_page_t fp;
 544         long i, page_count;
 545         int segind;
 546 #ifdef VM_PHYSSEG_DENSE
 547         long pi;
 548         boolean_t malloced;
 549 #endif
 550
 551         page_count = (end - start) / PAGE_SIZE;
 552
 553 #ifdef VM_PHYSSEG_DENSE
 554         pi = atop(start);
 555         if (pi >= first_page && atop(end) < vm_page_array_size) {
 556                 fp = &vm_page_array[pi - first_page];
 557                 malloced = FALSE;
 558         } else
 559 #endif
 560         {
 561                 fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
 562                     M_WAITOK | M_ZERO);
 563 #ifdef VM_PHYSSEG_DENSE
 564                 malloced = TRUE;
 565 #endif
 566         }
 567         for (i = 0; i < page_count; i++) {
 568                 vm_page_initfake(&fp[i], start + PAGE_SIZE * i, memattr);
 569                 fp[i].oflags &= ~VPO_UNMANAGED;
 570                 fp[i].busy_lock = VPB_UNBUSIED;
 571         }
 572         mtx_lock(&vm_phys_fictitious_reg_mtx);
 573         for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
 574                 seg = &vm_phys_fictitious_segs[segind];
 575                 if (seg->start == 0 && seg->end == 0) {
 576                         seg->start = start;
 577                         seg->end = end;
 578                         seg->first_page = fp;
 579                         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 580                         return (0);
 581                 }
 582         }
 583         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 584 #ifdef VM_PHYSSEG_DENSE
 585         if (malloced)
 586 #endif
 587                 free(fp, M_FICT_PAGES);
 588         return (EBUSY);
 589 }
 590
 591 void
 592 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
 593 {
 594         struct vm_phys_fictitious_seg *seg;
 595         vm_page_t fp;
 596         int segind;
 597 #ifdef VM_PHYSSEG_DENSE
 598         long pi;
 599 #endif
 600
 601 #ifdef VM_PHYSSEG_DENSE
 602         pi = atop(start);
 603 #endif
 604
 605         mtx_lock(&vm_phys_fictitious_reg_mtx);
 606         for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
 607                 seg = &vm_phys_fictitious_segs[segind];
 608                 if (seg->start == start && seg->end == end) {
 609                         seg->start = seg->end = 0;
 610                         fp = seg->first_page;
 611                         seg->first_page = NULL;
 612                         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 613 #ifdef VM_PHYSSEG_DENSE
 614                         if (pi < first_page || atop(end) >= vm_page_array_size)
 615 #endif
 616                                 free(fp, M_FICT_PAGES);
 617                         return;
 618                 }
 619         }
 620         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 621         KASSERT(0, ("Unregistering not registered fictitious range"));
 622 }
 623
 624 /*
 625  * Find the segment containing the given physical address.
 626  */
 627 static int
 628 vm_phys_paddr_to_segind(vm_paddr_t pa)
 629 {
 630         struct vm_phys_seg *seg;
 631         int segind;
 632
 633         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 634                 seg = &vm_phys_segs[segind];
 635                 if (pa >= seg->start && pa < seg->end)
 636                         return (segind);
 637         }
 638         panic("vm_phys_paddr_to_segind: paddr %#jx is not in any segment" ,
 639             (uintmax_t)pa);
 640 }
 641
 642 /*
 643  * Free a contiguous, power of two-sized set of physical pages.
 644  *
 645  * The free page queues must be locked.
 646  */
 647 void
 648 vm_phys_free_pages(vm_page_t m, int order)
 649 {
 650         struct vm_freelist *fl;
 651         struct vm_phys_seg *seg;
 652         vm_paddr_t pa;
 653         vm_page_t m_buddy;
 654
 655         KASSERT(m->order == VM_NFREEORDER,
 656             ("vm_phys_free_pages: page %p has unexpected order %d",
 657             m, m->order));
 658         KASSERT(m->pool < VM_NFREEPOOL,
 659             ("vm_phys_free_pages: page %p has unexpected pool %d",
 660             m, m->pool));
 661         KASSERT(order < VM_NFREEORDER,
 662             ("vm_phys_free_pages: order %d is out of range", order));
 663         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 664         seg = &vm_phys_segs[m->segind];
 665         if (order < VM_NFREEORDER - 1) {
 666                 pa = VM_PAGE_TO_PHYS(m);
 667                 do {
 668                         pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
 669                         if (pa < seg->start || pa >= seg->end)
 670                                 break;
 671                         m_buddy = &seg->first_page[atop(pa - seg->start)];
 672                         if (m_buddy->order != order)
 673                                 break;
 674                         fl = (*seg->free_queues)[m_buddy->pool];
 675                         vm_freelist_rem(fl, m_buddy, order);
 676                         if (m_buddy->pool != m->pool)
 677                                 vm_phys_set_pool(m->pool, m_buddy, order);
 678                         order++;
 679                         pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
 680                         m = &seg->first_page[atop(pa - seg->start)];
 681                 } while (order < VM_NFREEORDER - 1);
 682         }
 683         fl = (*seg->free_queues)[m->pool];
 684         vm_freelist_add(fl, m, order, 1);
 685 }
 686
 687 /*
 688  * Free a contiguous, arbitrarily sized set of physical pages.
 689  *
 690  * The free page queues must be locked.
 691  */
 692 void
 693 vm_phys_free_contig(vm_page_t m, u_long npages)
 694 {
 695         u_int n;
 696         int order;
 697
 698         /*
 699          * Avoid unnecessary coalescing by freeing the pages in the largest
 700          * possible power-of-two-sized subsets.
 701          */
 702         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 703         for (;; npages -= n) {
 704                 /*
 705                  * Unsigned "min" is used here so that "order" is assigned
 706                  * "VM_NFREEORDER - 1" when "m"'s physical address is zero
 707                  * or the low-order bits of its physical address are zero
 708                  * because the size of a physical address exceeds the size of
 709                  * a long.
 710                  */
 711                 order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
 712                     VM_NFREEORDER - 1);
 713                 n = 1 << order;
 714                 if (npages < n)
 715                         break;
 716                 vm_phys_free_pages(m, order);
 717                 m += n;
 718         }
 719         /* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */
 720         for (; npages > 0; npages -= n) {
 721                 order = flsl(npages) - 1;
 722                 n = 1 << order;
 723                 vm_phys_free_pages(m, order);
 724                 m += n;
 725         }
 726 }
 727
 728 /*
 729  * Set the pool for a contiguous, power of two-sized set of physical pages.
 730  */
 731 void
 732 vm_phys_set_pool(int pool, vm_page_t m, int order)
 733 {
 734         vm_page_t m_tmp;
 735
 736         for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
 737                 m_tmp->pool = pool;
 738 }
 739
 740 /*
 741  * Search for the given physical page "m" in the free lists.  If the search
 742  * succeeds, remove "m" from the free lists and return TRUE.  Otherwise, return
 743  * FALSE, indicating that "m" is not in the free lists.
 744  *
 745  * The free page queues must be locked.
 746  */
 747 boolean_t
 748 vm_phys_unfree_page(vm_page_t m)
 749 {
 750         struct vm_freelist *fl;
 751         struct vm_phys_seg *seg;
 752         vm_paddr_t pa, pa_half;
 753         vm_page_t m_set, m_tmp;
 754         int order;
 755
 756         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 757
 758         /*
 759          * First, find the contiguous, power of two-sized set of free
 760          * physical pages containing the given physical page "m" and
 761          * assign it to "m_set".
 762          */
 763         seg = &vm_phys_segs[m->segind];
 764         for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
 765             order < VM_NFREEORDER - 1; ) {
 766                 order++;
 767                 pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
 768                 if (pa >= seg->start)
 769                         m_set = &seg->first_page[atop(pa - seg->start)];
 770                 else
 771                         return (FALSE);
 772         }
 773         if (m_set->order < order)
 774                 return (FALSE);
 775         if (m_set->order == VM_NFREEORDER)
 776                 return (FALSE);
 777         KASSERT(m_set->order < VM_NFREEORDER,
 778             ("vm_phys_unfree_page: page %p has unexpected order %d",
 779             m_set, m_set->order));
 780
 781         /*
 782          * Next, remove "m_set" from the free lists.  Finally, extract
 783          * "m" from "m_set" using an iterative algorithm: While "m_set"
 784          * is larger than a page, shrink "m_set" by returning the half
 785          * of "m_set" that does not contain "m" to the free lists.
 786          */
 787         fl = (*seg->free_queues)[m_set->pool];
 788         order = m_set->order;
 789         vm_freelist_rem(fl, m_set, order);
 790         while (order > 0) {
 791                 order--;
 792                 pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
 793                 if (m->phys_addr < pa_half)
 794                         m_tmp = &seg->first_page[atop(pa_half - seg->start)];
 795                 else {
 796                         m_tmp = m_set;
 797                         m_set = &seg->first_page[atop(pa_half - seg->start)];
 798                 }
 799                 vm_freelist_add(fl, m_tmp, order, 0);
 800         }
 801         KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
 802         return (TRUE);
 803 }
 804
 805 /*
 806  * Try to zero one physical page.  Used by an idle priority thread.
 807  */
 808 boolean_t
 809 vm_phys_zero_pages_idle(void)
 810 {
 811         static struct vm_freelist *fl;
 812         static int flind, oind, pind;
 813         vm_page_t m, m_tmp;
 814         int domain;
 815
 816         domain = vm_rr_selectdomain();
 817         fl = vm_phys_free_queues[domain][0][0];
 818         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 819         for (;;) {
 820                 TAILQ_FOREACH_REVERSE(m, &fl[oind].pl, pglist, plinks.q) {
 821                         for (m_tmp = m; m_tmp < &m[1 << oind]; m_tmp++) {
 822                                 if ((m_tmp->flags & (PG_CACHED | PG_ZERO)) == 0) {
 823                                         vm_phys_unfree_page(m_tmp);
 824                                         vm_phys_freecnt_adj(m, -1);
 825                                         mtx_unlock(&vm_page_queue_free_mtx);
 826                                         pmap_zero_page_idle(m_tmp);
 827                                         m_tmp->flags |= PG_ZERO;
 828                                         mtx_lock(&vm_page_queue_free_mtx);
 829                                         vm_phys_freecnt_adj(m, 1);
 830                                         vm_phys_free_pages(m_tmp, 0);
 831                                         vm_page_zero_count++;
 832                                         cnt_prezero++;
 833                                         return (TRUE);
 834                                 }
 835                         }
 836                 }
 837                 oind++;
 838                 if (oind == VM_NFREEORDER) {
 839                         oind = 0;
 840                         pind++;
 841                         if (pind == VM_NFREEPOOL) {
 842                                 pind = 0;
 843                                 flind++;
 844                                 if (flind == vm_nfreelists)
 845                                         flind = 0;
 846                         }
 847                         fl = vm_phys_free_queues[domain][flind][pind];
 848                 }
 849         }
 850 }
 851
 852 /*
 853  * Allocate a contiguous set of physical pages of the given size
 854  * "npages" from the free lists.  All of the physical pages must be at
 855  * or above the given physical address "low" and below the given
 856  * physical address "high".  The given value "alignment" determines the
 857  * alignment of the first physical page in the set.  If the given value
 858  * "boundary" is non-zero, then the set of physical pages cannot cross
 859  * any physical address boundary that is a multiple of that value.  Both
 860  * "alignment" and "boundary" must be a power of two.
 861  */
 862 vm_page_t
 863 vm_phys_alloc_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
 864     u_long alignment, vm_paddr_t boundary)
 865 {
 866         struct vm_freelist *fl;
 867         struct vm_phys_seg *seg;
 868         vm_paddr_t pa, pa_last, size;
 869         vm_page_t m, m_ret;
 870         u_long npages_end;
 871         int dom, domain, flind, oind, order, pind;
 872
 873         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 874         size = npages << PAGE_SHIFT;
 875         KASSERT(size != 0,
 876             ("vm_phys_alloc_contig: size must not be 0"));
 877         KASSERT((alignment & (alignment - 1)) == 0,
 878             ("vm_phys_alloc_contig: alignment must be a power of 2"));
 879         KASSERT((boundary & (boundary - 1)) == 0,
 880             ("vm_phys_alloc_contig: boundary must be a power of 2"));
 881         /* Compute the queue that is the best fit for npages. */
 882         for (order = 0; (1 << order) < npages; order++);
 883         dom = 0;
 884 restartdom:
 885         domain = vm_rr_selectdomain();
 886         for (flind = 0; flind < vm_nfreelists; flind++) {
 887                 for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER; oind++) {
 888                         for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 889                                 fl = &vm_phys_free_queues[domain][flind][pind][0];
 890                                 TAILQ_FOREACH(m_ret, &fl[oind].pl, plinks.q) {
 891                                         /*
 892                                          * A free list may contain physical pages
 893                                          * from one or more segments.
 894                                          */
 895                                         seg = &vm_phys_segs[m_ret->segind];
 896                                         if (seg->start > high ||
 897                                             low >= seg->end)
 898                                                 continue;
 899
 900                                         /*
 901                                          * Is the size of this allocation request
 902                                          * larger than the largest block size?
 903                                          */
 904                                         if (order >= VM_NFREEORDER) {
 905                                                 /*
 906                                                  * Determine if a sufficient number
 907                                                  * of subsequent blocks to satisfy
 908                                                  * the allocation request are free.
 909                                                  */
 910                                                 pa = VM_PAGE_TO_PHYS(m_ret);
 911                                                 pa_last = pa + size;
 912                                                 for (;;) {
 913                                                         pa += 1 << (PAGE_SHIFT + VM_NFREEORDER - 1);
 914                                                         if (pa >= pa_last)
 915                                                                 break;
 916                                                         if (pa < seg->start ||
 917                                                             pa >= seg->end)
 918                                                                 break;
 919                                                         m = &seg->first_page[atop(pa - seg->start)];
 920                                                         if (m->order != VM_NFREEORDER - 1)
 921                                                                 break;
 922                                                 }
 923                                                 /* If not, continue to the next block. */
 924                                                 if (pa < pa_last)
 925                                                         continue;
 926                                         }
 927
 928                                         /*
 929                                          * Determine if the blocks are within the given range,
 930                                          * satisfy the given alignment, and do not cross the
 931                                          * given boundary.
 932                                          */
 933                                         pa = VM_PAGE_TO_PHYS(m_ret);
 934                                         if (pa >= low &&
 935                                             pa + size <= high &&
 936                                             (pa & (alignment - 1)) == 0 &&
 937                                             ((pa ^ (pa + size - 1)) & ~(boundary - 1)) == 0)
 938                                                 goto done;
 939                                 }
 940                         }
 941                 }
 942         }
 943         if (++dom < vm_ndomains)
 944                 goto restartdom;
 945         return (NULL);
 946 done:
 947         for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
 948                 fl = (*seg->free_queues)[m->pool];
 949                 vm_freelist_rem(fl, m, m->order);
 950         }
 951         if (m_ret->pool != VM_FREEPOOL_DEFAULT)
 952                 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m_ret, oind);
 953         fl = (*seg->free_queues)[m_ret->pool];
 954         vm_phys_split_pages(m_ret, oind, fl, order);
 955         /* Return excess pages to the free lists. */
 956         npages_end = roundup2(npages, 1 << imin(oind, order));
 957         if (npages < npages_end)
 958                 vm_phys_free_contig(&m_ret[npages], npages_end - npages);
 959         return (m_ret);
 960 }
 961
 962 #ifdef DDB
 963 /*
 964  * Show the number of physical pages in each of the free lists.
 965  */
 966 DB_SHOW_COMMAND(freepages, db_show_freepages)
 967 {
 968         struct vm_freelist *fl;
 969         int flind, oind, pind, dom;
 970
 971         for (dom = 0; dom < vm_ndomains; dom++) {
 972                 db_printf("DOMAIN: %d\n", dom);
 973                 for (flind = 0; flind < vm_nfreelists; flind++) {
 974                         db_printf("FREE LIST %d:\n"
 975                             "\n  ORDER (SIZE)  |  NUMBER"
 976                             "\n              ", flind);
 977                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
 978                                 db_printf("  |  POOL %d", pind);
 979                         db_printf("\n--            ");
 980                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
 981                                 db_printf("-- --      ");
 982                         db_printf("--\n");
 983                         for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
 984                                 db_printf("  %2.2d (%6.6dK)", oind,
 985                                     1 << (PAGE_SHIFT - 10 + oind));
 986                                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 987                                 fl = vm_phys_free_queues[dom][flind][pind];
 988                                         db_printf("  |  %6.6d", fl[oind].lcnt);
 989                                 }
 990                                 db_printf("\n");
 991                         }
 992                         db_printf("\n");
 993                 }
 994                 db_printf("\n");
 995         }
 996 }
 997 #endif