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
 250         KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX,
 251             ("vm_phys_create_seg: increase VM_PHYSSEG_MAX"));
 252         KASSERT(domain < vm_ndomains,
 253             ("vm_phys_create_seg: invalid domain provided"));
 254         seg = &vm_phys_segs[vm_phys_nsegs++];
 255         while (seg > vm_phys_segs && (seg - 1)->start >= end) {
 256                 *seg = *(seg - 1);
 257                 seg--;
 258         }
 259         seg->start = start;
 260         seg->end = end;
 261         seg->domain = domain;
 262         seg->free_queues = &vm_phys_free_queues[domain][flind];
 263 }
 264
 265 static void
 266 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind)
 267 {
 268         int i;
 269
 270         if (mem_affinity == NULL) {
 271                 _vm_phys_create_seg(start, end, flind, 0);
 272                 return;
 273         }
 274
 275         for (i = 0;; i++) {
 276                 if (mem_affinity[i].end == 0)
 277                         panic("Reached end of affinity info");
 278                 if (mem_affinity[i].end <= start)
 279                         continue;
 280                 if (mem_affinity[i].start > start)
 281                         panic("No affinity info for start %jx",
 282                             (uintmax_t)start);
 283                 if (mem_affinity[i].end >= end) {
 284                         _vm_phys_create_seg(start, end, flind,
 285                             mem_affinity[i].domain);
 286                         break;
 287                 }
 288                 _vm_phys_create_seg(start, mem_affinity[i].end, flind,
 289                     mem_affinity[i].domain);
 290                 start = mem_affinity[i].end;
 291         }
 292 }
 293
 294 /*
 295  * Add a physical memory segment.
 296  */
 297 void
 298 vm_phys_add_seg(vm_paddr_t start, vm_paddr_t end)
 299 {
 300
 301         KASSERT((start & PAGE_MASK) == 0,
 302             ("vm_phys_define_seg: start is not page aligned"));
 303         KASSERT((end & PAGE_MASK) == 0,
 304             ("vm_phys_define_seg: end is not page aligned"));
 305 #ifdef  VM_FREELIST_ISADMA
 306         if (start < 16777216) {
 307                 if (end > 16777216) {
 308                         vm_phys_create_seg(start, 16777216,
 309                             VM_FREELIST_ISADMA);
 310                         vm_phys_create_seg(16777216, end, VM_FREELIST_DEFAULT);
 311                 } else
 312                         vm_phys_create_seg(start, end, VM_FREELIST_ISADMA);
 313                 if (VM_FREELIST_ISADMA >= vm_nfreelists)
 314                         vm_nfreelists = VM_FREELIST_ISADMA + 1;
 315         } else
 316 #endif
 317 #ifdef  VM_FREELIST_HIGHMEM
 318         if (end > VM_HIGHMEM_ADDRESS) {
 319                 if (start < VM_HIGHMEM_ADDRESS) {
 320                         vm_phys_create_seg(start, VM_HIGHMEM_ADDRESS,
 321                             VM_FREELIST_DEFAULT);
 322                         vm_phys_create_seg(VM_HIGHMEM_ADDRESS, end,
 323                             VM_FREELIST_HIGHMEM);
 324                 } else
 325                         vm_phys_create_seg(start, end, VM_FREELIST_HIGHMEM);
 326                 if (VM_FREELIST_HIGHMEM >= vm_nfreelists)
 327                         vm_nfreelists = VM_FREELIST_HIGHMEM + 1;
 328         } else
 329 #endif
 330         vm_phys_create_seg(start, end, VM_FREELIST_DEFAULT);
 331 }
 332
 333 /*
 334  * Initialize the physical memory allocator.
 335  */
 336 void
 337 vm_phys_init(void)
 338 {
 339         struct vm_freelist *fl;
 340         struct vm_phys_seg *seg;
 341 #ifdef VM_PHYSSEG_SPARSE
 342         long pages;
 343 #endif
 344         int dom, flind, oind, pind, segind;
 345
 346 #ifdef VM_PHYSSEG_SPARSE
 347         pages = 0;
 348 #endif
 349         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 350                 seg = &vm_phys_segs[segind];
 351 #ifdef VM_PHYSSEG_SPARSE
 352                 seg->first_page = &vm_page_array[pages];
 353                 pages += atop(seg->end - seg->start);
 354 #else
 355                 seg->first_page = PHYS_TO_VM_PAGE(seg->start);
 356 #endif
 357         }
 358         for (dom = 0; dom < vm_ndomains; dom++) {
 359                 for (flind = 0; flind < vm_nfreelists; flind++) {
 360                         for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 361                                 fl = vm_phys_free_queues[dom][flind][pind];
 362                                 for (oind = 0; oind < VM_NFREEORDER; oind++)
 363                                         TAILQ_INIT(&fl[oind].pl);
 364                         }
 365                 }
 366         }
 367         mtx_init(&vm_phys_fictitious_reg_mtx, "vmfctr", NULL, MTX_DEF);
 368 }
 369
 370 /*
 371  * Split a contiguous, power of two-sized set of physical pages.
 372  */
 373 static __inline void
 374 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order)
 375 {
 376         vm_page_t m_buddy;
 377
 378         while (oind > order) {
 379                 oind--;
 380                 m_buddy = &m[1 << oind];
 381                 KASSERT(m_buddy->order == VM_NFREEORDER,
 382                     ("vm_phys_split_pages: page %p has unexpected order %d",
 383                     m_buddy, m_buddy->order));
 384                 vm_freelist_add(fl, m_buddy, oind, 0);
 385         }
 386 }
 387
 388 /*
 389  * Initialize a physical page and add it to the free lists.
 390  */
 391 void
 392 vm_phys_add_page(vm_paddr_t pa)
 393 {
 394         vm_page_t m;
 395         struct vm_domain *vmd;
 396
 397         cnt.v_page_count++;
 398         m = vm_phys_paddr_to_vm_page(pa);
 399         m->phys_addr = pa;
 400         m->queue = PQ_NONE;
 401         m->segind = vm_phys_paddr_to_segind(pa);
 402         vmd = vm_phys_domain(m);
 403         vmd->vmd_page_count++;
 404         vmd->vmd_segs |= 1UL << m->segind;
 405         m->flags = PG_FREE;
 406         KASSERT(m->order == VM_NFREEORDER,
 407             ("vm_phys_add_page: page %p has unexpected order %d",
 408             m, m->order));
 409         m->pool = VM_FREEPOOL_DEFAULT;
 410         pmap_page_init(m);
 411         mtx_lock(&vm_page_queue_free_mtx);
 412         vm_phys_freecnt_adj(m, 1);
 413         vm_phys_free_pages(m, 0);
 414         mtx_unlock(&vm_page_queue_free_mtx);
 415 }
 416
 417 /*
 418  * Allocate a contiguous, power of two-sized set of physical pages
 419  * from the free lists.
 420  *
 421  * The free page queues must be locked.
 422  */
 423 vm_page_t
 424 vm_phys_alloc_pages(int pool, int order)
 425 {
 426         vm_page_t m;
 427         int dom, domain, flind;
 428
 429         KASSERT(pool < VM_NFREEPOOL,
 430             ("vm_phys_alloc_pages: pool %d is out of range", pool));
 431         KASSERT(order < VM_NFREEORDER,
 432             ("vm_phys_alloc_pages: order %d is out of range", order));
 433
 434         for (dom = 0; dom < vm_ndomains; dom++) {
 435                 domain = vm_rr_selectdomain();
 436                 for (flind = 0; flind < vm_nfreelists; flind++) {
 437                         m = vm_phys_alloc_domain_pages(domain, flind, pool,
 438                             order);
 439                         if (m != NULL)
 440                                 return (m);
 441                 }
 442         }
 443         return (NULL);
 444 }
 445
 446 /*
 447  * Find and dequeue a free page on the given free list, with the
 448  * specified pool and order
 449  */
 450 vm_page_t
 451 vm_phys_alloc_freelist_pages(int flind, int pool, int order)
 452 {
 453         vm_page_t m;
 454         int dom, domain;
 455
 456         KASSERT(flind < VM_NFREELIST,
 457             ("vm_phys_alloc_freelist_pages: freelist %d is out of range", flind));
 458         KASSERT(pool < VM_NFREEPOOL,
 459             ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
 460         KASSERT(order < VM_NFREEORDER,
 461             ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
 462
 463         for (dom = 0; dom < vm_ndomains; dom++) {
 464                 domain = vm_rr_selectdomain();
 465                 m = vm_phys_alloc_domain_pages(domain, flind, pool, order);
 466                 if (m != NULL)
 467                         return (m);
 468         }
 469         return (NULL);
 470 }
 471
 472 static vm_page_t
 473 vm_phys_alloc_domain_pages(int domain, int flind, int pool, int order)
 474 {
 475         struct vm_freelist *fl;
 476         struct vm_freelist *alt;
 477         int oind, pind;
 478         vm_page_t m;
 479
 480         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 481         fl = &vm_phys_free_queues[domain][flind][pool][0];
 482         for (oind = order; oind < VM_NFREEORDER; oind++) {
 483                 m = TAILQ_FIRST(&fl[oind].pl);
 484                 if (m != NULL) {
 485                         vm_freelist_rem(fl, m, oind);
 486                         vm_phys_split_pages(m, oind, fl, order);
 487                         return (m);
 488                 }
 489         }
 490
 491         /*
 492          * The given pool was empty.  Find the largest
 493          * contiguous, power-of-two-sized set of pages in any
 494          * pool.  Transfer these pages to the given pool, and
 495          * use them to satisfy the allocation.
 496          */
 497         for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
 498                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 499                         alt = &vm_phys_free_queues[domain][flind][pind][0];
 500                         m = TAILQ_FIRST(&alt[oind].pl);
 501                         if (m != NULL) {
 502                                 vm_freelist_rem(alt, m, oind);
 503                                 vm_phys_set_pool(pool, m, oind);
 504                                 vm_phys_split_pages(m, oind, fl, order);
 505                                 return (m);
 506                         }
 507                 }
 508         }
 509         return (NULL);
 510 }
 511
 512 /*
 513  * Find the vm_page corresponding to the given physical address.
 514  */
 515 vm_page_t
 516 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
 517 {
 518         struct vm_phys_seg *seg;
 519         int segind;
 520
 521         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 522                 seg = &vm_phys_segs[segind];
 523                 if (pa >= seg->start && pa < seg->end)
 524                         return (&seg->first_page[atop(pa - seg->start)]);
 525         }
 526         return (NULL);
 527 }
 528
 529 vm_page_t
 530 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
 531 {
 532         struct vm_phys_fictitious_seg *seg;
 533         vm_page_t m;
 534         int segind;
 535
 536         m = NULL;
 537         for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
 538                 seg = &vm_phys_fictitious_segs[segind];
 539                 if (pa >= seg->start && pa < seg->end) {
 540                         m = &seg->first_page[atop(pa - seg->start)];
 541                         KASSERT((m->flags & PG_FICTITIOUS) != 0,
 542                             ("%p not fictitious", m));
 543                         break;
 544                 }
 545         }
 546         return (m);
 547 }
 548
 549 int
 550 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
 551     vm_memattr_t memattr)
 552 {
 553         struct vm_phys_fictitious_seg *seg;
 554         vm_page_t fp;
 555         long i, page_count;
 556         int segind;
 557 #ifdef VM_PHYSSEG_DENSE
 558         long pi;
 559         boolean_t malloced;
 560 #endif
 561
 562         page_count = (end - start) / PAGE_SIZE;
 563
 564 #ifdef VM_PHYSSEG_DENSE
 565         pi = atop(start);
 566         if (pi >= first_page && pi < vm_page_array_size + first_page) {
 567                 if (atop(end) >= vm_page_array_size + first_page)
 568                         return (EINVAL);
 569                 fp = &vm_page_array[pi - first_page];
 570                 malloced = FALSE;
 571         } else
 572 #endif
 573         {
 574                 fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
 575                     M_WAITOK | M_ZERO);
 576 #ifdef VM_PHYSSEG_DENSE
 577                 malloced = TRUE;
 578 #endif
 579         }
 580         for (i = 0; i < page_count; i++) {
 581                 vm_page_initfake(&fp[i], start + PAGE_SIZE * i, memattr);
 582                 fp[i].oflags &= ~VPO_UNMANAGED;
 583                 fp[i].busy_lock = VPB_UNBUSIED;
 584         }
 585         mtx_lock(&vm_phys_fictitious_reg_mtx);
 586         for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
 587                 seg = &vm_phys_fictitious_segs[segind];
 588                 if (seg->start == 0 && seg->end == 0) {
 589                         seg->start = start;
 590                         seg->end = end;
 591                         seg->first_page = fp;
 592                         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 593                         return (0);
 594                 }
 595         }
 596         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 597 #ifdef VM_PHYSSEG_DENSE
 598         if (malloced)
 599 #endif
 600                 free(fp, M_FICT_PAGES);
 601         return (EBUSY);
 602 }
 603
 604 void
 605 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
 606 {
 607         struct vm_phys_fictitious_seg *seg;
 608         vm_page_t fp;
 609         int segind;
 610 #ifdef VM_PHYSSEG_DENSE
 611         long pi;
 612 #endif
 613
 614 #ifdef VM_PHYSSEG_DENSE
 615         pi = atop(start);
 616 #endif
 617
 618         mtx_lock(&vm_phys_fictitious_reg_mtx);
 619         for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
 620                 seg = &vm_phys_fictitious_segs[segind];
 621                 if (seg->start == start && seg->end == end) {
 622                         seg->start = seg->end = 0;
 623                         fp = seg->first_page;
 624                         seg->first_page = NULL;
 625                         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 626 #ifdef VM_PHYSSEG_DENSE
 627                         if (pi < first_page || atop(end) >= vm_page_array_size)
 628 #endif
 629                                 free(fp, M_FICT_PAGES);
 630                         return;
 631                 }
 632         }
 633         mtx_unlock(&vm_phys_fictitious_reg_mtx);
 634         KASSERT(0, ("Unregistering not registered fictitious range"));
 635 }
 636
 637 /*
 638  * Find the segment containing the given physical address.
 639  */
 640 static int
 641 vm_phys_paddr_to_segind(vm_paddr_t pa)
 642 {
 643         struct vm_phys_seg *seg;
 644         int segind;
 645
 646         for (segind = 0; segind < vm_phys_nsegs; segind++) {
 647                 seg = &vm_phys_segs[segind];
 648                 if (pa >= seg->start && pa < seg->end)
 649                         return (segind);
 650         }
 651         panic("vm_phys_paddr_to_segind: paddr %#jx is not in any segment" ,
 652             (uintmax_t)pa);
 653 }
 654
 655 /*
 656  * Free a contiguous, power of two-sized set of physical pages.
 657  *
 658  * The free page queues must be locked.
 659  */
 660 void
 661 vm_phys_free_pages(vm_page_t m, int order)
 662 {
 663         struct vm_freelist *fl;
 664         struct vm_phys_seg *seg;
 665         vm_paddr_t pa;
 666         vm_page_t m_buddy;
 667
 668         KASSERT(m->order == VM_NFREEORDER,
 669             ("vm_phys_free_pages: page %p has unexpected order %d",
 670             m, m->order));
 671         KASSERT(m->pool < VM_NFREEPOOL,
 672             ("vm_phys_free_pages: page %p has unexpected pool %d",
 673             m, m->pool));
 674         KASSERT(order < VM_NFREEORDER,
 675             ("vm_phys_free_pages: order %d is out of range", order));
 676         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 677         seg = &vm_phys_segs[m->segind];
 678         if (order < VM_NFREEORDER - 1) {
 679                 pa = VM_PAGE_TO_PHYS(m);
 680                 do {
 681                         pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
 682                         if (pa < seg->start || pa >= seg->end)
 683                                 break;
 684                         m_buddy = &seg->first_page[atop(pa - seg->start)];
 685                         if (m_buddy->order != order)
 686                                 break;
 687                         fl = (*seg->free_queues)[m_buddy->pool];
 688                         vm_freelist_rem(fl, m_buddy, order);
 689                         if (m_buddy->pool != m->pool)
 690                                 vm_phys_set_pool(m->pool, m_buddy, order);
 691                         order++;
 692                         pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
 693                         m = &seg->first_page[atop(pa - seg->start)];
 694                 } while (order < VM_NFREEORDER - 1);
 695         }
 696         fl = (*seg->free_queues)[m->pool];
 697         vm_freelist_add(fl, m, order, 1);
 698 }
 699
 700 /*
 701  * Free a contiguous, arbitrarily sized set of physical pages.
 702  *
 703  * The free page queues must be locked.
 704  */
 705 void
 706 vm_phys_free_contig(vm_page_t m, u_long npages)
 707 {
 708         u_int n;
 709         int order;
 710
 711         /*
 712          * Avoid unnecessary coalescing by freeing the pages in the largest
 713          * possible power-of-two-sized subsets.
 714          */
 715         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 716         for (;; npages -= n) {
 717                 /*
 718                  * Unsigned "min" is used here so that "order" is assigned
 719                  * "VM_NFREEORDER - 1" when "m"'s physical address is zero
 720                  * or the low-order bits of its physical address are zero
 721                  * because the size of a physical address exceeds the size of
 722                  * a long.
 723                  */
 724                 order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
 725                     VM_NFREEORDER - 1);
 726                 n = 1 << order;
 727                 if (npages < n)
 728                         break;
 729                 vm_phys_free_pages(m, order);
 730                 m += n;
 731         }
 732         /* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */
 733         for (; npages > 0; npages -= n) {
 734                 order = flsl(npages) - 1;
 735                 n = 1 << order;
 736                 vm_phys_free_pages(m, order);
 737                 m += n;
 738         }
 739 }
 740
 741 /*
 742  * Set the pool for a contiguous, power of two-sized set of physical pages.
 743  */
 744 void
 745 vm_phys_set_pool(int pool, vm_page_t m, int order)
 746 {
 747         vm_page_t m_tmp;
 748
 749         for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
 750                 m_tmp->pool = pool;
 751 }
 752
 753 /*
 754  * Search for the given physical page "m" in the free lists.  If the search
 755  * succeeds, remove "m" from the free lists and return TRUE.  Otherwise, return
 756  * FALSE, indicating that "m" is not in the free lists.
 757  *
 758  * The free page queues must be locked.
 759  */
 760 boolean_t
 761 vm_phys_unfree_page(vm_page_t m)
 762 {
 763         struct vm_freelist *fl;
 764         struct vm_phys_seg *seg;
 765         vm_paddr_t pa, pa_half;
 766         vm_page_t m_set, m_tmp;
 767         int order;
 768
 769         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 770
 771         /*
 772          * First, find the contiguous, power of two-sized set of free
 773          * physical pages containing the given physical page "m" and
 774          * assign it to "m_set".
 775          */
 776         seg = &vm_phys_segs[m->segind];
 777         for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
 778             order < VM_NFREEORDER - 1; ) {
 779                 order++;
 780                 pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
 781                 if (pa >= seg->start)
 782                         m_set = &seg->first_page[atop(pa - seg->start)];
 783                 else
 784                         return (FALSE);
 785         }
 786         if (m_set->order < order)
 787                 return (FALSE);
 788         if (m_set->order == VM_NFREEORDER)
 789                 return (FALSE);
 790         KASSERT(m_set->order < VM_NFREEORDER,
 791             ("vm_phys_unfree_page: page %p has unexpected order %d",
 792             m_set, m_set->order));
 793
 794         /*
 795          * Next, remove "m_set" from the free lists.  Finally, extract
 796          * "m" from "m_set" using an iterative algorithm: While "m_set"
 797          * is larger than a page, shrink "m_set" by returning the half
 798          * of "m_set" that does not contain "m" to the free lists.
 799          */
 800         fl = (*seg->free_queues)[m_set->pool];
 801         order = m_set->order;
 802         vm_freelist_rem(fl, m_set, order);
 803         while (order > 0) {
 804                 order--;
 805                 pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
 806                 if (m->phys_addr < pa_half)
 807                         m_tmp = &seg->first_page[atop(pa_half - seg->start)];
 808                 else {
 809                         m_tmp = m_set;
 810                         m_set = &seg->first_page[atop(pa_half - seg->start)];
 811                 }
 812                 vm_freelist_add(fl, m_tmp, order, 0);
 813         }
 814         KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
 815         return (TRUE);
 816 }
 817
 818 /*
 819  * Try to zero one physical page.  Used by an idle priority thread.
 820  */
 821 boolean_t
 822 vm_phys_zero_pages_idle(void)
 823 {
 824         static struct vm_freelist *fl;
 825         static int flind, oind, pind;
 826         vm_page_t m, m_tmp;
 827         int domain;
 828
 829         domain = vm_rr_selectdomain();
 830         fl = vm_phys_free_queues[domain][0][0];
 831         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 832         for (;;) {
 833                 TAILQ_FOREACH_REVERSE(m, &fl[oind].pl, pglist, plinks.q) {
 834                         for (m_tmp = m; m_tmp < &m[1 << oind]; m_tmp++) {
 835                                 if ((m_tmp->flags & (PG_CACHED | PG_ZERO)) == 0) {
 836                                         vm_phys_unfree_page(m_tmp);
 837                                         vm_phys_freecnt_adj(m, -1);
 838                                         mtx_unlock(&vm_page_queue_free_mtx);
 839                                         pmap_zero_page_idle(m_tmp);
 840                                         m_tmp->flags |= PG_ZERO;
 841                                         mtx_lock(&vm_page_queue_free_mtx);
 842                                         vm_phys_freecnt_adj(m, 1);
 843                                         vm_phys_free_pages(m_tmp, 0);
 844                                         vm_page_zero_count++;
 845                                         cnt_prezero++;
 846                                         return (TRUE);
 847                                 }
 848                         }
 849                 }
 850                 oind++;
 851                 if (oind == VM_NFREEORDER) {
 852                         oind = 0;
 853                         pind++;
 854                         if (pind == VM_NFREEPOOL) {
 855                                 pind = 0;
 856                                 flind++;
 857                                 if (flind == vm_nfreelists)
 858                                         flind = 0;
 859                         }
 860                         fl = vm_phys_free_queues[domain][flind][pind];
 861                 }
 862         }
 863 }
 864
 865 /*
 866  * Allocate a contiguous set of physical pages of the given size
 867  * "npages" from the free lists.  All of the physical pages must be at
 868  * or above the given physical address "low" and below the given
 869  * physical address "high".  The given value "alignment" determines the
 870  * alignment of the first physical page in the set.  If the given value
 871  * "boundary" is non-zero, then the set of physical pages cannot cross
 872  * any physical address boundary that is a multiple of that value.  Both
 873  * "alignment" and "boundary" must be a power of two.
 874  */
 875 vm_page_t
 876 vm_phys_alloc_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
 877     u_long alignment, vm_paddr_t boundary)
 878 {
 879         struct vm_freelist *fl;
 880         struct vm_phys_seg *seg;
 881         vm_paddr_t pa, pa_last, size;
 882         vm_page_t m, m_ret;
 883         u_long npages_end;
 884         int dom, domain, flind, oind, order, pind;
 885
 886         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
 887         size = npages << PAGE_SHIFT;
 888         KASSERT(size != 0,
 889             ("vm_phys_alloc_contig: size must not be 0"));
 890         KASSERT((alignment & (alignment - 1)) == 0,
 891             ("vm_phys_alloc_contig: alignment must be a power of 2"));
 892         KASSERT((boundary & (boundary - 1)) == 0,
 893             ("vm_phys_alloc_contig: boundary must be a power of 2"));
 894         /* Compute the queue that is the best fit for npages. */
 895         for (order = 0; (1 << order) < npages; order++);
 896         dom = 0;
 897 restartdom:
 898         domain = vm_rr_selectdomain();
 899         for (flind = 0; flind < vm_nfreelists; flind++) {
 900                 for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER; oind++) {
 901                         for (pind = 0; pind < VM_NFREEPOOL; pind++) {
 902                                 fl = &vm_phys_free_queues[domain][flind][pind][0];
 903                                 TAILQ_FOREACH(m_ret, &fl[oind].pl, plinks.q) {
 904                                         /*
 905                                          * A free list may contain physical pages
 906                                          * from one or more segments.
 907                                          */
 908                                         seg = &vm_phys_segs[m_ret->segind];
 909                                         if (seg->start > high ||
 910                                             low >= seg->end)
 911                                                 continue;
 912
 913                                         /*
 914                                          * Is the size of this allocation request
 915                                          * larger than the largest block size?
 916                                          */
 917                                         if (order >= VM_NFREEORDER) {
 918                                                 /*
 919                                                  * Determine if a sufficient number
 920                                                  * of subsequent blocks to satisfy
 921                                                  * the allocation request are free.
 922                                                  */
 923                                                 pa = VM_PAGE_TO_PHYS(m_ret);
 924                                                 pa_last = pa + size;
 925                                                 for (;;) {
 926                                                         pa += 1 << (PAGE_SHIFT + VM_NFREEORDER - 1);
 927                                                         if (pa >= pa_last)
 928                                                                 break;
 929                                                         if (pa < seg->start ||
 930                                                             pa >= seg->end)
 931                                                                 break;
 932                                                         m = &seg->first_page[atop(pa - seg->start)];
 933                                                         if (m->order != VM_NFREEORDER - 1)
 934                                                                 break;
 935                                                 }
 936                                                 /* If not, continue to the next block. */
 937                                                 if (pa < pa_last)
 938                                                         continue;
 939                                         }
 940
 941                                         /*
 942                                          * Determine if the blocks are within the given range,
 943                                          * satisfy the given alignment, and do not cross the
 944                                          * given boundary.
 945                                          */
 946                                         pa = VM_PAGE_TO_PHYS(m_ret);
 947                                         if (pa >= low &&
 948                                             pa + size <= high &&
 949                                             (pa & (alignment - 1)) == 0 &&
 950                                             ((pa ^ (pa + size - 1)) & ~(boundary - 1)) == 0)
 951                                                 goto done;
 952                                 }
 953                         }
 954                 }
 955         }
 956         if (++dom < vm_ndomains)
 957                 goto restartdom;
 958         return (NULL);
 959 done:
 960         for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
 961                 fl = (*seg->free_queues)[m->pool];
 962                 vm_freelist_rem(fl, m, m->order);
 963         }
 964         if (m_ret->pool != VM_FREEPOOL_DEFAULT)
 965                 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m_ret, oind);
 966         fl = (*seg->free_queues)[m_ret->pool];
 967         vm_phys_split_pages(m_ret, oind, fl, order);
 968         /* Return excess pages to the free lists. */
 969         npages_end = roundup2(npages, 1 << imin(oind, order));
 970         if (npages < npages_end)
 971                 vm_phys_free_contig(&m_ret[npages], npages_end - npages);
 972         return (m_ret);
 973 }
 974
 975 #ifdef DDB
 976 /*
 977  * Show the number of physical pages in each of the free lists.
 978  */
 979 DB_SHOW_COMMAND(freepages, db_show_freepages)
 980 {
 981         struct vm_freelist *fl;
 982         int flind, oind, pind, dom;
 983
 984         for (dom = 0; dom < vm_ndomains; dom++) {
 985                 db_printf("DOMAIN: %d\n", dom);
 986                 for (flind = 0; flind < vm_nfreelists; flind++) {
 987                         db_printf("FREE LIST %d:\n"
 988                             "\n  ORDER (SIZE)  |  NUMBER"
 989                             "\n              ", flind);
 990                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
 991                                 db_printf("  |  POOL %d", pind);
 992                         db_printf("\n--            ");
 993                         for (pind = 0; pind < VM_NFREEPOOL; pind++)
 994                                 db_printf("-- --      ");
 995                         db_printf("--\n");
 996                         for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
 997                                 db_printf("  %2.2d (%6.6dK)", oind,
 998                                     1 << (PAGE_SHIFT - 10 + oind));
 999                                 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1000                                 fl = vm_phys_free_queues[dom][flind][pind];
1001                                         db_printf("  |  %6.6d", fl[oind].lcnt);
1002                                 }
1003                                 db_printf("\n");
1004                         }
1005                         db_printf("\n");
1006                 }
1007                 db_printf("\n");
1008         }
1009 }
1010 #endif