sys/vm/vm_radix.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2013 EMC Corp.
   5  * Copyright (c) 2011 Jeffrey Roberson <jeff@freebsd.org>
   6  * Copyright (c) 2008 Mayur Shardul <mayur.shardul@gmail.com>
   7  * All rights reserved.
   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 AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  28  * SUCH DAMAGE.
  29  *
  30  */
  31
  32 /*
  33  * Path-compressed radix trie implementation.
  34  * The following code is not generalized into a general purpose library
  35  * because there are way too many parameters embedded that should really
  36  * be decided by the library consumers.  At the same time, consumers
  37  * of this code must achieve highest possible performance.
  38  *
  39  * The implementation takes into account the following rationale:
  40  * - Size of the nodes should be as small as possible but still big enough
  41  *   to avoid a large maximum depth for the trie.  This is a balance
  42  *   between the necessity to not wire too much physical memory for the nodes
  43  *   and the necessity to avoid too much cache pollution during the trie
  44  *   operations.
  45  * - There is not a huge bias toward the number of lookup operations over
  46  *   the number of insert and remove operations.  This basically implies
  47  *   that optimizations supposedly helping one operation but hurting the
  48  *   other might be carefully evaluated.
  49  * - On average not many nodes are expected to be fully populated, hence
  50  *   level compression may just complicate things.
  51  */
  52
  53 #include <sys/cdefs.h>
  54 __FBSDID("$FreeBSD$");
  55
  56 #include "opt_ddb.h"
  57
  58 #include <sys/param.h>
  59 #include <sys/systm.h>
  60 #include <sys/kernel.h>
  61 #include <sys/proc.h>
  62 #include <sys/vmmeter.h>
  63 #include <sys/smr.h>
  64 #include <sys/smr_types.h>
  65
  66 #include <vm/uma.h>
  67 #include <vm/vm.h>
  68 #include <vm/vm_param.h>
  69 #include <vm/vm_object.h>
  70 #include <vm/vm_page.h>
  71 #include <vm/vm_radix.h>
  72
  73 #ifdef DDB
  74 #include <ddb/ddb.h>
  75 #endif
  76
  77 /*
  78  * These widths should allow the pointers to a node's children to fit within
  79  * a single cache line.  The extra levels from a narrow width should not be
  80  * a problem thanks to path compression.
  81  */
  82 #ifdef __LP64__
  83 #define VM_RADIX_WIDTH  4
  84 #else
  85 #define VM_RADIX_WIDTH  3
  86 #endif
  87
  88 #define VM_RADIX_COUNT  (1 << VM_RADIX_WIDTH)
  89 #define VM_RADIX_MASK   (VM_RADIX_COUNT - 1)
  90 #define VM_RADIX_LIMIT                                                  \
  91         (howmany(sizeof(vm_pindex_t) * NBBY, VM_RADIX_WIDTH) - 1)
  92
  93 /* Flag bits stored in node pointers. */
  94 #define VM_RADIX_ISLEAF 0x1
  95 #define VM_RADIX_FLAGS  0x1
  96 #define VM_RADIX_PAD    VM_RADIX_FLAGS
  97
  98 /* Returns one unit associated with specified level. */
  99 #define VM_RADIX_UNITLEVEL(lev)                                         \
 100         ((vm_pindex_t)1 << ((lev) * VM_RADIX_WIDTH))
 101
 102 enum vm_radix_access { SMR, LOCKED, UNSERIALIZED };
 103
 104 struct vm_radix_node;
 105 typedef SMR_POINTER(struct vm_radix_node *) smrnode_t;
 106
 107 struct vm_radix_node {
 108         vm_pindex_t     rn_owner;                       /* Owner of record. */
 109         uint16_t        rn_count;                       /* Valid children. */
 110         uint8_t         rn_clev;                        /* Current level. */
 111         int8_t          rn_last;                        /* zero last ptr. */
 112         smrnode_t       rn_child[VM_RADIX_COUNT];       /* Child nodes. */
 113 };
 114
 115 static uma_zone_t vm_radix_node_zone;
 116 static smr_t vm_radix_smr;
 117
 118 static void vm_radix_node_store(smrnode_t *p, struct vm_radix_node *v,
 119     enum vm_radix_access access);
 120
 121 /*
 122  * Allocate a radix node.
 123  */
 124 static struct vm_radix_node *
 125 vm_radix_node_get(vm_pindex_t owner, uint16_t count, uint16_t clevel)
 126 {
 127         struct vm_radix_node *rnode;
 128
 129         rnode = uma_zalloc_smr(vm_radix_node_zone, M_NOWAIT);
 130         if (rnode == NULL)
 131                 return (NULL);
 132
 133         /*
 134          * We want to clear the last child pointer after the final section
 135          * has exited so lookup can not return false negatives.  It is done
 136          * here because it will be cache-cold in the dtor callback.
 137          */
 138         if (rnode->rn_last != 0) {
 139                 vm_radix_node_store(&rnode->rn_child[rnode->rn_last - 1],
 140                     NULL, UNSERIALIZED);
 141                 rnode->rn_last = 0;
 142         }
 143         rnode->rn_owner = owner;
 144         rnode->rn_count = count;
 145         rnode->rn_clev = clevel;
 146         return (rnode);
 147 }
 148
 149 /*
 150  * Free radix node.
 151  */
 152 static __inline void
 153 vm_radix_node_put(struct vm_radix_node *rnode, int8_t last)
 154 {
 155 #ifdef INVARIANTS
 156         int slot;
 157
 158         KASSERT(rnode->rn_count == 0,
 159             ("vm_radix_node_put: rnode %p has %d children", rnode,
 160             rnode->rn_count));
 161         for (slot = 0; slot < VM_RADIX_COUNT; slot++) {
 162                 if (slot == last)
 163                         continue;
 164                 KASSERT(smr_unserialized_load(&rnode->rn_child[slot], true) ==
 165                     NULL, ("vm_radix_node_put: rnode %p has a child", rnode));
 166         }
 167 #endif
 168         /* Off by one so a freshly zero'd node is not assigned to. */
 169         rnode->rn_last = last + 1;
 170         uma_zfree_smr(vm_radix_node_zone, rnode);
 171 }
 172
 173 /*
 174  * Return the position in the array for a given level.
 175  */
 176 static __inline int
 177 vm_radix_slot(vm_pindex_t index, uint16_t level)
 178 {
 179
 180         return ((index >> (level * VM_RADIX_WIDTH)) & VM_RADIX_MASK);
 181 }
 182
 183 /* Trims the key after the specified level. */
 184 static __inline vm_pindex_t
 185 vm_radix_trimkey(vm_pindex_t index, uint16_t level)
 186 {
 187         vm_pindex_t ret;
 188
 189         ret = index;
 190         if (level > 0) {
 191                 ret >>= level * VM_RADIX_WIDTH;
 192                 ret <<= level * VM_RADIX_WIDTH;
 193         }
 194         return (ret);
 195 }
 196
 197 /*
 198  * Fetch a node pointer from a slot in another node.
 199  */
 200 static __inline struct vm_radix_node *
 201 vm_radix_node_load(smrnode_t *p, enum vm_radix_access access)
 202 {
 203
 204         switch (access) {
 205         case UNSERIALIZED:
 206                 return (smr_unserialized_load(p, true));
 207         case LOCKED:
 208                 return (smr_serialized_load(p, true));
 209         case SMR:
 210                 return (smr_entered_load(p, vm_radix_smr));
 211         }
 212         __assert_unreachable();
 213 }
 214
 215 static __inline void
 216 vm_radix_node_store(smrnode_t *p, struct vm_radix_node *v,
 217     enum vm_radix_access access)
 218 {
 219
 220
 221         switch (access) {
 222         case UNSERIALIZED:
 223                 smr_unserialized_store(p, v, true);
 224                 break;
 225         case LOCKED:
 226                 smr_serialized_store(p, v, true);
 227                 break;
 228         case SMR:
 229                 panic("vm_radix_node_store: Not supported in smr section.");
 230         }
 231 }
 232
 233 /*
 234  * Get the root node for a radix tree.
 235  */
 236 static __inline struct vm_radix_node *
 237 vm_radix_root_load(struct vm_radix *rtree, enum vm_radix_access access)
 238 {
 239
 240         return (vm_radix_node_load((smrnode_t *)&rtree->rt_root, access));
 241 }
 242
 243 /*
 244  * Set the root node for a radix tree.
 245  */
 246 static __inline void
 247 vm_radix_root_store(struct vm_radix *rtree, struct vm_radix_node *rnode,
 248     enum vm_radix_access access)
 249 {
 250
 251         vm_radix_node_store((smrnode_t *)&rtree->rt_root, rnode, access);
 252 }
 253
 254 /*
 255  * Returns TRUE if the specified radix node is a leaf and FALSE otherwise.
 256  */
 257 static __inline boolean_t
 258 vm_radix_isleaf(struct vm_radix_node *rnode)
 259 {
 260
 261         return (((uintptr_t)rnode & VM_RADIX_ISLEAF) != 0);
 262 }
 263
 264 /*
 265  * Returns the associated page extracted from rnode.
 266  */
 267 static __inline vm_page_t
 268 vm_radix_topage(struct vm_radix_node *rnode)
 269 {
 270
 271         return ((vm_page_t)((uintptr_t)rnode & ~VM_RADIX_FLAGS));
 272 }
 273
 274 /*
 275  * Adds the page as a child of the provided node.
 276  */
 277 static __inline void
 278 vm_radix_addpage(struct vm_radix_node *rnode, vm_pindex_t index, uint16_t clev,
 279     vm_page_t page, enum vm_radix_access access)
 280 {
 281         int slot;
 282
 283         slot = vm_radix_slot(index, clev);
 284         vm_radix_node_store(&rnode->rn_child[slot],
 285             (struct vm_radix_node *)((uintptr_t)page | VM_RADIX_ISLEAF), access);
 286 }
 287
 288 /*
 289  * Returns the slot where two keys differ.
 290  * It cannot accept 2 equal keys.
 291  */
 292 static __inline uint16_t
 293 vm_radix_keydiff(vm_pindex_t index1, vm_pindex_t index2)
 294 {
 295         uint16_t clev;
 296
 297         KASSERT(index1 != index2, ("%s: passing the same key value %jx",
 298             __func__, (uintmax_t)index1));
 299
 300         index1 ^= index2;
 301         for (clev = VM_RADIX_LIMIT;; clev--)
 302                 if (vm_radix_slot(index1, clev) != 0)
 303                         return (clev);
 304 }
 305
 306 /*
 307  * Returns TRUE if it can be determined that key does not belong to the
 308  * specified rnode.  Otherwise, returns FALSE.
 309  */
 310 static __inline boolean_t
 311 vm_radix_keybarr(struct vm_radix_node *rnode, vm_pindex_t idx)
 312 {
 313
 314         if (rnode->rn_clev < VM_RADIX_LIMIT) {
 315                 idx = vm_radix_trimkey(idx, rnode->rn_clev + 1);
 316                 return (idx != rnode->rn_owner);
 317         }
 318         return (FALSE);
 319 }
 320
 321 /*
 322  * Internal helper for vm_radix_reclaim_allnodes().
 323  * This function is recursive.
 324  */
 325 static void
 326 vm_radix_reclaim_allnodes_int(struct vm_radix_node *rnode)
 327 {
 328         struct vm_radix_node *child;
 329         int slot;
 330
 331         KASSERT(rnode->rn_count <= VM_RADIX_COUNT,
 332             ("vm_radix_reclaim_allnodes_int: bad count in rnode %p", rnode));
 333         for (slot = 0; rnode->rn_count != 0; slot++) {
 334                 child = vm_radix_node_load(&rnode->rn_child[slot], UNSERIALIZED);
 335                 if (child == NULL)
 336                         continue;
 337                 if (!vm_radix_isleaf(child))
 338                         vm_radix_reclaim_allnodes_int(child);
 339                 vm_radix_node_store(&rnode->rn_child[slot], NULL, UNSERIALIZED);
 340                 rnode->rn_count--;
 341         }
 342         vm_radix_node_put(rnode, -1);
 343 }
 344
 345 #ifndef UMA_MD_SMALL_ALLOC
 346 void vm_radix_reserve_kva(void);
 347 /*
 348  * Reserve the KVA necessary to satisfy the node allocation.
 349  * This is mandatory in architectures not supporting direct
 350  * mapping as they will need otherwise to carve into the kernel maps for
 351  * every node allocation, resulting into deadlocks for consumers already
 352  * working with kernel maps.
 353  */
 354 void
 355 vm_radix_reserve_kva(void)
 356 {
 357
 358         /*
 359          * Calculate the number of reserved nodes, discounting the pages that
 360          * are needed to store them.
 361          */
 362         if (!uma_zone_reserve_kva(vm_radix_node_zone,
 363             ((vm_paddr_t)vm_cnt.v_page_count * PAGE_SIZE) / (PAGE_SIZE +
 364             sizeof(struct vm_radix_node))))
 365                 panic("%s: unable to reserve KVA", __func__);
 366 }
 367 #endif
 368
 369 /*
 370  * Initialize the UMA slab zone.
 371  */
 372 void
 373 vm_radix_zinit(void)
 374 {
 375
 376         vm_radix_node_zone = uma_zcreate("RADIX NODE",
 377             sizeof(struct vm_radix_node), NULL, NULL, NULL, NULL,
 378             VM_RADIX_PAD, UMA_ZONE_VM | UMA_ZONE_SMR | UMA_ZONE_ZINIT);
 379         vm_radix_smr = uma_zone_get_smr(vm_radix_node_zone);
 380 }
 381
 382 /*
 383  * Inserts the key-value pair into the trie.
 384  * Panics if the key already exists.
 385  */
 386 int
 387 vm_radix_insert(struct vm_radix *rtree, vm_page_t page)
 388 {
 389         vm_pindex_t index, newind;
 390         struct vm_radix_node *rnode, *tmp;
 391         smrnode_t *parentp;
 392         vm_page_t m;
 393         int slot;
 394         uint16_t clev;
 395
 396         index = page->pindex;
 397
 398         /*
 399          * The owner of record for root is not really important because it
 400          * will never be used.
 401          */
 402         rnode = vm_radix_root_load(rtree, LOCKED);
 403         if (rnode == NULL) {
 404                 rtree->rt_root = (uintptr_t)page | VM_RADIX_ISLEAF;
 405                 return (0);
 406         }
 407         parentp = (smrnode_t *)&rtree->rt_root;
 408         for (;;) {
 409                 if (vm_radix_isleaf(rnode)) {
 410                         m = vm_radix_topage(rnode);
 411                         if (m->pindex == index)
 412                                 panic("%s: key %jx is already present",
 413                                     __func__, (uintmax_t)index);
 414                         clev = vm_radix_keydiff(m->pindex, index);
 415                         tmp = vm_radix_node_get(vm_radix_trimkey(index,
 416                             clev + 1), 2, clev);
 417                         if (tmp == NULL)
 418                                 return (ENOMEM);
 419                         /* These writes are not yet visible due to ordering. */
 420                         vm_radix_addpage(tmp, index, clev, page, UNSERIALIZED);
 421                         vm_radix_addpage(tmp, m->pindex, clev, m, UNSERIALIZED);
 422                         /* Synchronize to make leaf visible. */
 423                         vm_radix_node_store(parentp, tmp, LOCKED);
 424                         return (0);
 425                 } else if (vm_radix_keybarr(rnode, index))
 426                         break;
 427                 slot = vm_radix_slot(index, rnode->rn_clev);
 428                 parentp = &rnode->rn_child[slot];
 429                 tmp = vm_radix_node_load(parentp, LOCKED);
 430                 if (tmp == NULL) {
 431                         rnode->rn_count++;
 432                         vm_radix_addpage(rnode, index, rnode->rn_clev, page,
 433                             LOCKED);
 434                         return (0);
 435                 }
 436                 rnode = tmp;
 437         }
 438
 439         /*
 440          * A new node is needed because the right insertion level is reached.
 441          * Setup the new intermediate node and add the 2 children: the
 442          * new object and the older edge.
 443          */
 444         newind = rnode->rn_owner;
 445         clev = vm_radix_keydiff(newind, index);
 446         tmp = vm_radix_node_get(vm_radix_trimkey(index, clev + 1), 2, clev);
 447         if (tmp == NULL)
 448                 return (ENOMEM);
 449         slot = vm_radix_slot(newind, clev);
 450         /* These writes are not yet visible due to ordering. */
 451         vm_radix_addpage(tmp, index, clev, page, UNSERIALIZED);
 452         vm_radix_node_store(&tmp->rn_child[slot], rnode, UNSERIALIZED);
 453         /* Serializing write to make the above visible. */
 454         vm_radix_node_store(parentp, tmp, LOCKED);
 455
 456         return (0);
 457 }
 458
 459 /*
 460  * Returns TRUE if the specified radix tree contains a single leaf and FALSE
 461  * otherwise.
 462  */
 463 boolean_t
 464 vm_radix_is_singleton(struct vm_radix *rtree)
 465 {
 466         struct vm_radix_node *rnode;
 467
 468         rnode = vm_radix_root_load(rtree, LOCKED);
 469         if (rnode == NULL)
 470                 return (FALSE);
 471         return (vm_radix_isleaf(rnode));
 472 }
 473
 474 /*
 475  * Returns the value stored at the index.  If the index is not present,
 476  * NULL is returned.
 477  */
 478 static __always_inline vm_page_t
 479 _vm_radix_lookup(struct vm_radix *rtree, vm_pindex_t index,
 480     enum vm_radix_access access)
 481 {
 482         struct vm_radix_node *rnode;
 483         vm_page_t m;
 484         int slot;
 485
 486         rnode = vm_radix_root_load(rtree, access);
 487         while (rnode != NULL) {
 488                 if (vm_radix_isleaf(rnode)) {
 489                         m = vm_radix_topage(rnode);
 490                         if (m->pindex == index)
 491                                 return (m);
 492                         break;
 493                 }
 494                 if (vm_radix_keybarr(rnode, index))
 495                         break;
 496                 slot = vm_radix_slot(index, rnode->rn_clev);
 497                 rnode = vm_radix_node_load(&rnode->rn_child[slot], access);
 498         }
 499         return (NULL);
 500 }
 501
 502 /*
 503  * Returns the value stored at the index assuming there is an external lock.
 504  *
 505  * If the index is not present, NULL is returned.
 506  */
 507 vm_page_t
 508 vm_radix_lookup(struct vm_radix *rtree, vm_pindex_t index)
 509 {
 510
 511         return _vm_radix_lookup(rtree, index, LOCKED);
 512 }
 513
 514 /*
 515  * Returns the value stored at the index without requiring an external lock.
 516  *
 517  * If the index is not present, NULL is returned.
 518  */
 519 vm_page_t
 520 vm_radix_lookup_unlocked(struct vm_radix *rtree, vm_pindex_t index)
 521 {
 522         vm_page_t m;
 523
 524         smr_enter(vm_radix_smr);
 525         m = _vm_radix_lookup(rtree, index, SMR);
 526         smr_exit(vm_radix_smr);
 527
 528         return (m);
 529 }
 530
 531 /*
 532  * Look up the nearest entry at a position greater than or equal to index.
 533  */
 534 vm_page_t
 535 vm_radix_lookup_ge(struct vm_radix *rtree, vm_pindex_t index)
 536 {
 537         struct vm_radix_node *stack[VM_RADIX_LIMIT];
 538         vm_pindex_t inc;
 539         vm_page_t m;
 540         struct vm_radix_node *child, *rnode;
 541 #ifdef INVARIANTS
 542         int loops = 0;
 543 #endif
 544         int slot, tos;
 545
 546         rnode = vm_radix_root_load(rtree, LOCKED);
 547         if (rnode == NULL)
 548                 return (NULL);
 549         else if (vm_radix_isleaf(rnode)) {
 550                 m = vm_radix_topage(rnode);
 551                 if (m->pindex >= index)
 552                         return (m);
 553                 else
 554                         return (NULL);
 555         }
 556         tos = 0;
 557         for (;;) {
 558                 /*
 559                  * If the keys differ before the current bisection node,
 560                  * then the search key might rollback to the earliest
 561                  * available bisection node or to the smallest key
 562                  * in the current node (if the owner is greater than the
 563                  * search key).
 564                  */
 565                 if (vm_radix_keybarr(rnode, index)) {
 566                         if (index > rnode->rn_owner) {
 567 ascend:
 568                                 KASSERT(++loops < 1000,
 569                                     ("vm_radix_lookup_ge: too many loops"));
 570
 571                                 /*
 572                                  * Pop nodes from the stack until either the
 573                                  * stack is empty or a node that could have a
 574                                  * matching descendant is found.
 575                                  */
 576                                 do {
 577                                         if (tos == 0)
 578                                                 return (NULL);
 579                                         rnode = stack[--tos];
 580                                 } while (vm_radix_slot(index,
 581                                     rnode->rn_clev) == (VM_RADIX_COUNT - 1));
 582
 583                                 /*
 584                                  * The following computation cannot overflow
 585                                  * because index's slot at the current level
 586                                  * is less than VM_RADIX_COUNT - 1.
 587                                  */
 588                                 index = vm_radix_trimkey(index,
 589                                     rnode->rn_clev);
 590                                 index += VM_RADIX_UNITLEVEL(rnode->rn_clev);
 591                         } else
 592                                 index = rnode->rn_owner;
 593                         KASSERT(!vm_radix_keybarr(rnode, index),
 594                             ("vm_radix_lookup_ge: keybarr failed"));
 595                 }
 596                 slot = vm_radix_slot(index, rnode->rn_clev);
 597                 child = vm_radix_node_load(&rnode->rn_child[slot], LOCKED);
 598                 if (vm_radix_isleaf(child)) {
 599                         m = vm_radix_topage(child);
 600                         if (m->pindex >= index)
 601                                 return (m);
 602                 } else if (child != NULL)
 603                         goto descend;
 604
 605                 /*
 606                  * Look for an available edge or page within the current
 607                  * bisection node.
 608                  */
 609                 if (slot < (VM_RADIX_COUNT - 1)) {
 610                         inc = VM_RADIX_UNITLEVEL(rnode->rn_clev);
 611                         index = vm_radix_trimkey(index, rnode->rn_clev);
 612                         do {
 613                                 index += inc;
 614                                 slot++;
 615                                 child = vm_radix_node_load(&rnode->rn_child[slot],
 616                                     LOCKED);
 617                                 if (vm_radix_isleaf(child)) {
 618                                         m = vm_radix_topage(child);
 619                                         if (m->pindex >= index)
 620                                                 return (m);
 621                                 } else if (child != NULL)
 622                                         goto descend;
 623                         } while (slot < (VM_RADIX_COUNT - 1));
 624                 }
 625                 KASSERT(child == NULL || vm_radix_isleaf(child),
 626                     ("vm_radix_lookup_ge: child is radix node"));
 627
 628                 /*
 629                  * If a page or edge greater than the search slot is not found
 630                  * in the current node, ascend to the next higher-level node.
 631                  */
 632                 goto ascend;
 633 descend:
 634                 KASSERT(rnode->rn_clev > 0,
 635                     ("vm_radix_lookup_ge: pushing leaf's parent"));
 636                 KASSERT(tos < VM_RADIX_LIMIT,
 637                     ("vm_radix_lookup_ge: stack overflow"));
 638                 stack[tos++] = rnode;
 639                 rnode = child;
 640         }
 641 }
 642
 643 /*
 644  * Look up the nearest entry at a position less than or equal to index.
 645  */
 646 vm_page_t
 647 vm_radix_lookup_le(struct vm_radix *rtree, vm_pindex_t index)
 648 {
 649         struct vm_radix_node *stack[VM_RADIX_LIMIT];
 650         vm_pindex_t inc;
 651         vm_page_t m;
 652         struct vm_radix_node *child, *rnode;
 653 #ifdef INVARIANTS
 654         int loops = 0;
 655 #endif
 656         int slot, tos;
 657
 658         rnode = vm_radix_root_load(rtree, LOCKED);
 659         if (rnode == NULL)
 660                 return (NULL);
 661         else if (vm_radix_isleaf(rnode)) {
 662                 m = vm_radix_topage(rnode);
 663                 if (m->pindex <= index)
 664                         return (m);
 665                 else
 666                         return (NULL);
 667         }
 668         tos = 0;
 669         for (;;) {
 670                 /*
 671                  * If the keys differ before the current bisection node,
 672                  * then the search key might rollback to the earliest
 673                  * available bisection node or to the largest key
 674                  * in the current node (if the owner is smaller than the
 675                  * search key).
 676                  */
 677                 if (vm_radix_keybarr(rnode, index)) {
 678                         if (index > rnode->rn_owner) {
 679                                 index = rnode->rn_owner + VM_RADIX_COUNT *
 680                                     VM_RADIX_UNITLEVEL(rnode->rn_clev);
 681                         } else {
 682 ascend:
 683                                 KASSERT(++loops < 1000,
 684                                     ("vm_radix_lookup_le: too many loops"));
 685
 686                                 /*
 687                                  * Pop nodes from the stack until either the
 688                                  * stack is empty or a node that could have a
 689                                  * matching descendant is found.
 690                                  */
 691                                 do {
 692                                         if (tos == 0)
 693                                                 return (NULL);
 694                                         rnode = stack[--tos];
 695                                 } while (vm_radix_slot(index,
 696                                     rnode->rn_clev) == 0);
 697
 698                                 /*
 699                                  * The following computation cannot overflow
 700                                  * because index's slot at the current level
 701                                  * is greater than 0.
 702                                  */
 703                                 index = vm_radix_trimkey(index,
 704                                     rnode->rn_clev);
 705                         }
 706                         index--;
 707                         KASSERT(!vm_radix_keybarr(rnode, index),
 708                             ("vm_radix_lookup_le: keybarr failed"));
 709                 }
 710                 slot = vm_radix_slot(index, rnode->rn_clev);
 711                 child = vm_radix_node_load(&rnode->rn_child[slot], LOCKED);
 712                 if (vm_radix_isleaf(child)) {
 713                         m = vm_radix_topage(child);
 714                         if (m->pindex <= index)
 715                                 return (m);
 716                 } else if (child != NULL)
 717                         goto descend;
 718
 719                 /*
 720                  * Look for an available edge or page within the current
 721                  * bisection node.
 722                  */
 723                 if (slot > 0) {
 724                         inc = VM_RADIX_UNITLEVEL(rnode->rn_clev);
 725                         index |= inc - 1;
 726                         do {
 727                                 index -= inc;
 728                                 slot--;
 729                                 child = vm_radix_node_load(&rnode->rn_child[slot],
 730                                     LOCKED);
 731                                 if (vm_radix_isleaf(child)) {
 732                                         m = vm_radix_topage(child);
 733                                         if (m->pindex <= index)
 734                                                 return (m);
 735                                 } else if (child != NULL)
 736                                         goto descend;
 737                         } while (slot > 0);
 738                 }
 739                 KASSERT(child == NULL || vm_radix_isleaf(child),
 740                     ("vm_radix_lookup_le: child is radix node"));
 741
 742                 /*
 743                  * If a page or edge smaller than the search slot is not found
 744                  * in the current node, ascend to the next higher-level node.
 745                  */
 746                 goto ascend;
 747 descend:
 748                 KASSERT(rnode->rn_clev > 0,
 749                     ("vm_radix_lookup_le: pushing leaf's parent"));
 750                 KASSERT(tos < VM_RADIX_LIMIT,
 751                     ("vm_radix_lookup_le: stack overflow"));
 752                 stack[tos++] = rnode;
 753                 rnode = child;
 754         }
 755 }
 756
 757 /*
 758  * Remove the specified index from the trie, and return the value stored at
 759  * that index.  If the index is not present, return NULL.
 760  */
 761 vm_page_t
 762 vm_radix_remove(struct vm_radix *rtree, vm_pindex_t index)
 763 {
 764         struct vm_radix_node *rnode, *parent, *tmp;
 765         vm_page_t m;
 766         int i, slot;
 767
 768         rnode = vm_radix_root_load(rtree, LOCKED);
 769         if (vm_radix_isleaf(rnode)) {
 770                 m = vm_radix_topage(rnode);
 771                 if (m->pindex != index)
 772                         return (NULL);
 773                 vm_radix_root_store(rtree, NULL, LOCKED);
 774                 return (m);
 775         }
 776         parent = NULL;
 777         for (;;) {
 778                 if (rnode == NULL)
 779                         return (NULL);
 780                 slot = vm_radix_slot(index, rnode->rn_clev);
 781                 tmp = vm_radix_node_load(&rnode->rn_child[slot], LOCKED);
 782                 if (vm_radix_isleaf(tmp)) {
 783                         m = vm_radix_topage(tmp);
 784                         if (m->pindex != index)
 785                                 return (NULL);
 786                         vm_radix_node_store(&rnode->rn_child[slot], NULL, LOCKED);
 787                         rnode->rn_count--;
 788                         if (rnode->rn_count > 1)
 789                                 return (m);
 790                         for (i = 0; i < VM_RADIX_COUNT; i++)
 791                                 if (vm_radix_node_load(&rnode->rn_child[i],
 792                                     LOCKED) != NULL)
 793                                         break;
 794                         KASSERT(i != VM_RADIX_COUNT,
 795                             ("%s: invalid node configuration", __func__));
 796                         tmp = vm_radix_node_load(&rnode->rn_child[i], LOCKED);
 797                         if (parent == NULL)
 798                                 vm_radix_root_store(rtree, tmp, LOCKED);
 799                         else {
 800                                 slot = vm_radix_slot(index, parent->rn_clev);
 801                                 KASSERT(vm_radix_node_load(
 802                                     &parent->rn_child[slot], LOCKED) == rnode,
 803                                     ("%s: invalid child value", __func__));
 804                                 vm_radix_node_store(&parent->rn_child[slot],
 805                                     tmp, LOCKED);
 806                         }
 807                         /*
 808                          * The child is still valid and we can not zero the
 809                          * pointer until all smr references are gone.
 810                          */
 811                         rnode->rn_count--;
 812                         vm_radix_node_put(rnode, i);
 813                         return (m);
 814                 }
 815                 parent = rnode;
 816                 rnode = tmp;
 817         }
 818 }
 819
 820 /*
 821  * Remove and free all the nodes from the radix tree.
 822  * This function is recursive but there is a tight control on it as the
 823  * maximum depth of the tree is fixed.
 824  */
 825 void
 826 vm_radix_reclaim_allnodes(struct vm_radix *rtree)
 827 {
 828         struct vm_radix_node *root;
 829
 830         root = vm_radix_root_load(rtree, LOCKED);
 831         if (root == NULL)
 832                 return;
 833         vm_radix_root_store(rtree, NULL, UNSERIALIZED);
 834         if (!vm_radix_isleaf(root))
 835                 vm_radix_reclaim_allnodes_int(root);
 836 }
 837
 838 /*
 839  * Replace an existing page in the trie with another one.
 840  * Panics if there is not an old page in the trie at the new page's index.
 841  */
 842 vm_page_t
 843 vm_radix_replace(struct vm_radix *rtree, vm_page_t newpage)
 844 {
 845         struct vm_radix_node *rnode, *tmp;
 846         vm_page_t m;
 847         vm_pindex_t index;
 848         int slot;
 849
 850         index = newpage->pindex;
 851         rnode = vm_radix_root_load(rtree, LOCKED);
 852         if (rnode == NULL)
 853                 panic("%s: replacing page on an empty trie", __func__);
 854         if (vm_radix_isleaf(rnode)) {
 855                 m = vm_radix_topage(rnode);
 856                 if (m->pindex != index)
 857                         panic("%s: original replacing root key not found",
 858                             __func__);
 859                 rtree->rt_root = (uintptr_t)newpage | VM_RADIX_ISLEAF;
 860                 return (m);
 861         }
 862         for (;;) {
 863                 slot = vm_radix_slot(index, rnode->rn_clev);
 864                 tmp = vm_radix_node_load(&rnode->rn_child[slot], LOCKED);
 865                 if (vm_radix_isleaf(tmp)) {
 866                         m = vm_radix_topage(tmp);
 867                         if (m->pindex == index) {
 868                                 vm_radix_node_store(&rnode->rn_child[slot],
 869                                     (struct vm_radix_node *)((uintptr_t)newpage |
 870                                     VM_RADIX_ISLEAF), LOCKED);
 871                                 return (m);
 872                         } else
 873                                 break;
 874                 } else if (tmp == NULL || vm_radix_keybarr(tmp, index))
 875                         break;
 876                 rnode = tmp;
 877         }
 878         panic("%s: original replacing page not found", __func__);
 879 }
 880
 881 void
 882 vm_radix_wait(void)
 883 {
 884         uma_zwait(vm_radix_node_zone);
 885 }
 886
 887 #ifdef DDB
 888 /*
 889  * Show details about the given radix node.
 890  */
 891 DB_SHOW_COMMAND(radixnode, db_show_radixnode)
 892 {
 893         struct vm_radix_node *rnode, *tmp;
 894         int i;
 895
 896         if (!have_addr)
 897                 return;
 898         rnode = (struct vm_radix_node *)addr;
 899         db_printf("radixnode %p, owner %jx, children count %u, level %u:\n",
 900             (void *)rnode, (uintmax_t)rnode->rn_owner, rnode->rn_count,
 901             rnode->rn_clev);
 902         for (i = 0; i < VM_RADIX_COUNT; i++) {
 903                 tmp = vm_radix_node_load(&rnode->rn_child[i], UNSERIALIZED);
 904                 if (tmp != NULL)
 905                         db_printf("slot: %d, val: %p, page: %p, clev: %d\n",
 906                             i, (void *)tmp,
 907                             vm_radix_isleaf(tmp) ?  vm_radix_topage(tmp) : NULL,
 908                             rnode->rn_clev);
 909         }
 910 }
 911 #endif /* DDB */