sys/vm/vm_radix.c

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