sys/vm/vm_zone.c

   1 /*
   2  * (MPSAFE)
   3  *
   4  * Copyright (c) 1997, 1998 John S. Dyson
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *      notice immediately at the beginning of the file, without modification,
  12  *      this list of conditions, and the following disclaimer.
  13  * 2. Absolutely no warranty of function or purpose is made by the author
  14  *      John S. Dyson.
  15  *
  16  * $FreeBSD: src/sys/vm/vm_zone.c,v 1.30.2.6 2002/10/10 19:50:16 dillon Exp $
  17  */
  18
  19 #include <sys/param.h>
  20 #include <sys/queue.h>
  21 #include <sys/systm.h>
  22 #include <sys/kernel.h>
  23 #include <sys/lock.h>
  24 #include <sys/malloc.h>
  25 #include <sys/sysctl.h>
  26 #include <sys/vmmeter.h>
  27
  28 #include <vm/vm.h>
  29 #include <vm/vm_object.h>
  30 #include <vm/vm_page.h>
  31 #include <vm/vm_map.h>
  32 #include <vm/vm_kern.h>
  33 #include <vm/vm_extern.h>
  34 #include <vm/vm_zone.h>
  35
  36 #include <sys/spinlock2.h>
  37 #include <vm/vm_page2.h>
  38
  39 static MALLOC_DEFINE(M_ZONE, "ZONE", "Zone header");
  40
  41 #define ZONE_ERROR_INVALID 0
  42 #define ZONE_ERROR_NOTFREE 1
  43 #define ZONE_ERROR_ALREADYFREE 2
  44
  45 #define ZONE_ROUNDING   32
  46
  47 #define ZENTRY_FREE     0x12342378
  48
  49 long zone_burst = 128;
  50
  51 static void *zget(vm_zone_t z);
  52
  53 /*
  54  * Return an item from the specified zone.   This function is non-blocking for
  55  * ZONE_INTERRUPT zones.
  56  *
  57  * No requirements.
  58  */
  59 void *
  60 zalloc(vm_zone_t z)
  61 {
  62         globaldata_t gd = mycpu;
  63         vm_zpcpu_t *zpcpu;
  64         void *item;
  65         long n;
  66
  67 #ifdef INVARIANTS
  68         if (z == NULL)
  69                 zerror(ZONE_ERROR_INVALID);
  70 #endif
  71         zpcpu = &z->zpcpu[gd->gd_cpuid];
  72 retry:
  73         /*
  74          * Avoid spinlock contention by allocating from a per-cpu queue
  75          */
  76         if (zpcpu->zfreecnt > 0) {
  77                 crit_enter_gd(gd);
  78                 if (zpcpu->zfreecnt > 0) {
  79                         item = zpcpu->zitems;
  80 #ifdef INVARIANTS
  81                         KASSERT(item != NULL,
  82                                 ("zitems_pcpu unexpectedly NULL"));
  83                         if (((void **)item)[1] != (void *)ZENTRY_FREE)
  84                                 zerror(ZONE_ERROR_NOTFREE);
  85                         ((void **)item)[1] = NULL;
  86 #endif
  87                         zpcpu->zitems = ((void **) item)[0];
  88                         --zpcpu->zfreecnt;
  89                         ++zpcpu->znalloc;
  90                         crit_exit_gd(gd);
  91
  92                         return item;
  93                 }
  94                 crit_exit_gd(gd);
  95         }
  96
  97         /*
  98          * Per-zone spinlock for the remainder.  Always load at least one
  99          * item.
 100          */
 101         spin_lock(&z->zspin);
 102         if (z->zfreecnt > z->zfreemin) {
 103                 n = zone_burst;
 104                 do {
 105                         item = z->zitems;
 106 #ifdef INVARIANTS
 107                         KASSERT(item != NULL, ("zitems unexpectedly NULL"));
 108                         if (((void **)item)[1] != (void *)ZENTRY_FREE)
 109                                 zerror(ZONE_ERROR_NOTFREE);
 110 #endif
 111                         z->zitems = ((void **)item)[0];
 112                         --z->zfreecnt;
 113                         ((void **)item)[0] = zpcpu->zitems;
 114                         zpcpu->zitems = item;
 115                         ++zpcpu->zfreecnt;
 116                 } while (--n > 0 && z->zfreecnt > z->zfreemin);
 117                 spin_unlock(&z->zspin);
 118                 goto retry;
 119         } else {
 120                 spin_unlock(&z->zspin);
 121                 item = zget(z);
 122                 /*
 123                  * PANICFAIL allows the caller to assume that the zalloc()
 124                  * will always succeed.  If it doesn't, we panic here.
 125                  */
 126                 if (item == NULL && (z->zflags & ZONE_PANICFAIL))
 127                         panic("zalloc(%s) failed", z->zname);
 128         }
 129         return item;
 130 }
 131
 132 /*
 133  * Free an item to the specified zone.
 134  *
 135  * No requirements.
 136  */
 137 void
 138 zfree(vm_zone_t z, void *item)
 139 {
 140         globaldata_t gd = mycpu;
 141         vm_zpcpu_t *zpcpu;
 142         void *tail_item;
 143         long count;
 144         long zmax;
 145
 146         zpcpu = &z->zpcpu[gd->gd_cpuid];
 147
 148         /*
 149          * Avoid spinlock contention by freeing into a per-cpu queue
 150          */
 151         zmax = z->zmax_pcpu;
 152         if (zmax < 1024)
 153                 zmax = 1024;
 154
 155         /*
 156          * Add to pcpu cache
 157          */
 158         crit_enter_gd(gd);
 159         ((void **)item)[0] = zpcpu->zitems;
 160 #ifdef INVARIANTS
 161         if (((void **)item)[1] == (void *)ZENTRY_FREE)
 162                 zerror(ZONE_ERROR_ALREADYFREE);
 163         ((void **)item)[1] = (void *)ZENTRY_FREE;
 164 #endif
 165         zpcpu->zitems = item;
 166         ++zpcpu->zfreecnt;
 167
 168         if (zpcpu->zfreecnt < zmax) {
 169                 crit_exit_gd(gd);
 170                 return;
 171         }
 172
 173         /*
 174          * Hystereis, move (zmax) (calculated below) items to the pool.
 175          */
 176         zmax = zmax / 2;
 177         if (zmax > zone_burst)
 178                 zmax = zone_burst;
 179         tail_item = item;
 180         count = 1;
 181
 182         while (count < zmax) {
 183                 tail_item = ((void **)tail_item)[0];
 184                 ++count;
 185         }
 186         zpcpu->zitems = ((void **)tail_item)[0];
 187         zpcpu->zfreecnt -= count;
 188
 189         /*
 190          * Per-zone spinlock for the remainder.
 191          *
 192          * Also implement hysteresis by freeing a number of pcpu
 193          * entries.
 194          */
 195         spin_lock(&z->zspin);
 196         ((void **)tail_item)[0] = z->zitems;
 197         z->zitems = item;
 198         z->zfreecnt += count;
 199         spin_unlock(&z->zspin);
 200
 201         crit_exit_gd(gd);
 202 }
 203
 204 /*
 205  * This file comprises a very simple zone allocator.  This is used
 206  * in lieu of the malloc allocator, where needed or more optimal.
 207  *
 208  * Note that the initial implementation of this had coloring, and
 209  * absolutely no improvement (actually perf degradation) occurred.
 210  *
 211  * Note also that the zones are type stable.  The only restriction is
 212  * that the first two longwords of a data structure can be changed
 213  * between allocations.  Any data that must be stable between allocations
 214  * must reside in areas after the first two longwords.
 215  *
 216  * zinitna, zinit, zbootinit are the initialization routines.
 217  * zalloc, zfree, are the allocation/free routines.
 218  */
 219
 220 LIST_HEAD(zlist, vm_zone) zlist = LIST_HEAD_INITIALIZER(zlist);
 221 static int sysctl_vm_zone(SYSCTL_HANDLER_ARGS);
 222 static vm_pindex_t zone_kmem_pages, zone_kern_pages;
 223 static long zone_kmem_kvaspace;
 224
 225 /*
 226  * Create a zone, but don't allocate the zone structure.  If the
 227  * zone had been previously created by the zone boot code, initialize
 228  * various parts of the zone code.
 229  *
 230  * If waits are not allowed during allocation (e.g. during interrupt
 231  * code), a-priori allocate the kernel virtual space, and allocate
 232  * only pages when needed.
 233  *
 234  * Arguments:
 235  * z            pointer to zone structure.
 236  * obj          pointer to VM object (opt).
 237  * name         name of zone.
 238  * size         size of zone entries.
 239  * nentries     number of zone entries allocated (only ZONE_INTERRUPT.)
 240  * flags        ZONE_INTERRUPT -- items can be allocated at interrupt time.
 241  * zalloc       number of pages allocated when memory is needed.
 242  *
 243  * Note that when using ZONE_INTERRUPT, the size of the zone is limited
 244  * by the nentries argument.  The size of the memory allocatable is
 245  * unlimited if ZONE_INTERRUPT is not set.
 246  *
 247  * No requirements.
 248  */
 249 int
 250 zinitna(vm_zone_t z, char *name, size_t size, long nentries, uint32_t flags)
 251 {
 252         size_t totsize;
 253
 254         /*
 255          * Only zones created with zinit() are destroyable.
 256          */
 257         if (z->zflags & ZONE_DESTROYABLE)
 258                 panic("zinitna: can't create destroyable zone");
 259
 260         /*
 261          * NOTE: We can only adjust zsize if we previously did not
 262          *       use zbootinit().
 263          */
 264         if ((z->zflags & ZONE_BOOT) == 0) {
 265                 z->zsize = roundup2(size, ZONE_ROUNDING);
 266                 spin_init(&z->zspin, "zinitna");
 267                 z->zfreecnt = 0;
 268                 z->ztotal = 0;
 269                 z->zmax = 0;
 270                 z->zname = name;
 271                 z->zitems = NULL;
 272
 273                 lwkt_gettoken(&vm_token);
 274                 LIST_INSERT_HEAD(&zlist, z, zlink);
 275                 lwkt_reltoken(&vm_token);
 276
 277                 bzero(z->zpcpu, sizeof(z->zpcpu));
 278         }
 279
 280         z->zkmvec = NULL;
 281         z->zkmcur = z->zkmmax = 0;
 282         z->zflags |= flags;
 283
 284         /*
 285          * If we cannot wait, allocate KVA space up front, and we will fill
 286          * in pages as needed.  This is particularly required when creating
 287          * an allocation space for map entries in kernel_map, because we
 288          * do not want to go into a recursion deadlock with
 289          * vm_map_entry_reserve().
 290          */
 291         if (z->zflags & ZONE_INTERRUPT) {
 292                 totsize = round_page((size_t)z->zsize * nentries);
 293                 atomic_add_long(&zone_kmem_kvaspace, totsize);
 294
 295                 z->zkva = kmem_alloc_pageable(&kernel_map, totsize,
 296                                               VM_SUBSYS_ZALLOC);
 297                 if (z->zkva == 0) {
 298                         LIST_REMOVE(z, zlink);
 299                         return 0;
 300                 }
 301
 302                 z->zpagemax = totsize / PAGE_SIZE;
 303                 z->zallocflag = VM_ALLOC_SYSTEM | VM_ALLOC_INTERRUPT |
 304                                 VM_ALLOC_NORMAL | VM_ALLOC_RETRY;
 305                 z->zmax += nentries;
 306                 z->zmax_pcpu = z->zmax / ncpus / 16;
 307
 308                 /*
 309                  * Set reasonable pcpu cache bounds.  Low-memory systems
 310                  * might try to cache too little, large-memory systems
 311                  * might try to cache more than necessarsy.
 312                  *
 313                  * In particular, pvzone can wind up being excessive and
 314                  * waste memory unnecessarily.
 315                  */
 316                 if (z->zmax_pcpu < 1024)
 317                         z->zmax_pcpu = 1024;
 318                 if (z->zmax_pcpu * z->zsize > 16*1024*1024)
 319                         z->zmax_pcpu = 16*1024*1024 / z->zsize;
 320         } else {
 321                 z->zallocflag = VM_ALLOC_NORMAL | VM_ALLOC_SYSTEM;
 322                 z->zmax = 0;
 323                 z->zmax_pcpu = 8192;
 324         }
 325
 326
 327         if (z->zsize > PAGE_SIZE)
 328                 z->zfreemin = 1;
 329         else
 330                 z->zfreemin = PAGE_SIZE / z->zsize;
 331
 332         z->zpagecount = 0;
 333
 334         /*
 335          * Reduce kernel_map spam by allocating in chunks.
 336          */
 337         z->zalloc = ZONE_MAXPGLOAD;
 338
 339         /*
 340          * Populate the interrrupt zone at creation time rather than
 341          * on first allocation, as this is a potentially long operation.
 342          */
 343         if (z->zflags & ZONE_INTERRUPT) {
 344                 void *buf;
 345
 346                 buf = zget(z);
 347                 if (buf)
 348                         zfree(z, buf);
 349         }
 350
 351         return 1;
 352 }
 353
 354 /*
 355  * Subroutine same as zinitna, except zone data structure is allocated
 356  * automatically by malloc.  This routine should normally be used, except
 357  * in certain tricky startup conditions in the VM system -- then
 358  * zbootinit and zinitna can be used.  Zinit is the standard zone
 359  * initialization call.
 360  *
 361  * No requirements.
 362  */
 363 vm_zone_t
 364 zinit(char *name, size_t size, long nentries, uint32_t flags)
 365 {
 366         vm_zone_t z;
 367
 368         z = (vm_zone_t) kmalloc(sizeof (struct vm_zone), M_ZONE, M_NOWAIT);
 369         if (z == NULL)
 370                 return NULL;
 371
 372         z->zflags = 0;
 373         if (zinitna(z, name, size, nentries, flags & ~ZONE_DESTROYABLE) == 0) {
 374                 kfree(z, M_ZONE);
 375                 return NULL;
 376         }
 377
 378         if (flags & ZONE_DESTROYABLE)
 379                 z->zflags |= ZONE_DESTROYABLE;
 380
 381         return z;
 382 }
 383
 384 /*
 385  * Initialize a zone before the system is fully up.  This routine should
 386  * only be called before full VM startup.
 387  *
 388  * Called from the low level boot code only.
 389  */
 390 void
 391 zbootinit(vm_zone_t z, char *name, size_t size, void *item, long nitems)
 392 {
 393         long i;
 394
 395         spin_init(&z->zspin, "zbootinit");
 396         bzero(z->zpcpu, sizeof(z->zpcpu));
 397         z->zname = name;
 398         z->zsize = size;
 399         z->zpagemax = 0;
 400         z->zflags = ZONE_BOOT;
 401         z->zfreemin = 0;
 402         z->zallocflag = 0;
 403         z->zpagecount = 0;
 404         z->zalloc = 0;
 405
 406         bzero(item, (size_t)nitems * z->zsize);
 407         z->zitems = NULL;
 408         for (i = 0; i < nitems; i++) {
 409                 ((void **)item)[0] = z->zitems;
 410 #ifdef INVARIANTS
 411                 ((void **)item)[1] = (void *)ZENTRY_FREE;
 412 #endif
 413                 z->zitems = item;
 414                 item = (uint8_t *)item + z->zsize;
 415         }
 416         z->zfreecnt = nitems;
 417         z->zmax = nitems;
 418         z->ztotal = nitems;
 419
 420         lwkt_gettoken(&vm_token);
 421         LIST_INSERT_HEAD(&zlist, z, zlink);
 422         lwkt_reltoken(&vm_token);
 423 }
 424
 425 /*
 426  * Release all resources owned by zone created with zinit().
 427  *
 428  * No requirements.
 429  */
 430 void
 431 zdestroy(vm_zone_t z)
 432 {
 433         vm_pindex_t i;
 434
 435         if (z == NULL)
 436                 panic("zdestroy: null zone");
 437         if ((z->zflags & ZONE_DESTROYABLE) == 0)
 438                 panic("zdestroy: undestroyable zone");
 439
 440         lwkt_gettoken(&vm_token);
 441         LIST_REMOVE(z, zlink);
 442         lwkt_reltoken(&vm_token);
 443
 444         /*
 445          * Release virtual mappings, physical memory and update sysctl stats.
 446          */
 447         KKASSERT((z->zflags & ZONE_INTERRUPT) == 0);
 448         for (i = 0; i < z->zkmcur; i++) {
 449                 kmem_free(&kernel_map, z->zkmvec[i],
 450                           (size_t)z->zalloc * PAGE_SIZE);
 451                 atomic_subtract_long(&zone_kern_pages, z->zalloc);
 452         }
 453         if (z->zkmvec != NULL)
 454                 kfree(z->zkmvec, M_ZONE);
 455
 456         spin_uninit(&z->zspin);
 457         kfree(z, M_ZONE);
 458 }
 459
 460
 461 /*
 462  * void *zalloc(vm_zone_t zone) --
 463  *      Returns an item from a specified zone.  May not be called from a
 464  *      FAST interrupt or IPI function.
 465  *
 466  * void zfree(vm_zone_t zone, void *item) --
 467  *      Frees an item back to a specified zone.  May not be called from a
 468  *      FAST interrupt or IPI function.
 469  */
 470
 471 /*
 472  * Internal zone routine.  Not to be called from external (non vm_zone) code.
 473  *
 474  * This function may return NULL.
 475  *
 476  * No requirements.
 477  */
 478 static void *
 479 zget(vm_zone_t z)
 480 {
 481         vm_page_t pgs[ZONE_MAXPGLOAD];
 482         vm_page_t m;
 483         long nitems;
 484         long savezpc;
 485         size_t nbytes;
 486         size_t noffset;
 487         void *item;
 488         vm_pindex_t npages;
 489         vm_pindex_t nalloc;
 490         vm_pindex_t i;
 491
 492         if (z == NULL)
 493                 panic("zget: null zone");
 494
 495         if (z->zflags & ZONE_INTERRUPT) {
 496                 /*
 497                  * Interrupt zones do not mess with the kernel_map, they
 498                  * simply populate an existing mapping.
 499                  *
 500                  * First allocate as many pages as we can, stopping at
 501                  * our limit or if the page allocation fails.  Try to
 502                  * avoid exhausting the interrupt free minimum by backing
 503                  * off to normal page allocations after a certain point.
 504                  */
 505                 for (i = 0; i < ZONE_MAXPGLOAD && i < z->zalloc; ++i) {
 506                         if (i < 4) {
 507                                 m = vm_page_alloc(NULL,
 508                                                   mycpu->gd_rand_incr++,
 509                                                   z->zallocflag);
 510                         } else {
 511                                 m = vm_page_alloc(NULL,
 512                                                   mycpu->gd_rand_incr++,
 513                                                   VM_ALLOC_NORMAL |
 514                                                   VM_ALLOC_SYSTEM);
 515                         }
 516                         if (m == NULL)
 517                                 break;
 518                         pgs[i] = m;
 519                 }
 520                 nalloc = i;
 521
 522                 /*
 523                  * Account for the pages.
 524                  *
 525                  * NOTE! Do not allow overlap with a prior page as it
 526                  *       may still be undergoing allocation on another
 527                  *       cpu.
 528                  */
 529                 spin_lock(&z->zspin);
 530                 noffset = (size_t)z->zpagecount * PAGE_SIZE;
 531                 /* noffset -= noffset % z->zsize; */
 532                 savezpc = z->zpagecount;
 533                 if (z->zpagecount + nalloc > z->zpagemax)
 534                         z->zpagecount = z->zpagemax;
 535                 else
 536                         z->zpagecount += nalloc;
 537                 item = (char *)z->zkva + noffset;
 538                 npages = z->zpagecount - savezpc;
 539                 nitems = ((size_t)(savezpc + npages) * PAGE_SIZE - noffset) /
 540                          z->zsize;
 541                 atomic_add_long(&zone_kmem_pages, npages);
 542                 spin_unlock(&z->zspin);
 543
 544                 /*
 545                  * Enter the pages into the reserved KVA space.
 546                  */
 547                 for (i = 0; i < npages; ++i) {
 548                         vm_offset_t zkva;
 549
 550                         m = pgs[i];
 551                         KKASSERT(m->queue == PQ_NONE);
 552                         m->valid = VM_PAGE_BITS_ALL;
 553                         vm_page_wire(m);
 554                         vm_page_wakeup(m);
 555
 556                         zkva = z->zkva + (size_t)(savezpc + i) * PAGE_SIZE;
 557                         pmap_kenter(zkva, VM_PAGE_TO_PHYS(m));
 558                         bzero((void *)zkva, PAGE_SIZE);
 559                 }
 560                 for (i = npages; i < nalloc; ++i) {
 561                         m = pgs[i];
 562                         vm_page_free(m);
 563                 }
 564         } else if (z->zflags & ZONE_SPECIAL) {
 565                 /*
 566                  * The special zone is the one used for vm_map_entry_t's.
 567                  * We have to avoid an infinite recursion in
 568                  * vm_map_entry_reserve() by using vm_map_entry_kreserve()
 569                  * instead.  The map entries are pre-reserved by the kernel
 570                  * by vm_map_entry_reserve_cpu_init().
 571                  */
 572                 nbytes = (size_t)z->zalloc * PAGE_SIZE;
 573
 574                 item = (void *)kmem_alloc3(&kernel_map, nbytes,
 575                                            VM_SUBSYS_ZALLOC, KM_KRESERVE);
 576
 577                 /* note: z might be modified due to blocking */
 578                 if (item != NULL) {
 579                         atomic_add_long(&zone_kern_pages, z->zalloc);
 580                         bzero(item, nbytes);
 581                 } else {
 582                         nbytes = 0;
 583                 }
 584                 nitems = nbytes / z->zsize;
 585         } else {
 586                 /*
 587                  * Otherwise allocate KVA from the kernel_map.
 588                  */
 589                 nbytes = (size_t)z->zalloc * PAGE_SIZE;
 590
 591                 item = (void *)kmem_alloc3(&kernel_map, nbytes,
 592                                            VM_SUBSYS_ZALLOC, 0);
 593
 594                 /* note: z might be modified due to blocking */
 595                 if (item != NULL) {
 596                         atomic_add_long(&zone_kern_pages, z->zalloc);
 597                         bzero(item, nbytes);
 598
 599                         if (z->zflags & ZONE_DESTROYABLE) {
 600                                 if (z->zkmcur == z->zkmmax) {
 601                                         z->zkmmax =
 602                                                 z->zkmmax==0 ? 1 : z->zkmmax*2;
 603                                         z->zkmvec = krealloc(z->zkmvec,
 604                                             z->zkmmax * sizeof(z->zkmvec[0]),
 605                                             M_ZONE, M_WAITOK);
 606                                 }
 607                                 z->zkmvec[z->zkmcur++] = (vm_offset_t)item;
 608                         }
 609                 } else {
 610                         nbytes = 0;
 611                 }
 612                 nitems = nbytes / z->zsize;
 613         }
 614
 615         /*
 616          * Enter any new pages into the pool, reserving one, or get the
 617          * item from the existing pool.
 618          */
 619         spin_lock(&z->zspin);
 620         z->ztotal += nitems;
 621
 622         /*
 623          * The zone code may need to allocate kernel memory, which can
 624          * recurse zget() infinitely if we do not handle it properly.
 625          * We deal with this by directly repopulating the pcpu vm_map_entry
 626          * cache.
 627          */
 628         if (nitems > 1 && (z->zflags & ZONE_SPECIAL)) {
 629                 struct globaldata *gd = mycpu;
 630                 vm_map_entry_t entry;
 631
 632                 /*
 633                  * Make sure we have enough structures in gd_vme_base to handle
 634                  * the reservation request.
 635                  *
 636                  * The critical section protects access to the per-cpu gd.
 637                  */
 638                 crit_enter();
 639                 while (gd->gd_vme_avail < 2 && nitems > 1) {
 640                         entry = item;
 641                         entry->next = gd->gd_vme_base;
 642                         gd->gd_vme_base = entry;
 643                         atomic_add_int(&gd->gd_vme_avail, 1);
 644                         item = (uint8_t *)item + z->zsize;
 645                         --nitems;
 646                 }
 647                 crit_exit();
 648         }
 649
 650         if (nitems != 0) {
 651                 /*
 652                  * Enter pages into the pool saving one for immediate
 653                  * allocation.
 654                  */
 655                 nitems -= 1;
 656                 for (i = 0; i < nitems; i++) {
 657                         ((void **)item)[0] = z->zitems;
 658 #ifdef INVARIANTS
 659                         ((void **)item)[1] = (void *)ZENTRY_FREE;
 660 #endif
 661                         z->zitems = item;
 662                         item = (uint8_t *)item + z->zsize;
 663                 }
 664                 z->zfreecnt += nitems;
 665                 ++z->znalloc;
 666         } else if (z->zfreecnt > 0) {
 667                 /*
 668                  * Get an item from the existing pool.
 669                  */
 670                 item = z->zitems;
 671                 z->zitems = ((void **)item)[0];
 672 #ifdef INVARIANTS
 673                 if (((void **)item)[1] != (void *)ZENTRY_FREE)
 674                         zerror(ZONE_ERROR_NOTFREE);
 675                 ((void **) item)[1] = NULL;
 676 #endif
 677                 --z->zfreecnt;
 678                 ++z->znalloc;
 679         } else {
 680                 /*
 681                  * No items available.
 682                  */
 683                 item = NULL;
 684         }
 685         spin_unlock(&z->zspin);
 686
 687         return item;
 688 }
 689
 690 /*
 691  * No requirements.
 692  */
 693 static int
 694 sysctl_vm_zone(SYSCTL_HANDLER_ARGS)
 695 {
 696         vm_zone_t curzone;
 697         char tmpbuf[128];
 698         char tmpname[14];
 699         int error = 0;
 700
 701         ksnprintf(tmpbuf, sizeof(tmpbuf),
 702             "\nITEM            SIZE     LIMIT    USED    FREE  REQUESTS\n");
 703         error = SYSCTL_OUT(req, tmpbuf, strlen(tmpbuf));
 704         if (error)
 705                 return (error);
 706
 707         lwkt_gettoken(&vm_token);
 708         LIST_FOREACH(curzone, &zlist, zlink) {
 709                 size_t i;
 710                 size_t len;
 711                 int offset;
 712                 long freecnt;
 713                 long znalloc;
 714                 int n;
 715
 716                 len = strlen(curzone->zname);
 717                 if (len >= (sizeof(tmpname) - 1))
 718                         len = (sizeof(tmpname) - 1);
 719                 for(i = 0; i < sizeof(tmpname) - 1; i++)
 720                         tmpname[i] = ' ';
 721                 tmpname[i] = 0;
 722                 memcpy(tmpname, curzone->zname, len);
 723                 tmpname[len] = ':';
 724                 offset = 0;
 725                 if (curzone == LIST_FIRST(&zlist)) {
 726                         offset = 1;
 727                         tmpbuf[0] = '\n';
 728                 }
 729                 freecnt = curzone->zfreecnt;
 730                 znalloc = curzone->znalloc;
 731                 for (n = 0; n < ncpus; ++n) {
 732                         freecnt += curzone->zpcpu[n].zfreecnt;
 733                         znalloc += curzone->zpcpu[n].znalloc;
 734                 }
 735
 736                 ksnprintf(tmpbuf + offset, sizeof(tmpbuf) - offset,
 737                         "%s %6.6lu, %8.8lu, %6.6lu, %6.6lu, %8.8lu\n",
 738                         tmpname, curzone->zsize, curzone->zmax,
 739                         (curzone->ztotal - freecnt),
 740                         freecnt, znalloc);
 741
 742                 len = strlen((char *)tmpbuf);
 743                 if (LIST_NEXT(curzone, zlink) == NULL)
 744                         tmpbuf[len - 1] = 0;
 745
 746                 error = SYSCTL_OUT(req, tmpbuf, len);
 747
 748                 if (error)
 749                         break;
 750         }
 751         lwkt_reltoken(&vm_token);
 752         return (error);
 753 }
 754
 755 #if defined(INVARIANTS)
 756
 757 /*
 758  * Debugging only.
 759  */
 760 void
 761 zerror(int error)
 762 {
 763         char *msg;
 764
 765         switch (error) {
 766         case ZONE_ERROR_INVALID:
 767                 msg = "zone: invalid zone";
 768                 break;
 769         case ZONE_ERROR_NOTFREE:
 770                 msg = "zone: entry not free";
 771                 break;
 772         case ZONE_ERROR_ALREADYFREE:
 773                 msg = "zone: freeing free entry";
 774                 break;
 775         default:
 776                 msg = "zone: invalid error";
 777                 break;
 778         }
 779         panic("%s", msg);
 780 }
 781 #endif
 782
 783 SYSCTL_OID(_vm, OID_AUTO, zone, CTLTYPE_STRING|CTLFLAG_RD, \
 784         NULL, 0, sysctl_vm_zone, "A", "Zone Info");
 785
 786 SYSCTL_LONG(_vm, OID_AUTO, zone_kmem_pages,
 787         CTLFLAG_RD, &zone_kmem_pages, 0, "Number of interrupt safe pages allocated by zone");
 788 SYSCTL_LONG(_vm, OID_AUTO, zone_burst,
 789         CTLFLAG_RW, &zone_burst, 0, "Burst from depot to pcpu cache");
 790 SYSCTL_LONG(_vm, OID_AUTO, zone_kmem_kvaspace,
 791         CTLFLAG_RD, &zone_kmem_kvaspace, 0, "KVA space allocated by zone");
 792 SYSCTL_LONG(_vm, OID_AUTO, zone_kern_pages,
 793         CTLFLAG_RD, &zone_kern_pages, 0, "Number of non-interrupt safe pages allocated by zone");