The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_malloc.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1987, 1991, 1993
    3  *      The Regents of the University of California.
    4  * Copyright (c) 2005-2009 Robert N. M. Watson
    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, this list of conditions and the following disclaimer.
   12  * 2. Redistributions in binary form must reproduce the above copyright
   13  *    notice, this list of conditions and the following disclaimer in the
   14  *    documentation and/or other materials provided with the distribution.
   15  * 4. Neither the name of the University nor the names of its contributors
   16  *    may be used to endorse or promote products derived from this software
   17  *    without specific prior written permission.
   18  *
   19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   29  * SUCH DAMAGE.
   30  *
   31  *      @(#)kern_malloc.c       8.3 (Berkeley) 1/4/94
   32  */
   33 
   34 /*
   35  * Kernel malloc(9) implementation -- general purpose kernel memory allocator
   36  * based on memory types.  Back end is implemented using the UMA(9) zone
   37  * allocator.  A set of fixed-size buckets are used for smaller allocations,
   38  * and a special UMA allocation interface is used for larger allocations.
   39  * Callers declare memory types, and statistics are maintained independently
   40  * for each memory type.  Statistics are maintained per-CPU for performance
   41  * reasons.  See malloc(9) and comments in malloc.h for a detailed
   42  * description.
   43  */
   44 
   45 #include <sys/cdefs.h>
   46 __FBSDID("$FreeBSD$");
   47 
   48 #include "opt_ddb.h"
   49 #include "opt_kdtrace.h"
   50 #include "opt_vm.h"
   51 
   52 #include <sys/param.h>
   53 #include <sys/systm.h>
   54 #include <sys/kdb.h>
   55 #include <sys/kernel.h>
   56 #include <sys/lock.h>
   57 #include <sys/malloc.h>
   58 #include <sys/mbuf.h>
   59 #include <sys/mutex.h>
   60 #include <sys/vmmeter.h>
   61 #include <sys/proc.h>
   62 #include <sys/sbuf.h>
   63 #include <sys/sysctl.h>
   64 #include <sys/time.h>
   65 
   66 #include <vm/vm.h>
   67 #include <vm/pmap.h>
   68 #include <vm/vm_param.h>
   69 #include <vm/vm_kern.h>
   70 #include <vm/vm_extern.h>
   71 #include <vm/vm_map.h>
   72 #include <vm/vm_page.h>
   73 #include <vm/uma.h>
   74 #include <vm/uma_int.h>
   75 #include <vm/uma_dbg.h>
   76 
   77 #ifdef DEBUG_MEMGUARD
   78 #include <vm/memguard.h>
   79 #endif
   80 #ifdef DEBUG_REDZONE
   81 #include <vm/redzone.h>
   82 #endif
   83 
   84 #if defined(INVARIANTS) && defined(__i386__)
   85 #include <machine/cpu.h>
   86 #endif
   87 
   88 #include <ddb/ddb.h>
   89 
   90 #ifdef KDTRACE_HOOKS
   91 #include <sys/dtrace_bsd.h>
   92 
   93 dtrace_malloc_probe_func_t      dtrace_malloc_probe;
   94 #endif
   95 
   96 /*
   97  * When realloc() is called, if the new size is sufficiently smaller than
   98  * the old size, realloc() will allocate a new, smaller block to avoid
   99  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
  100  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
  101  */
  102 #ifndef REALLOC_FRACTION
  103 #define REALLOC_FRACTION        1       /* new block if <= half the size */
  104 #endif
  105 
  106 /*
  107  * Centrally define some common malloc types.
  108  */
  109 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
  110 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
  111 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
  112 
  113 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
  114 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
  115 
  116 static void kmeminit(void *);
  117 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL);
  118 
  119 static struct malloc_type *kmemstatistics;
  120 static vm_offset_t kmembase;
  121 static vm_offset_t kmemlimit;
  122 static int kmemcount;
  123 
  124 #define KMEM_ZSHIFT     4
  125 #define KMEM_ZBASE      16
  126 #define KMEM_ZMASK      (KMEM_ZBASE - 1)
  127 
  128 #define KMEM_ZMAX       PAGE_SIZE
  129 #define KMEM_ZSIZE      (KMEM_ZMAX >> KMEM_ZSHIFT)
  130 static u_int8_t kmemsize[KMEM_ZSIZE + 1];
  131 
  132 /*
  133  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
  134  * of various sizes.
  135  *
  136  * XXX: The comment here used to read "These won't be powers of two for
  137  * long."  It's possible that a significant amount of wasted memory could be
  138  * recovered by tuning the sizes of these buckets.
  139  */
  140 struct {
  141         int kz_size;
  142         char *kz_name;
  143         uma_zone_t kz_zone;
  144 } kmemzones[] = {
  145         {16, "16", NULL},
  146         {32, "32", NULL},
  147         {64, "64", NULL},
  148         {128, "128", NULL},
  149         {256, "256", NULL},
  150         {512, "512", NULL},
  151         {1024, "1024", NULL},
  152         {2048, "2048", NULL},
  153         {4096, "4096", NULL},
  154 #if PAGE_SIZE > 4096
  155         {8192, "8192", NULL},
  156 #if PAGE_SIZE > 8192
  157         {16384, "16384", NULL},
  158 #if PAGE_SIZE > 16384
  159         {32768, "32768", NULL},
  160 #if PAGE_SIZE > 32768
  161         {65536, "65536", NULL},
  162 #if PAGE_SIZE > 65536
  163 #error  "Unsupported PAGE_SIZE"
  164 #endif  /* 65536 */
  165 #endif  /* 32768 */
  166 #endif  /* 16384 */
  167 #endif  /* 8192 */
  168 #endif  /* 4096 */
  169         {0, NULL},
  170 };
  171 
  172 /*
  173  * Zone to allocate malloc type descriptions from.  For ABI reasons, memory
  174  * types are described by a data structure passed by the declaring code, but
  175  * the malloc(9) implementation has its own data structure describing the
  176  * type and statistics.  This permits the malloc(9)-internal data structures
  177  * to be modified without breaking binary-compiled kernel modules that
  178  * declare malloc types.
  179  */
  180 static uma_zone_t mt_zone;
  181 
  182 u_long vm_kmem_size;
  183 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
  184     "Size of kernel memory");
  185 
  186 static u_long vm_kmem_size_min;
  187 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
  188     "Minimum size of kernel memory");
  189 
  190 static u_long vm_kmem_size_max;
  191 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
  192     "Maximum size of kernel memory");
  193 
  194 static u_int vm_kmem_size_scale;
  195 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
  196     "Scale factor for kernel memory size");
  197 
  198 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
  199 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
  200     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
  201     sysctl_kmem_map_size, "LU", "Current kmem_map allocation size");
  202 
  203 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
  204 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
  205     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
  206     sysctl_kmem_map_free, "LU", "Largest contiguous free range in kmem_map");
  207 
  208 /*
  209  * The malloc_mtx protects the kmemstatistics linked list.
  210  */
  211 struct mtx malloc_mtx;
  212 
  213 #ifdef MALLOC_PROFILE
  214 uint64_t krequests[KMEM_ZSIZE + 1];
  215 
  216 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
  217 #endif
  218 
  219 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
  220 
  221 /*
  222  * time_uptime of the last malloc(9) failure (induced or real).
  223  */
  224 static time_t t_malloc_fail;
  225 
  226 /*
  227  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
  228  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
  229  */
  230 #ifdef MALLOC_MAKE_FAILURES
  231 SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
  232     "Kernel malloc debugging options");
  233 
  234 static int malloc_failure_rate;
  235 static int malloc_nowait_count;
  236 static int malloc_failure_count;
  237 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW,
  238     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
  239 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate);
  240 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
  241     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
  242 #endif
  243 
  244 static int
  245 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
  246 {
  247         u_long size;
  248 
  249         size = kmem_map->size;
  250         return (sysctl_handle_long(oidp, &size, 0, req));
  251 }
  252 
  253 static int
  254 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
  255 {
  256         u_long size;
  257 
  258         vm_map_lock_read(kmem_map);
  259         size = kmem_map->root != NULL ? kmem_map->root->max_free :
  260             kmem_map->max_offset - kmem_map->min_offset;
  261         vm_map_unlock_read(kmem_map);
  262         return (sysctl_handle_long(oidp, &size, 0, req));
  263 }
  264 
  265 int
  266 malloc_last_fail(void)
  267 {
  268 
  269         return (time_uptime - t_malloc_fail);
  270 }
  271 
  272 /*
  273  * An allocation has succeeded -- update malloc type statistics for the
  274  * amount of bucket size.  Occurs within a critical section so that the
  275  * thread isn't preempted and doesn't migrate while updating per-PCU
  276  * statistics.
  277  */
  278 static void
  279 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
  280     int zindx)
  281 {
  282         struct malloc_type_internal *mtip;
  283         struct malloc_type_stats *mtsp;
  284 
  285         critical_enter();
  286         mtip = mtp->ks_handle;
  287         mtsp = &mtip->mti_stats[curcpu];
  288         if (size > 0) {
  289                 mtsp->mts_memalloced += size;
  290                 mtsp->mts_numallocs++;
  291         }
  292         if (zindx != -1)
  293                 mtsp->mts_size |= 1 << zindx;
  294 
  295 #ifdef KDTRACE_HOOKS
  296         if (dtrace_malloc_probe != NULL) {
  297                 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
  298                 if (probe_id != 0)
  299                         (dtrace_malloc_probe)(probe_id,
  300                             (uintptr_t) mtp, (uintptr_t) mtip,
  301                             (uintptr_t) mtsp, size, zindx);
  302         }
  303 #endif
  304 
  305         critical_exit();
  306 }
  307 
  308 void
  309 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
  310 {
  311 
  312         if (size > 0)
  313                 malloc_type_zone_allocated(mtp, size, -1);
  314 }
  315 
  316 /*
  317  * A free operation has occurred -- update malloc type statistics for the
  318  * amount of the bucket size.  Occurs within a critical section so that the
  319  * thread isn't preempted and doesn't migrate while updating per-CPU
  320  * statistics.
  321  */
  322 void
  323 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
  324 {
  325         struct malloc_type_internal *mtip;
  326         struct malloc_type_stats *mtsp;
  327 
  328         critical_enter();
  329         mtip = mtp->ks_handle;
  330         mtsp = &mtip->mti_stats[curcpu];
  331         mtsp->mts_memfreed += size;
  332         mtsp->mts_numfrees++;
  333 
  334 #ifdef KDTRACE_HOOKS
  335         if (dtrace_malloc_probe != NULL) {
  336                 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
  337                 if (probe_id != 0)
  338                         (dtrace_malloc_probe)(probe_id,
  339                             (uintptr_t) mtp, (uintptr_t) mtip,
  340                             (uintptr_t) mtsp, size, 0);
  341         }
  342 #endif
  343 
  344         critical_exit();
  345 }
  346 
  347 /*
  348  *      malloc:
  349  *
  350  *      Allocate a block of memory.
  351  *
  352  *      If M_NOWAIT is set, this routine will not block and return NULL if
  353  *      the allocation fails.
  354  */
  355 void *
  356 malloc(unsigned long size, struct malloc_type *mtp, int flags)
  357 {
  358         int indx;
  359         caddr_t va;
  360         uma_zone_t zone;
  361 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
  362         unsigned long osize = size;
  363 #endif
  364 
  365 #ifdef INVARIANTS
  366         KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
  367         /*
  368          * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
  369          */
  370         indx = flags & (M_WAITOK | M_NOWAIT);
  371         if (indx != M_NOWAIT && indx != M_WAITOK) {
  372                 static  struct timeval lasterr;
  373                 static  int curerr, once;
  374                 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
  375                         printf("Bad malloc flags: %x\n", indx);
  376                         kdb_backtrace();
  377                         flags |= M_WAITOK;
  378                         once++;
  379                 }
  380         }
  381 #endif
  382 #ifdef MALLOC_MAKE_FAILURES
  383         if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
  384                 atomic_add_int(&malloc_nowait_count, 1);
  385                 if ((malloc_nowait_count % malloc_failure_rate) == 0) {
  386                         atomic_add_int(&malloc_failure_count, 1);
  387                         t_malloc_fail = time_uptime;
  388                         return (NULL);
  389                 }
  390         }
  391 #endif
  392         if (flags & M_WAITOK)
  393                 KASSERT(curthread->td_intr_nesting_level == 0,
  394                    ("malloc(M_WAITOK) in interrupt context"));
  395 
  396 #ifdef DEBUG_MEMGUARD
  397         if (memguard_cmp(mtp, size)) {
  398                 va = memguard_alloc(size, flags);
  399                 if (va != NULL)
  400                         return (va);
  401                 /* This is unfortunate but should not be fatal. */
  402         }
  403 #endif
  404 
  405 #ifdef DEBUG_REDZONE
  406         size = redzone_size_ntor(size);
  407 #endif
  408 
  409         if (size <= KMEM_ZMAX) {
  410                 if (size & KMEM_ZMASK)
  411                         size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
  412                 indx = kmemsize[size >> KMEM_ZSHIFT];
  413                 zone = kmemzones[indx].kz_zone;
  414 #ifdef MALLOC_PROFILE
  415                 krequests[size >> KMEM_ZSHIFT]++;
  416 #endif
  417                 va = uma_zalloc(zone, flags);
  418                 if (va != NULL)
  419                         size = zone->uz_size;
  420                 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
  421         } else {
  422                 size = roundup(size, PAGE_SIZE);
  423                 zone = NULL;
  424                 va = uma_large_malloc(size, flags);
  425                 malloc_type_allocated(mtp, va == NULL ? 0 : size);
  426         }
  427         if (flags & M_WAITOK)
  428                 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
  429         else if (va == NULL)
  430                 t_malloc_fail = time_uptime;
  431 #ifdef DIAGNOSTIC
  432         if (va != NULL && !(flags & M_ZERO)) {
  433                 memset(va, 0x70, osize);
  434         }
  435 #endif
  436 #ifdef DEBUG_REDZONE
  437         if (va != NULL)
  438                 va = redzone_setup(va, osize);
  439 #endif
  440         return ((void *) va);
  441 }
  442 
  443 /*
  444  *      free:
  445  *
  446  *      Free a block of memory allocated by malloc.
  447  *
  448  *      This routine may not block.
  449  */
  450 void
  451 free(void *addr, struct malloc_type *mtp)
  452 {
  453         uma_slab_t slab;
  454         u_long size;
  455 
  456         KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
  457 
  458         /* free(NULL, ...) does nothing */
  459         if (addr == NULL)
  460                 return;
  461 
  462 #ifdef DEBUG_MEMGUARD
  463         if (is_memguard_addr(addr)) {
  464                 memguard_free(addr);
  465                 return;
  466         }
  467 #endif
  468 
  469 #ifdef DEBUG_REDZONE
  470         redzone_check(addr);
  471         addr = redzone_addr_ntor(addr);
  472 #endif
  473 
  474         slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
  475 
  476         if (slab == NULL)
  477                 panic("free: address %p(%p) has not been allocated.\n",
  478                     addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
  479 
  480 
  481         if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
  482 #ifdef INVARIANTS
  483                 struct malloc_type **mtpp = addr;
  484 #endif
  485                 size = slab->us_keg->uk_size;
  486 #ifdef INVARIANTS
  487                 /*
  488                  * Cache a pointer to the malloc_type that most recently freed
  489                  * this memory here.  This way we know who is most likely to
  490                  * have stepped on it later.
  491                  *
  492                  * This code assumes that size is a multiple of 8 bytes for
  493                  * 64 bit machines
  494                  */
  495                 mtpp = (struct malloc_type **)
  496                     ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
  497                 mtpp += (size - sizeof(struct malloc_type *)) /
  498                     sizeof(struct malloc_type *);
  499                 *mtpp = mtp;
  500 #endif
  501                 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
  502         } else {
  503                 size = slab->us_size;
  504                 uma_large_free(slab);
  505         }
  506         malloc_type_freed(mtp, size);
  507 }
  508 
  509 /*
  510  *      realloc: change the size of a memory block
  511  */
  512 void *
  513 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
  514 {
  515         uma_slab_t slab;
  516         unsigned long alloc;
  517         void *newaddr;
  518 
  519         KASSERT(mtp->ks_magic == M_MAGIC,
  520             ("realloc: bad malloc type magic"));
  521 
  522         /* realloc(NULL, ...) is equivalent to malloc(...) */
  523         if (addr == NULL)
  524                 return (malloc(size, mtp, flags));
  525 
  526         /*
  527          * XXX: Should report free of old memory and alloc of new memory to
  528          * per-CPU stats.
  529          */
  530 
  531 #ifdef DEBUG_MEMGUARD
  532         if (is_memguard_addr(addr))
  533                 return (memguard_realloc(addr, size, mtp, flags));
  534 #endif
  535 
  536 #ifdef DEBUG_REDZONE
  537         slab = NULL;
  538         alloc = redzone_get_size(addr);
  539 #else
  540         slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
  541 
  542         /* Sanity check */
  543         KASSERT(slab != NULL,
  544             ("realloc: address %p out of range", (void *)addr));
  545 
  546         /* Get the size of the original block */
  547         if (!(slab->us_flags & UMA_SLAB_MALLOC))
  548                 alloc = slab->us_keg->uk_size;
  549         else
  550                 alloc = slab->us_size;
  551 
  552         /* Reuse the original block if appropriate */
  553         if (size <= alloc
  554             && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
  555                 return (addr);
  556 #endif /* !DEBUG_REDZONE */
  557 
  558         /* Allocate a new, bigger (or smaller) block */
  559         if ((newaddr = malloc(size, mtp, flags)) == NULL)
  560                 return (NULL);
  561 
  562         /* Copy over original contents */
  563         bcopy(addr, newaddr, min(size, alloc));
  564         free(addr, mtp);
  565         return (newaddr);
  566 }
  567 
  568 /*
  569  *      reallocf: same as realloc() but free memory on failure.
  570  */
  571 void *
  572 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
  573 {
  574         void *mem;
  575 
  576         if ((mem = realloc(addr, size, mtp, flags)) == NULL)
  577                 free(addr, mtp);
  578         return (mem);
  579 }
  580 
  581 /*
  582  * Initialize the kernel memory allocator
  583  */
  584 /* ARGSUSED*/
  585 static void
  586 kmeminit(void *dummy)
  587 {
  588         u_int8_t indx;
  589         u_long mem_size, tmp;
  590         int i;
  591  
  592         mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
  593 
  594         /*
  595          * Try to auto-tune the kernel memory size, so that it is
  596          * more applicable for a wider range of machine sizes.  The
  597          * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
  598          * available.
  599          *
  600          * Note that the kmem_map is also used by the zone allocator,
  601          * so make sure that there is enough space.
  602          */
  603         vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE;
  604         mem_size = cnt.v_page_count;
  605 
  606 #if defined(VM_KMEM_SIZE_SCALE)
  607         vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
  608 #endif
  609         TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
  610         if (vm_kmem_size_scale > 0 &&
  611             (mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE))
  612                 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
  613 
  614 #if defined(VM_KMEM_SIZE_MIN)
  615         vm_kmem_size_min = VM_KMEM_SIZE_MIN;
  616 #endif
  617         TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min);
  618         if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) {
  619                 vm_kmem_size = vm_kmem_size_min;
  620         }
  621 
  622 #if defined(VM_KMEM_SIZE_MAX)
  623         vm_kmem_size_max = VM_KMEM_SIZE_MAX;
  624 #endif
  625         TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
  626         if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
  627                 vm_kmem_size = vm_kmem_size_max;
  628 
  629         /* Allow final override from the kernel environment */
  630         TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size);
  631 
  632         /*
  633          * Limit kmem virtual size to twice the physical memory.
  634          * This allows for kmem map sparseness, but limits the size
  635          * to something sane.  Be careful to not overflow the 32bit
  636          * ints while doing the check or the adjustment.
  637          */
  638         if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
  639                 vm_kmem_size = 2 * mem_size * PAGE_SIZE;
  640 
  641         /*
  642          * Tune settings based on the kmem map's size at this time.
  643          */
  644         init_param3(vm_kmem_size / PAGE_SIZE);
  645 
  646 #ifdef DEBUG_MEMGUARD
  647         tmp = memguard_fudge(vm_kmem_size, kernel_map);
  648 #else
  649         tmp = vm_kmem_size;
  650 #endif
  651         kmem_map = kmem_suballoc(kernel_map, &kmembase, &kmemlimit,
  652             tmp, TRUE);
  653         kmem_map->system_map = 1;
  654 
  655 #ifdef DEBUG_MEMGUARD
  656         /*
  657          * Initialize MemGuard if support compiled in.  MemGuard is a
  658          * replacement allocator used for detecting tamper-after-free
  659          * scenarios as they occur.  It is only used for debugging.
  660          */
  661         memguard_init(kmem_map);
  662 #endif
  663 
  664         uma_startup2();
  665 
  666         mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
  667 #ifdef INVARIANTS
  668             mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
  669 #else
  670             NULL, NULL, NULL, NULL,
  671 #endif
  672             UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
  673         for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
  674                 int size = kmemzones[indx].kz_size;
  675                 char *name = kmemzones[indx].kz_name;
  676 
  677                 kmemzones[indx].kz_zone = uma_zcreate(name, size,
  678 #ifdef INVARIANTS
  679                     mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
  680 #else
  681                     NULL, NULL, NULL, NULL,
  682 #endif
  683                     UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
  684                     
  685                 for (;i <= size; i+= KMEM_ZBASE)
  686                         kmemsize[i >> KMEM_ZSHIFT] = indx;
  687                 
  688         }
  689 }
  690 
  691 void
  692 malloc_init(void *data)
  693 {
  694         struct malloc_type_internal *mtip;
  695         struct malloc_type *mtp;
  696 
  697         KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
  698 
  699         mtp = data;
  700         if (mtp->ks_magic != M_MAGIC)
  701                 panic("malloc_init: bad malloc type magic");
  702 
  703         mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
  704         mtp->ks_handle = mtip;
  705 
  706         mtx_lock(&malloc_mtx);
  707         mtp->ks_next = kmemstatistics;
  708         kmemstatistics = mtp;
  709         kmemcount++;
  710         mtx_unlock(&malloc_mtx);
  711 }
  712 
  713 void
  714 malloc_uninit(void *data)
  715 {
  716         struct malloc_type_internal *mtip;
  717         struct malloc_type_stats *mtsp;
  718         struct malloc_type *mtp, *temp;
  719         uma_slab_t slab;
  720         long temp_allocs, temp_bytes;
  721         int i;
  722 
  723         mtp = data;
  724         KASSERT(mtp->ks_magic == M_MAGIC,
  725             ("malloc_uninit: bad malloc type magic"));
  726         KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
  727 
  728         mtx_lock(&malloc_mtx);
  729         mtip = mtp->ks_handle;
  730         mtp->ks_handle = NULL;
  731         if (mtp != kmemstatistics) {
  732                 for (temp = kmemstatistics; temp != NULL;
  733                     temp = temp->ks_next) {
  734                         if (temp->ks_next == mtp) {
  735                                 temp->ks_next = mtp->ks_next;
  736                                 break;
  737                         }
  738                 }
  739                 KASSERT(temp,
  740                     ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
  741         } else
  742                 kmemstatistics = mtp->ks_next;
  743         kmemcount--;
  744         mtx_unlock(&malloc_mtx);
  745 
  746         /*
  747          * Look for memory leaks.
  748          */
  749         temp_allocs = temp_bytes = 0;
  750         for (i = 0; i < MAXCPU; i++) {
  751                 mtsp = &mtip->mti_stats[i];
  752                 temp_allocs += mtsp->mts_numallocs;
  753                 temp_allocs -= mtsp->mts_numfrees;
  754                 temp_bytes += mtsp->mts_memalloced;
  755                 temp_bytes -= mtsp->mts_memfreed;
  756         }
  757         if (temp_allocs > 0 || temp_bytes > 0) {
  758                 printf("Warning: memory type %s leaked memory on destroy "
  759                     "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
  760                     temp_allocs, temp_bytes);
  761         }
  762 
  763         slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
  764         uma_zfree_arg(mt_zone, mtip, slab);
  765 }
  766 
  767 struct malloc_type *
  768 malloc_desc2type(const char *desc)
  769 {
  770         struct malloc_type *mtp;
  771 
  772         mtx_assert(&malloc_mtx, MA_OWNED);
  773         for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
  774                 if (strcmp(mtp->ks_shortdesc, desc) == 0)
  775                         return (mtp);
  776         }
  777         return (NULL);
  778 }
  779 
  780 static int
  781 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
  782 {
  783         struct malloc_type_stream_header mtsh;
  784         struct malloc_type_internal *mtip;
  785         struct malloc_type_header mth;
  786         struct malloc_type *mtp;
  787         int buflen, count, error, i;
  788         struct sbuf sbuf;
  789         char *buffer;
  790 
  791         mtx_lock(&malloc_mtx);
  792 restart:
  793         mtx_assert(&malloc_mtx, MA_OWNED);
  794         count = kmemcount;
  795         mtx_unlock(&malloc_mtx);
  796         buflen = sizeof(mtsh) + count * (sizeof(mth) +
  797             sizeof(struct malloc_type_stats) * MAXCPU) + 1;
  798         buffer = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
  799         mtx_lock(&malloc_mtx);
  800         if (count < kmemcount) {
  801                 free(buffer, M_TEMP);
  802                 goto restart;
  803         }
  804 
  805         sbuf_new(&sbuf, buffer, buflen, SBUF_FIXEDLEN);
  806 
  807         /*
  808          * Insert stream header.
  809          */
  810         bzero(&mtsh, sizeof(mtsh));
  811         mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
  812         mtsh.mtsh_maxcpus = MAXCPU;
  813         mtsh.mtsh_count = kmemcount;
  814         if (sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)) < 0) {
  815                 mtx_unlock(&malloc_mtx);
  816                 error = ENOMEM;
  817                 goto out;
  818         }
  819 
  820         /*
  821          * Insert alternating sequence of type headers and type statistics.
  822          */
  823         for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
  824                 mtip = (struct malloc_type_internal *)mtp->ks_handle;
  825 
  826                 /*
  827                  * Insert type header.
  828                  */
  829                 bzero(&mth, sizeof(mth));
  830                 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
  831                 if (sbuf_bcat(&sbuf, &mth, sizeof(mth)) < 0) {
  832                         mtx_unlock(&malloc_mtx);
  833                         error = ENOMEM;
  834                         goto out;
  835                 }
  836 
  837                 /*
  838                  * Insert type statistics for each CPU.
  839                  */
  840                 for (i = 0; i < MAXCPU; i++) {
  841                         if (sbuf_bcat(&sbuf, &mtip->mti_stats[i],
  842                             sizeof(mtip->mti_stats[i])) < 0) {
  843                                 mtx_unlock(&malloc_mtx);
  844                                 error = ENOMEM;
  845                                 goto out;
  846                         }
  847                 }
  848         }
  849         mtx_unlock(&malloc_mtx);
  850         sbuf_finish(&sbuf);
  851         error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
  852 out:
  853         sbuf_delete(&sbuf);
  854         free(buffer, M_TEMP);
  855         return (error);
  856 }
  857 
  858 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
  859     0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
  860     "Return malloc types");
  861 
  862 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
  863     "Count of kernel malloc types");
  864 
  865 void
  866 malloc_type_list(malloc_type_list_func_t *func, void *arg)
  867 {
  868         struct malloc_type *mtp, **bufmtp;
  869         int count, i;
  870         size_t buflen;
  871 
  872         mtx_lock(&malloc_mtx);
  873 restart:
  874         mtx_assert(&malloc_mtx, MA_OWNED);
  875         count = kmemcount;
  876         mtx_unlock(&malloc_mtx);
  877 
  878         buflen = sizeof(struct malloc_type *) * count;
  879         bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
  880 
  881         mtx_lock(&malloc_mtx);
  882 
  883         if (count < kmemcount) {
  884                 free(bufmtp, M_TEMP);
  885                 goto restart;
  886         }
  887 
  888         for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
  889                 bufmtp[i] = mtp;
  890 
  891         mtx_unlock(&malloc_mtx);
  892 
  893         for (i = 0; i < count; i++)
  894                 (func)(bufmtp[i], arg);
  895 
  896         free(bufmtp, M_TEMP);
  897 }
  898 
  899 #ifdef DDB
  900 DB_SHOW_COMMAND(malloc, db_show_malloc)
  901 {
  902         struct malloc_type_internal *mtip;
  903         struct malloc_type *mtp;
  904         u_int64_t allocs, frees;
  905         u_int64_t alloced, freed;
  906         int i;
  907 
  908         db_printf("%18s %12s  %12s %12s\n", "Type", "InUse", "MemUse",
  909             "Requests");
  910         for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
  911                 mtip = (struct malloc_type_internal *)mtp->ks_handle;
  912                 allocs = 0;
  913                 frees = 0;
  914                 alloced = 0;
  915                 freed = 0;
  916                 for (i = 0; i < MAXCPU; i++) {
  917                         allocs += mtip->mti_stats[i].mts_numallocs;
  918                         frees += mtip->mti_stats[i].mts_numfrees;
  919                         alloced += mtip->mti_stats[i].mts_memalloced;
  920                         freed += mtip->mti_stats[i].mts_memfreed;
  921                 }
  922                 db_printf("%18s %12ju %12juK %12ju\n",
  923                     mtp->ks_shortdesc, allocs - frees,
  924                     (alloced - freed + 1023) / 1024, allocs);
  925                 if (db_pager_quit)
  926                         break;
  927         }
  928 }
  929 #endif
  930 
  931 #ifdef MALLOC_PROFILE
  932 
  933 static int
  934 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
  935 {
  936         int linesize = 64;
  937         struct sbuf sbuf;
  938         uint64_t count;
  939         uint64_t waste;
  940         uint64_t mem;
  941         int bufsize;
  942         int error;
  943         char *buf;
  944         int rsize;
  945         int size;
  946         int i;
  947 
  948         bufsize = linesize * (KMEM_ZSIZE + 1);
  949         bufsize += 128;         /* For the stats line */
  950         bufsize += 128;         /* For the banner line */
  951         waste = 0;
  952         mem = 0;
  953 
  954         buf = malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
  955         sbuf_new(&sbuf, buf, bufsize, SBUF_FIXEDLEN);
  956         sbuf_printf(&sbuf, 
  957             "\n  Size                    Requests  Real Size\n");
  958         for (i = 0; i < KMEM_ZSIZE; i++) {
  959                 size = i << KMEM_ZSHIFT;
  960                 rsize = kmemzones[kmemsize[i]].kz_size;
  961                 count = (long long unsigned)krequests[i];
  962 
  963                 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
  964                     (unsigned long long)count, rsize);
  965 
  966                 if ((rsize * count) > (size * count))
  967                         waste += (rsize * count) - (size * count);
  968                 mem += (rsize * count);
  969         }
  970         sbuf_printf(&sbuf,
  971             "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
  972             (unsigned long long)mem, (unsigned long long)waste);
  973         sbuf_finish(&sbuf);
  974 
  975         error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
  976 
  977         sbuf_delete(&sbuf);
  978         free(buf, M_TEMP);
  979         return (error);
  980 }
  981 
  982 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
  983     NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
  984 #endif /* MALLOC_PROFILE */

Cache object: 58f635da988e10ba52debd8779251ce7


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.