bsd/dev/unix_startup.c

   1 /*
   2  * Copyright (c) 2000-2008 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /*
  29  * Copyright (c) 1992,7 NeXT Computer, Inc.
  30  *
  31  * Unix data structure initialization.
  32  *
  33  */
  34
  35 #include <mach/mach_types.h>
  36
  37 #include <vm/vm_kern.h>
  38 #include <mach/vm_prot.h>
  39
  40 #include <sys/param.h>
  41 #include <sys/buf_internal.h>
  42 #include <sys/file_internal.h>
  43 #include <sys/proc_internal.h>
  44 #include <sys/clist.h>
  45 #include <sys/mbuf.h>
  46 #include <sys/systm.h>
  47 #include <sys/tty.h>
  48 #include <sys/vnode.h>
  49 #include <sys/sysctl.h>
  50 #include <machine/cons.h>
  51 #include <pexpert/pexpert.h>
  52 #include <sys/socketvar.h>
  53
  54 extern vm_map_t mb_map;
  55
  56 #if INET || INET6
  57 extern uint32_t   tcp_sendspace;
  58 extern uint32_t   tcp_recvspace;
  59 #endif
  60
  61 void            bsd_bufferinit(void) __attribute__((section("__TEXT, initcode")));
  62 extern void     md_prepare_for_shutdown(int, int, char *);
  63
  64 unsigned int    bsd_mbuf_cluster_reserve(boolean_t *);
  65 void bsd_srv_setup(int);
  66 void bsd_exec_setup(int);
  67
  68 /*
  69  * Declare these as initialized data so we can patch them.
  70  */
  71
  72 #ifdef  NBUF
  73 int             max_nbuf_headers = NBUF;
  74 int             niobuf_headers = NBUF / 2;
  75 int             nbuf_hashelements = NBUF;
  76 int             nbuf_headers = NBUF;
  77 #else
  78 int             max_nbuf_headers = 0;
  79 int             niobuf_headers = 0;
  80 int             nbuf_hashelements = 0;
  81 int             nbuf_headers = 0;
  82 #endif
  83
  84 SYSCTL_INT (_kern, OID_AUTO, nbuf, CTLFLAG_RD, &nbuf_headers, 0, "");
  85 SYSCTL_INT (_kern, OID_AUTO, maxnbuf, CTLFLAG_RW, &max_nbuf_headers, 0, "");
  86
  87 __private_extern__ int customnbuf = 0;
  88 int             srv = 0;        /* Flag indicates a server boot when set */
  89 int             ncl = 0;
  90 static unsigned int mbuf_poolsz;
  91
  92 vm_map_t        buffer_map;
  93 vm_map_t        bufferhdr_map;
  94 static int vnodes_sized = 0;
  95
  96 extern void     bsd_startupearly(void) __attribute__((section("__TEXT, initcode")));
  97
  98 void
  99 bsd_startupearly(void)
 100 {
 101         vm_offset_t     firstaddr;
 102         vm_size_t       size;
 103         kern_return_t   ret;
 104
 105         /* clip the number of buf headers upto 16k */
 106         if (max_nbuf_headers == 0)
 107                 max_nbuf_headers = atop_kernel(sane_size / 50); /* Get 2% of ram, but no more than we can map */
 108         if ((customnbuf == 0) && (max_nbuf_headers > 16384))
 109                 max_nbuf_headers = 16384;
 110         if (max_nbuf_headers < CONFIG_MIN_NBUF)
 111                 max_nbuf_headers = CONFIG_MIN_NBUF;
 112
 113         /* clip the number of hash elements  to 200000 */
 114         if ( (customnbuf == 0 ) && nbuf_hashelements == 0) {
 115                 nbuf_hashelements = atop_kernel(sane_size / 50);
 116                 if (nbuf_hashelements > 200000)
 117                         nbuf_hashelements = 200000;
 118         } else
 119                 nbuf_hashelements = max_nbuf_headers;
 120
 121         if (niobuf_headers == 0)
 122                 niobuf_headers = max_nbuf_headers;
 123         if (niobuf_headers > 4096)
 124                 niobuf_headers = 4096;
 125         if (niobuf_headers < CONFIG_MIN_NIOBUF)
 126                 niobuf_headers = CONFIG_MIN_NIOBUF;
 127
 128         size = (max_nbuf_headers + niobuf_headers) * sizeof(struct buf);
 129         size = round_page(size);
 130
 131         ret = kmem_suballoc(kernel_map,
 132                             &firstaddr,
 133                             size,
 134                             FALSE,
 135                             VM_FLAGS_ANYWHERE,
 136                             &bufferhdr_map);
 137
 138         if (ret != KERN_SUCCESS)
 139                 panic("Failed to create bufferhdr_map");
 140
 141         ret = kernel_memory_allocate(bufferhdr_map,
 142                                      &firstaddr,
 143                                      size,
 144                                      0,
 145                                      KMA_HERE | KMA_KOBJECT);
 146
 147         if (ret != KERN_SUCCESS)
 148                 panic("Failed to allocate bufferhdr_map");
 149
 150         buf_headers = (struct buf *) firstaddr;
 151         bzero(buf_headers, size);
 152
 153 #if SOCKETS
 154         {
 155 #if CONFIG_USESOCKTHRESHOLD
 156                 static const unsigned int       maxspace = 64 * 1024;
 157 #else
 158                 static const unsigned int       maxspace = 128 * 1024;
 159 #endif
 160                 int             scale;
 161
 162                 nmbclusters = bsd_mbuf_cluster_reserve(NULL) / MCLBYTES;
 163
 164 #if INET || INET6
 165                 if ((scale = nmbclusters / NMBCLUSTERS) > 1) {
 166                         tcp_sendspace *= scale;
 167                         tcp_recvspace *= scale;
 168
 169                         if (tcp_sendspace > maxspace)
 170                                 tcp_sendspace = maxspace;
 171                         if (tcp_recvspace > maxspace)
 172                                 tcp_recvspace = maxspace;
 173                 }
 174 #endif /* INET || INET6 */
 175         }
 176 #endif /* SOCKETS */
 177
 178         if (vnodes_sized == 0) {
 179         /*
 180          * Size vnodes based on memory
 181          * Number vnodes  is (memsize/64k) + 1024
 182          * This is the calculation that is used by launchd in tiger
 183          * we are clipping the max based on 16G
 184          * ie ((16*1024*1024*1024)/(64 *1024)) + 1024 = 263168;
 185          * CONFIG_VNODES is set to 263168 for "medium" configurations (the default)
 186          * but can be smaller or larger.
 187          */
 188         desiredvnodes  = (sane_size/65536) + 1024;
 189         if (desiredvnodes > CONFIG_VNODES)
 190                 desiredvnodes = CONFIG_VNODES;
 191         }
 192 }
 193
 194 void
 195 bsd_bufferinit(void)
 196 {
 197         kern_return_t   ret;
 198
 199         /*
 200          * Note: Console device initialized in kminit() from bsd_autoconf()
 201          * prior to call to us in bsd_init().
 202          */
 203
 204         bsd_startupearly();
 205
 206 #if SOCKETS
 207         ret = kmem_suballoc(kernel_map,
 208                             (vm_offset_t *) & mbutl,
 209                             (vm_size_t) (nmbclusters * MCLBYTES),
 210                             FALSE,
 211                             VM_FLAGS_ANYWHERE,
 212                             &mb_map);
 213
 214         if (ret != KERN_SUCCESS)
 215                 panic("Failed to allocate mb_map\n");
 216 #endif /* SOCKETS */
 217
 218         /*
 219          * Set up buffers, so they can be used to read disk labels.
 220          */
 221         bufinit();
 222 }
 223
 224 /* 512 MB (K32) or 2 GB (K64) hard limit on size of the mbuf pool */
 225 #if !defined(__LP64__)
 226 #define MAX_MBUF_POOL   (512 << MBSHIFT)
 227 #else
 228 #define MAX_MBUF_POOL   (2ULL << GBSHIFT)
 229 #endif /* !__LP64__ */
 230 #define MAX_NCL         (MAX_MBUF_POOL >> MCLSHIFT)
 231
 232 /*
 233  * this has been broken out into a separate routine that
 234  * can be called from the x86 early vm initialization to
 235  * determine how much lo memory to reserve on systems with
 236  * DMA hardware that can't fully address all of the physical
 237  * memory that is present.
 238  */
 239 unsigned int
 240 bsd_mbuf_cluster_reserve(boolean_t *overridden)
 241 {
 242         int mbuf_pool = 0;
 243         static boolean_t was_overridden = FALSE;
 244
 245         /* If called more than once, return the previously calculated size */
 246         if (mbuf_poolsz != 0)
 247                 goto done;
 248
 249         /*
 250          * Some of these are parsed in parse_bsd_args(), but for x86 we get
 251          * here early from i386_vm_init() and so we parse them now, in order
 252          * to correctly compute the size of the low-memory VM pool.  It is
 253          * redundant but rather harmless.
 254          */
 255         //(void) PE_parse_boot_argn("srv", &srv, sizeof (srv));
 256         (void) PE_parse_boot_argn("ncl", &ncl, sizeof (ncl));
 257         (void) PE_parse_boot_argn("mbuf_pool", &mbuf_pool, sizeof (mbuf_pool));
 258
 259         /*
 260          * Convert "mbuf_pool" from MB to # of 2KB clusters; it is
 261          * equivalent to "ncl", except that it uses different unit.
 262          */
 263         if (mbuf_pool != 0)
 264                 ncl = (mbuf_pool << MBSHIFT) >> MCLSHIFT;
 265
 266         if (sane_size > (64 * 1024 * 1024) || ncl != 0) {
 267
 268                 if (ncl || srv)
 269                         was_overridden = TRUE;
 270
 271                 if ((nmbclusters = ncl) == 0) {
 272                         /* Auto-configure the mbuf pool size */
 273                         nmbclusters = mbuf_default_ncl(srv, sane_size);
 274                 } else {
 275                         /* Make sure it's not odd in case ncl is manually set */
 276                         if (nmbclusters & 0x1)
 277                                 --nmbclusters;
 278
 279                         /* And obey the upper limit */
 280                         if (nmbclusters > MAX_NCL)
 281                                 nmbclusters = MAX_NCL;
 282                 }
 283         }
 284         mbuf_poolsz = nmbclusters << MCLSHIFT;
 285 done:
 286         if (overridden)
 287                 *overridden = was_overridden;
 288
 289         return (mbuf_poolsz);
 290 }
 291 #if defined(__LP64__)
 292 extern int tcp_tcbhashsize;
 293 extern int max_cached_sock_count;
 294 void IOSleep(int);
 295 #endif
 296
 297
 298 void
 299 bsd_srv_setup(int scale)
 300 {
 301 #if defined(__LP64__)
 302         /* if memory is more than 16G, then apply rules for processes */
 303         if (scale >  0) {
 304                 maxproc = 2500 * scale;
 305                 hard_maxproc = maxproc;
 306                 /* no fp usage */
 307                 maxprocperuid = (maxproc*3)/4;
 308                 maxfiles = (150000 * scale);
 309                 maxfilesperproc = maxfiles/2;
 310                 desiredvnodes = maxfiles;
 311                 vnodes_sized = 1;
 312                 if (scale > 4) {
 313                         /* clip them at 32G level */
 314                         somaxconn = 2048;
 315                         /* 64G or more the hash size is 32k */
 316                         if (scale > 7) {
 317                                 /* clip at 64G level */
 318                                 tcp_tcbhashsize = 16 *1024;
 319                                 max_cached_sock_count = 165000;
 320                         } else {
 321                                 tcp_tcbhashsize = 32 *1024;
 322                                 max_cached_sock_count = 60000 + ((scale-1) * 15000);
 323                         }
 324                 } else {
 325                         somaxconn = 512*scale;
 326                         tcp_tcbhashsize = 4*1024*scale;
 327                         max_cached_sock_count = 60000 + ((scale-1) * 15000);
 328                 }
 329         }
 330 #endif
 331         bsd_exec_setup(scale);
 332 }
 333