bsd/dev/unix_startup.c

   1 /*
   2  * Copyright (c) 2000-2010 Apple 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/mcache.h>
  46 #include <sys/mbuf.h>
  47 #include <sys/systm.h>
  48 #include <sys/tty.h>
  49 #include <sys/vnode.h>
  50 #include <sys/sysctl.h>
  51 #include <machine/cons.h>
  52 #include <pexpert/pexpert.h>
  53 #include <sys/socketvar.h>
  54 #include <pexpert/pexpert.h>
  55
  56 extern uint32_t kern_maxvnodes;
  57 extern vm_map_t mb_map;
  58
  59 #if INET || INET6
  60 extern uint32_t   tcp_sendspace;
  61 extern uint32_t   tcp_recvspace;
  62 #endif
  63
  64 void            bsd_bufferinit(void) __attribute__((section("__TEXT, initcode")));
  65 extern void     md_prepare_for_shutdown(int, int, char *);
  66
  67 unsigned int    bsd_mbuf_cluster_reserve(boolean_t *);
  68 void bsd_scale_setup(int);
  69 void bsd_exec_setup(int);
  70
  71 /*
  72  * Declare these as initialized data so we can patch them.
  73  */
  74
  75 #ifdef  NBUF
  76 int             max_nbuf_headers = NBUF;
  77 int             niobuf_headers = (NBUF / 2) + 2048;
  78 int             nbuf_hashelements = NBUF;
  79 int             nbuf_headers = NBUF;
  80 #else
  81 int             max_nbuf_headers = 0;
  82 int             niobuf_headers = 0;
  83 int             nbuf_hashelements = 0;
  84 int             nbuf_headers = 0;
  85 #endif
  86
  87 SYSCTL_INT (_kern, OID_AUTO, nbuf, CTLFLAG_RD | CTLFLAG_LOCKED, &nbuf_headers, 0, "");
  88 SYSCTL_INT (_kern, OID_AUTO, maxnbuf, CTLFLAG_RW | CTLFLAG_LOCKED, &max_nbuf_headers, 0, "");
  89
  90 __private_extern__ int customnbuf = 0;
  91 int             serverperfmode = 0;     /* Flag indicates a server boot when set */
  92 int             ncl = 0;
  93 static unsigned int mbuf_poolsz;
  94
  95 vm_map_t        buffer_map;
  96 vm_map_t        bufferhdr_map;
  97 static int vnodes_sized = 0;
  98
  99 extern void     bsd_startupearly(void) __attribute__((section("__TEXT, initcode")));
 100
 101 void
 102 bsd_startupearly(void)
 103 {
 104         vm_offset_t     firstaddr;
 105         vm_size_t       size;
 106         kern_return_t   ret;
 107
 108         /* clip the number of buf headers upto 16k */
 109         if (max_nbuf_headers == 0)
 110                 max_nbuf_headers = atop_kernel(sane_size / 50); /* Get 2% of ram, but no more than we can map */
 111         if ((customnbuf == 0) && (max_nbuf_headers > 16384))
 112                 max_nbuf_headers = 16384;
 113         if (max_nbuf_headers < CONFIG_MIN_NBUF)
 114                 max_nbuf_headers = CONFIG_MIN_NBUF;
 115
 116         /* clip the number of hash elements  to 200000 */
 117         if ( (customnbuf == 0 ) && nbuf_hashelements == 0) {
 118                 nbuf_hashelements = atop_kernel(sane_size / 50);
 119                 if (nbuf_hashelements > 200000)
 120                         nbuf_hashelements = 200000;
 121         } else
 122                 nbuf_hashelements = max_nbuf_headers;
 123
 124         if (niobuf_headers == 0) {
 125                 if (max_nbuf_headers < 4096)
 126                         niobuf_headers = max_nbuf_headers;
 127                 else
 128                         niobuf_headers = (max_nbuf_headers / 2) + 2048;
 129         }
 130         if (niobuf_headers < CONFIG_MIN_NIOBUF)
 131                 niobuf_headers = CONFIG_MIN_NIOBUF;
 132
 133         size = (max_nbuf_headers + niobuf_headers) * sizeof(struct buf);
 134         size = round_page(size);
 135
 136         ret = kmem_suballoc(kernel_map,
 137                             &firstaddr,
 138                             size,
 139                             FALSE,
 140                             VM_FLAGS_ANYWHERE,
 141                             &bufferhdr_map);
 142
 143         if (ret != KERN_SUCCESS)
 144                 panic("Failed to create bufferhdr_map");
 145
 146         ret = kernel_memory_allocate(bufferhdr_map,
 147                                      &firstaddr,
 148                                      size,
 149                                      0,
 150                                      KMA_HERE | KMA_KOBJECT);
 151
 152         if (ret != KERN_SUCCESS)
 153                 panic("Failed to allocate bufferhdr_map");
 154
 155         buf_headers = (struct buf *) firstaddr;
 156         bzero(buf_headers, size);
 157
 158 #if SOCKETS
 159         {
 160 #if CONFIG_USESOCKTHRESHOLD
 161                 static const unsigned int       maxspace = 64 * 1024;
 162 #else
 163                 static const unsigned int       maxspace = 128 * 1024;
 164 #endif
 165                 int             scale;
 166
 167                 nmbclusters = bsd_mbuf_cluster_reserve(NULL) / MCLBYTES;
 168
 169 #if INET || INET6
 170                 if ((scale = nmbclusters / NMBCLUSTERS) > 1) {
 171                         tcp_sendspace *= scale;
 172                         tcp_recvspace *= scale;
 173
 174                         if (tcp_sendspace > maxspace)
 175                                 tcp_sendspace = maxspace;
 176                         if (tcp_recvspace > maxspace)
 177                                 tcp_recvspace = maxspace;
 178                 }
 179 #endif /* INET || INET6 */
 180         }
 181 #endif /* SOCKETS */
 182
 183         if (vnodes_sized == 0) {
 184                 if (!PE_get_default("kern.maxvnodes", &desiredvnodes, sizeof(desiredvnodes))) {
 185                         /*
 186                          * Size vnodes based on memory
 187                          * Number vnodes  is (memsize/64k) + 1024
 188                          * This is the calculation that is used by launchd in tiger
 189                          * we are clipping the max based on 16G
 190                          * ie ((16*1024*1024*1024)/(64 *1024)) + 1024 = 263168;
 191                          * CONFIG_VNODES is set to 263168 for "medium" configurations (the default)
 192                          * but can be smaller or larger.
 193                          */
 194                         desiredvnodes  = (sane_size/65536) + 1024;
 195 #ifdef CONFIG_VNODES
 196                                 if (desiredvnodes > CONFIG_VNODES)
 197                                         desiredvnodes = CONFIG_VNODES;
 198 #endif
 199                 }
 200                 vnodes_sized = 1;
 201         }
 202 }
 203
 204 void
 205 bsd_bufferinit(void)
 206 {
 207         kern_return_t   ret;
 208
 209         /*
 210          * Note: Console device initialized in kminit() from bsd_autoconf()
 211          * prior to call to us in bsd_init().
 212          */
 213
 214         bsd_startupearly();
 215
 216 #if SOCKETS
 217         ret = kmem_suballoc(kernel_map,
 218                             (vm_offset_t *) & mbutl,
 219                             (vm_size_t) (nmbclusters * MCLBYTES),
 220                             FALSE,
 221                             VM_FLAGS_ANYWHERE,
 222                             &mb_map);
 223
 224         if (ret != KERN_SUCCESS)
 225                 panic("Failed to allocate mb_map\n");
 226 #endif /* SOCKETS */
 227
 228         /*
 229          * Set up buffers, so they can be used to read disk labels.
 230          */
 231         bufinit();
 232 }
 233
 234 /* 512 MB (K32) or 2 GB (K64) hard limit on size of the mbuf pool */
 235 #if !defined(__LP64__)
 236 #define MAX_MBUF_POOL   (512 << MBSHIFT)
 237 #else
 238 #define MAX_MBUF_POOL   (2ULL << GBSHIFT)
 239 #endif /* !__LP64__ */
 240 #define MAX_NCL         (MAX_MBUF_POOL >> MCLSHIFT)
 241
 242 /*
 243  * this has been broken out into a separate routine that
 244  * can be called from the x86 early vm initialization to
 245  * determine how much lo memory to reserve on systems with
 246  * DMA hardware that can't fully address all of the physical
 247  * memory that is present.
 248  */
 249 unsigned int
 250 bsd_mbuf_cluster_reserve(boolean_t *overridden)
 251 {
 252         int mbuf_pool = 0;
 253         static boolean_t was_overridden = FALSE;
 254
 255         /* If called more than once, return the previously calculated size */
 256         if (mbuf_poolsz != 0)
 257                 goto done;
 258
 259         /*
 260          * Some of these are parsed in parse_bsd_args(), but for x86 we get
 261          * here early from i386_vm_init() and so we parse them now, in order
 262          * to correctly compute the size of the low-memory VM pool.  It is
 263          * redundant but rather harmless.
 264          */
 265         (void) PE_parse_boot_argn("ncl", &ncl, sizeof (ncl));
 266         (void) PE_parse_boot_argn("mbuf_pool", &mbuf_pool, sizeof (mbuf_pool));
 267
 268         /*
 269          * Convert "mbuf_pool" from MB to # of 2KB clusters; it is
 270          * equivalent to "ncl", except that it uses different unit.
 271          */
 272         if (mbuf_pool != 0)
 273                 ncl = (mbuf_pool << MBSHIFT) >> MCLSHIFT;
 274
 275         if (sane_size > (64 * 1024 * 1024) || ncl != 0) {
 276
 277                 if (ncl || serverperfmode)
 278                         was_overridden = TRUE;
 279
 280                 if ((nmbclusters = ncl) == 0) {
 281                         /* Auto-configure the mbuf pool size */
 282                         nmbclusters = mbuf_default_ncl(serverperfmode, sane_size);
 283                 } else {
 284                         /* Make sure it's not odd in case ncl is manually set */
 285                         if (nmbclusters & 0x1)
 286                                 --nmbclusters;
 287
 288                         /* And obey the upper limit */
 289                         if (nmbclusters > MAX_NCL)
 290                                 nmbclusters = MAX_NCL;
 291                 }
 292
 293                 /* Round it down to nearest multiple of 4KB clusters */
 294                 nmbclusters = P2ROUNDDOWN(nmbclusters, NCLPBG);
 295         }
 296         mbuf_poolsz = nmbclusters << MCLSHIFT;
 297 done:
 298         if (overridden)
 299                 *overridden = was_overridden;
 300
 301         return (mbuf_poolsz);
 302 }
 303 #if defined(__LP64__)
 304 extern int tcp_tcbhashsize;
 305 extern int max_cached_sock_count;
 306 void IOSleep(int);
 307 #endif
 308
 309
 310 void
 311 bsd_scale_setup(int scale)
 312 {
 313 #if defined(__LP64__)
 314         if ((scale > 0) && (serverperfmode == 0)) {
 315                 maxproc *= scale;
 316                 maxprocperuid = (maxproc * 2) / 3;
 317         }
 318         /* Apply server scaling rules */
 319         if ((scale >  0) && (serverperfmode !=0)) {
 320                 maxproc = 2500 * scale;
 321                 hard_maxproc = maxproc;
 322                 /* no fp usage */
 323                 maxprocperuid = (maxproc*3)/4;
 324                 maxfiles = (150000 * scale);
 325                 maxfilesperproc = maxfiles/2;
 326                 desiredvnodes = maxfiles;
 327                 vnodes_sized = 1;
 328                 if (scale > 4) {
 329                         /* clip them at 32G level */
 330                         somaxconn = 2048;
 331                         /* 64G or more the hash size is 32k */
 332                         if (scale > 7) {
 333                                 /* clip at 64G level */
 334                                 tcp_tcbhashsize = 16 *1024;
 335                                 max_cached_sock_count = 165000;
 336                         } else {
 337                                 tcp_tcbhashsize = 32 *1024;
 338                                 max_cached_sock_count = 60000 + ((scale-1) * 15000);
 339                         }
 340                 } else {
 341                         somaxconn = 512*scale;
 342                         tcp_tcbhashsize = 4*1024*scale;
 343                         max_cached_sock_count = 60000 + ((scale-1) * 15000);
 344                 }
 345         }
 346 #endif
 347         bsd_exec_setup(scale);
 348 }
 349