bsd/dev/unix_startup.c

   1 /*
   2  * Copyright (c) 2000-2014 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 <kern/startup.h>
  38 #include <vm/vm_kern.h>
  39 #include <mach/vm_prot.h>
  40
  41 #include <sys/param.h>
  42 #include <sys/buf_internal.h>
  43 #include <sys/file_internal.h>
  44 #include <sys/proc_internal.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 #include <netinet/tcp_var.h>
  56
  57 extern uint32_t kern_maxvnodes;
  58 extern vm_map_t mb_map;
  59
  60 #if INET
  61 extern uint32_t   tcp_sendspace;
  62 extern uint32_t   tcp_recvspace;
  63 #endif
  64
  65 void            bsd_bufferinit(void);
  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 | CTLFLAG_KERN, &max_nbuf_headers, 0, "");
  89
  90 __private_extern__ int customnbuf = 0;
  91 int             ncl = 0;
  92
  93 /* Indicates a server boot when set */
  94 TUNABLE(int, serverperfmode, "serverperfmode", 0);
  95
  96 #if SOCKETS
  97 static unsigned int mbuf_poolsz;
  98 #endif
  99
 100 vm_map_t        buffer_map;
 101 vm_map_t        bufferhdr_map;
 102 static int vnodes_sized = 0;
 103
 104 extern void     bsd_startupearly(void);
 105
 106 void
 107 bsd_startupearly(void)
 108 {
 109         vm_offset_t     firstaddr;
 110         vm_size_t       size;
 111         kern_return_t   ret;
 112
 113         /* clip the number of buf headers upto 16k */
 114         if (max_nbuf_headers == 0) {
 115                 max_nbuf_headers = (int)atop_kernel(sane_size / 50); /* Get 2% of ram, but no more than we can map */
 116         }
 117         if ((customnbuf == 0) && ((unsigned int)max_nbuf_headers > 16384)) {
 118                 max_nbuf_headers = 16384;
 119         }
 120         if (max_nbuf_headers < CONFIG_MIN_NBUF) {
 121                 max_nbuf_headers = CONFIG_MIN_NBUF;
 122         }
 123
 124         /* clip the number of hash elements  to 200000 */
 125         if ((customnbuf == 0) && nbuf_hashelements == 0) {
 126                 nbuf_hashelements = (int)atop_kernel(sane_size / 50);
 127                 if ((unsigned int)nbuf_hashelements > 200000) {
 128                         nbuf_hashelements = 200000;
 129                 }
 130         } else {
 131                 nbuf_hashelements = max_nbuf_headers;
 132         }
 133
 134         if (niobuf_headers == 0) {
 135                 if (max_nbuf_headers < 4096) {
 136                         niobuf_headers = max_nbuf_headers;
 137                 } else {
 138                         niobuf_headers = (max_nbuf_headers / 2) + 2048;
 139                 }
 140         }
 141         if (niobuf_headers < CONFIG_MIN_NIOBUF) {
 142                 niobuf_headers = CONFIG_MIN_NIOBUF;
 143         }
 144
 145         size = (max_nbuf_headers + niobuf_headers) * sizeof(struct buf);
 146         size = round_page(size);
 147
 148         ret = kmem_suballoc(kernel_map,
 149             &firstaddr,
 150             size,
 151             FALSE,
 152             VM_FLAGS_ANYWHERE,
 153             VM_MAP_KERNEL_FLAGS_NONE,
 154             VM_KERN_MEMORY_FILE,
 155             &bufferhdr_map);
 156
 157         if (ret != KERN_SUCCESS) {
 158                 panic("Failed to create bufferhdr_map");
 159         }
 160
 161         ret = kernel_memory_allocate(bufferhdr_map,
 162             &firstaddr,
 163             size,
 164             0,
 165             KMA_HERE | KMA_KOBJECT,
 166             VM_KERN_MEMORY_FILE);
 167
 168         if (ret != KERN_SUCCESS) {
 169                 panic("Failed to allocate bufferhdr_map");
 170         }
 171
 172         buf_headers = (struct buf *) firstaddr;
 173         bzero(buf_headers, size);
 174
 175 #if SOCKETS
 176         {
 177                 static const unsigned int       maxspace = 128 * 1024;
 178                 int             scale;
 179
 180                 nmbclusters = bsd_mbuf_cluster_reserve(NULL) / MCLBYTES;
 181
 182 #if INET
 183                 if ((scale = nmbclusters / NMBCLUSTERS) > 1) {
 184                         tcp_sendspace *= scale;
 185                         tcp_recvspace *= scale;
 186
 187                         if (tcp_sendspace > maxspace) {
 188                                 tcp_sendspace = maxspace;
 189                         }
 190                         if (tcp_recvspace > maxspace) {
 191                                 tcp_recvspace = maxspace;
 192                         }
 193                 }
 194 #endif /* INET */
 195         }
 196 #endif /* SOCKETS */
 197
 198         if (vnodes_sized == 0) {
 199                 if (!PE_get_default("kern.maxvnodes", &desiredvnodes, sizeof(desiredvnodes))) {
 200                         /*
 201                          * Size vnodes based on memory
 202                          * Number vnodes  is (memsize/64k) + 1024
 203                          * This is the calculation that is used by launchd in tiger
 204                          * we are clipping the max based on 16G
 205                          * ie ((16*1024*1024*1024)/(64 *1024)) + 1024 = 263168;
 206                          * CONFIG_VNODES is set to 263168 for "medium" configurations (the default)
 207                          * but can be smaller or larger.
 208                          */
 209                         desiredvnodes  = (int)(sane_size / 65536) + 1024;
 210 #ifdef CONFIG_VNODES
 211                         if (desiredvnodes > CONFIG_VNODES) {
 212                                 desiredvnodes = CONFIG_VNODES;
 213                         }
 214 #endif
 215                 }
 216                 vnodes_sized = 1;
 217         }
 218 }
 219
 220 void
 221 bsd_bufferinit(void)
 222 {
 223 #if SOCKETS
 224         kern_return_t   ret;
 225 #endif
 226         /*
 227          * Note: Console device initialized in kminit() from bsd_autoconf()
 228          * prior to call to us in bsd_init().
 229          */
 230
 231         bsd_startupearly();
 232
 233 #if SOCKETS
 234         ret = kmem_suballoc(kernel_map,
 235             (vm_offset_t *) &mbutl,
 236             (vm_size_t) (nmbclusters * MCLBYTES),
 237             FALSE,
 238             VM_FLAGS_ANYWHERE,
 239             VM_MAP_KERNEL_FLAGS_NONE,
 240             VM_KERN_MEMORY_MBUF,
 241             &mb_map);
 242
 243         if (ret != KERN_SUCCESS) {
 244                 panic("Failed to allocate mb_map\n");
 245         }
 246 #endif /* SOCKETS */
 247
 248         /*
 249          * Set up buffers, so they can be used to read disk labels.
 250          */
 251         bufinit();
 252 }
 253
 254 /* 512 MB (K32) or 2 GB (K64) hard limit on size of the mbuf pool */
 255 #if !defined(__LP64__)
 256 #define MAX_MBUF_POOL   (512 << MBSHIFT)
 257 #else
 258 #define MAX_MBUF_POOL   (2ULL << GBSHIFT)
 259 #endif /* !__LP64__ */
 260 #define MAX_NCL         (MAX_MBUF_POOL >> MCLSHIFT)
 261
 262 #if SOCKETS
 263 /*
 264  * this has been broken out into a separate routine that
 265  * can be called from the x86 early vm initialization to
 266  * determine how much lo memory to reserve on systems with
 267  * DMA hardware that can't fully address all of the physical
 268  * memory that is present.
 269  */
 270 unsigned int
 271 bsd_mbuf_cluster_reserve(boolean_t *overridden)
 272 {
 273         int mbuf_pool = 0;
 274         static boolean_t was_overridden = FALSE;
 275
 276         /* If called more than once, return the previously calculated size */
 277         if (mbuf_poolsz != 0) {
 278                 goto done;
 279         }
 280
 281         /*
 282          * Some of these are parsed in parse_bsd_args(), but for x86 we get
 283          * here early from i386_vm_init() and so we parse them now, in order
 284          * to correctly compute the size of the low-memory VM pool.  It is
 285          * redundant but rather harmless.
 286          */
 287         (void) PE_parse_boot_argn("ncl", &ncl, sizeof(ncl));
 288         (void) PE_parse_boot_argn("mbuf_pool", &mbuf_pool, sizeof(mbuf_pool));
 289
 290         /*
 291          * Convert "mbuf_pool" from MB to # of 2KB clusters; it is
 292          * equivalent to "ncl", except that it uses different unit.
 293          */
 294         if (mbuf_pool != 0) {
 295                 ncl = (mbuf_pool << MBSHIFT) >> MCLSHIFT;
 296         }
 297
 298         if (sane_size > (64 * 1024 * 1024) || ncl != 0) {
 299                 if (ncl || serverperfmode) {
 300                         was_overridden = TRUE;
 301                 }
 302
 303                 if ((nmbclusters = ncl) == 0) {
 304                         /* Auto-configure the mbuf pool size */
 305                         nmbclusters = mbuf_default_ncl(mem_actual);
 306                 } else {
 307                         /* Make sure it's not odd in case ncl is manually set */
 308                         if (nmbclusters & 0x1) {
 309                                 --nmbclusters;
 310                         }
 311
 312                         /* And obey the upper limit */
 313                         if (nmbclusters > MAX_NCL) {
 314                                 nmbclusters = MAX_NCL;
 315                         }
 316                 }
 317
 318                 /* Round it down to nearest multiple of PAGE_SIZE */
 319                 nmbclusters = (unsigned int)P2ROUNDDOWN(nmbclusters, NCLPG);
 320         }
 321         mbuf_poolsz = nmbclusters << MCLSHIFT;
 322 done:
 323         if (overridden) {
 324                 *overridden = was_overridden;
 325         }
 326
 327         return mbuf_poolsz;
 328 }
 329 #endif
 330
 331 #if defined(__LP64__)
 332 extern int tcp_tcbhashsize;
 333 extern int max_cached_sock_count;
 334 #endif
 335
 336
 337 void
 338 bsd_scale_setup(int scale)
 339 {
 340 #if defined(__LP64__)
 341         if ((scale > 0) && (serverperfmode == 0)) {
 342                 maxproc *= scale;
 343                 maxprocperuid = (maxproc * 2) / 3;
 344                 if (scale > 2) {
 345                         maxfiles *= scale;
 346                         maxfilesperproc = maxfiles / 2;
 347                 }
 348         }
 349         /* Apply server scaling rules */
 350         if ((scale > 0) && (serverperfmode != 0)) {
 351                 maxproc = 2500 * scale;
 352                 hard_maxproc = maxproc;
 353                 /* no fp usage */
 354                 maxprocperuid = (maxproc * 3) / 4;
 355                 maxfiles = (150000 * scale);
 356                 maxfilesperproc = maxfiles / 2;
 357                 desiredvnodes = maxfiles;
 358                 vnodes_sized = 1;
 359                 tcp_tfo_backlog = 100 * scale;
 360                 if (scale > 4) {
 361                         /* clip somaxconn at 32G level */
 362                         somaxconn = 2048;
 363                         /*
 364                          * For scale > 4 (> 32G), clip
 365                          * tcp_tcbhashsize to 32K
 366                          */
 367                         tcp_tcbhashsize = 32 * 1024;
 368
 369                         if (scale > 7) {
 370                                 /* clip at 64G level */
 371                                 max_cached_sock_count = 165000;
 372                         } else {
 373                                 max_cached_sock_count = 60000 + ((scale - 1) * 15000);
 374                         }
 375                 } else {
 376                         somaxconn = 512 * scale;
 377                         tcp_tcbhashsize = 4 * 1024 * scale;
 378                         max_cached_sock_count = 60000 + ((scale - 1) * 15000);
 379                 }
 380         }
 381
 382         if (maxproc > hard_maxproc) {
 383                 hard_maxproc = maxproc;
 384         }
 385 #endif
 386         bsd_exec_setup(scale);
 387 }