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 <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 #include <netinet/tcp_var.h>
  56
  57 extern uint32_t kern_maxvnodes;
  58 extern vm_map_t mb_map;
  59
  60 #if INET || INET6
  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, &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
  94 #if SOCKETS
  95 static unsigned int mbuf_poolsz;
  96 #endif
  97
  98 vm_map_t        buffer_map;
  99 vm_map_t        bufferhdr_map;
 100 static int vnodes_sized = 0;
 101
 102 extern void     bsd_startupearly(void);
 103
 104 void
 105 bsd_startupearly(void)
 106 {
 107         vm_offset_t     firstaddr;
 108         vm_size_t       size;
 109         kern_return_t   ret;
 110
 111         /* clip the number of buf headers upto 16k */
 112         if (max_nbuf_headers == 0)
 113                 max_nbuf_headers = atop_kernel(sane_size / 50); /* Get 2% of ram, but no more than we can map */
 114         if ((customnbuf == 0) && (max_nbuf_headers > 16384))
 115                 max_nbuf_headers = 16384;
 116         if (max_nbuf_headers < CONFIG_MIN_NBUF)
 117                 max_nbuf_headers = CONFIG_MIN_NBUF;
 118
 119         /* clip the number of hash elements  to 200000 */
 120         if ( (customnbuf == 0 ) && nbuf_hashelements == 0) {
 121                 nbuf_hashelements = atop_kernel(sane_size / 50);
 122                 if (nbuf_hashelements > 200000)
 123                         nbuf_hashelements = 200000;
 124         } else
 125                 nbuf_hashelements = max_nbuf_headers;
 126
 127         if (niobuf_headers == 0) {
 128                 if (max_nbuf_headers < 4096)
 129                         niobuf_headers = max_nbuf_headers;
 130                 else
 131                         niobuf_headers = (max_nbuf_headers / 2) + 2048;
 132         }
 133         if (niobuf_headers < CONFIG_MIN_NIOBUF)
 134                 niobuf_headers = CONFIG_MIN_NIOBUF;
 135
 136         size = (max_nbuf_headers + niobuf_headers) * sizeof(struct buf);
 137         size = round_page(size);
 138
 139         ret = kmem_suballoc(kernel_map,
 140                             &firstaddr,
 141                             size,
 142                             FALSE,
 143                             VM_FLAGS_ANYWHERE | VM_MAKE_TAG(VM_KERN_MEMORY_FILE),
 144                             &bufferhdr_map);
 145
 146         if (ret != KERN_SUCCESS)
 147                 panic("Failed to create bufferhdr_map");
 148
 149         ret = kernel_memory_allocate(bufferhdr_map,
 150                                      &firstaddr,
 151                                      size,
 152                                      0,
 153                                      KMA_HERE | KMA_KOBJECT,
 154                                      VM_KERN_MEMORY_FILE);
 155
 156         if (ret != KERN_SUCCESS)
 157                 panic("Failed to allocate bufferhdr_map");
 158
 159         buf_headers = (struct buf *) firstaddr;
 160         bzero(buf_headers, size);
 161
 162 #if SOCKETS
 163         {
 164                 static const unsigned int       maxspace = 128 * 1024;
 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 #if SOCKETS
 208         kern_return_t   ret;
 209 #endif
 210         /*
 211          * Note: Console device initialized in kminit() from bsd_autoconf()
 212          * prior to call to us in bsd_init().
 213          */
 214
 215         bsd_startupearly();
 216
 217 #if SOCKETS
 218         ret = kmem_suballoc(kernel_map,
 219                             (vm_offset_t *) &mbutl,
 220                             (vm_size_t) (nmbclusters * MCLBYTES),
 221                             FALSE,
 222                             VM_FLAGS_ANYWHERE | VM_MAKE_TAG(VM_KERN_MEMORY_MBUF),
 223                             &mb_map);
 224
 225         if (ret != KERN_SUCCESS)
 226                 panic("Failed to allocate mb_map\n");
 227 #endif /* SOCKETS */
 228
 229         /*
 230          * Set up buffers, so they can be used to read disk labels.
 231          */
 232         bufinit();
 233 }
 234
 235 /* 512 MB (K32) or 2 GB (K64) hard limit on size of the mbuf pool */
 236 #if !defined(__LP64__)
 237 #define MAX_MBUF_POOL   (512 << MBSHIFT)
 238 #else
 239 #define MAX_MBUF_POOL   (2ULL << GBSHIFT)
 240 #endif /* !__LP64__ */
 241 #define MAX_NCL         (MAX_MBUF_POOL >> MCLSHIFT)
 242
 243 #if SOCKETS
 244 /*
 245  * this has been broken out into a separate routine that
 246  * can be called from the x86 early vm initialization to
 247  * determine how much lo memory to reserve on systems with
 248  * DMA hardware that can't fully address all of the physical
 249  * memory that is present.
 250  */
 251 unsigned int
 252 bsd_mbuf_cluster_reserve(boolean_t *overridden)
 253 {
 254         int mbuf_pool = 0;
 255         static boolean_t was_overridden = FALSE;
 256
 257         /* If called more than once, return the previously calculated size */
 258         if (mbuf_poolsz != 0)
 259                 goto done;
 260
 261         /*
 262          * Some of these are parsed in parse_bsd_args(), but for x86 we get
 263          * here early from i386_vm_init() and so we parse them now, in order
 264          * to correctly compute the size of the low-memory VM pool.  It is
 265          * redundant but rather harmless.
 266          */
 267         (void) PE_parse_boot_argn("ncl", &ncl, sizeof (ncl));
 268         (void) PE_parse_boot_argn("mbuf_pool", &mbuf_pool, sizeof (mbuf_pool));
 269
 270         /*
 271          * Convert "mbuf_pool" from MB to # of 2KB clusters; it is
 272          * equivalent to "ncl", except that it uses different unit.
 273          */
 274         if (mbuf_pool != 0)
 275                 ncl = (mbuf_pool << MBSHIFT) >> MCLSHIFT;
 276
 277         if (sane_size > (64 * 1024 * 1024) || ncl != 0) {
 278
 279                 if (ncl || serverperfmode)
 280                         was_overridden = TRUE;
 281
 282                 if ((nmbclusters = ncl) == 0) {
 283                         /* Auto-configure the mbuf pool size */
 284                         nmbclusters = mbuf_default_ncl(serverperfmode, sane_size);
 285                 } else {
 286                         /* Make sure it's not odd in case ncl is manually set */
 287                         if (nmbclusters & 0x1)
 288                                 --nmbclusters;
 289
 290                         /* And obey the upper limit */
 291                         if (nmbclusters > MAX_NCL)
 292                                 nmbclusters = MAX_NCL;
 293                 }
 294
 295                 /* Round it down to nearest multiple of PAGE_SIZE */
 296                 nmbclusters = P2ROUNDDOWN(nmbclusters, NCLPG);
 297         }
 298         mbuf_poolsz = nmbclusters << MCLSHIFT;
 299 done:
 300         if (overridden)
 301                 *overridden = was_overridden;
 302
 303         return (mbuf_poolsz);
 304 }
 305 #endif
 306
 307 #if defined(__LP64__)
 308 extern int tcp_tcbhashsize;
 309 extern int max_cached_sock_count;
 310 #endif
 311
 312
 313 void
 314 bsd_scale_setup(int scale)
 315 {
 316 #if defined(__LP64__)
 317         if ((scale > 0) && (serverperfmode == 0)) {
 318                 maxproc *= scale;
 319                 maxprocperuid = (maxproc * 2) / 3;
 320         }
 321         /* Apply server scaling rules */
 322         if ((scale >  0) && (serverperfmode !=0)) {
 323                 maxproc = 2500 * scale;
 324                 hard_maxproc = maxproc;
 325                 /* no fp usage */
 326                 maxprocperuid = (maxproc*3)/4;
 327                 maxfiles = (150000 * scale);
 328                 maxfilesperproc = maxfiles/2;
 329                 desiredvnodes = maxfiles;
 330                 vnodes_sized = 1;
 331                 tcp_tfo_backlog = 100 * scale;
 332                 if (scale > 4) {
 333                         /* clip somaxconn at 32G level */
 334                         somaxconn = 2048;
 335                         /*
 336                          * For scale > 4 (> 32G), clip
 337                          * tcp_tcbhashsize to 32K
 338                          */
 339                         tcp_tcbhashsize = 32 *1024;
 340
 341                         if (scale > 7) {
 342                                 /* clip at 64G level */
 343                                 max_cached_sock_count = 165000;
 344                         } else {
 345                                 max_cached_sock_count = 60000 + ((scale-1) * 15000);
 346                         }
 347                 } else {
 348                         somaxconn = 512*scale;
 349                         tcp_tcbhashsize = 4*1024*scale;
 350                         max_cached_sock_count = 60000 + ((scale-1) * 15000);
 351                 }
 352         }
 353
 354         if(maxproc > hard_maxproc) {
 355                 hard_maxproc = maxproc;
 356         }
 357 #endif
 358         bsd_exec_setup(scale);
 359 }
 360