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 | CTLFLAG_KERN, &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         }
 115         if ((customnbuf == 0) && (max_nbuf_headers > 16384)) {
 116                 max_nbuf_headers = 16384;
 117         }
 118         if (max_nbuf_headers < CONFIG_MIN_NBUF) {
 119                 max_nbuf_headers = CONFIG_MIN_NBUF;
 120         }
 121
 122         /* clip the number of hash elements  to 200000 */
 123         if ((customnbuf == 0) && nbuf_hashelements == 0) {
 124                 nbuf_hashelements = atop_kernel(sane_size / 50);
 125                 if (nbuf_hashelements > 200000) {
 126                         nbuf_hashelements = 200000;
 127                 }
 128         } else {
 129                 nbuf_hashelements = max_nbuf_headers;
 130         }
 131
 132         if (niobuf_headers == 0) {
 133                 if (max_nbuf_headers < 4096) {
 134                         niobuf_headers = max_nbuf_headers;
 135                 } else {
 136                         niobuf_headers = (max_nbuf_headers / 2) + 2048;
 137                 }
 138         }
 139         if (niobuf_headers < CONFIG_MIN_NIOBUF) {
 140                 niobuf_headers = CONFIG_MIN_NIOBUF;
 141         }
 142
 143         size = (max_nbuf_headers + niobuf_headers) * sizeof(struct buf);
 144         size = round_page(size);
 145
 146         ret = kmem_suballoc(kernel_map,
 147             &firstaddr,
 148             size,
 149             FALSE,
 150             VM_FLAGS_ANYWHERE,
 151             VM_MAP_KERNEL_FLAGS_NONE,
 152             VM_KERN_MEMORY_FILE,
 153             &bufferhdr_map);
 154
 155         if (ret != KERN_SUCCESS) {
 156                 panic("Failed to create bufferhdr_map");
 157         }
 158
 159         ret = kernel_memory_allocate(bufferhdr_map,
 160             &firstaddr,
 161             size,
 162             0,
 163             KMA_HERE | KMA_KOBJECT,
 164             VM_KERN_MEMORY_FILE);
 165
 166         if (ret != KERN_SUCCESS) {
 167                 panic("Failed to allocate bufferhdr_map");
 168         }
 169
 170         buf_headers = (struct buf *) firstaddr;
 171         bzero(buf_headers, size);
 172
 173 #if SOCKETS
 174         {
 175                 static const unsigned int       maxspace = 128 * 1024;
 176                 int             scale;
 177
 178                 nmbclusters = bsd_mbuf_cluster_reserve(NULL) / MCLBYTES;
 179
 180 #if INET || INET6
 181                 if ((scale = nmbclusters / NMBCLUSTERS) > 1) {
 182                         tcp_sendspace *= scale;
 183                         tcp_recvspace *= scale;
 184
 185                         if (tcp_sendspace > maxspace) {
 186                                 tcp_sendspace = maxspace;
 187                         }
 188                         if (tcp_recvspace > maxspace) {
 189                                 tcp_recvspace = maxspace;
 190                         }
 191                 }
 192 #endif /* INET || INET6 */
 193         }
 194 #endif /* SOCKETS */
 195
 196         if (vnodes_sized == 0) {
 197                 if (!PE_get_default("kern.maxvnodes", &desiredvnodes, sizeof(desiredvnodes))) {
 198                         /*
 199                          * Size vnodes based on memory
 200                          * Number vnodes  is (memsize/64k) + 1024
 201                          * This is the calculation that is used by launchd in tiger
 202                          * we are clipping the max based on 16G
 203                          * ie ((16*1024*1024*1024)/(64 *1024)) + 1024 = 263168;
 204                          * CONFIG_VNODES is set to 263168 for "medium" configurations (the default)
 205                          * but can be smaller or larger.
 206                          */
 207                         desiredvnodes  = (sane_size / 65536) + 1024;
 208 #ifdef CONFIG_VNODES
 209                         if (desiredvnodes > CONFIG_VNODES) {
 210                                 desiredvnodes = CONFIG_VNODES;
 211                         }
 212 #endif
 213                 }
 214                 vnodes_sized = 1;
 215         }
 216 }
 217
 218 void
 219 bsd_bufferinit(void)
 220 {
 221 #if SOCKETS
 222         kern_return_t   ret;
 223 #endif
 224         /*
 225          * Note: Console device initialized in kminit() from bsd_autoconf()
 226          * prior to call to us in bsd_init().
 227          */
 228
 229         bsd_startupearly();
 230
 231 #if SOCKETS
 232         ret = kmem_suballoc(kernel_map,
 233             (vm_offset_t *) &mbutl,
 234             (vm_size_t) (nmbclusters * MCLBYTES),
 235             FALSE,
 236             VM_FLAGS_ANYWHERE,
 237             VM_MAP_KERNEL_FLAGS_NONE,
 238             VM_KERN_MEMORY_MBUF,
 239             &mb_map);
 240
 241         if (ret != KERN_SUCCESS) {
 242                 panic("Failed to allocate mb_map\n");
 243         }
 244 #endif /* SOCKETS */
 245
 246         /*
 247          * Set up buffers, so they can be used to read disk labels.
 248          */
 249         bufinit();
 250 }
 251
 252 /* 512 MB (K32) or 2 GB (K64) hard limit on size of the mbuf pool */
 253 #if !defined(__LP64__)
 254 #define MAX_MBUF_POOL   (512 << MBSHIFT)
 255 #else
 256 #define MAX_MBUF_POOL   (2ULL << GBSHIFT)
 257 #endif /* !__LP64__ */
 258 #define MAX_NCL         (MAX_MBUF_POOL >> MCLSHIFT)
 259
 260 #if SOCKETS
 261 /*
 262  * this has been broken out into a separate routine that
 263  * can be called from the x86 early vm initialization to
 264  * determine how much lo memory to reserve on systems with
 265  * DMA hardware that can't fully address all of the physical
 266  * memory that is present.
 267  */
 268 unsigned int
 269 bsd_mbuf_cluster_reserve(boolean_t *overridden)
 270 {
 271         int mbuf_pool = 0;
 272         static boolean_t was_overridden = FALSE;
 273
 274         /* If called more than once, return the previously calculated size */
 275         if (mbuf_poolsz != 0) {
 276                 goto done;
 277         }
 278
 279         /*
 280          * Some of these are parsed in parse_bsd_args(), but for x86 we get
 281          * here early from i386_vm_init() and so we parse them now, in order
 282          * to correctly compute the size of the low-memory VM pool.  It is
 283          * redundant but rather harmless.
 284          */
 285         (void) PE_parse_boot_argn("ncl", &ncl, sizeof(ncl));
 286         (void) PE_parse_boot_argn("mbuf_pool", &mbuf_pool, sizeof(mbuf_pool));
 287
 288         /*
 289          * Convert "mbuf_pool" from MB to # of 2KB clusters; it is
 290          * equivalent to "ncl", except that it uses different unit.
 291          */
 292         if (mbuf_pool != 0) {
 293                 ncl = (mbuf_pool << MBSHIFT) >> MCLSHIFT;
 294         }
 295
 296         if (sane_size > (64 * 1024 * 1024) || ncl != 0) {
 297                 if (ncl || serverperfmode) {
 298                         was_overridden = TRUE;
 299                 }
 300
 301                 if ((nmbclusters = ncl) == 0) {
 302                         /* Auto-configure the mbuf pool size */
 303                         nmbclusters = mbuf_default_ncl(serverperfmode, sane_size);
 304                 } else {
 305                         /* Make sure it's not odd in case ncl is manually set */
 306                         if (nmbclusters & 0x1) {
 307                                 --nmbclusters;
 308                         }
 309
 310                         /* And obey the upper limit */
 311                         if (nmbclusters > MAX_NCL) {
 312                                 nmbclusters = MAX_NCL;
 313                         }
 314                 }
 315
 316                 /* Round it down to nearest multiple of PAGE_SIZE */
 317                 nmbclusters = P2ROUNDDOWN(nmbclusters, NCLPG);
 318         }
 319         mbuf_poolsz = nmbclusters << MCLSHIFT;
 320 done:
 321         if (overridden) {
 322                 *overridden = was_overridden;
 323         }
 324
 325         return mbuf_poolsz;
 326 }
 327 #endif
 328
 329 #if defined(__LP64__)
 330 extern int tcp_tcbhashsize;
 331 extern int max_cached_sock_count;
 332 #endif
 333
 334
 335 void
 336 bsd_scale_setup(int scale)
 337 {
 338 #if defined(__LP64__)
 339         if ((scale > 0) && (serverperfmode == 0)) {
 340                 maxproc *= scale;
 341                 maxprocperuid = (maxproc * 2) / 3;
 342                 if (scale > 2) {
 343                         maxfiles *= scale;
 344                         maxfilesperproc = maxfiles / 2;
 345                 }
 346         }
 347         /* Apply server scaling rules */
 348         if ((scale > 0) && (serverperfmode != 0)) {
 349                 maxproc = 2500 * scale;
 350                 hard_maxproc = maxproc;
 351                 /* no fp usage */
 352                 maxprocperuid = (maxproc * 3) / 4;
 353                 maxfiles = (150000 * scale);
 354                 maxfilesperproc = maxfiles / 2;
 355                 desiredvnodes = maxfiles;
 356                 vnodes_sized = 1;
 357                 tcp_tfo_backlog = 100 * scale;
 358                 if (scale > 4) {
 359                         /* clip somaxconn at 32G level */
 360                         somaxconn = 2048;
 361                         /*
 362                          * For scale > 4 (> 32G), clip
 363                          * tcp_tcbhashsize to 32K
 364                          */
 365                         tcp_tcbhashsize = 32 * 1024;
 366
 367                         if (scale > 7) {
 368                                 /* clip at 64G level */
 369                                 max_cached_sock_count = 165000;
 370                         } else {
 371                                 max_cached_sock_count = 60000 + ((scale - 1) * 15000);
 372                         }
 373                 } else {
 374                         somaxconn = 512 * scale;
 375                         tcp_tcbhashsize = 4 * 1024 * scale;
 376                         max_cached_sock_count = 60000 + ((scale - 1) * 15000);
 377                 }
 378         }
 379
 380         if (maxproc > hard_maxproc) {
 381                 hard_maxproc = maxproc;
 382         }
 383 #endif
 384         bsd_exec_setup(scale);
 385 }