]> git.saurik.com Git - apple/xnu.git/blame - bsd/dev/unix_startup.c
xnu-7195.81.3.tar.gz
[apple/xnu.git] / bsd / dev / unix_startup.c
CommitLineData
1c79356b 1/*
fe8ab488 2 * Copyright (c) 2000-2014 Apple Inc. All rights reserved.
5d5c5d0d 3 *
2d21ac55 4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
0a7de745 5 *
2d21ac55
A
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.
0a7de745 14 *
2d21ac55
A
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
0a7de745 17 *
2d21ac55
A
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
8f6c56a5
A
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
2d21ac55
A
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.
0a7de745 25 *
2d21ac55 26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
1c79356b
A
27 */
28/*
29 * Copyright (c) 1992,7 NeXT Computer, Inc.
30 *
31 * Unix data structure initialization.
91447636 32 *
1c79356b
A
33 */
34
35#include <mach/mach_types.h>
36
f427ee49 37#include <kern/startup.h>
1c79356b
A
38#include <vm/vm_kern.h>
39#include <mach/vm_prot.h>
40
41#include <sys/param.h>
91447636 42#include <sys/buf_internal.h>
b0d623f7
A
43#include <sys/file_internal.h>
44#include <sys/proc_internal.h>
6d2010ae 45#include <sys/mcache.h>
1c79356b
A
46#include <sys/mbuf.h>
47#include <sys/systm.h>
48#include <sys/tty.h>
0c530ab8
A
49#include <sys/vnode.h>
50#include <sys/sysctl.h>
b0d623f7 51#include <machine/cons.h>
c910b4d9 52#include <pexpert/pexpert.h>
b0d623f7 53#include <sys/socketvar.h>
6d2010ae 54#include <pexpert/pexpert.h>
3e170ce0 55#include <netinet/tcp_var.h>
1c79356b 56
6d2010ae 57extern uint32_t kern_maxvnodes;
91447636
A
58extern vm_map_t mb_map;
59
f427ee49 60#if INET
b0d623f7
A
61extern uint32_t tcp_sendspace;
62extern uint32_t tcp_recvspace;
2d21ac55 63#endif
91447636 64
39236c6e 65void bsd_bufferinit(void);
1c79356b 66
0a7de745 67unsigned int bsd_mbuf_cluster_reserve(boolean_t *);
6d2010ae 68void bsd_scale_setup(int);
b0d623f7 69void bsd_exec_setup(int);
0c530ab8 70
1c79356b
A
71/*
72 * Declare these as initialized data so we can patch them.
73 */
74
0a7de745 75#ifdef NBUF
0c530ab8 76int max_nbuf_headers = NBUF;
6d2010ae 77int niobuf_headers = (NBUF / 2) + 2048;
0a7de745
A
78int nbuf_hashelements = NBUF;
79int nbuf_headers = NBUF;
1c79356b 80#else
0c530ab8 81int max_nbuf_headers = 0;
2d21ac55 82int niobuf_headers = 0;
0a7de745
A
83int nbuf_hashelements = 0;
84int nbuf_headers = 0;
6601e61a 85#endif
4452a7af 86
0a7de745
A
87SYSCTL_INT(_kern, OID_AUTO, nbuf, CTLFLAG_RD | CTLFLAG_LOCKED, &nbuf_headers, 0, "");
88SYSCTL_INT(_kern, OID_AUTO, maxnbuf, CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_KERN, &max_nbuf_headers, 0, "");
0c530ab8
A
89
90__private_extern__ int customnbuf = 0;
91447636 91int ncl = 0;
39236c6e 92
f427ee49
A
93/* Indicates a server boot when set */
94TUNABLE(int, serverperfmode, "serverperfmode", 0);
95
39236c6e 96#if SOCKETS
c910b4d9 97static unsigned int mbuf_poolsz;
39236c6e 98#endif
91447636
A
99
100vm_map_t buffer_map;
101vm_map_t bufferhdr_map;
b0d623f7 102static int vnodes_sized = 0;
91447636 103
39236c6e 104extern void bsd_startupearly(void);
1c79356b
A
105
106void
91447636 107bsd_startupearly(void)
1c79356b 108{
91447636
A
109 vm_offset_t firstaddr;
110 vm_size_t size;
111 kern_return_t ret;
1c79356b 112
0c530ab8 113 /* clip the number of buf headers upto 16k */
0a7de745 114 if (max_nbuf_headers == 0) {
f427ee49 115 max_nbuf_headers = (int)atop_kernel(sane_size / 50); /* Get 2% of ram, but no more than we can map */
0a7de745 116 }
f427ee49 117 if ((customnbuf == 0) && ((unsigned int)max_nbuf_headers > 16384)) {
0c530ab8 118 max_nbuf_headers = 16384;
0a7de745
A
119 }
120 if (max_nbuf_headers < CONFIG_MIN_NBUF) {
2d21ac55 121 max_nbuf_headers = CONFIG_MIN_NBUF;
0a7de745 122 }
0c530ab8
A
123
124 /* clip the number of hash elements to 200000 */
0a7de745 125 if ((customnbuf == 0) && nbuf_hashelements == 0) {
f427ee49
A
126 nbuf_hashelements = (int)atop_kernel(sane_size / 50);
127 if ((unsigned int)nbuf_hashelements > 200000) {
0c530ab8 128 nbuf_hashelements = 200000;
0a7de745
A
129 }
130 } else {
0c530ab8 131 nbuf_hashelements = max_nbuf_headers;
0a7de745 132 }
1c79356b 133
6d2010ae 134 if (niobuf_headers == 0) {
0a7de745 135 if (max_nbuf_headers < 4096) {
6d2010ae 136 niobuf_headers = max_nbuf_headers;
0a7de745 137 } else {
6d2010ae 138 niobuf_headers = (max_nbuf_headers / 2) + 2048;
0a7de745 139 }
6d2010ae 140 }
0a7de745 141 if (niobuf_headers < CONFIG_MIN_NIOBUF) {
2d21ac55 142 niobuf_headers = CONFIG_MIN_NIOBUF;
0a7de745 143 }
1c79356b 144
2d21ac55 145 size = (max_nbuf_headers + niobuf_headers) * sizeof(struct buf);
91447636 146 size = round_page(size);
1c79356b
A
147
148 ret = kmem_suballoc(kernel_map,
0a7de745
A
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) {
1c79356b 158 panic("Failed to create bufferhdr_map");
0a7de745 159 }
91447636 160
1c79356b 161 ret = kernel_memory_allocate(bufferhdr_map,
0a7de745
A
162 &firstaddr,
163 size,
164 0,
165 KMA_HERE | KMA_KOBJECT,
166 VM_KERN_MEMORY_FILE);
1c79356b 167
0a7de745 168 if (ret != KERN_SUCCESS) {
1c79356b 169 panic("Failed to allocate bufferhdr_map");
0a7de745 170 }
1c79356b 171
2d21ac55
A
172 buf_headers = (struct buf *) firstaddr;
173 bzero(buf_headers, size);
1c79356b 174
2d21ac55 175#if SOCKETS
0c530ab8 176 {
0a7de745 177 static const unsigned int maxspace = 128 * 1024;
91447636 178 int scale;
1c79356b 179
0b4c1975 180 nmbclusters = bsd_mbuf_cluster_reserve(NULL) / MCLBYTES;
0c530ab8 181
f427ee49 182#if INET
1c79356b
A
183 if ((scale = nmbclusters / NMBCLUSTERS) > 1) {
184 tcp_sendspace *= scale;
185 tcp_recvspace *= scale;
186
0a7de745 187 if (tcp_sendspace > maxspace) {
593a1d5f 188 tcp_sendspace = maxspace;
0a7de745
A
189 }
190 if (tcp_recvspace > maxspace) {
593a1d5f 191 tcp_recvspace = maxspace;
0a7de745 192 }
1c79356b 193 }
f427ee49 194#endif /* INET */
1c79356b 195 }
2d21ac55 196#endif /* SOCKETS */
0c530ab8 197
b0d623f7 198 if (vnodes_sized == 0) {
6d2010ae
A
199 if (!PE_get_default("kern.maxvnodes", &desiredvnodes, sizeof(desiredvnodes))) {
200 /*
0a7de745
A
201 * Size vnodes based on memory
202 * Number vnodes is (memsize/64k) + 1024
6d2010ae 203 * This is the calculation that is used by launchd in tiger
0a7de745 204 * we are clipping the max based on 16G
6d2010ae
A
205 * ie ((16*1024*1024*1024)/(64 *1024)) + 1024 = 263168;
206 * CONFIG_VNODES is set to 263168 for "medium" configurations (the default)
0a7de745 207 * but can be smaller or larger.
6d2010ae 208 */
f427ee49 209 desiredvnodes = (int)(sane_size / 65536) + 1024;
6d2010ae 210#ifdef CONFIG_VNODES
0a7de745
A
211 if (desiredvnodes > CONFIG_VNODES) {
212 desiredvnodes = CONFIG_VNODES;
213 }
6d2010ae
A
214#endif
215 }
216 vnodes_sized = 1;
b0d623f7 217 }
1c79356b
A
218}
219
220void
91447636 221bsd_bufferinit(void)
1c79356b 222{
39236c6e 223#if SOCKETS
91447636 224 kern_return_t ret;
39236c6e 225#endif
b0d623f7
A
226 /*
227 * Note: Console device initialized in kminit() from bsd_autoconf()
228 * prior to call to us in bsd_init().
229 */
1c79356b
A
230
231 bsd_startupearly();
232
2d21ac55 233#if SOCKETS
91447636 234 ret = kmem_suballoc(kernel_map,
0a7de745
A
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) {
1c79356b 244 panic("Failed to allocate mb_map\n");
0a7de745 245 }
2d21ac55 246#endif /* SOCKETS */
0b4e3aa0 247
91447636
A
248 /*
249 * Set up buffers, so they can be used to read disk labels.
250 */
251 bufinit();
0b4e3aa0 252}
0c530ab8 253
b0d623f7
A
254/* 512 MB (K32) or 2 GB (K64) hard limit on size of the mbuf pool */
255#if !defined(__LP64__)
0a7de745 256#define MAX_MBUF_POOL (512 << MBSHIFT)
b0d623f7 257#else
0a7de745 258#define MAX_MBUF_POOL (2ULL << GBSHIFT)
b0d623f7 259#endif /* !__LP64__ */
0a7de745 260#define MAX_NCL (MAX_MBUF_POOL >> MCLSHIFT)
2d21ac55 261
39236c6e 262#if SOCKETS
0c530ab8
A
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 */
b0d623f7 270unsigned int
0b4c1975 271bsd_mbuf_cluster_reserve(boolean_t *overridden)
0c530ab8 272{
b0d623f7 273 int mbuf_pool = 0;
0b4c1975 274 static boolean_t was_overridden = FALSE;
b0d623f7 275
c910b4d9 276 /* If called more than once, return the previously calculated size */
0a7de745 277 if (mbuf_poolsz != 0) {
b0d623f7 278 goto done;
0a7de745 279 }
c910b4d9 280
b0d623f7
A
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 */
0a7de745
A
287 (void) PE_parse_boot_argn("ncl", &ncl, sizeof(ncl));
288 (void) PE_parse_boot_argn("mbuf_pool", &mbuf_pool, sizeof(mbuf_pool));
b0d623f7
A
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 */
0a7de745 294 if (mbuf_pool != 0) {
b0d623f7 295 ncl = (mbuf_pool << MBSHIFT) >> MCLSHIFT;
0a7de745 296 }
0c530ab8 297
0a7de745
A
298 if (sane_size > (64 * 1024 * 1024) || ncl != 0) {
299 if (ncl || serverperfmode) {
0b4c1975 300 was_overridden = TRUE;
0a7de745 301 }
0b4c1975 302
0a7de745 303 if ((nmbclusters = ncl) == 0) {
b0d623f7 304 /* Auto-configure the mbuf pool size */
f427ee49 305 nmbclusters = mbuf_default_ncl(mem_actual);
b0d623f7
A
306 } else {
307 /* Make sure it's not odd in case ncl is manually set */
0a7de745 308 if (nmbclusters & 0x1) {
b0d623f7 309 --nmbclusters;
0a7de745 310 }
b0d623f7
A
311
312 /* And obey the upper limit */
0a7de745 313 if (nmbclusters > MAX_NCL) {
b0d623f7 314 nmbclusters = MAX_NCL;
0a7de745 315 }
0c530ab8 316 }
6d2010ae 317
3e170ce0 318 /* Round it down to nearest multiple of PAGE_SIZE */
f427ee49 319 nmbclusters = (unsigned int)P2ROUNDDOWN(nmbclusters, NCLPG);
c910b4d9
A
320 }
321 mbuf_poolsz = nmbclusters << MCLSHIFT;
322done:
0a7de745 323 if (overridden) {
0b4c1975 324 *overridden = was_overridden;
0a7de745 325 }
0b4c1975 326
0a7de745 327 return mbuf_poolsz;
0c530ab8 328}
39236c6e
A
329#endif
330
b0d623f7
A
331#if defined(__LP64__)
332extern int tcp_tcbhashsize;
333extern int max_cached_sock_count;
0a7de745 334#endif
b0d623f7
A
335
336
337void
6d2010ae 338bsd_scale_setup(int scale)
b0d623f7
A
339{
340#if defined(__LP64__)
6d2010ae
A
341 if ((scale > 0) && (serverperfmode == 0)) {
342 maxproc *= scale;
343 maxprocperuid = (maxproc * 2) / 3;
5ba3f43e
A
344 if (scale > 2) {
345 maxfiles *= scale;
0a7de745 346 maxfilesperproc = maxfiles / 2;
5ba3f43e 347 }
6d2010ae
A
348 }
349 /* Apply server scaling rules */
0a7de745 350 if ((scale > 0) && (serverperfmode != 0)) {
b0d623f7
A
351 maxproc = 2500 * scale;
352 hard_maxproc = maxproc;
353 /* no fp usage */
0a7de745 354 maxprocperuid = (maxproc * 3) / 4;
b0d623f7 355 maxfiles = (150000 * scale);
0a7de745 356 maxfilesperproc = maxfiles / 2;
b0d623f7
A
357 desiredvnodes = maxfiles;
358 vnodes_sized = 1;
3e170ce0 359 tcp_tfo_backlog = 100 * scale;
b0d623f7 360 if (scale > 4) {
fe8ab488 361 /* clip somaxconn at 32G level */
b0d623f7 362 somaxconn = 2048;
3e170ce0
A
363 /*
364 * For scale > 4 (> 32G), clip
fe8ab488
A
365 * tcp_tcbhashsize to 32K
366 */
0a7de745 367 tcp_tcbhashsize = 32 * 1024;
3e170ce0 368
b0d623f7
A
369 if (scale > 7) {
370 /* clip at 64G level */
b0d623f7
A
371 max_cached_sock_count = 165000;
372 } else {
0a7de745 373 max_cached_sock_count = 60000 + ((scale - 1) * 15000);
b0d623f7
A
374 }
375 } else {
0a7de745
A
376 somaxconn = 512 * scale;
377 tcp_tcbhashsize = 4 * 1024 * scale;
378 max_cached_sock_count = 60000 + ((scale - 1) * 15000);
b0d623f7
A
379 }
380 }
39037602 381
0a7de745 382 if (maxproc > hard_maxproc) {
39037602
A
383 hard_maxproc = maxproc;
384 }
b0d623f7
A
385#endif
386 bsd_exec_setup(scale);
387}