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