xref: /freebsd/sys/vm/vm_init.c (revision ab40f58ccfe6c07ebefddc72f4661a52fe746353)
1 /*-
2  * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
3  *
4  * Copyright (c) 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * The Mach Operating System project at Carnegie-Mellon University.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from: @(#)vm_init.c	8.1 (Berkeley) 6/11/93
35  *
36  *
37  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38  * All rights reserved.
39  *
40  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41  *
42  * Permission to use, copy, modify and distribute this software and
43  * its documentation is hereby granted, provided that both the copyright
44  * notice and this permission notice appear in all copies of the
45  * software, derivative works or modified versions, and any portions
46  * thereof, and that both notices appear in supporting documentation.
47  *
48  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51  *
52  * Carnegie Mellon requests users of this software to return to
53  *
54  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
55  *  School of Computer Science
56  *  Carnegie Mellon University
57  *  Pittsburgh PA 15213-3890
58  *
59  * any improvements or extensions that they make and grant Carnegie the
60  * rights to redistribute these changes.
61  */
62 
63 /*
64  *	Initialize the Virtual Memory subsystem.
65  */
66 
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
69 
70 #include <sys/param.h>
71 #include <sys/kernel.h>
72 #include <sys/lock.h>
73 #include <sys/proc.h>
74 #include <sys/rwlock.h>
75 #include <sys/malloc.h>
76 #include <sys/sysctl.h>
77 #include <sys/systm.h>
78 #include <sys/selinfo.h>
79 #include <sys/smp.h>
80 #include <sys/pipe.h>
81 #include <sys/bio.h>
82 #include <sys/buf.h>
83 #include <sys/vmem.h>
84 #include <sys/vmmeter.h>
85 
86 #include <vm/vm.h>
87 #include <vm/vm_param.h>
88 #include <vm/vm_kern.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_phys.h>
92 #include <vm/vm_pagequeue.h>
93 #include <vm/vm_map.h>
94 #include <vm/vm_pager.h>
95 #include <vm/vm_extern.h>
96 
97 extern void	uma_startup1(void);
98 extern void	uma_startup2(void);
99 extern void	vm_radix_reserve_kva(void);
100 
101 #if VM_NRESERVLEVEL > 0
102 #define	KVA_QUANTUM	(1 << (VM_LEVEL_0_ORDER + PAGE_SHIFT))
103 #else
104 	/* On non-superpage architectures want large import sizes. */
105 #define	KVA_QUANTUM	(PAGE_SIZE * 1024)
106 #endif
107 long physmem;
108 
109 /*
110  * System initialization
111  */
112 static void vm_mem_init(void *);
113 SYSINIT(vm_mem, SI_SUB_VM, SI_ORDER_FIRST, vm_mem_init, NULL);
114 
115 /*
116  * Import kva into the kernel arena.
117  */
118 static int
119 kva_import(void *unused, vmem_size_t size, int flags, vmem_addr_t *addrp)
120 {
121 	vm_offset_t addr;
122 	int result;
123 
124 	KASSERT((size % KVA_QUANTUM) == 0,
125 	    ("kva_import: Size %jd is not a multiple of %d",
126 	    (intmax_t)size, (int)KVA_QUANTUM));
127 	addr = vm_map_min(kernel_map);
128 	result = vm_map_find(kernel_map, NULL, 0, &addr, size, 0,
129 	    VMFS_SUPER_SPACE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
130 	if (result != KERN_SUCCESS)
131                 return (ENOMEM);
132 
133 	*addrp = addr;
134 
135 	return (0);
136 }
137 
138 #if VM_NRESERVLEVEL > 0
139 /*
140  * Import a superpage from the normal kernel arena into the special
141  * arena for allocations with different permissions.
142  */
143 static int
144 kernel_rwx_alloc(void *arena, vmem_size_t size, int flags, vmem_addr_t *addrp)
145 {
146 
147 	KASSERT((size % KVA_QUANTUM) == 0,
148 	    ("kernel_rwx_alloc: Size %jd is not a multiple of %d",
149 	    (intmax_t)size, (int)KVA_QUANTUM));
150 	return (vmem_xalloc(arena, size, KVA_QUANTUM, 0, 0, VMEM_ADDR_MIN,
151 	    VMEM_ADDR_MAX, flags, addrp));
152 }
153 #endif
154 
155 /*
156  *	vm_init initializes the virtual memory system.
157  *	This is done only by the first cpu up.
158  *
159  *	The start and end address of physical memory is passed in.
160  */
161 /* ARGSUSED*/
162 static void
163 vm_mem_init(dummy)
164 	void *dummy;
165 {
166 	int domain;
167 
168 	/*
169 	 * Initializes resident memory structures. From here on, all physical
170 	 * memory is accounted for, and we use only virtual addresses.
171 	 */
172 	vm_set_page_size();
173 	virtual_avail = vm_page_startup(virtual_avail);
174 
175 #ifdef	UMA_MD_SMALL_ALLOC
176 	/* Announce page availability to UMA. */
177 	uma_startup1();
178 #endif
179 	/*
180 	 * Initialize other VM packages
181 	 */
182 	vmem_startup();
183 	vm_object_init();
184 	vm_map_startup();
185 	kmem_init(virtual_avail, virtual_end);
186 
187 	/*
188 	 * Initialize the kernel_arena.  This can grow on demand.
189 	 */
190 	vmem_init(kernel_arena, "kernel arena", 0, 0, PAGE_SIZE, 0, 0);
191 	vmem_set_import(kernel_arena, kva_import, NULL, NULL, KVA_QUANTUM);
192 
193 #if VM_NRESERVLEVEL > 0
194 	/*
195 	 * In an architecture with superpages, maintain a separate arena
196 	 * for allocations with permissions that differ from the "standard"
197 	 * read/write permissions used for memory in the kernel_arena.
198 	 */
199 	kernel_rwx_arena = vmem_create("kernel rwx arena", 0, 0, PAGE_SIZE,
200 	    0, M_WAITOK);
201 	vmem_set_import(kernel_rwx_arena, kernel_rwx_alloc,
202 	    (vmem_release_t *)vmem_xfree, kernel_arena, KVA_QUANTUM);
203 #endif
204 
205 	for (domain = 0; domain < vm_ndomains; domain++) {
206 		vm_dom[domain].vmd_kernel_arena = vmem_create(
207 		    "kernel arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
208 		vmem_set_import(vm_dom[domain].vmd_kernel_arena,
209 		    (vmem_import_t *)vmem_alloc, NULL, kernel_arena,
210 		    KVA_QUANTUM);
211 #if VM_NRESERVLEVEL > 0
212 		vm_dom[domain].vmd_kernel_rwx_arena = vmem_create(
213 		    "kernel rwx arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
214 		vmem_set_import(vm_dom[domain].vmd_kernel_rwx_arena,
215 		    kernel_rwx_alloc, (vmem_release_t *)vmem_xfree,
216 		    vm_dom[domain].vmd_kernel_arena, KVA_QUANTUM);
217 #endif
218 	}
219 
220 #ifndef	UMA_MD_SMALL_ALLOC
221 	/* Set up radix zone to use noobj_alloc. */
222 	vm_radix_reserve_kva();
223 #endif
224 	/* Announce full page availability to UMA. */
225 	uma_startup2();
226 	kmem_init_zero_region();
227 	pmap_init();
228 	vm_pager_init();
229 }
230 
231 void
232 vm_ksubmap_init(struct kva_md_info *kmi)
233 {
234 	vm_offset_t firstaddr;
235 	caddr_t v;
236 	vm_size_t size = 0;
237 	long physmem_est;
238 	vm_offset_t minaddr;
239 	vm_offset_t maxaddr;
240 
241 	/*
242 	 * Allocate space for system data structures.
243 	 * The first available kernel virtual address is in "v".
244 	 * As pages of kernel virtual memory are allocated, "v" is incremented.
245 	 * As pages of memory are allocated and cleared,
246 	 * "firstaddr" is incremented.
247 	 */
248 
249 	/*
250 	 * Make two passes.  The first pass calculates how much memory is
251 	 * needed and allocates it.  The second pass assigns virtual
252 	 * addresses to the various data structures.
253 	 */
254 	firstaddr = 0;
255 again:
256 	v = (caddr_t)firstaddr;
257 
258 	/*
259 	 * Discount the physical memory larger than the size of kernel_map
260 	 * to avoid eating up all of KVA space.
261 	 */
262 	physmem_est = lmin(physmem, btoc(kernel_map->max_offset -
263 	    kernel_map->min_offset));
264 
265 	v = kern_vfs_bio_buffer_alloc(v, physmem_est);
266 
267 	/*
268 	 * End of first pass, size has been calculated so allocate memory
269 	 */
270 	if (firstaddr == 0) {
271 		size = (vm_size_t)v;
272 #ifdef VM_FREELIST_DMA32
273 		/*
274 		 * Try to protect 32-bit DMAable memory from the largest
275 		 * early alloc of wired mem.
276 		 */
277 		firstaddr = kmem_alloc_attr(kernel_arena, size,
278 		    M_ZERO | M_NOWAIT, (vm_paddr_t)1 << 32,
279 		    ~(vm_paddr_t)0, VM_MEMATTR_DEFAULT);
280 		if (firstaddr == 0)
281 #endif
282 			firstaddr = kmem_malloc(kernel_arena, size,
283 			    M_ZERO | M_WAITOK);
284 		if (firstaddr == 0)
285 			panic("startup: no room for tables");
286 		goto again;
287 	}
288 
289 	/*
290 	 * End of second pass, addresses have been assigned
291 	 */
292 	if ((vm_size_t)((char *)v - firstaddr) != size)
293 		panic("startup: table size inconsistency");
294 
295 	/*
296 	 * Allocate the clean map to hold all of the paging and I/O virtual
297 	 * memory.
298 	 */
299 	size = (long)nbuf * BKVASIZE + (long)nswbuf * MAXPHYS +
300 	    (long)bio_transient_maxcnt * MAXPHYS;
301 	kmi->clean_sva = firstaddr = kva_alloc(size);
302 	kmi->clean_eva = firstaddr + size;
303 
304 	/*
305 	 * Allocate the buffer arena.
306 	 *
307 	 * Enable the quantum cache if we have more than 4 cpus.  This
308 	 * avoids lock contention at the expense of some fragmentation.
309 	 */
310 	size = (long)nbuf * BKVASIZE;
311 	kmi->buffer_sva = firstaddr;
312 	kmi->buffer_eva = kmi->buffer_sva + size;
313 	vmem_init(buffer_arena, "buffer arena", kmi->buffer_sva, size,
314 	    PAGE_SIZE, (mp_ncpus > 4) ? BKVASIZE * 8 : 0, 0);
315 	firstaddr += size;
316 
317 	/*
318 	 * Now swap kva.
319 	 */
320 	swapbkva = firstaddr;
321 	size = (long)nswbuf * MAXPHYS;
322 	firstaddr += size;
323 
324 	/*
325 	 * And optionally transient bio space.
326 	 */
327 	if (bio_transient_maxcnt != 0) {
328 		size = (long)bio_transient_maxcnt * MAXPHYS;
329 		vmem_init(transient_arena, "transient arena",
330 		    firstaddr, size, PAGE_SIZE, 0, 0);
331 		firstaddr += size;
332 	}
333 	if (firstaddr != kmi->clean_eva)
334 		panic("Clean map calculation incorrect");
335 
336 	/*
337 	 * Allocate the pageable submaps.  We may cache an exec map entry per
338 	 * CPU, so we therefore need to reserve space for at least ncpu+1
339 	 * entries to avoid deadlock.  The exec map is also used by some image
340 	 * activators, so we leave a fixed number of pages for their use.
341 	 */
342 #ifdef __LP64__
343 	exec_map_entries = 8 * mp_ncpus;
344 #else
345 	exec_map_entries = 2 * mp_ncpus + 4;
346 #endif
347 	exec_map_entry_size = round_page(PATH_MAX + ARG_MAX);
348 	exec_map = kmem_suballoc(kernel_map, &minaddr, &maxaddr,
349 	    exec_map_entries * exec_map_entry_size + 64 * PAGE_SIZE, FALSE);
350 	pipe_map = kmem_suballoc(kernel_map, &minaddr, &maxaddr, maxpipekva,
351 	    FALSE);
352 }
353