xref: /freebsd/sys/vm/vm_kern.c (revision 540da48d83fc148c1d34136da34ca8a2985a2a27)
160727d8bSWarner Losh /*-
2796df753SPedro F. Giffuni  * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
351369649SPedro F. Giffuni  *
4df8bae1dSRodney W. Grimes  * Copyright (c) 1991, 1993
5df8bae1dSRodney W. Grimes  *	The Regents of the University of California.  All rights reserved.
6df8bae1dSRodney W. Grimes  *
7df8bae1dSRodney W. Grimes  * This code is derived from software contributed to Berkeley by
8df8bae1dSRodney W. Grimes  * The Mach Operating System project at Carnegie-Mellon University.
9df8bae1dSRodney W. Grimes  *
10df8bae1dSRodney W. Grimes  * Redistribution and use in source and binary forms, with or without
11df8bae1dSRodney W. Grimes  * modification, are permitted provided that the following conditions
12df8bae1dSRodney W. Grimes  * are met:
13df8bae1dSRodney W. Grimes  * 1. Redistributions of source code must retain the above copyright
14df8bae1dSRodney W. Grimes  *    notice, this list of conditions and the following disclaimer.
15df8bae1dSRodney W. Grimes  * 2. Redistributions in binary form must reproduce the above copyright
16df8bae1dSRodney W. Grimes  *    notice, this list of conditions and the following disclaimer in the
17df8bae1dSRodney W. Grimes  *    documentation and/or other materials provided with the distribution.
18fbbd9655SWarner Losh  * 3. Neither the name of the University nor the names of its contributors
19df8bae1dSRodney W. Grimes  *    may be used to endorse or promote products derived from this software
20df8bae1dSRodney W. Grimes  *    without specific prior written permission.
21df8bae1dSRodney W. Grimes  *
22df8bae1dSRodney W. Grimes  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23df8bae1dSRodney W. Grimes  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24df8bae1dSRodney W. Grimes  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25df8bae1dSRodney W. Grimes  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26df8bae1dSRodney W. Grimes  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27df8bae1dSRodney W. Grimes  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28df8bae1dSRodney W. Grimes  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29df8bae1dSRodney W. Grimes  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30df8bae1dSRodney W. Grimes  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31df8bae1dSRodney W. Grimes  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32df8bae1dSRodney W. Grimes  * SUCH DAMAGE.
33df8bae1dSRodney W. Grimes  *
343c4dd356SDavid Greenman  *	from: @(#)vm_kern.c	8.3 (Berkeley) 1/12/94
35df8bae1dSRodney W. Grimes  *
36df8bae1dSRodney W. Grimes  *
37df8bae1dSRodney W. Grimes  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38df8bae1dSRodney W. Grimes  * All rights reserved.
39df8bae1dSRodney W. Grimes  *
40df8bae1dSRodney W. Grimes  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41df8bae1dSRodney W. Grimes  *
42df8bae1dSRodney W. Grimes  * Permission to use, copy, modify and distribute this software and
43df8bae1dSRodney W. Grimes  * its documentation is hereby granted, provided that both the copyright
44df8bae1dSRodney W. Grimes  * notice and this permission notice appear in all copies of the
45df8bae1dSRodney W. Grimes  * software, derivative works or modified versions, and any portions
46df8bae1dSRodney W. Grimes  * thereof, and that both notices appear in supporting documentation.
47df8bae1dSRodney W. Grimes  *
48df8bae1dSRodney W. Grimes  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49df8bae1dSRodney W. Grimes  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50df8bae1dSRodney W. Grimes  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51df8bae1dSRodney W. Grimes  *
52df8bae1dSRodney W. Grimes  * Carnegie Mellon requests users of this software to return to
53df8bae1dSRodney W. Grimes  *
54df8bae1dSRodney W. Grimes  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
55df8bae1dSRodney W. Grimes  *  School of Computer Science
56df8bae1dSRodney W. Grimes  *  Carnegie Mellon University
57df8bae1dSRodney W. Grimes  *  Pittsburgh PA 15213-3890
58df8bae1dSRodney W. Grimes  *
59df8bae1dSRodney W. Grimes  * any improvements or extensions that they make and grant Carnegie the
60df8bae1dSRodney W. Grimes  * rights to redistribute these changes.
61df8bae1dSRodney W. Grimes  */
62df8bae1dSRodney W. Grimes 
63df8bae1dSRodney W. Grimes /*
64df8bae1dSRodney W. Grimes  *	Kernel memory management.
65df8bae1dSRodney W. Grimes  */
66df8bae1dSRodney W. Grimes 
67874651b1SDavid E. O'Brien #include <sys/cdefs.h>
68874651b1SDavid E. O'Brien __FBSDID("$FreeBSD$");
69874651b1SDavid E. O'Brien 
707a469c8eSJeff Roberson #include "opt_vm.h"
717a469c8eSJeff Roberson 
72df8bae1dSRodney W. Grimes #include <sys/param.h>
73df8bae1dSRodney W. Grimes #include <sys/systm.h>
742b914b85SMark Johnston #include <sys/asan.h>
757a469c8eSJeff Roberson #include <sys/domainset.h>
760f2c2ce0SPawel Jakub Dawidek #include <sys/eventhandler.h>
772b914b85SMark Johnston #include <sys/kernel.h>
78fb919e4dSMark Murray #include <sys/lock.h>
79a1f6d91cSDavid Greenman #include <sys/malloc.h>
80*540da48dSMark Johnston #include <sys/msan.h>
812b914b85SMark Johnston #include <sys/proc.h>
8289f6b863SAttilio Rao #include <sys/rwlock.h>
8386f08737SRobert Watson #include <sys/sysctl.h>
845df87b21SJeff Roberson #include <sys/vmem.h>
857a469c8eSJeff Roberson #include <sys/vmmeter.h>
86df8bae1dSRodney W. Grimes 
87df8bae1dSRodney W. Grimes #include <vm/vm.h>
88efeaf95aSDavid Greenman #include <vm/vm_param.h>
897a469c8eSJeff Roberson #include <vm/vm_domainset.h>
905df87b21SJeff Roberson #include <vm/vm_kern.h>
91efeaf95aSDavid Greenman #include <vm/pmap.h>
92efeaf95aSDavid Greenman #include <vm/vm_map.h>
93efeaf95aSDavid Greenman #include <vm/vm_object.h>
94df8bae1dSRodney W. Grimes #include <vm/vm_page.h>
95df8bae1dSRodney W. Grimes #include <vm/vm_pageout.h>
96e2068d0bSJeff Roberson #include <vm/vm_pagequeue.h>
97431fb8abSMark Johnston #include <vm/vm_phys.h>
9833fff5d5SMark Johnston #include <vm/vm_radix.h>
999b4288a3SBruce Evans #include <vm/vm_extern.h>
1000f2c2ce0SPawel Jakub Dawidek #include <vm/uma.h>
101df8bae1dSRodney W. Grimes 
1027dd979dfSMark Johnston struct vm_map kernel_map_store;
1037dd979dfSMark Johnston struct vm_map exec_map_store;
1047dd979dfSMark Johnston struct vm_map pipe_map_store;
105f23b4c91SGarrett Wollman 
10689cb2a19SMatthew D Fleming const void *zero_region;
10789cb2a19SMatthew D Fleming CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
10889cb2a19SMatthew D Fleming 
109645743eaSJohn Baldwin /* NB: Used by kernel debuggers. */
110645743eaSJohn Baldwin const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS;
111645743eaSJohn Baldwin 
112ec492b13SMark Johnston u_int exec_map_entry_size;
113ec492b13SMark Johnston u_int exec_map_entries;
114ec492b13SMark Johnston 
11594bfd5b1SMarius Strobl SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
116f0188618SHans Petter Selasky     SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
11794bfd5b1SMarius Strobl 
11894bfd5b1SMarius Strobl SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
11958aa35d4SWarner Losh #if defined(__arm__)
12094bfd5b1SMarius Strobl     &vm_max_kernel_address, 0,
12194bfd5b1SMarius Strobl #else
122f0188618SHans Petter Selasky     SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
12394bfd5b1SMarius Strobl #endif
12494bfd5b1SMarius Strobl     "Max kernel address");
12594bfd5b1SMarius Strobl 
1261aed6d48SMark Johnston #if VM_NRESERVLEVEL > 0
12725ed23cfSMark Johnston #define	KVA_QUANTUM_SHIFT	(VM_LEVEL_0_ORDER + PAGE_SHIFT)
1281aed6d48SMark Johnston #else
1292b601070SMark Johnston /* On non-superpage architectures we want large import sizes. */
1302b601070SMark Johnston #define	KVA_QUANTUM_SHIFT	(8 + PAGE_SHIFT)
1311aed6d48SMark Johnston #endif
13223e875fdSMark Johnston #define	KVA_QUANTUM		(1ul << KVA_QUANTUM_SHIFT)
133aea9103eSMark Johnston #define	KVA_NUMA_IMPORT_QUANTUM	(KVA_QUANTUM * 128)
1341aed6d48SMark Johnston 
135a81c400eSJeff Roberson extern void     uma_startup2(void);
136a81c400eSJeff Roberson 
137df8bae1dSRodney W. Grimes /*
1385df87b21SJeff Roberson  *	kva_alloc:
139a839bdc8SDmitrij Tejblum  *
140b77c2bcdSAlan Cox  *	Allocate a virtual address range with no underlying object and
141b77c2bcdSAlan Cox  *	no initial mapping to physical memory.  Any mapping from this
142b77c2bcdSAlan Cox  *	range to physical memory must be explicitly created prior to
143b77c2bcdSAlan Cox  *	its use, typically with pmap_qenter().  Any attempt to create
144b77c2bcdSAlan Cox  *	a mapping on demand through vm_fault() will result in a panic.
145a839bdc8SDmitrij Tejblum  */
146a839bdc8SDmitrij Tejblum vm_offset_t
1476e309d75SEd Maste kva_alloc(vm_size_t size)
148a839bdc8SDmitrij Tejblum {
149a839bdc8SDmitrij Tejblum 	vm_offset_t addr;
150a839bdc8SDmitrij Tejblum 
151a839bdc8SDmitrij Tejblum 	size = round_page(size);
1525df87b21SJeff Roberson 	if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr))
153a839bdc8SDmitrij Tejblum 		return (0);
1545df87b21SJeff Roberson 
155a839bdc8SDmitrij Tejblum 	return (addr);
156a839bdc8SDmitrij Tejblum }
157a839bdc8SDmitrij Tejblum 
158a839bdc8SDmitrij Tejblum /*
1595df87b21SJeff Roberson  *	kva_free:
160ca596a25SJuli Mallett  *
1615df87b21SJeff Roberson  *	Release a region of kernel virtual memory allocated
1625df87b21SJeff Roberson  *	with kva_alloc, and return the physical pages
1635df87b21SJeff Roberson  *	associated with that region.
1645df87b21SJeff Roberson  *
1655df87b21SJeff Roberson  *	This routine may not block on kernel maps.
166ca596a25SJuli Mallett  */
1675df87b21SJeff Roberson void
1686e309d75SEd Maste kva_free(vm_offset_t addr, vm_size_t size)
169df8bae1dSRodney W. Grimes {
170df8bae1dSRodney W. Grimes 
171df8bae1dSRodney W. Grimes 	size = round_page(size);
1725df87b21SJeff Roberson 	vmem_free(kernel_arena, addr, size);
173df8bae1dSRodney W. Grimes }
174df8bae1dSRodney W. Grimes 
175*540da48dSMark Johnston /*
176*540da48dSMark Johnston  * Update sanitizer shadow state to reflect a new allocation.  Force inlining to
177*540da48dSMark Johnston  * help make KMSAN origin tracking more precise.
178*540da48dSMark Johnston  */
179*540da48dSMark Johnston static __always_inline void
180*540da48dSMark Johnston kmem_alloc_san(vm_offset_t addr, vm_size_t size, vm_size_t asize, int flags)
181*540da48dSMark Johnston {
182*540da48dSMark Johnston 	if ((flags & M_ZERO) == 0) {
183*540da48dSMark Johnston 		kmsan_mark((void *)addr, asize, KMSAN_STATE_UNINIT);
184*540da48dSMark Johnston 		kmsan_orig((void *)addr, asize, KMSAN_TYPE_KMEM,
185*540da48dSMark Johnston 		    KMSAN_RET_ADDR);
186*540da48dSMark Johnston 	} else {
187*540da48dSMark Johnston 		kmsan_mark((void *)addr, asize, KMSAN_STATE_INITED);
188*540da48dSMark Johnston 	}
189*540da48dSMark Johnston 	kasan_mark((void *)addr, size, asize, KASAN_KMEM_REDZONE);
190*540da48dSMark Johnston }
191*540da48dSMark Johnston 
19233655d95SMark Johnston static vm_page_t
19333655d95SMark Johnston kmem_alloc_contig_pages(vm_object_t object, vm_pindex_t pindex, int domain,
19433655d95SMark Johnston     int pflags, u_long npages, vm_paddr_t low, vm_paddr_t high,
19533655d95SMark Johnston     u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
19633655d95SMark Johnston {
19733655d95SMark Johnston 	vm_page_t m;
19833655d95SMark Johnston 	int tries;
199660344caSRyan Stone 	bool wait, reclaim;
20033655d95SMark Johnston 
20133655d95SMark Johnston 	VM_OBJECT_ASSERT_WLOCKED(object);
20233655d95SMark Johnston 
20333655d95SMark Johnston 	wait = (pflags & VM_ALLOC_WAITOK) != 0;
204660344caSRyan Stone 	reclaim = (pflags & VM_ALLOC_NORECLAIM) == 0;
20533655d95SMark Johnston 	pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
20633655d95SMark Johnston 	pflags |= VM_ALLOC_NOWAIT;
20733655d95SMark Johnston 	for (tries = wait ? 3 : 1;; tries--) {
20833655d95SMark Johnston 		m = vm_page_alloc_contig_domain(object, pindex, domain, pflags,
20933655d95SMark Johnston 		    npages, low, high, alignment, boundary, memattr);
210660344caSRyan Stone 		if (m != NULL || tries == 0 || !reclaim)
21133655d95SMark Johnston 			break;
21233655d95SMark Johnston 
21333655d95SMark Johnston 		VM_OBJECT_WUNLOCK(object);
21433655d95SMark Johnston 		if (!vm_page_reclaim_contig_domain(domain, pflags, npages,
21533655d95SMark Johnston 		    low, high, alignment, boundary) && wait)
21633655d95SMark Johnston 			vm_wait_domain(domain);
21733655d95SMark Johnston 		VM_OBJECT_WLOCK(object);
21833655d95SMark Johnston 	}
21933655d95SMark Johnston 	return (m);
22033655d95SMark Johnston }
22133655d95SMark Johnston 
222df8bae1dSRodney W. Grimes /*
2230ff0fc84SAlan Cox  *	Allocates a region from the kernel address map and physical pages
2240ff0fc84SAlan Cox  *	within the specified address range to the kernel object.  Creates a
2250ff0fc84SAlan Cox  *	wired mapping from this region to these pages, and returns the
2260ff0fc84SAlan Cox  *	region's starting virtual address.  The allocated pages are not
2270ff0fc84SAlan Cox  *	necessarily physically contiguous.  If M_ZERO is specified through the
2280ff0fc84SAlan Cox  *	given flags, then the pages are zeroed before they are mapped.
2290ff0fc84SAlan Cox  */
2309978bd99SMark Johnston static vm_offset_t
2317a469c8eSJeff Roberson kmem_alloc_attr_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
2320ff0fc84SAlan Cox     vm_paddr_t high, vm_memattr_t memattr)
2330ff0fc84SAlan Cox {
2347a469c8eSJeff Roberson 	vmem_t *vmem;
23533655d95SMark Johnston 	vm_object_t object;
236d1780e8dSKonstantin Belousov 	vm_offset_t addr, i, offset;
2370ff0fc84SAlan Cox 	vm_page_t m;
2382b914b85SMark Johnston 	vm_size_t asize;
23933655d95SMark Johnston 	int pflags;
240d22ff6e6SKonstantin Belousov 	vm_prot_t prot;
2410ff0fc84SAlan Cox 
24233655d95SMark Johnston 	object = kernel_object;
2432b914b85SMark Johnston 	asize = round_page(size);
2447a469c8eSJeff Roberson 	vmem = vm_dom[domain].vmd_kernel_arena;
2452b914b85SMark Johnston 	if (vmem_alloc(vmem, asize, M_BESTFIT | flags, &addr))
2460ff0fc84SAlan Cox 		return (0);
2470ff0fc84SAlan Cox 	offset = addr - VM_MIN_KERNEL_ADDRESS;
24891e31c3cSJeff Roberson 	pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
249d22ff6e6SKonstantin Belousov 	prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW;
25089f6b863SAttilio Rao 	VM_OBJECT_WLOCK(object);
2512b914b85SMark Johnston 	for (i = 0; i < asize; i += PAGE_SIZE) {
25233655d95SMark Johnston 		m = kmem_alloc_contig_pages(object, atop(offset + i),
2537a469c8eSJeff Roberson 		    domain, pflags, 1, low, high, PAGE_SIZE, 0, memattr);
2540ff0fc84SAlan Cox 		if (m == NULL) {
25589f6b863SAttilio Rao 			VM_OBJECT_WUNLOCK(object);
2569e829b22SAlan Cox 			kmem_unback(object, addr, i);
2572b914b85SMark Johnston 			vmem_free(vmem, addr, asize);
2580ff0fc84SAlan Cox 			return (0);
2590ff0fc84SAlan Cox 		}
260431fb8abSMark Johnston 		KASSERT(vm_page_domain(m) == domain,
2617a469c8eSJeff Roberson 		    ("kmem_alloc_attr_domain: Domain mismatch %d != %d",
262431fb8abSMark Johnston 		    vm_page_domain(m), domain));
2630ff0fc84SAlan Cox 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
2640ff0fc84SAlan Cox 			pmap_zero_page(m);
26591e31c3cSJeff Roberson 		vm_page_valid(m);
266d22ff6e6SKonstantin Belousov 		pmap_enter(kernel_pmap, addr + i, m, prot,
267d22ff6e6SKonstantin Belousov 		    prot | PMAP_ENTER_WIRED, 0);
2680ff0fc84SAlan Cox 	}
26989f6b863SAttilio Rao 	VM_OBJECT_WUNLOCK(object);
270*540da48dSMark Johnston 	kmem_alloc_san(addr, size, asize, flags);
2710ff0fc84SAlan Cox 	return (addr);
2720ff0fc84SAlan Cox }
2730ff0fc84SAlan Cox 
2747a469c8eSJeff Roberson vm_offset_t
275db7c2a48SAlan Cox kmem_alloc_attr(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high,
276db7c2a48SAlan Cox     vm_memattr_t memattr)
2777a469c8eSJeff Roberson {
2789978bd99SMark Johnston 
2799978bd99SMark Johnston 	return (kmem_alloc_attr_domainset(DOMAINSET_RR(), size, flags, low,
2809978bd99SMark Johnston 	    high, memattr));
2819978bd99SMark Johnston }
2829978bd99SMark Johnston 
2839978bd99SMark Johnston vm_offset_t
2849978bd99SMark Johnston kmem_alloc_attr_domainset(struct domainset *ds, vm_size_t size, int flags,
2859978bd99SMark Johnston     vm_paddr_t low, vm_paddr_t high, vm_memattr_t memattr)
2869978bd99SMark Johnston {
2877a469c8eSJeff Roberson 	struct vm_domainset_iter di;
2887a469c8eSJeff Roberson 	vm_offset_t addr;
2895fee468eSBryan Drewery 	int domain;
2907a469c8eSJeff Roberson 
2915fee468eSBryan Drewery 	vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
2927a469c8eSJeff Roberson 	do {
2937a469c8eSJeff Roberson 		addr = kmem_alloc_attr_domain(domain, size, flags, low, high,
2947a469c8eSJeff Roberson 		    memattr);
2957a469c8eSJeff Roberson 		if (addr != 0)
2967a469c8eSJeff Roberson 			break;
2974c29d2deSMark Johnston 	} while (vm_domainset_iter_policy(&di, &domain) == 0);
2987a469c8eSJeff Roberson 
2997a469c8eSJeff Roberson 	return (addr);
3007a469c8eSJeff Roberson }
3017a469c8eSJeff Roberson 
3020ff0fc84SAlan Cox /*
3030ff0fc84SAlan Cox  *	Allocates a region from the kernel address map and physically
3040ff0fc84SAlan Cox  *	contiguous pages within the specified address range to the kernel
3050ff0fc84SAlan Cox  *	object.  Creates a wired mapping from this region to these pages, and
3060ff0fc84SAlan Cox  *	returns the region's starting virtual address.  If M_ZERO is specified
3070ff0fc84SAlan Cox  *	through the given flags, then the pages are zeroed before they are
3080ff0fc84SAlan Cox  *	mapped.
3090ff0fc84SAlan Cox  */
3109978bd99SMark Johnston static vm_offset_t
3117a469c8eSJeff Roberson kmem_alloc_contig_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
3120ff0fc84SAlan Cox     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
3130ff0fc84SAlan Cox     vm_memattr_t memattr)
3140ff0fc84SAlan Cox {
3157a469c8eSJeff Roberson 	vmem_t *vmem;
31633655d95SMark Johnston 	vm_object_t object;
317d1780e8dSKonstantin Belousov 	vm_offset_t addr, offset, tmp;
3180ff0fc84SAlan Cox 	vm_page_t end_m, m;
3192b914b85SMark Johnston 	vm_size_t asize;
320c869e672SAlan Cox 	u_long npages;
32133655d95SMark Johnston 	int pflags;
3220ff0fc84SAlan Cox 
32333655d95SMark Johnston 	object = kernel_object;
3242b914b85SMark Johnston 	asize = round_page(size);
3257a469c8eSJeff Roberson 	vmem = vm_dom[domain].vmd_kernel_arena;
3262b914b85SMark Johnston 	if (vmem_alloc(vmem, asize, flags | M_BESTFIT, &addr))
3270ff0fc84SAlan Cox 		return (0);
3280ff0fc84SAlan Cox 	offset = addr - VM_MIN_KERNEL_ADDRESS;
32991e31c3cSJeff Roberson 	pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
3302b914b85SMark Johnston 	npages = atop(asize);
33189f6b863SAttilio Rao 	VM_OBJECT_WLOCK(object);
33233655d95SMark Johnston 	m = kmem_alloc_contig_pages(object, atop(offset), domain,
33333655d95SMark Johnston 	    pflags, npages, low, high, alignment, boundary, memattr);
3340ff0fc84SAlan Cox 	if (m == NULL) {
33589f6b863SAttilio Rao 		VM_OBJECT_WUNLOCK(object);
3362b914b85SMark Johnston 		vmem_free(vmem, addr, asize);
3370ff0fc84SAlan Cox 		return (0);
3380ff0fc84SAlan Cox 	}
339431fb8abSMark Johnston 	KASSERT(vm_page_domain(m) == domain,
3407a469c8eSJeff Roberson 	    ("kmem_alloc_contig_domain: Domain mismatch %d != %d",
341431fb8abSMark Johnston 	    vm_page_domain(m), domain));
342c869e672SAlan Cox 	end_m = m + npages;
3435df87b21SJeff Roberson 	tmp = addr;
3440ff0fc84SAlan Cox 	for (; m < end_m; m++) {
3450ff0fc84SAlan Cox 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
3460ff0fc84SAlan Cox 			pmap_zero_page(m);
34791e31c3cSJeff Roberson 		vm_page_valid(m);
3480766f278SJonathan T. Looney 		pmap_enter(kernel_pmap, tmp, m, VM_PROT_RW,
3490766f278SJonathan T. Looney 		    VM_PROT_RW | PMAP_ENTER_WIRED, 0);
3505df87b21SJeff Roberson 		tmp += PAGE_SIZE;
3510ff0fc84SAlan Cox 	}
35289f6b863SAttilio Rao 	VM_OBJECT_WUNLOCK(object);
353*540da48dSMark Johnston 	kmem_alloc_san(addr, size, asize, flags);
3540ff0fc84SAlan Cox 	return (addr);
3550ff0fc84SAlan Cox }
3560ff0fc84SAlan Cox 
3577a469c8eSJeff Roberson vm_offset_t
35844d0efb2SAlan Cox kmem_alloc_contig(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high,
35944d0efb2SAlan Cox     u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
3607a469c8eSJeff Roberson {
3619978bd99SMark Johnston 
3629978bd99SMark Johnston 	return (kmem_alloc_contig_domainset(DOMAINSET_RR(), size, flags, low,
3639978bd99SMark Johnston 	    high, alignment, boundary, memattr));
3649978bd99SMark Johnston }
3659978bd99SMark Johnston 
3669978bd99SMark Johnston vm_offset_t
3679978bd99SMark Johnston kmem_alloc_contig_domainset(struct domainset *ds, vm_size_t size, int flags,
3689978bd99SMark Johnston     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
3699978bd99SMark Johnston     vm_memattr_t memattr)
3709978bd99SMark Johnston {
3717a469c8eSJeff Roberson 	struct vm_domainset_iter di;
3727a469c8eSJeff Roberson 	vm_offset_t addr;
3735fee468eSBryan Drewery 	int domain;
3747a469c8eSJeff Roberson 
3755fee468eSBryan Drewery 	vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
3767a469c8eSJeff Roberson 	do {
3777a469c8eSJeff Roberson 		addr = kmem_alloc_contig_domain(domain, size, flags, low, high,
3787a469c8eSJeff Roberson 		    alignment, boundary, memattr);
3797a469c8eSJeff Roberson 		if (addr != 0)
3807a469c8eSJeff Roberson 			break;
3814c29d2deSMark Johnston 	} while (vm_domainset_iter_policy(&di, &domain) == 0);
3827a469c8eSJeff Roberson 
3837a469c8eSJeff Roberson 	return (addr);
3847a469c8eSJeff Roberson }
3857a469c8eSJeff Roberson 
3860ff0fc84SAlan Cox /*
3877dd979dfSMark Johnston  *	kmem_subinit:
388df8bae1dSRodney W. Grimes  *
3897dd979dfSMark Johnston  *	Initializes a map to manage a subrange
390df8bae1dSRodney W. Grimes  *	of the kernel virtual address space.
391df8bae1dSRodney W. Grimes  *
392df8bae1dSRodney W. Grimes  *	Arguments are as follows:
393df8bae1dSRodney W. Grimes  *
394df8bae1dSRodney W. Grimes  *	parent		Map to take range from
395df8bae1dSRodney W. Grimes  *	min, max	Returned endpoints of map
396030f2369SAlfred Perlstein  *	size		Size of range to find
3973202ed75SAlan Cox  *	superpage_align	Request that min is superpage aligned
398df8bae1dSRodney W. Grimes  */
3997dd979dfSMark Johnston void
4007dd979dfSMark Johnston kmem_subinit(vm_map_t map, vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
4017dd979dfSMark Johnston     vm_size_t size, bool superpage_align)
402df8bae1dSRodney W. Grimes {
4036e4f51d1SAlfred Perlstein 	int ret;
40423955314SAlfred Perlstein 
405df8bae1dSRodney W. Grimes 	size = round_page(size);
406df8bae1dSRodney W. Grimes 
4072bc24aa9SAlan Cox 	*min = vm_map_min(parent);
408edb572a3SJohn Baldwin 	ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
4095aa60b6fSJohn Baldwin 	    VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
4103364c323SKonstantin Belousov 	    MAP_ACC_NO_CHARGE);
41124dedba9SAlan Cox 	if (ret != KERN_SUCCESS)
4127dd979dfSMark Johnston 		panic("kmem_subinit: bad status return of %d", ret);
413df8bae1dSRodney W. Grimes 	*max = *min + size;
4147dd979dfSMark Johnston 	vm_map_init(map, vm_map_pmap(parent), *min, *max);
4157dd979dfSMark Johnston 	if (vm_map_submap(parent, *min, *max, map) != KERN_SUCCESS)
4167dd979dfSMark Johnston 		panic("kmem_subinit: unable to change range to submap");
417df8bae1dSRodney W. Grimes }
418df8bae1dSRodney W. Grimes 
419df8bae1dSRodney W. Grimes /*
4209978bd99SMark Johnston  *	kmem_malloc_domain:
4211c7c3c6aSMatthew Dillon  *
4225df87b21SJeff Roberson  *	Allocate wired-down pages in the kernel's address space.
423df8bae1dSRodney W. Grimes  */
4249978bd99SMark Johnston static vm_offset_t
425067fd858SAlan Cox kmem_malloc_domain(int domain, vm_size_t size, int flags)
426df8bae1dSRodney W. Grimes {
4270766f278SJonathan T. Looney 	vmem_t *arena;
428df8bae1dSRodney W. Grimes 	vm_offset_t addr;
4292b914b85SMark Johnston 	vm_size_t asize;
4305df87b21SJeff Roberson 	int rv;
431df8bae1dSRodney W. Grimes 
432067fd858SAlan Cox 	if (__predict_true((flags & M_EXEC) == 0))
4330766f278SJonathan T. Looney 		arena = vm_dom[domain].vmd_kernel_arena;
4340766f278SJonathan T. Looney 	else
4350766f278SJonathan T. Looney 		arena = vm_dom[domain].vmd_kernel_rwx_arena;
4362b914b85SMark Johnston 	asize = round_page(size);
4372b914b85SMark Johnston 	if (vmem_alloc(arena, asize, flags | M_BESTFIT, &addr))
4385df87b21SJeff Roberson 		return (0);
439df8bae1dSRodney W. Grimes 
4402b914b85SMark Johnston 	rv = kmem_back_domain(domain, kernel_object, addr, asize, flags);
4415df87b21SJeff Roberson 	if (rv != KERN_SUCCESS) {
4422b914b85SMark Johnston 		vmem_free(arena, addr, asize);
443f31c239dSAlan Cox 		return (0);
444df8bae1dSRodney W. Grimes 	}
4452b914b85SMark Johnston 	kasan_mark((void *)addr, size, asize, KASAN_KMEM_REDZONE);
4465df87b21SJeff Roberson 	return (addr);
447e3813573SMatthew D Fleming }
448e3813573SMatthew D Fleming 
4497a469c8eSJeff Roberson vm_offset_t
45083a90bffSAlan Cox kmem_malloc(vm_size_t size, int flags)
4517a469c8eSJeff Roberson {
4529978bd99SMark Johnston 
4539978bd99SMark Johnston 	return (kmem_malloc_domainset(DOMAINSET_RR(), size, flags));
4549978bd99SMark Johnston }
4559978bd99SMark Johnston 
4569978bd99SMark Johnston vm_offset_t
4579978bd99SMark Johnston kmem_malloc_domainset(struct domainset *ds, vm_size_t size, int flags)
4589978bd99SMark Johnston {
4597a469c8eSJeff Roberson 	struct vm_domainset_iter di;
4607a469c8eSJeff Roberson 	vm_offset_t addr;
4617a469c8eSJeff Roberson 	int domain;
4627a469c8eSJeff Roberson 
4639978bd99SMark Johnston 	vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
4647a469c8eSJeff Roberson 	do {
465067fd858SAlan Cox 		addr = kmem_malloc_domain(domain, size, flags);
4667a469c8eSJeff Roberson 		if (addr != 0)
4677a469c8eSJeff Roberson 			break;
4684c29d2deSMark Johnston 	} while (vm_domainset_iter_policy(&di, &domain) == 0);
4697a469c8eSJeff Roberson 
4707a469c8eSJeff Roberson 	return (addr);
4717a469c8eSJeff Roberson }
4727a469c8eSJeff Roberson 
473e3813573SMatthew D Fleming /*
47425ed23cfSMark Johnston  *	kmem_back_domain:
475e3813573SMatthew D Fleming  *
47625ed23cfSMark Johnston  *	Allocate physical pages from the specified domain for the specified
47725ed23cfSMark Johnston  *	virtual address range.
478e3813573SMatthew D Fleming  */
479e3813573SMatthew D Fleming int
4807a469c8eSJeff Roberson kmem_back_domain(int domain, vm_object_t object, vm_offset_t addr,
4817a469c8eSJeff Roberson     vm_size_t size, int flags)
482e3813573SMatthew D Fleming {
483e3813573SMatthew D Fleming 	vm_offset_t offset, i;
48433fff5d5SMark Johnston 	vm_page_t m, mpred;
4850766f278SJonathan T. Looney 	vm_prot_t prot;
486e3813573SMatthew D Fleming 	int pflags;
487e3813573SMatthew D Fleming 
4882e47807cSJeff Roberson 	KASSERT(object == kernel_object,
4897a469c8eSJeff Roberson 	    ("kmem_back_domain: only supports kernel object."));
4905df87b21SJeff Roberson 
4910891ef4cSJohn Dyson 	offset = addr - VM_MIN_KERNEL_ADDRESS;
49291e31c3cSJeff Roberson 	pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
4938d6fbbb8SJeff Roberson 	pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
4948d6fbbb8SJeff Roberson 	if (flags & M_WAITOK)
4958d6fbbb8SJeff Roberson 		pflags |= VM_ALLOC_WAITFAIL;
4960766f278SJonathan T. Looney 	prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW;
497df8bae1dSRodney W. Grimes 
49833fff5d5SMark Johnston 	i = 0;
49933fff5d5SMark Johnston 	VM_OBJECT_WLOCK(object);
5008d6fbbb8SJeff Roberson retry:
50133fff5d5SMark Johnston 	mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i));
50233fff5d5SMark Johnston 	for (; i < size; i += PAGE_SIZE, mpred = m) {
5037a469c8eSJeff Roberson 		m = vm_page_alloc_domain_after(object, atop(offset + i),
5047a469c8eSJeff Roberson 		    domain, pflags, mpred);
505df8bae1dSRodney W. Grimes 
506df8bae1dSRodney W. Grimes 		/*
5070d94caffSDavid Greenman 		 * Ran out of space, free everything up and return. Don't need
5080d94caffSDavid Greenman 		 * to lock page queues here as we know that the pages we got
5090d94caffSDavid Greenman 		 * aren't on any queues.
510df8bae1dSRodney W. Grimes 		 */
511df8bae1dSRodney W. Grimes 		if (m == NULL) {
5128d6fbbb8SJeff Roberson 			if ((flags & M_NOWAIT) == 0)
513b18bfc3dSJohn Dyson 				goto retry;
5148d6fbbb8SJeff Roberson 			VM_OBJECT_WUNLOCK(object);
5159e829b22SAlan Cox 			kmem_unback(object, addr, i);
516e3813573SMatthew D Fleming 			return (KERN_NO_SPACE);
517df8bae1dSRodney W. Grimes 		}
518431fb8abSMark Johnston 		KASSERT(vm_page_domain(m) == domain,
5197a469c8eSJeff Roberson 		    ("kmem_back_domain: Domain mismatch %d != %d",
520431fb8abSMark Johnston 		    vm_page_domain(m), domain));
5211e081f88SJeff Roberson 		if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
522fff6062aSAlan Cox 			pmap_zero_page(m);
523d98d0ce2SKonstantin Belousov 		KASSERT((m->oflags & VPO_UNMANAGED) != 0,
5249f5c801bSAlan Cox 		    ("kmem_malloc: page %p is managed", m));
52591e31c3cSJeff Roberson 		vm_page_valid(m);
5260766f278SJonathan T. Looney 		pmap_enter(kernel_pmap, addr + i, m, prot,
5270766f278SJonathan T. Looney 		    prot | PMAP_ENTER_WIRED, 0);
52849bfa624SAlan Cox 		if (__predict_false((prot & VM_PROT_EXECUTE) != 0))
52949bfa624SAlan Cox 			m->oflags |= VPO_KMEM_EXEC;
530df8bae1dSRodney W. Grimes 	}
5315df87b21SJeff Roberson 	VM_OBJECT_WUNLOCK(object);
532*540da48dSMark Johnston 	kmem_alloc_san(addr, size, size, flags);
533e3813573SMatthew D Fleming 	return (KERN_SUCCESS);
534df8bae1dSRodney W. Grimes }
535df8bae1dSRodney W. Grimes 
53625ed23cfSMark Johnston /*
53725ed23cfSMark Johnston  *	kmem_back:
53825ed23cfSMark Johnston  *
53925ed23cfSMark Johnston  *	Allocate physical pages for the specified virtual address range.
54025ed23cfSMark Johnston  */
5417a469c8eSJeff Roberson int
5427a469c8eSJeff Roberson kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
5437a469c8eSJeff Roberson {
54425ed23cfSMark Johnston 	vm_offset_t end, next, start;
54525ed23cfSMark Johnston 	int domain, rv;
5467a469c8eSJeff Roberson 
5477a469c8eSJeff Roberson 	KASSERT(object == kernel_object,
5487a469c8eSJeff Roberson 	    ("kmem_back: only supports kernel object."));
5497a469c8eSJeff Roberson 
55025ed23cfSMark Johnston 	for (start = addr, end = addr + size; addr < end; addr = next) {
55125ed23cfSMark Johnston 		/*
55225ed23cfSMark Johnston 		 * We must ensure that pages backing a given large virtual page
55325ed23cfSMark Johnston 		 * all come from the same physical domain.
55425ed23cfSMark Johnston 		 */
55525ed23cfSMark Johnston 		if (vm_ndomains > 1) {
55625ed23cfSMark Johnston 			domain = (addr >> KVA_QUANTUM_SHIFT) % vm_ndomains;
55730c5525bSAndrew Gallatin 			while (VM_DOMAIN_EMPTY(domain))
55830c5525bSAndrew Gallatin 				domain++;
55925ed23cfSMark Johnston 			next = roundup2(addr + 1, KVA_QUANTUM);
56025ed23cfSMark Johnston 			if (next > end || next < start)
56125ed23cfSMark Johnston 				next = end;
5623d14a7bbSMark Johnston 		} else {
5633d14a7bbSMark Johnston 			domain = 0;
56425ed23cfSMark Johnston 			next = end;
5653d14a7bbSMark Johnston 		}
56625ed23cfSMark Johnston 		rv = kmem_back_domain(domain, object, addr, next - addr, flags);
56725ed23cfSMark Johnston 		if (rv != KERN_SUCCESS) {
56825ed23cfSMark Johnston 			kmem_unback(object, start, addr - start);
5697a469c8eSJeff Roberson 			break;
57025ed23cfSMark Johnston 		}
57125ed23cfSMark Johnston 	}
57225ed23cfSMark Johnston 	return (rv);
5737a469c8eSJeff Roberson }
5747a469c8eSJeff Roberson 
5759e829b22SAlan Cox /*
5769e829b22SAlan Cox  *	kmem_unback:
5779e829b22SAlan Cox  *
5789e829b22SAlan Cox  *	Unmap and free the physical pages underlying the specified virtual
5799e829b22SAlan Cox  *	address range.
5809e829b22SAlan Cox  *
5819e829b22SAlan Cox  *	A physical page must exist within the specified object at each index
5829e829b22SAlan Cox  *	that is being unmapped.
5839e829b22SAlan Cox  */
58449bfa624SAlan Cox static struct vmem *
5857a469c8eSJeff Roberson _kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
5865df87b21SJeff Roberson {
58749bfa624SAlan Cox 	struct vmem *arena;
5887e05ffa6SMark Johnston 	vm_page_t m, next;
5897e05ffa6SMark Johnston 	vm_offset_t end, offset;
5907a469c8eSJeff Roberson 	int domain;
5915df87b21SJeff Roberson 
5922e47807cSJeff Roberson 	KASSERT(object == kernel_object,
5932e47807cSJeff Roberson 	    ("kmem_unback: only supports kernel object."));
5945df87b21SJeff Roberson 
5957a469c8eSJeff Roberson 	if (size == 0)
59649bfa624SAlan Cox 		return (NULL);
597fa2f411cSAlan Cox 	pmap_remove(kernel_pmap, addr, addr + size);
5985df87b21SJeff Roberson 	offset = addr - VM_MIN_KERNEL_ADDRESS;
5997e05ffa6SMark Johnston 	end = offset + size;
6005df87b21SJeff Roberson 	VM_OBJECT_WLOCK(object);
6017a469c8eSJeff Roberson 	m = vm_page_lookup(object, atop(offset));
602431fb8abSMark Johnston 	domain = vm_page_domain(m);
60349bfa624SAlan Cox 	if (__predict_true((m->oflags & VPO_KMEM_EXEC) == 0))
60449bfa624SAlan Cox 		arena = vm_dom[domain].vmd_kernel_arena;
60549bfa624SAlan Cox 	else
60649bfa624SAlan Cox 		arena = vm_dom[domain].vmd_kernel_rwx_arena;
6077a469c8eSJeff Roberson 	for (; offset < end; offset += PAGE_SIZE, m = next) {
6087e05ffa6SMark Johnston 		next = vm_page_next(m);
609e9ceb9ddSJeff Roberson 		vm_page_xbusy_claim(m);
61088ea538aSMark Johnston 		vm_page_unwire_noq(m);
6115df87b21SJeff Roberson 		vm_page_free(m);
6125df87b21SJeff Roberson 	}
6135df87b21SJeff Roberson 	VM_OBJECT_WUNLOCK(object);
6147a469c8eSJeff Roberson 
61549bfa624SAlan Cox 	return (arena);
6167a469c8eSJeff Roberson }
6177a469c8eSJeff Roberson 
6187a469c8eSJeff Roberson void
6197a469c8eSJeff Roberson kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
6207a469c8eSJeff Roberson {
6217a469c8eSJeff Roberson 
62249bfa624SAlan Cox 	(void)_kmem_unback(object, addr, size);
6235df87b21SJeff Roberson }
6245df87b21SJeff Roberson 
625df8bae1dSRodney W. Grimes /*
6265df87b21SJeff Roberson  *	kmem_free:
6275df87b21SJeff Roberson  *
6285df87b21SJeff Roberson  *	Free memory allocated with kmem_malloc.  The size must match the
6295df87b21SJeff Roberson  *	original allocation.
6305df87b21SJeff Roberson  */
6315df87b21SJeff Roberson void
63249bfa624SAlan Cox kmem_free(vm_offset_t addr, vm_size_t size)
6335df87b21SJeff Roberson {
6340766f278SJonathan T. Looney 	struct vmem *arena;
6350766f278SJonathan T. Looney 
6365df87b21SJeff Roberson 	size = round_page(size);
6372b914b85SMark Johnston 	kasan_mark((void *)addr, size, size, 0);
63849bfa624SAlan Cox 	arena = _kmem_unback(kernel_object, addr, size);
63949bfa624SAlan Cox 	if (arena != NULL)
6400766f278SJonathan T. Looney 		vmem_free(arena, addr, size);
6415df87b21SJeff Roberson }
6425df87b21SJeff Roberson 
6435df87b21SJeff Roberson /*
6445df87b21SJeff Roberson  *	kmap_alloc_wait:
645df8bae1dSRodney W. Grimes  *
646df8bae1dSRodney W. Grimes  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
647df8bae1dSRodney W. Grimes  *	has no room, the caller sleeps waiting for more memory in the submap.
648df8bae1dSRodney W. Grimes  *
6491c7c3c6aSMatthew Dillon  *	This routine may block.
650df8bae1dSRodney W. Grimes  */
6510d94caffSDavid Greenman vm_offset_t
6526e309d75SEd Maste kmap_alloc_wait(vm_map_t map, vm_size_t size)
653df8bae1dSRodney W. Grimes {
654df8bae1dSRodney W. Grimes 	vm_offset_t addr;
65523955314SAlfred Perlstein 
656df8bae1dSRodney W. Grimes 	size = round_page(size);
6573364c323SKonstantin Belousov 	if (!swap_reserve(size))
6583364c323SKonstantin Belousov 		return (0);
659df8bae1dSRodney W. Grimes 
660df8bae1dSRodney W. Grimes 	for (;;) {
661df8bae1dSRodney W. Grimes 		/*
6620d94caffSDavid Greenman 		 * To make this work for more than one map, use the map's lock
6630d94caffSDavid Greenman 		 * to lock out sleepers/wakers.
664df8bae1dSRodney W. Grimes 		 */
665df8bae1dSRodney W. Grimes 		vm_map_lock(map);
6669f701172SKonstantin Belousov 		addr = vm_map_findspace(map, vm_map_min(map), size);
6679f701172SKonstantin Belousov 		if (addr + size <= vm_map_max(map))
668df8bae1dSRodney W. Grimes 			break;
669df8bae1dSRodney W. Grimes 		/* no space now; see if we can ever get space */
670df8bae1dSRodney W. Grimes 		if (vm_map_max(map) - vm_map_min(map) < size) {
671df8bae1dSRodney W. Grimes 			vm_map_unlock(map);
6723364c323SKonstantin Belousov 			swap_release(size);
673df8bae1dSRodney W. Grimes 			return (0);
674df8bae1dSRodney W. Grimes 		}
6759688f931SAlan Cox 		map->needs_wakeup = TRUE;
6768ce2d00aSPawel Jakub Dawidek 		vm_map_unlock_and_wait(map, 0);
677df8bae1dSRodney W. Grimes 	}
67807702f72SMark Johnston 	vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_RW, VM_PROT_RW,
67907702f72SMark Johnston 	    MAP_ACC_CHARGED);
680df8bae1dSRodney W. Grimes 	vm_map_unlock(map);
681df8bae1dSRodney W. Grimes 	return (addr);
682df8bae1dSRodney W. Grimes }
683df8bae1dSRodney W. Grimes 
684df8bae1dSRodney W. Grimes /*
6855df87b21SJeff Roberson  *	kmap_free_wakeup:
686df8bae1dSRodney W. Grimes  *
68724a1cce3SDavid Greenman  *	Returns memory to a submap of the kernel, and wakes up any processes
688df8bae1dSRodney W. Grimes  *	waiting for memory in that map.
689df8bae1dSRodney W. Grimes  */
6900d94caffSDavid Greenman void
6916e309d75SEd Maste kmap_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
692df8bae1dSRodney W. Grimes {
69323955314SAlfred Perlstein 
694df8bae1dSRodney W. Grimes 	vm_map_lock(map);
695655c3490SKonstantin Belousov 	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
6969688f931SAlan Cox 	if (map->needs_wakeup) {
6979688f931SAlan Cox 		map->needs_wakeup = FALSE;
6989688f931SAlan Cox 		vm_map_wakeup(map);
6999688f931SAlan Cox 	}
700df8bae1dSRodney W. Grimes 	vm_map_unlock(map);
701df8bae1dSRodney W. Grimes }
702df8bae1dSRodney W. Grimes 
7035df87b21SJeff Roberson void
70489cb2a19SMatthew D Fleming kmem_init_zero_region(void)
70589cb2a19SMatthew D Fleming {
706cfb00e5aSMatthew D Fleming 	vm_offset_t addr, i;
70789cb2a19SMatthew D Fleming 	vm_page_t m;
70889cb2a19SMatthew D Fleming 
709cfb00e5aSMatthew D Fleming 	/*
710cfb00e5aSMatthew D Fleming 	 * Map a single physical page of zeros to a larger virtual range.
711cfb00e5aSMatthew D Fleming 	 * This requires less looping in places that want large amounts of
712cfb00e5aSMatthew D Fleming 	 * zeros, while not using much more physical resources.
713cfb00e5aSMatthew D Fleming 	 */
7145df87b21SJeff Roberson 	addr = kva_alloc(ZERO_REGION_SIZE);
715a4667e09SMark Johnston 	m = vm_page_alloc_noobj(VM_ALLOC_WIRED | VM_ALLOC_ZERO);
71689cb2a19SMatthew D Fleming 	for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
71789cb2a19SMatthew D Fleming 		pmap_qenter(addr + i, &m, 1);
7185df87b21SJeff Roberson 	pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
71989cb2a19SMatthew D Fleming 
72089cb2a19SMatthew D Fleming 	zero_region = (const void *)addr;
72189cb2a19SMatthew D Fleming }
72289cb2a19SMatthew D Fleming 
723df8bae1dSRodney W. Grimes /*
724969e147aSMark Johnston  * Import KVA from the kernel map into the kernel arena.
7251aed6d48SMark Johnston  */
7261aed6d48SMark Johnston static int
7271aed6d48SMark Johnston kva_import(void *unused, vmem_size_t size, int flags, vmem_addr_t *addrp)
7281aed6d48SMark Johnston {
7291aed6d48SMark Johnston 	vm_offset_t addr;
7301aed6d48SMark Johnston 	int result;
7311aed6d48SMark Johnston 
7321aed6d48SMark Johnston 	KASSERT((size % KVA_QUANTUM) == 0,
7331aed6d48SMark Johnston 	    ("kva_import: Size %jd is not a multiple of %d",
7341aed6d48SMark Johnston 	    (intmax_t)size, (int)KVA_QUANTUM));
7351aed6d48SMark Johnston 	addr = vm_map_min(kernel_map);
7361aed6d48SMark Johnston 	result = vm_map_find(kernel_map, NULL, 0, &addr, size, 0,
7371aed6d48SMark Johnston 	    VMFS_SUPER_SPACE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
7381aed6d48SMark Johnston 	if (result != KERN_SUCCESS)
7391aed6d48SMark Johnston                 return (ENOMEM);
7401aed6d48SMark Johnston 
7411aed6d48SMark Johnston 	*addrp = addr;
7421aed6d48SMark Johnston 
7431aed6d48SMark Johnston 	return (0);
7441aed6d48SMark Johnston }
7451aed6d48SMark Johnston 
7461aed6d48SMark Johnston /*
747969e147aSMark Johnston  * Import KVA from a parent arena into a per-domain arena.  Imports must be
748969e147aSMark Johnston  * KVA_QUANTUM-aligned and a multiple of KVA_QUANTUM in size.
7491aed6d48SMark Johnston  */
7501aed6d48SMark Johnston static int
751969e147aSMark Johnston kva_import_domain(void *arena, vmem_size_t size, int flags, vmem_addr_t *addrp)
7521aed6d48SMark Johnston {
7531aed6d48SMark Johnston 
7541aed6d48SMark Johnston 	KASSERT((size % KVA_QUANTUM) == 0,
755969e147aSMark Johnston 	    ("kva_import_domain: Size %jd is not a multiple of %d",
7561aed6d48SMark Johnston 	    (intmax_t)size, (int)KVA_QUANTUM));
7571aed6d48SMark Johnston 	return (vmem_xalloc(arena, size, KVA_QUANTUM, 0, 0, VMEM_ADDR_MIN,
7581aed6d48SMark Johnston 	    VMEM_ADDR_MAX, flags, addrp));
7591aed6d48SMark Johnston }
7601aed6d48SMark Johnston 
7611aed6d48SMark Johnston /*
7621c7c3c6aSMatthew Dillon  * 	kmem_init:
7631c7c3c6aSMatthew Dillon  *
7641c7c3c6aSMatthew Dillon  *	Create the kernel map; insert a mapping covering kernel text,
7651c7c3c6aSMatthew Dillon  *	data, bss, and all space allocated thus far (`boostrap' data).  The
7661c7c3c6aSMatthew Dillon  *	new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
7671c7c3c6aSMatthew Dillon  *	`start' as allocated, and the range between `start' and `end' as free.
7681aed6d48SMark Johnston  *	Create the kernel vmem arena and its per-domain children.
769df8bae1dSRodney W. Grimes  */
7700d94caffSDavid Greenman void
7716e309d75SEd Maste kmem_init(vm_offset_t start, vm_offset_t end)
772df8bae1dSRodney W. Grimes {
773aea9103eSMark Johnston 	vm_size_t quantum;
7741aed6d48SMark Johnston 	int domain;
775df8bae1dSRodney W. Grimes 
7767efe14cbSMark Johnston 	vm_map_init(kernel_map, kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
7777efe14cbSMark Johnston 	kernel_map->system_map = 1;
7787efe14cbSMark Johnston 	vm_map_lock(kernel_map);
779df8bae1dSRodney W. Grimes 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
7807efe14cbSMark Johnston 	(void)vm_map_insert(kernel_map, NULL, 0,
7815cfa90e9SAlan Cox #ifdef __amd64__
7825cfa90e9SAlan Cox 	    KERNBASE,
7835cfa90e9SAlan Cox #else
7845cfa90e9SAlan Cox 	    VM_MIN_KERNEL_ADDRESS,
7855cfa90e9SAlan Cox #endif
7865cfa90e9SAlan Cox 	    start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
787df8bae1dSRodney W. Grimes 	/* ... and ending with the completion of the above `insert' */
7889d75f0dcSMark Johnston 
7899d75f0dcSMark Johnston #ifdef __amd64__
7909d75f0dcSMark Johnston 	/*
7919d75f0dcSMark Johnston 	 * Mark KVA used for the page array as allocated.  Other platforms
7929d75f0dcSMark Johnston 	 * that handle vm_page_array allocation can simply adjust virtual_avail
7939d75f0dcSMark Johnston 	 * instead.
7949d75f0dcSMark Johnston 	 */
7957dd979dfSMark Johnston 	(void)vm_map_insert(kernel_map, NULL, 0, (vm_offset_t)vm_page_array,
7969d75f0dcSMark Johnston 	    (vm_offset_t)vm_page_array + round_2mpage(vm_page_array_size *
7979d75f0dcSMark Johnston 	    sizeof(struct vm_page)),
7989d75f0dcSMark Johnston 	    VM_PROT_RW, VM_PROT_RW, MAP_NOFAULT);
7999d75f0dcSMark Johnston #endif
8007dd979dfSMark Johnston 	vm_map_unlock(kernel_map);
8011aed6d48SMark Johnston 
8021aed6d48SMark Johnston 	/*
803aea9103eSMark Johnston 	 * Use a large import quantum on NUMA systems.  This helps minimize
804aea9103eSMark Johnston 	 * interleaving of superpages, reducing internal fragmentation within
805aea9103eSMark Johnston 	 * the per-domain arenas.
806aea9103eSMark Johnston 	 */
807aea9103eSMark Johnston 	if (vm_ndomains > 1 && PMAP_HAS_DMAP)
808aea9103eSMark Johnston 		quantum = KVA_NUMA_IMPORT_QUANTUM;
809aea9103eSMark Johnston 	else
810aea9103eSMark Johnston 		quantum = KVA_QUANTUM;
811aea9103eSMark Johnston 
812aea9103eSMark Johnston 	/*
8131aed6d48SMark Johnston 	 * Initialize the kernel_arena.  This can grow on demand.
8141aed6d48SMark Johnston 	 */
8151aed6d48SMark Johnston 	vmem_init(kernel_arena, "kernel arena", 0, 0, PAGE_SIZE, 0, 0);
816aea9103eSMark Johnston 	vmem_set_import(kernel_arena, kva_import, NULL, NULL, quantum);
8171aed6d48SMark Johnston 
8181aed6d48SMark Johnston 	for (domain = 0; domain < vm_ndomains; domain++) {
819969e147aSMark Johnston 		/*
820969e147aSMark Johnston 		 * Initialize the per-domain arenas.  These are used to color
821969e147aSMark Johnston 		 * the KVA space in a way that ensures that virtual large pages
822969e147aSMark Johnston 		 * are backed by memory from the same physical domain,
823969e147aSMark Johnston 		 * maximizing the potential for superpage promotion.
824969e147aSMark Johnston 		 */
8251aed6d48SMark Johnston 		vm_dom[domain].vmd_kernel_arena = vmem_create(
8261aed6d48SMark Johnston 		    "kernel arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
8271aed6d48SMark Johnston 		vmem_set_import(vm_dom[domain].vmd_kernel_arena,
828aea9103eSMark Johnston 		    kva_import_domain, NULL, kernel_arena, quantum);
829969e147aSMark Johnston 
830969e147aSMark Johnston 		/*
831969e147aSMark Johnston 		 * In architectures with superpages, maintain separate arenas
832969e147aSMark Johnston 		 * for allocations with permissions that differ from the
833969e147aSMark Johnston 		 * "standard" read/write permissions used for kernel memory,
834969e147aSMark Johnston 		 * so as not to inhibit superpage promotion.
835aea9103eSMark Johnston 		 *
836aea9103eSMark Johnston 		 * Use the base import quantum since this arena is rarely used.
837969e147aSMark Johnston 		 */
8381aed6d48SMark Johnston #if VM_NRESERVLEVEL > 0
8391aed6d48SMark Johnston 		vm_dom[domain].vmd_kernel_rwx_arena = vmem_create(
8401aed6d48SMark Johnston 		    "kernel rwx arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
8411aed6d48SMark Johnston 		vmem_set_import(vm_dom[domain].vmd_kernel_rwx_arena,
842969e147aSMark Johnston 		    kva_import_domain, (vmem_release_t *)vmem_xfree,
843969e147aSMark Johnston 		    kernel_arena, KVA_QUANTUM);
844303b7702SMark Johnston #else
845303b7702SMark Johnston 		vm_dom[domain].vmd_kernel_rwx_arena =
846303b7702SMark Johnston 		    vm_dom[domain].vmd_kernel_arena;
8471aed6d48SMark Johnston #endif
8481aed6d48SMark Johnston 	}
849a81c400eSJeff Roberson 
850a81c400eSJeff Roberson 	/*
851a81c400eSJeff Roberson 	 * This must be the very first call so that the virtual address
852a81c400eSJeff Roberson 	 * space used for early allocations is properly marked used in
853a81c400eSJeff Roberson 	 * the map.
854a81c400eSJeff Roberson 	 */
855a81c400eSJeff Roberson 	uma_startup2();
856df8bae1dSRodney W. Grimes }
85786f08737SRobert Watson 
858483f692eSMark Johnston /*
859483f692eSMark Johnston  *	kmem_bootstrap_free:
860483f692eSMark Johnston  *
861483f692eSMark Johnston  *	Free pages backing preloaded data (e.g., kernel modules) to the
862483f692eSMark Johnston  *	system.  Currently only supported on platforms that create a
863483f692eSMark Johnston  *	vm_phys segment for preloaded data.
864483f692eSMark Johnston  */
865483f692eSMark Johnston void
866483f692eSMark Johnston kmem_bootstrap_free(vm_offset_t start, vm_size_t size)
867483f692eSMark Johnston {
868483f692eSMark Johnston #if defined(__i386__) || defined(__amd64__)
869483f692eSMark Johnston 	struct vm_domain *vmd;
8706c85795aSMark Johnston 	vm_offset_t end, va;
871483f692eSMark Johnston 	vm_paddr_t pa;
872483f692eSMark Johnston 	vm_page_t m;
873483f692eSMark Johnston 
874483f692eSMark Johnston 	end = trunc_page(start + size);
875483f692eSMark Johnston 	start = round_page(start);
876483f692eSMark Johnston 
877f822c9e2SMark Johnston #ifdef __amd64__
878f822c9e2SMark Johnston 	/*
879f822c9e2SMark Johnston 	 * Preloaded files do not have execute permissions by default on amd64.
880f822c9e2SMark Johnston 	 * Restore the default permissions to ensure that the direct map alias
881f822c9e2SMark Johnston 	 * is updated.
882f822c9e2SMark Johnston 	 */
883f822c9e2SMark Johnston 	pmap_change_prot(start, end - start, VM_PROT_RW);
884f822c9e2SMark Johnston #endif
8856c85795aSMark Johnston 	for (va = start; va < end; va += PAGE_SIZE) {
8866c85795aSMark Johnston 		pa = pmap_kextract(va);
887483f692eSMark Johnston 		m = PHYS_TO_VM_PAGE(pa);
888483f692eSMark Johnston 
889483f692eSMark Johnston 		vmd = vm_pagequeue_domain(m);
890483f692eSMark Johnston 		vm_domain_free_lock(vmd);
891483f692eSMark Johnston 		vm_phys_free_pages(m, 0);
892483f692eSMark Johnston 		vm_domain_free_unlock(vmd);
893c16bd872SMark Johnston 
894c16bd872SMark Johnston 		vm_domain_freecnt_inc(vmd, 1);
895c16bd872SMark Johnston 		vm_cnt.v_page_count++;
896483f692eSMark Johnston 	}
8976c85795aSMark Johnston 	pmap_remove(kernel_pmap, start, end);
8986c85795aSMark Johnston 	(void)vmem_add(kernel_arena, start, end - start, M_WAITOK);
899483f692eSMark Johnston #endif
900483f692eSMark Johnston }
901483f692eSMark Johnston 
90286f08737SRobert Watson /*
90386f08737SRobert Watson  * Allow userspace to directly trigger the VM drain routine for testing
90486f08737SRobert Watson  * purposes.
90586f08737SRobert Watson  */
90686f08737SRobert Watson static int
90786f08737SRobert Watson debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
90886f08737SRobert Watson {
90986f08737SRobert Watson 	int error, i;
91086f08737SRobert Watson 
91186f08737SRobert Watson 	i = 0;
91286f08737SRobert Watson 	error = sysctl_handle_int(oidp, &i, 0, req);
91351a7be5fSKonstantin Belousov 	if (error != 0)
91486f08737SRobert Watson 		return (error);
9159b43bc27SAndriy Gapon 	if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
9169b43bc27SAndriy Gapon 		return (EINVAL);
9179b43bc27SAndriy Gapon 	if (i != 0)
9189b43bc27SAndriy Gapon 		EVENTHANDLER_INVOKE(vm_lowmem, i);
91986f08737SRobert Watson 	return (0);
92086f08737SRobert Watson }
92151a7be5fSKonstantin Belousov SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem,
92251a7be5fSKonstantin Belousov     CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0, debug_vm_lowmem, "I",
92351a7be5fSKonstantin Belousov     "set to trigger vm_lowmem event with given flags");
92489619b74SKonstantin Belousov 
92589619b74SKonstantin Belousov static int
92689619b74SKonstantin Belousov debug_uma_reclaim(SYSCTL_HANDLER_ARGS)
92789619b74SKonstantin Belousov {
92889619b74SKonstantin Belousov 	int error, i;
92989619b74SKonstantin Belousov 
93089619b74SKonstantin Belousov 	i = 0;
93189619b74SKonstantin Belousov 	error = sysctl_handle_int(oidp, &i, 0, req);
932aabe13f1SMark Johnston 	if (error != 0 || req->newptr == NULL)
93389619b74SKonstantin Belousov 		return (error);
93489619b74SKonstantin Belousov 	if (i != UMA_RECLAIM_TRIM && i != UMA_RECLAIM_DRAIN &&
93589619b74SKonstantin Belousov 	    i != UMA_RECLAIM_DRAIN_CPU)
93689619b74SKonstantin Belousov 		return (EINVAL);
93789619b74SKonstantin Belousov 	uma_reclaim(i);
93889619b74SKonstantin Belousov 	return (0);
93989619b74SKonstantin Belousov }
94089619b74SKonstantin Belousov SYSCTL_PROC(_debug, OID_AUTO, uma_reclaim,
94189619b74SKonstantin Belousov     CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0, debug_uma_reclaim, "I",
94289619b74SKonstantin Belousov     "set to generate request to reclaim uma caches");
943aabe13f1SMark Johnston 
944aabe13f1SMark Johnston static int
945aabe13f1SMark Johnston debug_uma_reclaim_domain(SYSCTL_HANDLER_ARGS)
946aabe13f1SMark Johnston {
947aabe13f1SMark Johnston 	int domain, error, request;
948aabe13f1SMark Johnston 
949aabe13f1SMark Johnston 	request = 0;
950aabe13f1SMark Johnston 	error = sysctl_handle_int(oidp, &request, 0, req);
951aabe13f1SMark Johnston 	if (error != 0 || req->newptr == NULL)
952aabe13f1SMark Johnston 		return (error);
953aabe13f1SMark Johnston 
954aabe13f1SMark Johnston 	domain = request >> 4;
955aabe13f1SMark Johnston 	request &= 0xf;
956aabe13f1SMark Johnston 	if (request != UMA_RECLAIM_TRIM && request != UMA_RECLAIM_DRAIN &&
957aabe13f1SMark Johnston 	    request != UMA_RECLAIM_DRAIN_CPU)
958aabe13f1SMark Johnston 		return (EINVAL);
959aabe13f1SMark Johnston 	if (domain < 0 || domain >= vm_ndomains)
960aabe13f1SMark Johnston 		return (EINVAL);
961aabe13f1SMark Johnston 	uma_reclaim_domain(request, domain);
962aabe13f1SMark Johnston 	return (0);
963aabe13f1SMark Johnston }
964aabe13f1SMark Johnston SYSCTL_PROC(_debug, OID_AUTO, uma_reclaim_domain,
965aabe13f1SMark Johnston     CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0,
966aabe13f1SMark Johnston     debug_uma_reclaim_domain, "I",
967aabe13f1SMark Johnston     "");
968