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_init.c 8.1 (Berkeley) 6/11/93 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 * Initialize the Virtual Memory subsystem. 65df8bae1dSRodney W. Grimes */ 66df8bae1dSRodney W. Grimes 67874651b1SDavid E. O'Brien #include <sys/cdefs.h> 68874651b1SDavid E. O'Brien __FBSDID("$FreeBSD$"); 69874651b1SDavid E. O'Brien 70df8bae1dSRodney W. Grimes #include <sys/param.h> 712b14f991SJulian Elischer #include <sys/kernel.h> 72fb919e4dSMark Murray #include <sys/lock.h> 73fb919e4dSMark Murray #include <sys/proc.h> 7489f6b863SAttilio Rao #include <sys/rwlock.h> 755df87b21SJeff Roberson #include <sys/malloc.h> 76f5fca0d8SKris Kennaway #include <sys/sysctl.h> 7705f0fdd2SPoul-Henning Kamp #include <sys/systm.h> 78cebde069SMike Silbersack #include <sys/selinfo.h> 7921fae961SJeff Roberson #include <sys/smp.h> 80cebde069SMike Silbersack #include <sys/pipe.h> 81219d632cSMatthew Dillon #include <sys/bio.h> 82219d632cSMatthew Dillon #include <sys/buf.h> 835f518366SJeff Roberson #include <sys/vmem.h> 847a469c8eSJeff Roberson #include <sys/vmmeter.h> 85df8bae1dSRodney W. Grimes 86df8bae1dSRodney W. Grimes #include <vm/vm.h> 87219d632cSMatthew Dillon #include <vm/vm_param.h> 88219d632cSMatthew Dillon #include <vm/vm_kern.h> 89efeaf95aSDavid Greenman #include <vm/vm_object.h> 90df8bae1dSRodney W. Grimes #include <vm/vm_page.h> 917a469c8eSJeff Roberson #include <vm/vm_phys.h> 92*e2068d0bSJeff Roberson #include <vm/vm_pagequeue.h> 93efeaf95aSDavid Greenman #include <vm/vm_map.h> 9424a1cce3SDavid Greenman #include <vm/vm_pager.h> 95efeaf95aSDavid Greenman #include <vm/vm_extern.h> 96df8bae1dSRodney W. Grimes 97ae941b1bSGleb Smirnoff extern void uma_startup1(void); 98ae941b1bSGleb Smirnoff extern void vm_radix_reserve_kva(void); 997a469c8eSJeff Roberson 1007a469c8eSJeff Roberson #if VM_NRESERVLEVEL > 0 1017a469c8eSJeff Roberson #define KVA_QUANTUM (1 << (VM_LEVEL_0_ORDER + PAGE_SHIFT)) 1027a469c8eSJeff Roberson #else 1037a469c8eSJeff Roberson /* On non-superpage architectures want large import sizes. */ 1047a469c8eSJeff Roberson #define KVA_QUANTUM (PAGE_SIZE * 1024) 1057a469c8eSJeff Roberson #endif 1060fca57b8SThomas Moestl long physmem; 1070fca57b8SThomas Moestl 108df8bae1dSRodney W. Grimes /* 1092b14f991SJulian Elischer * System initialization 1102b14f991SJulian Elischer */ 11111caded3SAlfred Perlstein static void vm_mem_init(void *); 112237fdd78SRobert Watson SYSINIT(vm_mem, SI_SUB_VM, SI_ORDER_FIRST, vm_mem_init, NULL); 1132b14f991SJulian Elischer 1142b14f991SJulian Elischer /* 1155df87b21SJeff Roberson * Import kva into the kernel arena. 1165df87b21SJeff Roberson */ 1175df87b21SJeff Roberson static int 1185df87b21SJeff Roberson kva_import(void *unused, vmem_size_t size, int flags, vmem_addr_t *addrp) 1195df87b21SJeff Roberson { 1205df87b21SJeff Roberson vm_offset_t addr; 1215df87b21SJeff Roberson int result; 1225df87b21SJeff Roberson 1237a469c8eSJeff Roberson KASSERT((size % KVA_QUANTUM) == 0, 1247a469c8eSJeff Roberson ("kva_import: Size %jd is not a multiple of %d", 1257a469c8eSJeff Roberson (intmax_t)size, (int)KVA_QUANTUM)); 1265df87b21SJeff Roberson addr = vm_map_min(kernel_map); 127edb572a3SJohn Baldwin result = vm_map_find(kernel_map, NULL, 0, &addr, size, 0, 1285aa60b6fSJohn Baldwin VMFS_SUPER_SPACE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT); 1295df87b21SJeff Roberson if (result != KERN_SUCCESS) 1305df87b21SJeff Roberson return (ENOMEM); 1315df87b21SJeff Roberson 1325df87b21SJeff Roberson *addrp = addr; 1335df87b21SJeff Roberson 1345df87b21SJeff Roberson return (0); 1355df87b21SJeff Roberson } 1365df87b21SJeff Roberson 1375df87b21SJeff Roberson /* 138df8bae1dSRodney W. Grimes * vm_init initializes the virtual memory system. 139df8bae1dSRodney W. Grimes * This is done only by the first cpu up. 140df8bae1dSRodney W. Grimes * 141df8bae1dSRodney W. Grimes * The start and end address of physical memory is passed in. 142df8bae1dSRodney W. Grimes */ 1432b14f991SJulian Elischer /* ARGSUSED*/ 1442b14f991SJulian Elischer static void 145d841aaa7SBruce Evans vm_mem_init(dummy) 146d841aaa7SBruce Evans void *dummy; 147df8bae1dSRodney W. Grimes { 1487a469c8eSJeff Roberson int domain; 1495df87b21SJeff Roberson 150df8bae1dSRodney W. Grimes /* 1510d94caffSDavid Greenman * Initializes resident memory structures. From here on, all physical 1520d94caffSDavid Greenman * memory is accounted for, and we use only virtual addresses. 153df8bae1dSRodney W. Grimes */ 15426f9a767SRodney W. Grimes vm_set_page_size(); 155889eb0fcSAlan Cox virtual_avail = vm_page_startup(virtual_avail); 156e7841165SDag-Erling Smørgrav 157ae941b1bSGleb Smirnoff #ifdef UMA_MD_SMALL_ALLOC 158ae941b1bSGleb Smirnoff /* Announce page availability to UMA. */ 159ae941b1bSGleb Smirnoff uma_startup1(); 160ae941b1bSGleb Smirnoff #endif 161df8bae1dSRodney W. Grimes /* 162df8bae1dSRodney W. Grimes * Initialize other VM packages 163df8bae1dSRodney W. Grimes */ 1645f518366SJeff Roberson vmem_startup(); 165a316d390SJohn Dyson vm_object_init(); 166df8bae1dSRodney W. Grimes vm_map_startup(); 167df8bae1dSRodney W. Grimes kmem_init(virtual_avail, virtual_end); 1685df87b21SJeff Roberson 1695df87b21SJeff Roberson /* 1705df87b21SJeff Roberson * Initialize the kernel_arena. This can grow on demand. 1715df87b21SJeff Roberson */ 1725df87b21SJeff Roberson vmem_init(kernel_arena, "kernel arena", 0, 0, PAGE_SIZE, 0, 0); 1737a469c8eSJeff Roberson vmem_set_import(kernel_arena, kva_import, NULL, NULL, KVA_QUANTUM); 1747a469c8eSJeff Roberson 1757a469c8eSJeff Roberson for (domain = 0; domain < vm_ndomains; domain++) { 1767a469c8eSJeff Roberson vm_dom[domain].vmd_kernel_arena = vmem_create( 1777a469c8eSJeff Roberson "kernel arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK); 1787a469c8eSJeff Roberson vmem_set_import(vm_dom[domain].vmd_kernel_arena, 1797a469c8eSJeff Roberson (vmem_import_t *)vmem_alloc, NULL, kernel_arena, 1807a469c8eSJeff Roberson KVA_QUANTUM); 1817a469c8eSJeff Roberson } 1825df87b21SJeff Roberson 183ae941b1bSGleb Smirnoff #ifndef UMA_MD_SMALL_ALLOC 184ae941b1bSGleb Smirnoff /* Set up radix zone to use noobj_alloc. */ 185ae941b1bSGleb Smirnoff vm_radix_reserve_kva(); 186ae941b1bSGleb Smirnoff /* Announce page availability to UMA. */ 187ae941b1bSGleb Smirnoff uma_startup1(); 188ae941b1bSGleb Smirnoff #endif 1895df87b21SJeff Roberson kmem_init_zero_region(); 190bdb93eb2SAlan Cox pmap_init(); 191df8bae1dSRodney W. Grimes vm_pager_init(); 192df8bae1dSRodney W. Grimes } 193219d632cSMatthew Dillon 194219d632cSMatthew Dillon void 195219d632cSMatthew Dillon vm_ksubmap_init(struct kva_md_info *kmi) 196219d632cSMatthew Dillon { 197219d632cSMatthew Dillon vm_offset_t firstaddr; 198219d632cSMatthew Dillon caddr_t v; 199219d632cSMatthew Dillon vm_size_t size = 0; 200447b3772SPeter Wemm long physmem_est; 201219d632cSMatthew Dillon vm_offset_t minaddr; 202219d632cSMatthew Dillon vm_offset_t maxaddr; 203219d632cSMatthew Dillon 204219d632cSMatthew Dillon /* 205219d632cSMatthew Dillon * Allocate space for system data structures. 206219d632cSMatthew Dillon * The first available kernel virtual address is in "v". 207219d632cSMatthew Dillon * As pages of kernel virtual memory are allocated, "v" is incremented. 208219d632cSMatthew Dillon * As pages of memory are allocated and cleared, 209219d632cSMatthew Dillon * "firstaddr" is incremented. 210219d632cSMatthew Dillon */ 211219d632cSMatthew Dillon 212219d632cSMatthew Dillon /* 213219d632cSMatthew Dillon * Make two passes. The first pass calculates how much memory is 214219d632cSMatthew Dillon * needed and allocates it. The second pass assigns virtual 215219d632cSMatthew Dillon * addresses to the various data structures. 216219d632cSMatthew Dillon */ 217219d632cSMatthew Dillon firstaddr = 0; 218219d632cSMatthew Dillon again: 219219d632cSMatthew Dillon v = (caddr_t)firstaddr; 220219d632cSMatthew Dillon 221219d632cSMatthew Dillon /* 222219d632cSMatthew Dillon * Discount the physical memory larger than the size of kernel_map 223219d632cSMatthew Dillon * to avoid eating up all of KVA space. 224219d632cSMatthew Dillon */ 225447b3772SPeter Wemm physmem_est = lmin(physmem, btoc(kernel_map->max_offset - 226219d632cSMatthew Dillon kernel_map->min_offset)); 227219d632cSMatthew Dillon 228219d632cSMatthew Dillon v = kern_vfs_bio_buffer_alloc(v, physmem_est); 229219d632cSMatthew Dillon 230219d632cSMatthew Dillon /* 231219d632cSMatthew Dillon * End of first pass, size has been calculated so allocate memory 232219d632cSMatthew Dillon */ 233219d632cSMatthew Dillon if (firstaddr == 0) { 234857961d9SRobert Drehmel size = (vm_size_t)v; 235edb2994aSAndrew Gallatin #ifdef VM_FREELIST_DMA32 236edb2994aSAndrew Gallatin /* 237edb2994aSAndrew Gallatin * Try to protect 32-bit DMAable memory from the largest 238edb2994aSAndrew Gallatin * early alloc of wired mem. 239edb2994aSAndrew Gallatin */ 240edb2994aSAndrew Gallatin firstaddr = kmem_alloc_attr(kernel_arena, size, 241edb2994aSAndrew Gallatin M_ZERO | M_NOWAIT, (vm_paddr_t)1 << 32, 242edb2994aSAndrew Gallatin ~(vm_paddr_t)0, VM_MEMATTR_DEFAULT); 243edb2994aSAndrew Gallatin if (firstaddr == 0) 244edb2994aSAndrew Gallatin #endif 245edb2994aSAndrew Gallatin firstaddr = kmem_malloc(kernel_arena, size, 2465df87b21SJeff Roberson M_ZERO | M_WAITOK); 247219d632cSMatthew Dillon if (firstaddr == 0) 248219d632cSMatthew Dillon panic("startup: no room for tables"); 249219d632cSMatthew Dillon goto again; 250219d632cSMatthew Dillon } 251219d632cSMatthew Dillon 252219d632cSMatthew Dillon /* 253219d632cSMatthew Dillon * End of second pass, addresses have been assigned 254219d632cSMatthew Dillon */ 255219d632cSMatthew Dillon if ((vm_size_t)((char *)v - firstaddr) != size) 256219d632cSMatthew Dillon panic("startup: table size inconsistency"); 257219d632cSMatthew Dillon 2585df87b21SJeff Roberson /* 2595df87b21SJeff Roberson * Allocate the clean map to hold all of the paging and I/O virtual 2605df87b21SJeff Roberson * memory. 2615df87b21SJeff Roberson */ 2625f518366SJeff Roberson size = (long)nbuf * BKVASIZE + (long)nswbuf * MAXPHYS + 2635f518366SJeff Roberson (long)bio_transient_maxcnt * MAXPHYS; 2645df87b21SJeff Roberson kmi->clean_sva = firstaddr = kva_alloc(size); 2655df87b21SJeff Roberson kmi->clean_eva = firstaddr + size; 2665f518366SJeff Roberson 2675df87b21SJeff Roberson /* 2685df87b21SJeff Roberson * Allocate the buffer arena. 26921fae961SJeff Roberson * 27021fae961SJeff Roberson * Enable the quantum cache if we have more than 4 cpus. This 27121fae961SJeff Roberson * avoids lock contention at the expense of some fragmentation. 2725df87b21SJeff Roberson */ 2735f518366SJeff Roberson size = (long)nbuf * BKVASIZE; 2745df87b21SJeff Roberson kmi->buffer_sva = firstaddr; 2755f518366SJeff Roberson kmi->buffer_eva = kmi->buffer_sva + size; 2765f518366SJeff Roberson vmem_init(buffer_arena, "buffer arena", kmi->buffer_sva, size, 27721fae961SJeff Roberson PAGE_SIZE, (mp_ncpus > 4) ? BKVASIZE * 8 : 0, 0); 2785df87b21SJeff Roberson firstaddr += size; 2795f518366SJeff Roberson 2805df87b21SJeff Roberson /* 2815df87b21SJeff Roberson * Now swap kva. 2825df87b21SJeff Roberson */ 2835df87b21SJeff Roberson swapbkva = firstaddr; 2845f518366SJeff Roberson size = (long)nswbuf * MAXPHYS; 2855df87b21SJeff Roberson firstaddr += size; 2865f518366SJeff Roberson 2875df87b21SJeff Roberson /* 2885df87b21SJeff Roberson * And optionally transient bio space. 2895df87b21SJeff Roberson */ 2907db07e1cSKonstantin Belousov if (bio_transient_maxcnt != 0) { 2915f518366SJeff Roberson size = (long)bio_transient_maxcnt * MAXPHYS; 2925f518366SJeff Roberson vmem_init(transient_arena, "transient arena", 2935df87b21SJeff Roberson firstaddr, size, PAGE_SIZE, 0, 0); 2945df87b21SJeff Roberson firstaddr += size; 2957db07e1cSKonstantin Belousov } 2965df87b21SJeff Roberson if (firstaddr != kmi->clean_eva) 2975df87b21SJeff Roberson panic("Clean map calculation incorrect"); 2985df87b21SJeff Roberson 2995df87b21SJeff Roberson /* 300ec492b13SMark Johnston * Allocate the pageable submaps. We may cache an exec map entry per 301ec492b13SMark Johnston * CPU, so we therefore need to reserve space for at least ncpu+1 302ec492b13SMark Johnston * entries to avoid deadlock. The exec map is also used by some image 303ec492b13SMark Johnston * activators, so we leave a fixed number of pages for their use. 3045df87b21SJeff Roberson */ 305ec492b13SMark Johnston #ifdef __LP64__ 306ec492b13SMark Johnston exec_map_entries = 8 * mp_ncpus; 307ec492b13SMark Johnston #else 3088d65cba2SMark Johnston exec_map_entries = 2 * mp_ncpus + 4; 309ec492b13SMark Johnston #endif 310ec492b13SMark Johnston exec_map_entry_size = round_page(PATH_MAX + ARG_MAX); 311219d632cSMatthew Dillon exec_map = kmem_suballoc(kernel_map, &minaddr, &maxaddr, 312ec492b13SMark Johnston exec_map_entries * exec_map_entry_size + 64 * PAGE_SIZE, FALSE); 3133202ed75SAlan Cox pipe_map = kmem_suballoc(kernel_map, &minaddr, &maxaddr, maxpipekva, 3143202ed75SAlan Cox FALSE); 315219d632cSMatthew Dillon } 316