pmap.h revision 282065
1/*- 2 * Copyright (c) 1991 Regents of the University of California. 3 * All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * the Systems Programming Group of the University of Utah Computer 7 * Science Department and William Jolitz of UUNET Technologies Inc. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * Derived from hp300 version by Mike Hibler, this version by William 34 * Jolitz uses a recursive map [a pde points to the page directory] to 35 * map the page tables using the pagetables themselves. This is done to 36 * reduce the impact on kernel virtual memory for lots of sparse address 37 * space, and to reduce the cost of memory to each process. 38 * 39 * from: hp300: @(#)pmap.h 7.2 (Berkeley) 12/16/90 40 * from: @(#)pmap.h 7.4 (Berkeley) 5/12/91 41 * $FreeBSD: stable/10/sys/i386/include/pmap.h 282065 2015-04-27 08:02:12Z kib $ 42 */ 43 44#ifndef _MACHINE_PMAP_H_ 45#define _MACHINE_PMAP_H_ 46 47/* 48 * Page-directory and page-table entries follow this format, with a few 49 * of the fields not present here and there, depending on a lot of things. 50 */ 51 /* ---- Intel Nomenclature ---- */ 52#define PG_V 0x001 /* P Valid */ 53#define PG_RW 0x002 /* R/W Read/Write */ 54#define PG_U 0x004 /* U/S User/Supervisor */ 55#define PG_NC_PWT 0x008 /* PWT Write through */ 56#define PG_NC_PCD 0x010 /* PCD Cache disable */ 57#define PG_A 0x020 /* A Accessed */ 58#define PG_M 0x040 /* D Dirty */ 59#define PG_PS 0x080 /* PS Page size (0=4k,1=4M) */ 60#define PG_PTE_PAT 0x080 /* PAT PAT index */ 61#define PG_G 0x100 /* G Global */ 62#define PG_AVAIL1 0x200 /* / Available for system */ 63#define PG_AVAIL2 0x400 /* < programmers use */ 64#define PG_AVAIL3 0x800 /* \ */ 65#define PG_PDE_PAT 0x1000 /* PAT PAT index */ 66#if defined(PAE) || defined(PAE_TABLES) 67#define PG_NX (1ull<<63) /* No-execute */ 68#endif 69 70 71/* Our various interpretations of the above */ 72#define PG_W PG_AVAIL1 /* "Wired" pseudoflag */ 73#define PG_MANAGED PG_AVAIL2 74#if defined(PAE) || defined(PAE_TABLES) 75#define PG_FRAME (0x000ffffffffff000ull) 76#define PG_PS_FRAME (0x000fffffffe00000ull) 77#else 78#define PG_FRAME (~PAGE_MASK) 79#define PG_PS_FRAME (0xffc00000) 80#endif 81#define PG_PROT (PG_RW|PG_U) /* all protection bits . */ 82#define PG_N (PG_NC_PWT|PG_NC_PCD) /* Non-cacheable */ 83 84/* Page level cache control fields used to determine the PAT type */ 85#define PG_PDE_CACHE (PG_PDE_PAT | PG_NC_PWT | PG_NC_PCD) 86#define PG_PTE_CACHE (PG_PTE_PAT | PG_NC_PWT | PG_NC_PCD) 87 88/* 89 * Promotion to a 2 or 4MB (PDE) page mapping requires that the corresponding 90 * 4KB (PTE) page mappings have identical settings for the following fields: 91 */ 92#define PG_PTE_PROMOTE (PG_MANAGED | PG_W | PG_G | PG_PTE_PAT | \ 93 PG_M | PG_A | PG_NC_PCD | PG_NC_PWT | PG_U | PG_RW | PG_V) 94 95/* 96 * Page Protection Exception bits 97 */ 98 99#define PGEX_P 0x01 /* Protection violation vs. not present */ 100#define PGEX_W 0x02 /* during a Write cycle */ 101#define PGEX_U 0x04 /* access from User mode (UPL) */ 102#define PGEX_RSV 0x08 /* reserved PTE field is non-zero */ 103#define PGEX_I 0x10 /* during an instruction fetch */ 104 105/* 106 * Size of Kernel address space. This is the number of page table pages 107 * (4MB each) to use for the kernel. 256 pages == 1 Gigabyte. 108 * This **MUST** be a multiple of 4 (eg: 252, 256, 260, etc). 109 * For PAE, the page table page unit size is 2MB. This means that 512 pages 110 * is 1 Gigabyte. Double everything. It must be a multiple of 8 for PAE. 111 */ 112#ifndef KVA_PAGES 113#if defined(PAE) || defined(PAE_TABLES) 114#define KVA_PAGES 512 115#else 116#define KVA_PAGES 256 117#endif 118#endif 119 120/* 121 * Pte related macros 122 */ 123#define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<<PDRSHIFT)|((pti)<<PAGE_SHIFT))) 124 125/* 126 * The initial number of kernel page table pages that are constructed 127 * by locore must be sufficient to map vm_page_array. That number can 128 * be calculated as follows: 129 * max_phys / PAGE_SIZE * sizeof(struct vm_page) / NBPDR 130 * PAE: max_phys 16G, sizeof(vm_page) 76, NBPDR 2M, 152 page table pages. 131 * PAE_TABLES: max_phys 4G, sizeof(vm_page) 68, NBPDR 2M, 36 page table pages. 132 * Non-PAE: max_phys 4G, sizeof(vm_page) 68, NBPDR 4M, 18 page table pages. 133 */ 134#ifndef NKPT 135#if defined(PAE) 136#define NKPT 240 137#elif defined(PAE_TABLES) 138#define NKPT 60 139#else 140#define NKPT 30 141#endif 142#endif 143 144#ifndef NKPDE 145#define NKPDE (KVA_PAGES) /* number of page tables/pde's */ 146#endif 147 148/* 149 * The *PTDI values control the layout of virtual memory 150 * 151 * XXX This works for now, but I am not real happy with it, I'll fix it 152 * right after I fix locore.s and the magic 28K hole 153 */ 154#define KPTDI (NPDEPTD-NKPDE) /* start of kernel virtual pde's */ 155#define PTDPTDI (KPTDI-NPGPTD) /* ptd entry that points to ptd! */ 156 157/* 158 * XXX doesn't really belong here I guess... 159 */ 160#define ISA_HOLE_START 0xa0000 161#define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START) 162 163#ifndef LOCORE 164 165#include <sys/queue.h> 166#include <sys/_cpuset.h> 167#include <sys/_lock.h> 168#include <sys/_mutex.h> 169 170#include <vm/_vm_radix.h> 171 172#if defined(PAE) || defined(PAE_TABLES) 173 174typedef uint64_t pdpt_entry_t; 175typedef uint64_t pd_entry_t; 176typedef uint64_t pt_entry_t; 177 178#define PTESHIFT (3) 179#define PDESHIFT (3) 180 181#else 182 183typedef uint32_t pd_entry_t; 184typedef uint32_t pt_entry_t; 185 186#define PTESHIFT (2) 187#define PDESHIFT (2) 188 189#endif 190 191/* 192 * Address of current address space page table maps and directories. 193 */ 194#ifdef _KERNEL 195extern pt_entry_t PTmap[]; 196extern pd_entry_t PTD[]; 197extern pd_entry_t PTDpde[]; 198 199#if defined(PAE) || defined(PAE_TABLES) 200extern pdpt_entry_t *IdlePDPT; 201#endif 202extern pd_entry_t *IdlePTD; /* physical address of "Idle" state directory */ 203 204/* 205 * Translate a virtual address to the kernel virtual address of its page table 206 * entry (PTE). This can be used recursively. If the address of a PTE as 207 * previously returned by this macro is itself given as the argument, then the 208 * address of the page directory entry (PDE) that maps the PTE will be 209 * returned. 210 * 211 * This macro may be used before pmap_bootstrap() is called. 212 */ 213#define vtopte(va) (PTmap + i386_btop(va)) 214 215/* 216 * Translate a virtual address to its physical address. 217 * 218 * This macro may be used before pmap_bootstrap() is called. 219 */ 220#define vtophys(va) pmap_kextract((vm_offset_t)(va)) 221 222#if defined(XEN) 223#include <sys/param.h> 224 225#include <xen/xen-os.h> 226 227#include <machine/xen/xenvar.h> 228#include <machine/xen/xenpmap.h> 229 230extern pt_entry_t pg_nx; 231 232#define PG_KERNEL (PG_V | PG_A | PG_RW | PG_M) 233 234#define MACH_TO_VM_PAGE(ma) PHYS_TO_VM_PAGE(xpmap_mtop((ma))) 235#define VM_PAGE_TO_MACH(m) xpmap_ptom(VM_PAGE_TO_PHYS((m))) 236 237#define VTOM(va) xpmap_ptom(VTOP(va)) 238 239static __inline vm_paddr_t 240pmap_kextract_ma(vm_offset_t va) 241{ 242 vm_paddr_t ma; 243 if ((ma = PTD[va >> PDRSHIFT]) & PG_PS) { 244 ma = (ma & ~(NBPDR - 1)) | (va & (NBPDR - 1)); 245 } else { 246 ma = (*vtopte(va) & PG_FRAME) | (va & PAGE_MASK); 247 } 248 return ma; 249} 250 251static __inline vm_paddr_t 252pmap_kextract(vm_offset_t va) 253{ 254 return xpmap_mtop(pmap_kextract_ma(va)); 255} 256#define vtomach(va) pmap_kextract_ma(((vm_offset_t) (va))) 257 258vm_paddr_t pmap_extract_ma(struct pmap *pmap, vm_offset_t va); 259 260void pmap_kenter_ma(vm_offset_t va, vm_paddr_t pa); 261void pmap_map_readonly(struct pmap *pmap, vm_offset_t va, int len); 262void pmap_map_readwrite(struct pmap *pmap, vm_offset_t va, int len); 263 264static __inline pt_entry_t 265pte_load_store(pt_entry_t *ptep, pt_entry_t v) 266{ 267 pt_entry_t r; 268 269 r = *ptep; 270 PT_SET_VA(ptep, v, TRUE); 271 return (r); 272} 273 274static __inline pt_entry_t 275pte_load_store_ma(pt_entry_t *ptep, pt_entry_t v) 276{ 277 pt_entry_t r; 278 279 r = *ptep; 280 PT_SET_VA_MA(ptep, v, TRUE); 281 return (r); 282} 283 284#define pte_load_clear(ptep) pte_load_store((ptep), (pt_entry_t)0ULL) 285 286#define pte_store(ptep, pte) pte_load_store((ptep), (pt_entry_t)pte) 287#define pte_store_ma(ptep, pte) pte_load_store_ma((ptep), (pt_entry_t)pte) 288#define pde_store_ma(ptep, pte) pte_load_store_ma((ptep), (pt_entry_t)pte) 289 290#elif !defined(XEN) 291 292/* 293 * KPTmap is a linear mapping of the kernel page table. It differs from the 294 * recursive mapping in two ways: (1) it only provides access to kernel page 295 * table pages, and not user page table pages, and (2) it provides access to 296 * a kernel page table page after the corresponding virtual addresses have 297 * been promoted to a 2/4MB page mapping. 298 * 299 * KPTmap is first initialized by locore to support just NPKT page table 300 * pages. Later, it is reinitialized by pmap_bootstrap() to allow for 301 * expansion of the kernel page table. 302 */ 303extern pt_entry_t *KPTmap; 304 305/* 306 * Extract from the kernel page table the physical address that is mapped by 307 * the given virtual address "va". 308 * 309 * This function may be used before pmap_bootstrap() is called. 310 */ 311static __inline vm_paddr_t 312pmap_kextract(vm_offset_t va) 313{ 314 vm_paddr_t pa; 315 316 if ((pa = PTD[va >> PDRSHIFT]) & PG_PS) { 317 pa = (pa & PG_PS_FRAME) | (va & PDRMASK); 318 } else { 319 /* 320 * Beware of a concurrent promotion that changes the PDE at 321 * this point! For example, vtopte() must not be used to 322 * access the PTE because it would use the new PDE. It is, 323 * however, safe to use the old PDE because the page table 324 * page is preserved by the promotion. 325 */ 326 pa = KPTmap[i386_btop(va)]; 327 pa = (pa & PG_FRAME) | (va & PAGE_MASK); 328 } 329 return (pa); 330} 331#endif 332 333#if !defined(XEN) 334#define PT_UPDATES_FLUSH() 335#endif 336 337#if (defined(PAE) || defined(PAE_TABLES)) && !defined(XEN) 338 339#define pde_cmpset(pdep, old, new) atomic_cmpset_64_i586(pdep, old, new) 340#define pte_load_store(ptep, pte) atomic_swap_64_i586(ptep, pte) 341#define pte_load_clear(ptep) atomic_swap_64_i586(ptep, 0) 342#define pte_store(ptep, pte) atomic_store_rel_64_i586(ptep, pte) 343 344extern pt_entry_t pg_nx; 345 346#elif !defined(PAE) && !defined(PAE_TABLES) && !defined(XEN) 347 348#define pde_cmpset(pdep, old, new) atomic_cmpset_int(pdep, old, new) 349#define pte_load_store(ptep, pte) atomic_swap_int(ptep, pte) 350#define pte_load_clear(ptep) atomic_swap_int(ptep, 0) 351#define pte_store(ptep, pte) do { \ 352 *(u_int *)(ptep) = (u_int)(pte); \ 353} while (0) 354 355#endif /* PAE */ 356 357#define pte_clear(ptep) pte_store(ptep, 0) 358 359#define pde_store(pdep, pde) pte_store(pdep, pde) 360 361#endif /* _KERNEL */ 362 363/* 364 * Pmap stuff 365 */ 366struct pv_entry; 367struct pv_chunk; 368 369struct md_page { 370 TAILQ_HEAD(,pv_entry) pv_list; 371 int pat_mode; 372}; 373 374struct pmap { 375 struct mtx pm_mtx; 376 pd_entry_t *pm_pdir; /* KVA of page directory */ 377 TAILQ_HEAD(,pv_chunk) pm_pvchunk; /* list of mappings in pmap */ 378 cpuset_t pm_active; /* active on cpus */ 379 struct pmap_statistics pm_stats; /* pmap statistics */ 380 LIST_ENTRY(pmap) pm_list; /* List of all pmaps */ 381#if defined(PAE) || defined(PAE_TABLES) 382 pdpt_entry_t *pm_pdpt; /* KVA of page directory pointer 383 table */ 384#endif 385 struct vm_radix pm_root; /* spare page table pages */ 386}; 387 388typedef struct pmap *pmap_t; 389 390#ifdef _KERNEL 391extern struct pmap kernel_pmap_store; 392#define kernel_pmap (&kernel_pmap_store) 393 394#define PMAP_LOCK(pmap) mtx_lock(&(pmap)->pm_mtx) 395#define PMAP_LOCK_ASSERT(pmap, type) \ 396 mtx_assert(&(pmap)->pm_mtx, (type)) 397#define PMAP_LOCK_DESTROY(pmap) mtx_destroy(&(pmap)->pm_mtx) 398#define PMAP_LOCK_INIT(pmap) mtx_init(&(pmap)->pm_mtx, "pmap", \ 399 NULL, MTX_DEF | MTX_DUPOK) 400#define PMAP_LOCKED(pmap) mtx_owned(&(pmap)->pm_mtx) 401#define PMAP_MTX(pmap) (&(pmap)->pm_mtx) 402#define PMAP_TRYLOCK(pmap) mtx_trylock(&(pmap)->pm_mtx) 403#define PMAP_UNLOCK(pmap) mtx_unlock(&(pmap)->pm_mtx) 404#endif 405 406/* 407 * For each vm_page_t, there is a list of all currently valid virtual 408 * mappings of that page. An entry is a pv_entry_t, the list is pv_list. 409 */ 410typedef struct pv_entry { 411 vm_offset_t pv_va; /* virtual address for mapping */ 412 TAILQ_ENTRY(pv_entry) pv_next; 413} *pv_entry_t; 414 415/* 416 * pv_entries are allocated in chunks per-process. This avoids the 417 * need to track per-pmap assignments. 418 */ 419#define _NPCM 11 420#define _NPCPV 336 421struct pv_chunk { 422 pmap_t pc_pmap; 423 TAILQ_ENTRY(pv_chunk) pc_list; 424 uint32_t pc_map[_NPCM]; /* bitmap; 1 = free */ 425 TAILQ_ENTRY(pv_chunk) pc_lru; 426 struct pv_entry pc_pventry[_NPCPV]; 427}; 428 429#ifdef _KERNEL 430 431extern caddr_t CADDR3; 432extern pt_entry_t *CMAP3; 433extern vm_paddr_t phys_avail[]; 434extern vm_paddr_t dump_avail[]; 435extern int pseflag; 436extern int pgeflag; 437extern char *ptvmmap; /* poor name! */ 438extern vm_offset_t virtual_avail; 439extern vm_offset_t virtual_end; 440 441#define pmap_page_get_memattr(m) ((vm_memattr_t)(m)->md.pat_mode) 442#define pmap_page_is_write_mapped(m) (((m)->aflags & PGA_WRITEABLE) != 0) 443#define pmap_unmapbios(va, sz) pmap_unmapdev((va), (sz)) 444 445/* 446 * Only the following functions or macros may be used before pmap_bootstrap() 447 * is called: pmap_kenter(), pmap_kextract(), pmap_kremove(), vtophys(), and 448 * vtopte(). 449 */ 450void pmap_bootstrap(vm_paddr_t); 451int pmap_cache_bits(int mode, boolean_t is_pde); 452int pmap_change_attr(vm_offset_t, vm_size_t, int); 453void pmap_init_pat(void); 454void pmap_kenter(vm_offset_t va, vm_paddr_t pa); 455void *pmap_kenter_temporary(vm_paddr_t pa, int i); 456void pmap_kremove(vm_offset_t); 457void *pmap_mapbios(vm_paddr_t, vm_size_t); 458void *pmap_mapdev(vm_paddr_t, vm_size_t); 459void *pmap_mapdev_attr(vm_paddr_t, vm_size_t, int); 460boolean_t pmap_page_is_mapped(vm_page_t m); 461void pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma); 462void pmap_unmapdev(vm_offset_t, vm_size_t); 463pt_entry_t *pmap_pte(pmap_t, vm_offset_t) __pure2; 464void pmap_invalidate_page(pmap_t, vm_offset_t); 465void pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t); 466void pmap_invalidate_all(pmap_t); 467void pmap_invalidate_cache(void); 468void pmap_invalidate_cache_pages(vm_page_t *pages, int count); 469void pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva, 470 boolean_t force); 471 472#endif /* _KERNEL */ 473 474#endif /* !LOCORE */ 475 476#endif /* !_MACHINE_PMAP_H_ */ 477