]> git.saurik.com Git - apple/xnu.git/blob - osfmk/i386/pmap.h
xnu-1504.15.3.tar.gz
[apple/xnu.git] / osfmk / i386 / pmap.h
1 /*
2 * Copyright (c) 2000-2007 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /*
29 * @OSF_COPYRIGHT@
30 */
31 /*
32 * Mach Operating System
33 * Copyright (c) 1991,1990,1989,1988 Carnegie Mellon University
34 * All Rights Reserved.
35 *
36 * Permission to use, copy, modify and distribute this software and its
37 * documentation is hereby granted, provided that both the copyright
38 * notice and this permission notice appear in all copies of the
39 * software, derivative works or modified versions, and any portions
40 * thereof, and that both notices appear in supporting documentation.
41 *
42 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
43 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
44 * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
45 *
46 * Carnegie Mellon requests users of this software to return to
47 *
48 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
49 * School of Computer Science
50 * Carnegie Mellon University
51 * Pittsburgh PA 15213-3890
52 *
53 * any improvements or extensions that they make and grant Carnegie Mellon
54 * the rights to redistribute these changes.
55 */
56 /*
57 */
58
59 /*
60 * File: pmap.h
61 *
62 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
63 * Date: 1985
64 *
65 * Machine-dependent structures for the physical map module.
66 */
67 #ifdef KERNEL_PRIVATE
68 #ifndef _PMAP_MACHINE_
69 #define _PMAP_MACHINE_ 1
70
71 #ifndef ASSEMBLER
72
73 #include <platforms.h>
74
75 #include <mach/kern_return.h>
76 #include <mach/machine/vm_types.h>
77 #include <mach/vm_prot.h>
78 #include <mach/vm_statistics.h>
79 #include <mach/machine/vm_param.h>
80 #include <kern/kern_types.h>
81 #include <kern/thread.h>
82 #include <kern/lock.h>
83
84 #include <i386/mp.h>
85 #include <i386/proc_reg.h>
86
87 /*
88 * Define the generic in terms of the specific
89 */
90
91 #define INTEL_PGBYTES I386_PGBYTES
92 #define INTEL_PGSHIFT I386_PGSHIFT
93 #define intel_btop(x) i386_btop(x)
94 #define intel_ptob(x) i386_ptob(x)
95 #define intel_round_page(x) i386_round_page(x)
96 #define intel_trunc_page(x) i386_trunc_page(x)
97 #define trunc_intel_to_vm(x) trunc_i386_to_vm(x)
98 #define round_intel_to_vm(x) round_i386_to_vm(x)
99 #define vm_to_intel(x) vm_to_i386(x)
100
101 /*
102 * i386/i486/i860 Page Table Entry
103 */
104
105 #endif /* ASSEMBLER */
106
107 #define NPGPTD 4
108 #define PDESHIFT 21
109 #define PTEMASK 0x1ff
110 #define PTEINDX 3
111
112 #define PTESHIFT 12
113
114
115 #define INITPT_SEG_BASE 0x100000
116 #define INITGDT_SEG_BASE 0x106000
117 #define SLEEP_SEG_BASE 0x107000
118
119 #ifdef __x86_64__
120 #define LOW_4GB_MASK ((vm_offset_t)0x00000000FFFFFFFFUL)
121 #endif
122
123 #define PDESIZE sizeof(pd_entry_t) /* for assembly files */
124 #define PTESIZE sizeof(pt_entry_t) /* for assembly files */
125
126 #define INTEL_OFFMASK (I386_PGBYTES - 1)
127 #define INTEL_LOFFMASK (I386_LPGBYTES - 1)
128 #define PG_FRAME 0x000FFFFFFFFFF000ULL
129 #define NPTEPG (PAGE_SIZE/(sizeof (pt_entry_t)))
130 #define NPTDPG (PAGE_SIZE/(sizeof (pd_entry_t)))
131
132 #define NBPTD (NPGPTD << PAGE_SHIFT)
133 #define NPDEPTD (NBPTD / (sizeof (pd_entry_t)))
134 #define NPDEPG (PAGE_SIZE/(sizeof (pd_entry_t)))
135 #define NBPDE (1 << PDESHIFT)
136 #define PDEMASK (NBPDE - 1)
137
138 #define PTE_PER_PAGE 512 /* number of PTE's per page on any level */
139
140 /* cleanly define parameters for all the page table levels */
141 typedef uint64_t pml4_entry_t;
142 #define NPML4PG (PAGE_SIZE/(sizeof (pml4_entry_t)))
143 #define PML4SHIFT 39
144 #define PML4PGSHIFT 9
145 #define NBPML4 (1ULL << PML4SHIFT)
146 #define PML4MASK (NBPML4-1)
147 #define PML4_ENTRY_NULL ((pml4_entry_t *) 0)
148
149 typedef uint64_t pdpt_entry_t;
150 #define NPDPTPG (PAGE_SIZE/(sizeof (pdpt_entry_t)))
151 #define PDPTSHIFT 30
152 #define PDPTPGSHIFT 9
153 #define NBPDPT (1 << PDPTSHIFT)
154 #define PDPTMASK (NBPDPT-1)
155 #define PDPT_ENTRY_NULL ((pdpt_entry_t *) 0)
156
157 typedef uint64_t pd_entry_t;
158 #define NPDPG (PAGE_SIZE/(sizeof (pd_entry_t)))
159 #define PDSHIFT 21
160 #define PDPGSHIFT 9
161 #define NBPD (1 << PDSHIFT)
162 #define PDMASK (NBPD-1)
163 #define PD_ENTRY_NULL ((pd_entry_t *) 0)
164
165 typedef uint64_t pt_entry_t;
166 #define NPTPG (PAGE_SIZE/(sizeof (pt_entry_t)))
167 #define PTSHIFT 12
168 #define PTPGSHIFT 9
169 #define NBPT (1 << PTSHIFT)
170 #define PTMASK (NBPT-1)
171 #define PT_ENTRY_NULL ((pt_entry_t *) 0)
172
173 typedef uint64_t pmap_paddr_t;
174
175 /* superpages */
176 #ifdef __x86_64__
177 #define SUPERPAGE_NBASEPAGES 512
178 #else
179 #define SUPERPAGE_NBASEPAGES 1 /* we don't support superpages on i386 */
180 #endif
181
182 /*
183 * Atomic 64-bit store of a page table entry.
184 */
185 static inline void
186 pmap_store_pte(pt_entry_t *entryp, pt_entry_t value)
187 {
188 #ifdef __i386__
189 /*
190 * Load the new value into %ecx:%ebx
191 * Load the old value into %edx:%eax
192 * Compare-exchange-8bytes at address entryp (loaded in %edi)
193 * If the compare succeeds, the new value will have been stored.
194 * Otherwise, the old value changed and reloaded, so try again.
195 */
196 __asm__ volatile(
197 " movl (%0), %%eax \n\t"
198 " movl 4(%0), %%edx \n\t"
199 "1: \n\t"
200 " cmpxchg8b (%0) \n\t"
201 " jnz 1b"
202 :
203 : "D" (entryp),
204 "b" ((uint32_t)value),
205 "c" ((uint32_t)(value >> 32))
206 : "eax", "edx", "memory");
207 #else
208 /*
209 * In the 32-bit kernel a compare-and-exchange loop was
210 * required to provide atomicity. For K64, life is easier:
211 */
212 *entryp = value;
213 #endif
214 }
215
216 /*
217 * Atomic 64-bit compare and exchange of a page table entry.
218 */
219 static inline boolean_t
220 pmap_cmpx_pte(pt_entry_t *entryp, pt_entry_t old, pt_entry_t new)
221 {
222 boolean_t ret;
223
224 #ifdef __i386__
225 /*
226 * Load the old value into %edx:%eax
227 * Load the new value into %ecx:%ebx
228 * Compare-exchange-8bytes at address entryp (loaded in %edi)
229 * If the compare succeeds, the new value is stored, return TRUE.
230 * Otherwise, no swap is made, return FALSE.
231 */
232 asm volatile(
233 " lock; cmpxchg8b (%1) \n\t"
234 " setz %%al \n\t"
235 " movzbl %%al,%0"
236 : "=a" (ret)
237 : "D" (entryp),
238 "a" ((uint32_t)old),
239 "d" ((uint32_t)(old >> 32)),
240 "b" ((uint32_t)new),
241 "c" ((uint32_t)(new >> 32))
242 : "memory");
243 #else
244 /*
245 * Load the old value into %rax
246 * Load the new value into another register
247 * Compare-exchange-quad at address entryp
248 * If the compare succeeds, the new value is stored, return TRUE.
249 * Otherwise, no swap is made, return FALSE.
250 */
251 asm volatile(
252 " lock; cmpxchgq %2,(%3) \n\t"
253 " setz %%al \n\t"
254 " movzbl %%al,%0"
255 : "=a" (ret)
256 : "a" (old),
257 "r" (new),
258 "r" (entryp)
259 : "memory");
260 #endif
261 return ret;
262 }
263
264 #define pmap_update_pte(entryp, old, new) \
265 while (!pmap_cmpx_pte((entryp), (old), (new)))
266
267
268 /* in 64 bit spaces, the number of each type of page in the page tables */
269 #define NPML4PGS (1ULL * (PAGE_SIZE/(sizeof (pml4_entry_t))))
270 #define NPDPTPGS (NPML4PGS * (PAGE_SIZE/(sizeof (pdpt_entry_t))))
271 #define NPDEPGS (NPDPTPGS * (PAGE_SIZE/(sizeof (pd_entry_t))))
272 #define NPTEPGS (NPDEPGS * (PAGE_SIZE/(sizeof (pt_entry_t))))
273
274 #ifdef __i386__
275 /*
276 * The 64-bit kernel is remapped in uber-space which is at the base
277 * the highest 4th-level directory (KERNEL_UBER_PML4_INDEX). That is,
278 * 512GB from the top of virtual space (or zero).
279 */
280 #define KERNEL_UBER_PML4_INDEX 511
281 #define KERNEL_UBER_BASE (0ULL - NBPML4)
282 #define KERNEL_UBER_BASE_HI32 ((uint32_t)(KERNEL_UBER_BASE >> 32))
283 #else
284 #define KERNEL_PML4_INDEX 511
285 #define KERNEL_KEXTS_INDEX 510 /* Home of KEXTs - the basement */
286 #define KERNEL_PHYSMAP_INDEX 509 /* virtual to physical map */
287 #define KERNEL_BASE (0ULL - NBPML4)
288 #define KERNEL_BASEMENT (KERNEL_BASE - NBPML4)
289 #endif
290
291 #define VM_WIMG_COPYBACK VM_MEM_COHERENT
292 #define VM_WIMG_DEFAULT VM_MEM_COHERENT
293 /* ?? intel ?? */
294 #define VM_WIMG_IO (VM_MEM_COHERENT | \
295 VM_MEM_NOT_CACHEABLE | VM_MEM_GUARDED)
296 #define VM_WIMG_WTHRU (VM_MEM_WRITE_THROUGH | VM_MEM_COHERENT | VM_MEM_GUARDED)
297 /* write combining mode, aka store gather */
298 #define VM_WIMG_WCOMB (VM_MEM_NOT_CACHEABLE | VM_MEM_COHERENT)
299
300 /*
301 * Pte related macros
302 */
303 #ifdef __i386__
304 #define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<<PDESHIFT)|((pti)<<PTESHIFT)))
305 #define VADDR64(pmi, pdi, pti) ((vm_offset_t)(((pmi)<<PLM4SHIFT))((pdi)<<PDESHIFT)|((pti)<<PTESHIFT))
306 #else
307 #define KVADDR(pmi, pdpi, pdi, pti) \
308 ((vm_offset_t) \
309 ((uint64_t) -1 << 47) | \
310 ((uint64_t)(pmi) << PML4SHIFT) | \
311 ((uint64_t)(pdpi) << PDPTSHIFT) | \
312 ((uint64_t)(pdi) << PDESHIFT) | \
313 ((uint64_t)(pti) << PTESHIFT))
314 #endif
315
316 /*
317 * Size of Kernel address space. This is the number of page table pages
318 * (4MB each) to use for the kernel. 256 pages == 1 Gigabyte.
319 * This **MUST** be a multiple of 4 (eg: 252, 256, 260, etc).
320 */
321 #ifndef KVA_PAGES
322 #define KVA_PAGES 1024
323 #endif
324
325 #ifndef NKPT
326 #define NKPT 500 /* actual number of kernel page tables */
327 #endif
328 #ifndef NKPDE
329 #define NKPDE (KVA_PAGES - 1) /* addressable number of page tables/pde's */
330 #endif
331
332
333 #ifdef __i386__
334 enum high_cpu_types {
335 HIGH_CPU_ISS0,
336 HIGH_CPU_ISS1,
337 HIGH_CPU_DESC,
338 HIGH_CPU_LDT_BEGIN,
339 HIGH_CPU_LDT_END = HIGH_CPU_LDT_BEGIN + (LDTSZ / 512) - 1,
340 HIGH_CPU_END
341 };
342
343 enum high_fixed_addresses {
344 HIGH_FIXED_TRAMPS, /* must be first */
345 HIGH_FIXED_TRAMPS_END,
346 HIGH_FIXED_GDT,
347 HIGH_FIXED_IDT,
348 HIGH_FIXED_LDT_BEGIN,
349 HIGH_FIXED_LDT_END = HIGH_FIXED_LDT_BEGIN + (LDTSZ / 512) - 1,
350 HIGH_FIXED_KTSS,
351 HIGH_FIXED_DFTSS,
352 HIGH_FIXED_DBTSS,
353 HIGH_FIXED_CPUS_BEGIN,
354 HIGH_FIXED_CPUS_END = HIGH_FIXED_CPUS_BEGIN + (HIGH_CPU_END * MAX_CPUS) - 1,
355 };
356
357
358 /* XXX64 below PTDI values need cleanup */
359 /*
360 * The *PTDI values control the layout of virtual memory
361 *
362 */
363 #define KPTDI (0x000)/* start of kernel virtual pde's */
364 #define PTDPTDI (0x7F4) /* ptd entry that points to ptd! */
365 #define APTDPTDI (0x7F8) /* alt ptd entry that points to APTD */
366 #define UMAXPTDI (0x7F8) /* ptd entry for user space end */
367 #define UMAXPTEOFF (NPTEPG) /* pte entry for user space end */
368
369 #define KERNBASE VADDR(KPTDI,0)
370
371 /*
372 * Convert address offset to directory address
373 * containing the page table pointer - legacy
374 */
375 /*#define pmap_pde(m,v) (&((m)->dirbase[(vm_offset_t)(v) >> PDESHIFT]))*/
376
377 #define HIGH_MEM_BASE ((uint32_t)( -NBPDE) ) /* shared gdt etc seg addr */ /* XXX64 ?? */
378 #define pmap_index_to_virt(x) (HIGH_MEM_BASE | ((unsigned)(x) << PAGE_SHIFT))
379 #endif
380
381 /*
382 * Convert address offset to page descriptor index
383 */
384 #define pdptnum(pmap, a) (((vm_offset_t)(a) >> PDPTSHIFT) & PDPTMASK)
385 #define pdenum(pmap, a) (((vm_offset_t)(a) >> PDESHIFT) & PDEMASK)
386 #define PMAP_INVALID_PDPTNUM (~0ULL)
387
388 #ifdef __i386__
389 #define pdeidx(pmap, a) (((a) >> PDSHIFT) & ((1ULL<<(48 - PDSHIFT)) -1))
390 #define pdptidx(pmap, a) (((a) >> PDPTSHIFT) & ((1ULL<<(48 - PDPTSHIFT)) -1))
391 #define pml4idx(pmap, a) (((a) >> PML4SHIFT) & ((1ULL<<(48 - PML4SHIFT)) -1))
392 #else
393 #define VAMASK ((1ULL<<48)-1)
394 #define pml4idx(pmap, a) ((((a) & VAMASK) >> PML4SHIFT) & \
395 ((1ULL<<(48 - PML4SHIFT))-1))
396 #define pdptidx(pmap, a) ((((a) & PML4MASK) >> PDPTSHIFT) & \
397 ((1ULL<<(48 - PDPTSHIFT))-1))
398 #define pdeidx(pmap, a) ((((a) & PML4MASK) >> PDSHIFT) & \
399 ((1ULL<<(48 - PDSHIFT)) - 1))
400 #endif
401
402 /*
403 * Convert page descriptor index to user virtual address
404 */
405 #define pdetova(a) ((vm_offset_t)(a) << PDESHIFT)
406
407 /*
408 * Convert address offset to page table index
409 */
410 #define ptenum(a) (((vm_offset_t)(a) >> PTESHIFT) & PTEMASK)
411
412 /*
413 * Hardware pte bit definitions (to be used directly on the ptes
414 * without using the bit fields).
415 */
416
417 #define INTEL_PTE_VALID 0x00000001
418 #define INTEL_PTE_WRITE 0x00000002
419 #define INTEL_PTE_RW 0x00000002
420 #define INTEL_PTE_USER 0x00000004
421 #define INTEL_PTE_WTHRU 0x00000008
422 #define INTEL_PTE_NCACHE 0x00000010
423 #define INTEL_PTE_REF 0x00000020
424 #define INTEL_PTE_MOD 0x00000040
425 #define INTEL_PTE_PS 0x00000080
426 #define INTEL_PTE_PTA 0x00000080
427 #define INTEL_PTE_GLOBAL 0x00000100
428 #define INTEL_PTE_WIRED 0x00000200
429 #define INTEL_PDPTE_NESTED 0x00000400
430 #define INTEL_PTE_PFN PG_FRAME
431
432 #define INTEL_PTE_NX (1ULL << 63)
433
434 #define INTEL_PTE_INVALID 0
435 /* This is conservative, but suffices */
436 #define INTEL_PTE_RSVD ((1ULL << 8) | (1ULL << 9) | (1ULL << 10) | (1ULL << 11) | (0x1FFULL << 54))
437 #define pa_to_pte(a) ((a) & INTEL_PTE_PFN) /* XXX */
438 #define pte_to_pa(p) ((p) & INTEL_PTE_PFN) /* XXX */
439 #define pte_increment_pa(p) ((p) += INTEL_OFFMASK+1)
440
441 #define pte_kernel_rw(p) ((pt_entry_t)(pa_to_pte(p) | INTEL_PTE_VALID|INTEL_PTE_RW))
442 #define pte_kernel_ro(p) ((pt_entry_t)(pa_to_pte(p) | INTEL_PTE_VALID))
443 #define pte_user_rw(p) ((pt_entry)t)(pa_to_pte(p) | INTEL_PTE_VALID|INTEL_PTE_USER|INTEL_PTE_RW))
444 #define pte_user_ro(p) ((pt_entry_t)(pa_to_pte(p) | INTEL_PTE_VALID|INTEL_PTE_USER))
445
446 #define PMAP_DEFAULT_CACHE 0
447 #define PMAP_INHIBIT_CACHE 1
448 #define PMAP_GUARDED_CACHE 2
449 #define PMAP_ACTIVATE_CACHE 4
450 #define PMAP_NO_GUARD_CACHE 8
451
452
453 #ifndef ASSEMBLER
454
455 #include <sys/queue.h>
456
457 /*
458 * Address of current and alternate address space page table maps
459 * and directories.
460 */
461
462 #ifdef __i386__
463 extern pt_entry_t PTmap[], APTmap[], Upte;
464 extern pd_entry_t PTD[], APTD[], PTDpde[], APTDpde[], Upde;
465 extern pmap_paddr_t lo_kernel_cr3;
466 extern pdpt_entry_t *IdlePDPT64;
467 #else
468 extern pt_entry_t *PTmap;
469 #endif
470 extern boolean_t no_shared_cr3;
471 extern addr64_t kernel64_cr3;
472 extern pd_entry_t *IdlePTD; /* physical addr of "Idle" state PTD */
473 extern pdpt_entry_t IdlePDPT[];
474 extern pml4_entry_t IdlePML4[];
475
476 extern uint64_t pmap_pv_hashlist_walks;
477 extern uint64_t pmap_pv_hashlist_cnts;
478 extern uint32_t pmap_pv_hashlist_max;
479 extern uint32_t pmap_kernel_text_ps;
480
481 #ifdef __i386__
482 /*
483 * ** i386 **
484 * virtual address to page table entry and
485 * to physical address. Likewise for alternate address space.
486 * Note: these work recursively, thus vtopte of a pte will give
487 * the corresponding pde that in turn maps it.
488 */
489
490 #define vtopte(va) (PTmap + i386_btop((vm_offset_t)va))
491 #endif
492
493 #ifdef __x86_64__
494 #define ID_MAP_VTOP(x) ((void *)(((uint64_t)(x)) & LOW_4GB_MASK))
495
496 #define PHYSMAP_BASE KVADDR(KERNEL_PHYSMAP_INDEX,0,0,0)
497 #define PHYSMAP_PTOV(x) ((void *)(((uint64_t)(x)) + PHYSMAP_BASE))
498 #endif
499
500 typedef volatile long cpu_set; /* set of CPUs - must be <= 32 */
501 /* changed by other processors */
502 struct md_page {
503 int pv_list_count;
504 TAILQ_HEAD(,pv_entry) pv_list;
505 };
506
507 #include <vm/vm_page.h>
508
509 /*
510 * For each vm_page_t, there is a list of all currently
511 * valid virtual mappings of that page. An entry is
512 * a pv_entry_t; the list is the pv_table.
513 */
514
515 struct pmap {
516 pd_entry_t *dirbase; /* page directory pointer */
517 #ifdef __i386__
518 pmap_paddr_t pdirbase; /* phys. address of dirbase */
519 #endif
520 vm_object_t pm_obj; /* object to hold pde's */
521 int ref_count; /* reference count */
522 int nx_enabled;
523 task_map_t pm_task_map;
524 decl_simple_lock_data(,lock) /* lock on map */
525 struct pmap_statistics stats; /* map statistics */
526 #ifdef __i386__
527 vm_offset_t pm_hold; /* true pdpt zalloc addr */
528 #endif
529 pmap_paddr_t pm_cr3; /* physical addr */
530 pdpt_entry_t *pm_pdpt; /* KVA of 3rd level page */
531 pml4_entry_t *pm_pml4; /* VKA of top level */
532 vm_object_t pm_obj_pdpt; /* holds pdpt pages */
533 vm_object_t pm_obj_pml4; /* holds pml4 pages */
534 vm_object_t pm_obj_top; /* holds single top level page */
535 boolean_t pm_shared;
536 };
537
538
539 #if NCOPY_WINDOWS > 0
540 #define PMAP_PDPT_FIRST_WINDOW 0
541 #define PMAP_PDPT_NWINDOWS 4
542 #define PMAP_PDE_FIRST_WINDOW (PMAP_PDPT_NWINDOWS)
543 #define PMAP_PDE_NWINDOWS 4
544 #define PMAP_PTE_FIRST_WINDOW (PMAP_PDE_FIRST_WINDOW + PMAP_PDE_NWINDOWS)
545 #define PMAP_PTE_NWINDOWS 4
546
547 #define PMAP_NWINDOWS_FIRSTFREE (PMAP_PTE_FIRST_WINDOW + PMAP_PTE_NWINDOWS)
548 #define PMAP_WINDOW_SIZE 8
549 #define PMAP_NWINDOWS (PMAP_NWINDOWS_FIRSTFREE + PMAP_WINDOW_SIZE)
550
551 typedef struct {
552 pt_entry_t *prv_CMAP;
553 caddr_t prv_CADDR;
554 } mapwindow_t;
555
556 typedef struct cpu_pmap {
557 int pdpt_window_index;
558 int pde_window_index;
559 int pte_window_index;
560 mapwindow_t mapwindow[PMAP_NWINDOWS];
561 } cpu_pmap_t;
562
563
564 extern mapwindow_t *pmap_get_mapwindow(pt_entry_t pentry);
565 extern void pmap_put_mapwindow(mapwindow_t *map);
566 #endif
567
568 typedef struct pmap_memory_regions {
569 ppnum_t base;
570 ppnum_t end;
571 ppnum_t alloc;
572 uint32_t type;
573 } pmap_memory_region_t;
574
575 extern unsigned pmap_memory_region_count;
576 extern unsigned pmap_memory_region_current;
577
578 #define PMAP_MEMORY_REGIONS_SIZE 128
579
580 extern pmap_memory_region_t pmap_memory_regions[];
581
582 static inline void
583 set_dirbase(pmap_t tpmap, __unused thread_t thread) {
584 current_cpu_datap()->cpu_task_cr3 = tpmap->pm_cr3;
585 current_cpu_datap()->cpu_task_map = tpmap->pm_task_map;
586 #ifndef __i386__
587 /*
588 * Switch cr3 if necessary
589 * - unless running with no_shared_cr3 debugging mode
590 * and we're not on the kernel's cr3 (after pre-empted copyio)
591 */
592 if (!no_shared_cr3) {
593 if (get_cr3() != tpmap->pm_cr3)
594 set_cr3(tpmap->pm_cr3);
595 } else {
596 if (get_cr3() != current_cpu_datap()->cpu_kernel_cr3)
597 set_cr3(current_cpu_datap()->cpu_kernel_cr3);
598 }
599 #endif
600 }
601
602 /*
603 * External declarations for PMAP_ACTIVATE.
604 */
605
606 extern void process_pmap_updates(void);
607 extern void pmap_update_interrupt(void);
608
609 /*
610 * Machine dependent routines that are used only for i386/i486/i860.
611 */
612
613 extern addr64_t (kvtophys)(
614 vm_offset_t addr);
615
616 extern void pmap_expand(
617 pmap_t pmap,
618 vm_map_offset_t addr);
619
620 extern pt_entry_t *pmap_pte(
621 struct pmap *pmap,
622 vm_map_offset_t addr);
623
624 extern pd_entry_t *pmap_pde(
625 struct pmap *pmap,
626 vm_map_offset_t addr);
627
628 extern pd_entry_t *pmap64_pde(
629 struct pmap *pmap,
630 vm_map_offset_t addr);
631
632 extern pdpt_entry_t *pmap64_pdpt(
633 struct pmap *pmap,
634 vm_map_offset_t addr);
635
636 extern vm_offset_t pmap_map(
637 vm_offset_t virt,
638 vm_map_offset_t start,
639 vm_map_offset_t end,
640 vm_prot_t prot,
641 unsigned int flags);
642
643 extern vm_offset_t pmap_map_bd(
644 vm_offset_t virt,
645 vm_map_offset_t start,
646 vm_map_offset_t end,
647 vm_prot_t prot,
648 unsigned int flags);
649
650 extern void pmap_bootstrap(
651 vm_offset_t load_start,
652 boolean_t IA32e);
653
654 extern boolean_t pmap_valid_page(
655 ppnum_t pn);
656
657 extern int pmap_list_resident_pages(
658 struct pmap *pmap,
659 vm_offset_t *listp,
660 int space);
661 extern void x86_filter_TLB_coherency_interrupts(boolean_t);
662 #ifdef __i386__
663 extern void pmap_commpage32_init(
664 vm_offset_t kernel,
665 vm_offset_t user,
666 int count);
667 extern void pmap_commpage64_init(
668 vm_offset_t kernel,
669 vm_map_offset_t user,
670 int count);
671
672 #endif
673
674 #if NCOPY_WINDOWS > 0
675 extern struct cpu_pmap *pmap_cpu_alloc(
676 boolean_t is_boot_cpu);
677 extern void pmap_cpu_free(
678 struct cpu_pmap *cp);
679 #endif
680
681 extern void pmap_map_block(
682 pmap_t pmap,
683 addr64_t va,
684 ppnum_t pa,
685 uint32_t size,
686 vm_prot_t prot,
687 int attr,
688 unsigned int flags);
689
690 extern void invalidate_icache(vm_offset_t addr, unsigned cnt, int phys);
691 extern void flush_dcache(vm_offset_t addr, unsigned count, int phys);
692 extern ppnum_t pmap_find_phys(pmap_t map, addr64_t va);
693
694 extern void pmap_cpu_init(void);
695 extern void pmap_disable_NX(pmap_t pmap);
696 #ifdef __i386__
697 extern void pmap_set_4GB_pagezero(pmap_t pmap);
698 extern void pmap_clear_4GB_pagezero(pmap_t pmap);
699 extern void pmap_load_kernel_cr3(void);
700 extern vm_offset_t pmap_cpu_high_map_vaddr(int, enum high_cpu_types);
701 extern vm_offset_t pmap_high_map_vaddr(enum high_cpu_types);
702 extern vm_offset_t pmap_high_map(pt_entry_t, enum high_cpu_types);
703 extern vm_offset_t pmap_cpu_high_shared_remap(int, enum high_cpu_types, vm_offset_t, int);
704 extern vm_offset_t pmap_high_shared_remap(enum high_fixed_addresses, vm_offset_t, int);
705 #endif
706
707 extern void pt_fake_zone_info(int *, vm_size_t *, vm_size_t *, vm_size_t *, vm_size_t *, int *, int *);
708 extern void pmap_pagetable_corruption_msg_log(int (*)(const char * fmt, ...)__printflike(1,2));
709
710
711 /*
712 * Macros for speed.
713 */
714
715
716 #include <kern/spl.h>
717
718
719 #define PMAP_ACTIVATE_MAP(map, thread) { \
720 register pmap_t tpmap; \
721 \
722 tpmap = vm_map_pmap(map); \
723 set_dirbase(tpmap, thread); \
724 }
725
726 #ifdef __i386__
727 #define PMAP_DEACTIVATE_MAP(map, thread) \
728 if (vm_map_pmap(map)->pm_task_map == TASK_MAP_64BIT_SHARED) \
729 pmap_load_kernel_cr3();
730 #else
731 #define PMAP_DEACTIVATE_MAP(map, my_cpu)
732 #endif
733
734 #if defined(__i386__)
735
736 #define PMAP_SWITCH_CONTEXT(old_th, new_th, my_cpu) { \
737 spl_t spl; \
738 pt_entry_t *kpdp; \
739 pt_entry_t *updp; \
740 int i; \
741 int need_flush; \
742 \
743 need_flush = 0; \
744 spl = splhigh(); \
745 if ((old_th->map != new_th->map) || (new_th->task != old_th->task)) { \
746 PMAP_DEACTIVATE_MAP(old_th->map, old_th); \
747 PMAP_ACTIVATE_MAP(new_th->map, new_th); \
748 } \
749 kpdp = current_cpu_datap()->cpu_copywindow_pdp; \
750 for (i = 0; i < NCOPY_WINDOWS; i++) { \
751 if (new_th->machine.copy_window[i].user_base != (user_addr_t)-1) { \
752 updp = pmap_pde(new_th->map->pmap, \
753 new_th->machine.copy_window[i].user_base);\
754 pmap_store_pte(kpdp, updp ? *updp : 0); \
755 } \
756 kpdp++; \
757 } \
758 splx(spl); \
759 if (new_th->machine.copyio_state == WINDOWS_OPENED) \
760 need_flush = 1; \
761 else \
762 new_th->machine.copyio_state = WINDOWS_DIRTY; \
763 if (new_th->machine.physwindow_pte) { \
764 pmap_store_pte((current_cpu_datap()->cpu_physwindow_ptep), \
765 new_th->machine.physwindow_pte); \
766 if (need_flush == 0) \
767 invlpg((uintptr_t)current_cpu_datap()->cpu_physwindow_base);\
768 } \
769 if (need_flush) \
770 flush_tlb(); \
771 }
772
773 #else /* __x86_64__ */
774 #define PMAP_SWITCH_CONTEXT(old_th, new_th, my_cpu) { \
775 spl_t spl; \
776 \
777 spl = splhigh(); \
778 if (old_th->map != new_th->map) { \
779 PMAP_DEACTIVATE_MAP(old_th->map, old_th); \
780 PMAP_ACTIVATE_MAP(new_th->map, new_th); \
781 } \
782 splx(spl); \
783 }
784 #endif /* __i386__ */
785
786 #ifdef __i386__
787 #define PMAP_SWITCH_USER(th, new_map, my_cpu) { \
788 spl_t spl; \
789 \
790 spl = splhigh(); \
791 PMAP_DEACTIVATE_MAP(th->map, th); \
792 th->map = new_map; \
793 PMAP_ACTIVATE_MAP(th->map, th); \
794 splx(spl); \
795 inval_copy_windows(th); \
796 }
797 #else
798 #define PMAP_SWITCH_USER(th, new_map, my_cpu) { \
799 spl_t spl; \
800 \
801 spl = splhigh(); \
802 PMAP_DEACTIVATE_MAP(th->map, th); \
803 th->map = new_map; \
804 PMAP_ACTIVATE_MAP(th->map, th); \
805 splx(spl); \
806 }
807 #endif
808
809 /*
810 * Marking the current cpu's cr3 inactive is achieved by setting its lsb.
811 * Marking the current cpu's cr3 active once more involves clearng this bit.
812 * Note that valid page tables are page-aligned and so the bottom 12 bits
813 * are noramlly zero.
814 * We can only mark the current cpu active/inactive but we can test any cpu.
815 */
816 #define CPU_CR3_MARK_INACTIVE() \
817 current_cpu_datap()->cpu_active_cr3 |= 1
818
819 #define CPU_CR3_MARK_ACTIVE() \
820 current_cpu_datap()->cpu_active_cr3 &= ~1
821
822 #define CPU_CR3_IS_ACTIVE(cpu) \
823 ((cpu_datap(cpu)->cpu_active_cr3 & 1) == 0)
824
825 #define CPU_GET_ACTIVE_CR3(cpu) \
826 (cpu_datap(cpu)->cpu_active_cr3 & ~1)
827
828 #define CPU_GET_TASK_CR3(cpu) \
829 (cpu_datap(cpu)->cpu_task_cr3)
830
831 /*
832 * Mark this cpu idle, and remove it from the active set,
833 * since it is not actively using any pmap. Signal_cpus
834 * will notice that it is idle, and avoid signaling it,
835 * but will queue the update request for when the cpu
836 * becomes active.
837 */
838 #if defined(__x86_64__)
839 #define MARK_CPU_IDLE(my_cpu) { \
840 int s = splhigh(); \
841 CPU_CR3_MARK_INACTIVE(); \
842 __asm__ volatile("mfence"); \
843 splx(s); \
844 }
845 #else /* __i386__ native */
846 #define MARK_CPU_IDLE(my_cpu) { \
847 /* \
848 * Mark this cpu idle, and remove it from the active set, \
849 * since it is not actively using any pmap. Signal_cpus \
850 * will notice that it is idle, and avoid signaling it, \
851 * but will queue the update request for when the cpu \
852 * becomes active. \
853 */ \
854 int s = splhigh(); \
855 if (!cpu_mode_is64bit() || no_shared_cr3) \
856 process_pmap_updates(); \
857 else \
858 pmap_load_kernel_cr3(); \
859 CPU_CR3_MARK_INACTIVE(); \
860 __asm__ volatile("mfence"); \
861 splx(s); \
862 }
863 #endif /* __i386__ */
864
865 #define MARK_CPU_ACTIVE(my_cpu) { \
866 \
867 int s = splhigh(); \
868 /* \
869 * If a kernel_pmap update was requested while this cpu \
870 * was idle, process it as if we got the interrupt. \
871 * Before doing so, remove this cpu from the idle set. \
872 * Since we do not grab any pmap locks while we flush \
873 * our TLB, another cpu may start an update operation \
874 * before we finish. Removing this cpu from the idle \
875 * set assures that we will receive another update \
876 * interrupt if this happens. \
877 */ \
878 CPU_CR3_MARK_ACTIVE(); \
879 __asm__ volatile("mfence"); \
880 \
881 if (current_cpu_datap()->cpu_tlb_invalid) \
882 process_pmap_updates(); \
883 splx(s); \
884 }
885
886 #define PMAP_CONTEXT(pmap, thread)
887
888 #define pmap_kernel_va(VA) \
889 ((((vm_offset_t) (VA)) >= vm_min_kernel_address) && \
890 (((vm_offset_t) (VA)) <= vm_max_kernel_address))
891
892
893 #define pmap_resident_count(pmap) ((pmap)->stats.resident_count)
894 #define pmap_resident_max(pmap) ((pmap)->stats.resident_max)
895 #define pmap_copy(dst_pmap,src_pmap,dst_addr,len,src_addr)
896 #define pmap_attribute(pmap,addr,size,attr,value) \
897 (KERN_INVALID_ADDRESS)
898 #define pmap_attribute_cache_sync(addr,size,attr,value) \
899 (KERN_INVALID_ADDRESS)
900
901 #define MACHINE_PMAP_IS_EMPTY 1
902 extern boolean_t pmap_is_empty(pmap_t pmap,
903 vm_map_offset_t start,
904 vm_map_offset_t end);
905
906
907 #endif /* ASSEMBLER */
908
909
910 #endif /* _PMAP_MACHINE_ */
911
912
913 #endif /* KERNEL_PRIVATE */