2 * This file is part of the Palacios Virtual Machine Monitor developed
3 * by the V3VEE Project with funding from the United States National
4 * Science Foundation and the Department of Energy.
6 * The V3VEE Project is a joint project between Northwestern University
7 * and the University of New Mexico. You can find out more at
10 * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu>
11 * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
12 * All rights reserved.
14 * Author: Jack Lange <jarusl@cs.northwestern.edu>
16 * This is free software. You are permitted to use,
17 * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
21 #ifndef __VMM_PAGING_H__
22 #define __VMM_PAGING_H__
27 #include <palacios/vmm_types.h>
28 #include <palacios/vmm_util.h>
33 In the following, when we say "page table", we mean the whole 2 or 4 layer
34 page table (PDEs, PTEs), etc.
37 guest-visible paging state
38 This is the state that the guest thinks the machine is using
40 - guest physical memory
41 The physical memory addresses the guest is allowed to use
42 (see shadow page maps, below)
44 (we care about when the current one changes)
45 - guest paging registers (these are never written to hardware)
51 This the state that the machine will actually use when the guest
52 is running. It consists of:
53 - current shadow page table
54 This is the page table actually useed when the guest is running.
55 It is changed/regenerated when the guest page table changes
56 It mostly reflects the guest page table, except that it restricts
57 physical addresses to those the VMM allocates to the guest.
59 This is a mapping from guest physical memory addresses to
60 the current location of the guest physical memory content.
61 It maps from regions of physical memory addresses to regions
62 located in physical memory or elsewhere.
63 (8192,16384) -> MEM(8912,...)
64 (0,8191) -> DISK(65536,..)
65 - guest paging registers (these are written to guest state)
70 This is the state we expect to be operative when the VMM is running.
71 Typically, this is set up by the host os into which we have embedded
72 the VMM, but we include the description here for clarity.
74 This is the page table we use when we are executing in
75 the VMM (or the host os)
81 The reason why the shadow paging state and the host paging state are
82 distinct is to permit the guest to use any virtual address it wants,
83 irrespective of the addresses the VMM or the host os use. These guest
84 virtual addresses are reflected in the shadow paging state. When we
85 exit from the guest, we switch to the host paging state so that any
86 virtual addresses that overlap between the guest and VMM/host now map
87 to the physical addresses epxected by the VMM/host. On AMD SVM, this
88 switch is done by the hardware. On Intel VT, the switch is done
89 by the hardware as well, but we are responsible for manually updating
90 the host state in the vmcs before entering the guest.
95 #define MAX_PDE32_ENTRIES 1024
96 #define MAX_PTE32_ENTRIES 1024
98 #define MAX_PDPE32PAE_ENTRIES 4
99 #define MAX_PDE32PAE_ENTRIES 512
100 #define MAX_PTE32PAE_ENTRIES 512
102 #define MAX_PML4E64_ENTRIES 512
103 #define MAX_PDPE64_ENTRIES 512
104 #define MAX_PDE64_ENTRIES 512
105 #define MAX_PTE64_ENTRIES 512
108 typedef enum {PAGE_4KB, PAGE_2MB, PAGE_4MB, PAGE_1GB,
110 PAGE_PT32, PAGE_PD32,
111 PAGE_PDP32PAE, PAGE_PD32PAE, PAGE_PT32PAE,
112 PAGE_PML464, PAGE_PDP64, PAGE_PD64, PAGE_PT64} page_type_t;
115 /* Converts an address into a page table index */
116 #define PDE32_INDEX(x) ((((uint_t)x) >> 22) & 0x3ff)
117 #define PTE32_INDEX(x) ((((uint_t)x) >> 12) & 0x3ff)
120 #define PDPE32PAE_INDEX(x) ((((uint_t)x) >> 30) & 0x3)
121 #define PDE32PAE_INDEX(x) ((((uint_t)x) >> 21) & 0x1ff)
122 #define PTE32PAE_INDEX(x) ((((uint_t)x) >> 12) & 0x1ff)
124 #define PML4E64_INDEX(x) ((((ullong_t)x) >> 39) & 0x1ff)
125 #define PDPE64_INDEX(x) ((((ullong_t)x) >> 30) & 0x1ff)
126 #define PDE64_INDEX(x) ((((ullong_t)x) >> 21) & 0x1ff)
127 #define PTE64_INDEX(x) ((((ullong_t)x) >> 12) & 0x1ff)
130 /* Gets the base address needed for a Page Table entry */
131 #define PAGE_BASE_ADDR(x) ((x) >> 12)
132 #define PAGE_BASE_ADDR_4KB(x) ((x) >> 12)
133 #define PAGE_BASE_ADDR_2MB(x) ((x) >> 21)
134 #define PAGE_BASE_ADDR_4MB(x) ((x) >> 22)
135 #define PAGE_BASE_ADDR_1GB(x) ((x) >> 30)
137 #define BASE_TO_PAGE_ADDR(x) (((addr_t)x) << 12)
138 #define BASE_TO_PAGE_ADDR_4KB(x) (((addr_t)x) << 12)
139 #define BASE_TO_PAGE_ADDR_2MB(x) (((addr_t)x) << 21)
140 #define BASE_TO_PAGE_ADDR_4MB(x) (((addr_t)x) << 22)
141 #define BASE_TO_PAGE_ADDR_1GB(x) (((addr_t)x) << 30)
145 #define PAGE_OFFSET(x) ((x) & 0xfff)
146 #define PAGE_OFFSET_4KB(x) ((x) & 0xfff)
147 #define PAGE_OFFSET_2MB(x) ((x) & 0x1fffff)
148 #define PAGE_OFFSET_4MB(x) ((x) & 0x3fffff)
149 #define PAGE_OFFSET_1GB(x) ((x) & 0x3fffffff)
151 #define PAGE_POWER 12
152 #define PAGE_POWER_4KB 12
153 #define PAGE_POWER_2MB 21
154 #define PAGE_POWER_4MB 22
155 #define PAGE_POWER_1GB 30
157 // We shift instead of mask because we don't know the address size
158 #define PAGE_ADDR(x) (((x) >> PAGE_POWER) << PAGE_POWER)
159 #define PAGE_ADDR_4KB(x) (((x) >> PAGE_POWER_4KB) << PAGE_POWER_4KB)
160 #define PAGE_ADDR_2MB(x) (((x) >> PAGE_POWER_2MB) << PAGE_POWER_2MB)
161 #define PAGE_ADDR_4MB(x) (((x) >> PAGE_POWER_4MB) << PAGE_POWER_4MB)
162 #define PAGE_ADDR_1GB(x) (((x) >> PAGE_POWER_1GB) << PAGE_POWER_1GB)
164 #define PAGE_SIZE 4096
165 #define PAGE_SIZE_4KB 4096
166 #define PAGE_SIZE_2MB (4096 * 512)
167 #define PAGE_SIZE_4MB (4096 * 1024)
168 #define PAGE_SIZE_1GB 0x40000000
176 #define CR3_TO_PDE32_PA(cr3) ((addr_t)(((uint_t)cr3) & 0xfffff000))
177 #define CR3_TO_PDPE32PAE_PA(cr3) ((addr_t)(((uint_t)cr3) & 0xffffffe0))
178 #define CR3_TO_PML4E64_PA(cr3) ((addr_t)(((ullong_t)cr3) & 0x000ffffffffff000LL))
180 #define CR3_TO_PDE32_VA(cr3) ((pde32_t *)V3_VAddr((void *)(addr_t)(((uint_t)cr3) & 0xfffff000)))
181 #define CR3_TO_PDPE32PAE_VA(cr3) ((pdpe32pae_t *)V3_VAddr((void *)(addr_t)(((uint_t)cr3) & 0xffffffe0)))
182 #define CR3_TO_PML4E64_VA(cr3) ((pml4e64_t *)V3_VAddr((void *)(addr_t)(((ullong_t)cr3) & 0x000ffffffffff000LL)))
188 /* We'll use the general form for now....
189 typedef enum {PDE32_ENTRY_NOT_PRESENT, PDE32_ENTRY_PTE32, PDE32_ENTRY_LARGE_PAGE} pde32_entry_type_t;
190 typedef enum {PTE32_ENTRY_NOT_PRESENT, PTE32_ENTRY_PAGE} pte32_entry_type_t;
192 typedef enum {PDPE32PAE_ENTRY_NOT_PRESENT, PDPE32PAE_ENTRY_PAGE} pdpe32pae_entry_type_t;
193 typedef enum {PDE32PAE_ENTRY_NOT_PRESENT, PDE32PAE_ENTRY_PTE32, PDE32PAE_ENTRY_LARGE_PAGE} pde32pae_entry_type_t;
194 typedef enum {PTE32PAE_ENTRY_NOT_PRESENT, PTE32PAE_ENTRY_PAGE} pte32pae_entry_type_t;
196 typedef enum {PML4E64_ENTRY_NOT_PRESENT, PML4E64_ENTRY_PAGE} pml4e64_entry_type_t;
197 typedef enum {PDPE64_ENTRY_NOT_PRESENT, PDPE64_ENTRY_PTE32, PDPE64_ENTRY_LARGE_PAGE} pdpe64_entry_type_t;
198 typedef enum {PDE64_ENTRY_NOT_PRESENT, PDE64_ENTRY_PTE32, PDE64_ENTRY_LARGE_PAGE} pde64_entry_type_t;
199 typedef enum {PTE64_ENTRY_NOT_PRESENT, PTE64_ENTRY_PAGE} pte64_entry_type_t;
203 typedef enum {PT_ENTRY_NOT_PRESENT, PT_ENTRY_LARGE_PAGE, PT_ENTRY_PAGE} pt_entry_type_t;
205 typedef enum {PT_ACCESS_OK, PT_ACCESS_NOT_PRESENT, PT_ACCESS_WRITE_ERROR, PT_ACCESS_USER_ERROR} pt_access_status_t;
207 /* Page table flag values */
208 #define V3_LARGE_PG 0x2
211 typedef struct gen_pt {
214 uint_t user_page : 1;
215 } __attribute__((packed)) gen_pt_t;
217 typedef struct pde32 {
220 uint_t user_page : 1;
221 uint_t write_through : 1;
222 uint_t cache_disable : 1;
225 uint_t large_page : 1;
226 uint_t global_page : 1;
228 uint_t pt_base_addr : 20;
229 } __attribute__((packed)) pde32_t;
231 typedef struct pde32_4MB {
234 uint_t user_page : 1;
235 uint_t write_through : 1;
236 uint_t cache_disable : 1;
239 uint_t large_page : 1;
240 uint_t global_page : 1;
244 uint_t page_base_addr : 10;
246 } __attribute__((packed)) pde32_4MB_t;
248 typedef struct pte32 {
251 uint_t user_page : 1;
252 uint_t write_through : 1;
253 uint_t cache_disable : 1;
257 uint_t global_page : 1;
259 uint_t page_base_addr : 20;
260 } __attribute__((packed)) pte32_t;
263 /* 32 bit PAE PAGE STRUCTURES */
264 typedef struct pdpe32pae {
266 uint_t rsvd : 2; // MBZ
267 uint_t write_through : 1;
268 uint_t cache_disable : 1;
271 uint_t rsvd2 : 2; // MBZ
273 uint_t pd_base_addr : 24;
274 uint_t rsvd3 : 28; // MBZ
275 } __attribute__((packed)) pdpe32pae_t;
279 typedef struct pde32pae {
282 uint_t user_page : 1;
283 uint_t write_through : 1;
284 uint_t cache_disable : 1;
287 uint_t large_page : 1;
288 uint_t global_page : 1;
290 uint_t pt_base_addr : 24;
292 } __attribute__((packed)) pde32pae_t;
294 typedef struct pde32pae_2MB {
297 uint_t user_page : 1;
298 uint_t write_through : 1;
299 uint_t cache_disable : 1;
303 uint_t global_page : 1;
307 uint_t page_base_addr : 15;
310 } __attribute__((packed)) pde32pae_2MB_t;
312 typedef struct pte32pae {
315 uint_t user_page : 1;
316 uint_t write_through : 1;
317 uint_t cache_disable : 1;
321 uint_t global_page : 1;
323 uint_t page_base_addr : 24;
325 } __attribute__((packed)) pte32pae_t;
334 /* LONG MODE 64 bit PAGE STRUCTURES */
335 typedef struct pml4e64 {
338 uint_t user_page : 1;
339 uint_t write_through : 1;
340 uint_t cache_disable : 1;
345 ullong_t pdp_base_addr : 40;
346 uint_t available : 11;
347 uint_t no_execute : 1;
348 } __attribute__((packed)) pml4e64_t;
351 typedef struct pdpe64 {
354 uint_t user_page : 1;
355 uint_t write_through : 1;
356 uint_t cache_disable : 1;
359 uint_t large_page : 1;
362 ullong_t pd_base_addr : 40;
363 uint_t available : 11;
364 uint_t no_execute : 1;
365 } __attribute__((packed)) pdpe64_t;
368 // We Don't support this
369 typedef struct pdpe64_1GB {
372 uint_t user_page : 1;
373 uint_t write_through : 1;
374 uint_t cache_disable : 1;
377 uint_t large_page : 1;
378 uint_t global_page : 1;
382 ullong_t page_base_addr : 22;
383 uint_t available : 11;
384 uint_t no_execute : 1;
385 } __attribute__((packed)) pdpe64_1GB_t;
389 typedef struct pde64 {
392 uint_t user_page : 1;
393 uint_t write_through : 1;
394 uint_t cache_disable : 1;
397 uint_t large_page : 1;
398 uint_t global_page : 1;
400 ullong_t pt_base_addr : 40;
401 uint_t available : 11;
402 uint_t no_execute : 1;
403 } __attribute__((packed)) pde64_t;
405 typedef struct pde64_2MB {
408 uint_t user_page : 1;
409 uint_t write_through : 1;
410 uint_t cache_disable : 1;
413 uint_t large_page : 1;
414 uint_t global_page : 1;
418 ullong_t page_base_addr : 31;
419 uint_t available : 11;
420 uint_t no_execute : 1;
421 } __attribute__((packed)) pde64_2MB_t;
424 typedef struct pte64 {
427 uint_t user_page : 1;
428 uint_t write_through : 1;
429 uint_t cache_disable : 1;
433 uint_t global_page : 1;
435 ullong_t page_base_addr : 40;
436 uint_t available : 11;
437 uint_t no_execute : 1;
438 } __attribute__((packed)) pte64_t;
440 /* *************** */
442 typedef struct pf_error_code {
443 uint_t present : 1; // if 0, fault due to page not present
444 uint_t write : 1; // if 1, faulting access was a write
445 uint_t user : 1; // if 1, faulting access was in user mode
446 uint_t rsvd_access : 1; // if 1, fault from reading a 1 from a reserved field (?)
447 uint_t ifetch : 1; // if 1, faulting access was an instr fetch (only with NX)
449 } __attribute__((packed)) pf_error_t;
457 int v3_translate_guest_pt_32(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr, addr_t * paddr);
458 int v3_translate_guest_pt_32pae(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr, addr_t * paddr);
459 int v3_translate_guest_pt_64(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr, addr_t * paddr);
461 int v3_translate_host_pt_32(struct guest_info * info, v3_reg_t host_cr3, addr_t vaddr, addr_t * paddr);
462 int v3_translate_host_pt_32pae(struct guest_info * info, v3_reg_t host_cr3, addr_t vaddr, addr_t * paddr);
463 int v3_translate_host_pt_64(struct guest_info * info, v3_reg_t host_cr3, addr_t vaddr, addr_t * paddr);
466 int v3_find_host_pt_32_page(struct guest_info * info, v3_reg_t host_cr3, page_type_t type, addr_t vaddr,
467 addr_t * page_ptr, addr_t * page_pa);
468 int v3_find_host_pt_32pae_page(struct guest_info * info, v3_reg_t host_cr3, page_type_t type, addr_t vaddr,
469 addr_t * page_ptr, addr_t * page_pa);
470 int v3_find_host_pt_64_page(struct guest_info * info, v3_reg_t host_cr3, page_type_t type, addr_t vaddr,
471 addr_t * page_ptr, addr_t * page_pa);
472 int v3_find_guest_pt_32_page(struct guest_info * info, v3_reg_t guest_cr3,
473 page_type_t type, addr_t vaddr,
474 addr_t * page_ptr, addr_t * page_pa);
475 int v3_find_guest_pt_32pae_page(struct guest_info * info, v3_reg_t guest_cr3,
476 page_type_t type, addr_t vaddr,
477 addr_t * page_ptr, addr_t * page_pa);
478 int v3_find_guest_pt_64_page(struct guest_info * info, v3_reg_t guest_cr3,
479 page_type_t type, addr_t vaddr,
480 addr_t * page_ptr, addr_t * page_pa);
484 pt_access_status_t inline v3_can_access_pde32(pde32_t * pde, addr_t addr, pf_error_t access_type);
485 pt_access_status_t inline v3_can_access_pte32(pte32_t * pte, addr_t addr, pf_error_t access_type);
487 pt_access_status_t inline v3_can_access_pdpe32pae(pdpe32pae_t * pdpe, addr_t addr, pf_error_t access_type);
488 pt_access_status_t inline v3_can_access_pde32pae(pde32pae_t * pde, addr_t addr, pf_error_t access_type);
489 pt_access_status_t inline v3_can_access_pte32pae(pte32pae_t * pte, addr_t addr, pf_error_t access_type);
491 pt_access_status_t inline v3_can_access_pml4e64(pml4e64_t * pmle, addr_t addr, pf_error_t access_type);
492 pt_access_status_t inline v3_can_access_pdpe64(pdpe64_t * pdpe, addr_t addr, pf_error_t access_type);
493 pt_access_status_t inline v3_can_access_pde64(pde64_t * pde, addr_t addr, pf_error_t access_type);
494 pt_access_status_t inline v3_can_access_pte64(pte64_t * pte, addr_t addr, pf_error_t access_type);
497 int v3_check_host_pt_32(struct guest_info * info, v3_reg_t host_cr3, addr_t vaddr,
498 pf_error_t access_type, pt_access_status_t * access_status);
499 int v3_check_host_pt_32pae(struct guest_info * info, v3_reg_t host_cr3, addr_t vaddr,
500 pf_error_t access_type, pt_access_status_t * access_status);
501 int v3_check_host_pt_64(struct guest_info * info, v3_reg_t host_cr3, addr_t vaddr,
502 pf_error_t access_type, pt_access_status_t * access_status);
503 int v3_check_guest_pt_32(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr,
504 pf_error_t access_type, pt_access_status_t * access_status);
505 int v3_check_guest_pt_32pae(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr,
506 pf_error_t access_type, pt_access_status_t * access_status);
507 int v3_check_guest_pt_64(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr,
508 pf_error_t access_type, pt_access_status_t * access_status);
512 page_type_t v3_get_guest_data_page_type_32(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr);
513 page_type_t v3_get_guest_data_page_type_32pae(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr);
514 page_type_t v3_get_guest_data_page_type_64(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr);
515 page_type_t v3_get_host_data_page_type_32(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr);
516 page_type_t v3_get_host_data_page_type_32pae(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr);
517 page_type_t v3_get_host_data_page_type_64(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr);
520 int v3_drill_host_pt_32(struct guest_info * info, v3_reg_t host_cr3, addr_t vaddr,
521 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_ptr, addr_t page_pa, void * private_data),
522 void * private_data);
523 int v3_drill_host_pt_32pae(struct guest_info * info, v3_reg_t host_cr3, addr_t vaddr,
524 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_ptr, addr_t page_pa, void * private_data),
525 void * private_data);
526 int v3_drill_host_pt_64(struct guest_info * info, v3_reg_t host_cr3, addr_t vaddr,
527 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_ptr, addr_t page_pa, void * private_data),
528 void * private_data);
530 int v3_drill_guest_pt_32(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr,
531 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_ptr, addr_t page_pa, void * private_data),
532 void * private_data);
533 int v3_drill_guest_pt_32pae(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr,
534 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_ptr, addr_t page_pa, void * private_data),
535 void * private_data);
536 int v3_drill_guest_pt_64(struct guest_info * info, v3_reg_t guest_cr3, addr_t vaddr,
537 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_ptr, addr_t page_pa, void * private_data),
538 void * private_data);
543 int v3_walk_host_pt_32(struct guest_info * info, v3_reg_t host_cr3,
544 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_va, addr_t page_pa, void * private_data),
545 void * private_data);
547 int v3_walk_host_pt_32pae(struct guest_info * info, v3_reg_t host_cr3,
548 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_va, addr_t page_pa, void * private_data),
549 void * private_data);
551 int v3_walk_host_pt_64(struct guest_info * info, v3_reg_t host_cr3,
552 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_va, addr_t page_pa, void * private_data),
553 void * private_data);
555 int v3_walk_guest_pt_32(struct guest_info * info, v3_reg_t guest_cr3,
556 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_va, addr_t page_pa, void * private_data),
557 void * private_data);
559 int v3_walk_guest_pt_32pae(struct guest_info * info, v3_reg_t guest_cr3,
560 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_va, addr_t page_pa, void * private_data),
561 void * private_data);
563 int v3_walk_guest_pt_64(struct guest_info * info, v3_reg_t guest_cr3,
564 int (*callback)(struct guest_info * info, page_type_t type, addr_t vaddr, addr_t page_va, addr_t page_pa, void * private_data),
565 void * private_data);
568 pde32_t * create_passthrough_pts_32(struct guest_info * guest_info);
569 pdpe32pae_t * create_passthrough_pts_32PAE(struct guest_info * guest_info);
570 pml4e64_t * create_passthrough_pts_64(struct guest_info * info);
573 void delete_page_tables_32(pde32_t * pde);
574 void delete_page_tables_32pae(pdpe32pae_t * pdpe);
575 void delete_page_tables_64(pml4e64_t * pml4);
579 const uchar_t * v3_page_type_to_str(page_type_t type);
583 void PrintPTEntry(struct guest_info * info, page_type_t type, addr_t vaddr, void * entry);
584 void PrintHostPageTables(struct guest_info * info, addr_t cr3);
585 void PrintGuestPageTables(struct guest_info * info, addr_t cr3);
586 void PrintHostPageTree(struct guest_info * info, addr_t virtual_addr, addr_t cr3);
587 void PrintGuestPageTree(struct guest_info * info, addr_t virtual_addr, addr_t cr3);