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 #include <palacios/vmm_shadow_paging.h>
24 #include <palacios/vmm.h>
25 #include <palacios/vm_guest_mem.h>
26 #include <palacios/vmm_decoder.h>
27 #include <palacios/vmm_ctrl_regs.h>
29 #ifndef DEBUG_SHADOW_PAGING
31 #define PrintDebug(fmt, args...)
43 DEFINE_HASHTABLE_INSERT(add_cr3_to_cache, addr_t, struct hashtable *);
44 DEFINE_HASHTABLE_SEARCH(find_cr3_in_cache, addr_t, struct hashtable *);
45 DEFINE_HASHTABLE_REMOVE(del_cr3_from_cache, addr_t, struct hashtable *, 0);
48 DEFINE_HASHTABLE_INSERT(add_pte_map, addr_t, addr_t);
49 DEFINE_HASHTABLE_SEARCH(find_pte_map, addr_t, addr_t);
50 DEFINE_HASHTABLE_REMOVE(del_pte_map, addr_t, addr_t, 0);
54 static uint_t pte_hash_fn(addr_t key) {
55 return hash_long(key, 32);
58 static int pte_equals(addr_t key1, addr_t key2) {
59 return (key1 == key2);
62 static uint_t cr3_hash_fn(addr_t key) {
63 return hash_long(key, 32);
66 static int cr3_equals(addr_t key1, addr_t key2) {
67 return (key1 == key2);
72 static int activate_shadow_pt_32(struct guest_info * info);
73 static int activate_shadow_pt_32pae(struct guest_info * info);
74 static int activate_shadow_pt_64(struct guest_info * info);
77 static int handle_shadow_pte32_fault(struct guest_info* info,
79 pf_error_t error_code,
83 static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code);
85 int v3_init_shadow_page_state(struct guest_info * info) {
86 struct shadow_page_state * state = &(info->shdw_pg_state);
91 state->cr3_cache = create_hashtable(0, &cr3_hash_fn, &cr3_equals);
93 state->cached_cr3 = 0;
94 state->cached_ptes = NULL;
103 For now we'll do something a little more lightweight
104 int cache_page_tables32(struct guest_info * info, addr_t pde) {
105 struct shadow_page_state * state = &(info->shdw_pg_state);
106 addr_t pde_host_addr;
108 struct hashtable * pte_cache = NULL;
112 pte_cache = (struct hashtable *)find_cr3_in_cache(state->cr3_cache, pde);
113 if (pte_cache != NULL) {
114 PrintError("CR3 already present in cache\n");
115 state->current_ptes = pte_cache;
118 PrintError("Creating new CR3 cache entry\n");
119 pte_cache = create_hashtable(0, &pte_hash_fn, &pte_equals);
120 state->current_ptes = pte_cache;
121 add_cr3_to_cache(state->cr3_cache, pde, pte_cache);
124 if (guest_pa_to_host_va(info, pde, &pde_host_addr) == -1) {
125 PrintError("Could not lookup host address of guest PDE\n");
129 tmp_pde = (pde32_t *)pde_host_addr;
131 add_pte_map(pte_cache, pde, pde_host_addr);
134 for (i = 0; i < MAX_PDE32_ENTRIES; i++) {
135 if ((tmp_pde[i].present) && (tmp_pde[i].large_page == 0)) {
136 addr_t pte_host_addr;
138 if (guest_pa_to_host_va(info, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), &pte_host_addr) == -1) {
139 PrintError("Could not lookup host address of guest PDE\n");
143 add_pte_map(pte_cache, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), pte_host_addr);
153 int v3_cache_page_tables(struct guest_info * info, addr_t cr3) {
154 switch(v3_get_cpu_mode(info)) {
156 return v3_cache_page_tables32(info, CR3_TO_PDE32_PA(cr3));
162 int v3_cache_page_tables32(struct guest_info * info, addr_t pde) {
163 struct shadow_page_state * state = &(info->shdw_pg_state);
164 addr_t pde_host_addr;
166 struct hashtable * pte_cache = NULL;
169 if (pde == state->cached_cr3) {
173 if (state->cached_ptes != NULL) {
174 hashtable_destroy(state->cached_ptes, 0, 0);
175 state->cached_ptes = NULL;
178 state->cached_cr3 = pde;
180 pte_cache = create_hashtable(0, &pte_hash_fn, &pte_equals);
181 state->cached_ptes = pte_cache;
183 if (guest_pa_to_host_va(info, pde, &pde_host_addr) == -1) {
184 PrintError("Could not lookup host address of guest PDE\n");
188 tmp_pde = (pde32_t *)pde_host_addr;
190 add_pte_map(pte_cache, pde, pde_host_addr);
193 for (i = 0; i < MAX_PDE32_ENTRIES; i++) {
194 if ((tmp_pde[i].present) && (tmp_pde[i].large_page == 0)) {
195 addr_t pte_host_addr;
197 if (guest_pa_to_host_va(info, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), &pte_host_addr) == -1) {
198 PrintError("Could not lookup host address of guest PDE\n");
202 add_pte_map(pte_cache, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), pte_host_addr);
212 int v3_replace_shdw_page32(struct guest_info * info, addr_t location, pte32_t * new_page, pte32_t * old_page) {
213 pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32_VA(info->ctrl_regs.cr3);
214 pde32_t * shadow_pde = (pde32_t *)&(shadow_pd[PDE32_INDEX(location)]);
216 if (shadow_pde->large_page == 0) {
217 pte32_t * shadow_pt = (pte32_t *)(addr_t)PDE32_T_ADDR((*shadow_pde));
218 pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(location)]);
220 //if (shadow_pte->present == 1) {
221 *(uint_t *)old_page = *(uint_t *)shadow_pte;
224 *(uint_t *)shadow_pte = *(uint_t *)new_page;
227 // currently unhandled
238 static int activate_shadow_pt_32(struct guest_info * info) {
239 struct cr3_32 * shadow_cr3 = (struct cr3_32 *)&(info->ctrl_regs.cr3);
240 struct cr3_32 * guest_cr3 = (struct cr3_32 *)&(info->shdw_pg_state.guest_cr3);
243 // Check if shadow page tables are in the cache
244 cached = v3_cache_page_tables32(info, CR3_TO_PDE32_PA(*(addr_t *)guest_cr3));
247 PrintError("CR3 Cache failed\n");
249 } else if (cached == 0) {
252 PrintDebug("New CR3 is different - flushing shadow page table %p\n", shadow_cr3 );
253 delete_page_tables_32(CR3_TO_PDE32_VA(*(uint_t*)shadow_cr3));
255 shadow_pt = v3_create_new_shadow_pt();
257 shadow_cr3->pdt_base_addr = (addr_t)V3_PAddr((void *)(addr_t)PD32_BASE_ADDR(shadow_pt));
258 PrintDebug( "Created new shadow page table %p\n", (void *)(addr_t)shadow_cr3->pdt_base_addr );
260 PrintDebug("Reusing cached shadow Page table\n");
263 shadow_cr3->pwt = guest_cr3->pwt;
264 shadow_cr3->pcd = guest_cr3->pcd;
269 static int activate_shadow_pt_32pae(struct guest_info * info) {
270 PrintError("Activating 32 bit PAE page tables not implemented\n");
274 static int activate_shadow_pt_64(struct guest_info * info) {
275 PrintError("Activating 64 bit page tables not implemented\n");
280 // Reads the guest CR3 register
281 // creates new shadow page tables
282 // updates the shadow CR3 register to point to the new pts
283 int v3_activate_shadow_pt(struct guest_info * info) {
284 switch (info->cpu_mode) {
287 return activate_shadow_pt_32(info);
289 return activate_shadow_pt_32pae(info);
293 return activate_shadow_pt_64(info);
295 PrintError("Invalid CPU mode: %d\n", info->cpu_mode);
303 int v3_handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
305 if (info->mem_mode == PHYSICAL_MEM) {
306 // If paging is not turned on we need to handle the special cases
308 #ifdef DEBUG_SHADOW_PAGING
309 PrintPageTree(info->cpu_mode, fault_addr, info->ctrl_regs.cr3);
312 return handle_special_page_fault(info, fault_addr, fault_addr, error_code);
313 } else if (info->mem_mode == VIRTUAL_MEM) {
315 switch (info->cpu_mode) {
317 return handle_shadow_pagefault32(info, fault_addr, error_code);
322 PrintError("Unhandled CPU Mode\n");
326 PrintError("Invalid Memory mode\n");
331 addr_t v3_create_new_shadow_pt() {
334 host_pde = V3_VAddr(V3_AllocPages(1));
335 memset(host_pde, 0, PAGE_SIZE);
337 return (addr_t)host_pde;
341 static void inject_guest_pf(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
342 info->ctrl_regs.cr2 = fault_addr;
343 v3_raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code);
347 static int is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shadow_access) {
348 /* basically the reasoning is that there can be multiple reasons for a page fault:
349 If there is a permissions failure for a page present in the guest _BUT_
350 the reason for the fault was that the page is not present in the shadow,
351 _THEN_ we have to map the shadow page in and reexecute, this will generate
352 a permissions fault which is _THEN_ valid to send to the guest
353 _UNLESS_ both the guest and shadow have marked the page as not present
357 if (guest_access != PT_ACCESS_OK) {
358 // Guest Access Error
360 if ((shadow_access != PT_ENTRY_NOT_PRESENT) &&
361 (guest_access != PT_ENTRY_NOT_PRESENT)) {
362 // aka (guest permission error)
366 if ((shadow_access == PT_ENTRY_NOT_PRESENT) &&
367 (guest_access == PT_ENTRY_NOT_PRESENT)) {
368 // Page tables completely blank, handle guest first
372 // Otherwise we'll handle the guest fault later...?
381 /* The guest status checks have already been done,
382 * only special case shadow checks remain
384 static int handle_large_pagefault32(struct guest_info * info,
385 addr_t fault_addr, pf_error_t error_code,
386 pte32_t * shadow_pt, pde32_4MB_t * large_guest_pde)
388 pt_access_status_t shadow_pte_access = can_access_pte32(shadow_pt, fault_addr, error_code);
389 pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(fault_addr)]);
391 if (shadow_pte_access == PT_ACCESS_OK) {
392 // Inconsistent state...
393 // Guest Re-Entry will flush tables and everything should now workd
394 PrintDebug("Inconsistent state... Guest re-entry should flush tlb\n");
399 if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) {
400 // Get the guest physical address of the fault
401 addr_t guest_fault_pa = PDE32_4MB_T_ADDR(*large_guest_pde) + PD32_4MB_PAGE_OFFSET(fault_addr);
402 host_region_type_t host_page_type = get_shadow_addr_type(info, guest_fault_pa);
405 if (host_page_type == HOST_REGION_INVALID) {
406 // Inject a machine check in the guest
407 PrintDebug("Invalid Guest Address in page table (0x%p)\n", (void *)guest_fault_pa);
408 v3_raise_exception(info, MC_EXCEPTION);
412 if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) {
413 struct shadow_page_state * state = &(info->shdw_pg_state);
414 addr_t shadow_pa = get_shadow_addr(info, guest_fault_pa);
416 shadow_pte->page_base_addr = PT32_BASE_ADDR(shadow_pa);
418 shadow_pte->present = 1;
420 /* We are assuming that the PDE entry has precedence
421 * so the Shadow PDE will mirror the guest PDE settings,
422 * and we don't have to worry about them here
425 shadow_pte->user_page = 1;
427 if (find_pte_map(state->cached_ptes, PT32_PAGE_ADDR(guest_fault_pa)) != NULL) {
428 // Check if the entry is a page table...
429 PrintDebug("Marking page as Guest Page Table (large page)\n");
430 shadow_pte->vmm_info = PT32_GUEST_PT;
431 shadow_pte->writable = 0;
433 shadow_pte->writable = 1;
437 //set according to VMM policy
438 shadow_pte->write_through = 0;
439 shadow_pte->cache_disable = 0;
440 shadow_pte->global_page = 0;
444 // Handle hooked pages as well as other special pages
445 if (handle_special_page_fault(info, fault_addr, guest_fault_pa, error_code) == -1) {
446 PrintError("Special Page Fault handler returned error for address: %p\n", (void *)fault_addr);
450 } else if ((shadow_pte_access == PT_WRITE_ERROR) &&
451 (shadow_pte->vmm_info == PT32_GUEST_PT)) {
453 struct shadow_page_state * state = &(info->shdw_pg_state);
454 PrintDebug("Write operation on Guest PAge Table Page (large page)\n");
455 state->cached_cr3 = 0;
456 shadow_pte->writable = 1;
459 PrintError("Error in large page fault handler...\n");
460 PrintError("This case should have been handled at the top level handler\n");
464 PrintDebug("Returning from large page fault handler\n");
469 static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
470 pde32_t * guest_pd = NULL;
471 pde32_t * shadow_pd = CR3_TO_PDE32_VA(info->ctrl_regs.cr3);
472 addr_t guest_cr3 = CR3_TO_PDE32_PA(info->shdw_pg_state.guest_cr3);
473 pt_access_status_t guest_pde_access;
474 pt_access_status_t shadow_pde_access;
475 pde32_t * guest_pde = NULL;
476 pde32_t * shadow_pde = (pde32_t *)&(shadow_pd[PDE32_INDEX(fault_addr)]);
478 PrintDebug("Shadow page fault handler: %p\n", (void*) fault_addr );
480 if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pd) == -1) {
481 PrintError("Invalid Guest PDE Address: 0x%p\n", (void *)guest_cr3);
485 guest_pde = (pde32_t *)&(guest_pd[PDE32_INDEX(fault_addr)]);
488 // Check the guest page permissions
489 guest_pde_access = can_access_pde32(guest_pd, fault_addr, error_code);
491 // Check the shadow page permissions
492 shadow_pde_access = can_access_pde32(shadow_pd, fault_addr, error_code);
494 /* Was the page fault caused by the Guest's page tables? */
495 if (is_guest_pf(guest_pde_access, shadow_pde_access) == 1) {
496 PrintDebug("Injecting PDE pf to guest: (guest access error=%d) (pf error code=%d)\n",
497 *(uint_t *)&guest_pde_access, *(uint_t *)&error_code);
498 inject_guest_pf(info, fault_addr, error_code);
503 if (shadow_pde_access == PT_ENTRY_NOT_PRESENT)
505 pte32_t * shadow_pt = (pte32_t *)v3_create_new_shadow_pt();
507 shadow_pde->present = 1;
508 shadow_pde->user_page = guest_pde->user_page;
509 // shadow_pde->large_page = guest_pde->large_page;
510 shadow_pde->large_page = 0;
513 // VMM Specific options
514 shadow_pde->write_through = 0;
515 shadow_pde->cache_disable = 0;
516 shadow_pde->global_page = 0;
519 guest_pde->accessed = 1;
521 shadow_pde->pt_base_addr = PD32_BASE_ADDR((addr_t)V3_PAddr(shadow_pt));
523 if (guest_pde->large_page == 0) {
524 shadow_pde->writable = guest_pde->writable;
526 // ?? What if guest pde is dirty a this point?
527 ((pde32_4MB_t *)guest_pde)->dirty = 0;
528 shadow_pde->writable = 0;
531 else if (shadow_pde_access == PT_ACCESS_OK)
536 pte32_t * shadow_pt = (pte32_t *)V3_VAddr( (void*)(addr_t) PDE32_T_ADDR(*shadow_pde) );
538 if (guest_pde->large_page == 0) {
539 pte32_t * guest_pt = NULL;
540 if (guest_pa_to_host_va(info, PDE32_T_ADDR((*guest_pde)), (addr_t*)&guest_pt) == -1) {
541 // Machine check the guest
542 PrintDebug("Invalid Guest PTE Address: 0x%x\n", PDE32_T_ADDR((*guest_pde)));
543 v3_raise_exception(info, MC_EXCEPTION);
547 if (handle_shadow_pte32_fault(info, fault_addr, error_code, shadow_pt, guest_pt) == -1) {
548 PrintError("Error handling Page fault caused by PTE\n");
551 } else if (guest_pde->large_page == 1) {
552 if (handle_large_pagefault32(info, fault_addr, error_code, shadow_pt, (pde32_4MB_t *)guest_pde) == -1) {
553 PrintError("Error handling large pagefault\n");
558 else if ((shadow_pde_access == PT_WRITE_ERROR) &&
559 (guest_pde->large_page == 1) &&
560 (((pde32_4MB_t *)guest_pde)->dirty == 0))
563 // Page Directory Entry marked read-only
564 // Its a large page and we need to update the dirty bit in the guest
567 PrintDebug("Large page write error... Setting dirty bit and returning\n");
568 ((pde32_4MB_t *)guest_pde)->dirty = 1;
569 shadow_pde->writable = guest_pde->writable;
573 else if (shadow_pde_access == PT_USER_ERROR)
576 // Page Directory Entry marked non-user
578 PrintDebug("Shadow Paging User access error (shadow_pde_access=0x%x, guest_pde_access=0x%x)\n",
579 shadow_pde_access, guest_pde_access);
580 inject_guest_pf(info, fault_addr, error_code);
585 // inject page fault in guest
586 inject_guest_pf(info, fault_addr, error_code);
587 PrintDebug("Unknown Error occurred (shadow_pde_access=%d)\n", shadow_pde_access);
588 PrintDebug("Manual Says to inject page fault into guest\n");
589 #ifdef DEBUG_SHADOW_PAGING
590 PrintDebug("Guest PDE: (access=%d)\n\t", guest_pde_access);
591 PrintPDE32(fault_addr, guest_pde);
592 PrintDebug("Shadow PDE: (access=%d)\n\t", shadow_pde_access);
593 PrintPDE32(fault_addr, shadow_pde);
599 PrintDebug("Returning end of PDE function (rip=%p)\n", (void *)(addr_t)(info->rip));
606 * We assume the the guest pte pointer has already been translated to a host virtual address
608 static int handle_shadow_pte32_fault(struct guest_info * info,
610 pf_error_t error_code,
612 pte32_t * guest_pt) {
614 pt_access_status_t guest_pte_access;
615 pt_access_status_t shadow_pte_access;
616 pte32_t * guest_pte = (pte32_t *)&(guest_pt[PTE32_INDEX(fault_addr)]);;
617 pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(fault_addr)]);
620 // Check the guest page permissions
621 guest_pte_access = can_access_pte32(guest_pt, fault_addr, error_code);
623 // Check the shadow page permissions
624 shadow_pte_access = can_access_pte32(shadow_pt, fault_addr, error_code);
626 #ifdef DEBUG_SHADOW_PAGING
627 PrintDebug("Guest PTE: (access=%d)\n\t", guest_pte_access);
628 PrintPTE32(fault_addr, guest_pte);
629 PrintDebug("Shadow PTE: (access=%d)\n\t", shadow_pte_access);
630 PrintPTE32(fault_addr, shadow_pte);
633 /* Was the page fault caused by the Guest's page tables? */
634 if (is_guest_pf(guest_pte_access, shadow_pte_access) == 1) {
635 PrintDebug("Access error injecting pf to guest (guest access error=%d) (pf error code=%d)\n",
636 guest_pte_access, *(uint_t*)&error_code);
637 inject_guest_pf(info, fault_addr, error_code);
642 if (shadow_pte_access == PT_ACCESS_OK) {
643 // Inconsistent state...
644 // Guest Re-Entry will flush page tables and everything should now work
645 PrintDebug("Inconsistent state... Guest re-entry should flush tlb\n");
650 if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) {
652 addr_t guest_pa = PTE32_T_ADDR((*guest_pte)) + PT32_PAGE_OFFSET(fault_addr);
654 // Page Table Entry Not Present
656 host_region_type_t host_page_type = get_shadow_addr_type(info, guest_pa);
658 if (host_page_type == HOST_REGION_INVALID) {
659 // Inject a machine check in the guest
660 PrintDebug("Invalid Guest Address in page table (0x%p)\n", (void *)guest_pa);
661 v3_raise_exception(info, MC_EXCEPTION);
667 if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) {
668 struct shadow_page_state * state = &(info->shdw_pg_state);
669 addr_t shadow_pa = get_shadow_addr(info, guest_pa);
671 shadow_pte->page_base_addr = PT32_BASE_ADDR(shadow_pa);
673 shadow_pte->present = guest_pte->present;
674 shadow_pte->user_page = guest_pte->user_page;
676 //set according to VMM policy
677 shadow_pte->write_through = 0;
678 shadow_pte->cache_disable = 0;
679 shadow_pte->global_page = 0;
682 guest_pte->accessed = 1;
684 if (find_pte_map(state->cached_ptes, PT32_PAGE_ADDR(guest_pa)) != NULL) {
685 // Check if the entry is a page table...
686 PrintDebug("Marking page as Guest Page Table %d\n", shadow_pte->writable);
687 shadow_pte->vmm_info = PT32_GUEST_PT;
690 if (guest_pte->dirty == 1) {
691 shadow_pte->writable = guest_pte->writable;
692 } else if ((guest_pte->dirty == 0) && (error_code.write == 1)) {
693 shadow_pte->writable = guest_pte->writable;
694 guest_pte->dirty = 1;
696 if (shadow_pte->vmm_info == PT32_GUEST_PT) {
697 // Well that was quick...
698 struct shadow_page_state * state = &(info->shdw_pg_state);
699 PrintDebug("Immediate Write operation on Guest PAge Table Page\n");
700 state->cached_cr3 = 0;
703 } else if ((guest_pte->dirty == 0) && (error_code.write == 0)) { // was =
704 shadow_pte->writable = 0;
710 // Page fault handled by hook functions
711 if (handle_special_page_fault(info, fault_addr, guest_pa, error_code) == -1) {
712 PrintError("Special Page fault handler returned error for address: %p\n", (void *)fault_addr);
717 } else if ((shadow_pte_access == PT_WRITE_ERROR) &&
718 (guest_pte->dirty == 0)) {
720 PrintDebug("Shadow PTE Write Error\n");
721 guest_pte->dirty = 1;
722 shadow_pte->writable = guest_pte->writable;
724 if (shadow_pte->vmm_info == PT32_GUEST_PT) {
725 struct shadow_page_state * state = &(info->shdw_pg_state);
726 PrintDebug("Write operation on Guest PAge Table Page\n");
727 state->cached_cr3 = 0;
733 // Inject page fault into the guest
734 inject_guest_pf(info, fault_addr, error_code);
735 PrintError("PTE Page fault fell through... Not sure if this should ever happen\n");
736 PrintError("Manual Says to inject page fault into guest\n");
740 PrintDebug("Returning end of function\n");
749 /* Currently Does not work with Segmentation!!! */
750 int v3_handle_shadow_invlpg(struct guest_info * info)
752 if (info->mem_mode != VIRTUAL_MEM) {
753 // Paging must be turned on...
754 // should handle with some sort of fault I think
755 PrintError("ERROR: INVLPG called in non paged mode\n");
760 if (info->cpu_mode != PROTECTED) {
767 int ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr);
769 PrintError("Could not read instruction 0x%p (ret=%d)\n", (void *)(addr_t)(info->rip), ret);
774 /* Can INVLPG work with Segments?? */
775 while (is_prefix_byte(instr[index])) {
780 if( (instr[index + 0] != (uchar_t) 0x0f) ||
781 (instr[index + 1] != (uchar_t) 0x01) ) {
782 PrintError("invalid Instruction Opcode\n");
783 PrintTraceMemDump(instr, 15);
787 addr_t first_operand;
788 addr_t second_operand;
789 addr_t guest_cr3 = CR3_TO_PDE32_PA(info->shdw_pg_state.guest_cr3);
791 pde32_t * guest_pd = NULL;
793 if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pd) == -1) {
794 PrintError("Invalid Guest PDE Address: 0x%p\n", (void *)guest_cr3);
800 v3_operand_type_t addr_type = decode_operands32(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32);
802 if (addr_type != MEM_OPERAND) {
803 PrintError("Invalid Operand type\n");
807 pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32_VA(info->ctrl_regs.cr3);
808 pde32_t * shadow_pde = (pde32_t *)&shadow_pd[PDE32_INDEX(first_operand)];
811 //PrintDebug("PDE Index=%d\n", PDE32_INDEX(first_operand));
812 //PrintDebug("FirstOperand = %x\n", first_operand);
814 PrintDebug("Invalidating page for %p\n", (void *)first_operand);
816 guest_pde = (pde32_t *)&(guest_pd[PDE32_INDEX(first_operand)]);
818 if (guest_pde->large_page == 1) {
819 shadow_pde->present = 0;
820 PrintDebug("Invalidating Large Page\n");
822 if (shadow_pde->present == 1) {
823 pte32_t * shadow_pt = (pte32_t *)(addr_t)PDE32_T_ADDR((*shadow_pde));
824 pte32_t * shadow_pte = (pte32_t *) V3_VAddr( (void*) &shadow_pt[PTE32_INDEX(first_operand)] );
826 #ifdef DEBUG_SHADOW_PAGING
827 PrintDebug("Setting not present\n");
828 PrintPTE32(first_operand, shadow_pte );
831 shadow_pte->present = 0;
843 static int create_pd32_nonaligned_4MB_page(struct guest_info * info, pte32_t * pt, addr_t guest_addr, pde32_4MB_t * large_shadow_pde) {
845 pte32_t * pte_cursor;
848 for (i = 0; i < 1024; i++) {
849 guest_pa = guest_addr + (PAGE_SIZE * i);
850 host_region_type_t host_page_type = get_shadow_addr_type(info, guest_pa);
852 pte_cursor = &(pt[i]);
854 if (host_page_type == HOST_REGION_INVALID) {
855 // Currently we don't support this, but in theory we could
856 PrintError("Invalid Host Memory Type\n");
858 } else if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) {
859 addr_t shadow_pa = get_shadow_addr(info, guest_pa);
862 pte_cursor->page_base_addr = PT32_BASE_ADDR(shadow_pa);
863 pte_cursor->present = 1;
864 pte_cursor->writable = large_shadow_pde->writable;
865 pte_cursor->user_page = large_shadow_pde->user_page;
866 pte_cursor->write_through = 0;
867 pte_cursor->cache_disable = 0;
868 pte_cursor->global_page = 0;
871 PrintError("Unsupported Host Memory Type\n");
879 static int handle_large_pagefault32(struct guest_info * info,
880 pde32_t * guest_pde, pde32_t * shadow_pde,
881 addr_t fault_addr, pf_error_t error_code ) {
882 struct shadow_region * mem_reg;
883 pde32_4MB_t * large_guest_pde = (pde32_4MB_t *)guest_pde;
884 pde32_4MB_t * large_shadow_pde = (pde32_4MB_t *)shadow_pde;
885 host_region_type_t host_page_type;
886 addr_t guest_start_addr = PDE32_4MB_T_ADDR(*large_guest_pde);
887 // addr_t guest_end_addr = guest_start_addr + PAGE_SIZE_4MB; // start address + 4MB
890 // Check that the Guest PDE entry points to valid memory
891 // else Machine Check the guest
892 PrintDebug("Large Page: Page Base Addr=%x\n", guest_start_addr);
894 host_page_type = get_shadow_addr_type(info, guest_start_addr);
896 if (host_page_type == HOST_REGION_INVALID) {
897 PrintError("Invalid guest address in large page (0x%x)\n", guest_start_addr);
898 v3_raise_exception(info, MC_EXCEPTION);
904 if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) {
906 addr_t host_start_addr = 0;
907 addr_t region_end_addr = 0;
909 // Check for a large enough region in host memory
910 mem_reg = get_shadow_region_by_addr(&(info->mem_map), guest_start_addr);
911 PrintDebug("Host region: host_addr=%x (guest_start=%x, end=%x)\n",
912 mem_reg->host_addr, mem_reg->guest_start, mem_reg->guest_end);
913 host_start_addr = mem_reg->host_addr + (guest_start_addr - mem_reg->guest_start);
914 region_end_addr = mem_reg->host_addr + (mem_reg->guest_end - mem_reg->guest_start);
916 PrintDebug("Host Start Addr=%x; Region End Addr=%x\n", host_start_addr, region_end_addr);
920 if (large_guest_pde->dirty == 1) { // dirty
921 large_shadow_pde->writable = guest_pde->writable;
922 } else if (error_code.write == 1) { // not dirty, access is write
923 large_shadow_pde->writable = guest_pde->writable;
924 large_guest_pde->dirty = 1;
925 } else { // not dirty, access is read
926 large_shadow_pde->writable = 0;
930 // Check if the region is at least an additional 4MB
934 if ((PD32_4MB_PAGE_OFFSET(host_start_addr) == 0) &&
935 (region_end_addr >= host_start_addr + PAGE_SIZE_4MB)) { // if 4MB boundary
936 large_shadow_pde->page_base_addr = PD32_4MB_BASE_ADDR(host_start_addr);
937 } else { // else generate 4k pages
938 pte32_t * shadow_pt = NULL;
939 PrintDebug("Handling non aligned large page\n");
941 shadow_pde->large_page = 0;
943 shadow_pt = create_new_shadow_pt32();
945 if (create_pd32_nonaligned_4MB_page(info, shadow_pt, guest_start_addr, large_shadow_pde) == -1) {
946 PrintError("Non Aligned Large Page Error\n");
952 #ifdef DEBUG_SHADOW_PAGING
953 PrintDebug("non-aligned Shadow PT\n");
954 PrintPT32(PT32_PAGE_ADDR(fault_addr), shadow_pt);
956 shadow_pde->pt_base_addr = PD32_BASE_ADDR(shadow_pt);
960 // Handle hooked pages as well as other special pages
961 if (handle_special_page_fault(info, fault_addr, guest_start_addr, error_code) == -1) {
962 PrintError("Special Page Fault handler returned error for address: %x\n", fault_addr);