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".
23 static inline int activate_shadow_pt_32(struct guest_info * info) {
24 struct cr3_32 * shadow_cr3 = (struct cr3_32 *)&(info->ctrl_regs.cr3);
25 struct cr3_32 * guest_cr3 = (struct cr3_32 *)&(info->shdw_pg_state.guest_cr3);
26 struct shadow_page_data * shdw_page = create_new_shadow_pt(info);
28 shdw_page->cr3 = shdw_page->page_pa;
30 shadow_cr3->pdt_base_addr = PAGE_BASE_ADDR_4KB(shdw_page->page_pa);
31 PrintDebug( "Created new shadow page table %p\n", (void *)BASE_TO_PAGE_ADDR(shadow_cr3->pdt_base_addr));
33 shadow_cr3->pwt = guest_cr3->pwt;
34 shadow_cr3->pcd = guest_cr3->pcd;
37 v3_swap_flush(info->vm_info);
48 * * 32 bit Page table fault handlers
52 static int handle_4MB_shadow_pagefault_32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code,
53 pte32_t * shadow_pt, pde32_4MB_t * large_guest_pde);
55 static int handle_pte_shadow_pagefault_32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code,
56 pte32_t * shadow_pt, pte32_t * guest_pt);
59 static inline int handle_shadow_pagefault_32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
60 pde32_t * guest_pd = NULL;
61 pde32_t * shadow_pd = CR3_TO_PDE32_VA(info->ctrl_regs.cr3);
62 addr_t guest_cr3 = CR3_TO_PDE32_PA(info->shdw_pg_state.guest_cr3);
63 pt_access_status_t guest_pde_access;
64 pt_access_status_t shadow_pde_access;
65 pde32_t * guest_pde = NULL;
66 pde32_t * shadow_pde = (pde32_t *)&(shadow_pd[PDE32_INDEX(fault_addr)]);
68 PrintDebug("Shadow page fault handler: %p\n", (void*) fault_addr );
69 PrintDebug("Handling PDE32 Fault\n");
71 if (v3_gpa_to_hva(info, guest_cr3, (addr_t*)&guest_pd) == -1) {
72 PrintError("Invalid Guest PDE Address: 0x%p\n", (void *)guest_cr3);
76 guest_pde = (pde32_t *)&(guest_pd[PDE32_INDEX(fault_addr)]);
79 // Check the guest page permissions
80 guest_pde_access = v3_can_access_pde32(guest_pd, fault_addr, error_code);
82 // Check the shadow page permissions
83 shadow_pde_access = v3_can_access_pde32(shadow_pd, fault_addr, error_code);
85 /* Was the page fault caused by the Guest's page tables? */
86 if (v3_is_guest_pf(guest_pde_access, shadow_pde_access) == 1) {
87 PrintDebug("Injecting PDE pf to guest: (guest access error=%d) (shdw access error=%d) (pf error code=%d)\n",
88 *(uint_t *)&guest_pde_access, *(uint_t *)&shadow_pde_access, *(uint_t *)&error_code);
89 if (v3_inject_guest_pf(info, fault_addr, error_code) == -1) {
90 PrintError("Could not inject guest page fault for vaddr %p\n", (void *)fault_addr);
98 if (shadow_pde_access == PT_ACCESS_USER_ERROR) {
100 // PDE Entry marked non user
102 PrintDebug("Shadow Paging User access error (shadow_pde_access=0x%x, guest_pde_access=0x%x)\n",
103 shadow_pde_access, guest_pde_access);
105 if (v3_inject_guest_pf(info, fault_addr, error_code) == -1) {
106 PrintError("Could not inject guest page fault for vaddr %p\n", (void *)fault_addr);
110 } else if ((shadow_pde_access == PT_ACCESS_WRITE_ERROR) &&
111 (guest_pde->large_page == 1)) {
113 ((pde32_4MB_t *)guest_pde)->dirty = 1;
114 shadow_pde->writable = guest_pde->writable;
116 } else if ((shadow_pde_access != PT_ACCESS_NOT_PRESENT) &&
117 (shadow_pde_access != PT_ACCESS_OK)) {
118 // inject page fault in guest
119 if (v3_inject_guest_pf(info, fault_addr, error_code) == -1) {
120 PrintError("Could not inject guest page fault for vaddr %p\n", (void *)fault_addr);
123 PrintDebug("Unknown Error occurred (shadow_pde_access=%d)\n", shadow_pde_access);
124 PrintDebug("Manual Says to inject page fault into guest\n");
129 pte32_t * shadow_pt = NULL;
130 pte32_t * guest_pt = NULL;
132 // Get the next shadow page level, allocate if not present
134 if (shadow_pde_access == PT_ACCESS_NOT_PRESENT) {
135 struct shadow_page_data * shdw_page = create_new_shadow_pt(info);
136 shadow_pt = (pte32_t *)V3_VAddr((void *)shdw_page->page_pa);
138 shadow_pde->present = 1;
139 shadow_pde->user_page = guest_pde->user_page;
142 if (guest_pde->large_page == 0) {
143 shadow_pde->writable = guest_pde->writable;
145 // This large page flag is temporary until we can get a working cache....
146 ((pde32_4MB_t *)guest_pde)->vmm_info = V3_LARGE_PG;
148 if (error_code.write) {
149 shadow_pde->writable = guest_pde->writable;
150 ((pde32_4MB_t *)guest_pde)->dirty = 1;
152 shadow_pde->writable = 0;
153 ((pde32_4MB_t *)guest_pde)->dirty = 0;
158 // VMM Specific options
159 shadow_pde->write_through = guest_pde->write_through;
160 shadow_pde->cache_disable = guest_pde->cache_disable;
161 shadow_pde->global_page = guest_pde->global_page;
164 guest_pde->accessed = 1;
169 shadow_pde->pt_base_addr = PAGE_BASE_ADDR(shdw_page->page_pa);
171 shadow_pt = (pte32_t *)V3_VAddr((void *)BASE_TO_PAGE_ADDR(shadow_pde->pt_base_addr));
176 if (guest_pde->large_page == 0) {
177 if (v3_gpa_to_hva(info, BASE_TO_PAGE_ADDR(guest_pde->pt_base_addr), (addr_t*)&guest_pt) == -1) {
178 // Machine check the guest
179 PrintDebug("Invalid Guest PTE Address: 0x%p\n", (void *)BASE_TO_PAGE_ADDR(guest_pde->pt_base_addr));
180 v3_raise_exception(info, MC_EXCEPTION);
184 if (handle_pte_shadow_pagefault_32(info, fault_addr, error_code, shadow_pt, guest_pt) == -1) {
185 PrintError("Error handling Page fault caused by PTE\n");
189 if (handle_4MB_shadow_pagefault_32(info, fault_addr, error_code, shadow_pt, (pde32_4MB_t *)guest_pde) == -1) {
190 PrintError("Error handling large pagefault\n");
199 static int handle_pte_shadow_pagefault_32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code,
200 pte32_t * shadow_pt, pte32_t * guest_pt) {
202 pt_access_status_t guest_pte_access;
203 pt_access_status_t shadow_pte_access;
204 pte32_t * guest_pte = (pte32_t *)&(guest_pt[PTE32_INDEX(fault_addr)]);;
205 pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(fault_addr)]);
206 addr_t guest_pa = BASE_TO_PAGE_ADDR((addr_t)(guest_pte->page_base_addr)) + PAGE_OFFSET(fault_addr);
208 struct v3_mem_region * shdw_reg = v3_get_mem_region(info->vm_info, info->vcpu_id, guest_pa);
210 if (shdw_reg == NULL) {
211 // Inject a machine check in the guest
212 PrintDebug("Invalid Guest Address in page table (0x%p)\n", (void *)guest_pa);
213 v3_raise_exception(info, MC_EXCEPTION);
217 // Check the guest page permissions
218 guest_pte_access = v3_can_access_pte32(guest_pt, fault_addr, error_code);
220 // Check the shadow page permissions
221 shadow_pte_access = v3_can_access_pte32(shadow_pt, fault_addr, error_code);
224 /* Was the page fault caused by the Guest's page tables? */
225 if (v3_is_guest_pf(guest_pte_access, shadow_pte_access) == 1) {
227 PrintDebug("Access error injecting pf to guest (guest access error=%d) (pf error code=%d)\n",
228 guest_pte_access, *(uint_t*)&error_code);
231 if (is_swapped_pte32(guest_pte)) {
233 pf_error_t swap_perms;
237 int sym_ret = v3_get_vaddr_perms(info, fault_addr, guest_pte, &swap_perms);
240 addr_t swp_pg_addr = 0;
244 #ifdef V3_CONFIG_SYMBIOTIC_SWAP_TELEMETRY
245 if (error_code.write == 0) {
246 info->vm_info->swap_state.read_faults++;
248 info->vm_info->swap_state.write_faults++;
253 swp_pg_addr = get_swapped_pg_addr(info->vm_info, guest_pte);
255 if (swp_pg_addr != 0) {
256 PrintDebug("Swapped page address=%p\n", (void *)swp_pg_addr);
259 if (info->cpl == 0) {
260 PrintError("Swapped Page fault in kernel mode.... bad...\n");
265 int sym_ret = get_vaddr_perms(info, fault_addr, guest_pte, &swap_perms);
268 PrintError("Symcall error...\n");
270 } else if (sym_ret == 0) {
273 if (swap_perms.present == 0) {
274 PrintError("Nonpresent swapped page\n");
277 // swap_perms.write ==1 || error_code.write == 0
278 // swap_perms.user == 0 || error_code.user == 1
280 // This checks for permissions violations that require a guest PF injection
281 if ( (swap_perms.present == 1) &&
282 ( (swap_perms.write == 1) ||
283 (error_code.write == 0) ) &&
284 ( (swap_perms.user == 1) ||
285 (error_code.user == 0) ) ) {
286 addr_t swp_pg_pa = 0;
288 swp_pg_pa = map_swp_page(info->vm_info, shadow_pte, guest_pte, (void *)swp_pg_addr);
290 PrintDebug("Page fault on swapped out page (vaddr=%p) (pte=%x) (error_code=%x)\n",
291 (void *)fault_addr, *(uint32_t *)guest_pte, *(uint32_t *)&error_code);
293 shadow_pte->writable = swap_perms.write;
294 shadow_pte->user_page = swap_perms.user;
296 shadow_pte->write_through = 0;
297 shadow_pte->cache_disable = 0;
298 shadow_pte->global_page = 0;
300 shadow_pte->present = 1;
302 shadow_pte->page_base_addr = swp_pg_pa;
304 #ifdef V3_CONFIG_SYMBIOTIC_SWAP_TELEMETRY
305 info->vm_info->swap_state.mapped_pages++;
307 // PrintError("Swap fault handled\n");
312 PrintDebug("Not a sym swappable page\n");
319 if (v3_inject_guest_pf(info, fault_addr, error_code) == -1) {
320 PrintError("Could not inject guest page fault for vaddr %p\n", (void *)fault_addr);
329 if (shadow_pte_access == PT_ACCESS_OK) {
330 // Inconsistent state...
331 // Guest Re-Entry will flush page tables and everything should now work
332 PrintDebug("Inconsistent state... Guest re-entry should flush tlb\n");
337 if (shadow_pte_access == PT_ACCESS_NOT_PRESENT) {
338 // Page Table Entry Not Present
339 PrintDebug("guest_pa =%p\n", (void *)guest_pa);
341 if ((shdw_reg->flags.alloced == 1) ||
342 (shdw_reg->flags.read == 1)) {
343 addr_t shadow_pa = 0;
345 if (v3_gpa_to_hpa(info, guest_pa, &shadow_pa) == -1) {
346 PrintError("could not translate page fault address (%p)\n", (void *)guest_pa);
350 shadow_pte->page_base_addr = PAGE_BASE_ADDR(shadow_pa);
352 PrintDebug("\tMapping shadow page (%p)\n", (void *)BASE_TO_PAGE_ADDR(shadow_pte->page_base_addr));
354 shadow_pte->present = guest_pte->present;
355 shadow_pte->user_page = guest_pte->user_page;
357 //set according to VMM policy
358 shadow_pte->write_through = guest_pte->write_through;
359 shadow_pte->cache_disable = guest_pte->cache_disable;
360 shadow_pte->global_page = guest_pte->global_page;
363 guest_pte->accessed = 1;
365 if (guest_pte->dirty == 1) {
366 shadow_pte->writable = guest_pte->writable;
367 } else if ((guest_pte->dirty == 0) && (error_code.write == 1)) {
368 shadow_pte->writable = guest_pte->writable;
369 guest_pte->dirty = 1;
370 } else if ((guest_pte->dirty == 0) && (error_code.write == 0)) {
371 shadow_pte->writable = 0;
375 if (shdw_reg->flags.write == 0) {
376 shadow_pte->writable = 0;
380 // Page fault handled by hook functions
382 if (shdw_reg->unhandled(info, fault_addr, guest_pa, shdw_reg, error_code) == -1) {
383 PrintError("Special Page fault handler returned error for address: %p\n", (void *)fault_addr);
387 } else if (shadow_pte_access == PT_ACCESS_WRITE_ERROR) {
388 guest_pte->dirty = 1;
391 if (shdw_reg->flags.write == 1) {
392 PrintDebug("Shadow PTE Write Error\n");
393 shadow_pte->writable = guest_pte->writable;
395 if (shdw_reg->unhandled(info, fault_addr, guest_pa, shdw_reg, error_code) == -1) {
396 PrintError("Special Page fault handler returned error for address: %p\n", (void *)fault_addr);
405 // Inject page fault into the guest
406 if (v3_inject_guest_pf(info, fault_addr, error_code) == -1) {
407 PrintError("Could not inject guest page fault for vaddr %p\n", (void *)fault_addr);
411 PrintError("PTE Page fault fell through... Not sure if this should ever happen\n");
412 PrintError("Manual Says to inject page fault into guest\n");
421 static int handle_4MB_shadow_pagefault_32(struct guest_info * info,
422 addr_t fault_addr, pf_error_t error_code,
423 pte32_t * shadow_pt, pde32_4MB_t * large_guest_pde)
425 pt_access_status_t shadow_pte_access = v3_can_access_pte32(shadow_pt, fault_addr, error_code);
426 pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(fault_addr)]);
427 addr_t guest_fault_pa = BASE_TO_PAGE_ADDR_4MB(large_guest_pde->page_base_addr) + PAGE_OFFSET_4MB(fault_addr);
430 PrintDebug("Handling 4MB fault (guest_fault_pa=%p) (error_code=%x)\n", (void *)guest_fault_pa, *(uint_t*)&error_code);
431 PrintDebug("ShadowPT=%p, LargeGuestPDE=%p\n", shadow_pt, large_guest_pde);
433 struct v3_mem_region * shdw_reg = v3_get_mem_region(info->vm_info, info->vcpu_id, guest_fault_pa);
436 if (shdw_reg == NULL) {
437 // Inject a machine check in the guest
438 PrintDebug("Invalid Guest Address in page table (0x%p)\n", (void *)guest_fault_pa);
439 v3_raise_exception(info, MC_EXCEPTION);
443 if (shadow_pte_access == PT_ACCESS_OK) {
444 // Inconsistent state...
445 // Guest Re-Entry will flush tables and everything should now workd
446 PrintDebug("Inconsistent state... Guest re-entry should flush tlb\n");
451 if (shadow_pte_access == PT_ACCESS_NOT_PRESENT) {
452 // Get the guest physical address of the fault
454 if ((shdw_reg->flags.alloced == 1) ||
455 (shdw_reg->flags.read == 1)) {
456 addr_t shadow_pa = 0;
458 if (v3_gpa_to_hpa(info, guest_fault_pa, &shadow_pa) == -1) {
459 PrintError("could not translate page fault address (%p)\n", (void *)guest_fault_pa);
462 shadow_pte->page_base_addr = PAGE_BASE_ADDR(shadow_pa);
464 PrintDebug("\tMapping shadow page (%p)\n", (void *)BASE_TO_PAGE_ADDR(shadow_pte->page_base_addr));
466 shadow_pte->present = 1;
468 /* We are assuming that the PDE entry has precedence
469 * so the Shadow PDE will mirror the guest PDE settings,
470 * and we don't have to worry about them here
473 shadow_pte->user_page = 1;
475 if (shdw_reg->flags.write == 0) {
476 shadow_pte->writable = 0;
478 shadow_pte->writable = 1;
481 //set according to VMM policy
482 shadow_pte->write_through = large_guest_pde->write_through;
483 shadow_pte->cache_disable = large_guest_pde->cache_disable;
484 shadow_pte->global_page = large_guest_pde->global_page;
488 if (shdw_reg->unhandled(info, fault_addr, guest_fault_pa, shdw_reg, error_code) == -1) {
489 PrintError("Special Page Fault handler returned error for address: %p\n", (void *)fault_addr);
493 } else if (shadow_pte_access == PT_ACCESS_WRITE_ERROR) {
494 if (shdw_reg->unhandled(info, fault_addr, guest_fault_pa, shdw_reg, error_code) == -1) {
495 PrintError("Special Page Fault handler returned error for address: %p\n", (void *)fault_addr);
499 PrintError("Error in large page fault handler...\n");
500 PrintError("This case should have been handled at the top level handler\n");
504 PrintDebug("Returning from large page fault handler\n");
519 /* If we start to optimize we should look up the guest pages in the cache... */
520 static inline int handle_shadow_invlpg_32(struct guest_info * info, addr_t vaddr) {
521 pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32_VA(info->ctrl_regs.cr3);
522 pde32_t * shadow_pde = (pde32_t *)&shadow_pd[PDE32_INDEX(vaddr)];
524 addr_t guest_cr3 = CR3_TO_PDE32_PA(info->shdw_pg_state.guest_cr3);
525 pde32_t * guest_pd = NULL;
528 if (v3_gpa_to_hva(info, guest_cr3, (addr_t*)&guest_pd) == -1) {
529 PrintError("Invalid Guest PDE Address: 0x%p\n", (void *)guest_cr3);
533 guest_pde = (pde32_t *)&(guest_pd[PDE32_INDEX(vaddr)]);
535 if (guest_pde->large_page == 1) {
536 shadow_pde->present = 0;
537 PrintDebug("Invalidating Large Page\n");
538 } else if (shadow_pde->present == 1) {
539 pte32_t * shadow_pt = (pte32_t *)(addr_t)BASE_TO_PAGE_ADDR_4KB(shadow_pde->pt_base_addr);
540 pte32_t * shadow_pte = (pte32_t *) V3_VAddr( (void*) &shadow_pt[PTE32_INDEX(vaddr)] );
542 PrintDebug("Setting not present\n");
544 shadow_pte->present = 0;