X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmm_shadow_paging.c;h=55920edadd4fa549226565ce24914a88c14fbbe3;hb=746934f539c53c6fc687890c02ebf886c6873342;hp=54347e351b5b3cc48a391c701115288a7a2818ce;hpb=5db3136d5331c355c188537c68005db8d2cb4eac;p=palacios.git diff --git a/palacios/src/palacios/vmm_shadow_paging.c b/palacios/src/palacios/vmm_shadow_paging.c index 54347e3..55920ed 100644 --- a/palacios/src/palacios/vmm_shadow_paging.c +++ b/palacios/src/palacios/vmm_shadow_paging.c @@ -24,6 +24,9 @@ #include #include #include +#include + +#include #ifndef DEBUG_SHADOW_PAGING #undef PrintDebug @@ -36,18 +39,37 @@ ***/ +struct guest_table { + addr_t cr3; + struct list_head link; +}; + + +struct backptr { + addr_t ptr; + struct list_head link; +}; + + +struct shadow_page_data { + addr_t ptr; + addr_t guest_addr; + + struct list_head backptrs; + struct list_head guest_tables; +}; + -DEFINE_HASHTABLE_INSERT(add_cr3_to_cache, addr_t, struct hashtable *); -DEFINE_HASHTABLE_SEARCH(find_cr3_in_cache, addr_t, struct hashtable *); -DEFINE_HASHTABLE_REMOVE(del_cr3_from_cache, addr_t, struct hashtable *, 0); +//DEFINE_HASHTABLE_INSERT(add_cr3_to_cache, addr_t, struct hashtable *); +//DEFINE_HASHTABLE_SEARCH(find_cr3_in_cache, addr_t, struct hashtable *); +//DEFINE_HASHTABLE_REMOVE(del_cr3_from_cache, addr_t, struct hashtable *, 0); DEFINE_HASHTABLE_INSERT(add_pte_map, addr_t, addr_t); DEFINE_HASHTABLE_SEARCH(find_pte_map, addr_t, addr_t); -DEFINE_HASHTABLE_REMOVE(del_pte_map, addr_t, addr_t, 0); - +//DEFINE_HASHTABLE_REMOVE(del_pte_map, addr_t, addr_t, 0); @@ -68,22 +90,25 @@ static int cr3_equals(addr_t key1, addr_t key2) { } -static int handle_shadow_pte32_fault(struct guest_info* info, - addr_t fault_addr, - pf_error_t error_code, - pte32_t * shadow_pte, - pte32_t * guest_pte); -static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code); +static int activate_shadow_pt_32(struct guest_info * info); +static int activate_shadow_pt_32pae(struct guest_info * info); +static int activate_shadow_pt_64(struct guest_info * info); + + +static int handle_shadow_pagefault_32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code); +static int handle_shadow_pagefault_32pae(struct guest_info * info, addr_t fault_addr, pf_error_t error_code); +static int handle_shadow_pagefault_64(struct guest_info * info, addr_t fault_addr, pf_error_t error_code); + -int init_shadow_page_state(struct guest_info * info) { +static int cache_page_tables_32(struct guest_info * info, addr_t pde); +static int cache_page_tables_64(struct guest_info * info, addr_t pde); + +int v3_init_shadow_page_state(struct guest_info * info) { struct shadow_page_state * state = &(info->shdw_pg_state); - state->guest_mode = PDE32; - state->shadow_mode = PDE32; state->guest_cr3 = 0; - state->shadow_cr3 = 0; - + state->guest_cr0 = 0; state->cr3_cache = create_hashtable(0, &cr3_hash_fn, &cr3_equals); @@ -96,57 +121,17 @@ int init_shadow_page_state(struct guest_info * info) { -/* - For now we'll do something a little more lightweight -int cache_page_tables32(struct guest_info * info, addr_t pde) { - struct shadow_page_state * state = &(info->shdw_pg_state); - addr_t pde_host_addr; - pde32_t * tmp_pde; - struct hashtable * pte_cache = NULL; - int i = 0; - - - pte_cache = (struct hashtable *)find_cr3_in_cache(state->cr3_cache, pde); - if (pte_cache != NULL) { - PrintError("CR3 already present in cache\n"); - state->current_ptes = pte_cache; - return 1; - } else { - PrintError("Creating new CR3 cache entry\n"); - pte_cache = create_hashtable(0, &pte_hash_fn, &pte_equals); - state->current_ptes = pte_cache; - add_cr3_to_cache(state->cr3_cache, pde, pte_cache); - } - if (guest_pa_to_host_va(info, pde, &pde_host_addr) == -1) { - PrintError("Could not lookup host address of guest PDE\n"); +int v3_cache_page_tables(struct guest_info * info, addr_t cr3) { + switch(v3_get_cpu_mode(info)) { + case PROTECTED: + return cache_page_tables_32(info, CR3_TO_PDE32_PA(cr3)); + default: return -1; } - - tmp_pde = (pde32_t *)pde_host_addr; - - add_pte_map(pte_cache, pde, pde_host_addr); - - - for (i = 0; i < MAX_PDE32_ENTRIES; i++) { - if ((tmp_pde[i].present) && (tmp_pde[i].large_page == 0)) { - addr_t pte_host_addr; - - if (guest_pa_to_host_va(info, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), &pte_host_addr) == -1) { - PrintError("Could not lookup host address of guest PDE\n"); - return -1; - } - - add_pte_map(pte_cache, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), pte_host_addr); - } - } - - - return 0; } -*/ -int cache_page_tables32(struct guest_info * info, addr_t pde) { +static int cache_page_tables_32(struct guest_info * info, addr_t pde) { struct shadow_page_state * state = &(info->shdw_pg_state); addr_t pde_host_addr; pde32_t * tmp_pde; @@ -167,7 +152,7 @@ int cache_page_tables32(struct guest_info * info, addr_t pde) { pte_cache = create_hashtable(0, &pte_hash_fn, &pte_equals); state->cached_ptes = pte_cache; - if (guest_pa_to_host_pa(info, pde, &pde_host_addr) == -1) { + if (guest_pa_to_host_va(info, pde, &pde_host_addr) == -1) { PrintError("Could not lookup host address of guest PDE\n"); return -1; } @@ -181,12 +166,12 @@ int cache_page_tables32(struct guest_info * info, addr_t pde) { if ((tmp_pde[i].present) && (tmp_pde[i].large_page == 0)) { addr_t pte_host_addr; - if (guest_pa_to_host_pa(info, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), &pte_host_addr) == -1) { + if (guest_pa_to_host_va(info, (addr_t)(BASE_TO_PAGE_ADDR(tmp_pde[i].pt_base_addr)), &pte_host_addr) == -1) { PrintError("Could not lookup host address of guest PDE\n"); return -1; } - add_pte_map(pte_cache, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), pte_host_addr); + add_pte_map(pte_cache, (addr_t)(BASE_TO_PAGE_ADDR(tmp_pde[i].pt_base_addr)), pte_host_addr); } } @@ -195,13 +180,17 @@ int cache_page_tables32(struct guest_info * info, addr_t pde) { } +static int cache_page_tables_64(struct guest_info * info, addr_t pde) { + return -1; +} + int v3_replace_shdw_page32(struct guest_info * info, addr_t location, pte32_t * new_page, pte32_t * old_page) { - pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32(info->shdw_pg_state.shadow_cr3); + pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32_VA(info->ctrl_regs.cr3); pde32_t * shadow_pde = (pde32_t *)&(shadow_pd[PDE32_INDEX(location)]); if (shadow_pde->large_page == 0) { - pte32_t * shadow_pt = (pte32_t *)(addr_t)PDE32_T_ADDR((*shadow_pde)); + pte32_t * shadow_pt = (pte32_t *)(addr_t)BASE_TO_PAGE_ADDR(shadow_pde->pt_base_addr); pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(location)]); //if (shadow_pte->present == 1) { @@ -212,30 +201,158 @@ int v3_replace_shdw_page32(struct guest_info * info, addr_t location, pte32_t * } else { // currently unhandled + PrintError("Replacing large shadow pages not implemented\n"); + return -1; + } + + return 0; +} + + + + + +// We assume that shdw_pg_state.guest_cr3 is pointing to the page tables we want to activate +// We also assume that the CPU mode has not changed during this page table transition +static int activate_shadow_pt_32(struct guest_info * info) { + struct cr3_32 * shadow_cr3 = (struct cr3_32 *)&(info->ctrl_regs.cr3); + struct cr3_32 * guest_cr3 = (struct cr3_32 *)&(info->shdw_pg_state.guest_cr3); + int cached = 0; + + // Check if shadow page tables are in the cache + cached = cache_page_tables_32(info, CR3_TO_PDE32_PA(*(addr_t *)guest_cr3)); + + if (cached == -1) { + PrintError("CR3 Cache failed\n"); + return -1; + } else if (cached == 0) { + addr_t shadow_pt; + + PrintDebug("New CR3 is different - flushing shadow page table %p\n", shadow_cr3 ); + delete_page_tables_32(CR3_TO_PDE32_VA(*(uint_t*)shadow_cr3)); + + shadow_pt = v3_create_new_shadow_pt(); + + shadow_cr3->pdt_base_addr = (addr_t)V3_PAddr((void *)(addr_t)PAGE_BASE_ADDR(shadow_pt)); + PrintDebug( "Created new shadow page table %p\n", (void *)(addr_t)shadow_cr3->pdt_base_addr ); + } else { + PrintDebug("Reusing cached shadow Page table\n"); + } + + shadow_cr3->pwt = guest_cr3->pwt; + shadow_cr3->pcd = guest_cr3->pcd; + + return 0; +} + +static int activate_shadow_pt_32pae(struct guest_info * info) { + PrintError("Activating 32 bit PAE page tables not implemented\n"); + return -1; +} + +static int activate_shadow_pt_64_cb(page_type_t type, addr_t vaddr, addr_t page_ptr, addr_t page_pa, void * private_data) { + PrintDebug("CB: Page: %p->%p (host_ptr=%p), Type: %s\n", (void *)vaddr, (void *)page_pa, (void *)page_ptr, v3_page_type_to_str(type)); + return 0; +} + + +static int activate_shadow_pt_64(struct guest_info * info) { + // struct cr3_64 * shadow_cr3 = (struct cr3_64 *)&(info->ctrl_regs.cr3); + struct cr3_64 * guest_cr3 = (struct cr3_64 *)&(info->shdw_pg_state.guest_cr3); + int cached = 0; + + v3_walk_guest_pt_64(info, info->shdw_pg_state.guest_cr3, activate_shadow_pt_64_cb, NULL); + + + + return -1; + + + // Check if shadow page tables are in the cache + cached = cache_page_tables_64(info, CR3_TO_PDE32_PA(*(addr_t *)guest_cr3)); + /* + if (cached == -1) { + PrintError("CR3 Cache failed\n"); return -1; + } else if (cached == 0) { + addr_t shadow_pt; + + PrintDebug("New CR3 is different - flushing shadow page table %p\n", shadow_cr3 ); + delete_page_tables_32(CR3_TO_PDE32_VA(*(uint_t*)shadow_cr3)); + + shadow_pt = v3_create_new_shadow_pt(); + + shadow_cr3->pml4t_base_addr = (addr_t)V3_PAddr((void *)(addr_t)PAGE_BASE_ADDR(shadow_pt)); + PrintDebug( "Created new shadow page table %p\n", (void *)(addr_t)shadow_cr3->pml4t_base_addr ); + } else { + PrintDebug("Reusing cached shadow Page table\n"); } + shadow_cr3->pwt = guest_cr3->pwt; + shadow_cr3->pcd = guest_cr3->pcd; + + return 0; + */ +} + + +// Reads the guest CR3 register +// creates new shadow page tables +// updates the shadow CR3 register to point to the new pts +int v3_activate_shadow_pt(struct guest_info * info) { + switch (info->cpu_mode) { + + case PROTECTED: + return activate_shadow_pt_32(info); + case PROTECTED_PAE: + return activate_shadow_pt_32pae(info); + case LONG: + case LONG_32_COMPAT: + case LONG_16_COMPAT: + return activate_shadow_pt_64(info); + default: + PrintError("Invalid CPU mode: %d\n", info->cpu_mode); + return -1; + } + return 0; } +int v3_activate_passthrough_pt(struct guest_info * info) { + // For now... But we need to change this.... + // As soon as shadow paging becomes active the passthrough tables are hosed + // So this will cause chaos if it is called at that time + info->ctrl_regs.cr3 = *(addr_t*)&(info->direct_map_pt); + //PrintError("Activate Passthrough Page tables not implemented\n"); + return 0; +} -int handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { +int v3_handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { if (info->mem_mode == PHYSICAL_MEM) { // If paging is not turned on we need to handle the special cases + +#ifdef DEBUG_SHADOW_PAGING + PrintHostPageTree(info->cpu_mode, fault_addr, info->ctrl_regs.cr3); + PrintGuestPageTree(info, fault_addr, info->shdw_pg_state.guest_cr3); +#endif + return handle_special_page_fault(info, fault_addr, fault_addr, error_code); } else if (info->mem_mode == VIRTUAL_MEM) { switch (info->cpu_mode) { case PROTECTED: - return handle_shadow_pagefault32(info, fault_addr, error_code); + return handle_shadow_pagefault_32(info, fault_addr, error_code); break; case PROTECTED_PAE: + return handle_shadow_pagefault_32pae(info, fault_addr, error_code); case LONG: + return handle_shadow_pagefault_64(info, fault_addr, error_code); + break; default: PrintError("Unhandled CPU Mode\n"); return -1; @@ -246,10 +363,10 @@ int handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_erro } } -addr_t create_new_shadow_pt32() { +addr_t v3_create_new_shadow_pt() { void * host_pde = 0; - host_pde = V3_AllocPages(1); + host_pde = V3_VAddr(V3_AllocPages(1)); memset(host_pde, 0, PAGE_SIZE); return (addr_t)host_pde; @@ -257,6 +374,10 @@ addr_t create_new_shadow_pt32() { static void inject_guest_pf(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { + if (info->enable_profiler) { + info->profiler.guest_pf_cnt++; + } + info->ctrl_regs.cr2 = fault_addr; v3_raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code); } @@ -275,14 +396,14 @@ static int is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shado if (guest_access != PT_ACCESS_OK) { // Guest Access Error - if ((shadow_access != PT_ENTRY_NOT_PRESENT) && - (guest_access != PT_ENTRY_NOT_PRESENT)) { + if ((shadow_access != PT_ACCESS_NOT_PRESENT) && + (guest_access != PT_ACCESS_NOT_PRESENT)) { // aka (guest permission error) return 1; } - if ((shadow_access == PT_ENTRY_NOT_PRESENT) && - (guest_access == PT_ENTRY_NOT_PRESENT)) { + if ((shadow_access == PT_ACCESS_NOT_PRESENT) && + (guest_access == PT_ACCESS_NOT_PRESENT)) { // Page tables completely blank, handle guest first return 1; } @@ -296,107 +417,84 @@ static int is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shado -/* The guest status checks have already been done, - * only special case shadow checks remain +/* + * * + * * + * * 64 bit Page table fault handlers + * * + * * */ -static int handle_large_pagefault32(struct guest_info * info, - addr_t fault_addr, pf_error_t error_code, - pte32_t * shadow_pt, pde32_4MB_t * large_guest_pde) -{ - pt_access_status_t shadow_pte_access = can_access_pte32(shadow_pt, fault_addr, error_code); - pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(fault_addr)]); - - if (shadow_pte_access == PT_ACCESS_OK) { - // Inconsistent state... - // Guest Re-Entry will flush tables and everything should now workd - PrintDebug("Inconsistent state... Guest re-entry should flush tlb\n"); - return 0; - } +static int handle_shadow_pagefault_64(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { + pt_access_status_t guest_access; + pt_access_status_t shadow_access; + int ret; + PrintDebug("64 bit shadow page fault\n"); + + ret = v3_check_guest_pt_32(info, info->shdw_pg_state.guest_cr3, fault_addr, error_code, &guest_access); + + PrintDebug("Guest Access Check: %d (access=%d)\n", ret, guest_access); + + ret = v3_check_host_pt_32(info->ctrl_regs.cr3, fault_addr, error_code, &shadow_access); + + PrintDebug("Shadow Access Check: %d (access=%d)\n", ret, shadow_access); - if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) { - // Get the guest physical address of the fault - addr_t guest_fault_pa = PDE32_4MB_T_ADDR(*large_guest_pde) + PD32_4MB_PAGE_OFFSET(fault_addr); - host_region_type_t host_page_type = get_shadow_addr_type(info, guest_fault_pa); - - if (host_page_type == HOST_REGION_INVALID) { - // Inject a machine check in the guest - PrintDebug("Invalid Guest Address in page table (0x%x)\n", guest_fault_pa); - v3_raise_exception(info, MC_EXCEPTION); - return 0; - } + PrintError("64 bit shadow paging not implemented\n"); + return -1; +} - if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) { - struct shadow_page_state * state = &(info->shdw_pg_state); - addr_t shadow_pa = get_shadow_addr(info, guest_fault_pa); - shadow_pte->page_base_addr = PT32_BASE_ADDR(shadow_pa); +/* + * * + * * + * * 32 bit PAE Page table fault handlers + * * + * * + */ - shadow_pte->present = 1; +static int handle_shadow_pagefault_32pae(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { + PrintError("32 bit PAE shadow paging not implemented\n"); + return -1; +} - /* We are assuming that the PDE entry has precedence - * so the Shadow PDE will mirror the guest PDE settings, - * and we don't have to worry about them here - * Allow everything - */ - shadow_pte->user_page = 1; - if (find_pte_map(state->cached_ptes, PT32_PAGE_ADDR(guest_fault_pa)) != NULL) { - // Check if the entry is a page table... - PrintDebug("Marking page as Guest Page Table (large page)\n"); - shadow_pte->vmm_info = PT32_GUEST_PT; - shadow_pte->writable = 0; - } else { - shadow_pte->writable = 1; - } - //set according to VMM policy - shadow_pte->write_through = 0; - shadow_pte->cache_disable = 0; - shadow_pte->global_page = 0; - // - - } else { - // Handle hooked pages as well as other special pages - if (handle_special_page_fault(info, fault_addr, guest_fault_pa, error_code) == -1) { - PrintError("Special Page Fault handler returned error for address: %x\n", fault_addr); - return -1; - } - } - } else if ((shadow_pte_access == PT_WRITE_ERROR) && - (shadow_pte->vmm_info == PT32_GUEST_PT)) { - struct shadow_page_state * state = &(info->shdw_pg_state); - PrintDebug("Write operation on Guest PAge Table Page (large page)\n"); - state->cached_cr3 = 0; - shadow_pte->writable = 1; - } else { - PrintError("Error in large page fault handler...\n"); - PrintError("This case should have been handled at the top level handler\n"); - return -1; - } - PrintDebug("Returning from large page fault handler\n"); - return 0; -} +/* + * * + * * + * * 32 bit Page table fault handlers + * * + * * + */ +static int handle_large_pagefault_32(struct guest_info * info, + addr_t fault_addr, pf_error_t error_code, + pte32_t * shadow_pt, pde32_4MB_t * large_guest_pde); + +static int handle_shadow_pte32_fault(struct guest_info * info, + addr_t fault_addr, + pf_error_t error_code, + pte32_t * shadow_pt, + pte32_t * guest_pt); -static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { +static int handle_shadow_pagefault_32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { pde32_t * guest_pd = NULL; - pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32(info->shdw_pg_state.shadow_cr3); - addr_t guest_cr3 = CR3_TO_PDE32(info->shdw_pg_state.guest_cr3); + pde32_t * shadow_pd = CR3_TO_PDE32_VA(info->ctrl_regs.cr3); + addr_t guest_cr3 = CR3_TO_PDE32_PA(info->shdw_pg_state.guest_cr3); pt_access_status_t guest_pde_access; pt_access_status_t shadow_pde_access; pde32_t * guest_pde = NULL; pde32_t * shadow_pde = (pde32_t *)&(shadow_pd[PDE32_INDEX(fault_addr)]); - PrintDebug("Shadow page fault handler\n"); + PrintDebug("Shadow page fault handler: %p\n", (void*) fault_addr ); if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pd) == -1) { - PrintError("Invalid Guest PDE Address: 0x%x\n", guest_cr3); + PrintError("Invalid Guest PDE Address: 0x%p\n", (void *)guest_cr3); return -1; } @@ -404,23 +502,23 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr // Check the guest page permissions - guest_pde_access = can_access_pde32(guest_pd, fault_addr, error_code); + guest_pde_access = v3_can_access_pde32(guest_pd, fault_addr, error_code); // Check the shadow page permissions - shadow_pde_access = can_access_pde32(shadow_pd, fault_addr, error_code); + shadow_pde_access = v3_can_access_pde32(shadow_pd, fault_addr, error_code); /* Was the page fault caused by the Guest's page tables? */ if (is_guest_pf(guest_pde_access, shadow_pde_access) == 1) { PrintDebug("Injecting PDE pf to guest: (guest access error=%d) (pf error code=%d)\n", - guest_pde_access, error_code); + *(uint_t *)&guest_pde_access, *(uint_t *)&error_code); inject_guest_pf(info, fault_addr, error_code); return 0; } - if (shadow_pde_access == PT_ENTRY_NOT_PRESENT) + if (shadow_pde_access == PT_ACCESS_NOT_PRESENT) { - pte32_t * shadow_pt = (pte32_t *)create_new_shadow_pt32(); + pte32_t * shadow_pt = (pte32_t *)v3_create_new_shadow_pt(); shadow_pde->present = 1; shadow_pde->user_page = guest_pde->user_page; @@ -436,11 +534,25 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr guest_pde->accessed = 1; - shadow_pde->pt_base_addr = PD32_BASE_ADDR((addr_t)shadow_pt); + shadow_pde->pt_base_addr = PAGE_BASE_ADDR((addr_t)V3_PAddr(shadow_pt)); if (guest_pde->large_page == 0) { + pte32_t * guest_pt = NULL; shadow_pde->writable = guest_pde->writable; + + if (guest_pa_to_host_va(info, BASE_TO_PAGE_ADDR(guest_pde->pt_base_addr), (addr_t*)&guest_pt) == -1) { + // Machine check the guest + PrintDebug("Invalid Guest PTE Address: 0x%p\n", (void *)BASE_TO_PAGE_ADDR(guest_pde->pt_base_addr)); + v3_raise_exception(info, MC_EXCEPTION); + return 0; + } + + if (handle_shadow_pte32_fault(info, fault_addr, error_code, shadow_pt, guest_pt) == -1) { + PrintError("Error handling Page fault caused by PTE\n"); + return -1; + } } else { + // ?? What if guest pde is dirty a this point? ((pde32_4MB_t *)guest_pde)->dirty = 0; shadow_pde->writable = 0; } @@ -450,13 +562,14 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr // // PTE fault // - pte32_t * shadow_pt = (pte32_t *)(addr_t)PDE32_T_ADDR((*shadow_pde)); + pte32_t * shadow_pt = (pte32_t *)V3_VAddr( (void*)(addr_t) BASE_TO_PAGE_ADDR(shadow_pde->pt_base_addr) ); if (guest_pde->large_page == 0) { pte32_t * guest_pt = NULL; - if (guest_pa_to_host_va(info, PDE32_T_ADDR((*guest_pde)), (addr_t*)&guest_pt) == -1) { + + if (guest_pa_to_host_va(info, BASE_TO_PAGE_ADDR(guest_pde->pt_base_addr), (addr_t*)&guest_pt) == -1) { // Machine check the guest - PrintDebug("Invalid Guest PTE Address: 0x%x\n", PDE32_T_ADDR((*guest_pde))); + PrintDebug("Invalid Guest PTE Address: 0x%p\n", (void *)BASE_TO_PAGE_ADDR(guest_pde->pt_base_addr)); v3_raise_exception(info, MC_EXCEPTION); return 0; } @@ -466,13 +579,13 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr return -1; } } else if (guest_pde->large_page == 1) { - if (handle_large_pagefault32(info, fault_addr, error_code, shadow_pt, (pde32_4MB_t *)guest_pde) == -1) { + if (handle_large_pagefault_32(info, fault_addr, error_code, shadow_pt, (pde32_4MB_t *)guest_pde) == -1) { PrintError("Error handling large pagefault\n"); return -1; } } } - else if ((shadow_pde_access == PT_WRITE_ERROR) && + else if ((shadow_pde_access == PT_ACCESS_WRITE_ERROR) && (guest_pde->large_page == 1) && (((pde32_4MB_t *)guest_pde)->dirty == 0)) { @@ -487,7 +600,7 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr return 0; } - else if (shadow_pde_access == PT_USER_ERROR) + else if (shadow_pde_access == PT_ACCESS_USER_ERROR) { // // Page Directory Entry marked non-user @@ -505,20 +618,123 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr PrintDebug("Manual Says to inject page fault into guest\n"); #ifdef DEBUG_SHADOW_PAGING PrintDebug("Guest PDE: (access=%d)\n\t", guest_pde_access); - PrintPDE32(fault_addr, guest_pde); + PrintPTEntry(PAGE_PD32, fault_addr, guest_pde); PrintDebug("Shadow PDE: (access=%d)\n\t", shadow_pde_access); - PrintPDE32(fault_addr, shadow_pde); + PrintPTEntry(PAGE_PD32, fault_addr, shadow_pde); #endif return 0; } - PrintDebug("Returning end of PDE function (rip=%x)\n", info->rip); + PrintDebug("Returning end of PDE function (rip=%p)\n", (void *)(addr_t)(info->rip)); return 0; } +/* The guest status checks have already been done, + * only special case shadow checks remain + */ +static int handle_large_pagefault_32(struct guest_info * info, + addr_t fault_addr, pf_error_t error_code, + pte32_t * shadow_pt, pde32_4MB_t * large_guest_pde) +{ + pt_access_status_t shadow_pte_access = v3_can_access_pte32(shadow_pt, fault_addr, error_code); + pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(fault_addr)]); + addr_t guest_fault_pa = BASE_TO_PAGE_ADDR_4MB(large_guest_pde->page_base_addr) + PAGE_OFFSET_4MB(fault_addr); + + struct v3_shadow_region * shdw_reg = v3_get_shadow_region(info, guest_fault_pa); + + + if ((shdw_reg == NULL) || + (shdw_reg->host_type == SHDW_REGION_INVALID)) { + // Inject a machine check in the guest + PrintDebug("Invalid Guest Address in page table (0x%p)\n", (void *)guest_fault_pa); + v3_raise_exception(info, MC_EXCEPTION); + return -1; + } + + if (shadow_pte_access == PT_ACCESS_OK) { + // Inconsistent state... + // Guest Re-Entry will flush tables and everything should now workd + PrintDebug("Inconsistent state... Guest re-entry should flush tlb\n"); + return 0; + } + + + if (shadow_pte_access == PT_ACCESS_NOT_PRESENT) { + // Get the guest physical address of the fault + + if ((shdw_reg->host_type == SHDW_REGION_ALLOCATED) || + (shdw_reg->host_type == SHDW_REGION_WRITE_HOOK)) { + struct shadow_page_state * state = &(info->shdw_pg_state); + addr_t shadow_pa = v3_get_shadow_addr(shdw_reg, guest_fault_pa); + + shadow_pte->page_base_addr = PAGE_BASE_ADDR(shadow_pa); + + shadow_pte->present = 1; + + /* We are assuming that the PDE entry has precedence + * so the Shadow PDE will mirror the guest PDE settings, + * and we don't have to worry about them here + * Allow everything + */ + shadow_pte->user_page = 1; + + if (find_pte_map(state->cached_ptes, PAGE_ADDR(guest_fault_pa)) != NULL) { + // Check if the entry is a page table... + PrintDebug("Marking page as Guest Page Table (large page)\n"); + shadow_pte->vmm_info = PT32_GUEST_PT; + shadow_pte->writable = 0; + } else if (shdw_reg->host_type == SHDW_REGION_WRITE_HOOK) { + shadow_pte->writable = 0; + } else { + shadow_pte->writable = 1; + } + + //set according to VMM policy + shadow_pte->write_through = 0; + shadow_pte->cache_disable = 0; + shadow_pte->global_page = 0; + // + + } else { + // Handle hooked pages as well as other special pages + // if (handle_special_page_fault(info, fault_addr, guest_fault_pa, error_code) == -1) { + + if (v3_handle_mem_full_hook(info, fault_addr, guest_fault_pa, shdw_reg, error_code) == -1) { + PrintError("Special Page Fault handler returned error for address: %p\n", (void *)fault_addr); + return -1; + } + } + } else if (shadow_pte_access == PT_ACCESS_WRITE_ERROR) { + + if (shdw_reg->host_type == SHDW_REGION_WRITE_HOOK) { + + if (v3_handle_mem_wr_hook(info, fault_addr, guest_fault_pa, shdw_reg, error_code) == -1) { + PrintError("Special Page Fault handler returned error for address: %p\n", (void *)fault_addr); + return -1; + } + } else if (shadow_pte->vmm_info == PT32_GUEST_PT) { + struct shadow_page_state * state = &(info->shdw_pg_state); + PrintDebug("Write operation on Guest PAge Table Page (large page)\n"); + state->cached_cr3 = 0; + shadow_pte->writable = 1; + } + + } else { + PrintError("Error in large page fault handler...\n"); + PrintError("This case should have been handled at the top level handler\n"); + return -1; + } + + PrintDebug("Returning from large page fault handler\n"); + return 0; +} + + + + /* * We assume the the guest pte pointer has already been translated to a host virtual address */ @@ -532,19 +748,29 @@ static int handle_shadow_pte32_fault(struct guest_info * info, pt_access_status_t shadow_pte_access; pte32_t * guest_pte = (pte32_t *)&(guest_pt[PTE32_INDEX(fault_addr)]);; pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(fault_addr)]); + addr_t guest_pa = BASE_TO_PAGE_ADDR((addr_t)(guest_pte->page_base_addr)) + PAGE_OFFSET(fault_addr); + + struct v3_shadow_region * shdw_reg = v3_get_shadow_region(info, guest_pa); + if ((shdw_reg == NULL) || + (shdw_reg->host_type == SHDW_REGION_INVALID)) { + // Inject a machine check in the guest + PrintDebug("Invalid Guest Address in page table (0x%p)\n", (void *)guest_pa); + v3_raise_exception(info, MC_EXCEPTION); + return 0; + } // Check the guest page permissions - guest_pte_access = can_access_pte32(guest_pt, fault_addr, error_code); + guest_pte_access = v3_can_access_pte32(guest_pt, fault_addr, error_code); // Check the shadow page permissions - shadow_pte_access = can_access_pte32(shadow_pt, fault_addr, error_code); + shadow_pte_access = v3_can_access_pte32(shadow_pt, fault_addr, error_code); #ifdef DEBUG_SHADOW_PAGING PrintDebug("Guest PTE: (access=%d)\n\t", guest_pte_access); - PrintPTE32(fault_addr, guest_pte); + PrintPTEntry(PAGE_PT32, fault_addr, guest_pte); PrintDebug("Shadow PTE: (access=%d)\n\t", shadow_pte_access); - PrintPTE32(fault_addr, shadow_pte); + PrintPTEntry(PAGE_PT32, fault_addr, shadow_pte); #endif /* Was the page fault caused by the Guest's page tables? */ @@ -564,28 +790,16 @@ static int handle_shadow_pte32_fault(struct guest_info * info, } - if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) { - - addr_t guest_pa = PTE32_T_ADDR((*guest_pte)) + PT32_PAGE_OFFSET(fault_addr); - + if (shadow_pte_access == PT_ACCESS_NOT_PRESENT) { // Page Table Entry Not Present + PrintDebug("guest_pa =%p\n", (void *)guest_pa); - host_region_type_t host_page_type = get_shadow_addr_type(info, guest_pa); - - if (host_page_type == HOST_REGION_INVALID) { - // Inject a machine check in the guest - PrintDebug("Invalid Guest Address in page table (0x%x)\n", guest_pa); - v3_raise_exception(info, MC_EXCEPTION); - return 0; - } - - // else... - - if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) { + if ((shdw_reg->host_type == SHDW_REGION_ALLOCATED) || + (shdw_reg->host_type == SHDW_REGION_WRITE_HOOK)) { struct shadow_page_state * state = &(info->shdw_pg_state); - addr_t shadow_pa = get_shadow_addr(info, guest_pa); + addr_t shadow_pa = v3_get_shadow_addr(shdw_reg, guest_pa); - shadow_pte->page_base_addr = PT32_BASE_ADDR(shadow_pa); + shadow_pte->page_base_addr = PAGE_BASE_ADDR(shadow_pa); shadow_pte->present = guest_pte->present; shadow_pte->user_page = guest_pte->user_page; @@ -598,13 +812,15 @@ static int handle_shadow_pte32_fault(struct guest_info * info, guest_pte->accessed = 1; - if (find_pte_map(state->cached_ptes, PT32_PAGE_ADDR(guest_pa)) != NULL) { + if (find_pte_map(state->cached_ptes, PAGE_ADDR(guest_pa)) != NULL) { // Check if the entry is a page table... - PrintDebug("Marking page as Guest Page Table\n", shadow_pte->writable); + PrintDebug("Marking page as Guest Page Table %d\n", shadow_pte->writable); shadow_pte->vmm_info = PT32_GUEST_PT; } - if (guest_pte->dirty == 1) { + if (shdw_reg->host_type == SHDW_REGION_WRITE_HOOK) { + shadow_pte->writable = 0; + } else if (guest_pte->dirty == 1) { shadow_pte->writable = guest_pte->writable; } else if ((guest_pte->dirty == 0) && (error_code.write == 1)) { shadow_pte->writable = guest_pte->writable; @@ -617,33 +833,41 @@ static int handle_shadow_pte32_fault(struct guest_info * info, state->cached_cr3 = 0; } - } else if ((guest_pte->dirty = 0) && (error_code.write == 0)) { + } else if ((guest_pte->dirty == 0) && (error_code.write == 0)) { // was = shadow_pte->writable = 0; } - - } else { // Page fault handled by hook functions - if (handle_special_page_fault(info, fault_addr, guest_pa, error_code) == -1) { - PrintError("Special Page fault handler returned error for address: %x\n", fault_addr); + + if (v3_handle_mem_full_hook(info, fault_addr, guest_pa, shdw_reg, error_code) == -1) { + PrintError("Special Page fault handler returned error for address: %p\n", (void *)fault_addr); return -1; } } - - } else if ((shadow_pte_access == PT_WRITE_ERROR) && + /* + } else if ((shadow_pte_access == PT_ACCESS_WRITE_ERROR) && (guest_pte->dirty == 0)) { - - PrintDebug("Shadow PTE Write Error\n"); + */ + } else if (shadow_pte_access == PT_ACCESS_WRITE_ERROR) { guest_pte->dirty = 1; - shadow_pte->writable = guest_pte->writable; + + if (shdw_reg->host_type == SHDW_REGION_WRITE_HOOK) { + if (v3_handle_mem_wr_hook(info, fault_addr, guest_pa, shdw_reg, error_code) == -1) { + PrintError("Special Page fault handler returned error for address: %p\n", (void *)fault_addr); + return -1; + } + } else { + PrintDebug("Shadow PTE Write Error\n"); + shadow_pte->writable = guest_pte->writable; + } if (shadow_pte->vmm_info == PT32_GUEST_PT) { struct shadow_page_state * state = &(info->shdw_pg_state); PrintDebug("Write operation on Guest PAge Table Page\n"); state->cached_cr3 = 0; } - + return 0; } else { @@ -664,229 +888,94 @@ static int handle_shadow_pte32_fault(struct guest_info * info, /* Currently Does not work with Segmentation!!! */ -int handle_shadow_invlpg(struct guest_info * info) { +int v3_handle_shadow_invlpg(struct guest_info * info) +{ if (info->mem_mode != VIRTUAL_MEM) { // Paging must be turned on... // should handle with some sort of fault I think PrintError("ERROR: INVLPG called in non paged mode\n"); return -1; } - - - if (info->cpu_mode == PROTECTED) { - uchar_t instr[15]; - int ret; - int index = 0; - - ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - if (ret != 15) { - PrintError("Could not read instruction 0x%x (ret=%d)\n", info->rip, ret); - return -1; - } - - - /* Can INVLPG work with Segments?? */ - while (is_prefix_byte(instr[index])) { - index++; - } - - - if ((instr[index] == (uchar_t)0x0f) && - (instr[index + 1] == (uchar_t)0x01)) { - - addr_t first_operand; - addr_t second_operand; - operand_type_t addr_type; - addr_t guest_cr3 = CR3_TO_PDE32(info->shdw_pg_state.guest_cr3); - - pde32_t * guest_pd = NULL; - - if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pd) == -1) { - PrintError("Invalid Guest PDE Address: 0x%x\n", guest_cr3); - return -1; - } - - - - - index += 2; - - addr_type = decode_operands32(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32); - - if (addr_type == MEM_OPERAND) { - pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32(info->shdw_pg_state.shadow_cr3); - pde32_t * shadow_pde = (pde32_t *)&shadow_pd[PDE32_INDEX(first_operand)]; - pde32_t * guest_pde; - - //PrintDebug("PDE Index=%d\n", PDE32_INDEX(first_operand)); - //PrintDebug("FirstOperand = %x\n", first_operand); - - PrintDebug("Invalidating page for %x\n", first_operand); - - guest_pde = (pde32_t *)&(guest_pd[PDE32_INDEX(first_operand)]); - - if (guest_pde->large_page == 1) { - shadow_pde->present = 0; - PrintDebug("Invalidating Large Page\n"); - } else { - - if (shadow_pde->present == 1) { - pte32_t * shadow_pt = (pte32_t *)(addr_t)PDE32_T_ADDR((*shadow_pde)); - pte32_t * shadow_pte = (pte32_t *)&shadow_pt[PTE32_INDEX(first_operand)]; - -#ifdef DEBUG_SHADOW_PAGING - PrintDebug("Setting not present\n"); - PrintPTE32(first_operand, shadow_pte); -#endif - - shadow_pte->present = 0; - } - } - - info->rip += index; - - } else { - PrintError("Invalid Operand type\n"); - return -1; - } - } else { - PrintError("invalid Instruction Opcode\n"); - PrintTraceMemDump(instr, 15); - return -1; - } - } - - return 0; -} - - -/* - - -static int create_pd32_nonaligned_4MB_page(struct guest_info * info, pte32_t * pt, addr_t guest_addr, pde32_4MB_t * large_shadow_pde) { - uint_t i = 0; - pte32_t * pte_cursor; - addr_t guest_pa = 0; - - for (i = 0; i < 1024; i++) { - guest_pa = guest_addr + (PAGE_SIZE * i); - host_region_type_t host_page_type = get_shadow_addr_type(info, guest_pa); - - pte_cursor = &(pt[i]); - - if (host_page_type == HOST_REGION_INVALID) { - // Currently we don't support this, but in theory we could - PrintError("Invalid Host Memory Type\n"); - return -1; - } else if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) { - addr_t shadow_pa = get_shadow_addr(info, guest_pa); - - - pte_cursor->page_base_addr = PT32_BASE_ADDR(shadow_pa); - pte_cursor->present = 1; - pte_cursor->writable = large_shadow_pde->writable; - pte_cursor->user_page = large_shadow_pde->user_page; - pte_cursor->write_through = 0; - pte_cursor->cache_disable = 0; - pte_cursor->global_page = 0; - - } else { - PrintError("Unsupported Host Memory Type\n"); - return -1; - } - } - return 0; -} - - -static int handle_large_pagefault32(struct guest_info * info, - pde32_t * guest_pde, pde32_t * shadow_pde, - addr_t fault_addr, pf_error_t error_code ) { - struct shadow_region * mem_reg; - pde32_4MB_t * large_guest_pde = (pde32_4MB_t *)guest_pde; - pde32_4MB_t * large_shadow_pde = (pde32_4MB_t *)shadow_pde; - host_region_type_t host_page_type; - addr_t guest_start_addr = PDE32_4MB_T_ADDR(*large_guest_pde); - // addr_t guest_end_addr = guest_start_addr + PAGE_SIZE_4MB; // start address + 4MB - // Check that the Guest PDE entry points to valid memory - // else Machine Check the guest - PrintDebug("Large Page: Page Base Addr=%x\n", guest_start_addr); + if (info->cpu_mode != PROTECTED) { + PrintError("Unsupported CPU mode (mode=%s)\n", v3_cpu_mode_to_str(info->cpu_mode)); + return -1; + } - host_page_type = get_shadow_addr_type(info, guest_start_addr); + uchar_t instr[15]; + int index = 0; - if (host_page_type == HOST_REGION_INVALID) { - PrintError("Invalid guest address in large page (0x%x)\n", guest_start_addr); - v3_raise_exception(info, MC_EXCEPTION); + int ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); + if (ret != 15) { + PrintError("Could not read instruction 0x%p (ret=%d)\n", (void *)(addr_t)(info->rip), ret); return -1; } - // else... - - if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) { - - addr_t host_start_addr = 0; - addr_t region_end_addr = 0; - - // Check for a large enough region in host memory - mem_reg = get_shadow_region_by_addr(&(info->mem_map), guest_start_addr); - PrintDebug("Host region: host_addr=%x (guest_start=%x, end=%x)\n", - mem_reg->host_addr, mem_reg->guest_start, mem_reg->guest_end); - host_start_addr = mem_reg->host_addr + (guest_start_addr - mem_reg->guest_start); - region_end_addr = mem_reg->host_addr + (mem_reg->guest_end - mem_reg->guest_start); - - PrintDebug("Host Start Addr=%x; Region End Addr=%x\n", host_start_addr, region_end_addr); - - - //4f - if (large_guest_pde->dirty == 1) { // dirty - large_shadow_pde->writable = guest_pde->writable; - } else if (error_code.write == 1) { // not dirty, access is write - large_shadow_pde->writable = guest_pde->writable; - large_guest_pde->dirty = 1; - } else { // not dirty, access is read - large_shadow_pde->writable = 0; - } - - - // Check if the region is at least an additional 4MB + + /* Can INVLPG work with Segments?? */ + while (is_prefix_byte(instr[index])) { + index++; + } - //4b. - if ((PD32_4MB_PAGE_OFFSET(host_start_addr) == 0) && - (region_end_addr >= host_start_addr + PAGE_SIZE_4MB)) { // if 4MB boundary - large_shadow_pde->page_base_addr = PD32_4MB_BASE_ADDR(host_start_addr); - } else { // else generate 4k pages - pte32_t * shadow_pt = NULL; - PrintDebug("Handling non aligned large page\n"); - - shadow_pde->large_page = 0; - - shadow_pt = create_new_shadow_pt32(); + if( (instr[index + 0] != (uchar_t) 0x0f) || + (instr[index + 1] != (uchar_t) 0x01) ) { + PrintError("invalid Instruction Opcode\n"); + PrintTraceMemDump(instr, 15); + return -1; + } + + addr_t first_operand; + addr_t second_operand; + addr_t guest_cr3 = CR3_TO_PDE32_PA(info->shdw_pg_state.guest_cr3); + + pde32_t * guest_pd = NULL; + + if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pd) == -1) { + PrintError("Invalid Guest PDE Address: 0x%p\n", (void *)guest_cr3); + return -1; + } + + index += 2; - if (create_pd32_nonaligned_4MB_page(info, shadow_pt, guest_start_addr, large_shadow_pde) == -1) { - PrintError("Non Aligned Large Page Error\n"); - V3_Free(shadow_pt); - return -1; - } - + v3_operand_type_t addr_type = decode_operands32(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32); + + if (addr_type != MEM_OPERAND) { + PrintError("Invalid Operand type\n"); + return -1; + } + + pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32_VA(info->ctrl_regs.cr3); + pde32_t * shadow_pde = (pde32_t *)&shadow_pd[PDE32_INDEX(first_operand)]; + pde32_t * guest_pde; + + //PrintDebug("PDE Index=%d\n", PDE32_INDEX(first_operand)); + //PrintDebug("FirstOperand = %x\n", first_operand); + + PrintDebug("Invalidating page for %p\n", (void *)first_operand); + + guest_pde = (pde32_t *)&(guest_pd[PDE32_INDEX(first_operand)]); + + if (guest_pde->large_page == 1) { + shadow_pde->present = 0; + PrintDebug("Invalidating Large Page\n"); + } else + if (shadow_pde->present == 1) { + pte32_t * shadow_pt = (pte32_t *)(addr_t)BASE_TO_PAGE_ADDR(shadow_pde->pt_base_addr); + pte32_t * shadow_pte = (pte32_t *) V3_VAddr( (void*) &shadow_pt[PTE32_INDEX(first_operand)] ); #ifdef DEBUG_SHADOW_PAGING - PrintDebug("non-aligned Shadow PT\n"); - PrintPT32(PT32_PAGE_ADDR(fault_addr), shadow_pt); + PrintDebug("Setting not present\n"); + PrintPTEntry(PAGE_PT32, first_operand, shadow_pte); #endif - shadow_pde->pt_base_addr = PD32_BASE_ADDR(shadow_pt); - } - - } else { - // Handle hooked pages as well as other special pages - if (handle_special_page_fault(info, fault_addr, guest_start_addr, error_code) == -1) { - PrintError("Special Page Fault handler returned error for address: %x\n", fault_addr); - return -1; + + shadow_pte->present = 0; } - } - + + info->rip += index; + return 0; } -*/ + +