X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmm_shadow_paging.c;h=63d0e92efc836327f9e9b9777b9f3ba3b14aea60;hb=9b31f917eae9b397cb21ff78d81084301b289e43;hp=faa48e1c26a0894d658de8318be87e538308943d;hpb=62406cf6b87a27a62921ce09a0aa44780ea25c06;p=palacios.git diff --git a/palacios/src/palacios/vmm_shadow_paging.c b/palacios/src/palacios/vmm_shadow_paging.c index faa48e1..63d0e92 100644 --- a/palacios/src/palacios/vmm_shadow_paging.c +++ b/palacios/src/palacios/vmm_shadow_paging.c @@ -24,6 +24,7 @@ #include #include #include +#include #ifndef DEBUG_SHADOW_PAGING #undef PrintDebug @@ -50,7 +51,6 @@ DEFINE_HASHTABLE_REMOVE(del_pte_map, addr_t, addr_t, 0); - static uint_t pte_hash_fn(addr_t key) { return hash_long(key, 32); } @@ -68,20 +68,21 @@ 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 v3_init_shadow_page_state(struct guest_info * info) { struct shadow_page_state * state = &(info->shdw_pg_state); state->guest_cr3 = 0; - state->shadow_cr3 = 0; - + state->guest_cr0 = 0; state->cr3_cache = create_hashtable(0, &cr3_hash_fn, &cr3_equals); @@ -144,6 +145,16 @@ int cache_page_tables32(struct guest_info * info, addr_t pde) { } */ + +int v3_cache_page_tables(struct guest_info * info, addr_t cr3) { + switch(v3_get_cpu_mode(info)) { + case PROTECTED: + return v3_cache_page_tables32(info, CR3_TO_PDE32_PA(cr3)); + default: + return -1; + } +} + int v3_cache_page_tables32(struct guest_info * info, addr_t pde) { struct shadow_page_state * state = &(info->shdw_pg_state); addr_t pde_host_addr; @@ -195,7 +206,7 @@ int v3_cache_page_tables32(struct guest_info * info, addr_t pde) { 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) { @@ -220,6 +231,84 @@ int v3_replace_shdw_page32(struct guest_info * info, addr_t location, pte32_t * +// 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 = v3_cache_page_tables32(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)PD32_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(struct guest_info * info) { + PrintError("Activating 64 bit page tables not implemented\n"); + return -1; +} + + +// 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 v3_handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { @@ -235,10 +324,13 @@ int v3_handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_e 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; @@ -299,98 +391,61 @@ 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; - } - - 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); - +static int handle_shadow_pagefault_64(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { + PrintError("64 bit shadow paging not implemented\n"); + return -1; +} - if (host_page_type == HOST_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 0; - } - 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); +/* + * * + * * + * * 32 bit PAE Page table fault handlers + * * + * * + */ - shadow_pte->page_base_addr = PT32_BASE_ADDR(shadow_pa); +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; +} - 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, 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: %p\n", (void *)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; - } +/* + * * + * * + * * 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); - PrintDebug("Returning from large page fault handler\n"); - return 0; -} +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 = (addr_t) V3_PAddr( 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; @@ -470,7 +525,7 @@ 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; } @@ -523,6 +578,96 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr +/* 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 = 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; + } + + + 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%p)\n", (void *)guest_fault_pa); + v3_raise_exception(info, MC_EXCEPTION); + return 0; + } + + 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); + + 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, 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: %p\n", (void *)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; +} + + + + /* * We assume the the guest pte pointer has already been translated to a host virtual address */ @@ -573,6 +718,7 @@ static int handle_shadow_pte32_fault(struct guest_info * info, addr_t guest_pa = PTE32_T_ADDR((*guest_pte)) + PT32_PAGE_OFFSET(fault_addr); // 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); @@ -670,92 +816,91 @@ static int handle_shadow_pte32_fault(struct guest_info * info, /* Currently Does not work with Segmentation!!! */ 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) - return 0; - - uchar_t instr[15]; - int index = 0; - - 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; - } - - - /* Can INVLPG work with Segments?? */ - while (is_prefix_byte(instr[index])) { - index++; - } + 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) { + return 0; + } + + uchar_t instr[15]; + int index = 0; + + 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; + } + + + /* Can INVLPG work with Segments?? */ + while (is_prefix_byte(instr[index])) { + index++; + } - 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 = (addr_t)V3_PAddr( (void*)(addr_t) 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%p\n", (void *)guest_cr3); - return -1; - } - - index += 2; - - 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(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 %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)PDE32_T_ADDR((*shadow_pde)); - pte32_t * shadow_pte = (pte32_t *) V3_VAddr( (void*) &shadow_pt[PTE32_INDEX(first_operand)] ); + 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; + 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)PDE32_T_ADDR((*shadow_pde)); + pte32_t * shadow_pte = (pte32_t *) V3_VAddr( (void*) &shadow_pt[PTE32_INDEX(first_operand)] ); + #ifdef DEBUG_SHADOW_PAGING - PrintDebug("Setting not present\n"); - PrintPTE32(first_operand, shadow_pte ); + PrintDebug("Setting not present\n"); + PrintPTE32(first_operand, shadow_pte ); #endif - - shadow_pte->present = 0; - } - - info->rip += index; - - return 0; + + shadow_pte->present = 0; + } + + info->rip += index; + + return 0; }