X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmm_shadow_paging.c;h=d3e4c931b37c05a8144b237706b4bd2ca52a35d7;hb=4d17b20a86eb8102a287263372458fcb9019c232;hp=6a0b3be9053632b01c9db701531a00d3428068ee;hpb=53f181096ad02cde9f79e30e2238ddad43ede574;p=palacios.git diff --git a/palacios/src/palacios/vmm_shadow_paging.c b/palacios/src/palacios/vmm_shadow_paging.c index 6a0b3be..d3e4c93 100644 --- a/palacios/src/palacios/vmm_shadow_paging.c +++ b/palacios/src/palacios/vmm_shadow_paging.c @@ -28,307 +28,326 @@ #include -#ifndef DEBUG_SHADOW_PAGING -#undef PrintDebug -#define PrintDebug(fmt, args...) +#include + + + + +#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY +#include #endif +#ifdef V3_CONFIG_SYMBIOTIC_SWAP +#include +#endif -/*** - *** There be dragons - ***/ +#ifndef V3_CONFIG_DEBUG_SHADOW_PAGING +#undef PrintDebug +#define PrintDebug(fmt, args...) +#endif -struct guest_table { - addr_t cr3; - struct list_head link; -}; +static const char default_strategy[] = "VTLB"; -struct backptr { - addr_t ptr; - struct list_head link; -}; +static struct hashtable * master_shdw_pg_table = NULL; +static uint_t shdw_pg_hash_fn(addr_t key) { + char * name = (char *)key; + return v3_hash_buffer((uint8_t *)name, strlen(name)); +} -struct shadow_page_data { - addr_t ptr; - addr_t guest_addr; +static int shdw_pg_eq_fn(addr_t key1, addr_t key2) { + char * name1 = (char *)key1; + char * name2 = (char *)key2; - struct list_head backptrs; - struct list_head guest_tables; -}; + return (strcmp(name1, name2) == 0); +} +int V3_init_shdw_paging() { + extern struct v3_shdw_pg_impl * __start__v3_shdw_pg_impls[]; + extern struct v3_shdw_pg_impl * __stop__v3_shdw_pg_impls[]; + struct v3_shdw_pg_impl ** tmp_impl = __start__v3_shdw_pg_impls; + int i = 0; + master_shdw_pg_table = v3_create_htable(0, shdw_pg_hash_fn, shdw_pg_eq_fn); -//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); + while (tmp_impl != __stop__v3_shdw_pg_impls) { + V3_Print("Registering Shadow Paging Impl (%s)\n", (*tmp_impl)->name); -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); + if (v3_htable_search(master_shdw_pg_table, (addr_t)((*tmp_impl)->name))) { + PrintError("Multiple instances of shadow paging impl (%s)\n", (*tmp_impl)->name); + return -1; + } + if (v3_htable_insert(master_shdw_pg_table, + (addr_t)((*tmp_impl)->name), + (addr_t)(*tmp_impl)) == 0) { + PrintError("Could not register shadow paging impl (%s)\n", (*tmp_impl)->name); + return -1; + } + tmp_impl = &(__start__v3_shdw_pg_impls[++i]); + } -static uint_t pte_hash_fn(addr_t key) { - return hash_long(key, 32); + return 0; } -static int pte_equals(addr_t key1, addr_t key2) { - return (key1 == key2); +int V3_deinit_shdw_paging() { + v3_free_htable(master_shdw_pg_table, 0, 0); + return 0; } -static addr_t create_new_shadow_pt(); -static void inject_guest_pf(struct guest_info * info, addr_t fault_addr, pf_error_t error_code); -static int is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shadow_access); - - -#include "vmm_shadow_paging_32.h" -#include "vmm_shadow_paging_32pae.h" -#include "vmm_shadow_paging_64.h" +/*** + *** There be dragons + ***/ -int v3_init_shadow_page_state(struct guest_info * info) { - struct shadow_page_state * state = &(info->shdw_pg_state); - - state->guest_cr3 = 0; - state->guest_cr0 = 0; - state->cached_ptes = NULL; +#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY +static void telemetry_cb(struct v3_vm_info * vm, void * private_data, char * hdr) { + int i = 0; + for (i = 0; i < vm->num_cores; i++) { + struct guest_info * core = &(vm->cores[i]); - return 0; + V3_Print("%s Guest Page faults: %d\n", hdr, core->shdw_pg_state.guest_faults); + } } +#endif +int v3_init_shdw_pg_state(struct guest_info * core) { + struct v3_shdw_pg_state * state = &(core->shdw_pg_state); + struct v3_shdw_pg_impl * impl = core->vm_info->shdw_impl.current_impl; + + state->guest_cr3 = 0; + state->guest_cr0 = 0; + state->guest_efer.value = 0x0LL; + if (impl->local_init(core) == -1) { + PrintError("Error in Shadow paging local initialization (%s)\n", impl->name); + 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; + +#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY + v3_add_telemetry_cb(core->vm_info, telemetry_cb, NULL); +#endif + + 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 +int v3_deinit_shdw_pg_state(struct guest_info * core) { + struct v3_shdw_pg_impl * impl = core->vm_info->shdw_impl.current_impl; - info->ctrl_regs.cr3 = *(addr_t*)&(info->direct_map_pt); - //PrintError("Activate Passthrough Page tables not implemented\n"); - return 0; + if (impl->local_deinit(core) == -1) { + PrintError("Error deinitializing shadow paging state\n"); + return -1; + } + + + return 0; } -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 +int v3_init_shdw_impl(struct v3_vm_info * vm) { + struct v3_shdw_impl_state * impl_state = &(vm->shdw_impl); + v3_cfg_tree_t * pg_cfg = v3_cfg_subtree(vm->cfg_data->cfg, "paging"); + char * pg_mode = v3_cfg_val(pg_cfg, "mode"); + char * pg_strat = v3_cfg_val(pg_cfg, "strategy"); + struct v3_shdw_pg_impl * impl = NULL; + + PrintDebug("Checking if shadow paging requested.\n"); + if ((pg_mode != NULL) && (strcasecmp(pg_mode, "nested") == 0)) { + PrintDebug("Nested paging specified - not initializing shadow paging.\n"); + return 0; + } -#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 + if (pg_strat == NULL) { + pg_strat = (char *)default_strategy; + } + + V3_Print("Initialization of Shadow Paging implementation\n"); + + impl = (struct v3_shdw_pg_impl *)v3_htable_search(master_shdw_pg_table, (addr_t)pg_strat); - 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_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; + if (impl == NULL) { + PrintError("Could not find shadow paging impl (%s)\n", pg_strat); + return -1; } - } else { - PrintError("Invalid Memory mode\n"); - return -1; - } + + impl_state->current_impl = impl; + + if (impl->init(vm, pg_cfg) == -1) { + PrintError("Could not initialize Shadow paging implemenation (%s)\n", impl->name); + return -1; + } + + return 0; } +int v3_deinit_shdw_impl(struct v3_vm_info * vm) { + struct v3_shdw_pg_impl * impl = vm->shdw_impl.current_impl; + if (impl == NULL) { + // Shadow paging not implemented + return 0; + } + + if (impl->deinit(vm) == -1) { + PrintError("Error deinitializing shadow paging implementation\n"); + return -1; + } -static addr_t create_new_shadow_pt() { - void * host_pde = 0; + return 0; +} - host_pde = V3_VAddr(V3_AllocPages(1)); - memset(host_pde, 0, PAGE_SIZE); - return (addr_t)host_pde; +// 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 * core) { + struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl); + struct v3_shdw_pg_impl * impl = state->current_impl; + return impl->activate_shdw_pt(core); } -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); +// This must flush any caches +// and reset the cr3 value to the correct value +int v3_invalidate_shadow_pts(struct guest_info * core) { + struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl); + struct v3_shdw_pg_impl * impl = state->current_impl; + return impl->invalidate_shdw_pt(core); } -static int is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shadow_access) { - /* basically the reasoning is that there can be multiple reasons for a page fault: - If there is a permissions failure for a page present in the guest _BUT_ - the reason for the fault was that the page is not present in the shadow, - _THEN_ we have to map the shadow page in and reexecute, this will generate - a permissions fault which is _THEN_ valid to send to the guest - _UNLESS_ both the guest and shadow have marked the page as not present - - whew... - */ - if (guest_access != PT_ACCESS_OK) { - // Guest Access Error - - if ((shadow_access != PT_ACCESS_NOT_PRESENT) && - (guest_access != PT_ACCESS_NOT_PRESENT)) { - // aka (guest permission error) - return 1; +int v3_handle_shadow_pagefault(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) { + + if (v3_get_vm_mem_mode(core) == PHYSICAL_MEM) { + // If paging is not turned on we need to handle the special cases + return v3_handle_passthrough_pagefault(core, fault_addr, error_code); + } else if (v3_get_vm_mem_mode(core) == VIRTUAL_MEM) { + struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl); + struct v3_shdw_pg_impl * impl = state->current_impl; + + return impl->handle_pagefault(core, fault_addr, error_code); + } else { + PrintError("Invalid Memory mode\n"); + return -1; } +} - if ((shadow_access == PT_ACCESS_NOT_PRESENT) && - (guest_access == PT_ACCESS_NOT_PRESENT)) { - // Page tables completely blank, handle guest first - return 1; - } - // Otherwise we'll handle the guest fault later...? - } +int v3_handle_shadow_invlpg(struct guest_info * core) { + uchar_t instr[15]; + struct x86_instr dec_instr; + int ret = 0; + addr_t vaddr = 0; - return 0; -} + if (v3_get_vm_mem_mode(core) != 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 (v3_get_vm_mem_mode(core) == PHYSICAL_MEM) { + ret = v3_read_gpa_memory(core, get_addr_linear(core, core->rip, &(core->segments.cs)), 15, instr); + } else { + ret = v3_read_gva_memory(core, get_addr_linear(core, core->rip, &(core->segments.cs)), 15, instr); + } + if (ret == -1) { + PrintError("Could not read instruction into buffer\n"); + return -1; + } + if (v3_decode(core, (addr_t)instr, &dec_instr) == -1) { + PrintError("Decoding Error\n"); + return -1; + } + + if ((dec_instr.op_type != V3_OP_INVLPG) || + (dec_instr.num_operands != 1) || + (dec_instr.dst_operand.type != MEM_OPERAND)) { + PrintError("Decoder Error: Not a valid INVLPG instruction...\n"); + return -1; + } + vaddr = dec_instr.dst_operand.operand; + core->rip += dec_instr.instr_length; + { + struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl); + struct v3_shdw_pg_impl * impl = state->current_impl; + return impl->handle_invlpg(core, vaddr); + } +} -/* 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) { - PrintError("Unsupported CPU mode (mode=%s)\n", v3_cpu_mode_to_str(info->cpu_mode)); - return -1; - } - - 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 = 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)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("Setting not present\n"); - PrintPTEntry(PAGE_PT32, first_operand, shadow_pte); +int v3_inject_guest_pf(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) { + core->ctrl_regs.cr2 = fault_addr; + +#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY + core->shdw_pg_state.guest_faults++; #endif - - shadow_pte->present = 0; + + return v3_raise_exception_with_error(core, PF_EXCEPTION, *(uint_t *)&error_code); +} + + +int v3_is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shadow_access) { + /* basically the reasoning is that there can be multiple reasons for a page fault: + If there is a permissions failure for a page present in the guest _BUT_ + the reason for the fault was that the page is not present in the shadow, + _THEN_ we have to map the shadow page in and reexecute, this will generate + a permissions fault which is _THEN_ valid to send to the guest + _UNLESS_ both the guest and shadow have marked the page as not present + + whew... + */ + if (guest_access != PT_ACCESS_OK) { + // Guest Access Error + + if ((shadow_access != PT_ACCESS_NOT_PRESENT) && + (guest_access != PT_ACCESS_NOT_PRESENT)) { + // aka (guest permission error) + return 1; + } + + /* + if ((shadow_access == PT_ACCESS_NOT_PRESENT) && + (guest_access == PT_ACCESS_NOT_PRESENT)) { + // Page tables completely blank, handle guest first + return 1; + } + */ + + if (guest_access == PT_ACCESS_NOT_PRESENT) { + // Page tables completely blank, handle guest first + return 1; + } + + // Otherwise we'll handle the guest fault later...? } - - info->rip += index; - - return 0; + + return 0; }