X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?p=palacios.git;a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmm_ctrl_regs.c;h=d3e9360e6d68294297ddd34edc1f3b5e0827aeed;hp=2fcd3b640ef6fd633440afe75ab4a6f383f9b49f;hb=0e097100a26bc43eb8964734fa43130fc4c71429;hpb=aad69eda8e2354cf146054e57ad25fe201a1cb52 diff --git a/palacios/src/palacios/vmm_ctrl_regs.c b/palacios/src/palacios/vmm_ctrl_regs.c index 2fcd3b6..d3e9360 100644 --- a/palacios/src/palacios/vmm_ctrl_regs.c +++ b/palacios/src/palacios/vmm_ctrl_regs.c @@ -1,829 +1,575 @@ +/* + * This file is part of the Palacios Virtual Machine Monitor developed + * by the V3VEE Project with funding from the United States National + * Science Foundation and the Department of Energy. + * + * The V3VEE Project is a joint project between Northwestern University + * and the University of New Mexico. You can find out more at + * http://www.v3vee.org + * + * Copyright (c) 2008, Jack Lange + * Copyright (c) 2008, The V3VEE Project + * All rights reserved. + * + * Author: Jack Lange + * + * This is free software. You are permitted to use, + * redistribute, and modify it as specified in the file "V3VEE_LICENSE". + */ + #include #include #include #include #include #include +#include +#ifndef CONFIG_DEBUG_CTRL_REGS +#undef PrintDebug +#define PrintDebug(fmt, args...) +#endif -/* Segmentation is a problem here... - * - * When we get a memory operand, presumably we use the default segment (which is?) - * unless an alternate segment was specfied in the prefix... - */ +static int handle_lmsw(struct guest_info * info, struct x86_instr * dec_instr); +static int handle_clts(struct guest_info * info, struct x86_instr * dec_instr); +static int handle_mov_to_cr0(struct guest_info * info, struct x86_instr * dec_instr); -#ifndef DEBUG_CTRL_REGS -#undef PrintDebug -#define PrintDebug(fmt, args...) -#endif +// First Attempt = 494 lines +// current = 106 lines +int v3_handle_cr0_write(struct guest_info * info) { + uchar_t instr[15]; + int ret; + struct x86_instr dec_instr; + + if (info->mem_mode == PHYSICAL_MEM) { + ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); + } else { + ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); + } + + if (v3_decode(info, (addr_t)instr, &dec_instr) == -1) { + PrintError("Could not decode instruction\n"); + return -1; + } + + if (dec_instr.op_type == V3_OP_LMSW) { + if (handle_lmsw(info, &dec_instr) == -1) { + return -1; + } + } else if (dec_instr.op_type == V3_OP_MOV2CR) { + if (handle_mov_to_cr0(info, &dec_instr) == -1) { + return -1; + } + } else if (dec_instr.op_type == V3_OP_CLTS) { + if (handle_clts(info, &dec_instr) == -1) { + return -1; + } + } else { + PrintError("Unhandled opcode in handle_cr0_write\n"); + return -1; + } + + info->rip += dec_instr.instr_length; + + return 0; +} +// The CR0 register only has flags in the low 32 bits +// The hardware does a format check to make sure the high bits are zero +// Because of this we can ignore the high 32 bits here +static int handle_mov_to_cr0(struct guest_info * info, struct x86_instr * dec_instr) { + // 32 bit registers + struct cr0_32 * shadow_cr0 = (struct cr0_32 *)&(info->ctrl_regs.cr0); + struct cr0_32 * new_cr0 = (struct cr0_32 *)(dec_instr->src_operand.operand); + struct cr0_32 * guest_cr0 = (struct cr0_32 *)&(info->shdw_pg_state.guest_cr0); + uint_t paging_transition = 0; + + PrintDebug("MOV2CR0 (MODE=%s)\n", v3_cpu_mode_to_str(info->cpu_mode)); + + PrintDebug("OperandVal = %x, length=%d\n", *(uint_t *)new_cr0, dec_instr->src_operand.size); + + PrintDebug("Old CR0=%x\n", *(uint_t *)shadow_cr0); + PrintDebug("Old Guest CR0=%x\n", *(uint_t *)guest_cr0); + + + // We detect if this is a paging transition + if (guest_cr0->pg != new_cr0->pg) { + paging_transition = 1; + } + + // Guest always sees the value they wrote + *guest_cr0 = *new_cr0; + + // This value must always be set to 1 + guest_cr0->et = 1; + + // Set the shadow register to catch non-virtualized flags + *shadow_cr0 = *guest_cr0; + + // Paging is always enabled + shadow_cr0->pg = 1; + + // Was there a paging transition + // Meaning we need to change the page tables + if (paging_transition) { + if (v3_get_vm_mem_mode(info) == VIRTUAL_MEM) { + + struct efer_64 * guest_efer = (struct efer_64 *)&(info->shdw_pg_state.guest_efer); + struct efer_64 * shadow_efer = (struct efer_64 *)&(info->ctrl_regs.efer); + + // Check long mode LME to set LME + if (guest_efer->lme == 1) { + PrintDebug("Enabing Long Mode\n"); + guest_efer->lma = 1; + + shadow_efer->lma = 1; + shadow_efer->lme = 1; + + PrintDebug("New EFER %p\n", (void *)*(addr_t *)(shadow_efer)); + } + + PrintDebug("Activating Shadow Page Tables\n"); + + if (v3_activate_shadow_pt(info) == -1) { + PrintError("Failed to activate shadow page tables\n"); + return -1; + } + } else { + shadow_cr0->wp = 1; + + if (v3_activate_passthrough_pt(info) == -1) { + PrintError("Failed to activate passthrough page tables\n"); + return -1; + } + } + } + + + PrintDebug("New Guest CR0=%x\n",*(uint_t *)guest_cr0); + PrintDebug("New CR0=%x\n", *(uint_t *)shadow_cr0); + + return 0; +} -int handle_cr0_write(struct guest_info * info) { - char instr[15]; - int ret; - struct x86_instr dec_instr; - if (info->mem_mode == PHYSICAL_MEM) { - ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - } else { - ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - } - if (ret != 15) { - // I think we should inject a GPF into the guest - PrintError("Could not read instruction (ret=%d)\n", ret); - return -1; - } - if (v3_decode(info, (addr_t)instr, &dec_instr) == -1) { - PrintError("Could not decode instruction\n"); - return -1; - } +static int handle_clts(struct guest_info * info, struct x86_instr * dec_instr) { + // CLTS + struct cr0_32 * real_cr0 = (struct cr0_32*)&(info->ctrl_regs.cr0); + + real_cr0->ts = 0; + + if (info->shdw_pg_mode == SHADOW_PAGING) { + struct cr0_32 * guest_cr0 = (struct cr0_32 *)&(info->shdw_pg_state.guest_cr0); + guest_cr0->ts = 0; + } + return 0; +} - if (opcode_cmp(V3_OPCODE_LMSW, (const uchar_t *)(dec_instr.opcode)) == 0) { - struct cr0_real *real_cr0 = (struct cr0_real*)&(info->ctrl_regs.cr0); - struct cr0_real *new_cr0 = (struct cr0_real *)(dec_instr.first_operand.operand); +static int handle_lmsw(struct guest_info * info, struct x86_instr * dec_instr) { + struct cr0_real * real_cr0 = (struct cr0_real *)&(info->ctrl_regs.cr0); + // XED is a mess, and basically reverses the operand order for an LMSW + struct cr0_real * new_cr0 = (struct cr0_real *)(dec_instr->dst_operand.operand); uchar_t new_cr0_val; - + PrintDebug("LMSW\n"); - + new_cr0_val = (*(char*)(new_cr0)) & 0x0f; PrintDebug("OperandVal = %x\n", new_cr0_val); - - PrintDebug("Old CR0=%x\n", *real_cr0); + + // We can just copy the new value through + // we don't need to virtualize the lower 4 bits + PrintDebug("Old CR0=%x\n", *(uint_t *)real_cr0); *(uchar_t*)real_cr0 &= 0xf0; *(uchar_t*)real_cr0 |= new_cr0_val; - PrintDebug("New CR0=%x\n", *real_cr0); - - + PrintDebug("New CR0=%x\n", *(uint_t *)real_cr0); + + + // If Shadow paging is enabled we push the changes to the virtualized copy of cr0 if (info->shdw_pg_mode == SHADOW_PAGING) { - struct cr0_real * shadow_cr0 = (struct cr0_real*)&(info->shdw_pg_state.guest_cr0); - - PrintDebug(" Old Shadow CR0=%x\n", *shadow_cr0); - *(uchar_t*)shadow_cr0 &= 0xf0; - *(uchar_t*)shadow_cr0 |= new_cr0_val; - PrintDebug("New Shadow CR0=%x\n", *shadow_cr0); - } - } else if (opcode_cmp(V3_OPCODE_MOV2CR, (const uchar_t *)(dec_instr.opcode)) == 0) { - PrintDebug("MOV2CR0\n"); - - if (info->cpu_mode == LONG) { - // 64 bit registers - } else { - // 32 bit registers - struct cr0_32 *real_cr0 = (struct cr0_32*)&(info->ctrl_regs.cr0); - struct cr0_32 *new_cr0= (struct cr0_32 *)(dec_instr.second_operand.operand); - - PrintDebug("OperandVal = %x, length=%d\n", *new_cr0, dec_instr.first_operand.size); - - - PrintDebug("Old CR0=%x\n", *real_cr0); - *real_cr0 = *new_cr0; + struct cr0_real * guest_cr0 = (struct cr0_real*)&(info->shdw_pg_state.guest_cr0); - - if (info->shdw_pg_mode == SHADOW_PAGING) { - struct cr0_32 * shadow_cr0 = (struct cr0_32 *)&(info->shdw_pg_state.guest_cr0); - - PrintDebug("Old Shadow CR0=%x\n", *shadow_cr0); - - real_cr0->et = 1; - - *shadow_cr0 = *new_cr0; - shadow_cr0->et = 1; - - if (get_mem_mode(info) == VIRTUAL_MEM) { - struct cr3_32 * shadow_cr3 = (struct cr3_32 *)&(info->shdw_pg_state.shadow_cr3); - - info->ctrl_regs.cr3 = *(addr_t*)shadow_cr3; - } else { - info->ctrl_regs.cr3 = *(addr_t*)&(info->direct_map_pt); - real_cr0->pg = 1; - } - - PrintDebug("New Shadow CR0=%x\n",*shadow_cr0); - } - PrintDebug("New CR0=%x\n", *real_cr0); - } - - } else if (opcode_cmp(V3_OPCODE_CLTS, (const uchar_t *)(dec_instr.opcode)) == 0) { - // CLTS - struct cr0_32 *real_cr0 = (struct cr0_32*)&(info->ctrl_regs.cr0); - - real_cr0->ts = 0; - - if (info->shdw_pg_mode == SHADOW_PAGING) { - struct cr0_32 * shadow_cr0 = (struct cr0_32 *)&(info->shdw_pg_state.guest_cr0); - shadow_cr0->ts = 0; + PrintDebug("Old Guest CR0=%x\n", *(uint_t *)guest_cr0); + *(uchar_t*)guest_cr0 &= 0xf0; + *(uchar_t*)guest_cr0 |= new_cr0_val; + PrintDebug("New Guest CR0=%x\n", *(uint_t *)guest_cr0); } - } - - info->rip += dec_instr.instr_length; - - return 0; + return 0; } -int handle_cr0_read(struct guest_info * info) { - char instr[15]; - - switch (info->cpu_mode) { - - case REAL: - { - int index = 0; - int ret; - PrintDebug("Real Mode read from CR0 at linear guest pa 0x%x\n",get_addr_linear(info,info->rip,&(info->segments.cs))); - //PrintV3Segments(info); - // The real rip address is actually a combination of the rip + CS base - ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - if (ret != 15) { - // I think we should inject a GPF into the guest - PrintDebug("Could not read Real Mode instruction (ret=%d)\n", ret); +// First attempt = 253 lines +// current = 51 lines +int v3_handle_cr0_read(struct guest_info * info) { + uchar_t instr[15]; + int ret; + struct x86_instr dec_instr; + + if (info->mem_mode == PHYSICAL_MEM) { + ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); + } else { + ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); + } + + + if (v3_decode(info, (addr_t)instr, &dec_instr) == -1) { + PrintError("Could not decode instruction\n"); return -1; - } - - - while (is_prefix_byte(instr[index])) { - switch(instr[index]) { - case PREFIX_CS_OVERRIDE: - case PREFIX_SS_OVERRIDE: - case PREFIX_DS_OVERRIDE: - case PREFIX_ES_OVERRIDE: - case PREFIX_FS_OVERRIDE: - case PREFIX_GS_OVERRIDE: - PrintDebug("Segment Override!!\n"); - return -1; - break; - default: - break; - } - index++; - } + } + + if (dec_instr.op_type == V3_OP_MOVCR2) { + PrintDebug("MOVCR2 (mode=%s)\n", v3_cpu_mode_to_str(info->cpu_mode)); - /* - while (is_prefix_byte(instr[index])) { - index++; - } - */ - - if ((instr[index] == cr_access_byte) && - (instr[index + 1] == smsw_byte) && - (MODRM_REG(instr[index + 2]) == smsw_reg_byte)) { - - // SMSW (store machine status word) - - addr_t first_operand; - addr_t second_operand; - struct cr0_real *cr0; - operand_type_t addr_type; - char cr0_val = 0; - - index += 2; - - cr0 = (struct cr0_real*)&(info->ctrl_regs.cr0); - - - addr_type = decode_operands16(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG16); - - if (addr_type == MEM_OPERAND) { - addr_t host_addr; - - if (guest_pa_to_host_va(info, first_operand + (info->segments.ds.base << 4), &host_addr) == -1) { - // gpf the guest - PrintDebug("Could not convert guest physical address to host virtual address\n"); - return -1; - } + if ((v3_get_vm_cpu_mode(info) == LONG) || + (v3_get_vm_cpu_mode(info) == LONG_32_COMPAT)) { + struct cr0_64 * dst_reg = (struct cr0_64 *)(dec_instr.dst_operand.operand); - first_operand = host_addr; + if (info->shdw_pg_mode == SHADOW_PAGING) { + struct cr0_64 * guest_cr0 = (struct cr0_64 *)&(info->shdw_pg_state.guest_cr0); + *dst_reg = *guest_cr0; + } else { + struct cr0_64 * shadow_cr0 = (struct cr0_64 *)&(info->ctrl_regs.cr0); + *dst_reg = *shadow_cr0; + } + + PrintDebug("returned CR0: %p\n", (void *)*(addr_t *)dst_reg); } else { - // Register operand - // Should be ok?? - } - - cr0_val = *(char*)cr0 & 0x0f; - - *(char *)first_operand &= 0xf0; - *(char *)first_operand |= cr0_val; - - PrintDebug("index = %d, rip = %x\n", index, (ulong_t)(info->rip)); - info->rip += index; - PrintDebug("new_rip = %x\n", (ulong_t)(info->rip)); - // success - - } else if ((instr[index] == cr_access_byte) && - (instr[index+1] == mov_from_cr_byte)) { - /* Mov from CR0 - * This can only take a 32 bit register argument in anything less than 64 bit mode. - */ - addr_t first_operand; - addr_t second_operand; - operand_type_t addr_type; - - struct cr0_32 * real_cr0 = (struct cr0_32 *)&(info->ctrl_regs.cr0); - - index += 2; - - addr_type = decode_operands16(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32); - - struct cr0_32 * virt_cr0 = (struct cr0_32 *)first_operand; - - if (addr_type != REG_OPERAND) { - // invalid opcode to guest - PrintDebug("Invalid operand type in mov from CR0\n"); - return -1; - } - - if (info->shdw_pg_mode == SHADOW_PAGING) { - *virt_cr0 = *(struct cr0_32 *)&(info->shdw_pg_state.guest_cr0); - } else { - *virt_cr0 = *real_cr0; + struct cr0_32 * dst_reg = (struct cr0_32 *)(dec_instr.dst_operand.operand); + + if (info->shdw_pg_mode == SHADOW_PAGING) { + struct cr0_32 * guest_cr0 = (struct cr0_32 *)&(info->shdw_pg_state.guest_cr0); + *dst_reg = *guest_cr0; + } else { + struct cr0_32 * shadow_cr0 = (struct cr0_32 *)&(info->ctrl_regs.cr0); + *dst_reg = *shadow_cr0; + } + + PrintDebug("returned CR0: %x\n", *(uint_t*)dst_reg); } - PrintDebug("Returning CR0: %x\n", *virt_cr0); - - info->rip += index; - - } else { - PrintDebug("Unknown read instr from CR0\n"); + } else if (dec_instr.op_type == V3_OP_SMSW) { + struct cr0_real * shadow_cr0 = (struct cr0_real *)&(info->ctrl_regs.cr0); + struct cr0_real * dst_reg = (struct cr0_real *)(dec_instr.dst_operand.operand); + char cr0_val = *(char*)shadow_cr0 & 0x0f; + + PrintDebug("SMSW\n"); + + // The lower 4 bits of the guest/shadow CR0 are mapped through + // We can treat nested and shadow paging the same here + *(char *)dst_reg &= 0xf0; + *(char *)dst_reg |= cr0_val; + + } else { + PrintError("Unhandled opcode in handle_cr0_read\n"); return -1; - } + } + + info->rip += dec_instr.instr_length; - } + return 0; +} - break; - case PROTECTED: - { - - int index = 0; - int ret; - PrintDebug("Protected %s Mode read from CR0 at guest %s linear rip 0x%x\n", - info->mem_mode == VIRTUAL_MEM ? "Paged" : "", - info->mem_mode == VIRTUAL_MEM ? "virtual" : "", - get_addr_linear(info, info->rip, &(info->segments.cs))); - // We need to read the instruction, which is at CS:IP, but that - // linear address is guest physical without PG and guest virtual with PG - if (info->cpu_mode == PHYSICAL_MEM) { - // The real rip address is actually a combination of the rip + CS base +// First Attempt = 256 lines +// current = 65 lines +int v3_handle_cr3_write(struct guest_info * info) { + int ret; + uchar_t instr[15]; + struct x86_instr dec_instr; + + if (info->mem_mode == PHYSICAL_MEM) { ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - } else { - // The real rip address is actually a combination of the rip + CS base + } else { ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - } - - - /* - PrintDebug("Instr (15 bytes) at %x:\n", instr); - PrintTraceMemDump((char*)instr, 15); - */ - - if (ret != 15) { - // I think we should inject a GPF into the guest - PrintDebug("Could not read Protected %s mode instruction (ret=%d)\n", - info->cpu_mode == VIRTUAL_MEM ? "Paged" : "", ret); + } + + if (v3_decode(info, (addr_t)instr, &dec_instr) == -1) { + PrintError("Could not decode instruction\n"); return -1; - } - - - while (is_prefix_byte(instr[index])) { - switch(instr[index]) { - case PREFIX_CS_OVERRIDE: - case PREFIX_SS_OVERRIDE: - case PREFIX_DS_OVERRIDE: - case PREFIX_ES_OVERRIDE: - case PREFIX_FS_OVERRIDE: - case PREFIX_GS_OVERRIDE: - PrintDebug("Segment Override!!\n"); - return -1; - break; - default: - break; - } - index++; - } - - - /* - while (is_prefix_byte(instr[index])) { - index++; - } - */ - - if ((instr[index] == cr_access_byte) && - (instr[index+1] == mov_from_cr_byte)) { + } + + if (dec_instr.op_type == V3_OP_MOV2CR) { + PrintDebug("MOV2CR3 (cpu_mode=%s)\n", v3_cpu_mode_to_str(info->cpu_mode)); - // MOV from CR0 to register - - addr_t first_operand; - addr_t second_operand; - operand_type_t addr_type; - struct cr0_32 * virt_cr0; - struct cr0_32 * real_cr0 = (struct cr0_32 *)&(info->ctrl_regs.cr0); - - index += 2; - - addr_type = decode_operands32(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32); - - if (addr_type != REG_OPERAND) { - PrintDebug("Invalid operand type in mov from CR0\n"); - return -1; - } - - virt_cr0 = (struct cr0_32 *)first_operand; - if (info->shdw_pg_mode == SHADOW_PAGING) { - *virt_cr0 = *(struct cr0_32 *)&(info->shdw_pg_state.guest_cr0); - - if (info->mem_mode == PHYSICAL_MEM) { - virt_cr0->pg = 0; // clear the pg bit because guest doesn't think it's on - } - - } else { - *virt_cr0 = *real_cr0; + PrintDebug("Old Shadow CR3=%p; Old Guest CR3=%p\n", + (void *)(addr_t)(info->ctrl_regs.cr3), + (void*)(addr_t)(info->shdw_pg_state.guest_cr3)); + + + // We update the guest CR3 + if (info->cpu_mode == LONG) { + struct cr3_64 * new_cr3 = (struct cr3_64 *)(dec_instr.src_operand.operand); + struct cr3_64 * guest_cr3 = (struct cr3_64 *)&(info->shdw_pg_state.guest_cr3); + *guest_cr3 = *new_cr3; + } else { + struct cr3_32 * new_cr3 = (struct cr3_32 *)(dec_instr.src_operand.operand); + struct cr3_32 * guest_cr3 = (struct cr3_32 *)&(info->shdw_pg_state.guest_cr3); + *guest_cr3 = *new_cr3; + } + + + // If Paging is enabled in the guest then we need to change the shadow page tables + if (info->mem_mode == VIRTUAL_MEM) { + if (v3_activate_shadow_pt(info) == -1) { + PrintError("Failed to activate 32 bit shadow page table\n"); + return -1; + } + } + + PrintDebug("New Shadow CR3=%p; New Guest CR3=%p\n", + (void *)(addr_t)(info->ctrl_regs.cr3), + (void*)(addr_t)(info->shdw_pg_state.guest_cr3)); + + } else if (info->shdw_pg_mode == NESTED_PAGING) { + + // This is just a passthrough operation which we probably don't need here + if (info->cpu_mode == LONG) { + struct cr3_64 * new_cr3 = (struct cr3_64 *)(dec_instr.src_operand.operand); + struct cr3_64 * guest_cr3 = (struct cr3_64 *)&(info->ctrl_regs.cr3); + *guest_cr3 = *new_cr3; + } else { + struct cr3_32 * new_cr3 = (struct cr3_32 *)(dec_instr.src_operand.operand); + struct cr3_32 * guest_cr3 = (struct cr3_32 *)&(info->ctrl_regs.cr3); + *guest_cr3 = *new_cr3; + } + } - - PrintDebug("real CR0: %x\n", *(uint_t*)real_cr0); - PrintDebug("returned CR0: %x\n", *(uint_t*)virt_cr0); - - info->rip += index; - - } else { - PrintDebug("Unknown read instruction from CR0\n"); + } else { + PrintError("Unhandled opcode in handle_cr3_write\n"); return -1; - } - } - break; - - case PROTECTED_PAE: - PrintDebug("Protected PAE Mode read to CR0 is UNIMPLEMENTED\n"); - return -1; - - case LONG: - PrintDebug("Protected Long Mode read to CR0 is UNIMPLEMENTED\n"); - return -1; - - - default: - { - PrintDebug("Unknown Mode read from CR0 (info->cpu_mode=0x%x)\n",info->cpu_mode); - return -1; } - break; - } - - - return 0; + + info->rip += dec_instr.instr_length; + + return 0; } - -int handle_cr3_write(struct guest_info * info) { - if (info->cpu_mode == REAL) { - // WHAT THE HELL DOES THIS EVEN MEAN????? - - int index = 0; +// first attempt = 156 lines +// current = 36 lines +int v3_handle_cr3_read(struct guest_info * info) { + uchar_t instr[15]; int ret; - char instr[15]; - - PrintDebug("Real Mode Write to CR3.\n"); - // We need to read the instruction, which is at CS:IP, but that - // linear address is guest physical without PG and guest virtual with PG - - ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - - if (ret != 15) { - PrintDebug("Could not read instruction (ret=%d)\n", ret); - return -1; + struct x86_instr dec_instr; + + if (info->mem_mode == PHYSICAL_MEM) { + ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); + } else { + ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); } - - while (is_prefix_byte(instr[index])) { - switch(instr[index]) { - case PREFIX_CS_OVERRIDE: - case PREFIX_SS_OVERRIDE: - case PREFIX_DS_OVERRIDE: - case PREFIX_ES_OVERRIDE: - case PREFIX_FS_OVERRIDE: - case PREFIX_GS_OVERRIDE: - PrintDebug("Segment Override!!\n"); + + if (v3_decode(info, (addr_t)instr, &dec_instr) == -1) { + PrintError("Could not decode instruction\n"); return -1; - break; - default: - break; - } - index++; } - - if ((instr[index] == cr_access_byte) && - (instr[index + 1] == mov_to_cr_byte)) { - - addr_t first_operand; - addr_t second_operand; - struct cr3_32 * new_cr3; - // struct cr3_32 * real_cr3; - operand_type_t addr_type; - - index += 2; - - addr_type = decode_operands16(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32); - - if (addr_type != REG_OPERAND) { - /* Mov to CR3 can only be a 32 bit register */ - return -1; - } - - new_cr3 = (struct cr3_32 *)first_operand; - - if (info->shdw_pg_mode == SHADOW_PAGING) { - int flushed=0; - addr_t shadow_pt; - struct cr3_32 * shadow_cr3 = (struct cr3_32 *)&(info->shdw_pg_state.shadow_cr3); - struct cr3_32 * guest_cr3 = (struct cr3_32 *)&(info->shdw_pg_state.guest_cr3); - - /* - - if (CR3_TO_PDE32(*(uint_t*)shadow_cr3) != 0) { - PrintDebug("Shadow Page Table\n"); - PrintDebugPageTables((pde32_t *)CR3_TO_PDE32(*(uint_t*)shadow_cr3)); - } - */ - - /* Delete the current Page Tables */ - if (!CR3_32_SAME_BASE(new_cr3,guest_cr3)) { - PrintDebug("New CR3 is different - flushing shadow page table\n"); - delete_page_tables_pde32((pde32_t *)CR3_TO_PDE32(*(uint_t*)shadow_cr3)); - flushed=1; - } else { - PrintDebug("New CR3 (0x%x) has same base as previous CR3 (0x%x) - reusing shadow page table\n", *((uint_t*)new_cr3), *((uint_t*)guest_cr3)); - } - - PrintDebug("Old Shadow CR3=%x; Old Guest CR3=%x\n", - *(uint_t*)shadow_cr3, *(uint_t*)guest_cr3); - - - *guest_cr3 = *new_cr3; - - - if (flushed) { - // Something like this - shadow_pt = create_new_shadow_pt32(info); - //shadow_pt = setup_shadow_pt32(info, CR3_TO_PDE32(*(addr_t *)new_cr3)); - } else { - shadow_pt = shadow_cr3->pdt_base_addr<<12; - } - - /* Copy Various flags */ - *shadow_cr3 = *new_cr3; - - /* - { - addr_t tmp_addr; - guest_pa_to_host_va(info, ((*(uint_t*)guest_cr3) & 0xfffff000), &tmp_addr); - PrintDebug("Guest PD\n"); - PrintPD32((pde32_t *)tmp_addr); - + if (dec_instr.op_type == V3_OP_MOVCR2) { + PrintDebug("MOVCR32 (mode=%s)\n", v3_cpu_mode_to_str(info->cpu_mode)); + + if (info->shdw_pg_mode == SHADOW_PAGING) { + + if ((v3_get_vm_cpu_mode(info) == LONG) || + (v3_get_vm_cpu_mode(info) == LONG_32_COMPAT)) { + struct cr3_64 * dst_reg = (struct cr3_64 *)(dec_instr.dst_operand.operand); + struct cr3_64 * guest_cr3 = (struct cr3_64 *)&(info->shdw_pg_state.guest_cr3); + *dst_reg = *guest_cr3; + } else { + struct cr3_32 * dst_reg = (struct cr3_32 *)(dec_instr.dst_operand.operand); + struct cr3_32 * guest_cr3 = (struct cr3_32 *)&(info->shdw_pg_state.guest_cr3); + *dst_reg = *guest_cr3; + } + + } else if (info->shdw_pg_mode == NESTED_PAGING) { + + // This is just a passthrough operation which we probably don't need here + if ((v3_get_vm_cpu_mode(info) == LONG) || + (v3_get_vm_cpu_mode(info) == LONG_32_COMPAT)) { + struct cr3_64 * dst_reg = (struct cr3_64 *)(dec_instr.dst_operand.operand); + struct cr3_64 * guest_cr3 = (struct cr3_64 *)&(info->ctrl_regs.cr3); + *dst_reg = *guest_cr3; + } else { + struct cr3_32 * dst_reg = (struct cr3_32 *)(dec_instr.dst_operand.operand); + struct cr3_32 * guest_cr3 = (struct cr3_32 *)&(info->ctrl_regs.cr3); + *dst_reg = *guest_cr3; + } } - */ - - shadow_cr3->pdt_base_addr = PD32_BASE_ADDR(shadow_pt); - - PrintDebug("New Shadow CR3=%x; New Guest CR3=%x\n", - *(uint_t*)shadow_cr3, *(uint_t*)guest_cr3); - - } - info->rip += index; - } else { - PrintDebug("Unknown Instruction\n"); - PrintTraceMemDump(instr,15); - return -1; + PrintError("Unhandled opcode in handle_cr3_read\n"); + return -1; } + + info->rip += dec_instr.instr_length; + + return 0; +} +// We don't need to virtualize CR4, all we need is to detect the activation of PAE +int v3_handle_cr4_read(struct guest_info * info) { + // PrintError("CR4 Read not handled\n"); + // Do nothing... + return 0; +} - } else if (info->cpu_mode == PROTECTED) { - int index = 0; +int v3_handle_cr4_write(struct guest_info * info) { + uchar_t instr[15]; int ret; - char instr[15]; - - PrintDebug("Protected %s mode write to CR3 at %s 0x%x\n", - info->cpu_mode==PROTECTED ? "" : "Paged", - info->cpu_mode==PROTECTED ? "guest physical" : "guest virtual", - get_addr_linear(info,info->rip,&(info->segments.cs))); - - // We need to read the instruction, which is at CS:IP, but that - // linear address is guest physical without PG and guest virtual with PG + int flush_tlb=0; + struct x86_instr dec_instr; + v3_cpu_mode_t cpu_mode = v3_get_vm_cpu_mode(info); + if (info->mem_mode == PHYSICAL_MEM) { - // The real rip address is actually a combination of the rip + CS base - //PrintDebug("Writing Guest CR3 Write (Physical Address)\n"); - ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); + ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); } else { - //PrintDebug("Writing Guest CR3 Write (Virtual Address)\n"); - // The real rip address is actually a combination of the rip + CS base - ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - } - - if (ret != 15) { - PrintDebug("Could not read instruction (ret=%d)\n", ret); - return -1; + ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); } - - while (is_prefix_byte(instr[index])) { - switch(instr[index]) { - case PREFIX_CS_OVERRIDE: - case PREFIX_SS_OVERRIDE: - case PREFIX_DS_OVERRIDE: - case PREFIX_ES_OVERRIDE: - case PREFIX_FS_OVERRIDE: - case PREFIX_GS_OVERRIDE: - PrintDebug("Segment Override!!\n"); + + if (v3_decode(info, (addr_t)instr, &dec_instr) == -1) { + PrintError("Could not decode instruction\n"); return -1; - break; - default: - break; - } - index++; } - /* - while (is_prefix_byte(instr[index])) { - index++; - } - */ - - if ((instr[index] == cr_access_byte) && - (instr[index + 1] == mov_to_cr_byte)) { - - addr_t first_operand; - addr_t second_operand; - struct cr3_32 * new_cr3; - // struct cr3_32 * real_cr3; - operand_type_t addr_type; - - index += 2; - - addr_type = decode_operands32(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32); - - if (addr_type != REG_OPERAND) { - /* Mov to CR3 can only be a 32 bit register */ + if (dec_instr.op_type != V3_OP_MOV2CR) { + PrintError("Invalid opcode in write to CR4\n"); return -1; - } - - new_cr3 = (struct cr3_32 *)first_operand; - - if (info->shdw_pg_mode == SHADOW_PAGING) { - int flushed=0; - addr_t shadow_pt; - struct cr3_32 * shadow_cr3 = (struct cr3_32 *)&(info->shdw_pg_state.shadow_cr3); - struct cr3_32 * guest_cr3 = (struct cr3_32 *)&(info->shdw_pg_state.guest_cr3); - - - /* - if (CR3_TO_PDE32(*(uint_t*)shadow_cr3) != 0) { - PrintDebug("Shadow Page Table\n"); - PrintDebugPageTables((pde32_t *)CR3_TO_PDE32(*(uint_t*)shadow_cr3)); - } - */ - - /* Delete the current Page Tables */ - if (!CR3_32_SAME_BASE(guest_cr3,new_cr3)) { - PrintDebug("New CR3 is different - flushing shadow page table\n"); - delete_page_tables_pde32((pde32_t *)CR3_TO_PDE32(*(uint_t*)shadow_cr3)); - flushed=1; - } else { - PrintDebug("New CR3 (0x%x) has same base as previous CR3 (0x%x) - reusing shadow page table\n",*((uint_t*)new_cr3), *((uint_t*)guest_cr3)); - } - - PrintDebug("Old Shadow CR3=%x; Old Guest CR3=%x\n", - *(uint_t*)shadow_cr3, *(uint_t*)guest_cr3); - - - *guest_cr3 = *new_cr3; - - if (flushed) { - // Something like this - shadow_pt = create_new_shadow_pt32(info); - //shadow_pt = setup_shadow_pt32(info, CR3_TO_PDE32(*(addr_t *)new_cr3)); - } else { - shadow_pt =shadow_cr3->pdt_base_addr << 12; + } + + // Check to see if we need to flush the tlb + + if (v3_get_vm_mem_mode(info) == VIRTUAL_MEM) { + struct cr4_32 * new_cr4 = (struct cr4_32 *)(dec_instr.src_operand.operand); + struct cr4_32 * cr4 = (struct cr4_32 *)&(info->ctrl_regs.cr4); + + // if pse, pge, or pae have changed while PG (in any mode) is on + // the side effect is a TLB flush, which means we need to + // toss the current shadow page tables too + // + // + // TODO - PAE FLAG needs to be special cased + if ((cr4->pse != new_cr4->pse) || + (cr4->pge != new_cr4->pge) || + (cr4->pae != new_cr4->pae)) { + PrintDebug("Handling PSE/PGE/PAE -> TLBFlush case, flag set\n"); + flush_tlb=1; + } + } + - - /* Copy Various flags */ - *shadow_cr3 = *new_cr3; - - /* - { - addr_t tmp_addr; - guest_pa_to_host_va(info, ((*(uint_t*)guest_cr3) & 0xfffff000), &tmp_addr); - PrintDebug("Guest PD\n"); - PrintPD32((pde32_t *)tmp_addr); - + if ((cpu_mode == PROTECTED) || (cpu_mode == PROTECTED_PAE)) { + struct cr4_32 * new_cr4 = (struct cr4_32 *)(dec_instr.src_operand.operand); + struct cr4_32 * cr4 = (struct cr4_32 *)&(info->ctrl_regs.cr4); + + PrintDebug("OperandVal = %x, length = %d\n", *(uint_t *)new_cr4, dec_instr.src_operand.size); + PrintDebug("Old CR4=%x\n", *(uint_t *)cr4); + + if ((info->shdw_pg_mode == SHADOW_PAGING)) { + if (v3_get_vm_mem_mode(info) == PHYSICAL_MEM) { + + if ((cr4->pae == 0) && (new_cr4->pae == 1)) { + PrintDebug("Creating PAE passthrough tables\n"); + + // create 32 bit PAE direct map page table + if (v3_reset_passthrough_pts(info) == -1) { + PrintError("Could not create 32 bit PAE passthrough pages tables\n"); + return -1; + } + + // reset cr3 to new page tables + info->ctrl_regs.cr3 = *(addr_t*)&(info->direct_map_pt); + + } else if ((cr4->pae == 1) && (new_cr4->pae == 0)) { + // Create passthrough standard 32bit pagetables + PrintError("Switching From PAE to Protected mode not supported\n"); + return -1; + } + } } - */ - - shadow_cr3->pdt_base_addr = PD32_BASE_ADDR(shadow_pt); - - PrintDebug("New Shadow CR3=%x; New Guest CR3=%x\n", - *(uint_t*)shadow_cr3, *(uint_t*)guest_cr3); - - - - if (info->mem_mode == VIRTUAL_MEM) { - // If we aren't in paged mode then we have to preserve the identity mapped CR3 - info->ctrl_regs.cr3 = *(addr_t*)shadow_cr3; + *cr4 = *new_cr4; + PrintDebug("New CR4=%x\n", *(uint_t *)cr4); + + } else if ((cpu_mode == LONG) || (cpu_mode == LONG_32_COMPAT)) { + struct cr4_64 * new_cr4 = (struct cr4_64 *)(dec_instr.src_operand.operand); + struct cr4_64 * cr4 = (struct cr4_64 *)&(info->ctrl_regs.cr4); + + PrintDebug("Old CR4=%p\n", (void *)*(addr_t *)cr4); + PrintDebug("New CR4=%p\n", (void *)*(addr_t *)new_cr4); + + if (new_cr4->pae == 0) { + // cannot turn off PAE in long mode GPF the guest + PrintError("Cannot disable PAE in long mode, should send GPF\n"); + return -1; } - } - - info->rip += index; - + + *cr4 = *new_cr4; + } else { - PrintDebug("Unknown Instruction\n"); - PrintTraceMemDump(instr,15); - return -1; + PrintError("CR4 write not supported in CPU_MODE: %s\n", v3_cpu_mode_to_str(cpu_mode)); + return -1; } - } else { - PrintDebug("Invalid operating Mode (0x%x)\n", info->cpu_mode); - return -1; - } - - return 0; -} - - - - -int handle_cr3_read(struct guest_info * info) { - - if (info->cpu_mode == REAL) { - char instr[15]; - int ret; - int index = 0; - addr_t linear_addr = 0; - - linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs)); - - //PrintDebug("RIP Linear: %x\n", linear_addr); - //PrintV3Segments(info); - ret = read_guest_pa_memory(info, linear_addr, 15, instr); - - if (ret != 15) { - PrintDebug("Could not read instruction (ret=%d)\n", ret); - return -1; + if (flush_tlb) { + PrintDebug("Handling PSE/PGE/PAE -> TLBFlush (doing flush now!)\n"); + if (v3_activate_shadow_pt(info) == -1) { + PrintError("Failed to activate shadow page tables when emulating TLB flush in handling cr4 write\n"); + return -1; + } } - while (is_prefix_byte(instr[index])) { - switch(instr[index]) { - case PREFIX_CS_OVERRIDE: - case PREFIX_SS_OVERRIDE: - case PREFIX_DS_OVERRIDE: - case PREFIX_ES_OVERRIDE: - case PREFIX_FS_OVERRIDE: - case PREFIX_GS_OVERRIDE: - PrintDebug("Segment Override!!\n"); - return -1; - break; - default: - break; - } - index++; - } - - - if ((instr[index] == cr_access_byte) && - (instr[index + 1] == mov_from_cr_byte)) { - addr_t first_operand; - addr_t second_operand; - struct cr3_32 * virt_cr3; - struct cr3_32 * real_cr3 = (struct cr3_32 *)&(info->ctrl_regs.cr3); - operand_type_t addr_type; - - index += 2; - - addr_type = decode_operands16(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32); - - if (addr_type != REG_OPERAND) { - /* Mov to CR3 can only be a 32 bit register */ - return -1; - } - - virt_cr3 = (struct cr3_32 *)first_operand; - - if (info->shdw_pg_mode == SHADOW_PAGING) { - *virt_cr3 = *(struct cr3_32 *)&(info->shdw_pg_state.guest_cr3); - } else { - *virt_cr3 = *real_cr3; - } - - info->rip += index; - } else { - PrintDebug("Unknown Instruction\n"); - PrintTraceMemDump(instr,15); - return -1; - } - - + + info->rip += dec_instr.instr_length; return 0; - } else if (info->cpu_mode == PROTECTED) { +} - int index = 0; - int ret; - char instr[15]; - - - // We need to read the instruction, which is at CS:IP, but that - // linear address is guest physical without PG and guest virtual with PG - if (info->cpu_mode == PHYSICAL_MEM) { - // The real rip address is actually a combination of the rip + CS base - ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - } else { - // The real rip address is actually a combination of the rip + CS base - ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - } - if (ret != 15) { - PrintDebug("Could not read instruction (ret=%d)\n", ret); - return -1; - } +int v3_handle_efer_read(uint_t msr, struct v3_msr * dst, void * priv_data) { + struct guest_info * info = (struct guest_info *)(priv_data); + PrintDebug("EFER Read HI=%x LO=%x\n", info->shdw_pg_state.guest_efer.hi, info->shdw_pg_state.guest_efer.lo); - while (is_prefix_byte(instr[index])) { - switch(instr[index]) { - case PREFIX_CS_OVERRIDE: - case PREFIX_SS_OVERRIDE: - case PREFIX_DS_OVERRIDE: - case PREFIX_ES_OVERRIDE: - case PREFIX_FS_OVERRIDE: - case PREFIX_GS_OVERRIDE: - PrintDebug("Segment Override!!\n"); - return -1; - break; - default: - break; - } - index++; - } - - /* - while (is_prefix_byte(instr[index])) { - index++; - } - */ - - if ((instr[index] == cr_access_byte) && - (instr[index + 1] == mov_from_cr_byte)) { - addr_t first_operand; - addr_t second_operand; - struct cr3_32 * virt_cr3; - struct cr3_32 * real_cr3 = (struct cr3_32 *)&(info->ctrl_regs.cr3); - operand_type_t addr_type; - - index += 2; - - addr_type = decode_operands32(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32); - - if (addr_type != REG_OPERAND) { - /* Mov to CR3 can only be a 32 bit register */ - return -1; - } + dst->value = info->shdw_pg_state.guest_efer.value; + + return 0; +} - virt_cr3 = (struct cr3_32 *)first_operand; - if (info->shdw_pg_mode == SHADOW_PAGING) { - *virt_cr3 = *(struct cr3_32 *)&(info->shdw_pg_state.guest_cr3); - } else { - *virt_cr3 = *real_cr3; - } - - info->rip += index; - } else { - PrintDebug("Unknown Instruction\n"); - PrintTraceMemDump(instr,15); - return -1; - } - } else { - PrintDebug("Invalid operating Mode (0x%x), control registers follow\n", info->cpu_mode); - PrintV3CtrlRegs(info); - return -1; - } - return 0; +// TODO: this is a disaster we need to clean this up... +int v3_handle_efer_write(uint_t msr, struct v3_msr src, void * priv_data) { + struct guest_info * info = (struct guest_info *)(priv_data); + //struct efer_64 * new_efer = (struct efer_64 *)&(src.value); + struct efer_64 * shadow_efer = (struct efer_64 *)&(info->ctrl_regs.efer); + struct v3_msr * guest_efer = &(info->shdw_pg_state.guest_efer); + + PrintDebug("EFER Write\n"); + PrintDebug("EFER Write Values: HI=%x LO=%x\n", src.hi, src.lo); + //PrintDebug("Old EFER=%p\n", (void *)*(addr_t*)(shadow_efer)); + + // We virtualize the guests efer to hide the SVME and LMA bits + guest_efer->value = src.value; + + + // Enable/Disable Syscall + shadow_efer->sce = src.value & 0x1; + + return 0; }