X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmx.c;h=ca875bb2ec806ff8aa2ab49e0952fa8d0dced4c0;hb=281d544b5352321631fb51c394e1ae0a9680b752;hp=a773191f82ab6554a338acd6216b9e54c58fd8a5;hpb=9e5edc5349bbd8d3e8e4f0dabe78e6543f32708a;p=palacios.git diff --git a/palacios/src/palacios/vmx.c b/palacios/src/palacios/vmx.c index a773191..ca875bb 100644 --- a/palacios/src/palacios/vmx.c +++ b/palacios/src/palacios/vmx.c @@ -7,13 +7,11 @@ * and the University of New Mexico. You can find out more at * http://www.v3vee.org * - * Copyright (c) 2008, Peter Dinda - * Copyright (c) 2008, Jack Lange - * Copyright (c) 2008, The V3VEE Project + * Copyright (c) 2011, Jack Lange + * Copyright (c) 2011, The V3VEE Project * All rights reserved. * - * Author: Peter Dinda - * Jack Lange + * Author: Jack Lange * * This is free software. You are permitted to use, * redistribute, and modify it as specified in the file "V3VEE_LICENSE". @@ -33,16 +31,29 @@ #include #include #include +#include +#include +#ifdef V3_CONFIG_CHECKPOINT +#include +#endif + +#include +#include +#include -#ifndef CONFIG_DEBUG_VMX +#ifndef V3_CONFIG_DEBUG_VMX #undef PrintDebug #define PrintDebug(fmt, args...) #endif -static addr_t host_vmcs_ptrs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0}; -static addr_t active_vmcs_ptrs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0}; +/* These fields contain the hardware feature sets supported by the local CPU */ +static struct vmx_hw_info hw_info; + +extern v3_cpu_arch_t v3_cpu_types[]; + +static addr_t host_vmcs_ptrs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0}; extern int v3_vmx_launch(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs); extern int v3_vmx_resume(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs); @@ -50,7 +61,7 @@ extern int v3_vmx_resume(struct v3_gprs * vm_regs, struct guest_info * info, str static int inline check_vmcs_write(vmcs_field_t field, addr_t val) { int ret = 0; - ret = vmcs_write(field,val); + ret = vmcs_write(field, val); if (ret != VMX_SUCCESS) { PrintError("VMWRITE error on %s!: %d\n", v3_vmcs_field_to_str(field), ret); @@ -72,54 +83,10 @@ static int inline check_vmcs_read(vmcs_field_t field, void * val) { return ret; } -#if 0 -// For the 32 bit reserved bit fields -// MB1s are in the low 32 bits, MBZs are in the high 32 bits of the MSR -static uint32_t sanitize_bits1(uint32_t msr_num, uint32_t val) { - v3_msr_t mask_msr; - - PrintDebug("sanitize_bits1 (MSR:%x)\n", msr_num); - - v3_get_msr(msr_num, &mask_msr.hi, &mask_msr.lo); - - PrintDebug("MSR %x = %x : %x \n", msr_num, mask_msr.hi, mask_msr.lo); - - val |= mask_msr.lo; - val |= mask_msr.hi; - - return val; -} - - - -static addr_t sanitize_bits2(uint32_t msr_num0, uint32_t msr_num1, addr_t val) { - v3_msr_t msr0, msr1; - addr_t msr0_val, msr1_val; - - PrintDebug("sanitize_bits2 (MSR0=%x, MSR1=%x)\n", msr_num0, msr_num1); - - v3_get_msr(msr_num0, &msr0.hi, &msr0.lo); - v3_get_msr(msr_num1, &msr1.hi, &msr1.lo); - - // This generates a mask that is the natural bit width of the CPU - msr0_val = msr0.value; - msr1_val = msr1.value; - - PrintDebug("MSR %x = %p, %x = %p \n", msr_num0, (void*)msr0_val, msr_num1, (void*)msr1_val); - - val |= msr0_val; - val |= msr1_val; - - return val; -} - -#endif - static addr_t allocate_vmcs() { - reg_ex_t msr; struct vmcs_data * vmcs_page = NULL; PrintDebug("Allocating page\n"); @@ -127,10 +94,8 @@ static addr_t allocate_vmcs() { vmcs_page = (struct vmcs_data *)V3_VAddr(V3_AllocPages(1)); memset(vmcs_page, 0, 4096); - v3_get_msr(VMX_BASIC_MSR, &(msr.e_reg.high), &(msr.e_reg.low)); - - vmcs_page->revision = ((struct vmx_basic_msr*)&msr)->revision; - PrintDebug("VMX Revision: 0x%x\n",vmcs_page->revision); + vmcs_page->revision = hw_info.basic_info.revision; + PrintDebug("VMX Revision: 0x%x\n", vmcs_page->revision); return (addr_t)V3_PAddr((void *)vmcs_page); } @@ -138,16 +103,14 @@ static addr_t allocate_vmcs() { -static int init_vmcs_bios(struct guest_info * info, struct vmx_data * vmx_state) { +static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) { int vmx_ret = 0; - struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data); // disable global interrupts for vm state initialization v3_disable_ints(); PrintDebug("Loading VMCS\n"); vmx_ret = vmcs_load(vmx_state->vmcs_ptr_phys); - active_vmcs_ptrs[V3_Get_CPU()] = vmx_info->vmcs_ptr_phys; vmx_state->state = VMX_UNLAUNCHED; if (vmx_ret != VMX_SUCCESS) { @@ -156,257 +119,311 @@ static int init_vmcs_bios(struct guest_info * info, struct vmx_data * vmx_state) } + /*** Setup default state from HW ***/ - /******* Setup Host State **********/ + vmx_state->pin_ctrls.value = hw_info.pin_ctrls.def_val; + vmx_state->pri_proc_ctrls.value = hw_info.proc_ctrls.def_val; + vmx_state->exit_ctrls.value = hw_info.exit_ctrls.def_val; + vmx_state->entry_ctrls.value = hw_info.entry_ctrls.def_val; + vmx_state->sec_proc_ctrls.value = hw_info.sec_proc_ctrls.def_val; - /* Cache GDTR, IDTR, and TR in host struct */ - addr_t gdtr_base; - struct { - uint16_t selector; - addr_t base; - } __attribute__((packed)) tmp_seg; - - - __asm__ __volatile__( - "sgdt (%0);" - : - : "q"(&tmp_seg) - : "memory" - ); - gdtr_base = tmp_seg.base; - vmx_state->host_state.gdtr.base = gdtr_base; - - __asm__ __volatile__( - "sidt (%0);" - : - : "q"(&tmp_seg) - : "memory" - ); - vmx_state->host_state.idtr.base = tmp_seg.base; - - __asm__ __volatile__( - "str (%0);" - : - : "q"(&tmp_seg) - : "memory" - ); - vmx_state->host_state.tr.selector = tmp_seg.selector; - - /* The GDTR *index* is bits 3-15 of the selector. */ - struct tss_descriptor * desc = NULL; - desc = (struct tss_descriptor *)(gdtr_base + (8 * (tmp_seg.selector >> 3))); - - tmp_seg.base = ((desc->base1) | - (desc->base2 << 16) | - (desc->base3 << 24) | -#ifdef __V3_64BIT__ - ((uint64_t)desc->base4 << 32) -#else - (0) -#endif - ); + /* Print Control MSRs */ + PrintDebug("CR0 MSR: %p\n", (void *)(addr_t)hw_info.cr0.value); + PrintDebug("CR4 MSR: %p\n", (void *)(addr_t)hw_info.cr4.value); - vmx_state->host_state.tr.base = tmp_seg.base; - - /********** Setup and VMX Control Fields from MSR ***********/ - /* Setup IO map */ + /******* Setup Host State **********/ + /* Cache GDTR, IDTR, and TR in host struct */ - struct v3_msr tmp_msr; - v3_get_msr(VMX_PINBASED_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo)); + /********** Setup VMX Control Fields ***********/ /* Add external interrupts, NMI exiting, and virtual NMI */ - vmx_state->pin_ctrls.value = tmp_msr.lo; vmx_state->pin_ctrls.nmi_exit = 1; vmx_state->pin_ctrls.ext_int_exit = 1; - v3_get_msr(VMX_PROCBASED_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo)); - vmx_state->pri_proc_ctrls.value = tmp_msr.lo; - vmx_state->pri_proc_ctrls.use_io_bitmap = 1; vmx_state->pri_proc_ctrls.hlt_exit = 1; - vmx_state->pri_proc_ctrls.invlpg_exit = 1; - vmx_state->pri_proc_ctrls.use_msr_bitmap = 1; - vmx_state->pri_proc_ctrls.pause_exit = 1; + + + vmx_state->pri_proc_ctrls.pause_exit = 0; vmx_state->pri_proc_ctrls.tsc_offset = 1; -#ifdef CONFIG_TIME_VIRTUALIZE_TSC +#ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC vmx_state->pri_proc_ctrls.rdtsc_exit = 1; #endif - vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_A_ADDR, (addr_t)V3_PAddr(info->vm_info->io_map.arch_data)); + /* Setup IO map */ + vmx_state->pri_proc_ctrls.use_io_bitmap = 1; + vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_A_ADDR, (addr_t)V3_PAddr(core->vm_info->io_map.arch_data)); vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_B_ADDR, - (addr_t)V3_PAddr(info->vm_info->io_map.arch_data) + PAGE_SIZE_4KB); - + (addr_t)V3_PAddr(core->vm_info->io_map.arch_data) + PAGE_SIZE_4KB); - vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(info->vm_info->msr_map.arch_data)); - v3_get_msr(VMX_EXIT_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo)); - vmx_state->exit_ctrls.value = tmp_msr.lo; - vmx_state->exit_ctrls.host_64_on = 1; - - if ((vmx_state->exit_ctrls.save_efer == 1) || (vmx_state->exit_ctrls.ld_efer == 1)) { - vmx_state->ia32e_avail = 1; - } + vmx_state->pri_proc_ctrls.use_msr_bitmap = 1; + vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data)); - v3_get_msr(VMX_ENTRY_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo)); - vmx_state->entry_ctrls.value = tmp_msr.lo; - { - struct vmx_exception_bitmap excp_bmap; - excp_bmap.value = 0; - - excp_bmap.pf = 1; - - vmx_ret |= check_vmcs_write(VMCS_EXCP_BITMAP, excp_bmap.value); - } - /******* Setup VMXAssist guest state ***********/ - info->rip = 0xd0000; - info->vm_regs.rsp = 0x80000; +#ifdef __V3_64BIT__ + // Ensure host runs in 64-bit mode at each VM EXIT + vmx_state->exit_ctrls.host_64_on = 1; +#endif - struct rflags * flags = (struct rflags *)&(info->ctrl_regs.rflags); - flags->rsvd1 = 1; + // Hook all accesses to EFER register + v3_hook_msr(core->vm_info, EFER_MSR, + &v3_handle_efer_read, + &v3_handle_efer_write, + core); - /* Print Control MSRs */ - v3_get_msr(VMX_CR0_FIXED0_MSR, &(tmp_msr.hi), &(tmp_msr.lo)); - PrintDebug("CR0 MSR: %p\n", (void *)(addr_t)tmp_msr.value); + // Restore host's EFER register on each VM EXIT + vmx_state->exit_ctrls.ld_efer = 1; - v3_get_msr(VMX_CR4_FIXED0_MSR, &(tmp_msr.hi), &(tmp_msr.lo)); - PrintDebug("CR4 MSR: %p\n", (void *)(addr_t)tmp_msr.value); + // Save/restore guest's EFER register to/from VMCS on VM EXIT/ENTRY + vmx_state->exit_ctrls.save_efer = 1; + vmx_state->entry_ctrls.ld_efer = 1; + // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written + vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE); -#define GUEST_CR0 0x80000031 -#define GUEST_CR4 0x00002000 - info->ctrl_regs.cr0 = GUEST_CR0; - info->ctrl_regs.cr4 = GUEST_CR4; - ((struct cr0_32 *)&(info->shdw_pg_state.guest_cr0))->pe = 1; - /* Setup paging */ - if (info->shdw_pg_mode == SHADOW_PAGING) { + if (core->shdw_pg_mode == SHADOW_PAGING) { PrintDebug("Creating initial shadow page table\n"); - if (v3_init_passthrough_pts(info) == -1) { + if (v3_init_passthrough_pts(core) == -1) { PrintError("Could not initialize passthrough page tables\n"); return -1; } #define CR0_PE 0x00000001 #define CR0_PG 0x80000000 +#define CR0_WP 0x00010000 // To ensure mem hooks work + vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP)); - - vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG) ); - vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE); - - info->ctrl_regs.cr3 = info->direct_map_pt; + core->ctrl_regs.cr3 = core->direct_map_pt; // vmx_state->pinbased_ctrls |= NMI_EXIT; /* Add CR exits */ vmx_state->pri_proc_ctrls.cr3_ld_exit = 1; vmx_state->pri_proc_ctrls.cr3_str_exit = 1; - } + + vmx_state->pri_proc_ctrls.invlpg_exit = 1; + + /* Add page fault exits */ + vmx_state->excp_bmap.pf = 1; + + // Setup VMX Assist + v3_vmxassist_init(core, vmx_state); + + } else if ((core->shdw_pg_mode == NESTED_PAGING) && + (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_CPU)) { + +#define CR0_PE 0x00000001 +#define CR0_PG 0x80000000 +#define CR0_WP 0x00010000 // To ensure mem hooks work + vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP)); + + // vmx_state->pinbased_ctrls |= NMI_EXIT; + + /* Disable CR exits */ + vmx_state->pri_proc_ctrls.cr3_ld_exit = 0; + vmx_state->pri_proc_ctrls.cr3_str_exit = 0; + + vmx_state->pri_proc_ctrls.invlpg_exit = 0; + + /* Add page fault exits */ + // vmx_state->excp_bmap.pf = 1; // This should never happen..., enabled to catch bugs + + // Setup VMX Assist + v3_vmxassist_init(core, vmx_state); + + /* Enable EPT */ + vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls + vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging - // Setup segment registers - { - struct v3_segment * seg_reg = (struct v3_segment *)&(info->segments); - int i; - for (i = 0; i < 10; i++) { - seg_reg[i].selector = 3 << 3; - seg_reg[i].limit = 0xffff; - seg_reg[i].base = 0x0; + if (v3_init_ept(core, &hw_info) == -1) { + PrintError("Error initializing EPT\n"); + return -1; } - info->segments.cs.selector = 2<<3; - - /* Set only the segment registers */ - for (i = 0; i < 6; i++) { - seg_reg[i].limit = 0xfffff; - seg_reg[i].granularity = 1; - seg_reg[i].type = 3; - seg_reg[i].system = 1; - seg_reg[i].dpl = 0; - seg_reg[i].present = 1; - seg_reg[i].db = 1; + } else if ((core->shdw_pg_mode == NESTED_PAGING) && + (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) { + int i = 0; + // For now we will assume that unrestricted guest mode is assured w/ EPT + + + core->vm_regs.rsp = 0x00; + core->rip = 0xfff0; + core->vm_regs.rdx = 0x00000f00; + core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1 + core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode + + + core->segments.cs.selector = 0xf000; + core->segments.cs.limit = 0xffff; + core->segments.cs.base = 0x0000000f0000LL; + + // (raw attributes = 0xf3) + core->segments.cs.type = 0xb; + core->segments.cs.system = 0x1; + core->segments.cs.dpl = 0x0; + core->segments.cs.present = 1; + + + + struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds), + &(core->segments.es), &(core->segments.fs), + &(core->segments.gs), NULL}; + + for ( i = 0; segregs[i] != NULL; i++) { + struct v3_segment * seg = segregs[i]; + + seg->selector = 0x0000; + // seg->base = seg->selector << 4; + seg->base = 0x00000000; + seg->limit = 0xffff; + + + seg->type = 0x3; + seg->system = 0x1; + seg->dpl = 0x0; + seg->present = 1; + // seg->granularity = 1; + } - info->segments.cs.type = 0xb; - info->segments.ldtr.selector = 0x20; - info->segments.ldtr.type = 2; - info->segments.ldtr.system = 0; - info->segments.ldtr.present = 1; - info->segments.ldtr.granularity = 0; + core->segments.gdtr.limit = 0x0000ffff; + core->segments.gdtr.base = 0x0000000000000000LL; - - /************* Map in GDT and vmxassist *************/ - - uint64_t gdt[] __attribute__ ((aligned(32))) = { - 0x0000000000000000ULL, /* 0x00: reserved */ - 0x0000830000000000ULL, /* 0x08: 32-bit TSS */ - //0x0000890000000000ULL, /* 0x08: 32-bit TSS */ - 0x00CF9b000000FFFFULL, /* 0x10: CS 32-bit */ - 0x00CF93000000FFFFULL, /* 0x18: DS 32-bit */ - 0x000082000000FFFFULL, /* 0x20: LDTR 32-bit */ - }; - -#define VMXASSIST_GDT 0x10000 - addr_t vmxassist_gdt = 0; - - if (v3_gpa_to_hva(info, VMXASSIST_GDT, &vmxassist_gdt) == -1) { - PrintError("Could not find VMXASSIST GDT destination\n"); + core->segments.idtr.limit = 0x0000ffff; + core->segments.idtr.base = 0x0000000000000000LL; + + core->segments.ldtr.selector = 0x0000; + core->segments.ldtr.limit = 0x0000ffff; + core->segments.ldtr.base = 0x0000000000000000LL; + core->segments.ldtr.type = 2; + core->segments.ldtr.present = 1; + + core->segments.tr.selector = 0x0000; + core->segments.tr.limit = 0x0000ffff; + core->segments.tr.base = 0x0000000000000000LL; + core->segments.tr.type = 0xb; + core->segments.tr.present = 1; + + // core->dbg_regs.dr6 = 0x00000000ffff0ff0LL; + core->dbg_regs.dr7 = 0x0000000000000400LL; + + /* Enable EPT */ + vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls + vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging + vmx_state->sec_proc_ctrls.unrstrct_guest = 1; // enable unrestricted guest operation + + + /* Disable shadow paging stuff */ + vmx_state->pri_proc_ctrls.cr3_ld_exit = 0; + vmx_state->pri_proc_ctrls.cr3_str_exit = 0; + + vmx_state->pri_proc_ctrls.invlpg_exit = 0; + + + if (v3_init_ept(core, &hw_info) == -1) { + PrintError("Error initializing EPT\n"); return -1; } - memcpy((void *)vmxassist_gdt, gdt, sizeof(uint64_t) * 5); - - info->segments.gdtr.base = VMXASSIST_GDT; - -#define VMXASSIST_TSS 0x40000 - uint64_t vmxassist_tss = VMXASSIST_TSS; - gdt[0x08 / sizeof(gdt[0])] |= - ((vmxassist_tss & 0xFF000000) << (56 - 24)) | - ((vmxassist_tss & 0x00FF0000) << (32 - 16)) | - ((vmxassist_tss & 0x0000FFFF) << (16)) | - (8392 - 1); - - info->segments.tr.selector = 0x08; - info->segments.tr.base = vmxassist_tss; - - //info->segments.tr.type = 0x9; - info->segments.tr.type = 0x3; - info->segments.tr.system = 0; - info->segments.tr.present = 1; - info->segments.tr.granularity = 0; + } else { + PrintError("Invalid Virtual paging mode\n"); + return -1; } - - // setup VMXASSIST - { -#define VMXASSIST_START 0x000d0000 - extern uint8_t v3_vmxassist_start[]; - extern uint8_t v3_vmxassist_end[]; - addr_t vmxassist_dst = 0; - - if (v3_gpa_to_hva(info, VMXASSIST_START, &vmxassist_dst) == -1) { - PrintError("Could not find VMXASSIST destination\n"); + + + // hook vmx msrs + + // Setup SYSCALL/SYSENTER MSRs in load/store area + + // save STAR, LSTAR, FMASK, KERNEL_GS_BASE MSRs in MSR load/store area + { +#define IA32_STAR 0xc0000081 +#define IA32_LSTAR 0xc0000082 +#define IA32_FMASK 0xc0000084 +#define IA32_KERN_GS_BASE 0xc0000102 + +#define IA32_CSTAR 0xc0000083 // Compatibility mode STAR (ignored for now... hopefully its not that important...) + + int msr_ret = 0; + + struct vmcs_msr_entry * exit_store_msrs = NULL; + struct vmcs_msr_entry * exit_load_msrs = NULL; + struct vmcs_msr_entry * entry_load_msrs = NULL;; + int max_msrs = (hw_info.misc_info.max_msr_cache_size + 1) * 4; + + V3_Print("Setting up MSR load/store areas (max_msr_count=%d)\n", max_msrs); + + if (max_msrs < 4) { + PrintError("Max MSR cache size is too small (%d)\n", max_msrs); return -1; } - memcpy((void *)vmxassist_dst, v3_vmxassist_start, v3_vmxassist_end - v3_vmxassist_start); + vmx_state->msr_area = V3_VAddr(V3_AllocPages(1)); + + if (vmx_state->msr_area == NULL) { + PrintError("could not allocate msr load/store area\n"); + return -1; + } + + msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_CNT, 4); + msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_CNT, 4); + msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_CNT, 4); + + + exit_store_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area); + exit_load_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area + (sizeof(struct vmcs_msr_entry) * 4)); + entry_load_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area + (sizeof(struct vmcs_msr_entry) * 8)); + + + exit_store_msrs[0].index = IA32_STAR; + exit_store_msrs[1].index = IA32_LSTAR; + exit_store_msrs[2].index = IA32_FMASK; + exit_store_msrs[3].index = IA32_KERN_GS_BASE; + + memcpy(exit_store_msrs, exit_load_msrs, sizeof(struct vmcs_msr_entry) * 4); + memcpy(exit_store_msrs, entry_load_msrs, sizeof(struct vmcs_msr_entry) * 4); + + + v3_get_msr(IA32_STAR, &(exit_load_msrs[0].hi), &(exit_load_msrs[0].lo)); + v3_get_msr(IA32_LSTAR, &(exit_load_msrs[1].hi), &(exit_load_msrs[1].lo)); + v3_get_msr(IA32_FMASK, &(exit_load_msrs[2].hi), &(exit_load_msrs[2].lo)); + v3_get_msr(IA32_KERN_GS_BASE, &(exit_load_msrs[3].hi), &(exit_load_msrs[3].lo)); + + msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_ADDR, (addr_t)V3_PAddr(exit_store_msrs)); + msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_ADDR, (addr_t)V3_PAddr(exit_load_msrs)); + msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_ADDR, (addr_t)V3_PAddr(entry_load_msrs)); + } - /*** Write all the info to the VMCS ***/ + /* Sanity check ctrl/reg fields against hw_defaults */ + -#define DEBUGCTL_MSR 0x1d9 - v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo)); - vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value); - info->dbg_regs.dr7 = 0x400; + + /*** Write all the info to the VMCS ***/ + + /* + { + // IS THIS NECESSARY??? +#define DEBUGCTL_MSR 0x1d9 + struct v3_msr tmp_msr; + v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo)); + vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value); + core->dbg_regs.dr7 = 0x400; + } + */ #ifdef __V3_64BIT__ vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL); @@ -415,32 +432,33 @@ static int init_vmcs_bios(struct guest_info * info, struct vmx_data * vmx_state) vmx_ret |= check_vmcs_write(VMCS_LINK_PTR_HIGH, (addr_t)0xffffffffUL); #endif - if (v3_update_vmcs_ctrl_fields(info)) { + + + + if (v3_update_vmcs_ctrl_fields(core)) { PrintError("Could not write control fields!\n"); return -1; } - if (v3_update_vmcs_host_state(info)) { + if (v3_update_vmcs_host_state(core)) { PrintError("Could not write host state\n"); return -1; } - - vmx_state->assist_state = VMXASSIST_DISABLED; - // reenable global interrupts for vm state initialization now // that the vm state is initialized. If another VM kicks us off, // it'll update our vmx state so that we know to reload ourself - v3_disable_ints(); + v3_enable_ints(); return 0; } -int v3_init_vmx_vmcs(struct guest_info * info, v3_vm_class_t vm_class) { +int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) { struct vmx_data * vmx_state = NULL; int vmx_ret = 0; vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data)); + memset(vmx_state, 0, sizeof(struct vmx_data)); PrintDebug("vmx_data pointer: %p\n", (void *)vmx_state); @@ -449,10 +467,10 @@ int v3_init_vmx_vmcs(struct guest_info * info, v3_vm_class_t vm_class) { PrintDebug("VMCS pointer: %p\n", (void *)(vmx_state->vmcs_ptr_phys)); - info->vmm_data = vmx_state; + core->vmm_data = vmx_state; vmx_state->state = VMX_UNLAUNCHED; - PrintDebug("Initializing VMCS (addr=%p)\n", info->vmm_data); + PrintDebug("Initializing VMCS (addr=%p)\n", core->vmm_data); // TODO: Fix vmcs fields so they're 32-bit @@ -466,12 +484,18 @@ int v3_init_vmx_vmcs(struct guest_info * info, v3_vm_class_t vm_class) { if (vm_class == V3_PC_VM) { PrintDebug("Initializing VMCS\n"); - init_vmcs_bios(info, vmx_state); + if (init_vmcs_bios(core, vmx_state) == -1) { + PrintError("Error initializing VMCS to BIOS state\n"); + return -1; + } } else { PrintError("Invalid VM Class\n"); return -1; } + PrintDebug("Serializing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys); + vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys); + return 0; } @@ -480,6 +504,7 @@ int v3_deinit_vmx_vmcs(struct guest_info * core) { struct vmx_data * vmx_state = core->vmm_data; V3_FreePages((void *)(vmx_state->vmcs_ptr_phys), 1); + V3_FreePages(V3_PAddr(vmx_state->msr_area), 1); V3_Free(vmx_state); @@ -487,14 +512,71 @@ int v3_deinit_vmx_vmcs(struct guest_info * core) { } + +#ifdef V3_CONFIG_CHECKPOINT +/* + * JRL: This is broken + */ +int v3_vmx_save_core(struct guest_info * core, void * ctx){ + uint64_t vmcs_ptr = vmcs_store(); + + v3_chkpt_save(ctx, "vmcs_data", PAGE_SIZE, (void *)vmcs_ptr); + + return 0; +} + +int v3_vmx_load_core(struct guest_info * core, void * ctx){ + struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data); + struct cr0_32 * shadow_cr0; + char vmcs[PAGE_SIZE_4KB]; + + v3_chkpt_load(ctx, "vmcs_data", PAGE_SIZE_4KB, vmcs); + + vmcs_clear(vmx_info->vmcs_ptr_phys); + vmcs_load((addr_t)vmcs); + + v3_vmx_save_vmcs(core); + + shadow_cr0 = (struct cr0_32 *)&(core->ctrl_regs.cr0); + + + /* Get the CPU mode to set the guest_ia32e entry ctrl */ + + if (core->shdw_pg_mode == SHADOW_PAGING) { + if (v3_get_vm_mem_mode(core) == VIRTUAL_MEM) { + if (v3_activate_shadow_pt(core) == -1) { + PrintError("Failed to activate shadow page tables\n"); + return -1; + } + } else { + if (v3_activate_passthrough_pt(core) == -1) { + PrintError("Failed to activate passthrough page tables\n"); + return -1; + } + } + } + + return 0; +} +#endif + + +void v3_flush_vmx_vm_core(struct guest_info * core) { + struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data); + vmcs_clear(vmx_info->vmcs_ptr_phys); + vmx_info->state = VMX_UNLAUNCHED; +} + + + static int update_irq_exit_state(struct guest_info * info) { struct vmx_exit_idt_vec_info idt_vec_info; check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value)); if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 0)) { -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("Calling v3_injecting_intr\n"); +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + V3_Print("Calling v3_injecting_intr\n"); #endif info->intr_core_state.irq_started = 0; v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ); @@ -525,15 +607,15 @@ static int update_irq_entry_state(struct guest_info * info) { check_vmcs_write(VMCS_ENTRY_EXCP_ERR, info->excp_state.excp_error_code); int_info.error_code = 1; -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("Injecting exception %d with error code %x\n", +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + V3_Print("Injecting exception %d with error code %x\n", int_info.vector, info->excp_state.excp_error_code); #endif } int_info.valid = 1; -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)(addr_t)info->rip); +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + V3_Print("Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)(addr_t)info->rip); #endif check_vmcs_write(VMCS_ENTRY_INT_INFO, int_info.value); @@ -544,8 +626,8 @@ static int update_irq_entry_state(struct guest_info * info) { if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 1)) { -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("IRQ pending from previous injection\n"); +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + V3_Print("IRQ pending from previous injection\n"); #endif // Copy the IDT vectoring info over to reinject the old interrupt @@ -571,8 +653,8 @@ static int update_irq_entry_state(struct guest_info * info) { ent_int.error_code = 0; ent_int.valid = 1; -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("Injecting Interrupt %d at exit %u(EIP=%p)\n", +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + V3_Print("Injecting Interrupt %d at exit %u(EIP=%p)\n", info->intr_core_state.irq_vector, (uint32_t)info->num_exits, (void *)(addr_t)info->rip); @@ -616,8 +698,8 @@ static int update_irq_entry_state(struct guest_info * info) { check_vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len); -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("Enabling Interrupt-Window exiting: %d\n", instr_len); +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + V3_Print("Enabling Interrupt-Window exiting: %d\n", instr_len); #endif vmx_info->pri_proc_ctrls.int_wndw_exit = 1; @@ -678,16 +760,25 @@ int v3_vmx_enter(struct guest_info * info) { // Perform any additional yielding needed for time adjustment v3_adjust_time(info); - // Update timer devices prior to entering VM. - v3_update_timers(info); - // disable global interrupts for vm state transition v3_disable_ints(); + // Update timer devices late after being in the VM so that as much + // of hte time in the VM is accounted for as possible. Also do it before + // updating IRQ entry state so that any interrupts the timers raise get + // handled on the next VM entry. Must be done with interrupts disabled. + v3_update_timers(info); + + if (vmcs_store() != vmx_info->vmcs_ptr_phys) { + vmcs_clear(vmx_info->vmcs_ptr_phys); + vmcs_load(vmx_info->vmcs_ptr_phys); + vmx_info->state = VMX_UNLAUNCHED; + } + v3_vmx_restore_vmcs(info); -#ifdef CONFIG_SYMCALL +#ifdef V3_CONFIG_SYMCALL if (info->sym_core_state.symcall_state.sym_call_active == 0) { update_irq_entry_state(info); } @@ -709,13 +800,16 @@ int v3_vmx_enter(struct guest_info * info) { check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high); check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low); - if (active_vmcs_ptrs[V3_Get_CPU()] != vmx_info->vmcs_ptr_phys) { - vmcs_load(vmx_info->vmcs_ptr_phys); - active_vmcs_ptrs[V3_Get_CPU()] = vmx_info->vmcs_ptr_phys; + if (v3_update_vmcs_host_state(info)) { + v3_enable_ints(); + PrintError("Could not write host state\n"); + return -1; } + if (vmx_info->state == VMX_UNLAUNCHED) { vmx_info->state = VMX_LAUNCHED; + info->vm_info->run_state = VM_RUNNING; ret = v3_vmx_launch(&(info->vm_regs), info, &(info->ctrl_regs)); } else { @@ -723,17 +817,22 @@ int v3_vmx_enter(struct guest_info * info) { ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs)); } + + // PrintDebug("VMX Exit: ret=%d\n", ret); if (ret != VMX_SUCCESS) { uint32_t error = 0; - vmcs_read(VMCS_INSTR_ERR, &error); - PrintError("VMENTRY Error: %d\n", error); + v3_enable_ints(); + + PrintError("VMENTRY Error: %d (launch_ret = %d)\n", error, ret); return -1; } + + // Immediate exit from VM time bookkeeping v3_time_exit_vm(info); @@ -756,12 +855,15 @@ int v3_vmx_enter(struct guest_info * info) { check_vmcs_read(VMCS_EXIT_INT_ERR, &(exit_info.int_err)); check_vmcs_read(VMCS_GUEST_LINEAR_ADDR, &(exit_info.guest_linear_addr)); + if (info->shdw_pg_mode == NESTED_PAGING) { + check_vmcs_read(VMCS_GUEST_PHYS_ADDR, &(exit_info.ept_fault_addr)); + } + //PrintDebug("VMX Exit taken, id-qual: %u-%lu\n", exit_info.exit_reason, exit_info.exit_qual); exit_log[info->num_exits % 10] = exit_info; - -#ifdef CONFIG_SYMCALL +#ifdef V3_CONFIG_SYMCALL if (info->sym_core_state.symcall_state.sym_call_active == 0) { update_irq_exit_state(info); } @@ -769,10 +871,15 @@ int v3_vmx_enter(struct guest_info * info) { update_irq_exit_state(info); #endif - // Handle any exits needed still in the atomic section - if (v3_handle_vmx_exit(info, &exit_info) == -1) { - PrintError("Error in atomic VMX exit handler\n"); - return -1; + if (exit_info.exit_reason == VMEXIT_INTR_WINDOW) { + // This is a special case whose only job is to inject an interrupt + vmcs_read(VMCS_PROC_CTRLS, &(vmx_info->pri_proc_ctrls.value)); + vmx_info->pri_proc_ctrls.int_wndw_exit = 0; + vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value); + +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + V3_Print("Interrupts available again! (RIP=%llx)\n", info->rip); +#endif } // reenable global interrupts after vm exit @@ -782,7 +889,7 @@ int v3_vmx_enter(struct guest_info * info) { v3_yield_cond(info); if (v3_handle_vmx_exit(info, &exit_info) == -1) { - PrintError("Error in VMX exit handler\n"); + PrintError("Error in VMX exit handler (Exit reason=%x)\n", exit_info.exit_reason); return -1; } @@ -792,30 +899,33 @@ int v3_vmx_enter(struct guest_info * info) { int v3_start_vmx_guest(struct guest_info * info) { - PrintDebug("Starting VMX core %u\n", info->cpu_id); + PrintDebug("Starting VMX core %u\n", info->vcpu_id); - if (info->cpu_id == 0) { + if (info->vcpu_id == 0) { info->core_run_state = CORE_RUNNING; info->vm_info->run_state = VM_RUNNING; } else { - PrintDebug("VMX core %u: Waiting for core initialization\n", info->cpu_id); + PrintDebug("VMX core %u: Waiting for core initialization\n", info->vcpu_id); while (info->core_run_state == CORE_STOPPED) { v3_yield(info); - //PrintDebug("VMX core %u: still waiting for INIT\n",info->cpu_id); + //PrintDebug("VMX core %u: still waiting for INIT\n",info->vcpu_id); } - PrintDebug("VMX core %u initialized\n", info->cpu_id); + PrintDebug("VMX core %u initialized\n", info->vcpu_id); + + // We'll be paranoid about race conditions here + v3_wait_at_barrier(info); } PrintDebug("VMX core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x), RIP=0x%p\n", - info->cpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base), + info->vcpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base), info->segments.cs.limit, (void *)(info->rip)); - PrintDebug("VMX core %u: Launching VMX VM\n", info->cpu_id); + PrintDebug("VMX core %u: Launching VMX VM on logical core %u\n", info->vcpu_id, info->pcpu_id); v3_start_time(info); @@ -827,11 +937,40 @@ int v3_start_vmx_guest(struct guest_info * info) { } if (v3_vmx_enter(info) == -1) { + + addr_t host_addr; + addr_t linear_addr = 0; + + info->vm_info->run_state = VM_ERROR; + + V3_Print("VMX core %u: VMX ERROR!!\n", info->vcpu_id); + + v3_print_guest_state(info); + + V3_Print("VMX core %u\n", info->vcpu_id); + + linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs)); + + if (info->mem_mode == PHYSICAL_MEM) { + v3_gpa_to_hva(info, linear_addr, &host_addr); + } else if (info->mem_mode == VIRTUAL_MEM) { + v3_gva_to_hva(info, linear_addr, &host_addr); + } + + V3_Print("VMX core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr); + + V3_Print("VMX core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr); + v3_dump_mem((uint8_t *)host_addr, 15); + + v3_print_stack(info); + + v3_print_vmcs(); print_exit_log(info); return -1; } + v3_wait_at_barrier(info); if (info->vm_info->run_state == VM_STOPPED) { @@ -850,6 +989,12 @@ int v3_start_vmx_guest(struct guest_info * info) { } + + +#define VMX_FEATURE_CONTROL_MSR 0x0000003a +#define CPUID_VMX_FEATURES 0x00000005 /* LOCK and VMXON */ +#define CPUID_1_ECX_VTXFLAG 0x00000020 + int v3_is_vmx_capable() { v3_msr_t feature_msr; uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0; @@ -863,7 +1008,7 @@ int v3_is_vmx_capable() { PrintDebug("MSRREGlow: 0x%.8x\n", feature_msr.lo); - if ((feature_msr.lo & FEATURE_CONTROL_VALID) != FEATURE_CONTROL_VALID) { + if ((feature_msr.lo & CPUID_VMX_FEATURES) != CPUID_VMX_FEATURES) { PrintDebug("VMX is locked -- enable in the BIOS\n"); return 0; } @@ -876,108 +1021,85 @@ int v3_is_vmx_capable() { return 1; } -static int has_vmx_nested_paging() { + +int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) { + // init vmcs bios + + if ((core->shdw_pg_mode == NESTED_PAGING) && + (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) { + // easy + core->rip = 0; + core->segments.cs.selector = rip << 8; + core->segments.cs.limit = 0xffff; + core->segments.cs.base = rip << 12; + } else { + core->vm_regs.rdx = core->vcpu_id; + core->vm_regs.rbx = rip; + } + return 0; } void v3_init_vmx_cpu(int cpu_id) { - extern v3_cpu_arch_t v3_cpu_types[]; - struct v3_msr tmp_msr; - uint64_t ret = 0; + addr_t vmx_on_region = 0; - v3_get_msr(VMX_CR4_FIXED0_MSR, &(tmp_msr.hi), &(tmp_msr.lo)); - -#ifdef __V3_64BIT__ - __asm__ __volatile__ ( - "movq %%cr4, %%rbx;" - "orq $0x00002000, %%rbx;" - "movq %%rbx, %0;" - : "=m"(ret) - : - : "%rbx" - ); - - if ((~ret & tmp_msr.value) == 0) { - __asm__ __volatile__ ( - "movq %0, %%cr4;" - : - : "q"(ret) - ); - } else { - PrintError("Invalid CR4 Settings!\n"); - return; - } - - __asm__ __volatile__ ( - "movq %%cr0, %%rbx; " - "orq $0x00000020,%%rbx; " - "movq %%rbx, %%cr0;" - : - : - : "%rbx" - ); -#elif __V3_32BIT__ - __asm__ __volatile__ ( - "movl %%cr4, %%ecx;" - "orl $0x00002000, %%ecx;" - "movl %%ecx, %0;" - : "=m"(ret) - : - : "%ecx" - ); - - if ((~ret & tmp_msr.value) == 0) { - __asm__ __volatile__ ( - "movl %0, %%cr4;" - : - : "q"(ret) - ); - } else { - PrintError("Invalid CR4 Settings!\n"); - return; + if (cpu_id == 0) { + if (v3_init_vmx_hw(&hw_info) == -1) { + PrintError("Could not initialize VMX hardware features on cpu %d\n", cpu_id); + return; + } } - __asm__ __volatile__ ( - "movl %%cr0, %%ecx; " - "orl $0x00000020,%%ecx; " - "movl %%ecx, %%cr0;" - : - : - : "%ecx" - ); - -#endif - - // - // Should check and return Error here.... + enable_vmx(); // Setup VMXON Region - host_vmcs_ptrs[cpu_id] = allocate_vmcs(); + vmx_on_region = allocate_vmcs(); - PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]); - if (v3_enable_vmx(host_vmcs_ptrs[cpu_id]) == VMX_SUCCESS) { - PrintDebug("VMX Enabled\n"); + if (vmx_on(vmx_on_region) == VMX_SUCCESS) { + V3_Print("VMX Enabled\n"); + host_vmcs_ptrs[cpu_id] = vmx_on_region; } else { - PrintError("VMX initialization failure\n"); - return; + V3_Print("VMX already enabled\n"); + V3_FreePages((void *)vmx_on_region, 1); } - - if (has_vmx_nested_paging() == 1) { - v3_cpu_types[cpu_id] = V3_VMX_EPT_CPU; - } else { - v3_cpu_types[cpu_id] = V3_VMX_CPU; - } + PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]); + { + struct vmx_sec_proc_ctrls sec_proc_ctrls; + sec_proc_ctrls.value = v3_vmx_get_ctrl_features(&(hw_info.sec_proc_ctrls)); + + if (sec_proc_ctrls.enable_ept == 0) { + V3_Print("VMX EPT (Nested) Paging not supported\n"); + v3_cpu_types[cpu_id] = V3_VMX_CPU; + } else if (sec_proc_ctrls.unrstrct_guest == 0) { + V3_Print("VMX EPT (Nested) Paging supported\n"); + v3_cpu_types[cpu_id] = V3_VMX_EPT_CPU; + } else { + V3_Print("VMX EPT (Nested) Paging + Unrestricted guest supported\n"); + v3_cpu_types[cpu_id] = V3_VMX_EPT_UG_CPU; + } + } } void v3_deinit_vmx_cpu(int cpu_id) { extern v3_cpu_arch_t v3_cpu_types[]; v3_cpu_types[cpu_id] = V3_INVALID_CPU; - V3_FreePages((void *)host_vmcs_ptrs[cpu_id], 1); + + if (host_vmcs_ptrs[cpu_id] != 0) { + V3_Print("Disabling VMX\n"); + + if (vmx_off() != VMX_SUCCESS) { + PrintError("Error executing VMXOFF\n"); + } + + V3_FreePages((void *)host_vmcs_ptrs[cpu_id], 1); + + host_vmcs_ptrs[cpu_id] = 0; + } }