X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmx.c;h=f26a1eae40cd2154b5878e9b6ef48bb6239cb701;hb=5d67a000f2d9e0043237af7c86eecd034119fe93;hp=a525e8735f28c2d5cfe87f8a317c1c06dc0dc2c0;hpb=fcc9962fe5d3b877f8f25de8745d0d4d1eaf394e;p=palacios.releases.git diff --git a/palacios/src/palacios/vmx.c b/palacios/src/palacios/vmx.c index a525e87..f26a1ea 100644 --- a/palacios/src/palacios/vmx.c +++ b/palacios/src/palacios/vmx.c @@ -36,7 +36,7 @@ #include #include -#ifndef CONFIG_DEBUG_VMX +#ifndef V3_CONFIG_DEBUG_VMX #undef PrintDebug #define PrintDebug(fmt, args...) #endif @@ -47,8 +47,7 @@ static struct vmx_hw_info hw_info; extern v3_cpu_arch_t v3_cpu_types[]; -static addr_t active_vmcs_ptrs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0}; -static addr_t host_vmcs_ptrs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0}; +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); @@ -106,7 +105,6 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) PrintDebug("Loading VMCS\n"); vmx_ret = vmcs_load(vmx_state->vmcs_ptr_phys); - active_vmcs_ptrs[V3_Get_CPU()] = vmx_state->vmcs_ptr_phys; vmx_state->state = VMX_UNLAUNCHED; if (vmx_ret != VMX_SUCCESS) { @@ -189,11 +187,11 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) vmx_state->pri_proc_ctrls.hlt_exit = 1; - vmx_state->pri_proc_ctrls.invlpg_exit = 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 @@ -209,24 +207,30 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) + + + #ifdef __V3_64BIT__ vmx_state->exit_ctrls.host_64_on = 1; #endif - /* Not sure how exactly to handle this... */ v3_hook_msr(core->vm_info, EFER_MSR, &v3_handle_efer_read, &v3_handle_efer_write, core); + // Or is it this??? + vmx_state->entry_ctrls.ld_efer = 1; + vmx_state->exit_ctrls.ld_efer = 1; + vmx_state->exit_ctrls.save_efer = 1; + /* *** */ vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE); - /* Setup paging */ if (core->shdw_pg_mode == SHADOW_PAGING) { PrintDebug("Creating initial shadow page table\n"); @@ -249,6 +253,8 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) 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; @@ -256,9 +262,7 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) v3_vmxassist_init(core, vmx_state); } else if ((core->shdw_pg_mode == NESTED_PAGING) && - (v3_cpu_types[core->cpu_id] == V3_VMX_EPT_CPU)) { - - // initialize 1to1 pts + (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_CPU)) { #define CR0_PE 0x00000001 #define CR0_PG 0x80000000 @@ -267,12 +271,14 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) // 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; + /* 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 + // vmx_state->excp_bmap.pf = 1; // This should never happen..., enabled to catch bugs // Setup VMX Assist v3_vmxassist_init(core, vmx_state); @@ -280,23 +286,20 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * 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 - // vmx_state->sec_proc_ctrls.unrstrct_guest = 1; // enable unrestricted guest operation - vmx_state->entry_ctrls.ld_efer = 1; - vmx_state->exit_ctrls.ld_efer = 1; - vmx_state->exit_ctrls.save_efer = 1; + if (v3_init_ept(core, &hw_info) == -1) { PrintError("Error initializing EPT\n"); return -1; } - } else if ((core->shdw_pg_mode == NESTED_PAGING) && - (v3_cpu_types[core->cpu_id] == V3_VMX_EPT_UG_CPU)) { + (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; @@ -364,9 +367,12 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging vmx_state->sec_proc_ctrls.unrstrct_guest = 1; // enable unrestricted guest operation - vmx_state->entry_ctrls.ld_efer = 1; - vmx_state->exit_ctrls.ld_efer = 1; - vmx_state->exit_ctrls.save_efer = 1; + + /* 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) { @@ -380,10 +386,69 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) } - // Hook the VMX msrs + // 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; + } + + 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)); + } /* Sanity check ctrl/reg fields against hw_defaults */ @@ -462,12 +527,18 @@ int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) { if (vm_class == V3_PC_VM) { PrintDebug("Initializing VMCS\n"); - init_vmcs_bios(core, 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; } @@ -476,6 +547,7 @@ 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(vmx_state->msr_area, 1); V3_Free(vmx_state); @@ -489,8 +561,8 @@ static int update_irq_exit_state(struct guest_info * 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); @@ -521,15 +593,15 @@ 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); @@ -540,8 +612,8 @@ 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 @@ -567,8 +639,8 @@ 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); @@ -612,8 +684,8 @@ 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; @@ -674,23 +746,22 @@ 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 prior to entering VM. Doing it here + // makes sure the guest sees any timers that fired while it + // was in the VMM + v3_update_timers(info); - if (active_vmcs_ptrs[V3_Get_CPU()] != vmx_info->vmcs_ptr_phys) { + if (vmcs_store() != vmx_info->vmcs_ptr_phys) { vmcs_load(vmx_info->vmcs_ptr_phys); - active_vmcs_ptrs[V3_Get_CPU()] = vmx_info->vmcs_ptr_phys; } - 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); } @@ -712,6 +783,12 @@ 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 (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; @@ -728,8 +805,10 @@ int v3_vmx_enter(struct guest_info * info) { uint32_t error = 0; vmcs_read(VMCS_INSTR_ERR, &error); - PrintError("VMENTRY Error: %d\n", error); + v3_enable_ints(); + + PrintError("VMENTRY Error: %d\n", error); return -1; } @@ -764,7 +843,7 @@ int v3_vmx_enter(struct guest_info * info) { 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); } @@ -778,8 +857,8 @@ int v3_vmx_enter(struct guest_info * info) { vmx_info->pri_proc_ctrls.int_wndw_exit = 0; vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value); -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("Interrupts available again! (RIP=%llx)\n", info->rip); +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + V3_Print("Interrupts available again! (RIP=%llx)\n", info->rip); #endif } @@ -800,30 +879,30 @@ 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); } 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); @@ -891,7 +970,23 @@ int v3_is_vmx_capable() { } +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; +} @@ -913,7 +1008,7 @@ void v3_init_vmx_cpu(int cpu_id) { PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]); if (vmx_on(host_vmcs_ptrs[cpu_id]) == VMX_SUCCESS) { - PrintDebug("VMX Enabled\n"); + V3_Print("VMX Enabled\n"); } else { PrintError("VMX initialization failure\n"); return;