X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fsvm.c;h=05fd183b5e55ffb8ff185be054fad77b6661e7d1;hb=60ad6a41c6d0ee08ed689e8505eb0c3df0c2a289;hp=f59bdc7ead2844fd3628452dd50cdb865fa1909b;hpb=29b7ff8811c82ab543f5c969e44ff217637a09c7;p=palacios.git diff --git a/palacios/src/palacios/svm.c b/palacios/src/palacios/svm.c index f59bdc7..05fd183 100644 --- a/palacios/src/palacios/svm.c +++ b/palacios/src/palacios/svm.c @@ -18,6 +18,7 @@ */ + #include #include @@ -35,6 +36,15 @@ #include #include +#include +#include + +#include + + +#ifdef V3_CONFIG_CHECKPOINT +#include +#endif #include @@ -43,17 +53,25 @@ #include +#ifdef V3_CONFIG_MEM_TRACK +#include +#endif -#ifndef CONFIG_DEBUG_SVM +#ifdef V3_CONFIG_TM_FUNC +#include +#endif + +#ifndef V3_CONFIG_DEBUG_SVM #undef PrintDebug #define PrintDebug(fmt, args...) #endif + uint32_t v3_last_exit; // This is a global pointer to the host's VMCB -static addr_t host_vmcbs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0}; +static addr_t host_vmcbs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0}; @@ -65,10 +83,10 @@ extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_ static vmcb_t * Allocate_VMCB() { vmcb_t * vmcb_page = NULL; - addr_t vmcb_pa = (addr_t)V3_AllocPages(1); + addr_t vmcb_pa = (addr_t)V3_AllocPages(1); // need not be shadow safe, not exposed to guest if ((void *)vmcb_pa == NULL) { - PrintError("Error allocating VMCB\n"); + PrintError(VM_NONE, VCORE_NONE, "Error allocating VMCB\n"); return NULL; } @@ -80,6 +98,25 @@ static vmcb_t * Allocate_VMCB() { } +static int v3_svm_handle_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data) +{ + int status; + + // Call arch-independent handler + if ((status = v3_handle_efer_write(core, msr, src, priv_data)) != 0) { + return status; + } + + // SVM-specific code + { + // Ensure that hardware visible EFER.SVME bit is set (SVM Enable) + struct efer_64 * hw_efer = (struct efer_64 *)&(core->ctrl_regs.efer); + hw_efer->svme = 1; + } + + return 0; +} + static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) { vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb); @@ -105,10 +142,9 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) { ctrl_area->instrs.HLT = 1; -#ifdef CONFIG_TIME_VIRTUALIZE_TSC - ctrl_area->instrs.RDTSC = 1; - ctrl_area->svm_instrs.RDTSCP = 1; -#endif + /* Set at VMM launch as needed */ + ctrl_area->instrs.RDTSC = 0; + ctrl_area->svm_instrs.RDTSCP = 0; // guest_state->cr0 = 0x00000001; // PE @@ -127,12 +163,16 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) { ctrl_area->exceptions.nmi = 1; */ + +#ifdef V3_CONFIG_TM_FUNC + v3_tm_set_excp_intercepts(ctrl_area); +#endif ctrl_area->instrs.NMI = 1; ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest ctrl_area->instrs.INIT = 1; - ctrl_area->instrs.PAUSE = 1; + // ctrl_area->instrs.PAUSE = 1; ctrl_area->instrs.shutdown_evts = 1; @@ -213,13 +253,30 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) { ctrl_area->instrs.MSR_PROT = 1; - PrintDebug("Exiting on interrupts\n"); + PrintDebug(core->vm_info, core, "Exiting on interrupts\n"); ctrl_area->guest_ctrl.V_INTR_MASKING = 1; ctrl_area->instrs.INTR = 1; + // The above also assures the TPR changes (CR8) are only virtual + + + // However, we need to see TPR writes since they will + // affect the virtual apic + // we reflect out cr8 to ctrl_regs->apic_tpr + ctrl_area->cr_reads.cr8 = 1; + ctrl_area->cr_writes.cr8 = 1; + // We will do all TPR comparisons in the virtual apic + // We also do not want the V_TPR to be able to mask the PIC + ctrl_area->guest_ctrl.V_IGN_TPR = 1; + + + v3_hook_msr(core->vm_info, EFER_MSR, + &v3_handle_efer_read, + &v3_svm_handle_efer_write, + core); if (core->shdw_pg_mode == SHADOW_PAGING) { - PrintDebug("Creating initial shadow page table\n"); + PrintDebug(core->vm_info, core, "Creating initial shadow page table\n"); /* JRL: This is a performance killer, and a simplistic solution */ /* We need to fix this */ @@ -228,28 +285,26 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) { if (v3_init_passthrough_pts(core) == -1) { - PrintError("Could not initialize passthrough page tables\n"); + PrintError(core->vm_info, core, "Could not initialize passthrough page tables\n"); return ; } core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL; - PrintDebug("Created\n"); + PrintDebug(core->vm_info, core, "Created\n"); core->ctrl_regs.cr0 |= 0x80000000; - core->ctrl_regs.cr3 = core->direct_map_pt; + + v3_activate_passthrough_pt(core); ctrl_area->cr_reads.cr0 = 1; ctrl_area->cr_writes.cr0 = 1; - //ctrl_area->cr_reads.cr4 = 1; + //intercept cr4 read so shadow pager can use PAE independently of guest + ctrl_area->cr_reads.cr4 = 1; ctrl_area->cr_writes.cr4 = 1; ctrl_area->cr_reads.cr3 = 1; ctrl_area->cr_writes.cr3 = 1; - v3_hook_msr(core->vm_info, EFER_MSR, - &v3_handle_efer_read, - &v3_handle_efer_write, - core); ctrl_area->instrs.INVLPG = 1; @@ -258,7 +313,6 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) { guest_state->g_pat = 0x7040600070406ULL; - } else if (core->shdw_pg_mode == NESTED_PAGING) { // Flush the TLB on entries/exits ctrl_area->TLB_CONTROL = 1; @@ -267,11 +321,11 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) { // Enable Nested Paging ctrl_area->NP_ENABLE = 1; - PrintDebug("NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE)); + PrintDebug(core->vm_info, core, "NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE)); // Set the Nested Page Table pointer if (v3_init_passthrough_pts(core) == -1) { - PrintError("Could not initialize Nested page tables\n"); + PrintError(core->vm_info, core, "Could not initialize Nested page tables\n"); return ; } @@ -285,37 +339,123 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) { &v3_handle_vm_cr_read, &v3_handle_vm_cr_write, core); + + + { +#define INT_PENDING_AMD_MSR 0xc0010055 + + v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL); + v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL); + v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL); + v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL); + v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL); + + v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL); + v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL); + v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL); + + + v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL); + v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL); + + // Passthrough read operations are ok. + v3_hook_msr(core->vm_info, INT_PENDING_AMD_MSR, NULL, v3_msr_unhandled_write, NULL); + } } int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) { - PrintDebug("Allocating VMCB\n"); - core->vmm_data = (void*)Allocate_VMCB(); + PrintDebug(core->vm_info, core, "Allocating VMCB\n"); + core->vmm_data = (void *)Allocate_VMCB(); if (core->vmm_data == NULL) { - PrintError("Could not allocate VMCB, Exiting...\n"); + PrintError(core->vm_info, core, "Could not allocate VMCB, Exiting...\n"); return -1; } if (vm_class == V3_PC_VM) { - PrintDebug("Initializing VMCB (addr=%p)\n", (void *)core->vmm_data); + PrintDebug(core->vm_info, core, "Initializing VMCB (addr=%p)\n", (void *)core->vmm_data); Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core); } else { - PrintError("Invalid VM class\n"); + PrintError(core->vm_info, core, "Invalid VM class\n"); return -1; } + core->core_run_state = CORE_STOPPED; + return 0; } int v3_deinit_svm_vmcb(struct guest_info * core) { - V3_FreePages(core->vmm_data, 1); + V3_FreePages(V3_PAddr(core->vmm_data), 1); return 0; } +#ifdef V3_CONFIG_CHECKPOINT +int v3_svm_save_core(struct guest_info * core, void * ctx){ + + vmcb_saved_state_t * guest_area = GET_VMCB_SAVE_STATE_AREA(core->vmm_data); + + // Special case saves of data we need immediate access to + // in some cases + V3_CHKPT_SAVE(ctx, "CPL", core->cpl, failout); + V3_CHKPT_SAVE(ctx,"STAR", guest_area->star, failout); + V3_CHKPT_SAVE(ctx,"CSTAR", guest_area->cstar, failout); + V3_CHKPT_SAVE(ctx,"LSTAR", guest_area->lstar, failout); + V3_CHKPT_SAVE(ctx,"SFMASK", guest_area->sfmask, failout); + V3_CHKPT_SAVE(ctx,"KERNELGSBASE", guest_area->KernelGsBase, failout); + V3_CHKPT_SAVE(ctx,"SYSENTER_CS", guest_area->sysenter_cs, failout); + V3_CHKPT_SAVE(ctx,"SYSENTER_ESP", guest_area->sysenter_esp, failout); + V3_CHKPT_SAVE(ctx,"SYSENTER_EIP", guest_area->sysenter_eip, failout); + +// and then we save the whole enchilada + if (v3_chkpt_save(ctx, "VMCB_DATA", PAGE_SIZE, core->vmm_data)) { + PrintError(core->vm_info, core, "Could not save SVM vmcb\n"); + goto failout; + } + + return 0; + + failout: + PrintError(core->vm_info, core, "Failed to save SVM state for core\n"); + return -1; + +} + +int v3_svm_load_core(struct guest_info * core, void * ctx){ + + + vmcb_saved_state_t * guest_area = GET_VMCB_SAVE_STATE_AREA(core->vmm_data); + + // Reload what we special cased, which we will overwrite in a minute + V3_CHKPT_LOAD(ctx, "CPL", core->cpl, failout); + V3_CHKPT_LOAD(ctx,"STAR", guest_area->star, failout); + V3_CHKPT_LOAD(ctx,"CSTAR", guest_area->cstar, failout); + V3_CHKPT_LOAD(ctx,"LSTAR", guest_area->lstar, failout); + V3_CHKPT_LOAD(ctx,"SFMASK", guest_area->sfmask, failout); + V3_CHKPT_LOAD(ctx,"KERNELGSBASE", guest_area->KernelGsBase, failout); + V3_CHKPT_LOAD(ctx,"SYSENTER_CS", guest_area->sysenter_cs, failout); + V3_CHKPT_LOAD(ctx,"SYSENTER_ESP", guest_area->sysenter_esp, failout); + V3_CHKPT_LOAD(ctx,"SYSENTER_EIP", guest_area->sysenter_eip, failout); + + // and then we load the whole enchilada + if (v3_chkpt_load(ctx, "VMCB_DATA", PAGE_SIZE, core->vmm_data)) { + PrintError(core->vm_info, core, "Could not load SVM vmcb\n"); + goto failout; + } + + return 0; + + failout: + PrintError(core->vm_info, core, "Failed to save SVM state for core\n"); + return -1; + +} +#endif + static int update_irq_exit_state(struct guest_info * info) { vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data)); @@ -324,8 +464,8 @@ static int update_irq_exit_state(struct guest_info * info) { if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) { -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector); +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + PrintDebug(info->vm_info, info, "INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector); #endif info->intr_core_state.irq_started = 1; @@ -335,16 +475,16 @@ static int update_irq_exit_state(struct guest_info * info) { } if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) { -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("Interrupt %d taken by guest\n", info->intr_core_state.irq_vector); +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + PrintDebug(info->vm_info, info, "Interrupt %d taken by guest\n", info->intr_core_state.irq_vector); #endif // Interrupt was taken fully vectored info->intr_core_state.irq_started = 0; } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) { -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector); +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + PrintDebug(info->vm_info, info, "EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector); #endif } @@ -369,8 +509,8 @@ static int update_irq_entry_state(struct guest_info * info) { if (info->excp_state.excp_error_code_valid) { guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code; guest_ctrl->EVENTINJ.ev = 1; -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code); +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + PrintDebug(info->vm_info, info, "Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code); #endif } @@ -378,8 +518,8 @@ static int update_irq_entry_state(struct guest_info * info) { guest_ctrl->EVENTINJ.valid = 1; -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n", +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + PrintDebug(info->vm_info, info, "<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n", (int)info->num_exits, guest_ctrl->EVENTINJ.vector, (void *)(addr_t)info->ctrl_regs.cr2, @@ -388,26 +528,36 @@ static int update_irq_entry_state(struct guest_info * info) { v3_injecting_excp(info, excp); } else if (info->intr_core_state.irq_started == 1) { -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("IRQ pending from previous injection\n"); +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + PrintDebug(info->vm_info, info, "IRQ pending from previous injection\n"); #endif guest_ctrl->guest_ctrl.V_IRQ = 1; guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector; + + // We ignore the virtual TPR on this injection + // TPR/PPR tests have already been done in the APIC. guest_ctrl->guest_ctrl.V_IGN_TPR = 1; - guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf; + guest_ctrl->guest_ctrl.V_INTR_PRIO = info->intr_core_state.irq_vector >> 4 ; // 0xf; } else { switch (v3_intr_pending(info)) { case V3_EXTERNAL_IRQ: { - uint32_t irq = v3_get_intr(info); + int irq = v3_get_intr(info); + + if (irq<0) { + break; + } guest_ctrl->guest_ctrl.V_IRQ = 1; guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq; + + // We ignore the virtual TPR on this injection + // TPR/PPR tests have already been done in the APIC. guest_ctrl->guest_ctrl.V_IGN_TPR = 1; - guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf; + guest_ctrl->guest_ctrl.V_INTR_PRIO = info->intr_core_state.irq_vector >> 4 ; // 0xf; -#ifdef CONFIG_DEBUG_INTERRUPTS - PrintDebug("Injecting Interrupt %d (EIP=%p)\n", +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + PrintDebug(info->vm_info, info, "Injecting Interrupt %d (EIP=%p)\n", guest_ctrl->guest_ctrl.V_INTR_VECTOR, (void *)(addr_t)info->rip); #endif @@ -422,6 +572,18 @@ static int update_irq_entry_state(struct guest_info * info) { break; case V3_SOFTWARE_INTR: guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR; + +#ifdef V3_CONFIG_DEBUG_INTERRUPTS + PrintDebug(info->vm_info, info, "Injecting software interrupt -- type: %d, vector: %d\n", + SVM_INJECTION_SOFT_INTR, info->intr_core_state.swintr_vector); +#endif + guest_ctrl->EVENTINJ.vector = info->intr_core_state.swintr_vector; + guest_ctrl->EVENTINJ.valid = 1; + + /* reset swintr state */ + info->intr_core_state.swintr_posted = 0; + info->intr_core_state.swintr_vector = 0; + break; case V3_VIRTUAL_IRQ: guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ; @@ -437,6 +599,26 @@ static int update_irq_entry_state(struct guest_info * info) { return 0; } +int +v3_svm_config_tsc_virtualization(struct guest_info * info) { + vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data)); + + + if (info->time_state.flags & VM_TIME_TRAP_RDTSC) { + ctrl_area->instrs.RDTSC = 1; + ctrl_area->svm_instrs.RDTSCP = 1; + } else { + ctrl_area->instrs.RDTSC = 0; + ctrl_area->svm_instrs.RDTSCP = 0; + + if (info->time_state.flags & VM_TIME_TSC_PASSTHROUGH) { + ctrl_area->TSC_OFFSET = 0; + } else { + ctrl_area->TSC_OFFSET = v3_tsc_host_offset(&info->time_state); + } + } + return 0; +} /* * CAUTION and DANGER!!! @@ -450,11 +632,21 @@ int v3_svm_enter(struct guest_info * info) { vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data)); vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data)); addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0; - - v3_adjust_time(info); + uint64_t guest_cycles = 0; // Conditionally yield the CPU if the timeslice has expired - v3_yield_cond(info); + v3_schedule(info); + +#ifdef V3_CONFIG_MEM_TRACK + v3_mem_track_entry(info); +#endif + + // Update timer devices after being in the VM before doing + // IRQ updates, so that any interrupts they raise get seen + // immediately. + v3_advance_time(info, NULL); + v3_update_timers(info); + // disable global interrupts for vm state transition v3_clgi(); @@ -466,10 +658,26 @@ int v3_svm_enter(struct guest_info * info) { guest_state->cr4 = info->ctrl_regs.cr4; guest_state->dr6 = info->dbg_regs.dr6; guest_state->dr7 = info->dbg_regs.dr7; - guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff; + + // CR8 is now updated by read/writes and it contains the APIC TPR + // the V_TPR should be just the class part of that. + // This update is here just for completeness. We currently + // are ignoring V_TPR on all injections and doing the priority logivc + // in the APIC. + // guest_ctrl->guest_ctrl.V_TPR = ((info->ctrl_regs.apic_tpr) >> 4) & 0xf; + + //guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff; + // + guest_state->rflags = info->ctrl_regs.rflags; guest_state->efer = info->ctrl_regs.efer; + /* Synchronize MSRs */ + guest_state->star = info->msrs.star; + guest_state->lstar = info->msrs.lstar; + guest_state->sfmask = info->msrs.sfmask; + guest_state->KernelGsBase = info->msrs.kern_gs_base; + guest_state->cpl = info->cpl; v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments)); @@ -478,7 +686,9 @@ int v3_svm_enter(struct guest_info * info) { guest_state->rip = info->rip; guest_state->rsp = info->vm_regs.rsp; -#ifdef CONFIG_SYMCALL + V3_FP_ENTRY_RESTORE(info); + +#ifdef V3_CONFIG_SYMCALL if (info->sym_core_state.symcall_state.sym_call_active == 0) { update_irq_entry_state(info); } @@ -486,38 +696,71 @@ int v3_svm_enter(struct guest_info * info) { update_irq_entry_state(info); #endif +#ifdef V3_CONFIG_TM_FUNC + v3_tm_check_intr_state(info, guest_ctrl, guest_state); +#endif + /* ** */ /* - PrintDebug("SVM Entry to CS=%p rip=%p...\n", + PrintDebug(info->vm_info, info, "SVM Entry to CS=%p rip=%p...\n", (void *)(addr_t)info->segments.cs.base, (void *)(addr_t)info->rip); */ -#ifdef CONFIG_SYMCALL +#ifdef V3_CONFIG_SYMCALL if (info->sym_core_state.symcall_state.sym_call_active == 1) { if (guest_ctrl->guest_ctrl.V_IRQ == 1) { - V3_Print("!!! Injecting Interrupt during Sym call !!!\n"); + V3_Print(info->vm_info, info, "!!! Injecting Interrupt during Sym call !!!\n"); } } #endif - v3_update_timers(info); - guest_ctrl->TSC_OFFSET = v3_tsc_host_offset(&info->time_state); + v3_svm_config_tsc_virtualization(info); + + //V3_Print(info->vm_info, info, "Calling v3_svm_launch\n"); + { + uint64_t entry_tsc = 0; + uint64_t exit_tsc = 0; + +#ifdef V3_CONFIG_PWRSTAT_TELEMETRY + v3_pwrstat_telemetry_enter(info); +#endif + +#ifdef V3_CONFIG_PMU_TELEMETRY + v3_pmu_telemetry_enter(info); +#endif + + + rdtscll(entry_tsc); + + v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]); - //V3_Print("Calling v3_svm_launch\n"); + rdtscll(exit_tsc); + +#ifdef V3_CONFIG_PMU_TELEMETRY + v3_pmu_telemetry_exit(info); +#endif + +#ifdef V3_CONFIG_PWRSTAT_TELEMETRY + v3_pwrstat_telemetry_exit(info); +#endif + + guest_cycles = exit_tsc - entry_tsc; + } - v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[info->cpu_id]); - //V3_Print("SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip); + //V3_Print(info->vm_info, info, "SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip); v3_last_exit = (uint32_t)(guest_ctrl->exit_code); - //PrintDebug("SVM Returned\n"); - + v3_advance_time(info, &guest_cycles); + info->num_exits++; + V3_FP_EXIT_SAVE(info); + // Save Guest state from VMCB info->rip = guest_state->rip; info->vm_regs.rsp = guest_state->rsp; @@ -531,23 +774,31 @@ int v3_svm_enter(struct guest_info * info) { info->ctrl_regs.cr4 = guest_state->cr4; info->dbg_regs.dr6 = guest_state->dr6; info->dbg_regs.dr7 = guest_state->dr7; - info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR; + // + // We do not track this anymore + // V_TPR is ignored and we do the logic in the APIC + //info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR; + // info->ctrl_regs.rflags = guest_state->rflags; info->ctrl_regs.efer = guest_state->efer; + /* Synchronize MSRs */ + info->msrs.star = guest_state->star; + info->msrs.lstar = guest_state->lstar; + info->msrs.sfmask = guest_state->sfmask; + info->msrs.kern_gs_base = guest_state->KernelGsBase; + v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments)); info->cpu_mode = v3_get_vm_cpu_mode(info); info->mem_mode = v3_get_vm_mem_mode(info); /* ** */ - // save exit info here exit_code = guest_ctrl->exit_code; exit_info1 = guest_ctrl->exit_info1; exit_info2 = guest_ctrl->exit_info2; - -#ifdef CONFIG_SYMCALL +#ifdef V3_CONFIG_SYMCALL if (info->sym_core_state.symcall_state.sym_call_active == 0) { update_irq_exit_state(info); } @@ -555,21 +806,38 @@ int v3_svm_enter(struct guest_info * info) { update_irq_exit_state(info); #endif - // reenable global interrupts after vm exit v3_stgi(); // Conditionally yield the CPU if the timeslice has expired - v3_yield_cond(info); + v3_schedule(info); + // This update timers is for time-dependent handlers + // if we're slaved to host time + v3_advance_time(info, NULL); + v3_update_timers(info); - if (v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2) != 0) { - PrintError("Error in SVM exit handler\n"); - return -1; + { + int ret = v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2); + + if (ret != 0) { + PrintError(info->vm_info, info, "Error in SVM exit handler (ret=%d)\n", ret); + PrintError(info->vm_info, info, " last Exit was %d (exit code=0x%llx)\n", v3_last_exit, (uint64_t) exit_code); + return -1; + } + } + + if (info->timeouts.timeout_active) { + /* Check to see if any timeouts have expired */ + v3_handle_timeouts(info, guest_cycles); } +#ifdef V3_CONFIG_MEM_TRACK + v3_mem_track_exit(info); +#endif + return 0; } @@ -579,39 +847,77 @@ int v3_start_svm_guest(struct guest_info * info) { // vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data)); // vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data)); - PrintDebug("Starting SVM core %u\n", info->cpu_id); + PrintDebug(info->vm_info, info, "Starting SVM core %u (on logical core %u)\n", info->vcpu_id, info->pcpu_id); - if (info->cpu_id == 0) { - info->core_run_state = CORE_RUNNING; - info->vm_info->run_state = VM_RUNNING; - } else { - PrintDebug("SVM core %u: Waiting for core initialization\n", info->cpu_id); - while (info->core_run_state == CORE_STOPPED) { - v3_yield(info); - //PrintDebug("SVM core %u: still waiting for INIT\n",info->cpu_id); - } +#ifdef V3_CONFIG_HVM + if (v3_setup_hvm_hrt_core_for_boot(info)) { + PrintError(info->vm_info, info, "Failed to setup HRT core...\n"); + return -1; + } +#endif + + + - PrintDebug("SVM core %u initialized\n", info->cpu_id); - } - PrintDebug("SVM 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->segments.cs.limit, (void *)(info->rip)); + while (1) { + if (info->core_run_state == CORE_STOPPED) { + if (info->vcpu_id == 0) { + info->core_run_state = CORE_RUNNING; + } else { + PrintDebug(info->vm_info, info, "SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id); - PrintDebug("SVM core %u: Launching SVM VM (vmcb=%p)\n", info->cpu_id, (void *)info->vmm_data); - //PrintDebugVMCB((vmcb_t*)(info->vmm_data)); - - v3_start_time(info); + V3_NO_WORK(info); - while (1) { + while (info->core_run_state == CORE_STOPPED) { + + if (info->vm_info->run_state == VM_STOPPED) { + // The VM was stopped before this core was initialized. + return 0; + } + + V3_STILL_NO_WORK(info); + //PrintDebug(info->vm_info, info, "SVM core %u: still waiting for INIT\n", info->vcpu_id); + } + + V3_HAVE_WORK_AGAIN(info); + + PrintDebug(info->vm_info, info, "SVM core %u(on %u) initialized\n", info->vcpu_id, info->pcpu_id); + + // We'll be paranoid about race conditions here + v3_wait_at_barrier(info); + } + + PrintDebug(info->vm_info, info, "SVM core %u(on %u): I am starting at CS=0x%x (base=0x%p, limit=0x%x), RIP=0x%p\n", + info->vcpu_id, info->pcpu_id, + info->segments.cs.selector, (void *)(info->segments.cs.base), + info->segments.cs.limit, (void *)(info->rip)); + + + + PrintDebug(info->vm_info, info, "SVM core %u: Launching SVM VM (vmcb=%p) (on cpu %u)\n", + info->vcpu_id, (void *)info->vmm_data, info->pcpu_id); + //PrintDebugVMCB((vmcb_t*)(info->vmm_data)); + + v3_start_time(info); + } + if (info->vm_info->run_state == VM_STOPPED) { info->core_run_state = CORE_STOPPED; break; } + +#ifdef V3_CONFIG_PMU_TELEMETRY + v3_pmu_telemetry_start(info); +#endif + +#ifdef V3_CONFIG_PWRSTAT_TELEMETRY + v3_pwrstat_telemetry_start(info); +#endif if (v3_svm_enter(info) == -1) { vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data)); @@ -620,17 +926,17 @@ int v3_start_svm_guest(struct guest_info * info) { info->vm_info->run_state = VM_ERROR; - V3_Print("SVM core %u: SVM ERROR!!\n", info->cpu_id); + V3_Print(info->vm_info, info, "SVM core %u: SVM ERROR!!\n", info->vcpu_id); v3_print_guest_state(info); - V3_Print("SVM core %u: SVM Exit Code: %p\n", info->cpu_id, (void *)(addr_t)guest_ctrl->exit_code); + V3_Print(info->vm_info, info, "SVM core %u: SVM Exit Code: %p\n", info->vcpu_id, (void *)(addr_t)guest_ctrl->exit_code); - V3_Print("SVM core %u: exit_info1 low = 0x%.8x\n", info->cpu_id, *(uint_t*)&(guest_ctrl->exit_info1)); - V3_Print("SVM core %u: exit_info1 high = 0x%.8x\n", info->cpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4)); + V3_Print(info->vm_info, info, "SVM core %u: exit_info1 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info1)); + V3_Print(info->vm_info, info, "SVM core %u: exit_info1 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4)); - V3_Print("SVM core %u: exit_info2 low = 0x%.8x\n", info->cpu_id, *(uint_t*)&(guest_ctrl->exit_info2)); - V3_Print("SVM core %u: exit_info2 high = 0x%.8x\n", info->cpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4)); + V3_Print(info->vm_info, info, "SVM core %u: exit_info2 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info2)); + V3_Print(info->vm_info, info, "SVM core %u: exit_info2 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4)); linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs)); @@ -640,16 +946,18 @@ int v3_start_svm_guest(struct guest_info * info) { v3_gva_to_hva(info, linear_addr, &host_addr); } - V3_Print("SVM core %u: Host Address of rip = 0x%p\n", info->cpu_id, (void *)host_addr); + V3_Print(info->vm_info, info, "SVM core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr); - V3_Print("SVM core %u: Instr (15 bytes) at %p:\n", info->cpu_id, (void *)host_addr); + V3_Print(info->vm_info, info, "SVM 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); - + break; } - + + v3_wait_at_barrier(info); + if (info->vm_info->run_state == VM_STOPPED) { info->core_run_state = CORE_STOPPED; @@ -657,14 +965,23 @@ int v3_start_svm_guest(struct guest_info * info) { } + /* - if ((info->num_exits % 5000) == 0) { - V3_Print("SVM Exit number %d\n", (uint32_t)info->num_exits); + if ((info->num_exits % 50000) == 0) { + V3_Print(info->vm_info, info, "SVM Exit number %d\n", (uint32_t)info->num_exits); + v3_print_guest_state(info); } */ } +#ifdef V3_CONFIG_PMU_TELEMETRY + v3_pmu_telemetry_end(info); +#endif + +#ifdef V3_CONFIG_PWRSTAT_TELEMETRY + v3_pwrstat_telemetry_end(info); +#endif // Need to take down the other cores on error... return 0; @@ -673,6 +990,31 @@ int v3_start_svm_guest(struct guest_info * info) { +int v3_reset_svm_vm_core(struct guest_info * core, addr_t rip) { + // init vmcb_bios + + // Write the RIP, CS, and descriptor + // assume the rest is already good to go + // + // vector VV -> rip at 0 + // CS = VV00 + // This means we start executing at linear address VV000 + // + // So the selector needs to be VV00 + // and the base needs to be VV000 + // + core->rip = 0; + core->segments.cs.selector = rip << 8; + core->segments.cs.limit = 0xffff; + core->segments.cs.base = rip << 12; + + return 0; +} + + + + + /* Checks machine SVM capability */ /* Implemented from: AMD Arch Manual 3, sect 15.4 */ @@ -682,38 +1024,38 @@ int v3_is_svm_capable() { v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx); - PrintDebug("CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx); + PrintDebug(VM_NONE, VCORE_NONE, "CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx); if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) { - V3_Print("SVM Not Available\n"); + V3_Print(VM_NONE, VCORE_NONE, "SVM Not Available\n"); return 0; } else { v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low); - PrintDebug("SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low); + PrintDebug(VM_NONE, VCORE_NONE, "SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low); if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) { - V3_Print("SVM is available but is disabled.\n"); + V3_Print(VM_NONE, VCORE_NONE, "SVM is available but is disabled.\n"); v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx); - PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx); + PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx); if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) { - V3_Print("SVM BIOS Disabled, not unlockable\n"); + V3_Print(VM_NONE, VCORE_NONE, "SVM BIOS Disabled, not unlockable\n"); } else { - V3_Print("SVM is locked with a key\n"); + V3_Print(VM_NONE, VCORE_NONE, "SVM is locked with a key\n"); } return 0; } else { - V3_Print("SVM is available and enabled.\n"); + V3_Print(VM_NONE, VCORE_NONE, "SVM is available and enabled.\n"); v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx); - PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax); - PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx); - PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx); - PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx); + PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax); + PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx); + PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx); + PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx); return 1; } @@ -722,19 +1064,20 @@ int v3_is_svm_capable() { static int has_svm_nested_paging() { uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0; - + v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx); - - //PrintDebug("CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx); - + + //PrintDebug(VM_NONE, VCORE_NONE, "CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx); + if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) { - V3_Print("SVM Nested Paging not supported\n"); + V3_Print(VM_NONE, VCORE_NONE, "SVM Nested Paging not supported\n"); return 0; } else { - V3_Print("SVM Nested Paging supported\n"); + V3_Print(VM_NONE, VCORE_NONE, "SVM Nested Paging supported\n"); return 1; } -} + } + void v3_init_svm_cpu(int cpu_id) { @@ -746,17 +1089,22 @@ void v3_init_svm_cpu(int cpu_id) { msr.e_reg.low |= EFER_MSR_svm_enable; v3_set_msr(EFER_MSR, 0, msr.e_reg.low); - V3_Print("SVM Enabled\n"); + V3_Print(VM_NONE, VCORE_NONE, "SVM Enabled\n"); // Setup the host state save area - host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4); + host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4); // need not be shadow-safe, not exposed to guest + + if (!host_vmcbs[cpu_id]) { + PrintError(VM_NONE, VCORE_NONE, "Failed to allocate VMCB\n"); + return; + } /* 64-BIT-ISSUE */ // msr.e_reg.high = 0; //msr.e_reg.low = (uint_t)host_vmcb; msr.r_reg = host_vmcbs[cpu_id]; - PrintDebug("Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]); + PrintDebug(VM_NONE, VCORE_NONE, "Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]); v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low); @@ -769,7 +1117,27 @@ void v3_init_svm_cpu(int cpu_id) { +void v3_deinit_svm_cpu(int cpu_id) { + reg_ex_t msr; + extern v3_cpu_arch_t v3_cpu_types[]; + + // reset SVM_VM_HSAVE_PA_MSR + // Does setting it to NULL disable?? + msr.r_reg = 0; + v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low); + // Disable SVM? + v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low)); + msr.e_reg.low &= ~EFER_MSR_svm_enable; + v3_set_msr(EFER_MSR, 0, msr.e_reg.low); + + v3_cpu_types[cpu_id] = V3_INVALID_CPU; + + V3_FreePages((void *)host_vmcbs[cpu_id], 4); + + V3_Print(VM_NONE, VCORE_NONE, "Host CPU %d host area freed, and SVM disabled\n", cpu_id); + return; +} @@ -850,7 +1218,7 @@ void v3_init_svm_cpu(int cpu_id) { end <<= 32; end += end_lo; - PrintDebug("VMSave Cycle Latency: %d\n", (uint32_t)(end - start)); + PrintDebug(core->vm_info, core, "VMSave Cycle Latency: %d\n", (uint32_t)(end - start)); __asm__ __volatile__ ( "rdtsc ; " @@ -872,7 +1240,7 @@ void v3_init_svm_cpu(int cpu_id) { end += end_lo; - PrintDebug("VMLoad Cycle Latency: %d\n", (uint32_t)(end - start)); + PrintDebug(core->vm_info, core, "VMLoad Cycle Latency: %d\n", (uint32_t)(end - start)); } /* End Latency Test */ @@ -969,7 +1337,12 @@ void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) { ctrl_area->instrs.IOIO_PROT = 1; - ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3); + ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3); // need not be shadow-safe, not exposed to guest + + if (!ctrl_area->IOPM_BASE_PA) { + PrintError(core->vm_info, core, "Cannot allocate IO bitmap\n"); + return; + } { reg_ex_t tmp_reg; @@ -994,12 +1367,12 @@ void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) { GetGDTR(gdt_buf); gdt_base = *(ulong_t*)((uchar_t*)gdt_buf + 2) & 0xffffffff; gdt_limit = *(ushort_t*)(gdt_buf) & 0xffff; - PrintDebug("GDT: base: %x, limit: %x\n", gdt_base, gdt_limit); + PrintDebug(core->vm_info, core, "GDT: base: %x, limit: %x\n", gdt_base, gdt_limit); GetIDTR(idt_buf); idt_base = *(ulong_t*)(idt_buf + 2) & 0xffffffff; idt_limit = *(ushort_t*)(idt_buf) & 0xffff; - PrintDebug("IDT: base: %x, limit: %x\n",idt_base, idt_limit); + PrintDebug(core->vm_info, core, "IDT: base: %x, limit: %x\n",idt_base, idt_limit); // gdt_base -= 0x2000; @@ -1024,7 +1397,7 @@ void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) { // Enable Nested Paging ctrl_area->NP_ENABLE = 1; - PrintDebug("NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE)); + PrintDebug(core->vm_info, core, "NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE)); // Set the Nested Page Table pointer ctrl_area->N_CR3 |= ((addr_t)vm_info.page_tables & 0xfffff000); @@ -1035,8 +1408,8 @@ void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) { guest_state->g_pat = 0x7040600070406ULL; - PrintDebug("Set Nested CR3: lo: 0x%x hi: 0x%x\n", (uint_t)*(&(ctrl_area->N_CR3)), (uint_t)*((unsigned char *)&(ctrl_area->N_CR3) + 4)); - PrintDebug("Set Guest CR3: lo: 0x%x hi: 0x%x\n", (uint_t)*(&(guest_state->cr3)), (uint_t)*((unsigned char *)&(guest_state->cr3) + 4)); + PrintDebug(core->vm_info, core, "Set Nested CR3: lo: 0x%x hi: 0x%x\n", (uint_t)*(&(ctrl_area->N_CR3)), (uint_t)*((unsigned char *)&(ctrl_area->N_CR3) + 4)); + PrintDebug(core->vm_info, core, "Set Guest CR3: lo: 0x%x hi: 0x%x\n", (uint_t)*(&(guest_state->cr3)), (uint_t)*((unsigned char *)&(guest_state->cr3) + 4)); // Enable Paging // guest_state->cr0 |= 0x80000000; }