#include <palacios/svm_msr.h>
#include <palacios/vmm_rbtree.h>
+#include <palacios/vmm_barrier.h>
+
+#ifdef V3_CONFIG_CHECKPOINT
+#include <palacios/vmm_checkpoint.h>
+#endif
#include <palacios/vmm_direct_paging.h>
#include <palacios/vmm_sprintf.h>
-#ifndef CONFIG_DEBUG_SVM
+#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};
static vmcb_t * Allocate_VMCB() {
- vmcb_t * vmcb_page = (vmcb_t *)V3_VAddr(V3_AllocPages(1));
+ vmcb_t * vmcb_page = NULL;
+ addr_t vmcb_pa = (addr_t)V3_AllocPages(1);
+
+ if ((void *)vmcb_pa == NULL) {
+ PrintError("Error allocating VMCB\n");
+ return NULL;
+ }
+
+ vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa);
memset(vmcb_page, 0, 4096);
}
+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);
ctrl_area->svm_instrs.STGI = 1;
ctrl_area->svm_instrs.CLGI = 1;
ctrl_area->svm_instrs.SKINIT = 1;
- ctrl_area->svm_instrs.RDTSCP = 1;
ctrl_area->svm_instrs.ICEBP = 1;
ctrl_area->svm_instrs.WBINVD = 1;
ctrl_area->svm_instrs.MONITOR = 1;
ctrl_area->instrs.HLT = 1;
-#ifdef CONFIG_TIME_VIRTUALIZE_TSC
- ctrl_area->instrs.rdtsc = 1;
- ctrl_area->svm_instrs.rdtscp = 1;
+#ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC
+ ctrl_area->instrs.RDTSC = 1;
+ ctrl_area->svm_instrs.RDTSCP = 1;
#endif
// guest_state->cr0 = 0x00000001; // PE
ctrl_area->instrs.INTR = 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");
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;
guest_state->g_pat = 0x7040600070406ULL;
}
+
+ /* tell the guest that we don't support SVM */
+ v3_hook_msr(core->vm_info, SVM_VM_CR_MSR,
+ &v3_handle_vm_cr_read,
+ &v3_handle_vm_cr_write,
+ core);
}
-int v3_init_svm_vmcb(struct guest_info * info, v3_vm_class_t vm_class) {
+int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) {
PrintDebug("Allocating VMCB\n");
- info->vmm_data = (void*)Allocate_VMCB();
+ core->vmm_data = (void *)Allocate_VMCB();
+ if (core->vmm_data == NULL) {
+ PrintError("Could not allocate VMCB, Exiting...\n");
+ return -1;
+ }
+
if (vm_class == V3_PC_VM) {
- PrintDebug("Initializing VMCB (addr=%p)\n", (void *)info->vmm_data);
- Init_VMCB_BIOS((vmcb_t*)(info->vmm_data), info);
+ PrintDebug("Initializing VMCB (addr=%p)\n", (void *)core->vmm_data);
+ Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
} else {
PrintError("Invalid VM class\n");
return -1;
}
+int v3_deinit_svm_vmcb(struct guest_info * core) {
+ 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){
+
+ v3_chkpt_save_8(ctx, "cpl", &(core->cpl));
+ v3_chkpt_save(ctx, "vmcb_data", PAGE_SIZE, core->vmm_data);
+
+ return 0;
+}
+
+int v3_svm_load_core(struct guest_info * core, void * ctx){
+
+ v3_chkpt_load_8(ctx, "cpl", &(core->cpl));
+
+ if (v3_chkpt_load(ctx, "vmcb_data", PAGE_SIZE, core->vmm_data) == -1) {
+ return -1;
+ }
+
+ return 0;
+}
+#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));
if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
-#ifdef CONFIG_DEBUG_INTERRUPTS
+#ifdef V3_CONFIG_DEBUG_INTERRUPTS
PrintDebug("INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
#endif
}
if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
-#ifdef CONFIG_DEBUG_INTERRUPTS
+#ifdef V3_CONFIG_DEBUG_INTERRUPTS
PrintDebug("Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
#endif
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
+#ifdef V3_CONFIG_DEBUG_INTERRUPTS
PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
#endif
}
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
+#ifdef V3_CONFIG_DEBUG_INTERRUPTS
PrintDebug("Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
#endif
}
guest_ctrl->EVENTINJ.valid = 1;
-#ifdef CONFIG_DEBUG_INTERRUPTS
+#ifdef V3_CONFIG_DEBUG_INTERRUPTS
PrintDebug("<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n",
(int)info->num_exits,
guest_ctrl->EVENTINJ.vector,
v3_injecting_excp(info, excp);
} else if (info->intr_core_state.irq_started == 1) {
-#ifdef CONFIG_DEBUG_INTERRUPTS
+#ifdef V3_CONFIG_DEBUG_INTERRUPTS
PrintDebug("IRQ pending from previous injection\n");
#endif
guest_ctrl->guest_ctrl.V_IRQ = 1;
guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
-#ifdef CONFIG_DEBUG_INTERRUPTS
+#ifdef V3_CONFIG_DEBUG_INTERRUPTS
PrintDebug("Injecting Interrupt %d (EIP=%p)\n",
guest_ctrl->guest_ctrl.V_INTR_VECTOR,
(void *)(addr_t)info->rip);
break;
case V3_SOFTWARE_INTR:
guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
+
+#ifdef V3_CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("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;
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;
+ sint64_t tsc_offset;
// Conditionally yield the CPU if the timeslice has expired
v3_yield_cond(info);
+ // Perform any additional yielding needed for time adjustment
+ v3_adjust_time(info);
+
// disable global interrupts for vm state transition
v3_clgi();
+ // Update timer devices after being in the VM, with interupts
+ // disabled, but before doing IRQ updates, so that any interrupts they
+ //raise get seen immediately.
+ v3_update_timers(info);
+
// Synchronize the guest state to the VMCB
guest_state->cr0 = info->ctrl_regs.cr0;
guest_state->cr2 = info->ctrl_regs.cr2;
guest_state->rip = info->rip;
guest_state->rsp = info->vm_regs.rsp;
-#ifdef CONFIG_SYMCALL
+#ifdef V3_CONFIG_SYMCALL
if (info->sym_core_state.symcall_state.sym_call_active == 0) {
update_irq_entry_state(info);
}
(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");
}
#endif
- v3_update_timers(info);
-
- /* If this guest is frequency-lagged behind host time, wait
- * for the appropriate host time before resuming the guest. */
- v3_adjust_time(info);
+ v3_time_enter_vm(info);
+ tsc_offset = v3_tsc_host_offset(&info->time_state);
+ guest_ctrl->TSC_OFFSET = tsc_offset;
- guest_ctrl->TSC_OFFSET = v3_tsc_host_offset(&info->time_state);
//V3_Print("Calling v3_svm_launch\n");
- v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[info->cpu_id]);
+ v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
//V3_Print("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");
-
+ // Immediate exit from VM time bookkeeping
+ v3_time_exit_vm(info);
+
info->num_exits++;
// Save Guest state from VMCB
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);
}
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);
-
-
- 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("Error in SVM exit handler (ret=%d)\n", ret);
+ PrintError(" last Exit was %d (exit code=0x%llx)\n", v3_last_exit, (uint64_t) exit_code);
+ return -1;
+ }
}
// 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("Starting SVM core %u (on logical core %u)\n", info->vcpu_id, info->pcpu_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("SVM core %u: Waiting for core initialization\n", info->cpu_id);
+ PrintDebug("SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id);
while (info->core_run_state == CORE_STOPPED) {
v3_yield(info);
- //PrintDebug("SVM core %u: still waiting for INIT\n",info->cpu_id);
+ //PrintDebug("SVM core %u: still waiting for INIT\n", info->vcpu_id);
}
- PrintDebug("SVM core %u initialized\n", info->cpu_id);
+ PrintDebug("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("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),
+ PrintDebug("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("SVM core %u: Launching SVM VM (vmcb=%p)\n", info->cpu_id, (void *)info->vmm_data);
+ PrintDebug("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);
while (1) {
+
+ if (info->vm_info->run_state == VM_STOPPED) {
+ info->core_run_state = CORE_STOPPED;
+ break;
+ }
+
if (v3_svm_enter(info) == -1) {
vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
addr_t host_addr;
info->vm_info->run_state = VM_ERROR;
- V3_Print("SVM core %u: SVM ERROR!!\n", info->cpu_id);
+ V3_Print("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("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("SVM core %u: exit_info1 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
+ V3_Print("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("SVM core %u: exit_info2 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
+ V3_Print("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));
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("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("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;
+ break;
+ }
+
+
/*
- if ((info->num_exits % 5000) == 0) {
+ if ((info->num_exits % 50000) == 0) {
V3_Print("SVM Exit number %d\n", (uint32_t)info->num_exits);
+ v3_print_guest_state(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 */
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);
-
+
if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
V3_Print("SVM Nested Paging not supported\n");
return 0;
V3_Print("SVM Nested Paging supported\n");
return 1;
}
-}
+ }
+
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("Host CPU %d host area freed, and SVM disabled\n", cpu_id);
+ return;
+}