#include <palacios/vmm_direct_paging.h>
#include <palacios/vmx_io.h>
#include <palacios/vmx_msr.h>
+#include <palacios/vmm_decoder.h>
+#include <palacios/vmm_barrier.h>
+
+#ifdef V3_CONFIG_CHECKPOINT
+#include <palacios/vmm_checkpoint.h>
+#endif
#include <palacios/vmx_ept.h>
#include <palacios/vmx_assist.h>
extern v3_cpu_arch_t v3_cpu_types[];
-static addr_t active_vmcs_ptrs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_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);
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) {
#ifdef __V3_64BIT__
+ // Ensure host runs in 64-bit mode at each VM EXIT
vmx_state->exit_ctrls.host_64_on = 1;
#endif
-
- /* Not sure how exactly to handle this... */
+ // Hook all accesses to EFER register
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;
+ // Restore host's EFER register on each VM EXIT
vmx_state->exit_ctrls.ld_efer = 1;
+
+ // 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;
- vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE);
+ // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
+ vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE);
/* Setup paging */
return -1;
}
+ PrintDebug("Serializing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
+ vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
+
return 0;
}
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_FreePages(V3_PAddr(vmx_state->msr_area), 1);
V3_Free(vmx_state);
}
+
+#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
+
+
static int update_irq_exit_state(struct guest_info * info) {
struct vmx_exit_idt_vec_info idt_vec_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 (active_vmcs_ptrs[V3_Get_CPU()] != vmx_info->vmcs_ptr_phys) {
+ if (vmcs_store() != vmx_info->vmcs_ptr_phys) {
+ vmcs_clear(vmx_info->vmcs_ptr_phys);
vmcs_load(vmx_info->vmcs_ptr_phys);
- active_vmcs_ptrs[V3_Get_CPU()] = vmx_info->vmcs_ptr_phys;
+ vmx_info->state = VMX_UNLAUNCHED;
}
-
v3_vmx_restore_vmcs(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");
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 {
exit_log[info->num_exits % 10] = exit_info;
-
#ifdef V3_CONFIG_SYMCALL
if (info->sym_core_state.symcall_state.sym_call_active == 0) {
update_irq_exit_state(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;
}
}
PrintDebug("VMX core %u initialized\n", info->vcpu_id);
+
+ // We'll be paranoid about race conditions here
+ v3_wait_at_barrier(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) {
}
-int v3_reset_vmx_cpu(struct guest_info * core, addr_t rip) {
+int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) {
// init vmcs bios
if ((core->shdw_pg_mode == NESTED_PAGING) &&
void v3_init_vmx_cpu(int cpu_id) {
+ addr_t vmx_on_region = 0;
if (cpu_id == 0) {
if (v3_init_vmx_hw(&hw_info) == -1) {
// 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 (vmx_on(host_vmcs_ptrs[cpu_id]) == VMX_SUCCESS) {
+ 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);
}
-
+
+ PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]);
{
struct vmx_sec_proc_ctrls sec_proc_ctrls;
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;
+ }
}