Palacios Public Git Repository

To checkout Palacios execute

  git clone http://v3vee.org/palacios/palacios.web/palacios.git
This will give you the master branch. You probably want the devel branch or one of the release branches. To switch to the devel branch, simply execute
  cd palacios
  git checkout --track -b devel origin/devel
The other branches are similar.


fixed vmcs guest pointer check location
[palacios.git] / palacios / src / palacios / vmx.c
index f71df0a..2c6b573 100644 (file)
@@ -28,6 +28,7 @@
 #include <palacios/vmm_lowlevel.h>
 #include <palacios/vmm_ctrl_regs.h>
 #include <palacios/vmm_config.h>
+#include <palacios/vmm_time.h>
 #include <palacios/vm_guest_mem.h>
 #include <palacios/vmm_direct_paging.h>
 #include <palacios/vmx_io.h>
@@ -41,8 +42,7 @@
 
 
 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};
 
 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);
@@ -72,51 +72,8 @@ 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;
@@ -140,9 +97,15 @@ static addr_t allocate_vmcs() {
 
 static int init_vmcs_bios(struct guest_info * info, 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) {
         PrintError("VMPTRLD failed\n");
@@ -420,7 +383,12 @@ static int init_vmcs_bios(struct guest_info * info, struct vmx_data * vmx_state)
     }
 
 
-    vmx_state->state = VMXASSIST_DISABLED;
+    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();
 
     return 0;
 }
@@ -439,6 +407,7 @@ 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;
+    vmx_state->state = VMX_UNLAUNCHED;
 
     PrintDebug("Initializing VMCS (addr=%p)\n", info->vmm_data);
     
@@ -658,19 +627,27 @@ int v3_vmx_enter(struct guest_info * info) {
     int ret = 0;
     uint32_t tsc_offset_low, tsc_offset_high;
     struct vmx_exit_info exit_info;
+    struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 
     // Conditionally yield the CPU if the timeslice has expired
     v3_yield_cond(info);
 
-    /* If this guest is frequency-lagged behind host time, wait 
-     * for the appropriate host time before resuming the guest. */
+    // Perform any additional yielding needed for time adjustment
     v3_adjust_time(info);
 
-    // v3_print_guest_state(info);
+    // Update timer devices prior to entering VM.
+    v3_update_timers(info);
 
     // disable global interrupts for vm state transition
     v3_disable_ints();
 
+
+    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;
+    }
+
+
     v3_vmx_restore_vmcs(info);
 
 
@@ -688,20 +665,24 @@ int v3_vmx_enter(struct guest_info * info) {
        vmcs_write(VMCS_GUEST_CR3, guest_cr3);
     }
 
-    v3_update_timers(info);
+    // Perform last-minute time bookkeeping prior to entering the VM
+    v3_time_enter_vm(info);
 
     tsc_offset_high = (uint32_t)((v3_tsc_host_offset(&info->time_state) >> 32) & 0xffffffff);
     tsc_offset_low = (uint32_t)(v3_tsc_host_offset(&info->time_state) & 0xffffffff);
     check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
     check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
 
-    if (info->vm_info->run_state == VM_STOPPED) {
+
+    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 {
+       V3_ASSERT(vmx_info->state != VMX_UNLAUNCHED);
        ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs));
     }
-
+    
     //  PrintDebug("VMX Exit: ret=%d\n", ret);
 
     if (ret != VMX_SUCCESS) {
@@ -713,6 +694,9 @@ int v3_vmx_enter(struct guest_info * info) {
        return -1;
     }
 
+    // Immediate exit from VM time bookkeeping
+    v3_time_exit_vm(info);
+
     info->num_exits++;
 
     /* Update guest state */
@@ -745,6 +729,17 @@ int v3_vmx_enter(struct guest_info * info) {
     update_irq_exit_state(info);
 #endif
 
+    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 CONFIG_DEBUG_INTERRUPTS
+        PrintDebug("Interrupts available again! (RIP=%llx)\n", info->rip);
+#endif
+    }
+
     // reenable global interrupts after vm exit
     v3_enable_ints();
 
@@ -847,6 +842,12 @@ int v3_is_vmx_capable() {
 }
 
 static int has_vmx_nested_paging() {
+    /* We assume that both EPT and unrestricted guest mode (Intel's Virtual Real Mode) 
+     * are mutually assured. i.e. We have either both or neither.
+     */
+
+    
+
     return 0;
 }