Basic HRT startup for HVM, plus assorted cleanup

[palacios.git] / palacios / src / palacios / vmx.c

diff --git a/palacios/src/palacios/vmx.c b/palacios/src/palacios/vmx.c
index 3a64842..f0823b4 100644 (file)
--- a/palacios/src/palacios/vmx.c
+++ b/palacios/src/palacios/vmx.c
@@ -44,6 +44,10 @@
 #include <palacios/vmx_assist.h>
 #include <palacios/vmx_hw_info.h>
 
+#ifdef V3_CONFIG_MEM_TRACK
+#include <palacios/vmm_mem_track.h>
+#endif 
+
 #ifndef V3_CONFIG_DEBUG_VMX
 #undef PrintDebug
 #define PrintDebug(fmt, args...)
@@ -97,7 +101,7 @@ static addr_t allocate_vmcs() {
 
     PrintDebug(VM_NONE, VCORE_NONE, "Allocating page\n");
 
-    temp = V3_AllocPages(1);
+    temp = V3_AllocPages(1); // need not be shadow-safe, not exposed to guest
     if (!temp) { 
        PrintError(VM_NONE, VCORE_NONE, "Cannot allocate VMCS\n");
        return -1;
@@ -270,16 +274,19 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state)
 
 
        // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
-       vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE);
+       vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE );
 
-        core->ctrl_regs.cr3 = core->direct_map_pt;
+       v3_activate_passthrough_pt(core);
 
         // 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;
-       
+
+       // Note that we intercept cr4.pae writes
+       // and we have cr4 read-shadowed to the shadow pager's cr4
+
        vmx_state->pri_proc_ctrls.invlpg_exit = 1;
        
        /* Add page fault exits */
@@ -326,7 +333,7 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state)
 
 
 
-       if (v3_init_ept(core, &hw_info) == -1) {
+       if (v3_init_nested_paging_core(core, &hw_info) == -1) {
            PrintError(core->vm_info, core, "Error initializing EPT\n");
            return -1;
        }
@@ -424,7 +431,7 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state)
        ((struct cr0_32 *)&(core->shdw_pg_state.guest_cr0))->ne = 1;
        ((struct cr0_32 *)&(core->shdw_pg_state.guest_cr0))->cd = 0;
 
-       if (v3_init_ept(core, &hw_info) == -1) {
+       if (v3_init_nested_paging_core(core, &hw_info) == -1) {
            PrintError(core->vm_info, core, "Error initializing EPT\n");
            return -1;
        }
@@ -456,7 +463,7 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state)
            return -1;
        }
 
-       vmx_state->msr_area_paddr = (addr_t)V3_AllocPages(1);
+       vmx_state->msr_area_paddr = (addr_t)V3_AllocPages(1); // need not be shadow-safe, not exposed to guest
        
        if (vmx_state->msr_area_paddr == (addr_t)NULL) {
            PrintError(core->vm_info, core, "could not allocate msr load/store area\n");
@@ -687,7 +694,7 @@ int v3_vmx_load_core(struct guest_info * core, void * ctx){
   struct cr0_32 * shadow_cr0;
   addr_t vmcs_page_paddr;  //HPA
   
-  vmcs_page_paddr = (addr_t) V3_AllocPages(1);
+  vmcs_page_paddr = (addr_t) V3_AllocPages(1); // need not be shadow-safe, not exposed to guest
   
   if (!vmcs_page_paddr) { 
     PrintError(core->vm_info, core, "Could not allocate space for a vmcs in VMX\n");
@@ -975,6 +982,10 @@ int v3_vmx_enter(struct guest_info * info) {
     // Conditionally yield the CPU if the timeslice has expired
     v3_schedule(info);
 
+#ifdef V3_CONFIG_MEM_TRACK
+    v3_mem_track_entry(info);
+#endif 
+
     // Update timer devices late after being in the VM so that as much 
     // of the 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 
@@ -1028,7 +1039,8 @@ int v3_vmx_enter(struct guest_info * info) {
        
        check_vmcs_write(VMCS_PREEMPT_TIMER, preempt_window);
     }
-   
+
+    V3_FP_ENTRY_RESTORE(info);
 
     {  
        uint64_t entry_tsc = 0;
@@ -1081,6 +1093,8 @@ int v3_vmx_enter(struct guest_info * info) {
 
     info->num_exits++;
 
+    V3_FP_EXIT_SAVE(info);
+
     /* If we have the preemption time, then use it to get more accurate guest time */
     if (vmx_info->pin_ctrls.active_preempt_timer) {
        uint32_t cycles_left = 0;
@@ -1172,6 +1186,10 @@ int v3_vmx_enter(struct guest_info * info) {
        v3_handle_timeouts(info, guest_cycles);
     }
 
+#ifdef V3_CONFIG_MEM_TRACK
+    v3_mem_track_exit(info);
+#endif 
+
     return 0;
 }
 
@@ -1180,6 +1198,13 @@ int v3_start_vmx_guest(struct guest_info * info) {
 
     PrintDebug(info->vm_info, info, "Starting VMX core %u\n", info->vcpu_id);
 
+#if V3_CONFIG_HVM
+    if (v3_setup_hvm_vm_for_boot(vm)) { 
+       PrintError(vm, VCORE_NONE, "HVM setup for boot failed\n");
+       return -1;
+    }
+#endif
+    
     while (1) {
        if (info->core_run_state == CORE_STOPPED) {
            if (info->vcpu_id == 0) {
@@ -1187,6 +1212,8 @@ int v3_start_vmx_guest(struct guest_info * info) {
            } else {
                
                PrintDebug(info->vm_info, info, "VMX core %u: Waiting for core initialization\n", info->vcpu_id);
+
+                V3_NO_WORK(info);
                
                while (info->core_run_state == CORE_STOPPED) {
                    
@@ -1194,11 +1221,13 @@ int v3_start_vmx_guest(struct guest_info * info) {
                        // The VM was stopped before this core was initialized. 
                        return 0;
                    }
-                   
-                   v3_yield(info,-1);
+
+                   V3_STILL_NO_WORK(info);
                    //PrintDebug(info->vm_info, info, "VMX core %u: still waiting for INIT\n",info->vcpu_id);
                }
-               
+
+               V3_HAVE_WORK_AGAIN(info);
+
                PrintDebug(info->vm_info, info, "VMX core %u initialized\n", info->vcpu_id);
                
                // We'll be paranoid about race conditions here