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.


Cleaned up configuration of time management; most advanced time features are
[palacios.git] / palacios / src / palacios / vmx.c
index f26a1ea..ca875bb 100644 (file)
 #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>
@@ -130,53 +136,6 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state)
     /******* Setup Host State **********/
 
     /* Cache GDTR, IDTR, and TR in host struct */
-    addr_t gdtr_base;
-    struct {
-        uint16_t selector;
-        addr_t   base;
-    } __attribute__((packed)) tmp_seg;
-    
-
-    __asm__ __volatile__(
-                        "sgdt (%0);"
-                        :
-                        : "q"(&tmp_seg)
-                        : "memory"
-                        );
-    gdtr_base = tmp_seg.base;
-    vmx_state->host_state.gdtr.base = gdtr_base;
-
-    __asm__ __volatile__(
-                        "sidt (%0);"
-                        :
-                        : "q"(&tmp_seg)
-                        : "memory"
-                        );
-    vmx_state->host_state.idtr.base = tmp_seg.base;
-
-    __asm__ __volatile__(
-                        "str (%0);"
-                        :
-                        : "q"(&tmp_seg)
-                        : "memory"
-                        );
-    vmx_state->host_state.tr.selector = tmp_seg.selector;
-
-    /* The GDTR *index* is bits 3-15 of the selector. */
-    struct tss_descriptor * desc = NULL;
-    desc = (struct tss_descriptor *)(gdtr_base + (8 * (tmp_seg.selector >> 3)));
-
-    tmp_seg.base = ((desc->base1) |
-                   (desc->base2 << 16) |
-                   (desc->base3 << 24) |
-#ifdef __V3_64BIT__
-                   ((uint64_t)desc->base4 << 32)
-#else 
-                   (0)
-#endif
-                   );
-
-    vmx_state->host_state.tr.base = tmp_seg.base;
 
 
     /********** Setup VMX Control Fields ***********/
@@ -207,28 +166,26 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state)
 
 
 
-    
-
-
-
 #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 */
@@ -547,7 +504,7 @@ int v3_deinit_vmx_vmcs(struct guest_info * core) {
     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);
 
@@ -555,6 +512,63 @@ int v3_deinit_vmx_vmcs(struct guest_info * core) {
 }
 
 
+
+#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
+
+
+void v3_flush_vmx_vm_core(struct guest_info * core) {
+    struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
+    vmcs_clear(vmx_info->vmcs_ptr_phys);
+    vmx_info->state = VMX_UNLAUNCHED;
+}
+
+
+
 static int update_irq_exit_state(struct guest_info * info) {
     struct vmx_exit_idt_vec_info idt_vec_info;
 
@@ -749,13 +763,16 @@ int v3_vmx_enter(struct guest_info * info) {
     // disable global interrupts for vm state transition
     v3_disable_ints();
 
-    // Update timer devices prior to entering VM.  Doing it here 
-    // makes sure the guest sees any timers that fired while it 
-    // was in the VMM
+    // 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 (vmcs_store() != vmx_info->vmcs_ptr_phys) {
+       vmcs_clear(vmx_info->vmcs_ptr_phys);
        vmcs_load(vmx_info->vmcs_ptr_phys);
+       vmx_info->state = VMX_UNLAUNCHED;
     }
 
     v3_vmx_restore_vmcs(info);
@@ -792,6 +809,7 @@ int v3_vmx_enter(struct guest_info * info) {
 
     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 {
@@ -799,19 +817,22 @@ int v3_vmx_enter(struct guest_info * info) {
        ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs));
     }
     
+
+
     //  PrintDebug("VMX Exit: ret=%d\n", ret);
 
     if (ret != VMX_SUCCESS) {
        uint32_t error = 0;
-
         vmcs_read(VMCS_INSTR_ERR, &error);
 
        v3_enable_ints();
 
-        PrintError("VMENTRY Error: %d\n", error);
+       PrintError("VMENTRY Error: %d (launch_ret = %d)\n", error, ret);
        return -1;
     }
 
+
+
     // Immediate exit from VM time bookkeeping
     v3_time_exit_vm(info);
 
@@ -842,7 +863,6 @@ int v3_vmx_enter(struct guest_info * info) {
 
     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);
@@ -869,7 +889,7 @@ int v3_vmx_enter(struct guest_info * 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;
     }
 
@@ -894,6 +914,9 @@ int v3_start_vmx_guest(struct guest_info * info) {
         }
        
        PrintDebug("VMX core %u initialized\n", info->vcpu_id);
+
+       // We'll be paranoid about race conditions here
+       v3_wait_at_barrier(info);
     }
 
 
@@ -914,11 +937,40 @@ int v3_start_vmx_guest(struct guest_info * 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) {
@@ -991,6 +1043,7 @@ int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) {
 
 
 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) {
@@ -1003,17 +1056,18 @@ void v3_init_vmx_cpu(int cpu_id) {
 
 
     // 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;
@@ -1036,5 +1090,16 @@ void v3_init_vmx_cpu(int cpu_id) {
 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;
+    }
 }