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.


enabled stopping a VM before the secondary cores have been initialized
[palacios.git] / palacios / src / palacios / svm.c
index 4b5b0a9..f61fadf 100644 (file)
 #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>
 
@@ -80,6 +85,25 @@ static vmcb_t * Allocate_VMCB() {
 }
 
 
+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);
@@ -220,7 +244,7 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
 
     v3_hook_msr(core->vm_info, EFER_MSR, 
                &v3_handle_efer_read,
-               &v3_handle_efer_write, 
+               &v3_svm_handle_efer_write, 
                core);
 
     if (core->shdw_pg_mode == SHADOW_PAGING) {
@@ -287,6 +311,19 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
        &v3_handle_vm_cr_read,
        &v3_handle_vm_cr_write, 
        core);
+
+
+    {
+       v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL);
+       v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL);
+       v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL);
+       v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL);
+       v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL);
+
+       v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL);
+       v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL);
+       v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL);
+    }
 }
 
 
@@ -318,6 +355,27 @@ int v3_deinit_svm_vmcb(struct guest_info * core) {
 }
 
 
+#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));
 
@@ -424,6 +482,18 @@ static int update_irq_entry_state(struct guest_info * info) {
                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;
@@ -452,6 +522,7 @@ int v3_svm_enter(struct guest_info * info) {
     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);
@@ -462,7 +533,9 @@ int v3_svm_enter(struct guest_info * info) {
     // disable global interrupts for vm state transition
     v3_clgi();
 
-    // Update timer devices prior to entering VM.
+    // 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
@@ -510,7 +583,9 @@ int v3_svm_enter(struct guest_info * info) {
 #endif
 
     v3_time_enter_vm(info);
-    guest_ctrl->TSC_OFFSET = v3_tsc_host_offset(&info->time_state);
+    tsc_offset = v3_tsc_host_offset(&info->time_state);
+    guest_ctrl->TSC_OFFSET = tsc_offset;
+
 
     //V3_Print("Calling v3_svm_launch\n");
 
@@ -547,13 +622,11 @@ int v3_svm_enter(struct guest_info * info) {
     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 V3_CONFIG_SYMCALL
     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
        update_irq_exit_state(info);
@@ -562,20 +635,20 @@ int v3_svm_enter(struct guest_info * 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");
-       PrintError("  last exit was %d\n", v3_last_exit);
-       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;
+       }
     }
 
 
@@ -596,11 +669,20 @@ int v3_start_svm_guest(struct guest_info * info) {
        PrintDebug("SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id);
 
        while (info->core_run_state == CORE_STOPPED) {
+           
+           if (info->vm_info->run_state == VM_STOPPED) {
+               // The VM was stopped before this core was initialized. 
+               return 0;
+           }
+
            v3_yield(info);
            //PrintDebug("SVM core %u: still waiting for INIT\n", info->vcpu_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(on %u): I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n", 
@@ -660,6 +742,8 @@ int v3_start_svm_guest(struct guest_info * info) {
            break;
        }
 
+       v3_wait_at_barrier(info);
+
 
        if (info->vm_info->run_state == VM_STOPPED) {
            info->core_run_state = CORE_STOPPED;
@@ -668,11 +752,12 @@ int v3_start_svm_guest(struct guest_info * info) {
 
        
 
-       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);
        }
-
+*/
        
     }
 
@@ -684,6 +769,31 @@ int v3_start_svm_guest(struct guest_info * 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 */ 
@@ -733,11 +843,11 @@ int v3_is_svm_capable() {
 
 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;
@@ -745,7 +855,8 @@ static int has_svm_nested_paging() {
        V3_Print("SVM Nested Paging supported\n");
        return 1;
     }
-}
+ }
 
 
 void v3_init_svm_cpu(int cpu_id) {