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.


Basic HRT startup for HVM, plus assorted cleanup
[palacios.git] / palacios / src / palacios / svm.c
index b54ab19..05fd183 100644 (file)
@@ -39,6 +39,7 @@
 #include <palacios/vmm_barrier.h>
 #include <palacios/vmm_debug.h>
 
+#include <palacios/vmm_perftune.h>
 
 
 #ifdef V3_CONFIG_CHECKPOINT
 
 #include <palacios/vmm_sprintf.h>
 
+#ifdef V3_CONFIG_MEM_TRACK
+#include <palacios/vmm_mem_track.h>
+#endif 
+
+#ifdef V3_CONFIG_TM_FUNC
+#include <extensions/trans_mem.h>
+#endif
 
 #ifndef V3_CONFIG_DEBUG_SVM
 #undef PrintDebug
@@ -59,6 +67,7 @@
 #endif
 
 
+
 uint32_t v3_last_exit;
 
 // This is a global pointer to the host's VMCB
@@ -74,7 +83,7 @@ extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_
 
 static vmcb_t * Allocate_VMCB() {
     vmcb_t * vmcb_page = NULL;
-    addr_t vmcb_pa = (addr_t)V3_AllocPages(1);
+    addr_t vmcb_pa = (addr_t)V3_AllocPages(1);   // need not be shadow safe, not exposed to guest
 
     if ((void *)vmcb_pa == NULL) {
       PrintError(VM_NONE, VCORE_NONE, "Error allocating VMCB\n");
@@ -154,6 +163,10 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
       
       ctrl_area->exceptions.nmi = 1;
     */
+
+#ifdef V3_CONFIG_TM_FUNC
+    v3_tm_set_excp_intercepts(ctrl_area);
+#endif
     
 
     ctrl_area->instrs.NMI = 1;
@@ -281,11 +294,13 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
        PrintDebug(core->vm_info, core, "Created\n");
        
        core->ctrl_regs.cr0 |= 0x80000000;
-       core->ctrl_regs.cr3 = core->direct_map_pt;
+
+        v3_activate_passthrough_pt(core);
 
        ctrl_area->cr_reads.cr0 = 1;
        ctrl_area->cr_writes.cr0 = 1;
-       //ctrl_area->cr_reads.cr4 = 1;
+       //intercept cr4 read so shadow pager can use PAE independently of guest
+       ctrl_area->cr_reads.cr4 = 1;
        ctrl_area->cr_writes.cr4 = 1;
        ctrl_area->cr_reads.cr3 = 1;
        ctrl_area->cr_writes.cr3 = 1;
@@ -527,7 +542,11 @@ static int update_irq_entry_state(struct guest_info * info) {
     } else {
        switch (v3_intr_pending(info)) {
            case V3_EXTERNAL_IRQ: {
-               uint32_t irq = v3_get_intr(info);
+               int irq = v3_get_intr(info); 
+
+               if (irq<0) {
+                 break;
+               }
 
                guest_ctrl->guest_ctrl.V_IRQ = 1;
                guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
@@ -618,12 +637,17 @@ int v3_svm_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 after being in the VM before doing 
     // IRQ updates, so that any interrupts they raise get seen 
     // immediately.
     v3_advance_time(info, NULL);
     v3_update_timers(info);
 
+
     // disable global interrupts for vm state transition
     v3_clgi();
 
@@ -662,6 +686,8 @@ int v3_svm_enter(struct guest_info * info) {
     guest_state->rip = info->rip;
     guest_state->rsp = info->vm_regs.rsp;
 
+    V3_FP_ENTRY_RESTORE(info);
+
 #ifdef V3_CONFIG_SYMCALL
     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
        update_irq_entry_state(info);
@@ -670,6 +696,10 @@ int v3_svm_enter(struct guest_info * info) {
     update_irq_entry_state(info);
 #endif
 
+#ifdef V3_CONFIG_TM_FUNC
+    v3_tm_check_intr_state(info, guest_ctrl, guest_state);
+#endif
+
 
     /* ** */
 
@@ -694,9 +724,15 @@ int v3_svm_enter(struct guest_info * info) {
        uint64_t entry_tsc = 0;
        uint64_t exit_tsc = 0;
        
+#ifdef V3_CONFIG_PWRSTAT_TELEMETRY
+       v3_pwrstat_telemetry_enter(info);
+#endif
+
 #ifdef V3_CONFIG_PMU_TELEMETRY
        v3_pmu_telemetry_enter(info);
 #endif
+
+
        rdtscll(entry_tsc);
 
        v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
@@ -706,6 +742,11 @@ int v3_svm_enter(struct guest_info * info) {
 #ifdef V3_CONFIG_PMU_TELEMETRY
        v3_pmu_telemetry_exit(info);
 #endif
+
+#ifdef V3_CONFIG_PWRSTAT_TELEMETRY
+       v3_pwrstat_telemetry_exit(info);
+#endif
+
        guest_cycles = exit_tsc - entry_tsc;
     }
 
@@ -718,6 +759,8 @@ int v3_svm_enter(struct guest_info * info) {
 
     info->num_exits++;
 
+    V3_FP_EXIT_SAVE(info);
+
     // Save Guest state from VMCB
     info->rip = guest_state->rip;
     info->vm_regs.rsp = guest_state->rsp;
@@ -765,6 +808,7 @@ int v3_svm_enter(struct guest_info * info) {
 
     // reenable global interrupts after vm exit
     v3_stgi();
+
  
     // Conditionally yield the CPU if the timeslice has expired
     v3_schedule(info);
@@ -774,6 +818,7 @@ int v3_svm_enter(struct guest_info * info) {
     v3_advance_time(info, NULL);
     v3_update_timers(info);
 
+
     {
        int ret = v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2);
        
@@ -789,6 +834,10 @@ int v3_svm_enter(struct guest_info * info) {
        v3_handle_timeouts(info, guest_cycles);
     }
 
+#ifdef V3_CONFIG_MEM_TRACK
+    v3_mem_track_exit(info);
+#endif 
+
 
     return 0;
 }
@@ -800,51 +849,75 @@ int v3_start_svm_guest(struct guest_info * info) {
 
     PrintDebug(info->vm_info, info, "Starting SVM core %u (on logical core %u)\n", info->vcpu_id, info->pcpu_id);
 
-    if (info->vcpu_id == 0) {
-       info->core_run_state = CORE_RUNNING;
-    } else  { 
-       PrintDebug(info->vm_info, info, "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,-1);
-           //PrintDebug(info->vm_info, info, "SVM core %u: still waiting for INIT\n", info->vcpu_id);
-       }
+#ifdef V3_CONFIG_HVM
+    if (v3_setup_hvm_hrt_core_for_boot(info)) { 
+       PrintError(info->vm_info, info, "Failed to setup HRT core...\n");
+       return -1;
+    }
+#endif
 
-       PrintDebug(info->vm_info, info, "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(info->vm_info, info, "SVM core %u(on %u): I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n", 
-              info->vcpu_id, info->pcpu_id, 
-              info->segments.cs.selector, (void *)(info->segments.cs.base), 
-              info->segments.cs.limit, (void *)(info->rip));
+    while (1) {
 
+       if (info->core_run_state == CORE_STOPPED) {
 
+           if (info->vcpu_id == 0) {
+               info->core_run_state = CORE_RUNNING;
+           } else  { 
+               PrintDebug(info->vm_info, info, "SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id);
 
-    PrintDebug(info->vm_info, info, "SVM core %u: Launching SVM VM (vmcb=%p) (on cpu %u)\n", 
-              info->vcpu_id, (void *)info->vmm_data, info->pcpu_id);
-    //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
-    
-    v3_start_time(info);
+               V3_NO_WORK(info);
 
-#ifdef V3_CONFIG_PMU_TELEMETRY
-    v3_pmu_telemetry_start(info);
-#endif
+               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_STILL_NO_WORK(info);
 
-    while (1) {
+                   //PrintDebug(info->vm_info, info, "SVM core %u: still waiting for INIT\n", info->vcpu_id);
+               }
 
+               V3_HAVE_WORK_AGAIN(info);
+               
+               PrintDebug(info->vm_info, info, "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(info->vm_info, info, "SVM core %u(on %u): I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n", 
+                      info->vcpu_id, info->pcpu_id, 
+                      info->segments.cs.selector, (void *)(info->segments.cs.base), 
+                      info->segments.cs.limit, (void *)(info->rip));
+           
+           
+           
+           PrintDebug(info->vm_info, info, "SVM core %u: Launching SVM VM (vmcb=%p) (on cpu %u)\n", 
+                      info->vcpu_id, (void *)info->vmm_data, info->pcpu_id);
+           //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
+           
+           v3_start_time(info);
+       }
+       
        if (info->vm_info->run_state == VM_STOPPED) {
            info->core_run_state = CORE_STOPPED;
            break;
        }
+
+#ifdef V3_CONFIG_PMU_TELEMETRY
+       v3_pmu_telemetry_start(info);
+#endif
+       
+#ifdef V3_CONFIG_PWRSTAT_TELEMETRY
+       v3_pwrstat_telemetry_start(info);
+#endif
        
        if (v3_svm_enter(info) == -1) {
            vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
@@ -879,12 +952,12 @@ int v3_start_svm_guest(struct guest_info * info) {
            v3_dump_mem((uint8_t *)host_addr, 15);
            
            v3_print_stack(info);
-
+           
            break;
        }
-
+       
        v3_wait_at_barrier(info);
-
+       
 
        if (info->vm_info->run_state == VM_STOPPED) {
            info->core_run_state = CORE_STOPPED;
@@ -905,6 +978,10 @@ int v3_start_svm_guest(struct guest_info * info) {
 #ifdef V3_CONFIG_PMU_TELEMETRY
     v3_pmu_telemetry_end(info);
 #endif
+
+#ifdef V3_CONFIG_PWRSTAT_TELEMETRY
+    v3_pwrstat_telemetry_end(info);
+#endif
     // Need to take down the other cores on error... 
 
     return 0;
@@ -1015,7 +1092,7 @@ void v3_init_svm_cpu(int cpu_id) {
     V3_Print(VM_NONE, VCORE_NONE,  "SVM Enabled\n");
 
     // Setup the host state save area
-    host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4);
+    host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4); // need not be shadow-safe, not exposed to guest
 
     if (!host_vmcbs[cpu_id]) {
        PrintError(VM_NONE, VCORE_NONE,  "Failed to allocate VMCB\n");
@@ -1260,7 +1337,7 @@ void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) {
   
 
   ctrl_area->instrs.IOIO_PROT = 1;
-  ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3);
+  ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3); // need not be shadow-safe, not exposed to guest
 
   if (!ctrl_area->IOPM_BASE_PA) { 
       PrintError(core->vm_info, core, "Cannot allocate IO bitmap\n");