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.


updated NMI exit handler
[palacios.git] / palacios / src / palacios / vmx_handler.c
index ee1824f..be86fe2 100644 (file)
 #include <palacios/vmx_ctrl_regs.h>
 #include <palacios/vmx_assist.h>
 #include <palacios/vmm_halt.h>
+#include <palacios/vmx_ept.h>
 
-#ifdef CONFIG_TELEMETRY
-#include <palacios/vmm_telemetry.h>
+#ifndef V3_CONFIG_DEBUG_VMX
+#undef PrintDebug
+#define PrintDebug(fmt, args...)
 #endif
 
-
-static int inline check_vmcs_write(vmcs_field_t field, addr_t val) {
-    int ret = 0;
-
-    ret = vmcs_write(field, val);
-
-    if (ret != VMX_SUCCESS) {
-        PrintError("VMWRITE error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
-    }
-
-    return ret;
-}
-
-static int inline check_vmcs_read(vmcs_field_t field, void * val) {
-    int ret = 0;
-
-    ret = vmcs_read(field, val);
-
-    if (ret != VMX_SUCCESS) {
-        PrintError("VMREAD error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
-    }
-
-    return ret;
-}
-
-static int inline handle_cr_access(struct guest_info * info, ulong_t exit_qual) {
-    struct vmx_exit_cr_qual * cr_qual = (struct vmx_exit_cr_qual *)&exit_qual;
-
-    // PrintDebug("Control register: %d\n", cr_qual->access_type);
-    switch(cr_qual->cr_id) {
-        case 0:
-           PrintDebug("Handling CR0 Access\n");
-            return v3_vmx_handle_cr0_access(info);
-        case 3:
-           PrintDebug("Handling CR3 Access\n");
-            return v3_vmx_handle_cr3_access(info);
-        default:
-            PrintError("Unhandled CR access: %d\n", cr_qual->cr_id);
-            return -1;
-    }
-    
-    return -1;
-}
-
+#ifdef V3_CONFIG_TELEMETRY
+#include <palacios/vmm_telemetry.h>
+#endif
 
 /* At this point the GPRs are already copied into the guest_info state */
-int v3_handle_vmx_exit(struct v3_gprs * gprs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs) {
-    uint64_t tmp_tsc = 0;
-    uint32_t exit_reason = 0;
-    addr_t exit_qual = 0;
-    struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
-    struct vmx_exit_idt_vec_info idt_vec_info;
-
-    rdtscll(tmp_tsc);
-    v3_update_time(info, tmp_tsc - info->time_state.cached_host_tsc);
-
-    v3_enable_ints();
-
-    check_vmcs_read(VMCS_EXIT_REASON, &exit_reason);
-    check_vmcs_read(VMCS_EXIT_QUAL, &exit_qual);
-
-    //PrintDebug("VMX Exit taken, id-qual: %u-%lu\n", exit_reason, exit_qual);
-
-    /* Update guest state */
-    v3_load_vmcs_guest_state(info);
-
-    // Load execution controls
-    check_vmcs_read(VMCS_PIN_CTRLS, &(vmx_info->pin_ctrls.value));
-    check_vmcs_read(VMCS_PROC_CTRLS, &(vmx_info->pri_proc_ctrls.value));
-
-    if (vmx_info->pri_proc_ctrls.sec_ctrls) {
-        check_vmcs_read(VMCS_SEC_PROC_CTRLS, &(vmx_info->sec_proc_ctrls.value));
-    }
-
-    info->mem_mode = v3_get_vm_mem_mode(info);
-    info->cpu_mode = v3_get_vm_cpu_mode(info);
-
-    // Check if we got interrupted while delivering interrupt
-    // Variable will be used later if this is true
-
-    check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
-
-    if ((info->intr_state.irq_started == 1) && (idt_vec_info.valid == 0)) {
-#ifdef CONFIG_DEBUG_INTERRUPTS
-        PrintDebug("Calling v3_injecting_intr\n");
-#endif
-        info->intr_state.irq_started = 0;
-        v3_injecting_intr(info, info->intr_state.irq_vector, V3_EXTERNAL_IRQ);
+int v3_handle_vmx_exit(struct guest_info * info, struct vmx_exit_info * exit_info) {
+    struct vmx_basic_exit_info * basic_info = (struct vmx_basic_exit_info *)&(exit_info->exit_reason);
+
+    /*
+      PrintError("Handling VMEXIT: %s (%u), %lu (0x%lx)\n", 
+      v3_vmx_exit_code_to_str(exit_info->exit_reason),
+      exit_info->exit_reason, 
+      exit_info->exit_qual, exit_info->exit_qual);
+      
+      v3_print_vmcs();
+    */
+
+
+    if (basic_info->entry_error == 1) {
+       switch (basic_info->reason) {
+           case VMEXIT_INVALID_GUEST_STATE:
+               PrintError("VM Entry failed due to invalid guest state\n");
+               PrintError("Printing VMCS: (NOTE: This VMCS may not belong to the correct guest)\n");
+               v3_print_vmcs();
+               break;
+           case VMEXIT_INVALID_MSR_LOAD:
+               PrintError("VM Entry failed due to error loading MSRs\n");
+               break;
+           default:
+               PrintError("Entry failed for unknown reason (%d)\n", basic_info->reason);
+               break;
+       }
+       
+       return -1;
     }
 
-    info->num_exits++;
-
 
-
-    if ((info->num_exits % 5000) == 0) {
-       PrintDebug("VMX Exit %d\n", (uint32_t)info->num_exits);
-    }
-
-#ifdef CONFIG_TELEMETRY
-    if (info->enable_telemetry) {
+#ifdef V3_CONFIG_TELEMETRY
+    if (info->vm_info->enable_telemetry) {
        v3_telemetry_start_exit(info);
     }
 #endif
 
-    switch (exit_reason) {
+    switch (basic_info->reason) {
         case VMEXIT_INFO_EXCEPTION_OR_NMI: {
-            uint32_t int_info;
-            pf_error_t error_code;
+            pf_error_t error_code = *(pf_error_t *)&(exit_info->int_err);
 
-            check_vmcs_read(VMCS_EXIT_INT_INFO, &int_info);
-            check_vmcs_read(VMCS_EXIT_INT_ERR, &error_code);
 
             // JRL: Change "0x0e" to a macro value
-            if ((uint8_t)int_info == 0x0e) {
-#ifdef CONFIG_DEBUG_SHADOW_PAGING
-                PrintDebug("Page Fault at %p error_code=%x\n", (void *)exit_qual, *(uint32_t *)&error_code);
+            if ((uint8_t)exit_info->int_info == 14) {
+#ifdef V3_CONFIG_DEBUG_SHADOW_PAGING
+                PrintDebug("Page Fault at %p error_code=%x\n", (void *)exit_info->exit_qual, *(uint32_t *)&error_code);
 #endif
 
                 if (info->shdw_pg_mode == SHADOW_PAGING) {
-                    if (v3_handle_shadow_pagefault(info, (addr_t)exit_qual, error_code) == -1) {
+                    if (v3_handle_shadow_pagefault(info, (addr_t)exit_info->exit_qual, error_code) == -1) {
                         PrintError("Error handling shadow page fault\n");
                         return -1;
                     }
+           
                 } else {
                     PrintError("Page fault in unimplemented paging mode\n");
                     return -1;
                 }
+           } else if ((uint8_t)exit_info->int_info == 2) {
+               // NMI. Don't do anything
+               V3_Print("NMI Exception Received\n");
             } else {
-                PrintError("Unknown exception: 0x%x\n", (uint8_t)int_info);
+                PrintError("Unknown exception: 0x%x\n", (uint8_t)exit_info->int_info);
                 v3_print_GPRs(info);
                 return -1;
             }
             break;
         }
 
+       case VMEXIT_EPT_VIOLATION: {
+           struct ept_exit_qual * ept_qual = (struct ept_exit_qual *)&(exit_info->exit_qual);
+
+           if (v3_handle_ept_fault(info, exit_info->ept_fault_addr, ept_qual) == -1) {
+               PrintError("Error handling EPT fault\n");
+               return -1;
+           }
+
+           break;
+       }
         case VMEXIT_INVLPG:
             if (info->shdw_pg_mode == SHADOW_PAGING) {
                 if (v3_handle_shadow_invlpg(info) == -1) {
@@ -180,6 +132,18 @@ int v3_handle_vmx_exit(struct v3_gprs * gprs, struct guest_info * info, struct v
             }
 
             break;
+
+        case VMEXIT_RDTSC:
+#ifdef V3_CONFIG_DEBUG_TIME
+           PrintDebug("RDTSC\n");
+#endif 
+           if (v3_handle_rdtsc(info) == -1) {
+               PrintError("Error Handling RDTSC instruction\n");
+               return -1;
+           }
+           
+           break;
+
         case VMEXIT_CPUID:
            if (v3_handle_cpuid(info) == -1) {
                PrintError("Error Handling CPUID instruction\n");
@@ -201,29 +165,42 @@ int v3_handle_vmx_exit(struct v3_gprs * gprs, struct guest_info * info, struct v
            }
 
             break;
+       case VMEXIT_VMCALL:
+           /* 
+            * Hypercall 
+            */
+
+           // VMCALL is a 3 byte op
+           // We do this early because some hypercalls can change the rip...
+           info->rip += 3;         
+
+           if (v3_handle_hypercall(info) == -1) {
+               return -1;
+           }
+           break;
         case VMEXIT_IO_INSTR: {
-           struct vmx_exit_io_qual * io_qual = (struct vmx_exit_io_qual *)&exit_qual;
+           struct vmx_exit_io_qual * io_qual = (struct vmx_exit_io_qual *)&(exit_info->exit_qual);
 
             if (io_qual->dir == 0) {
                 if (io_qual->string) {
-                    if (v3_handle_vmx_io_outs(info) == -1) {
+                    if (v3_handle_vmx_io_outs(info, exit_info) == -1) {
                         PrintError("Error in outs IO handler\n");
                         return -1;
                     }
                 } else {
-                    if (v3_handle_vmx_io_out(info) == -1) {
+                    if (v3_handle_vmx_io_out(info, exit_info) == -1) {
                         PrintError("Error in out IO handler\n");
                         return -1;
                     }
                 }
             } else {
                 if (io_qual->string) {
-                    if(v3_handle_vmx_io_ins(info) == -1) {
+                    if(v3_handle_vmx_io_ins(info, exit_info) == -1) {
                         PrintError("Error in ins IO handler\n");
                         return -1;
                     }
                 } else {
-                    if (v3_handle_vmx_io_in(info) == -1) {
+                    if (v3_handle_vmx_io_in(info, exit_info) == -1) {
                         PrintError("Error in in IO handler\n");
                         return -1;
                     }
@@ -231,13 +208,44 @@ int v3_handle_vmx_exit(struct v3_gprs * gprs, struct guest_info * info, struct v
             }
             break;
        }
-        case VMEXIT_CR_REG_ACCESSES:
-            if (handle_cr_access(info, exit_qual) != 0) {
-                PrintError("Error handling CR access\n");
-                return -1;
-            }
+        case VMEXIT_CR_REG_ACCESSES: {
+           struct vmx_exit_cr_qual * cr_qual = (struct vmx_exit_cr_qual *)&(exit_info->exit_qual);
+           
+           // PrintDebug("Control register: %d\n", cr_qual->access_type);
+           switch(cr_qual->cr_id) {
+               case 0:
+                   //PrintDebug("Handling CR0 Access\n");
+                   if (v3_vmx_handle_cr0_access(info, cr_qual, exit_info) == -1) {
+                       PrintError("Error in CR0 access handler\n");
+                       return -1;
+                   }
+                   break;
+               case 3:
+                   //PrintDebug("Handling CR3 Access\n");
+                   if (v3_vmx_handle_cr3_access(info, cr_qual) == -1) {
+                       PrintError("Error in CR3 access handler\n");
+                       return -1;
+                   }
+                   break;
+               case 4:
+                   //PrintDebug("Handling CR4 Access\n");
+                   if (v3_vmx_handle_cr4_access(info, cr_qual) == -1) {
+                       PrintError("Error in CR4 access handler\n");
+                       return -1;
+                   }
+                   break;
+               default:
+                   PrintError("Unhandled CR access: %d\n", cr_qual->cr_id);
+                   return -1;
+           }
+           
+           // TODO: move RIP increment into all of the above individual CR
+           //       handlers, not just v3_vmx_handle_cr4_access()
+           if (cr_qual->cr_id != 4)
+               info->rip += exit_info->instr_len;
 
-            break;
+           break;
+       }
         case VMEXIT_HLT:
             PrintDebug("Guest halted\n");
 
@@ -247,6 +255,9 @@ int v3_handle_vmx_exit(struct v3_gprs * gprs, struct guest_info * info, struct v
             }
 
             break;
+
+
+
         case VMEXIT_PAUSE:
             // Handled as NOP
             info->rip += 2;
@@ -256,148 +267,26 @@ int v3_handle_vmx_exit(struct v3_gprs * gprs, struct guest_info * info, struct v
             // Interrupts are handled outside switch
             break;
         case VMEXIT_INTR_WINDOW:
-
-            vmx_info->pri_proc_ctrls.int_wndw_exit = 0;
-            check_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
-
+           // This is handled in the atomic part of the vmx code,
+           // not in the generic (interruptable) vmx handler
             break;
+
+           
         default:
             PrintError("Unhandled VMEXIT: %s (%u), %lu (0x%lx)\n", 
-                      v3_vmx_exit_code_to_str(exit_reason),
-                      exit_reason, exit_qual, exit_qual);
+                      v3_vmx_exit_code_to_str(basic_info->reason),
+                      basic_info->reason, 
+                      exit_info->exit_qual, exit_info->exit_qual);
             return -1;
     }
 
-#ifdef CONFIG_TELEMETRY
-    if (info->enable_telemetry) {
-        v3_telemetry_end_exit(info, exit_reason);
-    }
-#endif
-
-
-    /* Check for pending exceptions to inject */
-    if (v3_excp_pending(info)) {
-        struct vmx_entry_int_info int_info;
-        int_info.value = 0;
-
-        // In VMX, almost every exception is hardware
-        // Software exceptions are pretty much only for breakpoint or overflow
-        int_info.type = 3;
-        int_info.vector = v3_get_excp_number(info);
-
-        if (info->excp_state.excp_error_code_valid) {
-            check_vmcs_write(VMCS_ENTRY_EXCP_ERR, info->excp_state.excp_error_code);
-            int_info.error_code = 1;
-
-            PrintDebug("Injecting exception %d with error code %x\n", 
-                    int_info.vector, info->excp_state.excp_error_code);
-        }
-
-        int_info.valid = 1;
-        PrintDebug("Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)info->rip);
-        check_vmcs_write(VMCS_ENTRY_INT_INFO, int_info.value);
-
-        v3_injecting_excp(info, int_info.vector);
-
-    } else if (((struct rflags *)&(info->ctrl_regs.rflags))->intr == 1) {
-       
-        if ((info->intr_state.irq_started == 1) && (idt_vec_info.valid == 1)) {
 
-#ifdef CONFIG_DEBUG_INTERRUPTS
-            PrintDebug("IRQ pending from previous injection\n");
-#endif
-
-            // Copy the IDT vectoring info over to reinject the old interrupt
-            if (idt_vec_info.error_code == 1) {
-                uint32_t err_code = 0;
-
-                check_vmcs_read(VMCS_IDT_VECTOR_ERR, &err_code);
-                check_vmcs_write(VMCS_ENTRY_EXCP_ERR, err_code);
-            }
-
-            idt_vec_info.undef = 0;
-            check_vmcs_write(VMCS_ENTRY_INT_INFO, idt_vec_info.value);
-
-        } else {
-            struct vmx_entry_int_info ent_int;
-            ent_int.value = 0;
-
-            switch (v3_intr_pending(info)) {
-                case V3_EXTERNAL_IRQ: {
-                    info->intr_state.irq_vector = v3_get_intr(info); 
-                    ent_int.vector = info->intr_state.irq_vector;
-                    ent_int.type = 0;
-                    ent_int.error_code = 0;
-                    ent_int.valid = 1;
-
-#ifdef CONFIG_DEBUG_INTERRUPTS
-                    PrintDebug("Injecting Interrupt %d at exit %u(EIP=%p)\n", 
-                              info->intr_state.irq_vector, 
-                              (uint32_t)info->num_exits, 
-                              (void *)info->rip);
-#endif
-
-                    check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
-                    info->intr_state.irq_started = 1;
-
-                    break;
-                }
-                case V3_NMI:
-                    PrintDebug("Injecting NMI\n");
-
-                    ent_int.type = 2;
-                    ent_int.vector = 2;
-                    ent_int.valid = 1;
-                    check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
-
-                    break;
-                case V3_SOFTWARE_INTR:
-                    PrintDebug("Injecting software interrupt\n");
-                    ent_int.type = 4;
-
-                    ent_int.valid = 1;
-                    check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
-
-                   break;
-                case V3_VIRTUAL_IRQ:
-                    // Not sure what to do here, Intel doesn't have virtual IRQs
-                    // May be the same as external interrupts/IRQs
-
-                   break;
-                case V3_INVALID_INTR:
-                default:
-                    break;
-            }
-        }
-    } else if ((v3_intr_pending(info)) && (vmx_info->pri_proc_ctrls.int_wndw_exit == 0)) {
-        // Enable INTR window exiting so we know when IF=1
-        uint32_t instr_len;
-
-        check_vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
-
-#ifdef CONFIG_DEBUG_INTERRUPTS
-        PrintDebug("Enabling Interrupt-Window exiting: %d\n", instr_len);
-#endif
-
-        vmx_info->pri_proc_ctrls.int_wndw_exit = 1;
-        check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
+#ifdef V3_CONFIG_TELEMETRY
+    if (info->vm_info->enable_telemetry) {
+        v3_telemetry_end_exit(info, exit_info->exit_reason);
     }
+#endif
 
-    check_vmcs_write(VMCS_GUEST_CR0, info->ctrl_regs.cr0);
-    check_vmcs_write(VMCS_GUEST_CR3, info->ctrl_regs.cr3);
-    check_vmcs_write(VMCS_GUEST_CR4, info->ctrl_regs.cr4);
-    check_vmcs_write(VMCS_GUEST_RIP, info->rip);
-    check_vmcs_write(VMCS_GUEST_RSP, info->vm_regs.rsp);
-
-    check_vmcs_write(VMCS_CR0_READ_SHDW, info->shdw_pg_state.guest_cr0);
-
-    v3_disable_ints();
-
-    rdtscll(info->time_state.cached_host_tsc);
 
     return 0;
 }
@@ -434,12 +323,12 @@ static const char VMEXIT_MOV_DR_STR[] = "VMEXIT_MOV_DR";
 static const char VMEXIT_IO_INSTR_STR[] = "VMEXIT_IO_INSTR";
 static const char VMEXIT_RDMSR_STR[] = "VMEXIT_RDMSR";
 static const char VMEXIT_WRMSR_STR[] = "VMEXIT_WRMSR";
-static const char VMEXIT_ENTRY_FAIL_INVALID_GUEST_STATE_STR[] = "VMEXIT_ENTRY_FAIL_INVALID_GUEST_STATE";
-static const char VMEXIT_ENTRY_FAIL_MSR_LOAD_STR[] = "VMEXIT_ENTRY_FAIL_MSR_LOAD";
+static const char VMEXIT_INVALID_GUEST_STATE_STR[] = "VMEXIT_INVALID_GUEST_STATE";
+static const char VMEXIT_INVALID_MSR_LOAD_STR[] = "VMEXIT_INVALID_MSR_LOAD";
 static const char VMEXIT_MWAIT_STR[] = "VMEXIT_MWAIT";
 static const char VMEXIT_MONITOR_STR[] = "VMEXIT_MONITOR";
 static const char VMEXIT_PAUSE_STR[] = "VMEXIT_PAUSE";
-static const char VMEXIT_ENTRY_FAILURE_MACHINE_CHECK_STR[] = "VMEXIT_ENTRY_FAILURE_MACHINE_CHECK";
+static const char VMEXIT_INVALID_MACHINE_CHECK_STR[] = "VMEXIT_INVALIDE_MACHINE_CHECK";
 static const char VMEXIT_TPR_BELOW_THRESHOLD_STR[] = "VMEXIT_TPR_BELOW_THRESHOLD";
 static const char VMEXIT_APIC_STR[] = "VMEXIT_APIC";
 static const char VMEXIT_GDTR_IDTR_STR[] = "VMEXIT_GDTR_IDTR";
@@ -520,18 +409,18 @@ const char * v3_vmx_exit_code_to_str(vmx_exit_t exit)
             return VMEXIT_RDMSR_STR;
         case VMEXIT_WRMSR:
             return VMEXIT_WRMSR_STR;
-        case VMEXIT_ENTRY_FAIL_INVALID_GUEST_STATE:
-            return VMEXIT_ENTRY_FAIL_INVALID_GUEST_STATE_STR;
-        case VMEXIT_ENTRY_FAIL_MSR_LOAD:
-            return VMEXIT_ENTRY_FAIL_MSR_LOAD_STR;
+        case VMEXIT_INVALID_GUEST_STATE:
+            return VMEXIT_INVALID_GUEST_STATE_STR;
+        case VMEXIT_INVALID_MSR_LOAD:
+            return VMEXIT_INVALID_MSR_LOAD_STR;
         case VMEXIT_MWAIT:
             return VMEXIT_MWAIT_STR;
         case VMEXIT_MONITOR:
             return VMEXIT_MONITOR_STR;
         case VMEXIT_PAUSE:
             return VMEXIT_PAUSE_STR;
-        case VMEXIT_ENTRY_FAILURE_MACHINE_CHECK:
-            return VMEXIT_ENTRY_FAILURE_MACHINE_CHECK_STR;
+        case VMEXIT_INVALID_MACHINE_CHECK:
+            return VMEXIT_INVALID_MACHINE_CHECK_STR;
         case VMEXIT_TPR_BELOW_THRESHOLD:
             return VMEXIT_TPR_BELOW_THRESHOLD_STR;
         case VMEXIT_APIC: