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.


VMX is working for a 32-bit Linux kernel. It should also work for a 64-bit kernel...
[palacios.git] / palacios / src / palacios / vmx_ctrl_regs.c
index 10503b4..8fb7a81 100644 (file)
 #include <palacios/vmx.h>
 #include <palacios/vmx_assist.h>
 #include <palacios/vm_guest_mem.h>
+#include <palacios/vmm_direct_paging.h>
+#include <palacios/vmx_handler.h>
 
-static int handle_mov_to_cr0(struct guest_info * info, v3_reg_t new_val);
+static v3_reg_t * get_reg_ptr(struct guest_info * info, struct vmx_exit_cr_qual cr_qual);
+static int handle_mov_to_cr0(struct guest_info * info, v3_reg_t * new_val);
+static int handle_mov_to_cr3(struct guest_info * info, v3_reg_t * cr3_reg);
+static int handle_mov_from_cr3(struct guest_info * info, v3_reg_t * cr3_reg);
 
-int v3_vmx_handle_cr0_write(struct guest_info * info, v3_reg_t new_val) {
-    return handle_mov_to_cr0(info, new_val);
+int v3_vmx_handle_cr0_access(struct guest_info * info) {
+    struct vmx_exit_cr_qual cr_qual;
+    
+    vmcs_read(VMCS_EXIT_QUAL, &(cr_qual.value));
+
+    if (cr_qual.access_type < 2) {
+        v3_reg_t * reg = get_reg_ptr(info, cr_qual);
+        
+        if (cr_qual.access_type == 0) {
+
+            if (handle_mov_to_cr0(info, reg) != 0) {
+                PrintError("Could not handle CR0 write\n");
+                return -1;
+            }
+        } else {
+            // Mov from cr
+           PrintError("Mov From CR0 not handled\n");
+           return -1;
+        }
+
+        return 0;
+    }
+
+    PrintError("Invalid CR0 Access type?? (type=%d)\n", cr_qual.access_type);
+    return -1;
+}
+
+int v3_vmx_handle_cr3_access(struct guest_info * info) {
+    struct vmx_exit_cr_qual cr_qual;
+
+    vmcs_read(VMCS_EXIT_QUAL, &(cr_qual.value));
+
+    if (cr_qual.access_type < 2) {
+        v3_reg_t * reg = get_reg_ptr(info, cr_qual);
+
+        if (cr_qual.access_type == 0) {
+            return handle_mov_to_cr3(info, reg);
+        } else {
+            return handle_mov_from_cr3(info, reg);
+        }
+    }
+
+    PrintError("Invalid CR3 Access type?? (type=%d)\n", cr_qual.access_type);
+    return -1;
 }
 
-static int handle_mov_to_cr0(struct guest_info * info, v3_reg_t new_val) {
-    PrintDebug("CR0 RIP: %p\n", (void *)info->rip);
+static int handle_mov_to_cr3(struct guest_info * info, v3_reg_t * cr3_reg) {
+    int instr_len = 0;
+
+    if (info->shdw_pg_mode == SHADOW_PAGING) {
+
+        PrintDebug("Old Guest CR3=%p, Old Shadow CR3=%p\n",
+                  (void *)info->ctrl_regs.cr3,
+                  (void *)info->shdw_pg_state.guest_cr3);
+
+        if (info->cpu_mode == LONG) {
+            info->shdw_pg_state.guest_cr3 = (uint64_t)*cr3_reg;
+        } else {
+            info->shdw_pg_state.guest_cr3 = (uint32_t)*cr3_reg;
+        }
+
+
+        if (v3_get_vm_mem_mode(info) == VIRTUAL_MEM) {
+            if (v3_activate_shadow_pt(info) == -1) {
+                PrintError("Failed to activate 32 bit shadow page table\n");
+                return -1;
+            }
+        }
+
+        PrintDebug("New guest CR3=%p, New shadow CR3=%p\n",
+                  (void *)info->ctrl_regs.cr3,
+                  (void *)info->shdw_pg_state.guest_cr3);
+
+    } else if (info->shdw_pg_mode == NESTED_PAGING) {
+        PrintError("Nested paging not available in VMX right now!\n");
+        return -1;
+    }
+
+
+    vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
+    info->rip += instr_len;
+
+    return 0;
+}
+
+static int handle_mov_from_cr3(struct guest_info * info, v3_reg_t * cr3_reg) {
+    int instr_len = 0;
+
+    if (info->shdw_pg_mode == SHADOW_PAGING) {
 
+        if ((v3_get_vm_cpu_mode(info) == LONG) ||
+           (v3_get_vm_cpu_mode(info) == LONG_32_COMPAT)) {
+
+            *cr3_reg = (uint64_t)info->shdw_pg_state.guest_cr3;
+        } else {
+            *cr3_reg = (uint32_t)info->shdw_pg_state.guest_cr3;
+        }
+
+    } else {
+        PrintError("Unhandled paging mode\n");
+        return -1;
+    }
+
+
+    vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
+    info->rip += instr_len;
+
+    return 0;
+}
+
+static int handle_mov_to_cr0(struct guest_info * info, v3_reg_t * new_cr0) {
     struct cr0_32 * guest_cr0 = (struct cr0_32 *)&(info->ctrl_regs.cr0);
-    struct cr0_32 * new_cr0 = (struct cr0_32 *)&new_val;
-    struct cr0_32 * shadow_cr0 = (struct cr0_32 *)&(info->shdw_pg_state.guest_cr0);
+    struct cr0_32 * shdw_cr0 = (struct cr0_32 *)&(info->shdw_pg_state.guest_cr0);
+    struct cr0_32 * new_shdw_cr0 = (struct cr0_32 *)new_cr0;
+    struct vmx_data * vmx_info = (struct vmx_data *)info->vmm_data;
+    uint_t paging_transition = 0;
+    int instr_len = 0;
 
-    // PG and PE are always enabled for VMX
+    PrintDebug("Old shadow CR0: 0x%x, New shadow CR0: 0x%x\n",
+              (uint32_t)info->shdw_pg_state.guest_cr0, (uint32_t)*new_cr0);
 
-    // Check if this is a paging transition
-    PrintDebug("Old CR0: 0x%x\n", *(uint32_t *)guest_cr0);
-    PrintDebug("Old shadow CR0: 0x%x\n", *(uint32_t *)shadow_cr0);
-    PrintDebug("New CR0: 0x%x\n", *(uint32_t *)new_cr0);
-            
-    if ( new_cr0->pe ) {
+
+    if (new_shdw_cr0->pe != shdw_cr0->pe) {
+        PrintDebug("Guest CR0: 0x%x\n", *(uint32_t *)guest_cr0);
+        PrintDebug("Old shadow CR0: 0x%x\n", *(uint32_t *)shdw_cr0);
+        PrintDebug("New shadow CR0: 0x%x\n", *(uint32_t *)new_shdw_cr0);
 
         if (v3_vmxassist_ctx_switch(info) != 0) {
             PrintError("Unable to execute VMXASSIST context switch!\n");
             return -1;
         }
 
-        ((struct vmx_data *)info->vmm_data)->state = VMXASSIST_DISABLED;
+        v3_load_vmcs_guest_state(info);
 
-        PrintDebug("New Shadow: 0x%x\n", *(uint32_t *)shadow_cr0);
-        PrintDebug("mem_mode: %s\n", v3_mem_mode_to_str(v3_get_vm_mem_mode(info))); 
+        if (vmx_info->state == VMXASSIST_ENABLED) {
+            PrintDebug("Loading VMXASSIST at RIP: %p\n", (void *)info->rip);
+        } else {
+            PrintDebug("Leaving VMXASSIST and entering protected mode at RIP: %p\n",
+                      (void *)info->rip);
+        }
 
+        // vmx assist sets the new cr values itself
         return 0;
     }
 
-    return -1;
+    if (new_shdw_cr0->pg != shdw_cr0->pg) {
+        paging_transition = 1;
+    }
+    // The shadow always reflects the new value
+    *shdw_cr0 = *new_shdw_cr0;
+
+    // We don't care about most of the flags, so lets go for it 
+    // and set them to the guest values
+    *guest_cr0 = *shdw_cr0;
+
+    // Except PG, PE, and NE, which are always set
+    guest_cr0->pe = 1;
+    guest_cr0->pg = 1;
+    guest_cr0->ne = 1;
+
+    if (paging_transition) {
+        // Paging transition
+
+        if (v3_get_vm_mem_mode(info) == VIRTUAL_MEM) {
+            struct efer_64 * guest_efer = (struct efer_64 *)&(info->ctrl_regs.efer);
+
+            if (guest_efer->lme == 1) {
+                PrintDebug("Enabling long mode\n");
+
+                guest_efer->lma = 1;
+                guest_efer->lme = 1;
+
+                vmx_info->entry_ctrls.guest_ia32e = 1;
+            }
+
+            PrintDebug("Activating Shadow Page tables\n");
+
+            if (v3_activate_shadow_pt(info) == -1) {
+                PrintError("Failed to activate shadow page tables\n");
+                return -1;
+            }
+
+        } else if (v3_activate_passthrough_pt(info) == -1) {
+            PrintError("Failed to activate passthrough page tables\n");
+            return -1;
+        }
+    }
+   
+    // PE loads its own RIP, otherwise we need to skip ahead an instruction
+
+    vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
+    info->rip += instr_len;
+   
+    return 0;
+}
+
+static v3_reg_t * get_reg_ptr(struct guest_info * info, struct vmx_exit_cr_qual cr_qual) {
+    v3_reg_t * reg = NULL;
+
+    switch (cr_qual.gpr) {
+       case 0:
+           reg = &(info->vm_regs.rax);
+           break;
+       case 1:
+           reg = &(info->vm_regs.rcx);
+           break;
+       case 2:
+           reg = &(info->vm_regs.rdx);
+           break;
+       case 3:
+           reg = &(info->vm_regs.rbx);
+           break;
+       case 4:
+           reg = &(info->vm_regs.rsp);
+           break;
+       case 5:
+           reg = &(info->vm_regs.rbp);
+           break;
+       case 6:
+           reg = &(info->vm_regs.rsi);
+           break;
+       case 7:
+           reg = &(info->vm_regs.rdi);
+           break;
+       case 8:
+           reg = &(info->vm_regs.r8);
+           break;
+       case 9:
+           reg = &(info->vm_regs.r9);
+           break;
+       case 10:
+           reg = &(info->vm_regs.r10);
+           break;
+       case 11:
+           reg = &(info->vm_regs.r11);
+           break;
+       case 12:
+           reg = &(info->vm_regs.r11);
+           break;
+       case 13:
+           reg = &(info->vm_regs.r13);
+           break;
+       case 14:
+           reg = &(info->vm_regs.r14);
+           break;
+       case 15:
+           reg = &(info->vm_regs.r15);
+           break;
+    }
+
+    return reg;
 }
 
+