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.


added write protect flag to realmode CR0, to enable memory hooks
[palacios.git] / palacios / src / palacios / svm.c
index 9c58c25..490e342 100644 (file)
@@ -36,8 +36,6 @@
 
 #include <palacios/vmm_rbtree.h>
 
-#include <palacios/vmm_profiler.h>
-
 #include <palacios/vmm_direct_paging.h>
 
 #include <palacios/vmm_ctrl_regs.h>
 #include <palacios/svm_io.h>
 
 
+
+// This is a global pointer to the host's VMCB
+static void * host_vmcb = NULL;
+
 extern void v3_stgi();
 extern void v3_clgi();
 //extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
-extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs);
+extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
 
 
 static vmcb_t * Allocate_VMCB() {
@@ -122,7 +124,8 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info *vm_info) {
 
     vm_info->vm_regs.rdx = 0x00000f00;
 
-    guest_state->cr0 = 0x60000010;
+
+    guest_state->cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
 
 
     guest_state->cs.selector = 0xf000;
@@ -262,11 +265,8 @@ static int init_svm_guest(struct guest_info * info, struct v3_vm_config * config
     return 0;
 }
 
-
-
-// can we start a kernel thread here...
 static int start_svm_guest(struct guest_info *info) {
-    vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
+    //    vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
     //  vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
     uint_t num_exits = 0;
 
@@ -276,39 +276,14 @@ static int start_svm_guest(struct guest_info *info) {
     //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
     
     info->run_state = VM_RUNNING;
-    
+    rdtscll(info->yield_start_cycle);
+
+
     while (1) {
        ullong_t tmp_tsc;
        
-
-#ifdef __V3_64BIT__
-
-#define MSR_LSTAR         0xc0000082
-#define MSR_CSTAR         0xc0000083
-#define MSR_SF_MASK       0xc0000084
-#define MSR_GS_BASE       0xc0000101
-#define MSR_KERNGS_BASE   0xc0000102
-       struct v3_msr host_cstar;
-       struct v3_msr host_lstar;
-       struct v3_msr host_syscall_mask;
-       struct v3_msr host_gs_base;
-       struct v3_msr host_kerngs_base;
-
-#else 
-
-#define MSR_SYSENTER_CS       0x00000174
-#define MSR_SYSENTER_ESP      0x00000175
-#define MSR_SYSENTER_EIP      0x00000176
-
-       struct v3_msr host_sysenter_cs;
-       struct v3_msr host_sysenter_esp;
-       struct v3_msr host_sysenter_eip;
-
-#endif
-
-#define MSR_STAR              0xc0000081
-       struct v3_msr host_star;
-
+       // Conditionally yield the CPU if the timeslice has expired
+       v3_yield_cond(info);
 
        /*
          PrintDebug("SVM Entry to CS=%p  rip=%p...\n", 
@@ -316,62 +291,36 @@ static int start_svm_guest(struct guest_info *info) {
          (void *)(addr_t)info->rip);
        */
 
+       // disable global interrupts for vm state transition
+       v3_clgi();
+
 
-#ifdef __V3_64BIT__
-       v3_get_msr(MSR_SF_MASK, &(host_syscall_mask.hi), &(host_syscall_mask.lo));
-       v3_get_msr(MSR_LSTAR, &(host_lstar.hi), &(host_lstar.lo));
-       v3_get_msr(MSR_CSTAR, &(host_cstar.hi), &(host_cstar.lo));
-       v3_get_msr(MSR_GS_BASE, &(host_gs_base.hi), &(host_gs_base.lo));
-       v3_get_msr(MSR_KERNGS_BASE, &(host_kerngs_base.hi), &(host_kerngs_base.lo));
-#else 
-       v3_get_msr(MSR_SYSENTER_CS, &(host_sysenter_cs.hi), &(host_sysenter_cs.lo));
-       v3_get_msr(MSR_SYSENTER_ESP, &(host_sysenter_esp.hi), &(host_sysenter_esp.lo));
-       v3_get_msr(MSR_SYSENTER_EIP, &(host_sysenter_eip.hi), &(host_sysenter_eip.lo));
-#endif
-       v3_get_msr(MSR_STAR, &(host_star.hi), &(host_star.lo));
 
        rdtscll(info->time_state.cached_host_tsc);
        //    guest_ctrl->TSC_OFFSET = info->time_state.guest_tsc - info->time_state.cached_host_tsc;
        
-       v3_svm_launch((vmcb_t*)V3_PAddr(info->vmm_data), &(info->vm_regs));
+       v3_svm_launch((vmcb_t*)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcb);
        
        rdtscll(tmp_tsc);
-       
-#ifdef __V3_64BIT__
-       v3_set_msr(MSR_SF_MASK, host_syscall_mask.hi, host_syscall_mask.lo);
-       v3_set_msr(MSR_LSTAR, host_lstar.hi, host_lstar.lo);
-       v3_set_msr(MSR_CSTAR, host_cstar.hi, host_cstar.lo);
-       v3_set_msr(MSR_GS_BASE, host_gs_base.hi, host_gs_base.lo);
-       v3_set_msr(MSR_KERNGS_BASE, host_kerngs_base.hi, host_kerngs_base.lo);
-#else 
-       v3_set_msr(MSR_SYSENTER_CS, host_sysenter_cs.hi, host_sysenter_cs.lo);
-       v3_set_msr(MSR_SYSENTER_ESP, host_sysenter_esp.hi, host_sysenter_esp.lo);
-       v3_set_msr(MSR_SYSENTER_EIP, host_sysenter_eip.hi, host_sysenter_eip.lo);
-#endif
-       v3_set_msr(MSR_STAR, host_star.hi, host_star.lo);
 
        
        //PrintDebug("SVM Returned\n");
 
+       // reenable global interrupts after vm exit
+       v3_stgi();
+
+
+       // Conditionally yield the CPU if the timeslice has expired
+       v3_yield_cond(info);
 
 
        v3_update_time(info, tmp_tsc - info->time_state.cached_host_tsc);
        num_exits++;
-
-       //PrintDebug("Turning on global interrupts\n");
-       v3_stgi();
-       v3_clgi();
        
        if ((num_exits % 5000) == 0) {
            PrintDebug("SVM Exit number %d\n", num_exits);
-
-           if (info->enable_profiler) {
-               v3_print_profile(info);
-           }
        }
 
-
-     
        if (v3_handle_svm_exit(info) != 0) {
            vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
            addr_t host_addr;
@@ -381,23 +330,7 @@ static int start_svm_guest(struct guest_info *info) {
            
            PrintDebug("SVM ERROR!!\n"); 
       
-           PrintDebug("RIP: %p\n", (void *)(addr_t)(guest_state->rip));
-
-
-           linear_addr = get_addr_linear(info, guest_state->rip, &(info->segments.cs));
-
-
-           PrintDebug("RIP Linear: %p\n", (void *)linear_addr);
-           v3_print_segments(info);
-           v3_print_ctrl_regs(info);
-           if (info->shdw_pg_mode == SHADOW_PAGING) {
-               PrintDebug("Shadow Paging Guest Registers:\n");
-               PrintDebug("\tGuest CR0=%p\n", (void *)(addr_t)(info->shdw_pg_state.guest_cr0));
-               PrintDebug("\tGuest CR3=%p\n", (void *)(addr_t)(info->shdw_pg_state.guest_cr3));
-               PrintDebug("\tGuest EFER=%p\n", (void *)(addr_t)(info->shdw_pg_state.guest_efer.value));
-               // CR4
-           }
-           v3_print_GPRs(info);
+           v3_print_guest_state(info);
 
            PrintDebug("SVM Exit Code: %p\n", (void *)(addr_t)guest_ctrl->exit_code); 
       
@@ -407,17 +340,18 @@ static int start_svm_guest(struct guest_info *info) {
            PrintDebug("exit_info2 low = 0x%.8x\n", *(uint_t*)&(guest_ctrl->exit_info2));
            PrintDebug("exit_info2 high = 0x%.8x\n", *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
       
+           linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
+
            if (info->mem_mode == PHYSICAL_MEM) {
                guest_pa_to_host_va(info, linear_addr, &host_addr);
            } else if (info->mem_mode == VIRTUAL_MEM) {
                guest_va_to_host_va(info, linear_addr, &host_addr);
            }
 
-
            PrintDebug("Host Address of rip = 0x%p\n", (void *)host_addr);
 
            PrintDebug("Instr (15 bytes) at %p:\n", (void *)host_addr);
-           PrintTraceMemDump((uchar_t *)host_addr, 15);
+           v3_dump_mem((uint8_t *)host_addr, 15);
 
            break;
        }
@@ -436,11 +370,11 @@ int v3_is_svm_capable() {
     uint_t vm_cr_low = 0, vm_cr_high = 0;
     addr_t eax = 0, ebx = 0, ecx = 0, edx = 0;
 
-    v3_cpuid(CPUID_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
+    v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
   
-    PrintDebug("CPUID_FEATURE_IDS_ecx=%p\n", (void *)ecx);
+    PrintDebug("CPUID_EXT_FEATURE_IDS_ecx=%p\n", (void *)ecx);
 
-    if ((ecx & CPUID_FEATURE_IDS_ecx_svm_avail) == 0) {
+    if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
       PrintDebug("SVM Not Available\n");
       return 0;
     }  else {
@@ -453,7 +387,7 @@ int v3_is_svm_capable() {
            
            v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
            
-           PrintDebug("CPUID_FEATURE_IDS_edx=%p\n", (void *)edx);
+           PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=%p\n", (void *)edx);
            
            if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
                PrintDebug("SVM BIOS Disabled, not unlockable\n");
@@ -466,10 +400,10 @@ int v3_is_svm_capable() {
            PrintDebug("SVM is available and  enabled.\n");
 
            v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
-           PrintDebug("CPUID_FEATURE_IDS_eax=%p\n", (void *)eax);
-           PrintDebug("CPUID_FEATURE_IDS_ebx=%p\n", (void *)ebx);
-           PrintDebug("CPUID_FEATURE_IDS_ecx=%p\n", (void *)ecx);
-           PrintDebug("CPUID_FEATURE_IDS_edx=%p\n", (void *)edx);
+           PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_eax=%p\n", (void *)eax);
+           PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ebx=%p\n", (void *)ebx);
+           PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ecx=%p\n", (void *)ecx);
+           PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=%p\n", (void *)edx);
 
 
            if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
@@ -488,7 +422,7 @@ static int has_svm_nested_paging() {
 
     v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 
-    //PrintDebug("CPUID_FEATURE_IDS_edx=0x%x\n", edx);
+    //PrintDebug("CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
 
     if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
        PrintDebug("SVM Nested Paging not supported\n");
@@ -503,7 +437,6 @@ static int has_svm_nested_paging() {
 
 void v3_init_SVM(struct v3_ctrl_ops * vmm_ops) {
     reg_ex_t msr;
-    void * host_state;
     extern v3_cpu_arch_t v3_cpu_type;
 
     // Enable SVM on the CPU
@@ -513,19 +446,20 @@ void v3_init_SVM(struct v3_ctrl_ops * vmm_ops) {
 
     PrintDebug("SVM Enabled\n");
 
-
     // Setup the host state save area
-    host_state = V3_AllocPages(4);
-
+    host_vmcb = V3_AllocPages(4);
 
     /* 64-BIT-ISSUE */
     //  msr.e_reg.high = 0;
-    //msr.e_reg.low = (uint_t)host_state;
-    msr.r_reg = (addr_t)host_state;
+    //msr.e_reg.low = (uint_t)host_vmcb;
+    msr.r_reg = (addr_t)host_vmcb;
 
-    PrintDebug("Host State being saved at %p\n", (void *)(addr_t)host_state);
+    PrintDebug("Host State being saved at %p\n", (void *)(addr_t)host_vmcb);
     v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
 
+
+
+
     if (has_svm_nested_paging() == 1) {
        v3_cpu_type = V3_SVM_REV3_CPU;
     } else {
@@ -591,126 +525,63 @@ void v3_init_SVM(struct v3_ctrl_ops * vmm_ops) {
 
 
 
-/*static void Init_VMCB(vmcb_t * vmcb, struct guest_info vm_info) {
-  vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
-  vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
-  uint_t i;
-
-
-  guest_state->rsp = vm_info.vm_regs.rsp;
-  guest_state->rip = vm_info.rip;
-
-
-  //ctrl_area->instrs.instrs.CR0 = 1;
-  ctrl_area->cr_reads.cr0 = 1;
-  ctrl_area->cr_writes.cr0 = 1;
-
-  guest_state->efer |= EFER_MSR_svm_enable;
-  guest_state->rflags = 0x00000002; // The reserved bit is always 1
-  ctrl_area->svm_instrs.VMRUN = 1;
-  // guest_state->cr0 = 0x00000001;    // PE 
-  ctrl_area->guest_ASID = 1;
-
-
-  ctrl_area->exceptions.de = 1;
-  ctrl_area->exceptions.df = 1;
-  ctrl_area->exceptions.pf = 1;
-  ctrl_area->exceptions.ts = 1;
-  ctrl_area->exceptions.ss = 1;
-  ctrl_area->exceptions.ac = 1;
-  ctrl_area->exceptions.mc = 1;
-  ctrl_area->exceptions.gp = 1;
-  ctrl_area->exceptions.ud = 1;
-  ctrl_area->exceptions.np = 1;
-  ctrl_area->exceptions.of = 1;
-  ctrl_area->exceptions.nmi = 1;
-
-  guest_state->cs.selector = 0x0000;
-  guest_state->cs.limit=~0u;
-  guest_state->cs.base = guest_state->cs.selector<<4;
-  guest_state->cs.attrib.raw = 0xf3;
-
-  
-  struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
-  for ( i = 0; segregs[i] != NULL; i++) {
-    struct vmcb_selector * seg = segregs[i];
+#if 0
+/* 
+ * Test VMSAVE/VMLOAD Latency 
+ */
+#define vmsave ".byte 0x0F,0x01,0xDB ; "
+#define vmload ".byte 0x0F,0x01,0xDA ; "
+{
+    uint32_t start_lo, start_hi;
+    uint32_t end_lo, end_hi;
+    uint64_t start, end;
     
-    seg->selector = 0x0000;
-    seg->base = seg->selector << 4;
-    seg->attrib.raw = 0xf3;
-    seg->limit = ~0u;
-  }
-  
-  if (vm_info.io_map.num_ports > 0) {
-    struct vmm_io_hook * iter;
-    addr_t io_port_bitmap;
+    __asm__ __volatile__ (
+                         "rdtsc ; "
+                         "movl %%eax, %%esi ; "
+                         "movl %%edx, %%edi ; "
+                         "movq  %%rcx, %%rax ; "
+                         vmsave
+                         "rdtsc ; "
+                         : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
+                         : "c"(host_vmcb), "0"(0), "1"(0), "2"(0), "3"(0)
+                         );
     
-    io_port_bitmap = (addr_t)V3_AllocPages(3);
-    memset((uchar_t*)io_port_bitmap, 0, PAGE_SIZE * 3);
+    start = start_hi;
+    start <<= 32;
+    start += start_lo;
     
-    ctrl_area->IOPM_BASE_PA = io_port_bitmap;
+    end = end_hi;
+    end <<= 32;
+    end += end_lo;
+    
+    PrintDebug("VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
+    
+    __asm__ __volatile__ (
+                         "rdtsc ; "
+                         "movl %%eax, %%esi ; "
+                         "movl %%edx, %%edi ; "
+                         "movq  %%rcx, %%rax ; "
+                         vmload
+                         "rdtsc ; "
+                         : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
+                             : "c"(host_vmcb), "0"(0), "1"(0), "2"(0), "3"(0)
+                             );
+       
+       start = start_hi;
+       start <<= 32;
+       start += start_lo;
 
-    //PrintDebug("Setting up IO Map at 0x%x\n", io_port_bitmap);
+       end = end_hi;
+       end <<= 32;
+       end += end_lo;
 
-    FOREACH_IO_HOOK(vm_info.io_map, iter) {
-      ushort_t port = iter->port;
-      uchar_t * bitmap = (uchar_t *)io_port_bitmap;
 
-      bitmap += (port / 8);
-      PrintDebug("Setting Bit in block %x\n", bitmap);
-      *bitmap |= 1 << (port % 8);
+       PrintDebug("VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
     }
+    /* End Latency Test */
 
-
-    //PrintDebugMemDump((uchar_t*)io_port_bitmap, PAGE_SIZE *2);
-
-    ctrl_area->instrs.IOIO_PROT = 1;
-  }
-
-  ctrl_area->instrs.INTR = 1;
-
-
-
-  if (vm_info.page_mode == SHADOW_PAGING) {
-    PrintDebug("Creating initial shadow page table\n");
-    vm_info.shdw_pg_state.shadow_cr3 |= ((addr_t)create_passthrough_pts_32(&vm_info) & ~0xfff);
-    PrintDebug("Created\n");
-
-    guest_state->cr3 = vm_info.shdw_pg_state.shadow_cr3;
-
-    ctrl_area->cr_reads.cr3 = 1;
-    ctrl_area->cr_writes.cr3 = 1;
-
-
-    ctrl_area->instrs.INVLPG = 1;
-    ctrl_area->instrs.INVLPGA = 1;
-
-    guest_state->g_pat = 0x7040600070406ULL;
-
-    guest_state->cr0 |= 0x80000000;
-  } else if (vm_info.page_mode == NESTED_PAGING) {
-    // Flush the TLB on entries/exits
-    //ctrl_area->TLB_CONTROL = 1;
-
-    // Enable Nested Paging
-    //ctrl_area->NP_ENABLE = 1;
-
-    //PrintDebug("NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
-
-        // Set the Nested Page Table pointer
-    //    ctrl_area->N_CR3 = ((addr_t)vm_info.page_tables);
-    // ctrl_area->N_CR3 = (addr_t)(vm_info.page_tables);
-
-    //   ctrl_area->N_CR3 = Get_CR3();
-    // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
-
-    //    guest_state->g_pat = 0x7040600070406ULL;
-  }
-
-
-
-}
-*/
+#endif