moved VM runstate settings to avoid race condition

[palacios.git] / palacios / src / palacios / vmx.c

diff --git a/palacios/src/palacios/vmx.c b/palacios/src/palacios/vmx.c
index 405bca8..5746e93 100644 (file)
--- a/palacios/src/palacios/vmx.c
+++ b/palacios/src/palacios/vmx.c
@@ -136,53 +136,6 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state)
     /******* Setup Host State **********/
 
     /* Cache GDTR, IDTR, and TR in host struct */
-    addr_t gdtr_base;
-    struct {
-        uint16_t selector;
-        addr_t   base;
-    } __attribute__((packed)) tmp_seg;
-    
-
-    __asm__ __volatile__(
-                        "sgdt (%0);"
-                        :
-                        : "q"(&tmp_seg)
-                        : "memory"
-                        );
-    gdtr_base = tmp_seg.base;
-    vmx_state->host_state.gdtr.base = gdtr_base;
-
-    __asm__ __volatile__(
-                        "sidt (%0);"
-                        :
-                        : "q"(&tmp_seg)
-                        : "memory"
-                        );
-    vmx_state->host_state.idtr.base = tmp_seg.base;
-
-    __asm__ __volatile__(
-                        "str (%0);"
-                        :
-                        : "q"(&tmp_seg)
-                        : "memory"
-                        );
-    vmx_state->host_state.tr.selector = tmp_seg.selector;
-
-    /* The GDTR *index* is bits 3-15 of the selector. */
-    struct tss_descriptor * desc = NULL;
-    desc = (struct tss_descriptor *)(gdtr_base + (8 * (tmp_seg.selector >> 3)));
-
-    tmp_seg.base = ((desc->base1) |
-                   (desc->base2 << 16) |
-                   (desc->base3 << 24) |
-#ifdef __V3_64BIT__
-                   ((uint64_t)desc->base4 << 32)
-#else 
-                   (0)
-#endif
-                   );
-
-    vmx_state->host_state.tr.base = tmp_seg.base;
 
 
     /********** Setup VMX Control Fields ***********/
@@ -213,10 +166,6 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state)
 
 
 
-    
-
-
-
 #ifdef __V3_64BIT__
     // Ensure host runs in 64-bit mode at each VM EXIT
     vmx_state->exit_ctrls.host_64_on = 1;
@@ -400,13 +349,6 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state)
     
     // save STAR, LSTAR, FMASK, KERNEL_GS_BASE MSRs in MSR load/store area
     {
-#define IA32_STAR 0xc0000081
-#define IA32_LSTAR 0xc0000082
-#define IA32_FMASK 0xc0000084
-#define IA32_KERN_GS_BASE 0xc0000102
-
-#define IA32_CSTAR 0xc0000083 // Compatibility mode STAR (ignored for now... hopefully its not that important...)
-
        int msr_ret = 0;
 
        struct vmcs_msr_entry * exit_store_msrs = NULL;
@@ -438,24 +380,42 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state)
        entry_load_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area + (sizeof(struct vmcs_msr_entry) * 8));
 
 
-       exit_store_msrs[0].index = IA32_STAR;
-       exit_store_msrs[1].index = IA32_LSTAR;
-       exit_store_msrs[2].index = IA32_FMASK;
-       exit_store_msrs[3].index = IA32_KERN_GS_BASE;
+       exit_store_msrs[0].index = IA32_STAR_MSR;
+       exit_store_msrs[1].index = IA32_LSTAR_MSR;
+       exit_store_msrs[2].index = IA32_FMASK_MSR;
+       exit_store_msrs[3].index = IA32_KERN_GS_BASE_MSR;
        
        memcpy(exit_store_msrs, exit_load_msrs, sizeof(struct vmcs_msr_entry) * 4);
        memcpy(exit_store_msrs, entry_load_msrs, sizeof(struct vmcs_msr_entry) * 4);
 
        
-       v3_get_msr(IA32_STAR, &(exit_load_msrs[0].hi), &(exit_load_msrs[0].lo));
-       v3_get_msr(IA32_LSTAR, &(exit_load_msrs[1].hi), &(exit_load_msrs[1].lo));
-       v3_get_msr(IA32_FMASK, &(exit_load_msrs[2].hi), &(exit_load_msrs[2].lo));
-       v3_get_msr(IA32_KERN_GS_BASE, &(exit_load_msrs[3].hi), &(exit_load_msrs[3].lo));
+       v3_get_msr(IA32_STAR_MSR, &(exit_load_msrs[0].hi), &(exit_load_msrs[0].lo));
+       v3_get_msr(IA32_LSTAR_MSR, &(exit_load_msrs[1].hi), &(exit_load_msrs[1].lo));
+       v3_get_msr(IA32_FMASK_MSR, &(exit_load_msrs[2].hi), &(exit_load_msrs[2].lo));
+       v3_get_msr(IA32_KERN_GS_BASE_MSR, &(exit_load_msrs[3].hi), &(exit_load_msrs[3].lo));
 
        msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_ADDR, (addr_t)V3_PAddr(exit_store_msrs));
        msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_ADDR, (addr_t)V3_PAddr(exit_load_msrs));
        msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_ADDR, (addr_t)V3_PAddr(entry_load_msrs));
 
+
+       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);
+
+
+       // IMPORTANT: These SYSCALL MSRs are currently not handled by hardware or cached
+       // We should really emulate these ourselves, or ideally include them in the MSR store area if there is room
+       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);
+       
+       v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL);
+       v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL);
+       
+
     }    
 
     /* Sanity check ctrl/reg fields against hw_defaults */
@@ -851,7 +811,6 @@ int v3_vmx_enter(struct guest_info * info) {
     check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
     check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
 
-
     if (v3_update_vmcs_host_state(info)) {
        v3_enable_ints();
         PrintError("Could not write host state\n");
@@ -861,8 +820,6 @@ int v3_vmx_enter(struct guest_info * info) {
 
     if (vmx_info->state == VMX_UNLAUNCHED) {
        vmx_info->state = VMX_LAUNCHED;
-
-       info->vm_info->run_state = VM_RUNNING;
        ret = v3_vmx_launch(&(info->vm_regs), info, &(info->ctrl_regs));
     } else {
        V3_ASSERT(vmx_info->state != VMX_UNLAUNCHED);
@@ -955,12 +912,17 @@ int v3_start_vmx_guest(struct guest_info * info) {
 
     if (info->vcpu_id == 0) {
        info->core_run_state = CORE_RUNNING;
-       info->vm_info->run_state = VM_RUNNING;
     } else {
 
         PrintDebug("VMX core %u: Waiting for core initialization\n", info->vcpu_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("VMX core %u: still waiting for INIT\n",info->vcpu_id);
         }