[v2,03/15] VMX: don't run with CR4.VMXE set when VMX could not be enabled

Commit Message

Jan Beulich Nov. 24, 2023, 8:37 a.m. UTC

While generally benign, doing so is still at best misleading.

Signed-off-by: Jan Beulich <[e-mail]>
---
Using set_in_cr4() seems favorable over updating mmu_cr4_features
despite the resulting redundant CR4 update. But I certainly could be
talked into going the alternative route.
---
v2: Actually clear CR4.VMXE for the BSP on the error path.

Patch

--- a/xen/arch/x86/hvm/vmx/vmx.c
+++ b/xen/arch/x86/hvm/vmx/vmx.c
@@ -2959,14 +2959,18 @@  static bool __init has_if_pschange_mc(vo
 
 const struct hvm_function_table * __init start_vmx(void)
 {
-    set_in_cr4(X86_CR4_VMXE);
+    write_cr4(read_cr4() | X86_CR4_VMXE);
 
     if ( vmx_vmcs_init() )
     {
+        write_cr4(read_cr4() & ~X86_CR4_VMXE);
         printk("VMX: failed to initialise.\n");
         return NULL;
     }
 
+    /* Arrange for APs to have CR4.VMXE set early on. */
+    set_in_cr4(X86_CR4_VMXE);
+
     vmx_function_table.singlestep_supported = cpu_has_monitor_trap_flag;
 
     if ( cpu_has_vmx_dt_exiting )