Changeset 71966 in vbox for trunk

Timestamp:

Apr 23, 2018 4:40:51 AM (7 years ago)

Author:

vboxsync

Message:

VMM/HMSVMAll: Some comments on the nuances of the VMCB caching with hardware-assisted SVM execution of nested-guests.

Location:

trunk

Files:

• include/VBox/vmm/hm_svm.h (modified) (1 diff)
• src/VBox/VMM/VMMAll/HMSVMAll.cpp (modified) (1 diff)

trunk/include/VBox/vmm/hm_svm.h

@@ -1019 +1019 @@
  * SVM R0 code for hardware-assisted SVM execution of a nested-guest.
  *
+ * A VMCB field needs to be cached when it needs to be modified for execution using
+ * hardware-assisted SVM and any of the following are true:
+ *   - If the original field needs to be inspected during execution of the
+ *     nested-guest or #VMEXIT processing.
+ *   - If the field is written back to memory on #VMEXIT by the physical CPU.
+ *
+ * A VMCB field needs to be restored only when the field is written back to
+ * memory on #VMEXIT by the physical CPU and thus would be visible to the guest.
+ *
  * @remarks Please update hmR3InfoSvmNstGstVmcbCache() when changes are made to
  *          this structure.

trunk/src/VBox/VMM/VMMAll/HMSVMAll.cpp

@@ -146 +146 @@
         /*
          * Restore fields as our own code might look at the VMCB controls as part
-         * of the #VMEXIT handling. Otherwise, we don't need to restore the current
-         * fields because none of them are written by a physical CPU on #VMEXIT.
+         * of the #VMEXIT handling in IEM. Otherwise, strictly speaking we don't need to
+         * restore these fields because currently none of them are written back to memory
+         * by a physical CPU on #VMEXIT.
          */
         pVmcbNstGstCtrl->u16InterceptRdCRx                 = pVmcbNstGstCache->u16InterceptRdCRx;