Changeset 97095 in vbox

Timestamp:

Oct 11, 2022 8:46:59 PM (2 years ago)

Author:

vboxsync

Message:

VMM/HMVMXR0: Some comments around importing INHIBIT_INT/NMI early.

File:

• trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp (modified) (2 diffs)

trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp

@@ -6310 +6310 @@
             /*
              * Always import the guest-interruptibility state as we need it while evaluating
-             * injecting events on re-entry.
+             * injecting events on re-entry.  We could in *theory* postpone reading it for
+             * exits that does not involve instruction emulation, but since most exits are
+             * for instruction emulation (exceptions being external interrupts, shadow
+             * paging building page faults and EPT violations, and interrupt window stuff)
+             * this is a reasonable simplification.
              *
              * We don't import CR0 (when unrestricted guest execution is unavailable) despite
@@ -6316 +6320 @@
              * mode changes wrt CR0 are intercepted.
              *
-             * Note! This mask _must_ match the default value for the a_fDonePostExit value
-             *       for the vmxHCImportGuestState template!
+             * Note! This mask _must_ match the default value for the default a_fDonePostExit
+             *       value for the vmxHCImportGuestState template!
              */
+            /** @todo r=bird: consider dropping the INHIBIT_XXX and fetch the state
+             * explicitly in the exit handlers and injection function.  That way we have
+             * fewer clusters of vmread spread around the code, because the EM history
+             * executor won't execute very many non-exiting instructions before stopping. */
             rc = vmxHCImportGuestState<  CPUMCTX_EXTRN_INHIBIT_INT
                                        | CPUMCTX_EXTRN_INHIBIT_NMI