[PATCH v3 2/4] x86/spec-ctrl: defer context-switch IBPB until guest entry

[Jan Beulich <jbeulich@suse.com>]

In order to avoid clobbering Xen's own predictions, defer the barrier as
much as possible. Merely mark the CPU as needing a barrier issued the
next time we're exiting to guest context.

Suggested-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
I couldn't find any sensible (central/unique) place where to move the
comment which is being deleted alongside spec_ctrl_new_guest_context().
---
v3: New.

--- a/xen/arch/x86/domain.c
+++ b/xen/arch/x86/domain.c
@@ -2038,7 +2038,7 @@ void context_switch(struct vcpu *prev, s
              */
             if ( *last_id != next_id )
             {
-                spec_ctrl_new_guest_context();
+                info->spec_ctrl_flags |= SCF_exit_ibpb;
                 *last_id = next_id;
             }
         }
--- a/xen/arch/x86/include/asm/spec_ctrl.h
+++ b/xen/arch/x86/include/asm/spec_ctrl.h
@@ -67,28 +67,6 @@
 void init_speculation_mitigations(void);
 void spec_ctrl_init_domain(struct domain *d);
 
-/*
- * Switch to a new guest prediction context.
- *
- * This flushes all indirect branch predictors (BTB, RSB/RAS), so guest code
- * which has previously run on this CPU can't attack subsequent guest code.
- *
- * As this flushes the RSB/RAS, it destroys the predictions of the calling
- * context.  For best performace, arrange for this to be used when we're going
- * to jump out of the current context, e.g. with reset_stack_and_jump().
- *
- * For hardware which mis-implements IBPB, fix up by flushing the RSB/RAS
- * manually.
- */
-static always_inline void spec_ctrl_new_guest_context(void)
-{
-    wrmsrl(MSR_PRED_CMD, PRED_CMD_IBPB);
-
-    /* (ab)use alternative_input() to specify clobbers. */
-    alternative_input("", "DO_OVERWRITE_RSB", X86_BUG_IBPB_NO_RET,
-                      : "rax", "rcx");
-}
-
 extern int8_t opt_ibpb_ctxt_switch;
 extern bool opt_ssbd;
 extern int8_t opt_eager_fpu;
--- a/xen/arch/x86/spec_ctrl.c
+++ b/xen/arch/x86/spec_ctrl.c
@@ -854,6 +854,11 @@ static void __init ibpb_calculations(voi
      */
     if ( opt_ibpb_ctxt_switch == -1 )
         opt_ibpb_ctxt_switch = !(opt_ibpb_entry_hvm && opt_ibpb_entry_pv);
+    if ( opt_ibpb_ctxt_switch )
+    {
+        setup_force_cpu_cap(X86_FEATURE_IBPB_EXIT_PV);
+        setup_force_cpu_cap(X86_FEATURE_IBPB_EXIT_HVM);
+    }
 }
 
 /* Calculate whether this CPU is vulnerable to L1TF. */