Branch Harden is enabled by default at compile and boot time.  Invert the code
to compile with lfence by default and nop out in the non-default case.

This has several advantages.  It removes 3829 patch points (in the random
build of Xen I have to hand) by default on boot, 70% (!) of the
.altinstr_replacement section.  For builds of Xen with a non-nops capable tool
chain, the code after `spec-ctrl=no-branch-harden` is better because Xen can
write long nops.

Most importantly however, it means the disassembly actually matches what runs
in the common case, with the ability to distinguish the lfences from other
uses of nops.

Finally, make opt_branch_harden local to spec_ctrl.c and __initdata.  It has
never been used externally, even at it's introduction in c/s 3860d5534df4
"spec: add l1tf-barrier".

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
---
CC: Jan Beulich <JBeulich@suse.com>
CC: Roger Pau Monné <roger.pau@citrix.com>
CC: Wei Liu <wl@xen.org>
---
 xen/arch/x86/spec_ctrl.c          | 7 ++++---
 xen/include/asm-x86/cpufeatures.h | 2 +-
 xen/include/asm-x86/nospec.h      | 2 +-
 xen/include/asm-x86/spec_ctrl.h   | 1 -
 4 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/xen/arch/x86/spec_ctrl.c b/xen/arch/x86/spec_ctrl.c
index 16d2a1d172b5..a5569c7f2b3f 100644
--- a/xen/arch/x86/spec_ctrl.c
+++ b/xen/arch/x86/spec_ctrl.c
@@ -52,7 +52,7 @@ bool __read_mostly opt_ibpb = true;
 bool __read_mostly opt_ssbd = false;
 int8_t __read_mostly opt_eager_fpu = -1;
 int8_t __read_mostly opt_l1d_flush = -1;
-bool __read_mostly opt_branch_harden = true;
+static bool __initdata opt_branch_harden = true;
 
 bool __initdata bsp_delay_spec_ctrl;
 uint8_t __read_mostly default_xen_spec_ctrl;
@@ -1121,8 +1121,9 @@ void __init init_speculation_mitigations(void)
     else if ( opt_l1d_flush == -1 )
         opt_l1d_flush = cpu_has_bug_l1tf && !(caps & ARCH_CAPS_SKIP_L1DFL);
 
-    if ( opt_branch_harden )
-        setup_force_cpu_cap(X86_FEATURE_SC_BRANCH_HARDEN);
+    /* We compile lfence's in by default, and nop them out if requested. */
+    if ( !opt_branch_harden )
+        setup_force_cpu_cap(X86_FEATURE_SC_NO_BRANCH_HARDEN);
 
     /*
      * We do not disable HT by default on affected hardware.
diff --git a/xen/include/asm-x86/cpufeatures.h b/xen/include/asm-x86/cpufeatures.h
index 72beb7babcce..b10154fc44bb 100644
--- a/xen/include/asm-x86/cpufeatures.h
+++ b/xen/include/asm-x86/cpufeatures.h
@@ -27,7 +27,7 @@ XEN_CPUFEATURE(XEN_SMAP,          X86_SYNTH(11)) /* SMAP gets used by Xen itself
 /* Bit 12 - unused. */
 XEN_CPUFEATURE(IND_THUNK_LFENCE,  X86_SYNTH(13)) /* Use IND_THUNK_LFENCE */
 XEN_CPUFEATURE(IND_THUNK_JMP,     X86_SYNTH(14)) /* Use IND_THUNK_JMP */
-XEN_CPUFEATURE(SC_BRANCH_HARDEN,  X86_SYNTH(15)) /* Conditional Branch Hardening */
+XEN_CPUFEATURE(SC_NO_BRANCH_HARDEN, X86_SYNTH(15)) /* (Disable) Conditional branch hardening */
 XEN_CPUFEATURE(SC_MSR_PV,         X86_SYNTH(16)) /* MSR_SPEC_CTRL used by Xen for PV */
 XEN_CPUFEATURE(SC_MSR_HVM,        X86_SYNTH(17)) /* MSR_SPEC_CTRL used by Xen for HVM */
 XEN_CPUFEATURE(SC_RSB_PV,         X86_SYNTH(18)) /* RSB overwrite needed for PV */
diff --git a/xen/include/asm-x86/nospec.h b/xen/include/asm-x86/nospec.h
index f6eb84eee554..5312ae4c6f31 100644
--- a/xen/include/asm-x86/nospec.h
+++ b/xen/include/asm-x86/nospec.h
@@ -10,7 +10,7 @@
 static always_inline bool barrier_nospec_true(void)
 {
 #ifdef CONFIG_SPECULATIVE_HARDEN_BRANCH
-    alternative("", "lfence", X86_FEATURE_SC_BRANCH_HARDEN);
+    alternative("lfence", "", X86_FEATURE_SC_NO_BRANCH_HARDEN);
 #endif
     return true;
 }
diff --git a/xen/include/asm-x86/spec_ctrl.h b/xen/include/asm-x86/spec_ctrl.h
index e671b6fd8d57..a803d16f9065 100644
--- a/xen/include/asm-x86/spec_ctrl.h
+++ b/xen/include/asm-x86/spec_ctrl.h
@@ -37,7 +37,6 @@ extern bool opt_ibpb;
 extern bool opt_ssbd;
 extern int8_t opt_eager_fpu;
 extern int8_t opt_l1d_flush;
-extern bool opt_branch_harden;
 
 extern bool bsp_delay_spec_ctrl;
 extern uint8_t default_xen_spec_ctrl;
-- 
2.11.0

On 05.10.2021 13:11, Andrew Cooper wrote:
> Branch Harden is enabled by default at compile and boot time.  Invert the code
> to compile with lfence by default and nop out in the non-default case.
> 
> This has several advantages.  It removes 3829 patch points (in the random
> build of Xen I have to hand) by default on boot, 70% (!) of the
> .altinstr_replacement section.  For builds of Xen with a non-nops capable tool
> chain, the code after `spec-ctrl=no-branch-harden` is better because Xen can
> write long nops.
> 
> Most importantly however, it means the disassembly actually matches what runs
> in the common case, with the ability to distinguish the lfences from other
> uses of nops.
> 
> Finally, make opt_branch_harden local to spec_ctrl.c and __initdata.  It has
> never been used externally, even at it's introduction in c/s 3860d5534df4
> "spec: add l1tf-barrier".
> 
> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>

Reviewed-by: Jan Beulich <jbeulich@suse.com>