[PATCH v3 12/16] x86/alternative: Try inline spectre_v2=retpoline,amd

[Peter Zijlstra]

Try and replace retpoline thunk calls with:

  LFENCE
  CALL    *%\reg

for spectre_v2=retpoline,amd.

Specifically, the sequence above is 5 bytes for the low 8 registers,
but 6 bytes for the high 8 registers. This means that unless the
compilers prefix stuff the call with higher registers this replacement
will fail.

Luckily GCC strongly favours RAX for the indirect calls and most (95%+
for defconfig-x86_64) will be converted. OTOH clang strongly favours
R11 and almost nothing gets converted.

Note: it will also generate a correct replacement for the Jcc.d32
case, except unless the compilers start to prefix stuff that, it'll
never fit. Specifically:

  Jncc.d8 1f
  LFENCE
  JMP     *%\reg
1:

is 7-8 bytes long, where the original instruction in unpadded form is
only 6 bytes.

Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
---
 arch/x86/kernel/alternative.c |   16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

--- a/arch/x86/kernel/alternative.c
+++ b/arch/x86/kernel/alternative.c
@@ -389,6 +389,7 @@ static int emit_indirect(int op, int reg
  *
  *   CALL *%\reg
  *
+ * It also tries to inline spectre_v2=retpoline,amd when size permits.
  */
 static int patch_retpoline(void *addr, struct insn *insn, u8 *bytes)
 {
@@ -405,7 +406,8 @@ static int patch_retpoline(void *addr, s
 	/* If anyone ever does: CALL/JMP *%rsp, we're in deep trouble. */
 	BUG_ON(reg == 4);
 
-	if (cpu_feature_enabled(X86_FEATURE_RETPOLINE))
+	if (cpu_feature_enabled(X86_FEATURE_RETPOLINE) &&
+	    !cpu_feature_enabled(X86_FEATURE_RETPOLINE_AMD))
 		return -1;
 
 	op = insn->opcode.bytes[0];
@@ -418,8 +420,9 @@ static int patch_retpoline(void *addr, s
 	 * into:
 	 *
 	 *   Jncc.d8 1f
+	 *   [ LFENCE ]
 	 *   JMP *%\reg
-	 *   NOP
+	 *   [ NOP ]
 	 * 1:
 	 */
 	if (op == 0x0f && (insn->opcode.bytes[1] & 0xf0) == 0x80) {
@@ -432,6 +435,15 @@ static int patch_retpoline(void *addr, s
 		op = JMP32_INSN_OPCODE;
 	}
 
+	/*
+	 * For RETPOLINE_AMD: prepend the indirect CALL/JMP with an LFENCE.
+	 */
+	if (cpu_feature_enabled(X86_FEATURE_RETPOLINE_AMD)) {
+		bytes[i++] = 0x0f;
+		bytes[i++] = 0xae;
+		bytes[i++] = 0xe8; /* LFENCE */
+	}
+
 	ret = emit_indirect(op, reg, bytes + i);
 	if (ret < 0)
 		return ret;