x86/bugs: Fix the SRSO mitigation on Zen3/4

The original version of the mitigation would patch in the calls to the
untraining routines directly.  That is, the alternative() in UNTRAIN_RET
will patch in the CALL to srso_alias_untrain_ret() directly.

However, even if commit e7c25c441e ("x86/cpu: Cleanup the untrain
mess") meant well in trying to clean up the situation, due to micro-
architectural reasons, the untraining routine srso_alias_untrain_ret()
must be the target of a CALL instruction and not of a JMP instruction as
it is done now.

Reshuffle the alternative macros to accomplish that.

Fixes: e7c25c441e ("x86/cpu: Cleanup the untrain mess")
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arch/x86/include/asm/asm-prototypes.h

@@ -14,6 +14,7 @@
 #include <asm/asm.h>
 #include <asm/fred.h>
 #include <asm/gsseg.h>
+#include <asm/nospec-branch.h>
 
 #ifndef CONFIG_X86_CMPXCHG64
 extern void cmpxchg8b_emu(void);

arch/x86/include/asm/nospec-branch.h

@@ -262,11 +262,20 @@
 .Lskip_rsb_\@:
 .endm
 
+/*
+ * The CALL to srso_alias_untrain_ret() must be patched in directly at
+ * the spot where untraining must be done, ie., srso_alias_untrain_ret()
+ * must be the target of a CALL instruction instead of indirectly
+ * jumping to a wrapper which then calls it. Therefore, this macro is
+ * called outside of __UNTRAIN_RET below, for the time being, before the
+ * kernel can support nested alternatives with arbitrary nesting.
+ */
+.macro CALL_UNTRAIN_RET
 #if defined(CONFIG_MITIGATION_UNRET_ENTRY) || defined(CONFIG_MITIGATION_SRSO)
-#define CALL_UNTRAIN_RET	"call entry_untrain_ret"
+	ALTERNATIVE_2 "", "call entry_untrain_ret", X86_FEATURE_UNRET, \
-#else
+		          "call srso_alias_untrain_ret", X86_FEATURE_SRSO_ALIAS
-#define CALL_UNTRAIN_RET	""
 #endif
+.endm
 
 /*
  * Mitigate RETBleed for AMD/Hygon Zen uarch. Requires KERNEL CR3 because the
@@ -282,8 +291,8 @@
 .macro __UNTRAIN_RET ibpb_feature, call_depth_insns
 #if defined(CONFIG_MITIGATION_RETHUNK) || defined(CONFIG_MITIGATION_IBPB_ENTRY)
 	VALIDATE_UNRET_END
-	ALTERNATIVE_3 "",						\
+	CALL_UNTRAIN_RET
-		      CALL_UNTRAIN_RET, X86_FEATURE_UNRET,		\
+	ALTERNATIVE_2 "",						\
 		      "call entry_ibpb", \ibpb_feature,			\
 		     __stringify(\call_depth_insns), X86_FEATURE_CALL_DEPTH
 #endif
@@ -342,6 +351,8 @@ extern void retbleed_return_thunk(void);
 static inline void retbleed_return_thunk(void) {}
 #endif
 
+extern void srso_alias_untrain_ret(void);
+
 #ifdef CONFIG_MITIGATION_SRSO
 extern void srso_return_thunk(void);
 extern void srso_alias_return_thunk(void);

arch/x86/lib/retpoline.S

@@ -163,6 +163,7 @@ SYM_CODE_START_NOALIGN(srso_alias_untrain_ret)
 	lfence
 	jmp srso_alias_return_thunk
 SYM_FUNC_END(srso_alias_untrain_ret)
+__EXPORT_THUNK(srso_alias_untrain_ret)
 	.popsection
 
 	.pushsection .text..__x86.rethunk_safe
@@ -224,10 +225,12 @@ SYM_CODE_START(srso_return_thunk)
 SYM_CODE_END(srso_return_thunk)
 
 #define JMP_SRSO_UNTRAIN_RET "jmp srso_untrain_ret"
-#define JMP_SRSO_ALIAS_UNTRAIN_RET "jmp srso_alias_untrain_ret"
 #else /* !CONFIG_MITIGATION_SRSO */
+/* Dummy for the alternative in CALL_UNTRAIN_RET. */
+SYM_CODE_START(srso_alias_untrain_ret)
+	RET
+SYM_FUNC_END(srso_alias_untrain_ret)
 #define JMP_SRSO_UNTRAIN_RET "ud2"
-#define JMP_SRSO_ALIAS_UNTRAIN_RET "ud2"
 #endif /* CONFIG_MITIGATION_SRSO */
 
 #ifdef CONFIG_MITIGATION_UNRET_ENTRY
@@ -319,9 +322,7 @@ SYM_FUNC_END(retbleed_untrain_ret)
 #if defined(CONFIG_MITIGATION_UNRET_ENTRY) || defined(CONFIG_MITIGATION_SRSO)
 
 SYM_FUNC_START(entry_untrain_ret)
-	ALTERNATIVE_2 JMP_RETBLEED_UNTRAIN_RET,				\
+	ALTERNATIVE JMP_RETBLEED_UNTRAIN_RET, JMP_SRSO_UNTRAIN_RET, X86_FEATURE_SRSO
-		      JMP_SRSO_UNTRAIN_RET, X86_FEATURE_SRSO,		\
-		      JMP_SRSO_ALIAS_UNTRAIN_RET, X86_FEATURE_SRSO_ALIAS
 SYM_FUNC_END(entry_untrain_ret)
 __EXPORT_THUNK(entry_untrain_ret)