From: "Borislav Petkov (AMD)" <bp@alien8.de>
Date: Thu, 28 Mar 2024 13:59:05 +0100
Subject: x86/bugs: Fix the SRSO mitigation on Zen3/4
Git-commit: 4535e1a4174c4111d92c5a9a21e542d232e0fcaa
Patch-mainline: v6.9-rc2
References: git-fixes

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 e7c25c441e9e ("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: e7c25c441e9e ("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>
Acked-by: Nikolay Borisov <nik.borisov@suse.com>
---
 arch/x86/include/asm/asm-prototypes.h |    1 +
 arch/x86/include/asm/nospec-branch.h  |   22 +++++++++++++++++++---
 arch/x86/lib/retpoline.S              |    4 +---
 3 files changed, 21 insertions(+), 6 deletions(-)

--- a/arch/x86/include/asm/asm-prototypes.h
+++ b/arch/x86/include/asm/asm-prototypes.h
@@ -11,6 +11,7 @@
 #include <asm/special_insns.h>
 #include <asm/preempt.h>
 #include <asm/asm.h>
+#include <asm/nospec-branch.h>
 
 #ifndef CONFIG_X86_CMPXCHG64
 extern void cmpxchg8b_emu(void);
--- a/arch/x86/include/asm/nospec-branch.h
+++ b/arch/x86/include/asm/nospec-branch.h
@@ -169,6 +169,22 @@
 #endif
 .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)
+	ALTERNATIVE_2 "", "call entry_untrain_ret", X86_FEATURE_UNRET, \
+		          "call srso_alias_untrain_ret", X86_FEATURE_SRSO_ALIAS
+ #endif
+.endm
 /*
  * Mitigate RETBleed for AMD/Hygon Zen uarch. Requires KERNEL CR3 because the
  * return thunk isn't mapped into the userspace tables (then again, AMD
@@ -182,9 +198,8 @@
  */
 .macro UNTRAIN_RET
 #ifdef CONFIG_RETPOLINE
-	ALTERNATIVE_2 "",						\
-	              "call entry_untrain_ret", X86_FEATURE_UNRET,	\
-		      "call entry_ibpb", X86_FEATURE_ENTRY_IBPB
+	CALL_UNTRAIN_RET
+	ALTERNATIVE "" ,"call entry_ibpb", X86_FEATURE_ENTRY_IBPB
 #endif
 .endm
 
@@ -234,6 +249,7 @@
 #ifdef CONFIG_RETPOLINE
 #ifdef CONFIG_X86_64
 
+
 extern void __x86_return_thunk(void);
 extern void entry_untrain_ret(void);
 extern void srso_untrain_ret(void);
--- a/arch/x86/lib/retpoline.S
+++ b/arch/x86/lib/retpoline.S
@@ -209,9 +209,7 @@ __EXPORT_THUNK(srso_untrain_ret)
 
 
 ENTRY(entry_untrain_ret)
-	ALTERNATIVE_2 "jmp retbleed_untrain_ret", \
-		      "jmp srso_untrain_ret", X86_FEATURE_SRSO, \
-		      "jmp srso_alias_untrain_ret", X86_FEATURE_SRSO_ALIAS
+	ALTERNATIVE "jmp retbleed_untrain_ret", "jmp srso_untrain_ret", X86_FEATURE_SRSO
 ENDPROC(entry_untrain_ret)
 __EXPORT_THUNK(entry_untrain_ret)