x86/ibt,ftrace: Make function-graph play nice

Return trampoline must not use indirect branch to return; while this
preserves the RSB, it is fundamentally incompatible with IBT. Instead
use a retpoline like ROP gadget that defeats IBT while not unbalancing
the RSB.

And since ftrace_stub is no longer a plain RET, don't use it to copy
from. Since RET is a trivial instruction, poke it directly.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Link: https://lore.kernel.org/r/20220308154318.347296408@infradead.org

arch/x86/kernel/ftrace.c

@@ -316,12 +316,12 @@ create_trampoline(struct ftrace_ops *ops, unsigned int *tramp_size)
 	unsigned long offset;
 	unsigned long npages;
 	unsigned long size;
-	unsigned long retq;
 	unsigned long *ptr;
 	void *trampoline;
 	void *ip;
 	/* 48 8b 15 <offset> is movq <offset>(%rip), %rdx */
 	unsigned const char op_ref[] = { 0x48, 0x8b, 0x15 };
+	unsigned const char retq[] = { RET_INSN_OPCODE, INT3_INSN_OPCODE };
 	union ftrace_op_code_union op_ptr;
 	int ret;
 
@@ -359,12 +359,7 @@ create_trampoline(struct ftrace_ops *ops, unsigned int *tramp_size)
 		goto fail;
 
 	ip = trampoline + size;
-
-	/* The trampoline ends with ret(q) */
-	retq = (unsigned long)ftrace_stub;
-	ret = copy_from_kernel_nofault(ip, (void *)retq, RET_SIZE);
-	if (WARN_ON(ret < 0))
-		goto fail;
+	memcpy(ip, retq, RET_SIZE);
 
 	/* No need to test direct calls on created trampolines */
 	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {

arch/x86/kernel/ftrace_64.S

@@ -176,10 +176,10 @@ SYM_FUNC_END(ftrace_caller);
 SYM_FUNC_START(ftrace_epilogue)
 /*
  * This is weak to keep gas from relaxing the jumps.
- * It is also used to copy the RET for trampolines.
  */
 SYM_INNER_LABEL_ALIGN(ftrace_stub, SYM_L_WEAK)
 	UNWIND_HINT_FUNC
+	ENDBR
 	RET
 SYM_FUNC_END(ftrace_epilogue)
 
@@ -284,6 +284,7 @@ SYM_FUNC_START(__fentry__)
 	jnz trace
 
 SYM_INNER_LABEL(ftrace_stub, SYM_L_GLOBAL)
+	ENDBR
 	RET
 
 trace:
@@ -307,7 +308,7 @@ EXPORT_SYMBOL(__fentry__)
 
 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 SYM_FUNC_START(return_to_handler)
-	subq  $24, %rsp
+	subq  $16, %rsp
 
 	/* Save the return values */
 	movq %rax, (%rsp)
@@ -319,7 +320,19 @@ SYM_FUNC_START(return_to_handler)
 	movq %rax, %rdi
 	movq 8(%rsp), %rdx
 	movq (%rsp), %rax
-	addq $24, %rsp
-	JMP_NOSPEC rdi
+
+	addq $16, %rsp
+	/*
+	 * Jump back to the old return address. This cannot be JMP_NOSPEC rdi
+	 * since IBT would demand that contain ENDBR, which simply isn't so for
+	 * return addresses. Use a retpoline here to keep the RSB balanced.
+	 */
+	ANNOTATE_INTRA_FUNCTION_CALL
+	call .Ldo_rop
+	int3
+.Ldo_rop:
+	mov %rdi, (%rsp)
+	UNWIND_HINT_FUNC
+	RET
 SYM_FUNC_END(return_to_handler)
 #endif