Re: [PATCH v5 04/17] x86,ftrace: Fix ftrace_regs_caller() unwind

[Steven Rostedt]

Hi Peter,

After looking at the code, I realized that the only possible way to do
the "direct call" part, is if the direct function helper is executed
there. All other ftrace_ops, will not call that path.

I added a trampoline that points to ftrace_regs_caller to the direct
ftrace_ops, to force it never to allocate its own trampoline, and thus
no created trampoline should ever do the jump for a direct caller.

By doing this, I was able to move the code around to be a bit simpler,
and not need the double modifications (the double ret).

Of course, if any created trampoline were to touch the ORIG_RAX, then
it would crash. We could always nop that compare in the created
trampoline if that is a concern.

Anyway, I added the below patch on top of your patches and it appears
to work. Thoughts?

-- Steve

diff --git a/arch/x86/kernel/ftrace.c b/arch/x86/kernel/ftrace.c
index 867c126ddabe..2e11250d5647 100644
--- a/arch/x86/kernel/ftrace.c
+++ b/arch/x86/kernel/ftrace.c
@@ -367,13 +367,6 @@ create_trampoline(struct ftrace_ops *ops, unsigned int *tramp_size)
 	if (WARN_ON(ret < 0))
 		goto fail;
 
-	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
-		ip = trampoline + (ftrace_regs_caller_ret - ftrace_regs_caller);
-		ret = probe_kernel_read(ip, (void *)retq, RET_SIZE);
-		if (WARN_ON(ret < 0))
-			goto fail;
-	}
-
 	/*
 	 * The address of the ftrace_ops that is used for this trampoline
 	 * is stored at the end of the trampoline. This will be used to
diff --git a/arch/x86/kernel/ftrace_64.S b/arch/x86/kernel/ftrace_64.S
index 9738ed23964e..1f2afaa8f71f 100644
--- a/arch/x86/kernel/ftrace_64.S
+++ b/arch/x86/kernel/ftrace_64.S
@@ -241,22 +241,9 @@ SYM_INNER_LABEL(ftrace_regs_call, SYM_L_GLOBAL)
 	 */
 	movq ORIG_RAX(%rsp), %rax
 	testq	%rax, %rax
-	jz	1f
+	jnz	1f
 
-	/* Swap the flags with orig_rax */
-	movq MCOUNT_REG_SIZE(%rsp), %rdi
-	movq %rdi, MCOUNT_REG_SIZE-8(%rsp)
-	movq %rax, MCOUNT_REG_SIZE(%rsp)
-
-	restore_mcount_regs 8
-	/* Restore flags */
-	popfq
-
-SYM_INNER_LABEL(ftrace_regs_caller_ret, SYM_L_GLOBAL);
-	UNWIND_HINT_RET_OFFSET
-	jmp	ftrace_epilogue
-
-1:	restore_mcount_regs
+	restore_mcount_regs
 	/* Restore flags */
 	popfq
 
@@ -267,8 +254,19 @@ SYM_INNER_LABEL(ftrace_regs_caller_ret, SYM_L_GLOBAL);
 	 * to the return.
 	 */
 SYM_INNER_LABEL(ftrace_regs_caller_end, SYM_L_GLOBAL)
+	UNWIND_HINT_RET_OFFSET
 	jmp ftrace_epilogue
 
+	/* Swap the flags with orig_rax */
+1:	movq MCOUNT_REG_SIZE(%rsp), %rdi
+	movq %rdi, MCOUNT_REG_SIZE-8(%rsp)
+	movq %rax, MCOUNT_REG_SIZE(%rsp)
+
+	restore_mcount_regs 8
+	/* Restore flags */
+	popfq
+	jmp	ftrace_epilogue
+
 SYM_FUNC_END(ftrace_regs_caller)
 
 
diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index 041694a1eb74..8f540eef7511 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -2399,6 +2399,13 @@ struct ftrace_ops direct_ops = {
 	.flags		= FTRACE_OPS_FL_IPMODIFY | FTRACE_OPS_FL_RECURSION_SAFE
 			  | FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_REGS
 			  | FTRACE_OPS_FL_PERMANENT,
+	/*
+	 * By declaring the main trampoline as this trampoline
+	 * it will never have one allocated for it. This is fine
+	 * as this should only be used if we are in the
+	 * ftrace_ops_list function.
+	 */
+	.trampoline	= FTRACE_REGS_ADDR,
 };
 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */