[PATCH v3] arm64: fix unwind_frame() for filtered out fn for function graph tracing

On 16/01/18 18:57, Catalin Marinas wrote:

On Fri, Jan 12, 2018 at 11:48:32AM +0100, Jerome Marchand wrote:

quoted

diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
index 76809ccd309c..5a528c58ef68 100644
--- a/arch/arm64/kernel/stacktrace.c
+++ b/arch/arm64/kernel/stacktrace.c

@@ -59,6 +59,10 @@ int notrace unwind_frame(struct task_struct *tsk, struct stackframe *frame)
 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 	if (tsk->ret_stack &&
 			(frame->pc == (unsigned long)return_to_handler)) {
+		WARN_ON(frame->graph == -1);
+		if (frame->graph < -1)
+			frame->graph += FTRACE_NOTRACE_DEPTH;
+
 		/*
 		 * This is a case where function graph tracer has
 		 * modified a return address (LR) in a stack frame

So do we still allow this to continue if graph == -1? The following line
doesn't seem safe:

		frame->pc = tsk->ret_stack[frame->graph--].ret;

You're right. We probably should return a error (-EINVAL I guess) if
this happens. Note that this shouldn't happen here and if we're
confident enough that profile_pc() was the only faulty caller, we could
just drop the warning.

Jerome

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: OpenPGP digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20180117/2bb52dbe/attachment.sig>