arm64: stacktrace: WARN_ON(state->common.pc == orig_pc) in kunwind_recover_return_address()

arm64: stacktrace: WARN_ON(state->common.pc == orig_pc) in kunwind_recover_return_address()

[Oleg Nesterov]

I don't know the details but people report that
kunwind_recover_return_address() hits this warning.

I know absolutely nothing about arm64 or kernel/trace/fgraph.c, but at
first glance this code looks broken:

	kunwind_recover_return_address:

		...

		orig_pc = ftrace_graph_ret_addr(state->task, NULL,
						state->common.pc,
						(void *)state->common.fp);
		if (WARN_ON_ONCE(state->common.pc == orig_pc))

	ftrace_graph_ret_addr:

		...

		if (!idx)
			return ret;


given that kunwind_recover_return_address() passes idx == NULL to
ftrace_graph_ret_addr(), it should always return ret == state->common.pc ?


Perhaps this connects to 29c1c24a27 ("function_graph: Fix up ftrace_graph_ret_addr()")
and I have no idea if something like the patch below makes any sense.

Oleg.

--- a/arch/arm64/kernel/stacktrace.c
+++ b/arch/arm64/kernel/stacktrace.c
@@ -106,7 +106,8 @@ kunwind_recover_return_address(struct kunwind_state *state)
 	if (state->task->ret_stack &&
 	    (state->common.pc == (unsigned long)return_to_handler)) {
 		unsigned long orig_pc;
-		orig_pc = ftrace_graph_ret_addr(state->task, NULL,
+		orig_pc = ftrace_graph_ret_addr(state->task,
+						&state->task->curr_ret_stack,
 						state->common.pc,
 						(void *)state->common.fp);
 		if (WARN_ON_ONCE(state->common.pc == orig_pc))

Re: arm64: stacktrace: WARN_ON(state->common.pc == orig_pc) in kunwind_recover_return_address()

[Puranjay Mohan]

[-- Attachment #1: Type: text/plain, Size: 1106 bytes --]

Oleg Nesterov <oleg@redhat.com> writes:

> I don't know the details but people report that
> kunwind_recover_return_address() hits this warning.
>
> I know absolutely nothing about arm64 or kernel/trace/fgraph.c, but at
> first glance this code looks broken:
>
> 	kunwind_recover_return_address:
>
> 		...
>
> 		orig_pc = ftrace_graph_ret_addr(state->task, NULL,
> 						state->common.pc,
> 						(void *)state->common.fp);
> 		if (WARN_ON_ONCE(state->common.pc == orig_pc))
>
> 	ftrace_graph_ret_addr:
>
> 		...
>
> 		if (!idx)
> 			return ret;
>
>
> given that kunwind_recover_return_address() passes idx == NULL to
> ftrace_graph_ret_addr(), it should always return ret == state->common.pc ?
>
>
> Perhaps this connects to 29c1c24a27 ("function_graph: Fix up ftrace_graph_ret_addr()")
> and I have no idea if something like the patch below makes any sense.
>

Your finding is accurate.

It has been fixed in upstream by c060f93253ca ("arm64: stacktrace: fix
the usage of ftrace_graph_ret_addr()")

See: https://elixir.bootlin.com/linux/v6.11/source/arch/arm64/kernel/stacktrace.c#L111

Thanks,
Puranjay

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 255 bytes --]

Re: arm64: stacktrace: WARN_ON(state->common.pc == orig_pc) in kunwind_recover_return_address()

[Oleg Nesterov]

Ah, sorry for the noise, I looked into the wrong tree, missed the recent
c060f93253ca ("arm64: stacktrace: fix the usage of ftrace_graph_ret_addr()")

On 09/16, Oleg Nesterov wrote:
>
> I don't know the details but people report that
> kunwind_recover_return_address() hits this warning.
>
> I know absolutely nothing about arm64 or kernel/trace/fgraph.c, but at
> first glance this code looks broken:
>
> 	kunwind_recover_return_address:
>
> 		...
>
> 		orig_pc = ftrace_graph_ret_addr(state->task, NULL,
> 						state->common.pc,
> 						(void *)state->common.fp);
> 		if (WARN_ON_ONCE(state->common.pc == orig_pc))
>
> 	ftrace_graph_ret_addr:
>
> 		...
>
> 		if (!idx)
> 			return ret;
>
>
> given that kunwind_recover_return_address() passes idx == NULL to
> ftrace_graph_ret_addr(), it should always return ret == state->common.pc ?
>
>
> Perhaps this connects to 29c1c24a27 ("function_graph: Fix up ftrace_graph_ret_addr()")
> and I have no idea if something like the patch below makes any sense.
>
> Oleg.
>
> --- a/arch/arm64/kernel/stacktrace.c
> +++ b/arch/arm64/kernel/stacktrace.c
> @@ -106,7 +106,8 @@ kunwind_recover_return_address(struct kunwind_state *state)
>  	if (state->task->ret_stack &&
>  	    (state->common.pc == (unsigned long)return_to_handler)) {
>  		unsigned long orig_pc;
> -		orig_pc = ftrace_graph_ret_addr(state->task, NULL,
> +		orig_pc = ftrace_graph_ret_addr(state->task,
> +						&state->task->curr_ret_stack,
>  						state->common.pc,
>  						(void *)state->common.fp);
>  		if (WARN_ON_ONCE(state->common.pc == orig_pc))

Re: arm64: stacktrace: WARN_ON(state->common.pc == orig_pc) in kunwind_recover_return_address()

[Oleg Nesterov]

On 09/16, Puranjay Mohan wrote:
>
> Oleg Nesterov <oleg@redhat.com> writes:
>
> > given that kunwind_recover_return_address() passes idx == NULL to
> > ftrace_graph_ret_addr(), it should always return ret == state->common.pc ?
> >
> >
> > Perhaps this connects to 29c1c24a27 ("function_graph: Fix up ftrace_graph_ret_addr()")
> > and I have no idea if something like the patch below makes any sense.
> >
>
> Your finding is accurate.
>
> It has been fixed in upstream by c060f93253ca ("arm64: stacktrace: fix
> the usage of ftrace_graph_ret_addr()")

Yes, I have already found that commit, see my next email ;)

Nevertheless, thanks a lot Puranjay for the confirmation!

Oleg.