Re: [PATCH] fgraph: record function return value

[Mark Rutland]

On Sat, Jan 12, 2019 at 02:57:01PM +0800, Changbin Du wrote:
> This patch adds a new trace option 'funcgraph-retval' and is disabled by
> default. When this option is enabled, fgraph tracer will show the return
> value of each function. This is useful to find/analyze a original error
> source in a call graph.
> 
> One limitation is that kernel doesn't know the prototype of functions. So
> fgraph assumes all functions have a retvalue of type int. You must ignore
> the value of *void* function. And if the retvalue looks like an error code
> then both hexadecimal and decimal number are displayed.

This sounds more confusing than helpful, and it sounds like this has
overlap with FTRACE_WITH_REGS functionality.

> diff --git a/arch/arm64/kernel/entry-ftrace.S b/arch/arm64/kernel/entry-ftrace.S
> index 81b8eb5c4633..223f4ad269d4 100644
> --- a/arch/arm64/kernel/entry-ftrace.S
> +++ b/arch/arm64/kernel/entry-ftrace.S
> @@ -202,6 +202,7 @@ ENTRY(return_to_handler)
>  	stp x4, x5, [sp, #32]
>  	stp x6, x7, [sp, #48]
>  
> +	mov	x1, x0			// return value
>  	mov	x0, x29			//     parent's fp
>  	bl	ftrace_return_to_handler// addr = ftrace_return_to_hander(fp);
>  	mov	x30, x0			// restore the original return address

What about indirect return values? Those go via x8.

Additionally, in some cases (e.g. static functions with cross-function
optimization), the compiler might not follow the usual PCS, so the
return value might not be in x0 regardless. Maybe such functions aren't
hooked by ftrace today?

Generally, I don't think that this is going to be reliable.

> +config HAVE_FTRACE_RETVAL
> +	bool
> +
>  config HAVE_DYNAMIC_FTRACE
>  	bool
>  	help
> @@ -160,6 +163,7 @@ config FUNCTION_GRAPH_TRACER
>  	depends on HAVE_FUNCTION_GRAPH_TRACER
>  	depends on FUNCTION_TRACER
>  	depends on !X86_32 || !CC_OPTIMIZE_FOR_SIZE
> +	select HAVE_FTRACE_RETVAL if (X86 || ARM)

... but not arm64?

Thanks,
Mark.