Re: [PATCH v3 1/4] DYNAMIC_FTRACE configurable with and without REGS

[Ard Biesheuvel]

Hi Torsten,

On 1 October 2018 at 16:16, Torsten Duwe <duwe@xxxxxx> wrote:
> In commit 06aeaaeabf69da4, many ftrace-related config options are
> consolidated. By accident, I guess, the choice about DYNAMIC_FTRACE
> and DYNAMIC_FTRACE_WITH_REGS is no longer available explicitly but
> determined by the sole availability on the architecture.
>
> This makes it hard to introduce DYNAMIC_FTRACE_WITH_REGS if it depends
> on new compiler features or other new properties of the toolchain
> without breaking existing configurations.
>
> This patch turns the def_bool into an actual choice. Should the toolchain
> not meet the requirements for _WITH_REGS it can be turned off.
>

I guess we now have Kbuild/Kconfig support for this, no? I mean, we
can now show/hide options depending on the capabilities of the
toolchain.

I am not saying it would be a better approach, though - I'd rather
have a warning than have things silently disabled, but other people
may think differently.


> Signed-off-by: Torsten Duwe <duwe@xxxxxxx>
>
>
> --- a/kernel/trace/Kconfig
> +++ b/kernel/trace/Kconfig
> @@ -508,9 +508,15 @@ config DYNAMIC_FTRACE
>           otherwise has native performance as long as no tracing is active.
>
>  config DYNAMIC_FTRACE_WITH_REGS
> -       def_bool y
> +       bool "Include register content tracking in dynamic ftrace facility"
> +       default y
>         depends on DYNAMIC_FTRACE
>         depends on HAVE_DYNAMIC_FTRACE_WITH_REGS
> +       help
> +         This architecture supports the inspection of register contents,
> +         as passed between functions, at the dynamic ftrace points.
> +         This is also a prerequisite for Kernel Live Patching (KLP).
> +         When in doubt, say Y.
>
>  config FUNCTION_PROFILER
>         bool "Kernel function profiler"