Re: [PATCH 5/8] ftrace: Make DIRECT_CALLS work WITH_ARGS and !WITH_REGS

[Mark Rutland]

On Wed, Feb 01, 2023 at 05:34:17PM +0100, Florent Revest wrote:
> Direct called trampolines can be called in two ways:
> - either from the ftrace callsite. In this case, they do not access any
>   struct ftrace_regs nor pt_regs
> - Or, if a ftrace ops is also attached, from the end of a ftrace
>   trampoline. In this case, the call_direct_funcs ops is in charge of
>   setting the direct call trampoline's address in a struct ftrace_regs
> 
> Since "ftrace: pass fregs to arch_ftrace_set_direct_caller()", the later
> case no longer requires a full pt_regs.

Minor nit, but could we please have the commit ID, e.g.

| Since commit:
| 
|   9705bc70960459ae ("ftrace: pass fregs to arch_ftrace_set_direct_caller()")
| 
| The latter case no longer requires a full pt_regs.

> It only needs a struct
> ftrace_regs so DIRECT_CALLS can work with both WITH_ARGS or WITH_REGS.
> With architectures like arm64 already abandoning WITH_REGS in favor of
> WITH_ARGS, it's important to have DIRECT_CALLS work WITH_ARGS only.
> 
> Signed-off-by: Florent Revest <revest@xxxxxxxxxxxx>
> ---
>  kernel/trace/Kconfig  | 2 +-
>  kernel/trace/ftrace.c | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
> index 5df427a2321d..4496a7c69810 100644
> --- a/kernel/trace/Kconfig
> +++ b/kernel/trace/Kconfig
> @@ -257,7 +257,7 @@ config DYNAMIC_FTRACE_WITH_REGS
>  
>  config DYNAMIC_FTRACE_WITH_DIRECT_CALLS
>  	def_bool y
> -	depends on DYNAMIC_FTRACE_WITH_REGS
> +	depends on DYNAMIC_FTRACE_WITH_REGS || DYNAMIC_FTRACE_WITH_ARGS
>  	depends on HAVE_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
>  
>  config DYNAMIC_FTRACE_WITH_CALL_OPS
> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> index b0426de11c45..73b6f6489ba1 100644
> --- a/kernel/trace/ftrace.c
> +++ b/kernel/trace/ftrace.c
> @@ -5282,7 +5282,7 @@ static LIST_HEAD(ftrace_direct_funcs);
>  
>  static int register_ftrace_function_nolock(struct ftrace_ops *ops);
>  
> -#define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_REGS)
> +#define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT)

Unfortunately, I think this is broken for architectures where:

* DYNAMIC_FTRACE_WITH_DIRECT_CALLS=y
* DYNAMIC_FTRACE_WITH_REGS=y
* DYNAMIC_FTRACE_WITH_ARGS=n

... since those might pass a NULL ftrace_regs around, and so when using the
list ops arch_ftrace_set_direct_caller() might blow up accessing an element of
ftrace_regs.

It looks like 32-bit x86 is the only case with that combination, and its
ftrace_caller implementation passes a NULL regs, so I reckon that'll blow up.
However, it looks like there aren't any FTRACE_DIRECT samples wired up for
32-bit x86, so I'm not aware of a test case we can use.

We could make the FTRACE_OPS_FL_SAVE_REGS flag conditional, or we could
implement DYNAMIC_FTRACE_WITH_ARGS for 32-bit x86 and have
DYNAMIC_FTRACE_WITH_DIRECT_CALLS depend solely on that.

Thanks,
Mark.