[PATCH] ftrace: shut up -Wcast-function-type warning for ftrace_ops_no_ops

[PATCH] ftrace: shut up -Wcast-function-type warning for ftrace_ops_no_ops

[Arnd Bergmann]

From: Arnd Bergmann <arnd@arndb.de>

With 'make W=1', gcc warns about casts between incompatible function
types:

kernel/trace/ftrace.c:128:31: error: cast between incompatible function types from 'void (*)(long unsigned int,  long unsigned int)' to 'void (*)(long unsigned int,  long unsigned int,  struct ftrace_ops *, struct ftrace_regs *)' [-Werror=cast-function-type]
  128 | #define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
      |                               ^

As the commet here explains, this one was intentional, so shut up the
warning harder by using a double cast.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 kernel/trace/ftrace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index 4d8e35575549..d8fc87a17421 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -125,7 +125,7 @@ static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
 #else
 /* See comment below, where ftrace_ops_list_func is defined */
 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
-#define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
+#define ftrace_ops_list_func ((ftrace_func_t)(void *)ftrace_ops_no_ops)
 #endif
 
 static inline void ftrace_ops_init(struct ftrace_ops *ops)
-- 
2.29.2

Re: [PATCH] ftrace: shut up -Wcast-function-type warning for ftrace_ops_no_ops

[Steven Rostedt]

On Mon, 22 Mar 2021 22:49:58 +0100
Arnd Bergmann <arnd@kernel.org> wrote:

> From: Arnd Bergmann <arnd@arndb.de>
> 
> With 'make W=1', gcc warns about casts between incompatible function
> types:
> 
> kernel/trace/ftrace.c:128:31: error: cast between incompatible function types from 'void (*)(long unsigned int,  long unsigned int)' to 'void (*)(long unsigned int,  long unsigned int,  struct ftrace_ops *, struct ftrace_regs *)' [-Werror=cast-function-type]
>   128 | #define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
>       |                               ^
> 
> As the commet here explains, this one was intentional, so shut up the
> warning harder by using a double cast.

Bonus points for reading the comment ;-)

I'll take this patch for the next merge window, thanks!

-- Steve

Re: [PATCH] ftrace: shut up -Wcast-function-type warning for ftrace_ops_no_ops

[Peter Zijlstra]

On Mon, Mar 22, 2021 at 10:49:58PM +0100, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> With 'make W=1', gcc warns about casts between incompatible function
> types:
> 
> kernel/trace/ftrace.c:128:31: error: cast between incompatible function types from 'void (*)(long unsigned int,  long unsigned int)' to 'void (*)(long unsigned int,  long unsigned int,  struct ftrace_ops *, struct ftrace_regs *)' [-Werror=cast-function-type]
>   128 | #define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
>       |                               ^
> 
> As the commet here explains, this one was intentional, so shut up the
> warning harder by using a double cast.
> 
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  kernel/trace/ftrace.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> index 4d8e35575549..d8fc87a17421 100644
> --- a/kernel/trace/ftrace.c
> +++ b/kernel/trace/ftrace.c
> @@ -125,7 +125,7 @@ static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
>  #else
>  /* See comment below, where ftrace_ops_list_func is defined */
>  static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
> -#define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
> +#define ftrace_ops_list_func ((ftrace_func_t)(void *)ftrace_ops_no_ops)

So now we're making perfectly fine code worse, because GCC is stupid?

Please just kill the warning, like we do with so many other warnings.
It's broken!

Re: [PATCH] ftrace: shut up -Wcast-function-type warning for ftrace_ops_no_ops

[Steven Rostedt]

On Tue, 23 Mar 2021 08:39:08 +0100
Peter Zijlstra <peterz@infradead.org> wrote:

> > index 4d8e35575549..d8fc87a17421 100644
> > --- a/kernel/trace/ftrace.c
> > +++ b/kernel/trace/ftrace.c
> > @@ -125,7 +125,7 @@ static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
> >  #else
> >  /* See comment below, where ftrace_ops_list_func is defined */
> >  static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
> > -#define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
> > +#define ftrace_ops_list_func ((ftrace_func_t)(void *)ftrace_ops_no_ops)  
> 
> So now we're making perfectly fine code worse, because GCC is stupid?
> 
> Please just kill the warning, like we do with so many other warnings.
> It's broken!

[ Adding Kees ]

Well, from what I understand is that typecasting functions to other types
of functions is one of the mistakes that cause security bugs to happen. And
there's been a few of them in the kernel. Which is why the compilers have
now been flagging them.

This particular instance of the typecasting is unique and there's a bit of
magic involved. Adding this little work around here is fine. Probably needs
a better comment though.

-- Steve

Re: [PATCH] ftrace: shut up -Wcast-function-type warning for ftrace_ops_no_ops

[Peter Zijlstra]

On Tue, Mar 23, 2021 at 08:57:42AM -0400, Steven Rostedt wrote:

> Well, from what I understand is that typecasting functions to other types
> of functions is one of the mistakes that cause security bugs to happen. And
> there's been a few of them in the kernel. Which is why the compilers have
> now been flagging them.

Doing the cast in the first pace is an indication something weird is
happening anyway. C will already warn you if you assign non-matching
pointer types.

So if there's bugs, audit all (function) pointer casts. Don't wreck
casts.

This warning will only make people add more (void *) intermediate casts
and then we have uglier code and an equal amount of hidden bugs. IE. we
really didn't win anything at all.

> This particular instance of the typecasting is unique and there's a bit of
> magic involved. Adding this little work around here is fine. Probably needs
> a better comment though.

Nah, it's not unique, static_call() has a whole bunch of them, including
the exact same you have. Yes, we're doing something C thinks is UB, but
we do that all day every day.

Re: [PATCH] ftrace: shut up -Wcast-function-type warning for ftrace_ops_no_ops

[Peter Zijlstra]

On Tue, Mar 23, 2021 at 04:25:07PM +0100, Peter Zijlstra wrote:
> On Tue, Mar 23, 2021 at 08:57:42AM -0400, Steven Rostedt wrote:
> 
> > Well, from what I understand is that typecasting functions to other types
> > of functions is one of the mistakes that cause security bugs to happen. And
> > there's been a few of them in the kernel. Which is why the compilers have
> > now been flagging them.
> 
> Doing the cast in the first pace is an indication something weird is
> happening anyway. C will already warn you if you assign non-matching
> pointer types.
> 
> So if there's bugs, audit all (function) pointer casts. Don't wreck
> casts.
> 
> This warning will only make people add more (void *) intermediate casts
> and then we have uglier code and an equal amount of hidden bugs. IE. we
> really didn't win anything at all.
> 
> > This particular instance of the typecasting is unique and there's a bit of
> > magic involved. Adding this little work around here is fine. Probably needs
> > a better comment though.
> 
> Nah, it's not unique, static_call() has a whole bunch of them, including
> the exact same you have. Yes, we're doing something C thinks is UB, but
> we do that all day every day.

Here's weakening static_call_update() to get around this new
fail^Wfeature:

https://lkml.kernel.org/r/YFoN7nCl8OfGtpeh@hirez.programming.kicks-ass.net