Message-ID: <CAK7LNATO+L4-0SpuWrCnYEUVbxmcQkjOQvMB6PEsFv1b-heJOA@mail.gmail.com>
Date:   Wed, 17 Oct 2018 13:48:46 +0900
From:   Masahiro Yamada <yamada.masahiro@...ionext.com>
To:     Nathan Chancellor <natechancellor@...il.com>
Cc:     Michal Marek <michal.lkml@...kovi.net>,
        Linux Kbuild mailing list <linux-kbuild@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Nick Desaulniers <ndesaulniers@...gle.com>,
        Kees Cook <keescook@...omium.org>
Subject: Re: [PATCH] Kbuild: Hide Clang's -Wempty-body behind W=1

Hi Nathan,


On Tue, Oct 16, 2018 at 11:15 AM Nathan Chancellor
<natechancellor@...il.com> wrote:
>
> There are only a few instances of this warning in an arm64 allyesconfig
> build but none of them appear useful. I believe the intention of the
> warning is to avoid situations like this:
>
> if (condition);
>         statement;
>
> where the user really intended
>
> if (condition)
>         statement;
>
> However, these instances have already been caught by GCC's warning about
> misleading indentation


Right, the example above is caught by -Wmisleading-indentation.

However, the following is not.

   if (condition)
          ;



So, -Wempty-body is a kind of different thing,
and still useful in my opinion.



> so the remaining warnings are about loops that
> fall into one of three categories:
>
> 1. Execute a function unconditionally (avoiding a useless variable to
>    hold the return value):
>
> drivers/isdn/hisax/hfc_pci.c:131:34: warning: if statement has empty body
> [-Wempty-body]
>         if (Read_hfc(cs, HFCPCI_INT_S1));


I think this is a real bug,
then -Wempty-body finally caught it.
(but -Wmisleading-indentation cannot catch it.)



It is wrong to enclose a non-effective statement with 'if ();'
just for suppressing another warning.



     Read_hfc(cs, HFCPCI_INT_S1);

would emit this warning.


In file included from drivers/isdn/hisax/hfc_pci.c:20:0:
drivers/isdn/hisax/hfc_pci.c: In function ‘reset_hfcpci’:
drivers/isdn/hisax/hfc_pci.h:232:25: warning: statement with no effect
[-Wunused-value]
 #define Read_hfc(a, b) (*(((u_char *)a->hw.hfcpci.pci_io) + b))
                        ~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/isdn/hisax/hfc_pci.c:131:2: note: in expansion of macro ‘Read_hfc’
  Read_hfc(cs, HFCPCI_INT_S1);
  ^~~~~~~~



The root cause is missing 'volatile'
while Read_hfc() is supposed to read out a HW register.



#define Read_hfc(a, b) (*(((volatile u_char *)a->hw.hfcpci.pci_io) + b))

will be a correct fix.
(or just use a standard accessor like readb(), ioread8(), etc.)




if (Read_hfc(cs, HFCPCI_INT_S1));

is optimized out by the compiler, so it is not working as expected.



>
> 2. Advancing a value to be used later on in the function like a pointer
>    or a count:
>
> drivers/atm/eni.c:244:48: warning: for loop has empty body
> [-Wempty-body]
>         for (order = 0; (1 << order) < *size; order++);
>                                                       ^

As you noted in the commit log,
Clang's -Wempty-body cares the location of a semi-colon,
while GCC's one does not.





   for (order = 0; (1 << order) < *size; order++)
             ;

is fine, and more readable in my opinion.




> 3. Busy waiting:
>
> drivers/atm/zatm.c:513:7: warning: while loop has empty body
> [-Wempty-body]
>         zwait;
>              ^


Again, Clang is fine with an empty body in while() loop,
but just picky about the semi-colon location.

For this particular case, how about something like this?


#define zwait  do {} while (zin(CMR) & uPD98401_BUSY)





I think an even better fix is

#define zwait()  do {} while (zin(CMR) & uPD98401_BUSY)



then, fix-up all

   zwait;

to

   zwait();






> None of these uses are problematic or need to be addressed.


The first pattern is really problematic, and need to be addressed.

I want to keep -Wempty-body enabled
to find out potential issues.

Please let me know if you see other patterns difficult to fix.





> Clang
> suggests moving the semi-colon to the next line to silence these
> warnings but that defeats the purpose of the compact nature of these
> constructs so just hide the warning behind W=1 so its use can still be
> audited but it won't polute a regular build.
>
> Link: https://github.com/ClangBuiltLinux/linux/issues/42
> Link: https://github.com/ClangBuiltLinux/linux/issues/66
> Signed-off-by: Nathan Chancellor <natechancellor@...il.com>
> ---
>  scripts/Makefile.extrawarn | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/scripts/Makefile.extrawarn b/scripts/Makefile.extrawarn
> index cf6cd0ef6975..8709d9d6faf1 100644
> --- a/scripts/Makefile.extrawarn
> +++ b/scripts/Makefile.extrawarn
> @@ -11,6 +11,7 @@
>  # are not supported by all versions of the compiler
>  # ==========================================================================
>
> +KBUILD_CFLAGS += $(call cc-disable-warning, empty-body)
>  KBUILD_CFLAGS += $(call cc-disable-warning, packed-not-aligned)
>
>  ifeq ("$(origin W)", "command line")
> @@ -32,6 +33,7 @@ warning-1 += $(call cc-option, -Wpacked-not-aligned)
>  warning-1 += $(call cc-option, -Wstringop-truncation)
>  warning-1 += $(call cc-disable-warning, missing-field-initializers)
>  warning-1 += $(call cc-disable-warning, sign-compare)
> +warning-1 += $(call cc-option, -Wempty-body)
>
>  warning-2 := -Waggregate-return
>  warning-2 += -Wcast-align
> --
> 2.19.1
>


-- 
Best Regards
Masahiro Yamada

Powered by blists - more mailing lists