Re: [PATCH] copy_{to,from}_user(): fix compile-time sanity checks with gcc -Og

From: Arnd Bergmann
Date: Sat Nov 03 2018 - 11:46:16 EST


On 11/3/18, Changbin Du <changbin.du@xxxxxxxxx> wrote:
> On Fri, Nov 02, 2018 at 04:56:58PM +0100, Arnd Bergmann wrote:
>> When CONFIG_CC_OPTIMIZE_FOR_DEBUGGING is set, we get countless warnings
>> like
>>
>> In function 'check_copy_size',
>> inlined from 'copy_from_user' at include/linux/uaccess.h:146:6,
>> inlined from '__copy_siginfo_from_user' at kernel/signal.c:3032:6:
>> include/linux/thread_info.h:147:4: error: call to '__bad_copy_to' declared
>> with attribute error: copy destination size is too small
>>
>> It seems that constant propagation doesn't work well enough to make
>> this code reliable, so turn it off for that configuration.
>>
> This is caused by __compiletime_warning() and fixed by below change
> already. Could you try the latest kbuild tree?
>
> --- a/include/linux/compiler-gcc.h
> +++ b/include/linux/compiler-gcc.h
> @@ -86,8 +86,10 @@
> #define __compiletime_object_size(obj) __builtin_object_size(obj, 0)
>
> #ifndef __CHECKER__
> +#ifndef CONFIG_CC_OPTIMIZE_FOR_DEBUGGING
> #define __compiletime_warning(message) __attribute__((warning(message)))
> #define __compiletime_error(message) __attribute__((error(message)))
> +#endif

Right, that works as well, but my version is a bit less invasive as it
only disables the __compiletime_error in the one file that introduced
the regression, and not all of them. I have built hundreds of randconfig
builds and found no other problem with __compiletime_error(), though
I did find a bunch of (correct) section mismatch warnings and
false-positive -Wformat-overflow= warnings with
CONFIG_CC_OPTIMIZE_FOR_DEBUGGING (patches pending).

I also got some link errors for which I already posted patches.

Arnd