On 5/27/20 4:03 PM, Jérôme Pouiller wrote:
> On Tuesday 26 May 2020 23:02:56 CEST Hauke Mehrtens wrote:
>> On 5/25/20 11:35 AM, Jérôme Pouiller wrote:
>>> On Thursday 21 May 2020 22:14:22 CEST Hauke Mehrtens wrote:
>>>> When compiling inside the kernel include linux/stddef.h instead of
>>>> stddef.h. When I compile this header file in backports for power PC I
>>>> run into a conflict with ptrdiff_t. I was unable to reproduce this in
>>>> mainline kernel. I still would like to fix this problem in the kernel.
>>>>
>>>> Fixes: 6989310f5d43 ("wireless: Use offsetof instead of custom macro.")
>>>> Signed-off-by: Hauke Mehrtens <hauke@xxxxxxxxxx>
>>>> ---
>>>> include/uapi/linux/wireless.h | 6 +++++-
>>>> 1 file changed, 5 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/include/uapi/linux/wireless.h b/include/uapi/linux/wireless.h
>>>> index a2c006a364e0..24f3371ad826 100644
>>>> --- a/include/uapi/linux/wireless.h
>>>> +++ b/include/uapi/linux/wireless.h
>>>> @@ -74,7 +74,11 @@
>>>> #include <linux/socket.h> /* for "struct sockaddr" et al */
>>>> #include <linux/if.h> /* for IFNAMSIZ and co... */
>>>>
>>>> -#include <stddef.h> /* for offsetof */
>>>> +#ifdef __KERNEL__
>>>> +# include <linux/stddef.h> /* for offsetof */
>>>> +#else
>>>> +# include <stddef.h> /* for offsetof */
>>>> +#endif
>>>
>>> Hello,
>>>
>>> This patch also solves a warning I encountered when I launched sparse
>>> (with make C=1):
>>>
>>> /usr/lib/gcc-cross/arm-linux-gnueabihf/9/include/stddef.h:406:9: warning: preprocessor token offsetof redefined
>>> ./include/linux/stddef.h:17:9: this was the original definition
>>>
>>> However, I wonder if it useful to keep 'include <stddef.h>' when this file
>>> is included from userspace. If it does not compile anymore, there is
>>> problem somewhere else, no?
>>>
>>> (Globally, I always find suspicious a kernel source that includes a system
>>> header)
>> Hi,
>>
>> The wireless.h makes use of offsetof and this is defined in system
>> stddef.h and in include/linux/stddef.h. When we only include
>> linux/stddef.h it will work fine when compiled inside the kernel, but it
>> could cause problems when used in user space, because this would include
>> ./include/uapi/linux/stddef.h which does not define offsetof.
>>
>> I think we need the #ifdef __KERNEL__ to make this work in the kernel
>> and in the user space.
>
> In this case, maybe the problem should be resolved directly in
> uapi/linux/stddef.h?
>
> I have found other headers that use offsetof and may have the same problem
> (currently, they do not include stddef.h. So, some of the macros do not
> compile):
> - include/uapi/linux/fuse.h
> - include/uapi/linux/genwqe/genwqe_card.h
> - include/uapi/linux/gfs2_ondisk.h
>
Hi,
Would you prefer to add this to include/uapi/linux/stddef.h:
#ifndef offsetof
#define offsetof(TYPE, MEMBER) ((size_t)&((TYPE *)0)->MEMBER)
#endif
and then include linux/stddef.h in all header files which use offsetof?
include/uapi/linux/fuse.h checks for __KERNEL__ and then includes
linux/types.h or stdint.h, my initial approach would be to do this in a
similar way for stddef.h.
What is the kernel policy of including system header files like stddef.h?
Hauke
Attachment:
signature.asc
Description: OpenPGP digital signature
