On 8/20/20 1:55 PM, Kees Cook wrote: > On Thu, Aug 20, 2020 at 09:37:46AM -0700, Scott Branden wrote: >> On 2020-08-19 12:00 a.m., Greg Kroah-Hartman wrote: >>> On Tue, Aug 18, 2020 at 05:35:04PM -0700, Scott Branden wrote: >>>> >>>> On 2020-08-18 10:44 a.m., Greg Kroah-Hartman wrote: >>>>> On Tue, Aug 18, 2020 at 10:23:42AM -0700, Scott Branden wrote: >>>>>>>> +#define VK_FWSTS_RELOCATION_ENTRY BIT(0) >>>>>>> <snip> >>>>>>> >>>>>>> I thought BIT() was not allowed in uapi .h files, this really works >>>>>>> properly??? >>>>>> I did some investigation and it looks like a few other header files in >>>>>> include/uapi also use the BIT() macro: >>>>>> include/uapi/misc/uacce/uacce.h >>>>>> include/uapi/linux/psci.h >>>>>> include/uapi/linux/v4l2-subdev.h >>>>> Does the header install test target now fail for these? >>>> I do not understand the question above. make headers_install works. >>>> But I guess the above headers would have similar issue with the BIT macro. >>> Try enabling CONFIG_UAPI_HEADER_TEST and see what happens :) >> I enabled CONFIG_UAPI_HEADER_TEST and then >> built using "make" and "make headers_install". >> >> There didn't appear to be any issue with the BIT macro in the headers. > > FWIW, other subsystems have not been so lucky: > > https://git.kernel.org/linus/23b2c96fad21886c53f5e1a4ffedd45ddd2e85ba > > It may just be better to avoid BIT(), even if it works "by accident"(?) > for some header combinations...
Since we ship a copy of unifdef.c in tree, we could certainly teach it to undo the BIT() definition and replace it with an appropriate constant definition. -- Florian
