Re: [kbuild-all] make[2]: *** No rule to make target 'tools/testing/nvdimm//config_check.o', needed by 'tools/testing/nvdimm//dax.o'.

[Fengguang Wu <fengguang.wu@intel.com>]

On Tue, Aug 16, 2016 at 09:47:52AM +0800, Yilong Ren wrote:
>On Tue, Aug 16, 2016 at 09:41:02AM +0800, Fengguang Wu wrote:
>> On Mon, Aug 15, 2016 at 06:30:48PM -0700, Dan Williams wrote:
>> >On Mon, Aug 15, 2016 at 6:26 PM, Fengguang Wu <fengguang.wu@intel.com> wrote:
>> >>On Mon, Aug 15, 2016 at 05:58:36PM -0700, Dan Williams wrote:
>> >>>
>> >>>On Mon, Aug 15, 2016 at 3:03 AM, kbuild test robot
>> >>><fengguang.wu@intel.com> wrote:
>> >>>>
>> >>>>tree:
>> >>>>https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
>> >>>>head:   694d0d0bb2030d2e36df73e2d23d5770511dbc8d
>> >>>>commit: ab68f26221366f92611650e8470e6a926801c7d4 /dev/dax, pmem: direct
>> >>>>access to persistent memory
>> >>>>date:   3 months ago
>> >>>>config: i386-randconfig-i1-201633 (attached as .config)
>> >>>>compiler: gcc-4.8 (Debian 4.8.4-1) 4.8.4
>> >>>>reproduce:
>> >>>>        git checkout ab68f26221366f92611650e8470e6a926801c7d4
>> >>>>        # save the attached .config to linux build tree
>> >>>>        make ARCH=i386
>> >>>>
>> >>>>All errors (new ones prefixed by >>):
>> >>>>
>> >>>>>>make[2]: *** No rule to make target
>> >>>>>>'tools/testing/nvdimm//config_check.o', needed by
>> >>>>>>'tools/testing/nvdimm//dax.o'.
>> >>>>
>> >>>>   make[2]: Target '__build' not remade because of errors.
>> >>>
>> >>>
>> >>>I think this is an invalid build test.  tools/testing/nvdimm/ uses a
>> >>>external module Kbuild environment, not Kconfig.  So, there's nothing
>> >>>I can do to prevent this compile error, unless there's some other way
>> >>>0-day could determine the configuration dependencies?
>> >>
>> >>
>> >>Yeah if you can offer a concrete rule for the dependency, we'll add
>> >>it to 0-day.
>> >
>> >Sounds good.  The config_check.c file itself lists the dependencies:
>> >
>> >void check(void)
>> >{
>> >       /*
>> >        * These kconfig symbols must be set to "m" for nfit_test to
>>
>> If "y" is not a valid option, we'll need to adjust 0-day's dependency
>> specification for ndctl test:
>>
>> wfg /c/lkp-tests% cat include/ndctl
>> need_kconfig:
>> - CONFIG_HAVE_DMA_CONTIGUOUS=y
>> - CONFIG_CMA=y
>> - CONFIG_DMA_CMA=y
>> - CONFIG_CMA_SIZE_MBYTES=200
>> - CONFIG_LIBNVDIMM
>> - CONFIG_BLK_DEV_PMEM
>> - CONFIG_ND_BLK
>> - CONFIG_BTT=y
>> - CONFIG_NVDIMM_PFN=y
>> - CONFIG_NVDIMM_DAX=y
>> - CONFIG_ZONE_DEVICE=y
>>
>> In the above list, a bare "CONFIG_BLK_DEV_PMEM" means "y" or "m" are
>> both acceptable.
>
>Yes, this is due to enable_module() can accept "y" and "m".
>How about forcing enable_module() to accept "m" ?

I think we could change 

        - CONFIG_BLK_DEV_PMEM
to
        - CONFIG_BLK_DEV_PMEM=m

The former will correspond to kernel's

        #define IS_ENABLED(option) __or(IS_BUILTIN(option), IS_MODULE(option))

while the latter correspond to

        #define IS_MODULE(option) config_enabled(option##_MODULE)

And add logic to handle the =m case. Currently we only have
enable_module() which corresponds to kernel's IS_ENABLED().

> 23 # CONFIG_XXX=m => unchange
> 24 # CONFIG_XXX=y => unchange
> 25 # CONFIG_XXX is not set => CONFIG_XXX=m
> 26 enable_module()

The behavior here is good for its current callers, except the "module"
in the function name might be a bit misleading.

Thanks,
Fengguang