Re: [PATCH v3 0/7] Remove unneeded build directory traversals

[Arnd Bergmann]

On Wed, Mar 15, 2017 at 10:15 PM, Andrew F. Davis <afd@xxxxxx> wrote:
> On 03/15/2017 04:03 PM, Arnd Bergmann wrote:
>> On Wed, Mar 15, 2017 at 5:37 PM, Andrew F. Davis <afd@xxxxxx> wrote:
>>> Hello all,
>>>
>>> I was building a kernel for x86 and noticed Make still descended into
>>> directories like drivers/gpu/drm/hisilicon, this seems kind of odd given
>>> nothing will be built here. It looks to be due to some directories being
>>> included in obj-y unconditionally instead of only when the relevant
>>> CONFIG_ is set.
>>>
>>> These patches are split by subsystem in-case, for some reason, a file in
>>> a directory does need to be built, I believe I have checked for all
>>> instances of this, but a quick review from some maintainers would be nice.
>>
>> I didn't see anything wrong with the patches, and made sure that there
>> are no tristate symbols controlling the subdirectory for anything that
>> requires a built-in driver (which would cause a link failure).
>>
>> I'm not sure about drivers/lguest, which has some special magic
>> in its Makefile, it's possible that this now fails with CONFIG_LGUEST=m.
>>
>
> lguest and mmc are the strange ones, so I put them last in the series in
> case they did need to be dropped.
>
> lguest was supposed to have been taken from v1:
> https://lkml.org/lkml/2016/6/20/1086
> but it looks like it didn't so I re-introduced it for v3.
>
> mmc caught some 0-day build warnings but I never got to the bottom of them.

Ah, I see now what happened to mmc:

obj-$(subst m,y,$(CONFIG_MMC))  += host/
tmio_mmc_core-$(subst m,y,$(CONFIG_MMC_SDHI))   += tmio_mmc_dma.o
obj-$(subst m,y,$(CONFIG_MMC_SDHCI_PCI))        += sdhci-pci-data.o

with CONFIG_MMC=m, this will fail to build the built-in files in
drivers/mmc/host. I suppose this could be expressed in a different
way these days, but dropping the patch would be easier.

     Arnd