Hi,

probably
4ce6efed48d736e3384c39ff87bda723e1f8e041
kbuild: soften modpost checks when doing cross builds

causes build failures on x86_64 kernel build on 32 bit machine:

make -f /local/xslaby/hid/scripts/Makefile.modpost
YYYYYYYYYYYYYY x86_64 x86_64
   scripts/mod/modpost -m -a -o /local/xslaby/buhid/Module.symvers    -S -K 
/local/xslaby/buhid/Module.markers -M /local/xslaby/buhid/Module.markers   -s
XXXXXXXXXXXXXXXXXX 20: 0 (2), 4 (4), 8 (4), 12 (8)
FATAL: drivers/hid/hid-apple: sizeof(struct hid_device_id)=20 is not a 
modulo of the size of section __mod_hid_device_table=648.
Fix definition of struct hid_device_id in mod_devicetable.h
make[2]: *** [__modpost] Error 1
make[1]: *** [modules] Error 2
make: *** [sub-make] Error 2

after sticking echo into:
  __modpost: $(modules:.ko=.o) FORCE
+       @echo YYYYYYYYYYYYYY $(KBUILD_BUILDHOST) $(ARCH)
         $(call cmd,modpost) $(wildcard vmlinux) $(filter-out FORCE,$^)

--

[ message continues ]

Well, let's fix sam's email.

--

[ message continues ]

Your machine claims it is a x86_64 and you are building for a x86_64

So it looks to me that you are building for 64 bit on a 64 bit machine.
So when you write:
 on x86_64 kernel build on 32 bit machine:

I assume you try to say that your HOSTCC is 32 bit and your
kernel cc is 64 bit (or something similar).

The commit you reference loosen up the check so we do not do the
check if we _know_ that the host and the target system differs.
The only thing I can imagine right now to solve your issue would
be to skip the check if we do not use the same gcc for target and host?

	Sam
--

[ message continues ]

So something like this (entirely untested):

--- a/scripts/Makefile.modpost
+++ b/scripts/Makefile.modpost
@@ -65,7 +65,7 @@ modules   := $(patsubst %.o,%.ko, $(wildcard $(__modules:.ko=.o)))
 # Stop after building .o files if NOFINAL is set. Makes compile tests quicker
 _modpost: $(if $(KBUILD_MODPOST_NOFINAL), $(modules:.ko:.o),$(modules))
 
-ifneq ($(KBUILD_BUILDHOST),$(ARCH))
+ifneq ($(CC),$(HOSTCC))
         cross_build := 1
 endif
 

Only powerpc (and ppc) does seems to do the wrong thing
and fiddle with CC:

powerpc/Makefile:override CC    += -m$(CONFIG_WORD_SIZE)

So for most users this is OK.
(If it works as expected).

	Sam
--

[ message continues ]

Well, the compiler is the same, but it's told to compile the kernel with 
-m64. It runs on x86_64 kernel and i386 userspace. It's debian i386 gcc 
package which understands x86_64 objects too. No SUBARCH option, it seems to 

--

[ message continues ]

Unless you manually fiddle with ARCH you can only build a kernel with 
the same 32<->64bit setting as your currently running kernel (no matter 
which userspace you use)...

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

--

[ message continues ]

echo -en '#!/bin/bash\nexec gcc -m32 "$@"\n' >/usr/local/bin/i386-linux-gcc
make CC=i386-linux-gcc
then the if($(CC),$(HOSTCC)) check should get true.
--

[ message continues ]

No doubt that you can work around the problem.

But that's not a solution.

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

--

[ message continues ]

Of course not. Ever since the arch unification, building for the
other bitness in bi- or multiarch systems has become harder.
--

[ message continues ]

It works fine for the unified powerpc tree.

-- 
dwmw2

--

[ message continues ]

That is good for ppc. But for x86, things seem to have changed.

Previously, one could just do `make ARCH=x86_64` on an i586
installation, now the required magic is a little bit tougher,
unless I am missing something (new).
--

[ message continues ]

In that cae modpost tells you to fix your typedefiniton so the
size used in the kernel matches what the userspace can see.
So what we have here is exactly the kind of issues this
check were introduced to catch.

Add a bit more alignment to  hid_device_id and userspace and
kernel should be happy.

	Sam
--

[ message continues ]