On 4 of February 2009 16:33:04 David Woodhouse wrote:

On Sun, 2009-02-01 at 11:05 -0800, Jesse Brandeburg wrote:

quoted

On Sat, Jan 31, 2009 at 7:59 AM, Andrey Borzenkov 

[off-list ref] wrote:

quoted

quoted

{pts/2}% LC_ALL=C make -C ~/src/linux-git
O=$HOME/build/linux-2.6.29 make: Entering directory
`/home/bor/src/linux-git'
 Using /home/bor/src/linux-git as source for kernel
ln -fsn /home/bor/src/linux-git source

I think (seem to recall something on the list) there have been
multiple people reporting the make O= option doesn't work with
CONFIG_FIRMWARE_IN_KERNEL,

That shouldn't be the case -- it works fine here.

Andrey, what version of make are you using?

OK, build error was likely result of something I did before (I suspect 
what, but after I cleaned up build firmware/e100 it rebuilds OK now). 
But even after I verified that firmware *was* built in kernel loading 
still failed. This is genuine bug caused by config option mismatch.

I had CONFIG_FW_LOADER=m. I do not remember ever setting it explicitly, 
so I suspect it was selected by some other driver in the past (I do make 
oldconfig for at least ten past releases if not more). This cannot work 
with CONFIG_FIRMWARE_IN_KERNEL=y because of this code in 
drivers/base/firmware_class.c:

#ifdef CONFIG_FW_LOADER
extern struct builtin_fw __start_builtin_fw[];
extern struct builtin_fw __end_builtin_fw[];
#else /* Module case. Avoid ifdefs later; it'll all optimise out */
static struct builtin_fw *__start_builtin_fw;
static struct builtin_fw *__end_builtin_fw;
#endif

IOW CONFIG_FIRMWARE_IN_KERNEL must force CONFIG_FW_LOADER=y. I do not 
understand Kconfig enough to offer a patch.