[yocto] how to build external firmware for kernel

[lothar at denx.de]

Zitat von Paul Eggleton <paul.eggleton at linux.intel.com>:

> Hi Lothar,
>
> On Monday 05 August 2013 11:42:22 lothar at denx.de wrote:
>> I have a linux kernel and a external firmware. Building manually,
>> first I have to checkout and build the firmware "native" to obtain a
>> firmware blob, then I need this firmware blob to build the kernel for
>> the target.
>>
>> So far I tried to do this in yocto with separate recipes: a
>> firmware.bb and a mylinux.bb. In firmware.bb, I check out the firmware
>> via git and let it run "make" on native (this is necessary, since the
>> Makefile builds a tool which then automatically gets called inside the
>> Makefile to create the firmware blob).
>>
>> In the kernel recipe, I declare a DEPEND on firmware.bb. I set the
>> .config options to load an extra firmware blob. But now, I have a
>> problem: I need actually to have the firmware blob within the kernel's
>> file hierarchy. Ideally I would copy/move it from the firmware's
>> destination directory into the kernel's source directory.
>
> Shouldn't firmware go into /lib/firmware?

Hi Paul, yes, so my first approach was indeed in the firmware.bb  
something like:
     install -d ${D}/lib/firmware/imx/sdma/
     install -m 755 sdma-imx53.bin ${D}/lib/firmware/imx/sdma/sdma-imx53.bin
...and to set in the kernel.bb (I'm aware of the existens of other  
ways to set .config using yocto)
     echo 'CONFIG_EXTRA_FIRMWARE="imx/sdma/sdma-imx53.bin"' >>  
${WORKDIR}/defconfig

But, compiling I got something as follows:
(...)
| DEBUG: Executing shell function do_compile
| NOTE: make -j 8 uImage CC=arm-linux-gnueabi-gcc   
-mno-thumb-interwork -marm LD=arm-linux-gnueabi-ld   LOADADDR=0x70008000
| scripts/kconfig/conf --silentoldconfig Kconfig
|   CHK     include/generated/uapi/linux/version.h
|   CHK     include/generated/utsrelease.h
| make[1]: `include/generated/mach-types.h' is up to date.
|   CC      scripts/mod/devicetable-offsets.s
|   CALL    scripts/checksyscalls.sh
|   GEN     scripts/mod/devicetable-offsets.h
|   HOSTCC  scripts/mod/file2alias.o
|   HOSTLD  scripts/mod/modpost
|   CHK     include/generated/compile.h
| make[1]: *** No rule to make target  
`/lib/firmware/imx/sdma/sdma-imx53.bin', needed by  
`firmware/imx/sdma/sdma-imx53.bin.gen.o'.  Stop.
| make[1]: *** Waiting for unfinished jobs....
|   MK_FW   firmware/imx/sdma/sdma-imx53.bin.gen.S
| make: *** [firmware] Error 2
| make: *** Waiting for unfinished jobs....
| ERROR: oe_runmake failed
| ERROR: Function failed: do_compile (see  
/work/lothar/BUILD__XXX__m53evk/tmp/work/m53evk-linux-gnueabi(...)

Honestly, I already expected some trouble, because of firmware.bb  
being built for "native", and the kernel not. I sort of played around  
with CONFIG_EXTRA_FIRMWARE and CONFIG_EXTRA_FIRMWARE_DIR without  
success.

I also tried to build it manually by the approach specified here:
https://community.freescale.com/thread/302880
...means, just copying the firmware blob into the kernel source's  
"firmware" folder and running make afterwards. Since this worked just  
perfectly, I thought of doing it like this, since it seemed easier  
than messing with /lib/firmware and yocto/bitbake. But with this  
approach I'm stuck at getting the firmware blob into the firmware  
folder.

Can you please tell me, is this the wrong approach?



>> Questions:
>> How can I achieve this kind of setup working out with yocto or
>> bitbake, respectively? What might be the best approach in my situation?
>
> You can get the kernel version name from ${STAGING_KERNEL_DIR}/kernel-
> abiversion if that helps, but I wouldn't have thought you'd need it if the
> firmware is installed to /lib/firmware instead.

How may I work with the STAGING_KERNEL_DIR, since I first run the  
firmware.bb and then the kernel.bb. I think I did not understand well,  
how I may achieve this. Do I need to install the firmware blob into  
something like $STAGING_KERNEL_DIR/firmware? How may I proceed here,  
could you please explain a bit more?