Re: [PATCH 1/2] of: Add support for linking device tree blobs into vmlinux

[Grant Likely <grant.likely@secretlab.ca>]

On Mon, Nov 15, 2010 at 10:28 PM, Dirk Brandewie
<dirk.brandewie@gmail.com> wrote:
> On 11/15/2010 09:17 PM, Grant Likely wrote:
>>
>> On Mon, Nov 15, 2010 at 10:06 PM, Dirk Brandewie
>> <dirk.brandewie@gmail.com> =A0wrote:
>>>
>>> On 11/15/2010 08:41 PM, Grant Likely wrote:
>>>>
>>>> On Mon, Nov 15, 2010 at 08:01:20PM -0800, dirk.brandewie@gmail.com
>>>> wrote:
>>>>>
>>>>> From: Dirk Brandewie<dirk.brandewie@gmail.com>
>>>>>
>>>>> This patch adds support for linking device tree blobs into
>>>>> vmlinux. The device tree blobs are placed in the init.data
>>>>> section.
>>>>>
>>>>> Signed-off-by: Dirk Brandewie<dirk.brandewie@gmail.com>
>>>>> ---
>>>>> =A0include/asm-generic/vmlinux.lds.h | =A0 19 +++++++++++++++++--
>>>>> =A0scripts/Makefile.lib =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 17 +++++++++=
++++++++
>>>>> =A02 files changed, 34 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/include/asm-generic/vmlinux.lds.h
>>>>> b/include/asm-generic/vmlinux.lds.h
>>>>> index bd69d79..ea671e7 100644
>>>>> --- a/include/asm-generic/vmlinux.lds.h
>>>>> +++ b/include/asm-generic/vmlinux.lds.h
>>>>> @@ -67,7 +67,14 @@
>>>>> =A0 * Align to a 32 byte boundary equal to the
>>>>> =A0 * alignment gcc 4.5 uses for a struct
>>>>> =A0 */
>>>>> -#define STRUCT_ALIGN() . =3D ALIGN(32)
>>>>> +#define STRUCT_ALIGNMENT 32
>>>>> +#define STRUCT_ALIGN() . =3D ALIGN(STRUCT_ALIGNMENT)
>>>>> +
>>>>> +/* Device tree blobs linked into the kernel need to have proper
>>>>> + * structure alignment to be parsed by the flat device tree library
>>>>> + * used in early boot
>>>>> +*/
>>>>> +#define DTB_ALIGNMENT STRUCT_ALIGNMENT
>>>>>
>>>>> =A0/* The actual configuration determine if the init/exit sections
>>>>> =A0 * are handled as text/data or they can be discarded (which
>>>>> @@ -146,6 +153,13 @@
>>>>> =A0#define TRACE_SYSCALLS()
>>>>> =A0#endif
>>>>>
>>>>> +
>>>>> +#define KERNEL_DTB()
>>>>> \
>>>>> + =A0 =A0 =A0 . =3D ALIGN(DTB_ALIGNMENT);
>>>>> \
>>>>> + =A0 =A0 =A0 VMLINUX_SYMBOL(__dtb_start) =3D .;
>>>>> =A0\
>>>>> + =A0 =A0 =A0 *(.dtb.init.rodata)
>>>>> \
>>>>> + =A0 =A0 =A0 VMLINUX_SYMBOL(__dtb_end) =3D .;
>>>>> +
>>>>> =A0/* .data section */
>>>>> =A0#define DATA_DATA
>>>>> \
>>>>> =A0 =A0 =A0 =A0*(.data)
>>>>> =A0\
>>>>> @@ -468,7 +482,8 @@
>>>>> =A0 =A0 =A0 =A0MCOUNT_REC()
>>>>> =A0\
>>>>> =A0 =A0 =A0 =A0DEV_DISCARD(init.rodata)
>>>>> =A0\
>>>>> =A0 =A0 =A0 =A0CPU_DISCARD(init.rodata)
>>>>> =A0\
>>>>> - =A0 =A0 =A0 MEM_DISCARD(init.rodata)
>>>>> + =A0 =A0 =A0 MEM_DISCARD(init.rodata)
>>>>> =A0\
>>>>> + =A0 =A0 =A0 KERNEL_DTB()
>>>>>
>>>>> =A0#define INIT_TEXT
>>>>> \
>>>>> =A0 =A0 =A0 =A0*(.init.text)
>>>>> \
>>>>> diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
>>>>> index 4c72c11..a8a4774 100644
>>>>> --- a/scripts/Makefile.lib
>>>>> +++ b/scripts/Makefile.lib
>>>>> @@ -200,6 +200,23 @@ quiet_cmd_gzip =3D GZIP =A0 =A0$@
>>>>> =A0cmd_gzip =3D (cat $(filter-out FORCE,$^) | gzip -f -9> =A0 =A0$@) =
|| \
>>>>> =A0 =A0 =A0 =A0(rm -f $@ ; false)
>>>>>
>>>>> +# DTC
>>>>> +#
>>>>>
>>>>> =A0------------------------------------------------------------------=
---------
>>>>> +$(obj)/%.dtb.S: $(obj)/%.dtb FORCE
>>>>> + =A0 =A0 =A0 @echo '#include<asm-generic/vmlinux.lds.h>'> =A0 =A0$@
>>>>> + =A0 =A0 =A0 @echo '.section .dtb.init.rodata,"a"'>> =A0 =A0$@
>>>>> + =A0 =A0 =A0 @echo '.balign DTB_ALIGNMENT'>> =A0 =A0$@
>>>>> + =A0 =A0 =A0 @echo '.global __dtb_$(*F)_begin'>> =A0 =A0$@
>>>>> + =A0 =A0 =A0 @echo '__dtb_$(*F)_begin:'>> =A0 =A0$@
>>>>> + =A0 =A0 =A0 @echo '.incbin "$<" '>> =A0 =A0$@
>>>>> + =A0 =A0 =A0 @echo '__dtb_$(*F)_end:'>> =A0 =A0$@
>>>>> + =A0 =A0 =A0 @echo '.global __dtb_$(*F)_end'>> =A0 =A0$@
>>>>> + =A0 =A0 =A0 @echo '.balign DTB_ALIGNMENT'>> =A0 =A0$@
>>>>> +
>>>>> +DTC =3D $(objtree)/scripts/dtc/dtc
>>>>> +
>>>>> +quiet_cmd_dtc =3D DTC =A0 =A0$@
>>>>> + =A0 =A0 =A0cmd_dtc =3D $(DTC) -O dtb -o $(obj)/$*.dtb -b 0 =A0$(src=
)/$*.dts
>>>>
>>>> Missing the %.dtb: %.dts rule, but otherwise looks okay. =A0You will
>>>> need to make sure this doesn't break powerpc or microblaze when the
>>>> dts->dtb rule is added.
>>>>
>>> I have the rule
>>> =A0$(obj)/%.dtb: $(src)/%.dts
>>> =A0 =A0 =A0 =A0 =A0$(call if_changed,dtc)
>>> in the arch/x86/kernel/Makefile to prevent this sneaking into other oth=
er
>>> architectures.
>>
>> This rule looks correct. =A0PowerPC and Microblaze need to be modified
>> to use it. =A0It should not be hard to do, give it a try. =A0Worst case,
>> your first attempt is wrong and the rest of us fix it up. =A0:-)
>>
>> Hint: PowerPC currently puts the .dtb file in a different directory
>> from the source .dts file. =A0It doesn't need to do it that way.
>> arch/powerpc/boot/Makefile will need to be modified.
>>
>
> I will give it a shot. =A0The only real difference except for the directo=
ry
> structures is powerpc and microblaze add padding to the dtb with the -p 1=
024
> command line argument to dtc.

You could use a $(DTCFLAGS) variable to pass in arch specific flags.

> =A0Is the padding needed when the blob are
> linked into vmlinux proper?

Maybe, maybe not.  Padding is required to be able to modify the .dtb
data in-place, which is important when adapting data from other
sources into the device tree structure.  However, when it is linked
into the kernel, the adding of additional data /possibly/ can be
deferred until after the tree is either copied or unflattened.

g.