Re: [PATCH] kbuild: Handle builtin dtb files containing hyphens

[Frank Rowand]

On 03/07/18 12:25, James Hogan wrote:
> On Wed, Mar 07, 2018 at 12:11:41PM -0800, Frank Rowand wrote:
>> I initially misread the patch description (and imagined an entirely
>> different problem).
>>
>>
>> On 03/07/18 06:06, James Hogan wrote:
>>> On dtb files which contain hyphens, the dt_S_dtb command to build the> dtb.S files (which allow DTB files to be built into the kernel) results> in errors like the following:> > bcm3368-netgear-cvg834g.dtb.S: Assembler messages:> bcm3368-netgear-cvg834g.dtb.S:5: Error: : no such section> bcm3368-netgear-cvg834g.dtb.S:5: Error: junk at end of line, first unrecognized character is `-'> bcm3368-netgear-cvg834g.dtb.S:6: Error: unrecognized opcode `__dtb_bcm3368-netgear-cvg834g_begin:'> bcm3368-netgear-cvg834g.dtb.S:8: Error: unrecognized opcode `__dtb_bcm3368-netgear-cvg834g_end:'> bcm3368-netgear-cvg834g.dtb.S:9: Error: : no such section> bcm3368-netgear-cvg834g.dtb.S:9: Error: junk at end of line, first unrecognized character is `-'
>> Please replace the following section:
>>
>>> This is due to the hyphen being used in symbol names. Replace all
>>> hyphens 
>>> with underscores in the dt_S_dtb command to avoid this problem.
>>>
>>> Quite a lot of dts files have hyphens, but its only a problem on MIPS
>>> where such files can be built into the kernel. For example when
>>> CONFIG_DT_NETGEAR_CVG834G=y, or on BMIPS kernels when the dtbs target is
>>> used (in the latter case it admitedly shouldn't really build all the
>>> dtb.o files, but thats a separate issue).
>>
>> with:
>>
>>    cmd_dt_S_dtb constructs the assembly source to incorporate a devicetree
>>    FDT (that is, the .dtb file) as binary data in the kernel image.
>>    This assembly source contains labels before and after the binary data.
>>    The label names incorporate the file name of the corresponding .dtb
>>    file.  Hyphens are not legal characters in labels, so transform all
>>    hyphens from the file name to underscores when constructing the labels.
> 
> Thanks, that is clearer.
> 
> I'll keep the paragraph about MIPS and the example configuration though,
> as I think its important information to reproduce the problem, and to
> justify why it wouldn't be appropriate to just rename the files (which
> was my first reaction).

Other than the part that says "its only a problem on MIPS".  That is
pedantically correct because no other architecture (that I am aware
of, not that I searched) currently has a devicetree source file name
with a hyphen in it, where that file is compiled into the kernel as
an asm file.  But it is potentially a problem on any architecture
to it is misleading to label it as MIPS only.


> 
>> Reviewed-by: Frank Rowand <frowand.list@xxxxxxxxx>
> 
> Thanks
> James
>