Re: [PATCH 1/1] OpenRISC: Fix kernel build problem on OpenRISC

[Guenter Roeck <linux@roeck-us.net>]

On 05/13/2015 02:58 AM, Paul Bolle wrote:
> On Tue, 2015-05-12 at 13:07 -0700, Guenter Roeck wrote:
>> On Tue, May 12, 2015 at 03:45:21PM +0200, anees wrote:
>>> Kernel build fails with error "target elf32-or32 not found"
>>> This is due to the change in OpenRISC compiler prefix from "or12"
>>> to "or1k". Add config options that set architecture output format
>>> default to "or1k" but also allow user to select older prefix.
>>>
>>> Tested with gcc version 4.9.1 uclibc
>>>
>>> Signed-off-by: anees <rean12is@gmail.com>
>>> ---
>>>   arch/openrisc/Kconfig              | 12 ++++++++++++
>>>   arch/openrisc/kernel/vmlinux.lds.S |  6 +++---
>>>   2 files changed, 15 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/arch/openrisc/Kconfig b/arch/openrisc/Kconfig
>>> index e5a693b..c7fe7c8 100644
>>> --- a/arch/openrisc/Kconfig
>>> +++ b/arch/openrisc/Kconfig
>>> @@ -75,6 +75,18 @@ config OPENRISC_BUILTIN_DTB
>>>           string "Builtin DTB"
>>>           default ""
>>>
>>> +config OUTPUT_FORMAT_TYPE
>>> +	string "Architecture output format type"
>>> +	default "or1k"
>>> +	help
>>> +	  Write "or12" if building kernel against an older toolchain
>>> +	  else leave the default (or1k)
>>> +
>> That means I'll have to use different toolchains for different kernel
>> versions, or hand-edit default configuration files to able to work
>> with the same toolchain. Both is, from a testing perspective,
>> quite annoying.
>
> Also quite annoying would be
>      $ grep OUTPUT_FORMAT .config
>      CONFIG_OUTPUT_FORMAT_TYPE="typo"
>
> (Note the missing CONFIG_OUTPUT_FORMAT= line.)
>
> What might be less annoying would be something (completely untested)
> like:
>      choice
> 	prompt "Architecture output format type"
> 	default OUTPUT_FORMAT_OR1K
> 	help
> 	  Choose "or12" if building kernel against an older toolchain
> 	  else leave the default "or1k".
>
>      config OUTPUT_FORMAT_OR1K
> 	bool "or1k"
>
>      config OUTPUT_FORMAT_OR12
> 	bool "or12"
>
>      enchoice
>
>      config OUTPUT_FORMAT
> 	string
> 	default "elf32-or1k" if OUTPUT_FORMAT_OR1K
> 	default "elf32-or12" if OUTPUT_FORMAT_OR12
>
>> Maybe I should simply stop testing older kernels for openrisc
>> after this patch has been merged; that would be much easier.
>> Jonas, is that ok with you ?
>
> Would testing be easier if you'd be provided with two defconfig files?
> Ie, one for "elf32-or1k" and "elf32-or12".
>

Just remembered ... or just apply https://lkml.org/lkml/2014/9/19/11 instead.

Guenter