[GIT PULL] Broadcom Cygnus SoC defconfig changes

[Olof Johansson]

On Sun, Nov 9, 2014 at 9:36 PM, Scott Branden <sbranden at broadcom.com> wrote:
> On 14-11-09 11:24 AM, Florian Fainelli wrote:
>>
>> On 11/08/14 21:45, Scott Branden wrote:
>>>
>>> Hi Olof,
>>>
>>> Thanks for the comments.  Please see my questions inline before we
>>> determine what needs to be done for the defconfig patchset.
>>>
>>> On 14-11-08 04:13 PM, Olof Johansson wrote:
>>>>
>>>> On Thu, Oct 30, 2014 at 12:50:09PM -0700, Florian Fainelli wrote:
>>>>>
>>>>> The following changes since commit
>>>>> f114040e3ea6e07372334ade75d1ee0775c355e1:
>>>>>
>>>>>     Linux 3.18-rc1 (2014-10-19 18:08:38 -0700)
>>>>>
>>>>> are available in the git repository at:
>>>>>
>>>>>     http://github.com/brcm/linux.git
>>>>> tags/arm-soc/for-3.18/cygnus-defconfig-v9
>>>>>
>>>>> for you to fetch changes up to
>>>>> aa0204801143ee2d47e9e8dd5f39cdddd5613b4b:
>>>>>
>>>>>     ARM: multi_v7_defconfig: Enable Broadcom Cygnus (2014-10-30
>>>>> 11:01:35 -0700)
>>>>
>>>>
>>>> Hi Florian,
>>>>
>>>> Apologies for the delay in dealing with this pull request.
>>>>
>>>>> ----------------------------------------------------------------
>>>>> This patchset contains initial support for Broadcom's Cygnus SoC
>>>>> based on our
>>>>> iProc architecture. Initial support is minimal and includes just the
>>>>> mach
>>>>> platform code, clock driver, and a basic device tree configuration.
>>>>> Peripheral
>>>>> drivers will be submitted soon, as will device tree configurations
>>>>> for other
>>>>> Cygnus board variants.
>>>>>
>>>>> These are the defconfig parts of the changes
>>>>>
>>>>> ----------------------------------------------------------------
>>>>> Jonathan Richardson (1):
>>>>>         ARM: cygnus defconfig : Initial defconfig for Broadcom Cygnus
>>>>> SoC
>>>>>
>>>>> Ray Jui (1):
>>>>>         ARM: multi_v7_defconfig: Enable Broadcom Cygnus
>>>>>
>>>>> Scott Branden (1):
>>>>>         ARM: bcm_defconfig/multi_v7_defconfig: remove one level of
>>>>> menu from Kconfig
>>>>>
>>>>>    arch/arm/configs/bcm_cygnus_defconfig | 237
>>>>> ++++++++++++++++++++++++++++++++++
>>>>>    arch/arm/configs/bcm_defconfig        |   3 +-
>>>>>    arch/arm/configs/multi_v7_defconfig   |  21 ++-
>>>>
>>>>
>>>> Please send multi_v7_defconfig updates as a patch instead of including
>>>> in
>>>> a merge request. We ask for this because it's a file that many
>>>> maintainers
>>>> touch and it's easy that we get lots of conflicts otherwise.
>>>
>>> OK, easy to do.  Are there other files that we need to be aware of that
>>> have these different requirements to merge as well?
>>>
>>>>
>>>> As far as the bcm_cygnus_defconfig: Since new platforms are
>>>> multiplatform
>>>> enabled by default, we normally only prefer to have one defconfig per
>>>> vendor.
>>>> We've made exceptions in cases where the families are very different,
>>>> but in
>>>> general one defconfig per vendor should be sufficient.
>>>>
>>>> I.e. please enable the cygnus options in bcm_defconfig instead of
>>>> adding a new
>>>> one.
>>>
>>>
>>> Hi Olof, we can do this.  But this makes bcm_defconfig unusable for us
>>> in a product and is not a simple configuration to start from for our
>>> customers.  bcm_defconfig is a different SoC family based around the
>>> Kona architecture. bcm_defconfig won't be what we build and test on
>>> Cygnus at all.  There is a script in scripts/kconfig/mergeconfig.pl that
>>> allows you to merge config files together if you wish to have a kernel
>>> support multiple SoCs.  I see the same problem with multi_v7_defconfig
>>> but even worse.  It isn't a configuration we will ever use.  Its only
>>> purpose seems to be to enable our drivers in the that build just in case
>>> somebody else happens to break something in the build.
>>
>>
>> The defconfigs shipped in the mainline kernel only serve the purpose of
>> building as many drivers/subsystems as possible and making sure they
>> work well together from a general distribution perspective like
>> Debian/Ubuntu/OE... That does not mean this is your configuration
>> template for a downstream kernel with only a specific set of SoCs to
>> support.
>>
>>>
>>> As it stands, multi_v7_defconfig boots noticably slower than
>>> bcm_cygnus_defconfig.  And, the kernel is larger than we want.  And, the
>>> kconfigs are more complicated than we need for our boards.  I also
>>> notice some of the other SoCs families enabled automatically turn on a
>>> lot of features we don't want on.
>>
>>
>> Right, you are also very likely to see code blow away that has been
>> enabled as part of the multi_v7_defconfig, but actually assumes that it
>> will only run on a a particular family of SoCs, of course, all of this
>> is good to fix to improve the multiplatform kernel, but does not
>> necessarily impact you directly.
>>
>>>
>>> Perhaps most of the multi_v7_defconfig merge issue you are facing could
>>> be solved if you used the mergconfig.pl script to merge multiple
>>> defconfigs together?
>>>
>>> What we wish to do is have a config file that only supports the Cygnus
>>> family of SoCs.  This allows others to easily see what applies to Cygnus
>>> without being confused by other configurations.  And, if somebody wishes
>>> to use scripts/kconfig/mergeconfig.pl to merge different SoC support
>>> together they are free to do so.
>>>
>>> So, we can do as you ask to get this upstreamed.  But it really serves
>>> little purpose for us other than verifying things build.
>>>
>>> Where do we upstream the defconfig that we actually use?  Or is this
>>> just not done and everyone keeps everything locally and has to continue
>>> to maintain it there?
>>
>>
>> FWIW, this is typically what we do for the brcmstb_defconfig, it is just
>> present in our downstream kernel, but we do use the multi_v7_defconfig
>> when testing upstream.
>
> Puzzling.  So dts files are allowed to be upstreamed per board, but the
> reference kernel configuration that is actually tested extensively on that
> board is not upstreamed?

Correct. Distros (desktop/server/embedded/etc) test the kernel too,
and they don't have their defconfigs in the kernel repo either.

The beauty of the defconfigs is that even though you enable more
options than the bare minimum, the kernel should still work (or it's
broken). Sure, it'll be larger, and might take a while longer to load
from disk, but things should work just fine.


-Olof