[I-PIPE] ipipe-core-4.14.89-x86-2 released

Philippe Gerum rpm at xenomai.org
Mon Jan 28 18:15:23 CET 2019


On 1/28/19 5:48 PM, Jan Kiszka wrote:
> On 28.01.19 16:31, Philippe Gerum wrote:
>> On 1/28/19 4:18 PM, Philippe Gerum via Xenomai wrote:
>>> On 1/28/19 12:22 PM, Jan Kiszka wrote:
>>>> On 27.12.18 17:17, Jan Kiszka via Xenomai wrote:
>>>>> On 24.12.18 11:06, Philippe Gerum via Xenomai wrote:
>>>>>> On 12/23/18 10:34 PM, Alec Ari via Xenomai wrote:
>>>>>>> Hello,
>>>>>>>
>>>>>>>
>>>>>>> I'm not sure if the patch generator was believed to be fixed yet or
>>>>>>> not, but files for other arches (arch/arm/include/asm/arch_timer.h)
>>>>>>> are still included. If this still being worked on, disregard this
>>>>>>> message.
>>>>>>>
>>>>>>>
>>>>>>
>>>>>> The upstream tree at https://gitlab.denx.de/Xenomai/ipipe-x86 still
>>>>>> exports I-pipe bits for multiple architectures (at least for x86 and
>>>>>> arm) but does not include scripts/ipipe/genpatches.sh, so the
>>>>>> generator
>>>>>> does a plain diff, assuming a single-arch tree.
>>>>>>
>>>>>> The ARM bits are likely left-overs and should be dropped to fix the
>>>>>> tree.
>>>>>>
>>>>>
>>>>> Going through the commits, it seems we have a couple of ARM-only
>>>>> driver and timer patches in the x86 tree but no general ARM enabling.
>>>>> Those should be removed, indeed.
>>>>
>>>> In fact, the problem is imported from ipipe-noarch: There are many
>>>> ARM-related driver changes that, strictly spoken, do not belong into
>>>> the
>>>> noarch tree. Some touch arch/arm, others just change drivers that are
>>>> practically ARM-only.
>>>>
>>>> Philippe, do we want to remove those commits from noarch and add
>>>> them to
>>>> ARM? Or should we filter out arch-unrelated diffs when generating the
>>>> patches?
>>>>
>>>
>>> The drivers you mention are shared between different archs (several arm/
>>> drivers are used by arm64) so instead of maintaining multiple copies,
>>> all these drivers have  been merged into noarch eventually.
>>>
>>> This is a patch generation issue for kernel series 4.14+, the easiest
>>> way may be to provide an exclude file to the generator on a per-arch
>>> basis. A filter rule excluding arm/ for anything but arm and arm64
>>> should be enough. I'll have a look at this.
>>>
>>> Please note that this does not apply to pre-4.14 trees; those should
>>> provide scripts/ipipe/genpatches.sh as the legacy I-pipe series normally
>>> does, which should filter contents appropriately.
>>>
>>
>> Those files should not be part of noarch though:
>>
>>   arch/arm/mach-imx/mach-imx51.c            |    5 +
>>   arch/arm/mach-imx/mach-imx53.c            |    5 +
>>   arch/arm/plat-omap/dmtimer.c              |   17 +-
>>   arch/arm/plat-omap/include/plat/dmtimer.h |    6 +
>>
> 
> So these three should go:
> 
> a1edb318ad047d0582de66482efa3b76b30b6e1a
> 09cafa0ee6b1547133eee438c37c8703e70a88d1
> 5fc1baae9cc27adc771c67aee131993612e88ff2
> 
> There is also 9f33ef39090f5f100f056e208733fef2e02bd373, which also
> contains a "FIXME" in the commit log.
> 
> If we get rid of these, there will be no further need for filtering of
> arch/.
> 

Actually, after a second look, if the idea is to strip anything an arch
does not use, there are many more commits to drop from arm64 and other
archs. Most of them live in drivers/irqchip and drivers/clocksource.
Either we go for a manifest of files to include, or an exclude filter.
genpatches.sh used to build the former dynamically, maybe this has to be
reinstated for 4.14+ eventually.

-- 
Philippe.



More information about the Xenomai mailing list