Re: [PATCH v4 00/21] platform/x86: Intel platform driver code movement

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 8/20/21 12:12 PM, Andy Shevchenko wrote:
> On Thu, Aug 19, 2021 at 08:46:14PM +0200, Hans de Goede wrote:
>> On 8/19/21 6:37 PM, Andy Shevchenko wrote:
>>> This is v4 of the Intel drivers move on in the source tree.
>>> v3 has been done by Kate:
>>> https://lore.kernel.org/platform-driver-x86/20210819033001.20136-1-hpa@redhat.com/
>>>
>>> I have taken the initial set from review-hans branch and removed Hans' SoB
>>> along with Link, while leaving others' tags.
>>>
>>> Changelog v4:
>>> - spelled fully the driver names in the commit messages
>>> - fixed SCU *.ko module names
>>
>> Actually I checked this during review and
>> CONFIG_INTEL_SCU_IPC, CONFIG_INTEL_SCU_PCI and CONFIG_INTEL_SCU_WDT
>> are booleans, iow if enabled they are always builtin so the
>> drivers/platform/x86/intel/scu/Makefile from v3 was correct.
>>
>> Since the v3 Makefile is more simple I prefer that version,
>> but if you prefer to keep the v4 version that is fine too.
> 
> I prefer mine, but we need to fix what kbuild bot complained about.

Right, so I've taken a look at that and I believe that
we need to change 2 things:

1. In drivers/platform/x86/intel/Makefile, use:

obj-$(CONFIG_INTEL_SCU_IPC)		+= scu/

Instead of:

obj-$(CONFIG_INTEL_SCU)			+= scu/

Since "config MFD_INTEL_PMC_BXT" from drivers/mfd/Kconfig
selects INTEL_SCU_IPC not INTEL_SCU. From a pdx86 pov thos
makes on difference since CONFIG_INTEL_SCU selects
INTEL_SCU_IPC too, so all pdx86 Kconfig options needing us to
dive into the scu subdir also end up selecting INTEL_SCU_IPC
(through INTEL_SC).

2. The "select INTEL_SCU_IPC" done by CONFIG_MFD_INTEL_PMC_BXT
only works if make will actually dive into the
drivers/platform/x86/intel subdir which requires
CONFIG_X86_PLATFORM_DRIVERS_INTEL to be set, so in the
"config MFD_INTEL_PMC_BXT" section of drivers/mfd/Kconfig

The following change should be made:

-	depends on X86_PLATFORM_DEVICES
+	depends on X86_PLATFORM_DRIVERS_INTEL

To ensure that we dive into the intel subdir of
drivers/platform/x86, note that X86_PLATFORM_DRIVERS_INTEL
depends on X86_PLATFORM_DEVICES (through being part of a
"if X86_PLATFORM_DEVICES" Kconfig section), so we can just
replace the depends on, rather then adding a new one.

With those 2 things changed that error should be gone.

This requires an ack from Lee to merge the drivers/mfd/Kconfig
change through pdx86, so as discussed I agree that it is best
to delay moving the scu bits till after 5.15-rc1 is out.

> 
>>> - dropped extra prefix in PMC files
>>> - dropped stray changes within the series
>>> - removed confusing comments in Makefile and Kconfig files
>>> - embedded a few of Kconfig improvements here and there (ordering, spelling)
>>> - split miscellaneous group
>>> - added a patch to fix kernel doc issue in SCU IPC code
>>
>> Thanks, I've not (re)reviewed things, but this all sounds good.
>>
>>> So, I have noticed the report about SCU and taking into account IPS
>>> header deferred move, I think those two should be excluded from the
>>> series and sent after rc1, it will also eliminate trampoline move for
>>> IPS header, since we may send one patch which includes DRM changes.
>>>
>>> Hans, what do you think?
>>
>> Fixing the SCU thing will require coordination with / an 
>> ack from the MFD maintainer (Lee) so yeah dropping that for
>> now and doing the SCU move next cycle is probably the best.
>>
>> Dropping the IPS move for now is fine with me too.
>>
>> Can you send a v5 with those 2 patches dropped ?> 
> Yes.

Great, thank you.

Regards,

Hans