Re: [PATCH 03/17] base: fix order of OF initialization

[Frank Rowand <frowand.list@gmail.com>]

On 06/08/17 02:05, Mark Rutland wrote:
> On Wed, Jun 07, 2017 at 08:49:43PM -0700, Frank Rowand wrote:
>> On 06/07/17 11:39, Wesley Terpstra wrote:
>>> It was a while ago that I debugged this. I already reported this bug
>>> to Benjamin Herrenschmidt (now in CC), and I believe he has a patch of
>>> his own to fix the same issue.
>>>
>>> As I understand it, of_core_init sets up the OF entries in
>>> /sys/firmware/devicetree. During platform bringup, when the system
>>> describes the cpu + cache hierarchy, it also makes an of_node symlink
>>> into that directory. However, if it doesn't exist yet, you get the
>>> warning.
>>>
>>> # ls -l /sys/devices/system/cpu/cpu3/of_node
>>> lrwxrwxrwx    1 root     root             0 Jan  1 00:00
>>> /sys/devices/system/cpu/cpu3/of_node ->
>>> ../../../../firmware/devicetree/base/cpus/cpu@3
>>>
>>> On Wed, Jun 7, 2017 at 2:35 AM, Mark Rutland <mark.rutland@arm.com> wrote:
>>>> On Wed, Jun 07, 2017 at 09:07:20AM +0200, Geert Uytterhoeven wrote:
>>>>> CC devicetree folks
>>>>>
>>>>> On Wed, Jun 7, 2017 at 12:59 AM, Palmer Dabbelt <palmer@dabbelt.com> wrote:
>>>>>> From: "Wesley W. Terpstra" <wesley@sifive.com>
>>>>>>
>>>>>> This fixes: [    0.010000] cpu cpu0: Error -2 creating of_node link
>>>>>> ... which you get for every CPU on all architectures with a OF cpu/ node.
>>>>
>>>> I take it this means a /cpus node? Or the /cpus/cpu@* nodes?
>>>>
>>>> I'm not seeing this on arm64 when booting v4.12-rc4 with DT, so clearly
>>>> this doesn't affect all such architectures.
>>>>
>>>> What path are these errors happening in?
>>
>> On the surface, the patch looks reasonable.  But it is not obvious to me why
>> the error message is occurring.  I would like to understand the cause before
>> saying the patch is good.
>>
>> What kernel version is showing the error?  For a specific architecture
>> (the patch lists 'riscv, nios, etc'), which config and device tree source?
>>
>> And again, what is the calling path?
> 
>>From having grepped around, I think this affects architectures which
> select CONFIG_GENERIC_CPU_DEVICES, which includes nios2.

Thanks Mark!  The "#ifdef CONFIG_GENERIC_CPU_DEVICES" in cpu_dev_register_generic()
explains why we don't see the error on ARM, ARM64, etc.  Without the CONFIG
option, register_cpu(cpu, i) is not called.


> In that case, driver_init() calls cpu_dev_init() before calling
> of_core_init(). Then we get the callchain:
> 
>    cpu_dev_init()
> -> cpu_dev_register_generic()
> -> register_cpu(cpu, i)
> -> device_register(&cpu->dev)
> -> device_add(dev)
> -> device_add_class_symlinks(dev)
> 
> ... in device_add_class_symlinks, we we dev->of_node, and call
> sysfs_create_link(), which fails because we haven't called
> of_core_init() to register the sysfs devicetree directory yet.
> 
> Given that, this patch makes sense to me.
> 
> FWIW, with the commit message updated to describe the particular
> ordering problem:
> 
> Acked-by: Mark Rutland <mark.rutland@arm.com>

Agree with Mark's request to update the commit message, and also
I would like to see this shaken out in the -next tree.

Acked-by: Frank Rowand <frowand.list@gmail.com>

> 
> Thanks,
> Mark.
>