Re: [PATCH -next] xtensa: Fix build error due to missing include file

From: Max Filippov
Date: Mon Jul 25 2016 - 16:02:24 EST


On Mon, Jul 25, 2016 at 4:22 PM, Rob Herring <robh@xxxxxxxxxx> wrote:
> On Mon, Jul 25, 2016 at 3:07 AM, Max Filippov <jcmvbkbc@xxxxxxxxx> wrote:
>> On Sun, Jul 24, 2016 at 7:27 PM, Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
>>> On 07/24/2016 08:18 AM, Rob Herring wrote:
>>>> On Sun, Jul 24, 2016 at 9:20 AM, Max Filippov <jcmvbkbc@xxxxxxxxx> wrote:
>>>>> On Sun, Jul 24, 2016 at 9:00 AM, Guenter Roeck <linux@xxxxxxxxxxxx>
>>>>> wrote:
>>>>>> Heads up: xtensa images build after this patch, but the image that
>>>>>> previously failed to build (generic_kc705_defconfig) doesn't boot,
>>>>>> at least not with qemu. I don't know if this is caused by 69d99e6c0d621f
>>>>>> or by some other patch.
>>>>>
>>>>> It seems to be booting, but serial console is non-functional.
>>>>> With commit 69d99e6c0d621f ("xtensa: Remove unnecessary
>>>>> of_platform_populate
>>>>> with default match table") I see the following message during boot:
>>>>>
>>>>> [ 6.375284] of_serial fd050020.serial: clk or clock-frequency not
>>>>> defined
>>>>>
>>>>> Reverting that commit fixes serial console for me.
>>>>> Any suggestions on what might go wrong?
>>>>
>>>> Probably the init order changed somehow and deferred probe is not
>>>> supported correctly.
>>>
>>> While the message is due to -EPROBE_DEFER (and there should really be no
>>> message
>>> in this case), I suspect that the underlying problem is that of_clk_init()
>>> is
>>> now called way too early, before the clock source is fully initialized.
>>> Specifically, the call now happens before the message "clocksource: Switched
>>> to clocksource ccount", which at least looks suspicious.
>>>
>>> If I move of_clk_init() back into xtensa_device_probe(), to be called as
>>> device_initcall(), everything is fine.
>>
>> The problem is in the way the main clock on xtfpga platform is set up:
>> it is done by updating corresponding OF node's clock-frequency attribute
>> and that must happen before the of_clk_init call.
>> Looks like it's time to create a clock provider for xtfpga boards, something
>> like the one attached.
>
> I don't think we should require a DT update to fix this. I think we
> should revert all but the removing of of_platform_populate line for
> now.

Fine with me too. Should I send the revert?

--
Thanks.
-- Max