Re: [PATCH v5 1/6] clk: hisilicon: add CRG driver for hi3519 soc

[xuejiancheng <xuejiancheng@huawei.com>]

On 2016/1/20 14:38, Tomeu Vizoso wrote:
> On 19 January 2016 at 19:20, Rob Herring <robh+dt@kernel.org> wrote:
>> On Fri, Jan 15, 2016 at 1:57 AM, xuejiancheng <xuejiancheng@huawei.com> wrote:
>>>
>>> On 2016/1/14 21:16, xuejiancheng wrote:
>>>> Hi Mike,
>>>>
>>>> On 2016/1/14 2:57, Michael Turquette wrote:
>>>>> Quoting xuejiancheng (2016-01-12 19:03:01)
>>>>>> Hi Stephen,
>>>>>>    Thank you very much for your reply.
>>>>>>
>>>>>> On 2016/1/13 6:12, Stephen Boyd wrote:
>>>>>>> On 01/08, Jiancheng Xue wrote:
>>>>>>>> diff --git a/drivers/clk/hisilicon/Kconfig b/drivers/clk/hisilicon/Kconfig
>>>>>>>> index e434854..b6baebf 100644
>>>>>>>> --- a/drivers/clk/hisilicon/Kconfig
>>>>>>>> +++ b/drivers/clk/hisilicon/Kconfig
>>>>>>>> @@ -1,3 +1,10 @@
>>>>>>>> +config COMMON_CLK_HI3519
>>>>>>>> +    tristate "Clock Driver for Hi3519"
>>>>>>>
>>>>>>> It looks like this has to be bool. Otherwise it needs to be a
>>>>>>> platform driver and the hisilicon APIs need to be exported and
>>>>>>> lose their __init markings.
>>>>>>>
>>>>>> Yes,it's a problem. I will fix it in next version. Thank you.
>>>>>
>>>>> The best solution would be to make this clock driver a real platform
>>>>> driver.
>>>>>
>>>> Now the work clock of the clocksource timer-sp804 is provided by this driver. So
>>>> it need to be registered early by CLK_OF_DECLARE. If the timer clock is treated
>>>> as a fixed-clock provider, this driver can be implemented as a platform driver.
>>>> Then the crg device must be registered before other clock consumer devices.Accordingly
>>>> the crg device node must be written above all other clock consumer devices node in dts files.
>>>> I think it is also a dependence.
>>>>
>>>> Can you help me understand why it is better to make this driver a platform driver?
>>>> Thank you very much!
>>>>
>>> arch_initcall(customize_machine)
>>> -->of_platform_populate
>>>    -->of_platform_bus_create
>>>      -->of_amba_device_create
>>>         -->amba_device_add
>>>            -->amba_get_enable_pclk
>>> The call sequence above shows that the clock of the amba device must be registered before
>>> amba_device_add. The clock of "arm,pl011" uart is registered in the probe function of the
>>> platform driver "hi3519-crg". So the platform device "hi3519-crg" must be created before
>>> the amba device "arm,pl011" uart.
>>
>> It is a problem, but Tomeu had a fix to support deferred probes here.
>> That was part of the on-demand probing series, but maybe it needs to
>> be applied separately if we are moving clock drivers to platform
>> drivers.
> 
> Hi,
> 
> Marek Szyprowski has kindly taken those two patches as part of a series of him:
> 
> http://lkml.kernel.org/g/1450868368-5650-1-git-send-email-m.szyprowski@samsung.com
> 
> I think it would be great if you could test them and report.
> 
Hi Tomeu,

I have applied the patch "https://lkml.org/lkml/2015/12/23/105" and tested on my hi3519-demb board.
It works even if the apb_pclk is registered later than the amba-pl011 device being registered.

But I think it is a problem if amba_read_periphid() returns -ENOMEM or -ENODEV when apb_pclk is available.
In this condition，amba_match() returns a non zero value which means the driver and device matches
and the amba_probe() will be called, but amba_device->periphid remains as 0. Then amba_lookup() called in
amba_probe() will return a null id pointer.The null pointer will be passed to amba_driver->probe() and
this may cause a segment fault.

Regards,

Jiancheng

> Thanks,
> 
> Tomeu
> 
> .
>