Re: [PATCH 2/2] PCI: meson: add the Amlogic Meson PCIe phy driver

[Hanjie Lin <hanjie.lin@amlogic.com>]

On 2018/8/17 16:09, Jerome Brunet wrote:
> On Fri, 2018-08-17 at 14:12 +0800, Hanjie Lin wrote:
>>
>> On 2018/8/16 16:33, Jerome Brunet wrote:
>>> On Thu, 2018-08-16 at 11:05 +0800, Hanjie Lin wrote:
>>>>
>>>> On 2018/8/14 18:41, Jerome Brunet wrote:
>>>>> On Tue, 2018-08-14 at 02:12 -0400, Hanjie Lin wrote:
>>>>>> From: Yue Wang <yue.wang@amlogic.com>
>>>>>>
>>>>>> The Meson-PCIE-PHY controller supports the 5-Gbps data rate
>>>>>> of the PCI Express Gen 2 specification and is backwardcompatible
>>>>>> with the 2.5-Gbps Gen 1.1 specification with only
>>>>>> inferred idle detection supported on AMLOGIC SoCs.
>>>>>
>>>>> It looks like the sole purpose of this driver is to provide the reset lines to
>>>>> pcie driver.
>>>>>
>>>>> I wonder why we need this ? Can't the pcie driver claim the reset lines itself.
>>>>>
>>>>> Also, an init of this phy will always reset both port. What will happen if the
>>>>> first port is in use and the 2nd port comes up ?? 
>>>>>
>>>>> Looks the the pcie driver should claim 'apb' and 'phy' reset lines as "shared"
>>>>> reset and the required 'port' as 'exclusive'
>>>>>
>>>>
>>>> Thank you for your response.
>>>>
>>>> Yes, 'apb' and 'phy' reset lines are shared, and ‘port' reset line is exclusive.
>>>> If we handle all reset lines during the first port initial sequence, 
>>>> and when the second port comes up, we will do nothing about the rest lines, 
>>>> and don't need a extra API to do ‘port' reset;
>>>
>>> With which other driver are your control shared ?
>>>
>>> Looks it is the answer is none since this phy driver will reset both port
>>> already, even if one is used.
>>>
>>> In this case the fact that you are using shared control is just abusing the
>>> framework to reset once.
>>>
>>> As far as I can tell, this driver makes no sense. The appropriate reset lines
>>> should be given directly to your pcie driver. 
>>>
>>>
>>>
>>> .
>>>
>>
>> Amlogic AXG SOC includes two pcie controllers and pipes when only one pcie phy: 
>>
>>                                     (port_a reset)
>>                       |PCIE_RC_A---->PCIE_PIPE_A------| 
>>     (apb_reset)       |                               |  (phy reset)
>>     APB BUS--->       |                               |   PCIE_PHY
>>                       |                               |
>>                       |             (port_b_reset)    |
>>                       |PCIE_RC_B---->PCIE_PIPE_B------|
>>
>> The phy_reset affect the PCIE_PHY.
>> The port_a_reset affect the PCIE_PIPE_A, port_b_reset affect the PCIE_PIPE_B. 
>>
>> As your suggestion we will move the 'port' reset to controller driver,
>> and keeping the phy driver to process the 'apb' and 'phy' reset or any
>> more changes of the phy in future.
> 
> As far as I can tell from this diagram, It would only make sense for the "phy"
> reset line to be controlled by your phy driver.
> 
> The apb and port is obviously related to the pcie device/driver itself, not the
> PHY. And whether you 1 or 2 reset lines in it, IMO it is overkill and
> unnecessary to make a phy driver for this. 
> 

Yeah, that makes sense.
We will move 'apb' reset to controller driver in next version too.

Thanks.

>>
> 
> 
> .
>