Re: [PATCH net-master 1/1] net: phy: dp83867: Add speed optimization feature

[Florian Fainelli]

On 1/31/20 11:54 AM, Dan Murphy wrote:
> Florian
> 
> On 1/31/20 1:29 PM, Florian Fainelli wrote:
>> On 1/31/20 11:14 AM, Dan Murphy wrote:
>>> Florian
>>>
>>> On 1/31/20 12:42 PM, Florian Fainelli wrote:
>>>> On 1/31/20 10:29 AM, Dan Murphy wrote:
>>>>> Florian
>>>>>
>>>>> On 1/31/20 11:49 AM, Florian Fainelli wrote:
>>>>>> On 1/31/20 7:11 AM, Dan Murphy wrote:
>>>>>>> Set the speed optimization bit on the DP83867 PHY.
>>>>>>> This feature can also be strapped on the 64 pin PHY devices
>>>>>>> but the 48 pin devices do not have the strap pin available to enable
>>>>>>> this feature in the hardware. PHY team suggests to have this bit
>>>>>>> set.
>>>>>> OK, but why and how does that optimization work exactly?
>>>>> I described this in the cover letter. And it is explained in the data
>>>>> sheet Section 8.4.6.6
>>>> Sorry I complete missed that and just focused on the patch, you should
>>>> consider not providing a cover letter for a single patch, and
>>>> especially
>>>> not when the cover letter contains more information than the patch
>>>> commit message itself.
>>> Sorry I usually give a cover letter to all my network related patches.
>>>
>>> Unless I misinterpreted David on his reply to me about cover letters.
>>>
>>> https://www.spinics.net/lists/netdev/msg617575.html
>> This was a 2 patches series, for which a cover letter is mandatory:
>>
>> but for single patches, there really is no need, and having to replicate
>> the same information in two places is just error prone.
>>
>>> And I seemed to have missed David on the --cc list so I will add him for
>>> v2.
>>>
>>> I was also asked not to provide the same information in the cover letter
>>> and the commit message.
>> The cover letter is meant to provide some background about choices you
>> have made, or how to merge the patches, or their dependencies, and
>> describe the changes in a big picture. The patches themselves are
>> supposed to be comprehensive.
> 
> As always thank you for the guidance. I will update the commit with
> better information and remove the cover letter.

No worries, every subsystem has its own rules, because why not :)

> 
> 
>>
>>> Either way I am ok with not providing a cover letter and updating the
>>> commit message with more information.
>>>
>>>
>>>>>> ÂÂÂ Departing from
>>>>>> the BMSR reads means you possibly are going to introduce bugs and/or
>>>>>> incomplete information. For instance, you set phydev->pause and
>>>>>> phydev->asym_pause to 0 now, is there no way to extract what the link
>>>>>> partner has advertised?
>>>>> I was using the marvel.c as my template as it appears to have a
>>>>> separate
>>>>> status register as well.
>>>>>
>>>>> Instead of setting those bits in the call back I can call the
>>>>> genphy_read_status then override the duplex and speed based on the
>>>>> physts register like below. This way link status and pause values can
>>>>> be updated and then we can update the speed and duplex settings.
>>>>>
>>>>> ÂÂÂ ÂÂÂ ret = genphy_read_status(phydev);
>>>>> ÂÂÂÂ Âif (ret)
>>>>> ÂÂÂÂ ÂÂÂ Âreturn ret;
>>>>>
>>>>> ÂÂÂÂ Âif (status < 0)
>>>>> ÂÂÂÂ ÂÂÂ Âreturn status;
>>>>>
>>>>> ÂÂÂÂ Âif (status & DP83867_PHYSTS_DUPLEX)
>>>>> ÂÂÂÂ ÂÂÂ Âphydev->duplex = DUPLEX_FULL;
>>>>> ÂÂÂÂ Âelse
>>>>> ÂÂÂÂ ÂÂÂ Âphydev->duplex = DUPLEX_HALF;
>>>>>
>>>>> ÂÂÂÂ Âif (status & DP83867_PHYSTS_1000)
>>>>> ÂÂÂÂ ÂÂÂ Âphydev->speed = SPEED_1000;
>>>>> ÂÂÂÂ Âelse if (status & DP83867_PHYSTS_100)
>>>>> ÂÂÂÂ ÂÂÂ Âphydev->speed = SPEED_100;
>>>>> ÂÂÂÂ Âelse
>>>>> ÂÂÂÂ ÂÂÂ Âphydev->speed = SPEED_10;
>>>>>
>>>> OK, but what if they disagree, are they consistently latched with
>>>> respect to one another?
>>> Well in parsing through the code for genphy read status when auto
>>> negotiation is set the phydev structure appears to be setup per what has
>>> been configured. I did not see any reading of speed or duplex when auto
>>> neg is set it is just taking the LPA register. But I am probably not
>>> right here. So we and our customers found that the phy was always
>>> reporting a 1Gbps connection when the 4 wire cable connected when using
>>> genphy_read_status. This PHYSTS register provides a single location
>>> within the register set for quick access to commonly accessed
>>> information.
>> That is the kind of information that you want to put in the commit
>> message, and that sounds like a Fix more than a feature to me. If the
>> BMSR is not reflecting the correct speed, clearly something is not quite
>> good. You may also consider reflecting whether downshift was in action
>> and that led to reducing the speed, something like
>> m88e1011_link_change_notify() does.
> 
> But what is it fixing? When the driver was originally submitted it was
> meant to be a Giga bit PHY. No requirement for any connection less then
> 1Gbps. Now we have a customer who wants to use this feature and they
> want it upstreamed. (YAY for them pushing upstream).
> 
> Do I reference my original commit? Because I am actually flipping the
> bits to turn it on as well only way this was a fix is if the user had
> the feature strapped to on. But to date we have not had any requests
> for this support.
> 
> So then it would be ok to do a genphy_read_status and then override the
> speed and duplex mode from the PHYSTS register?

I would think so yes, especially if that is needed for reporting the
actual link speed that ended up being negotiated, and not the one that
the link was initially trained at. That assumes I understand that the
problem is that you advertise and want Gigabit, but because of a 4-wire
cable being plugged in, you ended up at 100Mbits/sec.

> 
> I don't think that the link change notification is needed. The speed
> should not change once the cable is plugged in and the speed is negotiated.

The link change notification is just to signal to the user that the
speed may have been reduced due to downshifting, which would/could
happen with 4-wires instead of the expected 8-wires. Certainly not
strictly necessary right now, I agree.
-- 
Florian