Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH] net: phy: realtek: clear interrupt during init for
 rtl8211f
To:     Jisheng Zhang <Jisheng.Zhang@synaptics.com>,
        Florian Fainelli <f.fainelli@gmail.com>,
        Andrew Lunn <andrew@lunn.ch>,
        Russell King <linux@armlinux.org.uk>
Cc:     "David S. Miller" <davem@davemloft.net>, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
References: <20200512184601.40b1758a@xhacker.debian>
 <7735a257-21ff-e6c0-acdc-f5ee187b1f57@gmail.com>
 <20200513145151.04a6ee46@xhacker.debian>
 <a0aac85d-064f-2672-370b-404fd83e83f3@gmail.com>
 <20200514142537.63b478fd@xhacker.debian>
 <bbb70281-d477-d227-d1d2-aeecffdb6299@gmail.com>
 <20200515154128.41ee2afa@xhacker.debian>
 <18d7bdc7-b9f3-080f-f9df-a6ff61cd6a87@gmail.com>
 <e81ad573-ba30-a449-4529-d9a578ce0ee7@gmail.com>
 <20200617170926.5e582bad@xhacker.debian>
From:   Heiner Kallweit <hkallweit1@gmail.com>
Message-ID: <5657764c-3de8-667f-b4a7-5dbaf2e15303@gmail.com>
Date:   Sun, 21 Jun 2020 22:16:31 +0200
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:68.0) Gecko/20100101
 Thunderbird/68.9.0
MIME-Version: 1.0
In-Reply-To: <20200617170926.5e582bad@xhacker.debian>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 17.06.2020 11:09, Jisheng Zhang wrote:
> On Fri, 15 May 2020 19:30:38 +0200 Heiner Kallweit wrote:
> 
> 
>>
>>
>> On 15.05.2020 18:18, Florian Fainelli wrote:
>>>
>>>
>>> On 5/15/2020 12:41 AM, Jisheng Zhang wrote:  
>>>> On Thu, 14 May 2020 21:50:53 +0200 Heiner Kallweit wrote:
>>>>  
>>>>>
>>>>>
>>>>> On 14.05.2020 08:25, Jisheng Zhang wrote:  
>>>>>> On Wed, 13 May 2020 20:45:13 +0200 Heiner Kallweit wrote:
>>>>>>  
>>>>>>>
>>>>>>> On 13.05.2020 08:51, Jisheng Zhang wrote:  
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> On Tue, 12 May 2020 20:43:40 +0200 Heiner Kallweit wrote:
>>>>>>>>  
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On 12.05.2020 12:46, Jisheng Zhang wrote:  
>>>>>>>>>> The PHY Register Accessible Interrupt is enabled by default, so
>>>>>>>>>> there's such an interrupt during init. In PHY POLL mode case, the
>>>>>>>>>> INTB/PMEB pin is alway active, it is not good. Clear the interrupt by
>>>>>>>>>> calling rtl8211f_ack_interrupt().  
>>>>>>>>>
>>>>>>>>> As you say "it's not good" w/o elaborating a little bit more on it:
>>>>>>>>> Do you face any actual issue? Or do you just think that it's not nice?  
>>>>>>>>
>>>>>>>>
>>>>>>>> The INTB/PMEB pin can be used in two different modes:
>>>>>>>> INTB: used for interrupt
>>>>>>>> PMEB: special mode for Wake-on-LAN
>>>>>>>>
>>>>>>>> The PHY Register Accessible Interrupt is enabled by
>>>>>>>> default, there's always such an interrupt during the init. In PHY POLL mode
>>>>>>>> case, the pin is always active. If platforms plans to use the INTB/PMEB pin
>>>>>>>> as WOL, then the platform will see WOL active. It's not good.
>>>>>>>>  
>>>>>>> The platform should listen to this pin only once WOL has been configured and
>>>>>>> the pin has been switched to PMEB function. For the latter you first would
>>>>>>> have to implement the set_wol callback in the PHY driver.
>>>>>>> Or where in which code do you plan to switch the pin function to PMEB?  
>>>>>>
>>>>>> I think it's better to switch the pin function in set_wol callback. But this
>>>>>> is another story. No matter WOL has been configured or not, keeping the
>>>>>> INTB/PMEB pin active is not good. what do you think?
>>>>>>  
>>>>>
>>>>> It shouldn't hurt (at least it didn't hurt for the last years), because no
>>>>> listener should listen to the pin w/o having it configured before.
>>>>> So better extend the PHY driver first (set_wol, ..), and then do the follow-up
>>>>> platform changes (e.g. DT config of a connected GPIO).  
>>>>
>>>> There are two sides involved here: the listener, it should not listen to the pin
>>>> as you pointed out; the phy side, this patch tries to make the phy side
>>>> behave normally -- not keep the INTB/PMEB pin always active. The listener
>>>> side behaves correctly doesn't mean the phy side could keep the pin active.
>>>>
>>>> When .set_wol isn't implemented, this patch could make the system suspend/resume
>>>> work properly.
>>>>
>>>> PS: even with set_wol implemented as configure the pin mode, I think we
>>>> still need to clear the interrupt for phy poll mode either in set_wol
>>>> or as this patch does.  
>>>
>>> I agree with Jisheng here, Heiner, is there a reason you are pushing
>>> back on the change? Acknowledging prior interrupts while configuring the
>>> PHY is a common and established practice.
>>>  
>> First it's about the justification of the change as such, and second about the
>> question whether the change should be in the driver or in phylib.
>>
>> Acking interrupts we do already if the PHY is configured for interrupt mode,
>> we call phy_clear_interrupt() at the beginning of phy_enable_interrupts()
>> and at the end of phy_disable_interrupts().
>> When using polling mode there is no strict need to ack interrupts.
>> If we say however that interrupts should be acked in general, then I think
>> it's not specific to RTL8211F, but it's something for phylib. Most likely
>> we would have to add a call to phy_clear_interrupt() to phy_init_hw().
> 
> it's specific to RTL8211F from the following two PoV:
> 1. the PIN is shared between INTB and PMEB.
> 2. the PHY Register Accessible Interrupt is enabled by default
> 
If we clear the interrupt in config_init() and one interrupt source is
active per chip default, then wouldn't the irq pin be active soon again?

I was thinking about calling phy_disable_interrupts() in phy_init_hw(),
to have a defined init state, as we don't know in which state the PHY is
if the PHY driver is loaded. We shouldn't assume that it's the chip
power-on defaults, BIOS or boot loader could have changed this.
Or in case of dual-boot systems the other OS could leave the PHY in
whatever state.

This made me think about an issue we may have currently:
Interrupts are enabled in phy_request_interrupt() only. If the system
hibernates, then PHY may load power-on defaults on restore.
And mdio_bus_phy_restore() just calls phy_init_hw() and doesn't
care about interrupt config. Means after waking up from hibernation
we may have lost PHY interrupt config.

> I didn't see such behaviors with other PHYs.
> 
> Thanks
> 
Heiner