Re: [PATCH v2 1/2] net: phylink: init phydev on phylink_resume()

[Claudiu.Beznea]

On 12.12.2022 16:24, Russell King (Oracle) wrote:
> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
> 
> On Mon, Dec 12, 2022 at 01:26:54PM +0000, Claudiu.Beznea@xxxxxxxxxxxxx wrote:
>> On 12.12.2022 14:47, Russell King (Oracle) wrote:
>>> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
>>>
>>> On Mon, Dec 12, 2022 at 01:28:44PM +0200, Claudiu Beznea wrote:
>>>> There are scenarios where PHY power is cut off on system suspend.
>>>> There are also MAC drivers which handles themselves the PHY on
>>>> suspend/resume path. For such drivers the
>>>> struct phy_device::mac_managed_phy is set to true and thus the
>>>> mdio_bus_phy_suspend()/mdio_bus_phy_resume() wouldn't do the
>>>> proper PHY suspend/resume. For such scenarios call phy_init_hw()
>>>> from phylink_resume().
>>>>
>>>> Suggested-by: Russell King (Oracle) <linux@xxxxxxxxxxxxxxx>
>>>> Signed-off-by: Claudiu Beznea <claudiu.beznea@xxxxxxxxxxxxx>
>>>> ---
>>>>
>>>> Hi, Russel,
>>>>
>>>> I let phy_init_hw() to execute for all devices. I can restrict it only
>>>> for PHYs that has struct phy_device::mac_managed_phy = true.
>>>>
>>>> Please let me know what you think.
>>>
>>> I think it would be better to only do this in the path where we call
>>> phy_start() - if we do it in the WoL path (where the PHY remains
>>> running), then there is no phy_start() call, so phy_init_hw() could
>>> result in the PHY not working after a suspend/resume event.
>>
>> This will not work all the time for MACB usage on AT91 devices.
>>
>> As explained here [1] the scenario where:
>> - MACB is configured to handle WoL
>> - the system goes to a suspend mode (named backup and self-refresh (BSR) in
>>   our case) with power cut off on PHY and limited wake-up source (few pins
>>   and RTC alarms)
>>
>> is still valid. In this case MAC IP and MAC PHY are not powered. And in
>> those cases phylink_resume() will not hit phy_start().
> 
> If the MAC is handling WoL, how does the MAC receive the packet to
> wake up if the PHY has lost power?

Yes, this can't happen.

> 
> If the PHY loses power, the MAC won't be able to receive the magic
> packet, and so WoL will be non-functional, and therefore will be
> completely pointless to support in such a configuration.
> 
> What am I missing?

As explained in the previous version [1], we currently don't impose any
restriction like this in arch specific PM code (the only place where we can
decide if going to BSR and device_may_wakeup()). I had in mind to do it at
some point but though that user will have to re-do all the wakeup sources
reconfiguration when switching b/w suspend modes that cut or not PHY power.
To be consistent this will have to be done for all devices registered as
wakeup sources (except the few pins and RTC in case). And this may be a
pain for users.

[1]
https://lore.kernel.org/all/4375d733-ed49-869c-635f-0f0ba7304283@xxxxxxxxxxxxx/

> 
> --
> RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
> FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!