[PATCH net-next v5 3/3] net: stmmac: Add DWMAC glue layer for Renesas GBETH

[Biju Das]

Hi Andrew,

> -----Original Message-----
> From: Biju Das
> Sent: 21 April 2025 18:26
> Subject: RE: [PATCH net-next v5 3/3] net: stmmac: Add DWMAC glue layer for Renesas GBETH
> 
> Hi Andrew,
> 
> > -----Original Message-----
> > From: Andrew Lunn <andrew at lunn.ch>
> > Sent: 21 April 2025 16:12
> > Subject: Re: [PATCH net-next v5 3/3] net: stmmac: Add DWMAC glue layer
> > for Renesas GBETH
> >
> > On Mon, Apr 21, 2025 at 02:22:01PM +0000, Biju Das wrote:
> > > Hi Andrew,
> > >
> > > > -----Original Message-----
> > > > From: Andrew Lunn <andrew at lunn.ch>
> > > > Sent: 21 April 2025 15:02
> > > > Subject: Re: [PATCH net-next v5 3/3] net: stmmac: Add DWMAC glue
> > > > layer for Renesas GBETH
> > > >
> > > > > > On the RZ/G3E, the upstream support for testing S2R is not yet
> > > > > > in a usable state. So for now, I'll switch to using init/exit
> > > > > > callbacks and drop the PM
> > callback.
> > > > >
> > > > > FYI, On RZ/G3E, for STR to work with mainline, we need to reinitialize the PHY.
> > > > > I have done below changes on top of [1] to make STR working.
> > > >
> > > > Can you explain why you need to reinitialise the PHY? The MAC
> > > > driver should not need to do this, so something is wrong
> > > > somewhere. If we understand the 'Why?' we can probably tell you a better way to do this.
> > >
> > > Without this change bind/unbind works. But for the STR case, without
> > > reinitializing the PHY, even though the IP link is UP, I am not able
> > > to talk the NFS server or ping
> > the host properly.
> > >
> > > I checked clock/reset before and after reset everything set as expected.
> > >
> > > Only change during STR is, on wakeup we need to restore direction
> > > (MII/RGMII) of IO block for ET0/1_TXC_TXCLK (IO attribute) in the
> > > pin control driver. After that looks like PHY init is required to talk to server.
> >
> > When talking about suspend/resume, is this with or without WoL?
> 
> Without WoL.
> 
> >
> > If WoL is enabled for the PHY, the PHY is left running while the
> > system is suspended, and so all its clocks and reset lines also need
> > to be left enabled etc. On resume, there should not be any need to touch the PHY.
> 
> OK.
> 
> >
> > If WoL is not enabled in the PHY, it should get powered off. On
> > resume, phylib should be reinitializing the PHY.
> 
> OK.
> 
> >
> > Which of these two cases need the reinitialisation?
> >
> > The reasons i can think of for the PHY needing a reinitialization are:
> >
> > 1) It lost power when it did not expect to loose power
> > 2) Got reset when it did not expect to be reset
> > 3) Clock not ticking when it should of been ticking.
> 
> OK.
> 
> >
> > So you cannot just check clock/reset before and after, you need to
> > look at the order of events. The loss of power, or a reset after
> > phylib resumed the PHY would not be good. Similarly, if the needed clocks are not ticking while
> phylib resumes it would also not be good.
> >
> > I would also suggest you check the datasheet for the PHY, and does it
> > document anything about the clock input TXC_TXCLK is connected to?
> 
> It is connected to TX_CTL on micrel phy.
> 
> > Does it need to be ticking while configuring the PHY? Any action which
> > needs to be taken after this starts ticking? Is the pinctrl resume being called before the PHY
> resume?
> 
> Pinctrl resume is called before PHY resume
> 
> Previously STR did not work, because of not restoring direction (MII/RGMII) of IO block for
> ET0/1_TXC_TXCLK (IO attribute) .Because of this, the direction of IO block is set to IN (input) MII
> mode instead of OUT(output) RGMII mode.
> 
> Now everything works. Thanks for your detailed pointers.

I need to debug this issue further as without reinitializing the PHY, the STR wakeup is not stable
(Linkup, but not able to contact the server).

[   34.572232] dwmac4: Master AXI performs fixed burst length
[   34.572232] renesas-gbeth 15c30000.ethernet eth0: No Safety Features support found
[   34.572232] renesas-gbeth 15c30000.ethernet eth0: IEEE 1588-2008 Advanced Timestamp supported
[   34.572232] renesas-gbeth 15c30000.ethernet eth0: configuring for phy/rgmii-id link mode
[   34.572232] OOM killer enabled.
[   34.572232] Restarting tasks ... done.
[   34.572232] random: crng reseeded on system resumption
[   34.572232] PM: suspend exit

root at smarc-rzg3e:~#
root at smarc-rzg3e:~# ls
[   34.572232] renesas-gbeth 15c30000.ethernet eth0: Link is Up - 1Gbps/Full - flow control rx/tx
root at smarc-rzg3e:~# [   34.572232] mmc2: Skipping voltage switch

root at smarc-rzg3e:~#
root at smarc-rzg3e:~#
root at smarc-rzg3e:~# ping 192.168.10.1

[   34.572232] nfs: server 192.168.10.1 not responding, still trying


Cheers,
Biju