Re: [PATCH net-next v3 4/7] net: dsa: mt7530: move XTAL check to mt7530_setup()

["Russell King (Oracle)" <linux@armlinux.org.uk>]

On Sun, Feb 04, 2024 at 04:55:40PM +0300, Arınç ÜNAL wrote:
> On 2.02.2024 21:39, Daniel Golle wrote:
> > On Fri, Feb 02, 2024 at 09:16:02PM +0300, Arınç ÜNAL wrote:
> > > On 2.02.2024 14:48, Russell King (Oracle) wrote:
> > > > On Fri, Feb 02, 2024 at 12:19:10PM +0300, Arınç ÜNAL via B4 Relay wrote:
> > > > > From: Arınç ÜNAL <arinc.unal@arinc9.com>
> > > > > 
> > > > > The crystal frequency concerns the switch core. The frequency should be
> > > > > checked when the switch is being set up so the driver can reject the
> > > > > unsupported hardware earlier and without requiring port 6 to be used.
> > > > > 
> > > > > Move it to mt7530_setup(). Drop the unnecessary function printing.
> > > > > 
> > > > > Signed-off-by: Arınç ÜNAL <arinc.unal@arinc9.com>
> > > > > Reviewed-by: Andrew Lunn <andrew@lunn.ch>
> > > > > Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
> > > > 
> > > > I would prefer this to be earlier in the series, before patch 2 which
> > > > moves mt7530_setup_port6() to be called from mac_config(). mac_config()
> > > > is supposed to be configuration error-free - in other words, all state
> > > > should have been checked before hand.
> > > 
> > > I agree but mt7530_mac_config() is not a void function yet. The
> > > mac_port_config member of the mt753x_info structure points to this
> > > function. My next patch series gets rid of all useless error returns on the
> > > phylink path and change mac_port_config to void. So I don't think working
> > > on this patch series further will worth the effort. I'd rather have this
> > > version applied as is.
> > 
> > I agree regarding not changing the patch itself, but I also agree
> > with Russell regarding the patch ordering. I know it's a 10-minute
> > git headache to rebase the patches on top of each other in a different
> > order, but you can easily compare the end result being identical to
> > what you had before and hence don't need to retest.
> 
> This is not about laziness. This is before patch 2:
> 
> phylink_mac_ops :: mac_config() -> dsa_port_phylink_mac_config()
> -> dsa_switch_ops :: phylink_mac_config() -> mt753x_phylink_mac_config()
>    -> mt753x_mac_config()
>       -> mt753x_info :: mac_port_config() -> mt7530_mac_config()
>          -> mt7530_setup_port5()
>    -> mt753x_pad_setup()
>       -> mt753x_info :: pad_setup() -> mt7530_pad_clk_setup()
> 
> This is after:
> 
> phylink_mac_ops :: mac_config() -> dsa_port_phylink_mac_config()
> -> dsa_switch_ops :: phylink_mac_config() -> mt753x_phylink_mac_config()
>    -> mt753x_mac_config()
>       -> mt753x_info :: mac_port_config() -> mt7530_mac_config()
>          -> mt7530_setup_port5()
>          -> mt7530_setup_port6()
> 
> Patch 2 does not move mt7530_setup_port6() to be called from
> phylink_mac_ops :: mac_config(), it already is. There is no valid reason to
> reorder the patches.
> 
> My response to Russell should've stated this instead of focusing on his
> second sentence.

This patch moves the test for a 20MHz crystal to mt7530_setup(),
which is something that is entirely orthogonal to patch 2, which
can be done cleanly (I've just applied the patches in the original
order and then reordered them:

98c481f5d706 net: dsa: mt7530: do not clear config->supported_interfaces
93c6b53b17f4 net: dsa: mt7530: correct port capabilities of MT7988
c9c6d4c51a1d net: dsa: mt7530: simplify mt7530_setup_port6() and change to void
adfa948253e0 net: dsa: mt7530: remove pad_setup function pointer
57e21e6c2fc0 net: dsa: mt7530: call port 6 setup from mt7530_mac_config()
959a0f9323c8 net: dsa: mt7530: move XTAL check to mt7530_setup()
856ab64a22ef net: dsa: mt7530: empty default case on mt7530_setup_port5()

No problems. The end result is identical comparing the git tree at the
original "move XTAL" patch with adfa948253e0.

Now, if we look at "net: dsa: mt7530: remove pad_setup function pointer"
we can see that yes, the pad_setup() method was called from mac_confing,
but this is the exact contents of that patch removing the callsite:

-               mt753x_pad_setup(ds, state);

This returns an integer, which may be an error code, which is ignored.
Therefore, if the XTAL frequency check fires, and mt753x_pad_setup()
returns an error, it is ignored today.

After "net: dsa: mt7530: call port 6 setup from mt7530_mac_config()"
the renamed pad_setup() method is now called from mac_config() thusly:

+               ret = mt7530_setup_port6(priv->ds, interface);
+               if (ret)
+                       return ret;

So now the error checks cause mt7530_mac_config() to return an error
which in turn causes mt753x_mac_config() to fail, and therefore
mt753x_phylink_mac_config() has different behaviour.

So, patch 2 changes the driver behaviour in the case of a 20MHz XTAL,
which is then changed again by patch 4.

It would be better to have only one change of behaviour by moving
patch 4 before patch 2.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!