Message-ID: <20220901112721.GB2479@pengutronix.de>
Date:   Thu, 1 Sep 2022 13:27:21 +0200
From:   Oleksij Rempel <o.rempel@...gutronix.de>
To:     Arun.Ramadoss@...rochip.com
Cc:     olteanv@...il.com, andrew@...n.ch, linux-kernel@...r.kernel.org,
        UNGLinuxDriver@...rochip.com, vivien.didelot@...il.com,
        san@...v.dk, linux@...linux.org.uk, f.fainelli@...il.com,
        kuba@...nel.org, edumazet@...gle.com, pabeni@...hat.com,
        netdev@...r.kernel.org, Woojung.Huh@...rochip.com,
        davem@...emloft.net
Subject: Re: [Patch net-next v2 0/9] net: dsa: microchip: add support for
 phylink mac config and link up

On Thu, Sep 01, 2022 at 08:51:44AM +0000, Arun.Ramadoss@...rochip.com wrote:
> On Wed, 2022-08-31 at 18:10 +0200, Oleksij Rempel wrote:
> > EXTERNAL EMAIL: Do not click links or open attachments unless you
> > know the content is safe
> > 
> > On Wed, Aug 31, 2022 at 06:18:59PM +0300, Vladimir Oltean wrote:
> > > On Wed, Aug 31, 2022 at 09:43:24AM +0200, Oleksij Rempel wrote:
> > > > > > Should we point them to ksz8795_xmii_ctrl0 and
> > > > > > ksz8795_xmii_ctrl1? I don't know.
> > > > > > Could you find out what these should be set to?
> > > > > 
> > > > > xmii_ctrl0/1 are missing and it make no sense to add it.
> > > > > KSZ8873 switch is controlling CPU port MII configuration over
> > > > > global,
> > > > > not port based register.
> > > > > 
> > > > > I'll better define separate ops for this chip.
> > > > 
> > > > Hm, not only KSZ8830/KSZ8863/KSZ8873 are affected. ksz8795
> > > > compatible
> > > > series with defined .xmii_ctrl0/.xmii_ctrl1 are broken too.
> > > > Because it
> > > > is writing to the global config register over ksz_pwrite8
> > > > function. It
> > > > means, we are writing to 0xa6 instead of 0x06. And to 0xf6
> > > > instead of
> > > > 0x56.
> > > What do you mean by "global config register"? The Is_1Gbps bit is
> > > still
> > > part of a port register, it's just that this particular register is
> > > only
> > > defined for the 5th port (port #4, the only xMII port on KSZ7895
> > > AFAIU).
> > > That doesn't necessarily make it a "global" register.
> > > 
> > > Datasheet says:
> > > 
> > > Register 22 (0x16): Reserved
> > > Register 38 (0x26): Reserved
> > > Register 54 (0x36): Reserved
> > > Register 70 (0x46): Reserved
> > > Register 86 (0x56): Port 5 Interface Control 6
> > > 
> > > I wonder if it's ok to modify the regs table like this, because we
> > > should then only touch P_XMII_CTRL_1 using port 4:
> > > 
> > >  static const u16 ksz8795_regs[] = {
> > >       [REG_IND_CTRL_0]                = 0x6E,
> > >       [REG_IND_DATA_8]                = 0x70,
> > >       [REG_IND_DATA_CHECK]            = 0x72,
> > >       [REG_IND_DATA_HI]               = 0x71,
> > >       [REG_IND_DATA_LO]               = 0x75,
> > >       [REG_IND_MIB_CHECK]             = 0x74,
> > >       [REG_IND_BYTE]                  = 0xA0,
> > >       [P_FORCE_CTRL]                  = 0x0C,
> > >       [P_LINK_STATUS]                 = 0x0E,
> > >       [P_LOCAL_CTRL]                  = 0x07,
> > >       [P_NEG_RESTART_CTRL]            = 0x0D,
> > >       [P_REMOTE_STATUS]               = 0x08,
> > >       [P_SPEED_STATUS]                = 0x09,
> > >       [S_TAIL_TAG_CTRL]               = 0x0C,
> > >       [P_STP_CTRL]                    = 0x02,
> > >       [S_START_CTRL]                  = 0x01,
> > >       [S_BROADCAST_CTRL]              = 0x06,
> > >       [S_MULTICAST_CTRL]              = 0x04,
> > >       [P_XMII_CTRL_0]                 = 0x06,
> > > -     [P_XMII_CTRL_1]                 = 0x56,
> > > +     [P_XMII_CTRL_1]                 = 0x06,
> > >  };
> > > 
> > 
> > Speed configuration on ksz8795 is done over two registers:
> > Register 86 (0x56): Port 5 Interface Control 6: Is_1Gbps - BIT(6)
> > and
> > Register 6 (0x06): Global Control 4: Switch SW5-MII/RMII Speed
> > -BIT(4)
> > 
> > both are accessed on wrong offsets.
> > 
> > I would prefer to do following steps:
> > - remove everything from ksz_phylink_mac_link_up() except of
> >   dev->dev_ops->phylink_mac_link_up
> > - move ksz_duplex_flowctrl(), ksz_port_set_xmii_speed()... to
> > ksz9477.c
> >   and rename them. Assign ksz9477_phylink_mac_link_up()
> >   dev->dev_ops->phylink_mac_link_up
> > - create separate function ksz8795_phylink_mac_link_up()
> > - use documented, not generic register names.
> > 
> 
> In the original code, the ksz8795_cpu_interface_select (which does the
> xmii and speed configuration) is called only ksz87xx switch not for the
> ksz88x3.
> During refactoring, I intentionally did not add the P_XMII_CTRL0/1 for
> the ksz88xx in the chip_data structure.
> I added check that if switch belong to ksz88x3 then return in the
> phylink_mac_config function but I missed to add in the
> phylink_mac_link_up. Thats the mistake I have done. So, for the ksz88xx
> switch code breakage, it could be fixed by adding the check in the
> phylink_mac_link_up as well.
> 
> For the code breakage in ksz8795/ksz8794, the original code has only
> provision for configuring xmii mode and choosing the 100/1000Mbps speed
> selection which could be selected using is_1Gbps bit of
> port5_interface_control6 register (0x56). But it didn't have provision
> to select speed between 10/100Mbps, flow control and duplex.
> 
> As per vladimir suggestion, P_XMII_CTRL1 can be changed from 0x56 to
> 0x06. It fixes the problem for ksz_set_xmii, ksz_set_gbit since this is
> the port based register not the global register.
> 
> The global register 0x06 responsibilities are bit 4 for 10/100mbps
> speed selection, bit 5 for flow control and bit 6  for duplex
> operation. Since these three are new features added during refactoring
> I overlooked it. 
> To fix this, either I need to return from the ksz_set_100_10mbit &
> ksz_duplex_flowctrl function if the chip_id is ksz87xx or add dev-
> >dev_ops for this alone. 
> Kindly suggest on how to proceed.

Do not worry, every one makes mistakes ;)

I just scared how easy it is to make them in this driver. I assume, that
initially the good idea to make things more generic in this driver is
working against us. The register- and bit-mapping functionally is
already biting us:
- it is hard to review, due to different register naming schema for
  different switch families. Especially the old ones.
- we are using naming from one documentation to map to other
  documentation.
- some times we need to map some registers multiple times with different
  names: P_REMOTE_STATUS and P_LINK_STATUS

I would prefer to got ops way, to clean things up.

Regards,
Oleksij
-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

Powered by blists - more mailing lists