Re: [PATCH net-next 2/3] net: phy: add Broadcom BCM54140 support

[Michael Walle]

Hi Vladimir,

Am 2020-04-17 22:00, schrieb Vladimir Oltean:
> Hi Michael,
>
> On Fri, 17 Apr 2020 at 22:52, Michael Walle <michael@xxxxxxxx> wrote:
> > Hi Andrew,
> >
> > Am 2020-04-17 21:39, schrieb Andrew Lunn:
> > > On Fri, Apr 17, 2020 at 09:28:57PM +0200, Michael Walle wrote:
> > >
> > >> +static int bcm54140_get_base_addr_and_port(struct phy_device *phydev)
> > >> +{
> > >> +    struct bcm54140_phy_priv *priv = phydev->priv;
> > >> +    struct mii_bus *bus = phydev->mdio.bus;
> > >> +    int addr, min_addr, max_addr;
> > >> +    int step = 1;
> > >> +    u32 phy_id;
> > >> +    int tmp;
> > >> +
> > >> +    min_addr = phydev->mdio.addr;
> > >> +    max_addr = phydev->mdio.addr;
> > >> +    addr = phydev->mdio.addr;
> > >> +
> > >> +    /* We scan forward and backwards and look for PHYs which have the
> > >> +     * same phy_id like we do. Step 1 will scan forward, step 2
> > >> +     * backwards. Once we are finished, we have a min_addr and
> > >> +     * max_addr which resembles the range of PHY addresses of the same
> > >> +     * type of PHY. There is one caveat; there may be many PHYs of
> > >> +     * the same type, but we know that each PHY takes exactly 4
> > >> +     * consecutive addresses. Therefore we can deduce our offset
> > >> +     * to the base address of this quad PHY.
> > >> +     */
> > >
> > > Hi Michael
> > >
> > > How much flexibility is there in setting the base address using
> > > strapping etc? Is it limited to a multiple of 4?
> >
> > You can just set the base address to any address. Then the following
> > addresses are used:
> >    base, base + 1, base + 2, base + 3, (base + 4)*
> >
> > It is not specified what happens if you set the base so that it would
> > overflow. I guess that is a invalid strapping.
> >
> > * (base + 4) is some kind of special PHY address which maps some kind
> > of moving window to a QSGMII address space. It is enabled by default,
> > could be disabled in software, but it doesn't share the same PHY id
> > for which this scans.
> >
> > So yes, if you look at the addresses and the phy ids, there are
> > always 4 of this.
> >
> > -michael
>
> What does the reading of the global register give you, when accessed
> through the master PHY ID vs any other PHY ID? Could you use that as
> an indication of this being the correct PHY ID, and scan only to the
> left?

That was my first try, I thought it reads zero if you access a global
register by a PHY address which is not the base one. So I've looked
at registers which have at least one read-only 1 bit in it and scanned
only backwards. Well it turns out, my assumption was wrong and it
returns an old value of a successful read/write before. So it can just
return anything. And yes, its likely that you could read another
register and then probe the global register. But in the end I preferred
scanning the (known) phy id registers over strange hacks. Broadcom
could have just added a per-port register to actually read the base
address, but well.. ;)

-michael