Re: [PATCH net-next v4 2/2] of: net: fix of_get_mac_addr_nvmem() for non-platform devices

[Michael Walle]

Am 2021-04-16 17:19, schrieb Rob Herring:
> On Fri, Apr 16, 2021 at 2:30 AM Michael Walle <michael@xxxxxxxx> wrote:
> > Am 2021-04-16 05:24, schrieb Benjamin Herrenschmidt:
> > > On Mon, 2021-04-12 at 19:47 +0200, Michael Walle wrote:
> > >>
> > >>  /**
> > >>   * of_get_phy_mode - Get phy mode for given device_node
> > >> @@ -59,15 +60,39 @@ static int of_get_mac_addr(struct device_node *np,
> > >> const char *name, u8 *addr)
> > >>  static int of_get_mac_addr_nvmem(struct device_node *np, u8 *addr)
> > >>  {
> > >>         struct platform_device *pdev = of_find_device_by_node(np);
> > >> +       struct nvmem_cell *cell;
> > >> +       const void *mac;
> > >> +       size_t len;
> > >>         int ret;
> > >>
> > >> -       if (!pdev)
> > >> -               return -ENODEV;
> > >> +       /* Try lookup by device first, there might be a
> > >> nvmem_cell_lookup
> > >> +        * associated with a given device.
> > >> +        */
> > >> +       if (pdev) {
> > >> +               ret = nvmem_get_mac_address(&pdev->dev, addr);
> > >> +               put_device(&pdev->dev);
> > >> +               return ret;
> > >> +       }
> > >> +
> > >
> > > This smells like the wrong band aid :)
> > >
> > > Any struct device can contain an OF node pointer these days.
> >
> > But not all nodes might have an associated device, see DSA for 
> > example.
>
> I believe what Ben is saying and what I said earlier is going from dev
> -> OF node is right and OF node -> dev is wrong. If you only have an
> OF node, then use an of_* function.
>
> > And as the name suggests of_get_mac_address() operates on a node. So
> > if a driver calls of_get_mac_address() it should work on the node. 
> > What
> > is wrong IMHO, is that the ethernet drivers where the corresponding
> > board
> > has a nvmem_cell_lookup registered is calling 
> > of_get_mac_address(node).
> > It should rather call eth_get_mac_address(dev) in the first place.
> >
> > One would need to figure out if there is an actual device (with an
> > assiciated of_node), then call eth_get_mac_address(dev) and if there
> > isn't a device call of_get_mac_address(node).
>
> Yes, I think we're all in agreement.
>
> > But I don't know if that is easy to figure out. Well, one could start
> > with just the device where nvmem_cell_lookup is used. Then we could
> > drop the workaround above.
>
> Start with the ones just passing dev.of_node directly:
>
> $ git grep 'of_get_mac_address(.*of_node)'

[..]

Before I'll try to come up with a patch for this, I'd like to get
your opinion on it.

(1) replacing of_get_mac_address(node) with eth_get_mac_address(dev)
    might sometimes lead to confusing comments like in
    drivers/net/ethernet/allwinner/sun4i-emac.c:

    /* Read MAC-address from DT */
    ret = of_get_mac_address(np, ndev->dev_addr);

    Do we live with that or should the new name somehow reflect that
    it is taken from the device tree.

(2) What do you think of eth_get_mac_address(ndev). That is, the
    second argument is missing and ndev->dev_addr is used.
    I'm unsure about it. We'd still need a second function for drivers
    which don't write ndev->dev_addr directly, but have some custom
    logic in between. OTOH it would be like eth_hw_addr_random(ndev).

-michael