[PATCH 1/2] ARM: kirkwood: retain MAC address for DT ethernet

[Andrew Lunn]

On Thu, Oct 03, 2013 at 02:44:53PM +0200, Sebastian Hesselbarth wrote:
> Ethernet IP on Kirkwood SoCs loose their MAC address register content
> if clock gated. To allow modular ethernet driver setups and gated clocks
> also on non-DT capable bootloaders, we fixup port device nodes with no
> valid MAC address property. This patch copies MAC address register
> contents set up by bootloaders early, notably before ethernet clocks
> are gated. While at it, also reorder call sequence in _dt_init.
> 
> Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth at gmail.com>
> ---
> - Added Benjamin and Grant who where part of the discussion last time
>   and gave valuable input on DT fixups.
> - Patches are based on v3.12-rc1 and depend on latest mv643xx_eth
>   fixes applied yesterday by David Miller [1].
> 
> [1] https://git.kernel.org/cgit/linux/kernel/git/davem/net.git/commit/?id=569943d0639c85a451ea853087cbd5f738247dd9
> 
> Cc: Jason Cooper <jason at lakedaemon.net>
> Cc: Andrew Lunn <andrew at lunn.ch>
> Cc: Russell King <linux at arm.linux.org.uk>
> Cc: Grant Likely <grant.likely at linaro.org>
> Cc: Benjamin Herrenschmidt <benh at kernel.crashing.org>
> Cc: Jason Gunthorpe <jgunthorpe at obsidianresearch.com>
> Cc: Ezequiel Garcia <ezequiel.garcia at free-electrons.com>
> Cc: linux-arm-kernel at lists.infradead.org
> Cc: linux-kernel at vger.kernel.org
> ---
>  arch/arm/mach-kirkwood/board-dt.c |   72 +++++++++++++++++++++++++++++++++++--
>  1 file changed, 69 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/arm/mach-kirkwood/board-dt.c b/arch/arm/mach-kirkwood/board-dt.c
> index 82d3ad8..a2974ad 100644
> --- a/arch/arm/mach-kirkwood/board-dt.c
> +++ b/arch/arm/mach-kirkwood/board-dt.c
> @@ -13,6 +13,8 @@
>  #include <linux/kernel.h>
>  #include <linux/init.h>
>  #include <linux/of.h>
> +#include <linux/of_address.h>
> +#include <linux/of_net.h>
>  #include <linux/of_platform.h>
>  #include <linux/clk-provider.h>
>  #include <linux/clocksource.h>
> @@ -66,6 +68,71 @@ static void __init kirkwood_legacy_clk_init(void)
>  	clk_prepare_enable(clk);
>  }
>  
> +#define MV643XX_ETH_MAC_ADDR_LOW	0x0414
> +#define MV643XX_ETH_MAC_ADDR_HIGH	0x0418
> +
> +static void __init kirkwood_dt_eth_fixup(void)
> +{
> +	struct device_node *np;
> +
> +	/*
> +	 * The ethernet interfaces forget the MAC address assigned by u-boot
> +	 * if the clocks are turned off. Usually, u-boot on kirkwood boards
> +	 * has no DT support to properly set local-mac-address property.
> +	 * As a workaround, we get the MAC address from mv643xx_eth registers
> +	 * and update the port device node if no valid MAC address is set.
> +	 */
> +	for_each_compatible_node(np, NULL, "marvell,kirkwood-eth-port") {
> +		struct device_node *pnp = of_get_parent(np);
> +		struct property *pmac;
> +		void __iomem *io;
> +		u8 *macaddr;
> +		u32 reg;
> +
> +		if (!pnp || !of_device_is_available(pnp))
> +			continue;
> +
> +		if (of_get_mac_address(np))
> +			continue;
> +
> +		pr_err(FW_BUG "%s: local-mac-address is not set\n",
> +		       np->full_name);
> +
> +		io = of_iomap(pnp, 0);
> +		if (!io)
> +			continue;
> +
> +		pmac = kzalloc(sizeof(*pmac) + 6, GFP_KERNEL);
> +		if (!pmac) {
> +			iounmap(io);
> +			continue;
> +		}
> +
> +		pmac->value = pmac + 1;
> +		pmac->length = 6;
> +		pmac->name = kstrdup("local-mac-address", GFP_KERNEL);
> +		if (!pmac->name) {
> +			kfree(pmac);
> +			iounmap(io);
> +			continue;
> +		}
> +
> +		macaddr = pmac->value;
> +		reg = readl(io + MV643XX_ETH_MAC_ADDR_HIGH);
> +		macaddr[0] = (reg >> 24) & 0xff;
> +		macaddr[1] = (reg >> 16) & 0xff;
> +		macaddr[2] = (reg >> 8) & 0xff;
> +		macaddr[3] = reg & 0xff;

Hi Sebastian

What happens if at this point, the ethernet clock is gated off?  I
assume the CPU locks solid, requiring a power cycle.

It would be a bit of an odd situation, u-boot has disabled the clock,
yet we have a DT node for the device and no valid local-mac-address.
We do however know that Jason Gunthorpe u-boot does gate the second
ethernet for his board, so it is not completely impossible to hit this
situation, particularly when bringing up a new board.

Are we too early in the boot to actually use the clock information in
the node to find it, and clk_prepare_enable() it before making the
access?

	Andrew