net: designware: eqos: mimic Linux gmac4 mdio_{read, write} ...

The code was ported over from U-Boot and has some things different than
the Kernel driver:

- It reads ->mdio_address, but the kernel driver doesn't
- It doesn't write ->mdio_data in a read, but the kernel driver does
- It has delays in addition to spinning on MII_BUSY, but the kernel driver
  doesn't

Adopt the same flow as used by the kernel.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware: eqos: remove left-over PHY quirk work around ...

The dummy read was a left over from an abandoned merge with the existing
support for the SoCFPGA variant of the Designware MAC. It doesn't exist
in either the U-Boot or kernel drivers for the EQOS, thus drop it.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware: eqos: stm32: drop no longer needed syscfg-clk ...

The upstream device tree no longer lists "syscfg-clk" and the syscon node
contains a clocks property that the barebox syscon driver now
automatically clocks before access.

Thus drop the requirement that "syscfg-clk" is specified.

Fixes: 4aaceef58 ("dts: update to v5.6-rc1")
Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware: socfpga: fix possible invalid pointer deref ...

In cases where the reset controller specification in the device tree is
malformed, we get an error pointer back from reset_control_get. This
compares unequal to NULL and would cause an access violation when passed
to reset_control_(de)?assert.

Fix this by propagating the error. When the reset controller is missing,
reset_control_(de)?assert will be passed NULL pointers, rendering them
no-ops.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware: eqos: don't funnel all MDIO writes to register 0 ...

eqos_mdio_write ended up using the addr parameter instead of the
computed miiaddr variable, which would've factored in the reg parameter.

This had the effect that all writes went to PHY register 0, which was
fine as long as there were only register 0 writes. As soon there are more
writes, e.g. because a PHY driver was enabled, register 0 became
clobbered and erratic behavior ensued.
Fix the typo and while at it rename the val parameter to a more
descriptive name.

Fixes: a4f709bbb ("net: add Designware Ethernet QoS for STM32MP")
Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fsl-fman: Do not put hardware in reset before Linux start ...

Linux depends on the icid values configured in the DPAA. Do not put
the hardware into reset in order to preserve the register values.
Without this, the IOMMU doesn't work properly.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

Merge branch 'for-next/zynq'

Merge branch 'for-next/randfixes'

Merge branch 'for-next/misc'

Merge branch 'for-next/layerscape'

Merge branch 'for-next/fsl-fman'

net: designware: eqos: remove done TODO ...

The todo split comment is a left-over from refactoring. Remove it.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware-eqos: depend on COMMON_CLK ...

The clk_bulk_* functions are only available with common clock.
Also OFTREE should not be selected, but is a dependency.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fsl-fman: Disable unused network interfaces ...

The Linux dts files have all network interfaces enabled. We have
to disable the ones that are unused.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware: eqos: fix NULL pointer use in dev_printf ...

Inside mdio_register, a read of the PHY's id register is attempted.
If it fails, we print an error message with eqos_err, which uses the
ethernet device's unique name, but at this time there has been none set,
because eth_register was not yet called. Fix this by using the MDIO bus
device instead.

Signed-off-by: Ahmad Fatoum <ahmad@a3f.at>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware: eqos: enable clocks before mdio_register ...

We can't be using the MAC including the MDIO controller while the clocks
are off, but this is exactly the case when mdio_register is called and
the interface is not yet up. To allow reading the PHY id to succeed
before the interface is up, turn on the clocks as part of the
initialization in the probe.

This fixes following error at probe time:

    ERROR: <NULL>: MDIO not idle at entry

The NULL is fixed in a follow-up commit.

Signed-off-by: Ahmad Fatoum <ahmad@a3f.at>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware: eqos: properly stop DMA on halt ...

Specifying ->halt only means that it's called along with eth_unregister.
If we want to halt the DMA, we will have to call it ourselves in the
remove callback. Do this.

Signed-off-by: Ahmad Fatoum <ahmad@a3f.at>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fsl-fman: do not leave not transmitted DMA buffers mapped ...

When a packet can't be transmitted we should unmap it. This probably
won't change much since when we can't transmit a packet the fman
probably can't recover from it anyway, but still it is cleaner to not
leave the buffers mapped.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fsl-fman: simplify setting next offset ...

The offset of the current packet can be retrieved from the current txbd
index, so do this instead of increasing the offset with each new packet.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fsl-fman: Store index for rxbd ...

Like done for the txbd, also store the index of the current rxbd in the
driver private data. This makes the code easier to follow.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fsl-fman: move status read into loop ...

Instead of reading the status once before the loop and then again right
before the next iteration, just read the status in the loop.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fsl-fman: Store index for txbd ...

Instead of storing a pointer to the current txbd store the index of the
current buffer. This makes the code more straight forward.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fsl-fman: reset device before leaving ...

edev->halt is implemented, but not called by the core during shutdown.
We have to call it manually during device remove. With this and also
resetting the whole FMan during shutdown the driver now works properly
when started second stage from another barebox.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware: eqos: stop DMA on halt ...

designware_eqos.c contains an eqos_stop implementation to stop the NIC
when halting the interface. Unfortunately it wasn't used leading to
memory corruption on boot, possibly due to DMA. Fix this.

Signed-off-by: Ahmad Fatoum <ahmad@a3f.at>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: macb: add Zynq compatible ...

Add compatible for the GEM on the Zynq 7000 device.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: macb: handle more clocks ...

Both pclk and hclk are required clocks in the DT binding.
rx_clk and tx_clk are optional, but must be enabled if a system
has separate gates for them.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

Merge branch 'for-next/stm32'

net: eqos: extend Designware Ethernet QoS for Tegra 186 support ...

The U-Boot dwc_eth_qos driver ported in the previous commit had support
for both the Tegra 186/194 and STM32MP variants of the EQOS IP.

The barebox Tegra supported doesn't include the 186, but as the driver
was nevertheless ported along with the rest, lets include it in the
source tree.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: add Designware Ethernet QoS for STM32MP ...

We already have Designware NIC support in barebox, but for the DWMAC1000,
the DWMAC4 (also called GMAC4), no support was mainline so far.

The DWMAC4 is different enough that sharing code with the DWMAC1000 is
not really that helpful, because even basics like MDIO registers have
different layout. Instead of coding bit masks and shifts into the driver
data, like Linux does, we'll keep both driver kinds separate.

Nevertheless, we collect functions that are not SoC-specific into a
separate 'library' file.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware: socfpga: fix phy setup for Arria10 ...

Barebox-version of the Linux v5.2 patch:

    40ae255
    net: stmmac: socfpga: fix phy and ptp_ref setup for Arria10/Stratix10

    On the Arria10, Agilex, and Stratix10 SoC, there are a few differences from
    the Cyclone5 and Arria5:
     - The emac PHY setup bits are in separate registers.
     - The PTP reference clock select mask is different.
     - The register to enable the emac signal from FPGA is different.

    Thus, this patch creates a separate function for setting the phy modes on
    Arria10/Agilex/Stratix10. The separation is based a new DTS binding:
    "altr,socfpga-stmmac-a10-s10".

    Signed-off-by: Dinh Nguyen <dinguyen@kernel.org>
    Signed-off-by: David S. Miller <davem@davemloft.net>

The new DTS binding is already part of v2019.10.0 and the driver doesn't
probe on Arria10 without the new binding introduced in this patch.

Signed-off-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>