net: fec_imx: Do not use DMA coherent memory for Rx buffers ...

Trying to do unaligned access of coherent memory on AArch64 will lead
to an abort and some parts of our IP stack will do exactly that with
received packet buffer by using memcpy() to extracts parts of it.

Convert the driver to use regular memory for received data buffers to
prevent the issue from happening.

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fec_imx: Make use of readx_poll_timeout() macros ...

Condense a bit of code by using vairous readx_poll_timeout() macros.

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fec_imx: Drop frame_length ...

Drop frame_length in favour of just using len

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fec_imx: Drop struct fec_frame ...

Drop struct fec_frame since it doesn't have any real users in the
code. It is only referenced in fec_recv() and is used by that function
as a generic pointer.

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fec_imx: Read data_length only once ...

Data length of a given Rx data descriptor isn't going to change until
that descriptor is processed and given back to HW. Re-work the code to
save and re-use that value instead.

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fec_imx: Don't check bd_status & FEC_RBD_ERR twice ...

Re-shuffle comparisons in order to avoid having to check for
FEC_RBD_ERR in bd_status more than once.

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fec_imx: Drop extra indentation level by exiting early ...

Drop extra indentation level by exiting early which also allows us to
check for bd_status & FEC_RBD_EMPTY instead of !(bd_status &
FEC_RBD_EMPTY).

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: mdio: Do not warn about invalid PHY address ...

of_mdiobus_register is now called with nodes from ethernet devices.
Their child nodes are not necessarily phy nodes. For am335x devices
we get several warnings now:

miibus0: /ocp/ethernet@4a100000/mdio@4a101000 PHY address 1242566656 is too large
miibus0: /ocp/ethernet@4a100000/slave@4a100200 has invalid PHY address
miibus0: /ocp/ethernet@4a100000/slave@4a100300 has invalid PHY address
miibus0: /ocp/ethernet@4a100000/cpsw-phy-sel@44e10650 PHY address 1155597904 is too large

None of the nodes is a phy node, so silence the warnings and turn them
into dev_dbg.

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

Merge branch 'for-next/net'

Merge branch 'for-next/dma-mask'

net: fec_imx: Return EFAULT when DMA mapping fails ...

EFAULT seems to be much more appropriate error code for DMA mapping
failure, it also produces more informative message, compare:

ping 192.168.53.7
WARNING: eth0: No MAC address set. Using random address 92:4e:f1:18:96:67
eth0: 100Mbps full duplex link detected
ping failed: Bad address

versus old message:

ping 192.168.53.7
WARNING: eth0: No MAC address set. Using random address c6:b7:67:bb:4a:c3
eth0: 100Mbps full duplex link detected
ping failed: I/O error

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fec_imx: Specify that DMA is 32-bit only ...

FEC can only do a DMA transfer to 32-bit address space, so mark it as
such.

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

drivers: net: designware: select reset controller ...

The socfpga-specific designware driver uses reset_control_get().
Select the socfpga-reset driver to actually use this function instead of
a stub.

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

phylib: add support for reset-gpios ...

Add minimal support for reset-gpios in the PHY node.

Example DT that uses this:

	macb0: ethernet@fffbc000 {
		phy-mode = "rmii";
		#address-cells = <1>;
		#size-cells = <0>;

		ethphy0: ethernet-phy@1 {
			reg = <3>;
			reset-gpios = <&pioE 17 GPIO_ACTIVE_LOW>;
			reset-assert-us = <1000>;
			reset-deassert-us = <2000>;
		};
	};

The reset is required to get the Davicom PHY operational on
my proprietary board, and is assumed relevant for other boards too.

The PHY is reset when we read the info from DT,
before the phy_id is retreived.

The bindings are documented in dts/Bindings/net/phy.txt.

A side-effect of this patch is that we may see
phy devices created from the DT due to the extra call to:

    of_mdiobus_register()

with the ethernet node as argument.

The logic to determine if a child node is a PHY node
is a simpler version compared to the one used in the kernel.
The kernel have a whitelist of compatible strings
that is not included in the barebox version.
They can be added later if needed.

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

phylib: add Davicom PHY support ...

Based on driver from Linux kernel 4.18.0-rc4

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net/e1000: allow to overwrite flash size from device tree ...

When barebox probes the e1000 driver and the flash on the i210 device is
unprogrammed, the driver assumes the flash has a size of only 4 kiB.
This is annoying because to program the flash an image must be written
that is bigger than 4 kiB. So you first have to flash the first sector
to make barebox detect the right size on the next boot. Then reset the
board to be able to write the remaining data.

To work around that limitation, try to read the actual size from the
device tree. (Note however that barebox' pci code currently doesn't use
the device tree and so currently this try always fails without further
patching.)

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: designware: make sure DMA is disabled on kernel start. ...

Halt is not automatically executed if we start the kernel.
So, we may have potentially memory corruptions.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fec_imx: Make use of IS_ALIGNED ...

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: fec_imx: Use dma mapping functions ...

Rather than doing DMA on the input buffer address get a proper DMA
address from the mapping functions.

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

net: fec_imx: remove unnecessary DMA sync ops ...

The fec receive buffers are coherently mapped, no need to dma_sync on
them.

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

Merge branch 'for-next/socfpga'

Merge branch 'for-next/net'

usbnet: increase bulk message timeout ...

Timeout of 1ms is too low for case when device operates in Full Speed
mode. This causes most incoming frames to be lost.

Since timeout is only used to get out of polling loop in case of no
response from hardware, increasing it should be safe for all platforms.

Signed-off-by: Nikita Yushchenko <nikita.yoush@cogentembedded.com>
Tested-By: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: make SoCFPGA-specific designware driver work again ...

If MFD_SYSCON is disabled in .config then socfpga_designware_eth probe
fails with this message:

  socfpga_designware_eth ff702000.ethernet: Could not get sysmgr-syscon node

Thanks to Steffen for hint!

Cc: Steffen Trumtrar <s.trumtrar@pengutronix.de>
Signed-off-by: Antony Pavlov <antonynpavlov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: phy: micrel: fix too late link up ...

Since commit a4cee7207a1 ("net: phy: micrel: Add workaround for bad
autoneg") by Philipp Zabel <p.zabel@pengutronix.de>:

    Based on kernel commit d2fd719bcb0e ("net/phy: micrel: Add workaround
    for bad autoneg") by Nathan Sullivan <nathan.sullivan@ni.com>:

        Very rarely, the KSZ9031 will appear to complete autonegotiation, but
        will drop all traffic afterwards.  When this happens, the idle error
        count will read 0xFF after autonegotiation completes. Reset the PHY
        when in that state.

is it possible that we get the link to late and  barebox reports "Network is
down". The problem is that ksz9031_config_init is called again in an
error case. In this case autoneg will restart but waiting for finish is
missing.

Signed-off-by: Christian Hemp <c.hemp@phytec.de>
Tested-by: Stefan Müller-Klieser <s.mueller-klieser@phytec.de>

rename file_operations -> cdev_operations ...

Linux also has struct file_operations which are something different.
Rename our file_operations to cdev_operations which better matches
what we have.

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

net: cpsw: fix probe with fixed-link ...

While cpsw is probe dt, it accepts only slaves nodes with "phy_id" property.
In case of fixed-link there are no "phy_id" property and probe would be failed.
This patch avoid the failure due to missing "phy_id" in case of fixed-link.

Signed-off-by: Andreas Schmidt <mail@schmidt-andreas.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>

net: cpsw: Call phy_device_connect() earlier ...

We only want to register a slave when a valid phy is available.
Instead of manually calling mdiobus_scan() and phy_register_device()
we can let this do from phy_device_connect() which also works for
fixed phys.

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

Merge branch 'for-next/misc'

Merge branch 'for-next/e1000'