It turns out that in custom designs if the system is reset
multiple times in conjunction with a slight increase in external
temperature, the felix switch starts to behave in a strange way:
packets are no longer received on the ENECT interface connected
to the L2switch internal port (the TX side of internal port stops working
or the packets do not reach there. It is not very clear where
the packets remain blocked. None of the counters points to a disruption
in the L2switch)
The issue is not reproducible on NXP reference designs.
It was observed that by adding the switch core reset, the problem
goes aways, even if intensive testing in temperature chambers
is applied.
The current patch performs soft reset on the switch core to ensure proper
operation of the L2switch.
Signed-off-by: Radu Bulie <radu-andrei.bulie@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Instead of trying to guess which operating modes need in-band
negotiation to be active and which ones don't, parse the available
information from the device tree. That will be correct in the cases we
can already guess, and more.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
These functions can return errors, it's best to catch them and trigger
the driver unwind code path.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
strncpy() simply bails out when copying a source string whose size
exceeds the destination string size, potentially leaving the destination
string unterminated.
One possible way to address is to pass MDIO_NAME_LEN - 1 and a
previously zero-initialized destination string, but this is more
difficult to maintain.
The chosen alternative is to use strlcpy(), which properly limits the
copy len in the (srclen >= size) case to "size - 1", and which is also
more efficient than the strncpy() byte-by-byte implementation by using
memcpy. The destination string returned by strlcpy() is always NULL
terminated.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
The felix driver runs only on NXP LS1028A, which most definitely does
not support the parallel 10G interface, just USXGMII, and that only up
to 2.5Gbps (toned down from 10 Gbps via symbol replication).
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
After the discussion here:
https://lore.kernel.org/netdev/20210603143453.if7hgifupx5k433b@pali/
which resulted in this patch:
https://patchwork.kernel.org/project/netdevbpf/patch/20210704134325.24842-1-pali@kernel.org/
and many other discussions before it, notably:
https://patchwork.kernel.org/project/linux-arm-kernel/patch/1512016235-15909-1-git-send-email-Bhaskar.Upadhaya@nxp.com/
it became apparent that nobody really knows what "SGMII 2500" is.
Certainly, Freescale/NXP hardware engineers name this protocol
"SGMII 2500" in the reference manuals, but the PCS devices do not
support any "SGMII" specific features when operating at the speed of
2500 Mbps, no in-band autoneg and no speed change via symbol replication
. So that leaves a fixed speed of 2500 Mbps using a coding of 8b/10b
with a SERDES lane frequency of 3.125 GHz. In fact, "SGMII 2500 without
in-band autoneg and at a fixed speed" is indistinguishable from
"2500base-x without in-band autoneg", which is precisely what these NXP
devices support.
So it just appears that "SGMII 2500" is an unclear name with no clear
definition that stuck.
As such, in the Linux kernel, the drivers which use this SERDES protocol
use the 2500base-x phy-mode.
This patch converts U-Boot to use 2500base-x too, or at least, as much
as it can.
Note that I would have really liked to delete PHY_INTERFACE_MODE_SGMII_2500
completely, but the mvpp2 driver seems to even distinguish between SGMII
2500 and 2500base-X. Namely, it enables in-band autoneg for one but not
the other, and forces flow control for one but not the other. This goes
back to the idea that maybe 2500base-X is a fiber protocol and SGMII-2500
is an MII protocol (connects a MAC to a PHY such as Aquantia), but the
two are practically indistinguishable through everything except use case.
NXP devices can support both use cases through an identical configuration,
for example RX flow control can be unconditionally enabled in order to
support rate adaptation performed by an Aquantia PHY. At least I can
find no indication in online documents published by Cisco which would
point towards "SGMII-2500" being an actual standard with an actual
definition, so I cannot say "yes, NXP devices support it".
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
NXP Legal insists that the following are not fine:
- Saying "NXP Semiconductors" instead of "NXP", since the company's
registered name is "NXP"
- Putting a "(c)" sign in the copyright string
- Putting a comma in the copyright string
The only accepted copyright string format is "Copyright <year-range> NXP".
This patch changes the copyright headers in the networking files that
were sent by me, or derived from code sent by me.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Make sure that the link status returned by phy_startup() is propagated
to the .start() method of struct eth_ops.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Tested-by: Michael Walle <michael@walle.cc>
It is an unfortunate reality that some PHY settings done by U-Boot
persist even after the PHY is reset and taken over by Linux, and even
more unfortunate that Linux has come to depend on things being set in a
certain way.
For example, on the NXP LS1028A-RDB, the felix switch ports are
connected to a VSC8514 QSGMII PHY. Between the switch port PCS and the
PHY, the U-Boot drivers enable in-band auto-negotiation which makes the
copper-side negotiated speed and duplex be transmitted from the PHY to
the MAC automatically.
The PHY driver portion that does this is in vsc8514_config():
/* Enable Serdes Auto-negotiation */
phy_write(phydev, MDIO_DEVAD_NONE, PHY_EXT_PAGE_ACCESS,
PHY_EXT_PAGE_ACCESS_EXTENDED3);
val = phy_read(phydev, MDIO_DEVAD_NONE, MIIM_VSC8514_MAC_SERDES_CON);
val = val | MIIM_VSC8574_MAC_SERDES_ANEG;
phy_write(phydev, MDIO_DEVAD_NONE, MIIM_VSC8514_MAC_SERDES_CON, val);
The point is that in-band autoneg should be turned on in both the PHY
and the MAC, or off in both the PHY and the MAC, otherwise the QSGMII
link will be broken.
And because phy_config() is currently called at .port_enable() time, the
result is that ports on which traffic has been sent in U-Boot will have
in-band autoneg enabled, and the rest won't.
It can be argued that the Linux kernel should not assume one way or
another and just reinitialize everything according to what it expects,
and that is completely fair. In fact, I've already started an attempt to
remove this dependency, although admittedly I am making slow progress at
it:
https://patchwork.kernel.org/project/netdevbpf/cover/20210212172341.3489046-1-olteanv@gmail.com/
Nonetheless, the sad reality is that NXP also has, apart from kernel
drivers, some user space networking (DPDK), and for some reason, the
expectation there is that somebody else initializes the PHYs. The kernel
can't do it because the device ownership doesn't belong to the kernel,
so what remains is for the bootloader to do it (especially since other
drivers generally call phy_config() at probe time). This is a really
weak guarantee that might break at any time, but apparently that is
enough for some.
Since initializing the ports and PHYs at probe time does not break
anything, we can just do that.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Tested-by: Michael Walle <michael@walle.cc>
No one is calling this function from outside felix_switch.c.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Tested-by: Michael Walle <michael@walle.cc>
Removed call of ofnode_translate_address() after ofnode_read_resource
in luton_switch.c:luton_probe(); it is unnecessary since
the commit feb7ac457c ("dm: core: Add address translation in
fdt_get_resource").
Fixes: feb7ac457c ("dm: core: Add address translation in fdt_get_resource")
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Reported-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: Patrick Delaunay <patrick.delaunay@foss.st.com>
Today of_address_to_resource() is called only in
ofnode_read_resource() for livetree support and
fdt_get_resource() is called when livetree is not supported.
The fdt_get_resource() doesn't do the address translation
so when it is required, but the address translation is done
by ofnode_read_resource() caller, for example in
drivers/firmware/scmi/smt.c::scmi_dt_get_smt_buffer() {
...
ret = ofnode_read_resource(args.node, 0, &resource);
if (ret)
return ret;
faddr = cpu_to_fdt32(resource.start);
paddr = ofnode_translate_address(args.node, &faddr);
...
The both behavior should be aligned and the address translation
must be called in fdt_get_resource() and removed for each caller.
Fixes: a44810123f ("dm: core: Add dev_read_resource() to read device resources")
Signed-off-by: Patrick Delaunay <patrick.delaunay@foss.st.com>
Acked-by: Etienne Carriere <etienne.carriere@linaro.org>
Sometimes no traffic was getting out on the ports, the root cause was
a wrong configuration of the Serdes6G, which is used on jr2 pcb111.
This patch fixes this issue by applying the correct configuration.
Fixes: 5e1d417bec ("net: Add MSCC Jaguar2 network driver.")
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Make sure to reset the switch core at probe time.
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
This driver is used for the Ethernet switch integrated into LS1028A NXP.
Felix on LS1028A has 4 front panel ports and two internal ports, I/O
to/from the switch is done through an ENETC Ethernet interface.
The 4 front panel ports are available as Ethernet interfaces and can be
used with the typical network commands like tftp.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
We use 'priv' for private data but often use 'platdata' for platform data.
We can't really use 'pdata' since that is ambiguous (it could mean private
or platform data).
Rename some of the latter variables to end with 'plat' for consistency.
Signed-off-by: Simon Glass <sjg@chromium.org>
This construct is quite long-winded. In earlier days it made some sense
since auto-allocation was a strange concept. But with driver model now
used pretty universally, we can shorten this to 'auto'. This reduces
verbosity and makes it easier to read.
Coincidentally it also ensures that every declaration is on one line,
thus making dtoc's job easier.
Signed-off-by: Simon Glass <sjg@chromium.org>
At present dm/device.h includes the linux-compatible features. This
requires including linux/compat.h which in turn includes a lot of headers.
One of these is malloc.h which we thus end up including in every file in
U-Boot. Apart from the inefficiency of this, it is problematic for sandbox
which needs to use the system malloc() in some files.
Move the compatibility features into a separate header file.
Signed-off-by: Simon Glass <sjg@chromium.org>
Because all MSCC SoC use the same MDIO bus, put the implementation in
one common file(mscc_miim) and make all the other MSCC network drivers to
use these functions.
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
When serdes configuration was written in hardware there was a delay
of 100ms to be sure that configuration was written. But the delay is not
needed because already the function serdes_write it is checking that the
operation finished.
Therefore remove the mdelay. This improves the speed of configuring the
network driver.
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Update Luton network driver to have support also for pcb90. The pcb90
has 24 ports from which 12 ports are connected to SerDes6G.
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Add network driver for Microsemi Ethernet switch.
It is present on Serval SoCs.
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Add network driver for Microsemi Ethernet switch.
It is present on ServalT SoCs.
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Add network driver for Microsemi Ethernet switch.
It is present on Jaguar2 SoCs.
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
The function mscc_miim_reset resets all the phys, but it is called for
each phy separetely. One consequence of this is that the boot time
is increased by 2 seconds.
The fix consists for calling the mscc_miim_reset function only once for
all phys.
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
This functions can be reused by other MSCC SoCs therefore,
make them more generic and move them in separate files.
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Move file ocelot_switch to mscc_eswitch to prepare to add
new net drivers for other MSCC SoCs.
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
