Commit 4f0278dac5 ("net: sun8i-emac: Lower MDIO frequency") leads to
network failure on the OrangePi PC.
=> dhcp
sun8i_emac_eth_start: Timeout
According to the commit message the change of the MDIO frequency is only
required for external PHYs.
Fixes: 4f0278dac5 ("net: sun8i-emac: Lower MDIO frequency")
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Commit eb5a2b6710 ("net: sun8i-emac: Determine pinmux based on SoC,
not EMAC type") switched the pinmux setup over to look at
CONFIG_MACH_SUN* symbols, to find the appropriate mux value.
Unfortunately this patch missed to check for the H5, which is
pin-compatible to the H3, but uses a different Kconfig symbol (because
it has ARMv8 vs. ARMv7 cores).
Replace the pure SUN8I_H3 symbol with the joint SUNXI_H3_H5 one, which is
there to cover the peripherals common to both SoCs.
Also explicitly list each supported SoC, and have an error message in the
fallback case, to avoid those problems in the future.
This fixes Ethernet support on all H5 boards.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Samuel Holland <samuel@sholland.org> # Orange Pi PC2
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Move this out of the common header and include it only where needed. In
a number of cases this requires adding "struct udevice;" to avoid adding
another large header or in other cases replacing / adding missing header
files that had been pulled in, very indirectly. Finally, we have a few
cases where we did not need to include <asm/global_data.h> at all, so
remove that include.
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
The pinmux choice for the RMII/RGMII pins the EMAC is connected to is
not dependent on the EMAC IP, but on the SoC it is integrated in.
Deriving the pinmux from the DT compatible string (as we do at the
moment) will thus cause problems with certain EMAC IP / SoC combinations.
To avoid this exact issue with the H616, let's use our Kconfig MACH
symbols to choose the correct pinmux setup.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Jernej Skrabec <jernej.skrabec@siol.net>
At the moment we only consider the EPHY register for those SoCs were
we actually have an internal PHY to configure. However even other SoCs
have this register, an expect the EPHY select bit to be cleared for
proper operation with an external PHY.
Rework sun8i_emac_set_syscon_ephy() to be called regardless of the EMAC
model, and clear the H3_EPHY_SELECT bit if no internal PHY is used.
We get away without it so far because SoCs like the A64 clear this bit
on reset, but we need to explicitly clear it on the H616, for instance.
The Linux driver does so as well.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Jernej Skrabec <jernej.skrabec@siol.net>
So far all GBit users of the sun8i-emac driver were using the "rgmii"
PHY mode, even though this turns out to be wrong. It just worked because
the PHY driver doesn't do the proper setup (yet).
In fact for most boards the "rgmii-id" or "rgmii-txid" PHY modes are the
correct ones.
To allow the DTs to describe the phy-mode correctly, and to stay
compatible with Linux, at least allow those other RGMII modes in the
driver.
This avoids breakage if mainline DTs will be synced with U-Boot.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
This name is far too long. Rename it to remove the 'data' bits. This makes
it consistent with the platdata->plat rename.
Signed-off-by: Simon Glass <sjg@chromium.org>
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>
When sending a command via the MDIO bus, the Designware MAC expects some
bits in the CMD register to describe the clock divider value between
the main clock and the MDIO clock.
So far we were omitting these bits, resulting in setting "00", which
means "/ 16", so ending up with an MDIO frequency of either 18.75 or
12.5 MHz.
All the internal PHYs in the H3/H5/H6 SoCs as well as the Gbit Realtek
PHYs seem to be fine with that - although it looks like to be severly
overclocked (the MDIO spec limits the frequency to 2.5 MHz).
However the external 100Mbit PHY on the Pine64 (non-plus) board is
not happy with that, Ethernet was actually never working there, as the
PHY didn't probe.
As we set the EMAC clock (via AHB2) to 300 MHz in ATF (on the 64-bit
SoCs), and use 200 MHz on the H3, we need the highest divider of 128
to let the MDIO clock end up below the required 2.5 MHz.
This enables Ethernet on the Pine64(non-plus).
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
The current implementation of sun8i_get_ephy_nodes() makes quite some
assumptions, in general relying on DT path names is a bad idea.
I think the idea of the code was to determine if we are using the
internal PHY, for which there are simpler and more robust methods:
Rewrite (and rename) the existing function to simply lookup the DT node
that "phy-handle" points to, using the device's DT node.
Then check whether the parent of that PHY node is using an "H3 internal
MDIO" compatible string. If we ever get another internal MDIO bus
implementation, we will probably need code adjustments anyway, so this
is good enough for now.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
[jagan: rebase on master]
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
The error handling in recv() is somewhat broken, for instance
good_packet isn't really used, and it's hardly readable. Also we try
to check for short or too big packets, but those are actually filtered
out by the hardware.
Simplify the whole routine and improve the error handling:
- Bail out early if the current RX descriptor is not ready.
- Enable propagation of runt, huge and broken packets.
- Check for runt and huge packets, and return 0 to indicate this.
This will force the framework to call free_pkt for cleanup.
- Avoid aligning the packet buffer for invalidation again.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
The EMAC soft reset routine was subtly broken, using an open coded
timeout routine without any actual delay.
Remove the unneeded initial reset bit read, and call wait_for_bit_le32()
to handle the timeout correctly.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
When iterating over all RX/TX buffers, we were using a rather long "idx"
control variable, which lead to a nasty overlong line.
Replace "idx" with "i" to avoid this.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
To meet the current alignment requirements for our cache maintenance
functions, we were explicitly aligning the *arguments* to those calls.
This is not only ugly to read, but also wrong, as we need to make sure
we are not accidentally stepping on other data.
Provide wrapper functions for the common case of cleaning or
invalidating a descriptor, to make the cache maintenance calls more
readable. This fixes a good deal of the problematic calls.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
There is no reason to invalidate a TX descriptor before we are setting
it up, as we will only write to a field.
Remove the not needed invalidate_dcache_range() call.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
When we initialise the TX descriptors, there is no need yet to clean
them all to memory, as they don't contain any data yet. Later we will
touch and clean each descriptor anyway.
However we tell the MAC about the beginning of the chain, so we have to
clean at least the first descriptor, to make it clear that this is empty
and there are no packets to transfer yet.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
Before we initialise the RX descriptors, there is no need to *clean*
them from the cache, as we touch them for the first time.
However we should cover the case that those buffers contain dirty cache
lines, which could be evicted and written back to DRAM any time later,
in the worst case *after* the MAC has transferred a packet into them.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
The EMAC driver contains a lot of magic bits, although the manuals
and the Linux driver have all names for them.
Define those names and use them when programming the registers.
Also this replaces a lot of readl/mask/writel operations with the much
easier-to-read setbits_le32() macro.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
Apparently due to copying from some older or converted driver, the
sun8i_emac driver contains pointless wrapper functions to bridge
between a legacy driver and the driver model.
Since sun8i_emac is (and always was) driver model only, there is no
reason to have those confusing wrappers. Just remove them, and use
the driver model prototypes directly.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
When preparing the register value for the MDIO command register, we
start with a zeroed register, so there is no need to mask off certain
bits before setting them.
Simplify the sequence, and rename the variable to a more matching
mii_cmd on the way.
Also the open-coded time-out routine can be replaced with a much safer
and easier-to-read call to wait_for_bit_le32().
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
When initialising the TX DMA descriptors, we mostly chain them up,
but of course don't know about any data or its length yet.
That means they are still invalid, and the OWN bit should NOT be set
yet.
In fact when we later tell the MAC about the beginning of the chain,
and enable TX DMA in the start() routine, the MAC will start fetching
TX descriptors prematurely, as it can be seen by dumping the TX_DMA_STA
and TX_DMA_CUR_DESC registers.
Clear the owner bit, to not give the MAC the wrong illusion that it
owns the descriptors already.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
When phy_startup() returns with an error, because there is no link or
the user interrupted the process, we shall stop the _start() routine
and return with an error, instead of proceeding anyway.
This fixes pointless operations when there is no Ethernet cable
connected, for instance.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Tested-by: Amit Singh Tomar <amittomer25@gmail.com> # Pine64+
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
Pass a udevice into a few functions so `dev` is defined.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Tested-by: Patrick Delaunay <patrick.delaunay@st.com>
When you enable CONFIG_OF_LIVE, you will end up with a lot of
conversions.
To generate this commit, I used coccinelle excluding drivers/core/,
include/dm/, and test/
The semantic patch that makes this change is as follows:
<smpl>
@@
expression dev;
@@
-devfdt_get_addr(dev)
+dev_read_addr(dev)
</smpl>
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
When you enable CONFIG_OF_LIVE, you will end up with a lot of
conversions.
To generate this commit, I used coccinelle excluding drivers/core/,
include/dm/, and test/
The semantic patch that makes this change is as follows:
<smpl>
@@
expression dev;
@@
-devfdt_get_addr(dev)
+dev_read_addr(dev)
</smpl>
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
The H6 EMAC is very similar to the H3 variant, except that it uses the
same pinmux as R40. Add support for it.
Signed-off-by: Samuel Holland <samuel@sholland.org>
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
While the R40 uses a different register for EMAC clock configuration
than other chips, the register has a very similar layout. Reuse the
existing bitfield definitions in this file, since they match.
This allows the driver to compile on the H6 platform, where the
CCM_GMAC_CTRL definitions are not present.
Signed-off-by: Samuel Holland <samuel@sholland.org>
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
Move this header out of the common header. Network support is used in
quite a few places but it still does not warrant blanket inclusion.
Note that this net.h header itself has quite a lot in it. It could be
split into the driver-mode support, functions, structures, checksumming,
etc.
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>
At present if CONFIG_SPL_GPIO_SUPPORT is enabled then the GPIO uclass
is included in SPL/TPL without any control for boards. Some boards may
want to disable this to reduce code size where GPIOs are not needed in
SPL or TPL.
Add a new Kconfig option to permit this. Default it to 'y' so that
existing boards work correctly.
Change existing uses of CONFIG_DM_GPIO to CONFIG_IS_ENABLED(DM_GPIO) to
preserve the current behaviour. Also update the 74x164 GPIO driver since
it cannot build with SPL.
This allows us to remove the hacks in config_uncmd_spl.h and
Makefile.uncmd_spl (eventually those files should be removed).
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
These functions are CPU-related and do not use driver model. Move them to
cpu_func.h
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
H3/H5 can either use the internal phy or an external one.
Before getting clock and resets for the internal phy,
test that we are using it because otherwise it break emac
when using an external phy.
Tested-on: OrangePi PC2 (H5)
Fixes: 2348453c41 (net: sun8i_emac: Add EPHY CLK and RESET support)
Signed-off-by: Emmanuel Vadot <manu@freebsd.org>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Add EPHY CLK and RESET support for sun8i_emac driver to
enable EPHY TX clock and EPHY reset pins via CLK and RESET
framework.
Cc: Joe Hershberger <joe.hershberger@ni.com>
Cc: Lothar Felten <lothar.felten@gmail.com>
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Add CLK and RESET support for sun8i_emac driver to
enable TX clock and reset pins via CLK and RESET
framework.
Cc: Joe Hershberger <joe.hershberger@ni.com>
Cc: Lothar Felten <lothar.felten@gmail.com>
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Unlike other Allwinner SoC's R40 GMAC clock control register
is locate in CCU, but rest located via syscon itself. Since
the phandle property for current code look for 'syscon' and
it will grab the respective ccu or syscon base address based
on DT property defined in respective SoC dtsi.
So, use the existing 'syscon' code even for R40 for retrieving
GMAC clock via CCU and update the register directly in
sun8i_emac_set_syscon instead of writing it separately using
ccm base.
Cc: Joe Hershberger <joe.hershberger@ni.com>
Cc: Lothar Felten <lothar.felten@gmail.com>
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Some boards have the EMAC TX/RX lanes wired with a different length with
the clock lane, which can be workarounded by setting a TX/RX delay in
the EMAC.
This kind of delays are already defined in the newest device tree
binding of dwmac-sun8i, which has already entered linux-next.
Add support for setting these delays.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Jagan Teki <jagan@openedev.com>
Add support for the GMAC found in the Allwinner R40/V40 SoC.
The R40 GMAC interface is not controlled by the syscon register but
has a separate configuration register in the CCU.
The clock gate and reset bits are in a different register compared
to the other SoCs supported by this driver.
The driver uses the -gmac suffix for the R40 because the R40 also
has a different 100 MBit MAC (EMAC).
Signed-off-by: Lothar Felten <lothar.felten@gmail.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
Tested-by: Jagan Teki <jagan@openedev.com>
Use driver data->variant information to select device specific
pin mux and phy clock settings.
Suggested by Jagan Teki
Reviewed-by: Jagan Teki <jagan@openedev.com>
Tested-by: Jagan Teki <jagan@openedev.com>
Signed-off-by: Lothar Felten <lothar.felten@gmail.com>
If the variant is not set and therefore NULL, do not attempt to print
the variant.
Signed-off-by: Lothar Felten <lothar.felten@gmail.com>
Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
Tested-by: Jagan Teki <jagan@openedev.com>
When U-Boot started using SPDX tags we were among the early adopters and
there weren't a lot of other examples to borrow from. So we picked the
area of the file that usually had a full license text and replaced it
with an appropriate SPDX-License-Identifier: entry. Since then, the
Linux Kernel has adopted SPDX tags and they place it as the very first
line in a file (except where shebangs are used, then it's second line)
and with slightly different comment styles than us.
In part due to community overlap, in part due to better tag visibility
and in part for other minor reasons, switch over to that style.
This commit changes all instances where we have a single declared
license in the tag as both the before and after are identical in tag
contents. There's also a few places where I found we did not have a tag
and have introduced one.
Signed-off-by: Tom Rini <trini@konsulko.com>
The original DT binding used by U-Boot's sun8i-emac driver was not really
agreed upon, and deviated from the "official" binding now used by the
kernel. Since now all U-Boot users have been converted to the new
binding, we can remove support for the old DT nodes from the driver.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
The Ethernet MAC used in newer Allwinner SoCs (H3, A64, H5) got an
upstream Linux driver in v4.15.
This one uses a slightly different binding from the original one used
by the U-Boot driver.
The differences to the old binding are:
- The "syscon" address is held in a separate node, referenced via a
phandle in the "syscon" property.
- The reference to the PHY is held in a property called "phy-handle",
not "phy".
- The PHY register is at offset 0x30 in the syscon device, not at 0.
- The internal PHY is activated when the node, which phy-handle points
to, is a child node of an "allwinner,sun8i-h3-mdio-internal" node.
Teach the U-Boot driver how to find its resources in a "new-style" DT,
so that we can use a Linux kernel compatible DT for U-Boot as well.
This keeps support for the old binding for now, to allow a smooth
transition.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
The Linux kernel driver for the Allwinner pin controller gained support
for generic properties, which are now also used in the DTs.
The sun8i-emac Ethernet driver for new Allwinner MACs reads the pins from
the DT, but so far only supported the old binding.
Update the parsing routine to cope with both the old and new bindings,
so that the newer DTs can be used with U-Boot and its Ethernet driver.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
