out_be32 and in_be32 are actually #defined to little endian
writel/readl in arch/microblaze.
Just use __raw_writel/readl instead. That is also what is used
in the Linux kernel driver for this IP block
Tested on MIPSfpga. Can tftp a kernel.
Signed-off-by: Zubair Lutfullah Kakakhel <Zubair.Kakakhel@imgtec.com>
Reviewed-by: Paul Burton <paul.burton@imgtec.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Virtual to physical mapping isn't necessarily 1:1 for all architectures
Using ioremap_nocache allows for the arch code to translate the
physical address to a virtual address.
Signed-off-by: Zubair Lutfullah Kakakhel <Zubair.Kakakhel@imgtec.com>
Reviewed-by: Paul Burton <paul.burton@imgtec.com>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
On LD4 SoC or later, the pin-mux registers are 8bit wide, while 4bit
wide on sLD3 SoC. Support it for the sLD3 pinctrl driver.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
With sunxi-musb musb_lowlevel_init() can fail when a charger; or no cable
is plugged into the otg port.
To avoid leaking the struct musb allocated by musb_init_controller()
on repeated musb_usb_probe() calls, we were caching its result.
But musb_init_controller() does more, such as calling sunxi_musb_init()
which enables the clocks.
Not calling sunxi_musb_init() causes the musb controller to stop working
after a "usb reset" since that calls musb_usb_remove() which disables the
clocks.
This commit fixes this by removing the caching of the struct returned
from musb_init_controller(), it replaces this by free-ing the allocated
memory in musb_usb_remove() and calling musb_usb_remove() on
musb_usb_probe() errors to ensure proper cleanup.
While at it also make musb_usb_probe() and musb_usb_remove() static.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
The Linux kernel musb driver expects VBUS to be off while initializing
musb. Having it on results in a repeating string of warnings, followed
by an unusable peripheral. The peripheral is only usable after
physically removing the OTG adapter, letting musb reset its state.
This partially reverts commit c9f8947e66 ("sunxi: usb-phy: Never
power off the usb ports")
Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
When cold-booting the ldoio0/1 regulators are always off / the
gpios are always at tristate. But when re-booting from android these
are sometimes on. Disable them at axp_init time (iow as early as possible)
to remove this difference between a cold boot and a reboot.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Ian Campbell <ijc@hellion.org.uk>
At present TPL uses the same options as SPL support. In a few cases the board
config enables or disables the SPL options depending on whether
CONFIG_TPL_BUILD is defined.
With the move to Kconfig, options are determined for the whole build and
(without a hack like an #undef in a header file) cannot be controlled in this
way.
Create new TPL options for these and update users. This will allow Kconfig
conversion to proceed for these boards.
Signed-off-by: Simon Glass <sjg@chromium.org>
These macros are only referenced in pinctrl-uniphier-core.c, so
they need not reside in a header file.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
This is the state-of-the-art MMC driver implementation.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This driver has not been converted to Driver Model, and it is an
obstacle to migrate other block device drivers. Remove it for now.
The UniPhier SoCs already use a DM-based EHCI driver, so now
ARCH_UNIPHIER can select DM_USB.
These two changes must be done atomically because removing the
legacy driver causes a build error.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Marek Vasut <marex@denx.de>
Since the 'clk_client.h' doesn't exist, it should be 'clk.h'.
Signed-off-by: Wenyou Yang <wenyou.yang@atmel.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
The special handling of the chip address and register address must only
happen before we send the data buffer, otherwise we will end up
inserting both of these every 32 bytes.
Signed-off-by: John Keeping <john@metanate.com>
Acked-by: Simon Glass <sjg@chromium.org>
There is no point in writing intermediate values to the txdata
registers.
Also add padding to the debug logging to make it easier to read when
there are leading zeroes.
Signed-off-by: John Keeping <john@metanate.com>
Acked-by: Simon Glass <sjg@chromium.org>
Make it clear that we are using the same value in two adjacent lines.
Signed-off-by: John Keeping <john@metanate.com>
Acked-by: Simon Glass <sjg@chromium.org>
A previous patch (net: asix: fix operation without eeprom) added a
two-byte shift to the packet buffer when receiving a packet on the
AX88772B.
This shift was not included when the driver was updated to work with
DriverModel. Testing on a Marvell DB-88F6820-ACM showed that the adapter
was not functioning correctly (EHCI timeouts).
This patch brings the two-byte shift to the DriverModel implementation
of ops->recv (asix_eth_recv).
Testing on the same board, we were able to TFTP a file over and confirm
that the crc32 was correct.
Signed-off-by: Joshua Scott <joshua.scott@alliedtelesis.co.nz>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
When enabling a fixed regulator, it may take some time to rise to the
correct voltage. If we do not delay here then subsequent operations
will fail.
Signed-off-by: John Keeping <john@metanate.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Remove the device definition from board file, update the driver with
the new compatible property and update config with necessary options.
Signed-off-by: Beniamino Galvani <b.galvani@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Add a pin controller driver for Meson GXBB adapted from Linux kernel.
Signed-off-by: Beniamino Galvani <b.galvani@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
In cases where the pins and groups definitions are in a sub-node, as:
uart_a {
mux {
groups = "uart_tx_a", "uart_rx_a";
function = "uart_a";
};
};
pinctrl_generic_set_state_subnode() returns an error for the top-level
node and pinctrl_generic_set_state() fails. Instead, return success so
that the child nodes are tried.
Signed-off-by: Beniamino Galvani <b.galvani@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Masahiro Yamada <yamada.masahiro@socionext.com>
On the raspberry pi, you can disable the serial port to gain dynamic frequency
scaling which can get handy at times.
However, in such a configuration the serial controller gets its rx queue filled
up with zero bytes which then happily get transmitted on to whoever calls
getc() today.
This patch adds detection logic for that case by checking whether the RX pin is
mapped to GPIO15 and disables the mini uart if it is not mapped properly.
That way we can leave the driver enabled in the tree and can determine during
runtime whether serial is usable or not, having a single binary that allows for
uart and non-uart operation.
Signed-off-by: Alexander Graf <agraf@suse.de>
Acked-by: Stephen Warren <swarren@wwwdotorg.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
So far we could only tell the gpio framework that a GPIO was mapped as input or
output, not as alternative function.
This patch adds support for determining whether a function is mapped as
alternative.
Signed-off-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
Acked-by: Stephen Warren <swarren@wwwdotorg.org>
Currently the command buffer gets allocated with a size of 32 bytes.
This causes warning messages on systems with cache lines bigger than
32 bytes:
CACHE: Misaligned operation at range [9df17a00, 9df17a20]
Define command buffer to be at least 32 bytes, but more if cache
line is bigger.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
Reviewed-by: Stefano Babic <sbabic@denx.de>
Now that nand_info[] is an array of pointers we need to test the
pointer itself rather than using name as a proxy for NULLness.
Fixes: b616d9b0a7 ("nand: Embed mtd_info in struct nand_chip")
Signed-off-by: Scott Wood <oss@buserror.net>
Cc: Lukasz Majewski <l.majewski@samsung.com>
Cc: Tony Lindgren <tony@atomide.com>
Acked-by: Tony Lindgren <tony@atomide.com>
In commit 17cb4b8f32 ("mtd: nand: Add+use mtd_to/from_nand and
nand_get/set_controller_data") the assignment of mtd->priv was removed
but was not replaced. This adds the required nand_set_controller_data()
call.
Signed-off-by: Chris Packham <judge.packham@gmail.com>
This is very likely to be necessary for normal use cases.
Set its default to 'y' for shorter defconfig files.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
With the CONFIG_DM_MMC_OPS migration, the .set_ios callback can
return an integer now. Return an appropriate error value rather
than sudden death by BUG().
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
No more reason to define this function above the ops structure.
Move it near the caller. Also, change its return type to void
because it never fails.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Catch up with the DM migration.
As struct dm_mmc_ops does not have .init callback, call the init
function directly from the probe function.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
When the backlight's pwm input is connected to a pwm output of the SoC,
actually use pwm to drive the backlight.
The mean reason for doing this is to fix the backlight turning off
for aprox. 1 second while the kernel is booting. This is caused by
the kernel actually using pwm to drive the backlight, so that it
can dim the backlight. First the pwm driver loads and switches the
pinmux for the pin driving the backlight's pwm input to the pwm
controller. Then about 1s later the actual backlight driver loads
and tells the pwm driver to actually update the pwm settings, which
have a power-on-reset value of "off".
An additional advantage is that this allows us to initatiate the
backlight at 80%, which is the kernel default, avoiding a brightness
change while the kernel loads.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed by: Peter Korsgaard <peter@korsgaard.com>
Patch f8a10ed1 [i2c: mvtwsi: Make address length variable] accidentally
inverted the sequence of address bytes sent to the I2C device. This
patch corrects this by sending the highest byte first and the lowest
byte last again.
Tested on theadorable Armada-XP board.
Signed-off-by: Stefan Roese <sr@denx.de>
Cc: Mario Six <mario.six@gdsys.cc>
Cc: Heiko Schocher <hs@denx.de>
In tegra20_slink.c, the set_mode() function may be executed before the
SPI bus is claimed the first time, and hence the clocks to the SPI
controller may not be running. If so, any register read/write at this
time will hang the CPU. Fix this by ensuring the clock is running as soon
as the driver is probed. This is observed on the Tegra30 Beaver board.
Apply the same clock initialization fix to all other Tegra SPI drivers so
that if set_mode() is ever implemented there, the same bug will not appear.
Note that tegra114_spi.c already operates in this fashion.
The clock manipulation code is copied from claim_bus() to probe() rather
than moved. This ensures that any calls to set_speed() take effect; the
clock can't be set once during probe and left unchanged.
Fixes: 5cb1b7b395 ("spi: tegra20: Add support for mode selection")
Cc: Mirza Krak <mirza.krak@hostmobility.com>
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Tom Warren <twarren@nvidia.com>
The nvidia,bpmp property is left over from an old BPMP I2C binding, and
shouldn't be present. Remove it from the SoC DT file, and update the
I2C driver not to parse it; the value wasn't used for anything any more
anyway.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
The Tegra SDHCI binding dictates that the reseet name for the Tegra SDHCI
clock be "sdhci" not "sdmmc", and that the clock is accessed by index
rather than by name. Fix the Tegra186 DT and MMC driver to honor this.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
The Tegra I2C binding dictates that the clock name for the Tegra I2C clock
be "div-clk" not "i2c". Fix the Tegra186 DT and I2C driver to honor this.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
The spatch series converting legacy drivers from miiphy_register to
mdio_register changed the return convention of the davinci_emac internal
MDIO accessors, making the internal code relying on it misbehaving:
no mdiodev get registered and U-Boot crashes when using net cmds in the
context of the old legacy net API.
ATM davinci_emac_initialize and cpu_eth_init don't return a proper value
in that case but fixing them would not avoid the crash.
This change is just a follow-up to the spatch pass, the MDIO accessors
of the mdiodev introduced by the spatch pass retain their proper values.
Signed-off-by: Karl Beldan <karl.beldan+oss@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
The QMan is not used in FMan IM mode, so no QMI enqueue or QMI
dequeue are performed.
Signed-off-by: Hou Zhiqiang <Zhiqiang.Hou@nxp.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
ATM when receiving a packet the whole buffer is invalidated, this change
optimizes this behaviour.
Signed-off-by: Karl Beldan <karl.beldan+oss@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Mugunthan V N <mugunthanvnm@ti.com>
check_cache_range() warns that the top boundaries are not properly
aligned when flushing or invalidating the buffers and make these
operations fail.
This gets rid of the remaining warnings:
CACHE: Misaligned operation at range
Signed-off-by: Karl Beldan <karl.beldan+oss@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Mugunthan V N <mugunthanvnm@ti.com>
ATM the rx and tx descriptors are handled as cached memory while they
lie in a dedicated RAM of the SoCs, which is an uncached area.
Removing the said dcache ops, while optimizing the logic and clarifying
the code, also gets rid of most of the check_cache_range() incurred
warnings:
CACHE: Misaligned operation at range
Signed-off-by: Karl Beldan <karl.beldan+oss@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
