When doing "i2c dev 4; i2c probe" with ENET daughter card connected
on iMX8QXP MEK board, we met a i2c bus busy issue, that the BBF of
lpi2c always show busy, but the master is idle, and stop is detected
(SDF set).
This patch addes a handling to re-init the lpi2c master for this
case. Then the issue can be worked around.
Signed-off-by: Ye Li <ye.li@nxp.com>
Acked-by: Peng Fan <peng.fan@nxp.com>
Reviewed-by: Heiko Schocher <hs@denx.de>
In xfer function, both bus_i2c_read and bus_i2c_write will
send a STOP command. This causes a problem when reading register
data from i2c device.
Generally two operations comprise the register data reading:
1. Write the register address to i2c device.
START | chip_addr | W | ACK | register_addr | ACK |
2. Read the Data from i2c device.
START | chip_addr | R | ACK | DATA | NACK | STOP
The STOP command should happen at the end of the transfer, otherwise
we will always get data from register address 0
Signed-off-by: Ye Li <ye.li@nxp.com>
Acked-by: Peng Fan <peng.fan@nxp.com>
Reviewed-by: Heiko Schocher <hs@denx.de>
For LPI2C IP, NACK is detected by the rising edge of the ninth clock.
In current uboot driver, once NACK is detected, it will reset and then
disable LPI2C master. As a result, we can never see the falling edge
of the ninth clock.
Signed-off-by: Gao Pan <pandy.gao@nxp.com>
Signed-off-by: Peng Fan <peng.fan@nxp.com>
Reviewed-by: Heiko Schocher <hs@denx.de>
Add compatible string for i.MX8 and move imx_lpi2c.h from mx7ulp directory
to u-boot include directory as a common header file.
Signed-off-by: Ye Li <ye.li@nxp.com>
Signed-off-by: Peng Fan <peng.fan@nxp.com>
Reviewed-by: Heiko Schocher <hs@denx.de>
The SDRAM must first be rewritten by zeroes if ECC is used to initialize
the ECC metadata. Make the CPU overwrite the DRAM with zeroes in such a
case. This scrubbing implementation turns the caches on temporarily, then
overwrites the whole RAM with zeroes, flushes the caches and turns them
off again. This provides satisfactory performance.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Chin Liang See <chin.liang.see@intel.com>
Cc: Dinh Nguyen <dinguyen@kernel.org>
This function was never used in SPL and the default implementation of
dram_bank_mmu_setup() does the same thing. The only difference is the
part which configures OCRAM as cachable, which doesn't really work as
it covers more than the OCRAM.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Chin Liang See <chin.liang.see@intel.com>
Cc: Dinh Nguyen <dinguyen@kernel.org>
Use "%p" to print cmdbuf.
Compilation warning as below:
CC spl/drivers/spi/cadence_qspi_apb.o
LD spl/lib/built-in.o
drivers/spi/cadence_qspi_apb.c: In function ‘cadence_qspi_apb_indirect_write_setup’:
drivers/spi/cadence_qspi_apb.c:696:18: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]
cmdlen, (unsigned int)cmdbuf);
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Acked-by: Marek Vasut <marex@denx.de>
Use "%zu" for size_t data type.
Compilation warning as below:
In file included from include/linux/bug.h:7:0,
from include/common.h:26,
from drivers/spi/cadence_qspi.c:8:
drivers/spi/cadence_qspi.c: In function ‘cadence_spi_xfer’:
drivers/spi/cadence_qspi.c:211:8: warning: format ‘%d’ expects argument of type ‘int’, but argument 3 has type ‘size_t {aka long unsigned int}’ [-Wformat=]
debug("%s: len=%d [bytes]\n", __func__, data_bytes);
^
include/linux/printk.h:37:21: note: in definition of macro ‘pr_fmt’
#define pr_fmt(fmt) fmt
^~~
include/log.h:142:2: note: in expansion of macro ‘debug_cond’
debug_cond(_DEBUG, fmt, ##args)
^~~~~~~~~~
drivers/spi/cadence_qspi.c:211:2: note: in expansion of macro ‘debug’
debug("%s: len=%d [bytes]\n", __func__, data_bytes);
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Acked-by: Marek Vasut <marex@denx.de>
In the device tree, the address for the led is located
in the parent node (for exemple leds), not in the led node
(for exemple led@0).
The commit "led: bcm6328: convert to use live dt"
(sha1: 8994551760)
change this behaviour and read the address in the led node.
We fix this by reading the base address for led
in the parent node.
Signed-off-by: Philippe Reynes <philippe.reynes@softathome.com>
In the device tree, the address for cpu is located in
the node "cpus", not in the cpu node (for exemple cpu@0).
So when probing cpu, the cpu address must be read in the
cpu parent.
The commit "cpu: bmips: convert to use live dt"
(sha1: c444afbbef)
change this behaviour and read the address in the
cpu node when probing cpu.
We fix this by reading the address in the cpu parent.
Signed-off-by: Philippe Reynes <philippe.reynes@softathome.com>
The omap_gpio driver has a TODO that says when every board is converted
to DM and DT, the omap_gpio_bind can stop using calloc and switch
to auto-alloc.
This patch converts this driver to auto-calloc when DT is enabled.
Signed-off-by: Adam Ford <aford173@gmail.com>
Add support for loading U-Boot on the Broadcom 7445 SoC. This port
assumes Broadcom's BOLT bootloader is acting as the second stage
bootloader, and U-Boot is acting as the third stage bootloader, loaded
as an ELF program by BOLT.
Signed-off-by: Thomas Fitzsimmons <fitzsim@fitzsim.org>
Cc: Stefan Roese <sr@denx.de>
Cc: Tom Rini <trini@konsulko.com>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Add code to reset all reset signals as in Ethernet DT node. A reset
property is an optional feature, so only print out a warning and do not
fail if a reset property is not present.
If a reset property is discovered, then use it to deassert, thus
bringing the IP out of reset.
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Add code to reset all reset signals as in serial DT node. A reset
property is an optional feature, so do not fail if a reset property is
not present.
If a reset property is discovered, then use it to deassert, thus
bringing the IP out of reset.
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Reviewed-by: Marek Vasut <marex@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
Add code to reset all reset signals as in mmc DT node. A reset property
is an optional feature, so only print out a warning and do not fail if a
reset property is not present.
If a reset property is discovered, then use it to deassert, thus
bringing the IP out of reset.
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Rename CONFIG_SPL_RESET_SUPPORT to CONFIG_SPL_DM_RESET, so can use
CONFIG_IS_ENABLED(DM_RESET) checking in reset.h later.
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This commit adds Actions Semi OWL family UART support. This driver
relies on baudrate configured by primary bootloaders.
Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
This commit adds Actions Semi OWL family base clock and S900 SoC
specific clock support. For S900 peripheral clock support, only UART
clock has been added for now.
Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
Add a way to decode a memory region, including the memory type (sram or
sdram) and its start address and size.
Signed-off-by: Simon Glass <sjg@chromium.org>
Add a new device_bind_ofnode() function which can bind a device given its
ofnode. This allows binding devices more easily with livetree nodes.
Signed-off-by: Simon Glass <sjg@chromium.org>
We have a 32-bit version of this function. Add a 64-bit version as well so
we can easily read 64-bit ints from the device tree.
Signed-off-by: Simon Glass <sjg@chromium.org>
On the A64 the clock for the first USB controller is actually the parent
of the clock for the second controller, so turning them off in that order
makes the system hang.
Fix this by only turning off *both* clocks when the *last* OHCI controller
is brought down. This covers the case when only one controller is used.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
When using CONFIG_OF_BOARD on rpi to use the dtb provided by the
RaspberryPi Fundation, the compatible string isn't the same, resulting
in not-functional usb from u-boot.
Signed-off-by: Oleksandr Tymoshenko <gonzo@FreeBSD.org>
Signed-off-by: Emmanuel Vadot <manu@freebsd.org>
Make the initialization sequence consistent with the Linux kernel
driver.
Reviewed-by: Stefan Roese <sr@denx.de>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Signed-off-by: Rabeeh Khoury <rabeeh@solid-run.com>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
This fixes sporadic timeout on initial packet Tx (usually ARP), with an
error message like:
timeout: packet not sent
Reviewed-by: Stefan Roese <sr@denx.de>
Tested-by: Chris Packham <judge.packham@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Signed-off-by: Rabeeh Khoury <rabeeh@solid-run.com>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
phyread can timeout and val will contain random value. Initialize it to
zero not to report random value in case of error.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
When using CONFIG_OF_BOARD on rpi to use the dtb provided by the
RaspberryPi Fundation, the compatible string isn't the same, resulting
in not-functional video in u-boot.
Signed-off-by: Oleksandr Tymoshenko <gonzo@FreeBSD.org>
Signed-off-by: Emmanuel Vadot <manu@freebsd.org>
With the introduction of early timer support in the TSC driver,
the capability of getting clock rate from device tree was lost
unfortunately. Now we bring such functionality back, but with a
limitation that when TSC is used as early timer, specifying clock
rate from device tree does not work.
This fixes random boot failures seen on QEMU targets: printing "TSC
frequency is ZERO" and reset forever.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Masking clock gate, reset register bits based on the
probed controller is proper only due to the assumption
that masking should start with 0 even thought the controller
has separate PHY or shared between OTG.
unfortunately these are fixed due to lack of separate
clock, reset drivers.
Say for example EHCI1 - EHCI3 in the datasheet (EHCI0 is for the OTG)
so we need to start reg_mask 0 - 2.
This patch calculated the mask, based on the register base
so that we can get the proper bits to set with respect to
probed controller.
We even do this masking by using PHY index specifier from dt,
but dev_read_addr_size is failing for 64-bit boards.
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
For ohci, the maximam supported endpoint number is 32(in and out), and
now we have used (usb_pipeendpoint(pipe) << 1) to index the specified
endpoint descritor, usb_pipeendpoint(pipe) can reach 0xf, so we need
change the NUM_EDs from 8 to 32.
Signed-off-by: Zeng Tao <prime.zeng@hisilicon.com>
ohci-hcd casts priv_data pointer to (ohci_t *), thus it must be
the first member in private data struct.
Fixes 831cc98b1 ("usb: sunxi: Simplify ccm reg base code")
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
The N25Q256(A) datasheet clearly states that this device does have
a Flag Status Register and does update FSR PEC bit 7 during Program
and Erase cycles to indicate the cycle is in progress. Enable the
FSR PEC bit polling on this device to prevent data corruption.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Jagan Teki <jagan@openedev.com>
Cc: Tom Rini <trini@konsulko.com>
The claim/release bus function must not reset the whole SPI core because
settings regarding wordlen, clock-frequency and so on made by
set_wordlen, set_mode, set_speed get lost with this action. Resulting in
a non-functional SPI.
Without DM the failure didn't came up since after the spi_reset within
claim bus all the setup (wordlen, mode, ...) was called, in DM they are
called by the spi uclass.
We change now the things as following for having a working SPI instance
in DM:
- move the spi_reset(...) to the probe call in DM for having a known
hardware state after probe. Without DM we don't have a probe call, so we
issue the reset as before during the claim_bus call.
- in release bus we just reset the modulctrl to the reset-value (spi-
slave)
Signed-off-by: Hannes Schmelzer <oe5hpm@oevsv.at>
Reviewed-by: Jagan Teki <jagan@openedev.com>
This bug is the combination of dwc2 USB controller and lan78xx
USB ethernet controller, which is the combination in use on
the Raspberry Pi Model 3 B+.
When the host attempts to receive a packet, but a packet has not
arrived, the lan78xx controller responds by setting BIR
(Bulk-In Empty Response) to NAK. Unfortunately, this hangs
the USB controller and requires the USB controller to
be reset.
The fix proposed is to have the lan78xx controller respond
by setting BIR to ZLP.
Signed-off-by: Andrew Thomas <andrew.thomas@oracle.com>
Tested-by: Peter Robinson <pbrobinson@gmail.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
For now, the existing SPL MXS NAND driver only supports to identify
ONFi-compliant NAND chips. In order to allow identifying
non-ONFi-compliant chips add `mxs_flash_full_ident()` which uses the
`nand_get_flash_type()` functionality from `nand_base.c` to lookup
for supported NAND chips in the chip ID list.
For compatibility reason the full identification support is only
available if the config option `CONFIG_SPL_NAND_IDENT` is enabled.
The lookup was tested on a custom i.MX6ULL board with a Toshiba
TC58NVG1S3HTAI0 NAND chip.
Signed-off-by: Jörg Krause <joerg.krause@embedded.rocks>
The existing `mxs_flash_ident()` is limited to identify ONFi compliant
NAND chips only. In order to support non-ONFi NAND chips refactor the
function and rename it to `mxs_flash_onfi_ident()`.
A follow-up patch will add `mxs_flash_full_ident()` which allows to use
the chip ID list to lookup for supported NAND flashs.
Signed-off-by: Jörg Krause <joerg.krause@embedded.rocks>
Add the config option `CONFIG_SPL_NAND_IDENT` for using the NAND chip ID list
to identify the NAND flash in SPL.
Signed-off-by: Jörg Krause <joerg.krause@embedded.rocks>
`nand_get_flash_type()` allows identification of supported NAND flashs.
The function is useful in SPL (like mxs_nand_spl.c) to lookup for a NAND
flash (which does not support ONFi) instead of using nand_simple.c and
hard-coding all required NAND parameters.
Signed-off-by: Jörg Krause <joerg.krause@embedded.rocks>
The PFUZE3000 uses registers addresses up to 0xff.
The DM pfuze100 driver supports both pfuze100 and pfuze3000. Allow it
to use the device type to return the correct number of registers.
Also rename the too generic PMIC_NUM_OF_REGS enumeration value for
pfuze3000 to match the other "PFUZE3000_" prefixed enumerations and the
pfuze100 enumeration value PFUZE100_NUM_OF_REGS.
Cc: Peng Fan <Peng.Fan@freescale.com>
Cc: Jaehoon Chung <jh80.chung@samsung.com>
Cc: Stefano Babic <sbabic@denx.de>
Cc: Fabio Estevam <fabio.estevam@nxp.com>
Signed-off-by: Trent Piepho <tpiepho@impinj.com>
Reviewed-by: Fabio Estevam <fabio.estevam@nxp.com>
Add support for specified ECC strength/size using device tree
properties nand-ecc-strength/nand-ecc-step-size.
This aligns behavior with the mainline driver, such that:
- If fsl,use-minimal-ecc is requested it will use data from
data sheet/ONFI. If this is not available the driver will fail.
- If nand-ecc-strength/nand-ecc-step-size are specified those
value will be used.
- By default maximum possible ECC strength is used
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
Support driver data from device tree. Also support fsl,use-minimal-ecc
similar to Linux' GPMI NAND driver.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
In preparation for device tree support separate board init
from controller init similar to other raw NAND drivers.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
This function initializes DMA descriptors so mxs_nand_init_dma is
more precise. It also frees up the rather generic name mxs_nand_init.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
Move GPMI and BCH register structs to the driver struct mxs_nand_info
in prepartion for device tree support.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
Add support for minimum ECC strength supported by the NAND chip.
This aligns with the behavior when using the fsl,use-minimum-ecc
device tree property in Linux.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
Report correct ECC parameters back to the stack. Do not report
bytes as we have it not immeaditly available and the Linux version
also does not report it. It seems to have no aversive effect.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
Calculate BCH geometry at start and store the information in
a structure. This avoids recalculation on every page access
and allows to calculate ECC relevant information in one place.
This patch does not change ECC layout or driver behavior in
any way.
The patch aligns the driver somewhat with the Linux GPMI NAND
driver which drives the same IP.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
Add config option which allows to enable on flash bad block table
support. This has the same effect as when using the device tree
property "nand-on-flash-bbt" in Linux.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
Instead of completing initialization via scan_bbt callback use
NAND self init to initialize the GPMI (MXS) NAND controller.
Suggested-by: Scott Wood <oss@buserror.net>
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
In preparation to convert the driver to use NAND self init
provide a new minimal init for SPL builds. As a side effect
this also reduces size of SPL by about 4KiB.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
commit 8480792287
("spi: omap3: Skip set_mode, set_speed from claim") did break SPI
support on my AM335x board.
The named commit:
- ignored the responsible arguments (speed, mode)
The set speed/mode function must use the supplied function arguments to
work properly. With this commit we take those arguments and transfer
them to the priv-data.
- used wrong udevice pointer for getting priv data
the udevice-pointer within function argument is already the spi-bus
device, so it is wrong looking here for some parent (ocp-bus in this
case) and getting priv-pointer from there.
Signed-off-by: Hannes Schmelzer <oe5hpm@oevsv.at>
Reviewed-by: Jagan Teki <jagan@openedev.com>
Otherwise the frequency is zero and the clock divider cannot be setup by
'omap3_spi_set_speed' function.
Signed-off-by: Hannes Schmelzer <oe5hpm@oevsv.at>
Reviewed-by: Jagan Teki <jagan@openedev.com>
CS GPIO activation low/high is determinated by the device tree
so we don't need to take in accoung in cs_activate and cs_deactivate
Signed-off-by: Michael Trimarchi <michael@amarulasolutions.com>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
This patch replaced "return 0" with "return status" to fix the
incorrect return value error reported by the coverity.
Reviewed-by: Michal Simek <michal.simek@xilinx.com>
Signed-off-by: Vipul Kumar <vipul.kumar@xilinx.com>
[jagan: rebased on master]
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
In Linux, the clock rate of the UART is given by the clock driver.
If you try to follow that in U-Boot, you would end up with adding
more u-boot,dm-pre-reloc properties, and also the clock driver would
be too big for SPL, which is used for UniPhier ARMv7 platform.
The current solution is to add 'clock-frequency' property to the
UART nodes, but it does not exist in the DT files in Linux. I do
not want to let DT diverge for U-Boot.
Check the SoC compatible and set the clock rate according to it.
This will be helpful to sync DT between Linux and U-Boot.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Use the clk framework to initialize clocks from drivers that need them
instead of having hardcoded frequencies and initializations from board
code.
Signed-off-by: Beniamino Galvani <b.galvani@gmail.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Introduce a basic clock driver for Amlogic Meson SoCs which supports
enabling/disabling clock gates and getting their frequency.
Signed-off-by: Beniamino Galvani <b.galvani@gmail.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
The SARADC driver was merged after the following commit :
commit d35812368a ("regmap: change regmap_init_mem() to take ofnode instead udevice")
Thus breaking build, this patch fixes the regmap_init_mem accordingly.
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
At present the code overruns the bar[] array. Fix this.
At the same time, drop the leading / from the "/spl" path so that we can
run U-Boot SPL with:
spl/u-boot-spl
rather than requiring:
/path/to/spl/u-boot-spl
Reported-by: Coverity (CID: 131199)
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Move the strdup() call so that it is only done when we know we will bind
the device.
Reported-by: Coverity (CID: 131216)
Signed-off-by: Simon Glass <sjg@chromium.org>
The recent DTs have the PHY reset GPIO in the PHY node rather than
the ethernet MAC node, support extracting the PHY reset GPIO info
from both the PHY node and ethernet MAC node.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
Cc: Joe Hershberger <joe.hershberger@ni.com>
The recent DTs have the PHY reset GPIO in the PHY node rather than
the ethernet MAC node, support extracting the PHY reset GPIO info
from both the PHY node and ethernet MAC node.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
Cc: Joe Hershberger <joe.hershberger@ni.com>
The RAVB only supports 100Full and 1000Full operation, it does not support
10Full or any Half-duplex modes. The PHY could still advertise those features
though, so filter out the PHY features accordingly.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
Cc: Joe Hershberger <joe.hershberger@ni.com>
Do not stop the clock in the start callback in case of failure, keep
them running to also keep the PHY running. The failure could be ie.
PHY failing to negotiate link and if the clock get shut down, another
attempt at bringing the link up would fail. The clock right now are
started in probe function and stopped in remove function, which is
the correct behavior.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
Cc: Joe Hershberger <joe.hershberger@ni.com>
The sh_pfc_{read,write}() must operate on the register address directly
rather than on an offset, fix this to prevent illegal access.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
There are multiple GPIO banks with up to 32 pins / bank. When
using 'gpio status -a' to read the pins, this patch displays
both GPIO<bank>_<index> similar to how the device trees
display in addition to displaying gpio_#
Signed-off-by: Adam Ford <aford173@gmail.com>
When enabling BLOCK_CACHE on devices with limited RAM during SPL,
some devices may not boot. This creates an option to enable
block caching in SPL by defaults off. It is dependent on SPL_BLK
Fixes: 46960ad6d0 ("block: Have BLOCK_CACHE default to y in some cases")
Signed-off-by: Adam Ford <aford173@gmail.com>
The naming with "UART" is obviously wrong, we fix this here.
Signed-off-by: Hannes Schmelzer <oe5hpm@oevsv.at>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Fabio Estevam <fabio.estevam@nxp.com>
This adds a DM video driver for U-Boot as the EFI payload. The driver
makes use of all necessary information from the passed EFI GOP info
to create a linear framebuffer device, as if it were initialized by
U-Boot itself.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anatolij Gustschin <agust@denx.de>
This function is used only inside this driver that's why should be
static.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This issue was found when OF_LIVE was enabled that there are scrambled
chars on the console like this:
Chip ID: zu3eg
Watchdog: Started��j� sdhci@ff160000: 0, sdhci@ff170000: 1
In: serial@ff010000
I found a solution for this problem exactly the same as I found later in
serial_msm fixed by:
"serial: serial_msm: initialize uart only before relocation"
(sha1: 7e5ad796bc)
What it is happening is that output TX fifo still contains chars to be
sent and _uart_zynq_serial_init() resets TX fifo even in the middle of
transfer.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Change logic and put char to fifo till there is a space in output fifo.
Origin logic was that output fifo needs to be empty. It means only one
char was in output queue.
Also remove unused ZYNQ_UART_SR_TXEMPTY macro.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Coding style is checking to use BIT macros instead of shifts.
The patch is also fixing the rest of macros which should be BITs instead
of hex numbers.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Use u32 instead of int for max_bank, bank_min and bank_max. These values
can't be negative that's why no reason to use signed type.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Code around tuning_loop_counter variable expects to go below zero.
That's why this variable can't use unsigned type.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
This patch fixes the mmc tuning command failures
when tuning pattern data needs to read back for
comparision against the expected bit pattern.
Reported-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: Siva Durga Prasad Paladugu <siva.durga.paladugu@xilinx.com>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Tested-by: Masahiro Yamada <yamada.masahiro@socionext.com>
This function is required for adding bootstage support.
Also enable it directly for ZynqMP R5 configuration.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
struct sunxi_ccm_reg doesn't have ahb_reset0_cfg on sun4i and sun5i,
thus compilation fails with:
drivers/usb/host/ohci-sunxi.c:96:26: error: 'struct sunxi_ccm_reg' has
no member named 'ahb_reset0_cfg'
Access this reg using its offset to fix this issue.
Fixes commit 1ed9c1118 ("usb: sunxi: ehci: get rid of ifdefs")
and commit 56830cee3 ("usb: sunxi: ohci: get rid of ifdefs")
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Functions can return NULL in case of error that's why checking return
value is needed.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
In case of phyread()/phy_setup_op() timeout code is working with
uninitialized phyreg variable. Initialize this variable to make sure
that code it not working with random value.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
wait_for_bit_le32 returns negative value on failure. Fix phy...() to
handle these failures properly.
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Combine repeated code from smi_reg_read/smi_reg_write into a common
function smi_wait_ready.
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Chris Packham <judge.packham@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
On the SPEAr600 SoC, which has the dwmac1000 variant of the IP block,
the DMA reset never succeeds when a MII PHY is used (no problem with a
GMII PHY). The designware_eth_init() function sets the
DMAMAC_SRST bit in the DMA_BUS_MODE register, and then
polls until this bit clears. When a MII PHY is used, with the current
driver, this bit never clears and the driver therefore doesn't work.
The reason is that the PS bit of the GMAC_CONTROL register should be
correctly configured for the DMA reset to work. When the PS bit is 0,
it tells the MAC we have a GMII PHY, when the PS bit is 1, it tells
the MAC we have a MII PHY.
Doing a DMA reset clears all registers, so the PS bit is cleared as
well. This makes the DMA reset work fine with a GMII PHY. However,
with MII PHY, the PS bit should be set.
We have identified this issue thanks to two SPEAr600 platform:
- One equipped with a GMII PHY, with which the existing driver was
working fine.
- One equipped with a MII PHY, where the current driver fails because
the DMA reset times out.
Note: Taken from https://www.spinics.net/lists/netdev/msg432578.html
Signed-off-by: Quentin Schulz <quentin.schulz@bootlin.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
When connecting to from a CPU direct to a 88e6097 typically RGMII is
used. In order for traffic to actually pass we need to force the link up
so the CPU MAC on the other end will see the link.
Signed-off-by: Chris Packham <judge.packham@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
This makes sure the DMA buffers are properly aligned for the
hardware.
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Jon Nettleton <jon@solid-run.com>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Add driver for Socionext AVE ethernet controller that includes MAC and
MDIO bus supporting RGMII/RMII modes.
The driver behaves the ethernet driver model (DM_ETH) with devicetree.
Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
VCAP tables must be initialized even if no advanced classification
is used. If no initialization is performed, then ECC error will
be observed by the user when the first packet enters the l2switch.
The error is marked in MPIC_EISR0 -bit 29 which means - Internal RAM
multi-bit ECC error.
This patch fixes the aforementioned ECC error by performing the
initialization of VCAP tables.
Signed-off-by: Radu Bulie <radu-andrei.bulie@nxp.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
If you send a final packet just before stopping the interface (e.g. a final
ACK as part of the UDP fastboot protocol), then that packet isn't reliably
delivered onto the wire.
Reap packets prior to stopping the interface to ensure any which are
in-flight make it out. Also remove buffer and len from the call to
cpdma_process() as we weren't using them on their return.
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Add support for Cortina CS4223 10G PHY
- As per the CS4223 specs, an EEPROM module is
connected to the PHY. At startup the PHY reads
the firmware line and tries to load the firmware
into the internal memory.
- This driver reads the EEPROM status
and checks if firmware has been loaded
Signed-off-by: Vicentiu Galanopulo <vicentiu.galanopulo@nxp.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Put the enetaddr data in the same order as it was before the change in
commit ace1520cb5 ("net: sunxi-emac: Write HW address via function")
Reported-by: Udo Maslo <u.maslo@web.de>
Signed-off-by: Joe Hershberger <joe.hershberger@ni.com>
Bus translations should be applied when reading the address of the sgmii
phy registers from the DT. Use ofnode_get_addr_index instead of the
plain ofnode_read_u32_default to fix this.
Signed-off-by: Mario Six <mario.six@gdsys.cc>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
On A64 OHCI1 clock source is OHCI0 clock, so we need to enable OHCI0
clock when OHCI1 is in use.
Fixes commit dd3228170a ("usb: sunxi: Switch to use generic-phy")
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
If U-Boot gets used as coreboot payload all pci resources got
assigned by coreboot. If a dts without any pci ranges gets used
the dm is not able to access pci device memory. To get things
working make use of a 1:1 mapping for bus <-> phy addresses.
This change makes it possible to get the e1000 U-Boot driver
working on a sandybridge device where U-Boot is used as coreboot
payload.
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
[bmeng: fixed 'u-boot' in the commit message]
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
If we use U-Boot as coreboot payload with a generic dts without
any ranges specified we fail in pci pre_probe and our pci bus
is not usable.
So convert decode_regions(..) into a void function and do the simple
error handling there.
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
[bmeng: fixed 'u-boot' in the commit message and checkpatch warning]
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This fixes the following compiler warning:
"warning: cast from pointer to integer of different size
[-Wpointer-to-int-cast]"
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Starting with cpuid level 0x16 (Skylake-based processors)
it is possible to get CPU base freq via cpuid.
This fixes booting on a skylake based system.
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
[bmeng: fixed wrong indention of labels]
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Added the following:
1. defconfig for LS1012AFRWY Secure boot
2. PfE Validation support
Signed-off-by: Vinitha V Pillai <vinitha.pillai@nxp.com>
Reviewed-by: York Sun <york.sun@nxp.com>
Add WARN_ONCE definition to allow single time notification
of warnings to the user.
Taken from Linux kernel (4.17) with slight changes
(Removed __section(.data.once))
Signed-off-by: Ramon Fried <ramon.fried@gmail.com>
[trini: Drop the musb and dwc3 compat versions]
Signed-off-by: Tom Rini <trini@konsulko.com>
When dealing with filesystems that come from block devices we can get a
noticeable performance gain in some use cases from having the block
cache enabled. The code paths are valid in other cases when we have BLK
set and may provide wins in raw reads in some use cases, so have this be
default when BLK is enabled.
Signed-off-by: Tom Rini <trini@konsulko.com>
Equivalent code that disables the hidden i2c0 slave already exists in
the Turris Omnia platform specific code. But this hidden i2c0 slave that
interferes the i2c bus is not board specific. Armada 38x SoCs and at
least some Kirkwood variants are affected as well. Add code to disable
this slave to the i2c bus driver to make it work on all affected
hardware.
Use the bind callback because we want this to always run at boot,
regardless of whether U-Boot uses the i2c bus.
Cc: Rabeeh Khoury <rabeeh@solid-run.com>
Cc: Chris Packham <judge.packham@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Reviewed-by: Heiko Schocher <hs@denx.de>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
I just stumbled over some cluttered UBI messages. It seems some newline
chars are missing in the current U-Boot UBI source. Lets fix this
in U-Boot as well (Linux has those fixes already).
Signed-off-by: Stefan Roese <sr@denx.de>
Cc: Heiko Schocher <hs@denx.de>
Since commit 0e373c0ade ("spl: add SPL_RESET_SUPPORT"),
reset is supported in SPL, enable this flag for STM32F SoCs family.
This allows to remove a specific case in RCC mfd driver.
Signed-off-by: Patrice Chotard <patrice.chotard@st.com>
Split the rtc_{get,set,reset} functions so that the bodies can be used
in a DM driver.
Signed-off-by: Chris Packham <judge.packham@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
This flash IC is used in some chromebook models
manufactured by Bitland.
Signed-off-by: Carlo Caione <carlo@endlessm.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
The clean_bar() function resets the SPI NOR BAR register to 0, but
does not set the flash->curr_bar to 0 , therefore those two can get
out of sync, which could ultimatelly result in corrupted flash content.
The simplest test case is this:
=> mw 0x10000000 0x1234abcd 0x4000
=> sf probe
=> sf erase 0x1000000 0x10000
=> sf write 0x10000000 0x1000000 0x10000
=> sf probe ; sf read 0x12000000 0 0x10000 ; md 0x12000000
That is, erase a sector above the 16 MiB boundary and write it with
random pre-configured data. What will actually happen without this
patch is the sector will be erased, but the data will be written to
BAR 0 offset 0x0 in the flash.
This is because the erase command will call write_bar()+clean_bar(),
which will leave flash->bank_curr = 1 while the hardware BAR registers
will be set to 0 through clean_bar(). The subsequent write will also
trigger write_bar()+clean_bar(), but write_bar checks if the target
bank == flash->bank_curr and if so, does NOT reconfigure the BAR in
the SPI NOR. Since flash->bank_curr is still 1 and out of sync with
the HW, the condition matches, BAR programming is skipped and write
ends up at address 0x0, thus corrupting flash content.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Tom Rini <trini@konsulko.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
Change to use devfdt_get_addr_index() function to get fdt address.
Original code has compilation warning below:
drivers/spi/cadence_qspi.c: In function ‘cadence_spi_ofdata_to_platdata’:
drivers/spi/cadence_qspi.c:297:18: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
plat->regbase = (void *)data[0];
^
drivers/spi/cadence_qspi.c:298:18: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
plat->ahbbase = (void *)data[2];
^
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Acked-by: Marek Vasut <marex@denx.de>
Reviewed-by: Jagan Teki <jagan@openedev.com>
If compile with support for the efi loader we need to mark the pages
allocated for the framebuffer as reserved so the kernel won't attempt
to use them for other uses.
Signed-off-by: Emmanuel Vadot <manu@freebsd.org>
Acked-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Signed-off-by: Alexander Graf <agraf@suse.de>
Also fix bad accents in my name.
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Also fix bad accents in my name.
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Also fix bad accents in my name.
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Also fix bad accents in my name.
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
This patch adds support for loading secure bitstreams on ZynqMP
platforms. The secure bitstream images has to be generated using
Xilinx bootgen tool.
Signed-off-by: Siva Durga Prasad Paladugu <siva.durga.paladugu@xilinx.com>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
This patch adds support to load secure bitstreams(authenticated or
encrypted or both). As of now, this feature is added and tested only
for xilinx bitstreams and the secure bitstream was generated using
xilinx bootgen tool, but the command is defined in more generic way.
Command example to load authenticated and device key
encrypted bitstream is as follows
"fpga loads 0 100000 2000000 0 1"
Signed-off-by: Siva Durga Prasad Paladugu <siva.durga.paladugu@xilinx.com>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Add new compatible to the Ethernet AVB driver for R8A77990 E3 SoC.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Joe Hershberger <joe.hershberger@ni.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
Add new compatible to the Uniphier SD driver for R8A77990 E3 SoC.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Masahiro Yamada <yamada.masahiro@socionext.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
Add new compatible to the GPIO driver for R8A77990 E3 SoC.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
This patch adds initial pinctrl driver to support for the R8A77990 SoC.
Signed-off-by: Takeshi Kihara <takeshi.kihara.df@renesas.com>
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
This follows the style of existion PORT_GP_X macros and
will be used by a follow-up patch for the r8a77990 SoC.
Signed-off-by: Takeshi Kihara <takeshi.kihara.df@renesas.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
The PE clock have two parents, add support for picking the correct
one and deriving the clock from it.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
The PLL rate could be in the GHz range, which could overflow a 32bit
data type. Since the hardware is 64bit anyway, pass the clock rates
as 64bit number internally to avoid this.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
The mul and div arguments were reported in reverse order in the debug
message, swap them to fix this.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
This patch adds HS200 suuport for ZynqMP and enables
the same for ZC1751 DC1 board which has eMMC on it.
Signed-off-by: Siva Durga Prasad Paladugu <siva.durga.paladugu@xilinx.com>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Use live-tree functions.
Reported-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
There is no reason to specify header with full soc name.
Symlink is setup automatically (arch -> arch-zynqmp)
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Convert USB fastboot code to use shared fastboot protocol.
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Introduce 'oem format' which matches the USB implementation, guard this
with CONFIG_FASTBOOT_CMD_OEM_FORMAT so that you can configure it out.
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Introduce CONFIG_IMAGE_SPARSE and CONFIG_CMD_MMC_SWRITE so the "mmc
swrite" command is separated from the fastboot code.
Move image-sparse from common to lib so it's clear it's library code.
Rename CONFIG_FASTBOOT_FLASH_FILLBUF_SIZE to CONFIG_IMAGE_SPARSE_FILLBUF_SIZE
and migrate it to Kconfig.
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Acked-by: Jassi Brar <jaswinder.singh@linaro.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
Add FASTBOOT_FLASH_NAND_TRIMFFS to Kconfig; note there are no in-tree
users of it.
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Rename fb_mmc_flash_write/fb_mmc_erase/fb_nand_flash_write/fb_nand_erase to
fastboot_... as they form a public interface
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Move FASTBOOT_VERSION to include/fastboot.h so when we merge the UDP code
we only have one definition.
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Rename fb_set_reboot_flag to fastboot_set_reboot_flag so it matches
all other fastboot code in the global name space. Fix the guards around
them so that they're dependent on FASTBOOT, not just USB_FUNCTION_FASTBOOT.
Move the weak implementation of fastboot_set_reboot_flag to fb_common.c
so we can call it from non-USB fastboot code.
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Compiling on a 64 bit target the arguments to _fb_nand_write are
incompatible:
drivers/fastboot/fb_nand.c: In function ‘_fb_nand_write’:
drivers/fastboot/fb_nand.c:101:42: warning: passing argument 3 of ‘nand_write_skip_bad’ from incompatible pointer type [-Wincompatible-pointer-types]
return nand_write_skip_bad(mtd, offset, &length, written,
^
In file included from drivers/fastboot/fb_nand.c:16:0:
include/nand.h:107:5: note: expected ‘size_t * {aka long unsigned int *}’ but argument is of type ‘unsigned int *’
int nand_write_skip_bad(struct mtd_info *mtd, loff_t offset, size_t *length,
^~~~~~~~~~~~~~~~~~~
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
CONFIG_FASTBOOT_GPT_NAME and CONFIG_FASTBOOT_MBR_NAME are always defined
by Kconfig if you're compiling this code, so remove these redundant
defaults.
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Add newlines so we format our output correctly.
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
Reviewed-by: Jocelyn Bohr <bohr@google.com>
