If env is stored on SD card then U-Boot SPL automatically calls mmc_init()
before it is going to load proper U-Boot from SD card.
If env is not stored on SD card then U-Boot SPL fails to read proper U-Boot
from SD card due to missing mmc_init() call.
So add missing mmc_init() call into fsl_esdhc_spl's mmc_boot() function.
It fixes booting from SD card on P2020 boards without env support in SPL.
mmc_init() returns early if card was already initialized, so there is no
issue with calling this function more times.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Priyanka Jain <priyanka.jain@nxp.com>
In pre-PBL header is stored size of code which BootROM copies from SD card
to L2/SRAM. This size has upper limit of L2 cache size. In most cases this
is size of U-Boot SPL or size of L2 cache.
Therefore this size in pre-PBL header cannot be used for determining size
of proper U-Boot.
So always use CONFIG_SYS_MMC_U_BOOT_SIZE for determining size of proper
U-Boot which stored on SD card.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Priyanka Jain <priyanka.jain@nxp.com>
Pre-PBL BootROMs (MPC8536E, MPC8569E, P2020, P1011, P1012, P1013, P1020,
P1021, P1022) require custom BOOT signature on sector 0 and MBR/DBR
signature is not required at all.
So add check for BOOT signature and remove check for MBR/DBR.
This allows U-Boot SPL to load proper U-Boot on pre-PBL BootROMs platforms
also from SD cards which do not have MBR/DBR signature on sector 0.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Priyanka Jain <priyanka.jain@nxp.com>
The Rockchip SoCs rk3066/rk3188 have MMC DT nodes
with as compatible string "rockchip,rk2928-dw-mshc".
Add OF_PLATDATA support to the existing driver with
help of a DM_DRIVER_ALIAS.
This type needs a permanent enabled fifo.
The other Rockchip SoCs always have the property
"u-boot,spl-fifo-mode" in the MMC DT nodes,
because MMC to SRAM can't do DMA.
Make this property a requirement for MMC OF_PLATDATA
structures. The property "fifo-mode" must be added
for all other compile modes.
Signed-off-by: Johan Jonker <jbx6244@gmail.com>
Reviewed-by: Kever Yang <kever.yang@rock-chips.com>
The document rockchip-dw-mshc.yaml decribes a maximum of 4 clocks.
In the rockchip_dw_mmc driver the clock name in use was "fixed"
to "ciu" with index 1, but later reverted back to index 0.
The clock drivers can handle both, but the calling driver
should submit correct data as a standard practice.
Fix the "ciu" clock index by setting it back to 1.
clock-names:
minItems: 2
items:
- const: biu
- const: ciu
- const: ciu-drive
- const: ciu-sample
Signed-off-by: Johan Jonker <jbx6244@gmail.com>
Reviewed-by: Kever Yang <kever.yang@rock-chips.com>
A pointer can not be negative. Use macro IS_ERR_OR_NULL() for checking.
Signed-off-by: Haolin Li <li.haolin@qq.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Kever Yang <kever.yang@rock-chips.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Add compatible string for i.MX8MP, which permits i.MX8MP to use
HS400ES mode, just like all the other i.MX8M.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Fabio Estevam <festevam@denx.de>
Cc: Haibo Chen <haibo.chen@nxp.com>
Cc: Peng Fan <peng.fan@nxp.com>
Cc: Stefano Babic <sbabic@denx.de>
Reviewed-by: Fabio Estevam <festevam@denx.de>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
This macro currently supports only one parameter. Based on Linux iopoll,
let's extend read_poll_timeout common API to allow multiple variable
parameters.
Signed-off-by: Ariel D'Alessandro <ariel.dalessandro@collabora.com>
Every time a mmc bus/port is scanned and a new device is detected,
we want to call device_probe() as it will give us a chance to run
additional post-processings for some purposes.
In particular, support for creating partitions on a device will be added.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
The DLL mode supported SD reference clocks are 50 MHz, 100 MHz and
200 MHz. When user select SD frequency as 200MHz in the design, the
actual frequency is going to come around ~187MHz (<= 200MHz considering
the parent clock and divisor selection). We need to set SDx_BASECLK as
200 in this case, setting 187 will result in tuning failures in mmc.
Set SDx_BASECLK to exact value of 200, 100 or 50 based on the frequency
range.
Signed-off-by: Ashok Reddy Soma <ashok.reddy.soma@xilinx.com>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Link: https://lore.kernel.org/r/6c1e5eeeedd2864a0c85e6b409d182031d8c6c1a.1648210268.git.michal.simek@xilinx.com
microblaze:
- Add support for reserved memory
xilinx:
- Update FRU code with MAC reading
zynqmp:
- Remove double AMS setting
- DT updates (mostly for SOMs)
- Add support for zcu106 rev 1.0
zynq:
- Update nand binding
nand:
- Aligned zynq_nand to upstream DT binding
net:
- Add support for ethernet-phy-id
mmc:
- Workaround CD in zynq_sdhci driver also for ZynqMP
- Add support for dynamic/run-time SD config for SOMs
gpio:
- Add driver for slg7xl45106
firmware:
- Add support for dynamic SD config
power-domain:
- Update zynqmp driver with the latest firmware
video:
- Add skeleton driver for DP and DPDMA
i2c:
- Fix i2c to work with QEMU
pinctrl:
- Add driver for zynqmp pinctrl driver
-----BEGIN PGP SIGNATURE-----
iF0EABECAB0WIQQbPNTMvXmYlBPRwx7KSWXLKUoMIQUCYjIBkgAKCRDKSWXLKUoM
IYUwAKCYRBxb59BFq4MRvTbNcRZ7H4oGIQCfVUl3x9KZ7nFZaYAaBXFBxwna+PE=
=irCR
-----END PGP SIGNATURE-----
Merge tag 'xilinx-for-v2022.07-rc1' of https://source.denx.de/u-boot/custodians/u-boot-microblaze into next
Xilinx changes for v2022.07-rc1
microblaze:
- Add support for reserved memory
xilinx:
- Update FRU code with MAC reading
zynqmp:
- Remove double AMS setting
- DT updates (mostly for SOMs)
- Add support for zcu106 rev 1.0
zynq:
- Update nand binding
nand:
- Aligned zynq_nand to upstream DT binding
net:
- Add support for ethernet-phy-id
mmc:
- Workaround CD in zynq_sdhci driver also for ZynqMP
- Add support for dynamic/run-time SD config for SOMs
gpio:
- Add driver for slg7xl45106
firmware:
- Add support for dynamic SD config
power-domain:
- Update zynqmp driver with the latest firmware
video:
- Add skeleton driver for DP and DPDMA
i2c:
- Fix i2c to work with QEMU
pinctrl:
- Add driver for zynqmp pinctrl driver
On RK3568, a register bit must be set to enable Enhanced Strobe.
However, it appears that the address of this register may differ from
vendor to vendor and should be read from the underlying MMC IP. Let the
Rockchip SDHCI driver read this address and set the relevant bit when
Enhanced Strobe configuration is requested.
The IP uses a custom mode select value (0x7) for HS400, use that instead
of the common but non-standard SDHCI_CTRL_HS400 value (0x5). Also add
some necessary DLL_STRBIN and DLL_TXCLK configuration for HS400.
Additionally, a bit signifying that the connected hardware is an eMMC
chip must be set to enable Data Strobe for HS400 and HS400ES modes. Also
make the driver set this bit as appropriate.
This is partly ported from Linux's Synopsys DWC MSHC driver which
happens to be the underlying IP. (drivers/mmc/host/sdhci-of-dwcmshc.c in
Linux tree).
Co-developed-by: Yifeng Zhao <yifeng.zhao@rock-chips.com>
Signed-off-by: Yifeng Zhao <yifeng.zhao@rock-chips.com>
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Reviewed-by: Kever Yang <kever.yang@rock-chips.com>
On RK3399, a register bit must be set to enable Enhanced Strobe.
Let the Rockchip SDHCI driver set it when Enhanced Strobe configuration
is requested. However, having it set makes the lower-speed modes stop
working and makes reinitialization fail, so let it be unset as needed in
set_control_reg().
This is mostly ported from Linux's Arasan SDHCI driver which happens
to be the underlying IP. (drivers/mmc/host/sdhci-of-arasan.c in Linux
tree).
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Kever Yang <kever.yang@rock-chips.com>
Delegate setting the Enhanced Strobe configuration to individual drivers
if they set a function for it. Return -ENOTSUPP if they do not, like
what the MMC uclass does.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Kever Yang <kever.yang@rock-chips.com>
Generic SDHCI driver received support for checking the busy status by
polling the DAT[0] level instead of waiting for the worst MMC switch time.
Unfortunately, it appears that this does not work for Xenon controllers
despite being a part of the standard SDHCI registers and the Armada 3720
datasheet itself telling that BIT(20) is useful for detecting the DAT[0]
busy signal.
I have tried increasing the timeout value, but I have newer managed to
catch DAT_LEVEL bits change from 0 at all.
This issue appears to hit most if not all SoC-s supported by Xenon driver,
at least A3720, A8040 and CN9130 have non working eMMC currently.
So, until a better solution is found drop the wait_dat0 OP for Xenon.
I was able to only test it on A3720, but it should work for others as well.
Fixes: 40e6f52454 ("drivers: mmc: Add wait_dat0 support for sdhci driver")
Signed-off-by: Robert Marko <robert.marko@sartura.hr>
Reviewed-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Stefan Roese <sr@denx.de>
After commit f132aab403 ("Revert "mmc: fsl_esdhc_imx: use
VENDORSPEC_FRC_SDCLK_ON to control card clock output""), it
involve issue in mmc_switch_voltage(), because of the special
design of usdhc.
For FSL_USDHC, it do not implement VENDORSPEC_CKEN/PEREN/HCKEN/IPGEN,
these are reserved bits(Though RM contain the definition of these bits,
but actually internal IC logic do not implement, already confirm with
IC team). Instead, use VENDORSPEC_FRC_SDCLK_ON to gate on/off the card
clock output. Here is the definition of this bit in RM:
[8] FRC_SDCLK_ON
Force CLK output active
Do not set this bit to 1 unless it is necessary. Also, make sure that
this bit is cleared when uSDHC’s clock is about to be changed (frequency
change, clock source change, or delay chain tuning).
0b - CLK active or inactive is fully controlled by the hardware.
1b - Force CLK active
In default, the FRC_SDCLK_ON is 0. This means, when there is no command
or data transfer on bus, hardware will gate off the card clock. But in
some case, we need the card clock keep on. Take IO voltage 1.8v switch
as example, after IO voltage change to 1.8v, spec require gate off the
card clock for 5ms, and gate on the clock back, once detect the card
clock on, then the card will draw the dat0 to high immediately. If there
is not clock gate off/on behavior, some card will keep the dat0 to low
level. This is the reason we fail in mmc_switch_voltage().
To fix this issue, and concern that this is only the fsl usdhc hardware
design limitation, set the bit FRC_SDCLK_ON in the beginning of the
wait_dat0() and clear it in the end. To make sure the 1.8v IO voltage
switch process align with SD specification.
For standard tuning process, usdhc specification also require the card
clock keep on, so also add these behavior in fsl_esdhc_execute_tuning().
Reviewed-by: Marek Vasut <marex@denx.de>
Tested-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: Haibo Chen <haibo.chen@nxp.com>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
The Rockchip RK3399 eMMC PHY has to be power-cycled while changing its
clock speed to some higher speeds. This is dependent on the desired
SDHCI clock speed, and it looks like the PHY should be powered off while
setting the SDHCI clock in these cases.
Commit ac804143cf ("mmc: rockchip_sdhci: add phy and clock config for
rk3399") attempts to do this in the set_ios_post() hook by setting the
SDHCI clock once more while the PHY is turned off/on as necessary, as
the SDHCI framework does not provide a way to override how it sets its
clock. However, the commit breaks reinitializing the eMMC on a few
boards including chromebook_kevin and reportedly ROCKPro64.
This patch reworks the power cycling to utilize the SDHCI framework
slightly better (using the set_control_reg() hook to power off the PHY
and set_ios_post() hook to power it back on) which happens to fix the
issue, at least on a chromebook_kevin.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Add support for dynamic configuration which will takes care of
configuring the SD secure space configuration registers using firmware
APIs, performing SD reset assert and deassert.
High level sequence:
- Check for the PM dynamic configuration support, if no error proceed
with SD dynamic configurations(next steps) otherwise skip the dynamic
configuration.
- Put the SD Controller in reset.
- Configure SD Fixed configurations.
- Configure the SD Slot Type.
- Configure the BASE_CLOCK.
- Configure the 8-bit support.
- Bring the SD Controller out of reset.
In the above steps, apart from the Fixed configurations, remaining all
configurations are dynamic and they will be read from devicetree.
And also remove hardcoded secure register writes, as dynamic sd config
support is added.
Signed-off-by: Ashok Reddy Soma <ashok.reddy.soma@xilinx.com>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Link: https://lore.kernel.org/r/030a3ec041ff3efebd574b4d2b477ad85f12cbce.1645626962.git.michal.simek@xilinx.com
In the Linux kernel (v5.16) for this SoC there are two separate drivers
- namely sdhci-omap.c and omap_hsmmc.c which have separate set of
compatibles.
The U-Boot's drivers/mmc/omap_hsmmc.c driver supports both eMMC and
SD devices - hence the compatible for SDHCI can be added.
After this change the am335x DTS description can be easier ported
from Linux kernel.
Signed-off-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
In the workaround added with 'commit b6f44082d5 ("mmc: zynq_sdhci: Wait
till sd card detect state is stable")' the timeout variable has post
decrement. Whenever timeout happens, this post decrement is making
timeout=0xffffffff, so timeout error print and return statement are
never reached. Fix it by decrementing it inside the while loop.
Signed-off-by: Ashok Reddy Soma <ashok.reddy.soma@xilinx.com>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Link: https://lore.kernel.org/r/61fc1160ada0dd622cd29e381a74af7bf3d9a200.1645625609.git.michal.simek@xilinx.com
The original code logic can not show the correct card clock, and also
has one risk when the div is 0. Because there is div -=1 before.
So move the operation before div -=1, and also involve ddr_pre_div
to get the correct value.
Signed-off-by: Haibo Chen <haibo.chen@nxp.com>
Reviewed-by: Marek Vasut <marex@denx.de>
Now original fsl_esdhc.c are split as fsl_esdhc.c and fsl_esdhc_imx.c.
fsl_esdhc_imx.c only cover i.MX SoC. So ARCH_MXC is redundant.
Signed-off-by: Haibo Chen <haibo.chen@nxp.com>
Reviewed-by: Marek Vasut <marex@denx.de>
The Linux driver automatically can detect and enable UHS, HS200, HS400
and HS400_ES automatically without extra flags being placed into the
device tree.
Right now, for U-Boot to use UHS, HS200 or HS400, the extra flags are
needed in the device tree. Instead, go through the esdhc_soc_data
flags and enable the host caps where applicable to automatically
enable higher speeds.
Suggested-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: Adam Ford <aford173@gmail.com>
At present the IS_ENABLED() macro has extra brackets, making it possible
to write:
if IS_ENABLED(CONFIG_XXX)
but it is a bit confusing. Add the missing brackets.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Seems that we need the waterlevel setting not only for PIO mode as
without this at least the i.MX 8M Mini won't boot anymore when being
written by such a U-Boot. Corruption has also been observed both on
the i.MX 6 as well as i.MX 8M Mini when using ums on the eMMC. Fix
this by setting the watermark level again regardless of whether in
DMA or PIO mode.
Fixes: 41c6a22fc2 ("mmc: fsl_esdhc_imx: simplify esdhc_setup_data()")
Signed-off-by: Marcel Ziswiler <marcel.ziswiler@toradex.com>
Reviewed-by: Fabio Estevam <festevam@gmail.com>
Tested-by: Fabio Estevam <festevam@gmail.com>
Now that the sunxi_gpio driver handles pull-up/down via the driver
model, pin configuration does not need a platform-specific function.
Signed-off-by: Samuel Holland <samuel@sholland.org>
Tested-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
When failing to send a command because the hardware is busy, return
EBUSY to indicate the cause instead of just -1.
Signed-off-by: John Keeping <john@metanate.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
[ fsl_esdhc commit 263ddfc345 ]
The initial clock setting should be through sysctl register only,
while the mmc_set_clock() will call mmc_set_ios() introduce other
configurations like bus width, mode, and so on.
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
This builds on the previous patch by converting yet more preprocessor
macros to C ifs. This is split off so that the changes adapted from
Micheal's patch may be clearly distinguished from the ones I have
authored myself.
MMC_SUPPORTS_TUNING should really get a Kconfig conversion. And DM_GPIO
needs some -ENOSYS stubs when it isn't defined.
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
[ fsl_esdhc commit 52faec3182 ]
Make the code cleaner and drop the old-style #ifdef constructs where it is
possible.
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
[ fsl_esdhc commit 7e48a028a4 ]
First, we need the waterlevel setting for PIO mode only. Secondy, both DMA
setup code is identical for both directions, except for the data pointer.
Thus, unify them.
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
[ fsl_esdhc commit b1ba1460a4 ]
Use the dma_{map,unmap}_single() calls. These will take care of the
flushing and invalidation of caches.
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
[ fsl_esdhc commit da86e8cfcb ]
SDMA can only do DMA with 32 bit addresses. This is true for all
architectures (just doesn't apply to 32 bit ones). Simplify the code and
remove unnecessary CONFIG_FSL_LAYERSCAPE.
Also make the error message more concise.
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
[ fsl_esdhc commit 30f6444d02 ]
Fix mmc->clock with actual clock which is divided by the
controller, and record it with priv->clock.
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
[ fsl_esdhc commit commit 08197cb8df ]
Drop redundant code for non-removable feature. "non-removable" property
has been read in mmc_of_parse().
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
[ set MMC_CAP_NONREMOVABLE in plat->cfg.host_caps ]
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
[ fsl_esdhc commit 07bae1de38 ]
This patch is to clean up bus width setting code.
- For DM_MMC, remove getting "bus-width" from device tree.
This has been done in mmc_of_parse().
- For non-DM_MMC, move bus width configuration from fsl_esdhc_init()
to fsl_esdhc_initialize() which is non-DM_MMC specific.
And fix up bus width configuration to support only 1-bit, 4-bit,
or 8-bit. Keep using 8-bit if it's not set because many platforms
use driver without providing max bus width.
- Remove bus_width member from fsl_esdhc_priv structure.
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
[ converted if statement to switch ]
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
[ fsl_esdhc commit 5b05fc0310 ]
Voltage validation should be done by CMD8. Current comparison between
mmc_cfg voltages and host voltage capabilities is meaningless.
So drop current comparison and let voltage validation is through CMD8.
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
[ fsl_esdhc commit 2913926f3b ]
Remove redundant DM_MMC checking which is already in DM_MMC conditional
compile block.
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
U-boot prefers DM_MMC + BLK for MMC. Now eSDHC driver has already
support it, so let's force to use it.
- Drop non-BLK support for DM_MMC introduced by below patch.
66fa035 mmc: fsl_esdhc: fix probe issue without CONFIG_BLK enabled
- Support only DM_MMC + BLK (assuming BLK is always enabled for DM_MMC).
- Use DM_MMC instead of BLK for conditional compile.
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Add support for testing whether bus voltage level translator is present
and operational. This is useful on systems where the bus voltage level
translator is optional, as the translator can be auto-detected by the
driver and the feedback clock functionality can be disabled if it is
not present.
The translator test sets CMD high to avoid interfering with a card, and
then verifies whether signal set on CK is detected on CKIN. If the signal
is detected, translator is present, otherwise the CKIN feedback clock are
disabled.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Patrice Chotard <patrice.chotard@foss.st.com>
Cc: Patrick Delaunay <patrick.delaunay@foss.st.com>
Reviewed-by: Yann Gautier <yann.gautier@foss.st.com>
At present the MMC uclass assumes that the only child it can have is a
block device. Update this so we can add a bootmethod too.
Signed-off-by: Simon Glass <sjg@chromium.org>
Provide a way for sandbox MMC to present data from a backing file. This
allows a filesystem to be created on the host and easily served via an
emulated mmc device.
Signed-off-by: Simon Glass <sjg@chromium.org>
