The upstream git tree has disappeared, but a ZIP file is still hosted
somewhere at NXP. The content of that zip file has been verified to be
100% identical to the tarball we hosted on s.b.o.
As this is a zip file, we can't use the generic extract commands, and
must come up with our own. As such, it no longer makes sense to fix the
CRLF as a post-extract hook; this is moved to the extract command.
Add a hash file while at it.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
TF-A supports stack smashing protection (-fstack-protector-*).
However, that feature is currently silently disabled because
ENABLE_STACK_PROTECTOR is not set during build time.
As documented in the TF-A user guide, the flag ENABLE_STACK_PROTECTOR
is required to enable stack protection support. When enabled the symbols
for the stack protector (e.g. __stack_chk_guard) are built.
This needs to be done because TF-A does not link against an external
library that provides that symbols (e.g. libc).
So in case we see that BR2_SSP_* is enabled, let's enable the corresponding
ENABLE_STACK_PROTECTOR build flag for TF-A as documented in the TF-A user guide.
This patch also fixes a the following linker errors with older TF-A versions
if BR2_SSP_* is enabled (i.e. -fstack-protector-* is used as compiler flag)
and ENABLE_STACK_PROTECTOR is not set, which are caused by the missing
stack protector symbols:
[...]
params_setup.c:(.text.params_early_setup+0xc): undefined reference to `__stack_chk_guard'
aarch64-none-linux-gnu-ld: params_setup.c:(.text.params_early_setup+0x14): undefined reference to `__stack_chk_guard'
aarch64-none-linux-gnu-ld: params_setup.c:(.text.params_early_setup+0x104): undefined reference to `__stack_chk_guard'
aarch64-none-linux-gnu-ld: params_setup.c:(.text.params_early_setup+0x118): undefined reference to `__stack_chk_fail'
aarch64-none-linux-gnu-ld: ./build/px30/release/bl31/pmu.o: in function `rockchip_soc_sys_pwr_dm_suspend':
pmu.c:(.text.rockchip_soc_sys_pwr_dm_suspend+0xc): undefined reference to `__stack_chk_guard'
[...]
TF-A releases after Nov 2019, that include 7af195e29a4, will circumvent
these issue by explicitliy and silently disabling the stack protector
by appending '-fno-stack-protector' to the compiler flags in case
ENABLE_STACK_PROTECTOR is not set.
Tested on a Rockchip PX30 based system (TF-A v2.2 and upstream/master).
Signed-off-by: Christoph Müllner <christoph.muellner@theobroma-systems.com>
Reviewed-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
When using a custom git or mercurial repository for u-boot the error message
indicating a version had not been provided incorrectly stated that the URL was
missing. Update the error message to indicate that it's the version that's
missing.
Signed-off-by: Garret Kelly <garret.kelly@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Bump OP-TEE OS package version to OP-TEE release 3.11.0.
Remove patches since merged in OP-TEE OS 3.11.0
Signed-off-by: Etienne Carriere <etienne.carriere@linaro.org>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Add `.stm32` binary format that U-Boot generates for the trusted
configuration of STM32MP platforms.
Signed-off-by: Shlomi Vaknin <shlomi.39sd@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
For consistancy and dependencies between uboot and uboot-tools,
this patch migrates the script creation over in a similar way as
the env image creation.
Cc: Arnout Vandecappelle <arnout@mind.be>
Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Migrating the support for this feature to uboot-tools to gain the
ability to build env files when BR2_TARGET_UBOOT isn't selected.
When _ENVIMAGE_SOURCE is not set, we generate a default environment.
However, this default depends on the U-Boot configuration. Therefore,
this can only be done if uboot itself is built as well, and
host-uboot-tools needs to depend on uboot.
For the same reason, the commands for creating the environment have to
be adapted a little. Take this occasion to drastically simplify them.
Note: This patch creates a circular dependency with uboot until the
similar migration patch is merged for uboot scripts
Cc: Arnout Vandecappelle <arnout@mind.be>
Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
It's easier to locate given format when it's sorted.
Signed-off-by: Bartosz Bilas <b.bilas@grinn-global.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
The buildroot custom bareboxenv compile command misses the additional
include path 'scripts/include' to gain access to the local copy of the
kernel header files (which leads to compile error when using an older
toolchain).
This could be fixed by enhancing the custom bareboxenv compile command
(see [1]) or by using the barebox build system by simply enabling the
CONFIG_BAREBOXENV_TARGET option (available since April 2012, see [2])
instead (as suggested by Yann E. MORIN).
Fixes (with BR2_TARGET_BAREBOX_BAREBOXENV enabled):
build/barebox-2019.12.0/scripts/bareboxenv.c💯10: fatal error: linux/list.h: No such file or directory
[1] http://lists.busybox.net/pipermail/buildroot/2020-January/270942.html
[2] https://git.pengutronix.de/cgit/barebox/commit/?id=afb03d7a554a2911a3742e316f011319fcb416f1
Note: a user who would previously provide a barebox config file which
had CONFIG_BAREBOXENV_TARGET=y, but a Buildroot config file which did
not have BR2_TARGET_BAREBOX_BAREBOXENV=y, would have bareboxenv-target
built, but it would not be installed in the target. Now, and unset
BR2_TARGET_BAREBOX_BAREBOXENV will not even build it, but his is not a
regression: it was anyway previously not installed.
Reported-by: Frederick Gotham <cauldwell.thomas@gmail.com>
Signed-off-by: Peter Seiderer <ps.report@gmx.net>
[yann.morin.1998@free.fr:
- also explicitly disable it when not selected
- rewrap commit log
]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
This commit adds a patch that passes -fno-builtin to fix the build
with gcc 10. The patch has been submitted upstream.
There are no autobuilder failures, since it's a bootloader package,
and therefore not tested by the autobuilders. However, a build failure
was detected when building toolchains and a minimal ARMv7-M system at
https://gitlab.com/bootlin/toolchains-builder/-/jobs/729359529.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Recent versions of OP-TEE depend on Python 3. Currently, OP-TEE is
building with the Python interpreter provided by the user. This patch
includes an upstream patch that makes the interpreter configurable,
and makes use of this configuration with host-python3.
Signed-off-by: Dick Olsson <hi@senzilla.io>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
U-Boot must use $(BR2_MAKE) as it uses a Make feature from v4.0. We
already use $(BR2_MAKE) in the BUILD_CMDS, but the kconfig commands
still uses $(MAKE). Without this fix, building U-Boot with kconfig will
fail with the following cryptic error.
> Makefile:37: *** missing separator. Stop.
Signed-off-by: Brandon Maier <brandon.maier@rockwellcollins.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Enable this bootloader for cortex a7 based SoCs: support for the
sama7g5 SoC is now in upstream at91bootstrap3, and it is a Cortex-A7
based SoC.
Signed-off-by: Eugen Hristev <eugen.hristev@microchip.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Details: https://lists.gnu.org/archive/html/grub-devel/2020-07/msg00034.html
Fixes the following security issues:
* CVE-2020-10713
A flaw was found in grub2, prior to version 2.06. An attacker may
use the GRUB 2 flaw to hijack and tamper the GRUB verification
process. This flaw also allows the bypass of Secure Boot
protections. In order to load an untrusted or modified kernel, an
attacker would first need to establish access to the system such as
gaining physical access, obtain the ability to alter a pxe-boot
network, or have remote access to a networked system with root
access. With this access, an attacker could then craft a string to
cause a buffer overflow by injecting a malicious payload that leads
to arbitrary code execution within GRUB. The highest threat from
this vulnerability is to data confidentiality and integrity as well
as system availability.
* CVE-2020-14308
In grub2 versions before 2.06 the grub memory allocator doesn't
check for possible arithmetic overflows on the requested allocation
size. This leads the function to return invalid memory allocations
which can be further used to cause possible integrity,
confidentiality and availability impacts during the boot process.
* CVE-2020-14309
There's an issue with grub2 in all versions before 2.06 when
handling squashfs filesystems containing a symbolic link with name
length of UINT32 bytes in size. The name size leads to an
arithmetic overflow leading to a zero-size allocation further
causing a heap-based buffer overflow with attacker controlled data.
* CVE-2020-14310
An integer overflow in read_section_from_string may lead to a heap
based buffer overflow.
* CVE-2020-14311
An integer overflow in grub_ext2_read_link may lead to a heap-based
buffer overflow.
* CVE-2020-15706
GRUB2 contains a race condition in grub_script_function_create()
leading to a use-after-free vulnerability which can be triggered by
redefining a function whilst the same function is already
executing, leading to arbitrary code execution and secure boot
restriction bypass
* CVE-2020-15707
Integer overflows were discovered in the functions grub_cmd_initrd
and grub_initrd_init in the efilinux component of GRUB2, as shipped
in Debian, Red Hat, and Ubuntu (the functionality is not included
in GRUB2 upstream), leading to a heap-based buffer overflow. These
could be triggered by an extremely large number of arguments to the
initrd command on 32-bit architectures, or a crafted filesystem
with very large files on any architecture. An attacker could use
this to execute arbitrary code and bypass UEFI Secure Boot
restrictions. This issue affects GRUB2 version 2.04 and prior
versions.
Signed-off-by: Stefan Sørensen <stefan.sorensen@spectralink.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Bump OP-TEE OS package version to OP-TEE release 3.9.0.
Update patch on pydrypto/pycryptodome to match 3.9.0.
Add patch on CFG_OPTEE_REVISION_MINOR that was not updated in release
3.9.0 and fixed only few commits above.
Signed-off-by: Etienne Carriere <etienne.carriere@linaro.org>
Tested-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Before now, U-Boot SPL could only load the Platform Management Unit
(PMU) by patching the board-specific pm_cfg_obj.c file into the generic
PMU firmware, but that then requires generating a new PMU firmware for
every board configuration. To fix that, Luca Ceresoli added support to
U-Boot to load the pm_cfg_obj[1].
Like the PMU firmware, we need a way to pass the PMU cfg to U-Boot
during build. U-Boot only accepts the binary format of the cfg, so we
must convert the source file with the tool provided with U-Boot.
[1] https://lucaceresoli.net/zynqmp-uboot-spl-pmufw-cfg-load/
Signed-off-by: Brandon Maier <brandon.maier@rockwellcollins.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
The OpenSBI platform 'qemu/virt' has been removed in v0.8. Builds for
the QEMU 'virt' machine can use the 'generic' platform instead.
Tested with qemu_riscv32_virt_defconfig and
qemu_riscv64_virt_defconfig using Buildroot host-qemu 5.0.0.
Signed-off-by: Mark Corbin <mark@dibsco.co.uk>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
[yann.morin.1998@free.fr: two spaces in hash file]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Backport a patch from upstream to fix the build on certain versions of
gsc, notably:
Ubuntu 19.10 with gcc (Ubuntu 8.3.0-26ubuntu1~19.10) 8.3.0
Ubuntu 19.10 with gcc (Ubuntu 9.2.1-9ubuntu2) 9.2.1 20191008
The upstream patch is simply a change in the gentpl.py script, which is
used to generate parts of the automake machinery, so if we just backport
the upstream patch, we need to call the script to regenerate those files.
However, the modified script is a python script, so we would need to add
a dependency on host-python (2 or 3), which is not so nice.
Furthermore, calling the script is not enough: it needs a specific set
of optionss for each file it is to generate. That set of options is not
static; it is constructed in the convoluted autogen.sh. Calling
autogen.sh is usally not so good an idea in the Buildroot context, and
indeed this fails becasue it calls to autoreconf, but without our
carefuly crafted options and environment variables.
There was a little light in the tunnel, in that autogen.sh can be told
not to run autoreconf, by setting the environemnt variable
FROM_BOOTSTRAP to an non-=empty string, but this is fraught with various
other side-effects, as in that cause, autogen.sh expects to be valled by
an upper sciopt, bootstrap, which is not provided in the tarball
distribution...
So, between all those issues, autogen, bootstrap, and a host-python (2
or 3) dependency, we choose another route: path the script *and* the one
generated file affected by the change. Since that patched file is a .am
file, we also patch the corresponding .in file
However, we're faced with another issue: the other generated file is
now older than the script, so the automake machinery will now want to
re-run autoconf et al during the build step, which is still not a good
idea for us. So we touch the other generated file so it is mopre recent
than the script.
This is still not sufficient, because the patched file also has a
dependency on the generated file, so we need to touch as well.
Fixes:
- https://bugs.buildroot.org/show_bug.cgi?id=12946
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
[yann.morin.1998@free.fr:
- keep the hunk about patching gentpl.py
- make it a git-formatted patch
- add the touch
- drastically expand the commit log
]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
The HTTPS URL seems to be more reliable and quicker for download than
FTP. FTP may also be a blocked protocol on some sites and in CI/CD
setups.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
A few conflicts had to be resolved:
- Version number and hash for mesa3d-headers/mesa3d
- Patches added in qemu, and the qemu version number
- The gnuconfig README.buildroot
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
The commit [1] "licensing info is only valid for v1.4" fixed the legal-info
issues when a custom ATF tarball or a version from git is used.
But we need to ignore licencing for a used defined official ATF version.
Althougt the ATF version are licensed under BSD-3-Clause, the license
file can be updated between version (for example between v1.4 and v2.0).
Ignore the licencing check if the user provide a custom official version.
[1] d1a61703f7
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Cc: Yann E. MORIN <yann.morin.1998@free.fr>
[yann.morin.1998@free.fr: use positive logic with the _LATEST option]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
>From [1] included in optee-os release 3.7.0:
"PyCryptodome is a fork of PyCrypto, which is not maintained any more
(the last release dates back to 2013 [2]). It exposes almost the same
API, but there are a few incompatibilities [3]."
pem_to_pub_c.py/sign.py scripts still use pycrypto that is replaced
by pycryptodomex. Add a patch to use pycryptodomex but don't use
upstream commit since it also switches from the algorithm
TEE_ALG_RSASSA_PKCS1_V1_5_SHA256 to TEE_ALG_RSASSA_PKCS1_PSS_MGF1_SHA256
when replacing pycrypto to pycryptodomex [4].
[1] 90ad245043
[2] https://pypi.org/project/pycrypto/#history
[3] https://pycryptodome.readthedocs.io/en/latest/src/vs_pycrypto.html
[4] ababd72d2f
Fixes:
https://gitlab.com/buildroot.org/buildroot/-/jobs/526035730
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Cc: James Hilliard <james.hilliard1@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
U-Boot versions newer than 2020.01 use Python 3.x instead of Python
2.x in various scripts.
We already had the BR2_TARGET_UBOOT_NEEDS_PYLIBFDT and
BR2_TARGET_UBOOT_NEEDS_PYELFTOOLS options, but depending on the U-Boot
version, we now need to indicate if Python 2.x or Python 3.x should be
used.
In addition, it turns out that some U-Boot configurations need a
Python interpreter, without needing pylibfdt or pyelftools. Some of
our defconfigs were abusing the BR2_TARGET_UBOOT_NEEDS_PYLIBFDT option
to make sure a Python interpreter was built.
To solve this issue, we introduce a choice, that let the users specify
what, if any, host python version is needed. The default is 'no', to
preserve the previous behaviour, unless any of the pylibfdt or the
pyelftools options is enabled, in which case we hide the 'no' option,
and use python 2 by default. This dfault is guaranteed by the order of
options in the choice.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
[yann.morin.1998@free.fr:
- explicitly make the choice a bool
- make BR2_TARGET_UBOOT_NEEDS_PYTHON a blind option
- introduce the 'no' option in the choice
- reword the commit log accordingly
]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
The following defconfig:
BR2_x86_i686=y
BR2_TOOLCHAIN_EXTERNAL=y
BR2_TOOLCHAIN_EXTERNAL_DOWNLOAD=y
BR2_TOOLCHAIN_EXTERNAL_URL="http://toolchains.bootlin.com/downloads/releases/toolchains/x86-i686/tarballs/x86-i686--glibc--bleeding-edge-2018.11-1.tar.bz2"
BR2_TOOLCHAIN_EXTERNAL_GCC_8=y
BR2_TOOLCHAIN_EXTERNAL_HEADERS_4_14=y
BR2_TOOLCHAIN_EXTERNAL_CUSTOM_GLIBC=y
BR2_TOOLCHAIN_EXTERNAL_CXX=y
BR2_INIT_NONE=y
BR2_TARGET_SYSLINUX=y
BR2_TARGET_SYSLINUX_EFI=y
fails to build due to missing setjmp/longjmp definitions, which is a
consequence of a change introduced between gnu-efi 3.0.9 and 3.0.10.
This build failure is fixed by adding another syslinux paytch, which
has been submitted upstream.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Some ATF configurations, require a pre-built bare metal toolchain to
build some platforms which host cortex-m series core, for instance
rockchip rk3399 has a cortex-m0 core. Without a pre-built bare metal
toolchain, the build fails:
make[3]: arm-none-eabi-gcc: Command not found
To solve this, this commit implements a
BR2_TARGET_ARM_TRUSTED_FIRMWARE_NEEDS_ARM32_TOOLCHAIN
option. Platforms which have such requirement should enable this
config option.
Signed-off-by: Suniel Mahesh <sunil@amarulasolutions.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
When barebox, and thus barebox-aux, are downloaded from a git tree, then
barebox-aux download fails because a hash check is attempted on the
downloaded archive:
Could not fetch special ref 'v2020.03.0'; assuming it is not special.
ERROR: No hash found for barebox-aux-v2020.03.0.tar.gz
This is because we only exclude from the check the archive of the bare
barebox:
BR_NO_CHECK_HASH_FOR += $(BAREBOX_SOURCE)
However, the default name of an archive is based on the package name,
which for barebox-aux is not 'barebox'.
Since barebox-aux really uses the exact same source as the bare barebox,
it should also share the archive name.
This has two direct consequences and advantages:
- the hash check is completely avoided for the barebox-aux archive;
- the barebox-aux archive is not downloaded as it is already
downloaded for barebox.
Reported-by: Yegor Yefremov <yegorslists@googlemail.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Tested-by: Yegor Yefremov <yegorslists@googlemail.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
barebox and barebox-aux are really the same package, from the same URL
and the same version. They deserve being stored in the same directory.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Yegor Yefremov <yegorslists@googlemail.com>
Tested-by: Yegor Yefremov <yegorslists@googlemail.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Some ATF configurations, such as the ones for the STM32MP1 processor
family, require DTC during the build as Device Tree files are
used. Without dtc, the build fails:
/bin/sh: 1: dtc: not found
dtc version too old (), you need at least version 1.4.4
plat/st/stm32mp1/platform.mk:239: recipe for target 'check_dtc_version' failed
make[1]: *** [check_dtc_version] Error 1
To solve this, this commit implements a
BR2_TARGET_ARM_TRUSTED_FIRMWARE_NEEDS_DTC option, in a way that mimics
the BR2_TARGET_UBOOT_NEEDS_DTC option we already have for the U-Boot
package.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Tested with qemu_riscv32_virt_defconfig and
qemu_riscv64_virt_defconfig using Buildroot host-qemu 4.2.0.
Signed-off-by: Mark Corbin <mark@dibsco.co.uk>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Add missing qstrip wrapping to the new
BR2_TARGET_ARM_TRUSTED_FIRMWARE_ADDITIONAL_TARGETS option.
Signed-off-by: Francois Gervais <fgervais@distech-controls.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
We can only know the details of the license files for known versions. For
custom, older or newer versions, the license files may change, or may be
moved around.
So, do for optee-os as was done for other packages in the recent past,
and only define the list of license files for the latest version.
Reported-by: Peter Korsgaard <peter@korsgaard.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Markus Mayer <mmayer@broadcom.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
We can only know the details of the license files for known versions. For
custom, older or newer versions, the license files may change, or may be
moved around.
So, do for at91bootstrap3 as was done for other packages in the recent
past, and only define the list of license files for the latest version.
Reported-by: Peter Korsgaard <peter@korsgaard.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Markus Mayer <mmayer@broadcom.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
We can only know the details of the license files for known versions. For
custom, older or newer versions, the license files may change, or may be
moved around.
So, do for Barebox as was done for ATF, linux, and linux-headers, and
only define the list of license files for the latest version.
Add the hash for that license file, and align hashes to the new spacing
convention.
Reported-by: Peter Korsgaard <peter@korsgaard.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Markus Mayer <mmayer@broadcom.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
We can only know the details of the license files for known versions. For
custom, older or newer versions, the license files may change, or may be
moved around.
So, do for U-Boot as was done for ATF, linux, and linux-headers, and only
define the list of license files for the latest version.
Reported-by: Peter Korsgaard <peter@korsgaard.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Markus Mayer <mmayer@broadcom.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
