The configs were enabling SeaBIOS payload, but this is to be
handled by lbmk, not coreboot.
Further, they were enabling VGA ROM execution in coreboot, but
this should be handled by SeaBIOS.
This board should not have a GRUB payload enabled either; this
will be checked and fixed if necessary in the next commit.
Add U-Boot to the source release script's modules list so that it is
included in source release tarballs. Don't include the unused upstream
source and .git directories.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Copy the resources/scripts/build/clean/crossgcc script and adapt it to
run "make distclean" on U-Boot build trees. Some build artifacts persist
after the run, so also run "git clean -fdx" if we can.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
U-Boot build dependencies are listed on their online documentation [1],
but the listed Debian packages also include test-only dependencies.
While installing dependencies, install the packages necessary to build
U-Boot, except for the test-only ones I could identify.
[1] https://u-boot.readthedocs.io/en/latest/build/gcc.html
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Add a build for QEMU AArch64 virtual machine using U-Boot as payload.
Coreboot config is based on the following defconfig:
CONFIG_CBFS_SIZE=0x00c00000
CONFIG_BOARD_EMULATION_QEMU_AARCH64=y
CONFIG_CONSOLE_CBMEM_BUFFER_SIZE=0x20000
CONFIG_COREBOOT_ROMSIZE_KB_12288=y
CONFIG_UART_PCI_ADDR=0x0
The resulting ROM can be booted with a command line like:
qemu-system-aarch64 \
-machine virt,secure=on,virtualization=on \
-cpu cortex-a53 -m 1G \
-vga none -display none -serial stdio \
-bios bin/qemu_arm64_12mb/uboot_*.rom
However, this is little more than a proof of concept because U-Boot
upstream is missing coreboot integration on non-x86 boards, which could
have been useful for e.g. a framebuffer.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Add a U-Boot payload build for the QEMU AArch64 virtual machine. The
config is same as upstream "qemu-arm64" defconfig, but SYS_TEXT_BASE is
set to 0x50000000 so that it doesn't conflict with coreboot. QEMU
auto-generates and passes a device-tree file to U-Boot at runtime,
there's no compile-time canonical version, so there's no need to set
REMAKE_ELF or OF_EMBED.
It's not immediately obvious if QEMU-specific drivers are available to
support display output, but most coreboot integration is unavailable
(depends on x86) and entire video subsystem is disabled in the U-Boot
upstream defconfig.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Add a U-Boot build for the qemu_x86_12mb board. The config is a copy of
the upstream "coreboot" defconfig, but with OF_EMBED=y.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
U-Boot runtime configuration is done with a device-tree file, which is
built alongside the executable in the upstream build system, and must be
available to U-Boot at runtime.
This device-tree is normally not linked into the default "u-boot" ELF
file. So far we have been handling it by re-creating a "u-boot.elf" from
the raw binary parts by setting REMAKE_ELF, and using that as the
coreboot payload. Unfortunately, that fails to build for x86 boards,
more specificly the "coreboot" boards upstream.
It's also possible (but discouraged) to set OF_EMBED to embed the
device-tree file into the U-Boot itself, in which case we could use the
"u-boot" file as the payload on the "coreboot" boards. Add support for
using the "u-boot" file as the payload if "u-boot.elf" doesn't exist.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Add a series posted to upstream mailing lists that makes the GRUB
text-mode console faster by implementing video damage tracking [1].
Refresh the config files to include its new VIDEO_DAMAGE Kconfig.
Patch 7/7 upstream has a tiny conflict with "Improve UEFI experience"
series we already have, but it's only in the diff context. No changes
other than fixing that.
[1] https://lore.kernel.org/u-boot/20220609225921.62462-1-agraf@csgraf.de/
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Set revision to the commit hash of the v2022.10 release, and run "make
olddefconfig" for all boards to refresh the configs.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Merge all boards into a common "default" tree, currently for v2022.07.
This ends up applying the "Improve UEFI experience on DM_VIDEO" series
to everything, so refresh the configs for the new options.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
The roms_helper script skips building crossgcc-i386 if its target
directory exists. Skip it for other architectures as well.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Add the coreboot-built cross-architecture toolchains to the PATH so that
modules and payloads can use them. When building for a foreign-arch
board, also export CROSS_COMPILE pointing to the appropriate prefix.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
This re-applies commit a69855f7e4 ("Build 32-bit crossgcc for AArch64
as well") which was inexplicably reverted along with unrelated changes.
Mention in a comment that building crossgcc-arm is necessary for
AArch64.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
When overriding which payloads will be built with the -p command line
argument, the roms_helper script builds the Memtest86+ payload before
checking if it should be disabled. Move the build command after the
command line override.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
When overriding which payloads will be built with the -p command line
argument, the roms_helper script doesn't disable the U-Boot payload.
Disable it in this case.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
The U-Boot download script does its work from the repository root
instead going into the newly created dirs, unlike the coreboot
counterpart. It should run the board-specific extra.sh files with the
downloaded paths as their working directory. Do so by a subshell.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
The no-argument form of the U-Boot download script prepare trees for all
boards when run with no arguments, like the corresponding script for
coreboot. The usage text for this case was removed without any changes
to the corresponding code, assume it was by mistake and add it back.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Removing the git dirs was part of deblobbing, which Libreboot no longer
cares about. The variable that triggers it is no more. Remove the dead
code.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
*This condition will probably never be met, but it is theoretically
possible that the code could still fail at this point. Catch all errors,
and exit, ruthlessly.
The code was only checking whether all of the bytes were read,
but there are other errors that can be caught via errno.
Enforce strict errno handling, when generating random
numbers for command `setmac`.
this is a hangover from pre-osboot-merge libreboot. the idea
was to distribute fsdg uboot archives
lbmk has uboot support, and releases will simply
include uboot in the main src archive like with everything else
the --nuke option in ifdtool will be used instead, to nuke
the ME regions in specific rom sets (and cbfstool will be
used to delete mrc.bin files from rom sets)
the new method being implemented is heavier on disk io, but
simplifies lbmk, and disk io could still be optimised in
the following ways:
* when copying roms from boards with ME in them, use
ifdtool --nuke to get filename.rom.new, and *move* (not copy)
filename.rom.new to the new destination (for use with tar)
* possibly modify ifdtool to make efficient use of mmap for
disk i/o; it currently loads entire roms into an allocated
buffer in memory
previously, it was always initialised, but now it's only
initialised if '?' is used on a mac address character in
command `setmac`
this is done by simply moving mac address character
randomisation to a separate function
If one of the checksums was valid, but the other was not,
errno would be set to E_CANCELED, but then the buffer would
be modified anyway; this is acceptable behaviour, and errno
would later be reset writing the GBE file, which is done
only on the condition that the buffer was modified, but
it's also a good idea to reset it here just in case.
This is not a bugfix, and no behavioural changes will be
observed by the user, but this may *prevent* a bug in the
future, so let's pre-fix that bug now.
Leah Rowe
qeeg