Recently, gru boards were migrated to use common stack addresses with
U-Boot commit 5e7cd8a11995 ("rockchip: Use common bss and stack
addresses on RK3399") and commit 49f8131e5594 ("rockchip: rk3399-gru:
Use TPL with common bss and stack addresses"). This is done with the
ROCKCHIP_COMMON_STACK_ADDR config.
With POSITION_INDEPENDENT, INIT_SP_RELATIVE defaults to enabled as well.
However, ROCKCHIP_COMMON_STACK_ADDR selects HAS_CUSTOM_SYS_INIT_SP_ADDR,
which depends on INIT_SP_RELATIVE being disabled. So this results in a
configuration warning:
WARNING: unmet direct dependencies detected for HAS_CUSTOM_SYS_INIT_SP_ADDR
Depends on [n]: ARM [=y] && ARCH_KIRKWOOD [=n] || ARC [=n] || ARM [=y] && !INIT_SP_RELATIVE [=y] || MIPS [=n] || PPC [=n] || RISCV [=n]
Selected by [y]:
- ROCKCHIP_COMMON_STACK_ADDR [=y] && ARM [=y] && ARCH_ROCKCHIP [=y] && SPL_SHARES_INIT_SP_ADDR [=y]
I'm not sure if adhering to the Rockchip values means we can't be
position-independent. Disabling INIT_SP_RELATIVE still appears to keep
my kevin board working, so let's do that for now.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Apply our preserved changes to the new U-Boot defconfigs. Upstream
rearranged memory layouts for Rockchip boards to a unified layout, which
got rid of CUSTOM_SYS_INIT_SP_ADDR and HAS_CUSTOM_SYS_INIT_SP_ADDR, and
will need a change to a related INIT_SP_RELATIVE later.
Normalize the positions of each line in the config by regenerating the
defconfig by `./mk -l u-boot` and then `./mk -s u-boot`, so that the
diff looks all green when we actually expand it to the full config.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Set the U-Boot revision to the commit hash for v2025.04, and rebase the
patches for the default U-Boot tree to accommodate for upstream changes:
- The SPL/TPL/VPL phases are being unified under the xPL name, so
there's a config rename.
- Some test macros were renamed, for the video-related patches.
- Add some missing hunks for video damage series.
- Upstream Makefile adds another argument to the binman call.
- The SWIG related patch is merged upstream, drop it.
I'm not sure if src/u-boot/* directories are regenerated on new builds,
so it may be necessary to remove them manually after applying this.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Run diffconfig from Linux to track our modifications to the old upstream
defconfigs, so we can apply them to the new ones. Restore the original
defconfigs to highlight our changes here, and upstream changes in the
next commit. Done manually, but something like:
do_diff() {
ours="$1"
theirs="$2"
tree="$3"
diffconfig \
src/u-boot/${tree}/configs/${theirs}_defconfig \
config/u-boot/${ours}/config/default \
>config/u-boot/${ours}/config/diffconfig
cp src/u-boot/${tree}/configs/${theirs}_defconfig \
config/u-boot/${ours}/config/default
}
do_diff amd64coreboot coreboot64 x86_64
do_diff i386coreboot coreboot x86
do_diff gru_bob chromebook_bob default
do_diff gru_kevin chromebook_kevin default
do_diff qemu_arm64_12mb qemu_arm64 default
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Run `./mk -s u-boot` to convert our configs into defconfigs, so we can
keep our changes to the old upstream defconfigs and re-apply them to the
new upstream defconfigs.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
it's not necessary, and was the cause of a recent issue,
which i mitigated, but why mitigate it?
prevent bugs. don't use eval unless absolutely necessary.
Signed-off-by: Leah Rowe <leah@libreboot.org>
see:
commit f0c629dcc6c3f2242a735567cdd6855fa108d1a5
Author: Leah Rowe <leah@libreboot.org>
Date: Sat Apr 12 13:51:49 2025 +0100
lib.sh: write version/versiondate to dotfiles
and this bug report:
https://codeberg.org/libreboot/lbmk/issues/284
The report indicates that the above commit broke bash,
when sh (on the user's system) is bash.
I know sometimes when using bash, I need to use the
back slash when dealing with dots, e.g. when grepping
something.
Also double quote references to dotfiles, e.g. when
directing the output of printf.
I never noticed the issue myself, since I use dash.
Signed-off-by: Leah Rowe <leah@libreboot.org>
If the mode string is empty, then it's a build command.
See commit:
commit b1ea4165754f04660d3c7282274c4b12355f88a0
Author: Leah Rowe <leah@libreboot.org>
Date: Wed Apr 23 03:54:08 2025 +0100
mk: remove mkhelp() and use x_() instead
This commit removed the following check:
If mode isn't set, run an mkhelper, otherwise don't.
Because this simplification removed that behaviour,
running e.g. "./mk -m coreboot x200_8mb" would result
in the mkcorebootbin function being executed, which is
normally putting the coreboot rom together.
Since it wasn't built in this case, an error is thrown.
This change therefore restores the previous behaviour,
fixing the bug.
First reported in this error report:
https://codeberg.org/libreboot/lbmk/issues/306
This commit fixes the issue.
Signed-off-by: Leah Rowe <leah@libreboot.org>
the command -v check has been removed, since this function
already calls git immediately, which would accomplish the
same thing since that causes an error if git isn't there.
Signed-off-by: Leah Rowe <leah@libreboot.org>
setvars is always invoked with eval, so make the error
condition a message for eval, to ensure that it is reliably
handled, in case of error condition.
Signed-off-by: Leah Rowe <leah@libreboot.org>
one function, for one task. skeleton functions for
performing multiple tasks. that is the basic coding
style guideline for lbmk.
Signed-off-by: Leah Rowe <leah@libreboot.org>
On initialisation of the child instance, ./mk is
executed, but an error from it won't reveal what
command was actually executed.
This change makes that the case, since x_ does
print the command that caused an error.
This is useful for debugging. However, we don't
want x_ to cause a real exit, because we still
need to handle the lock file from the parent
instance.
Therefore, the first child instance is executed
inside a subshell, and xbmk_rval is set if that
subshell returns non-zero.
Signed-off-by: Leah Rowe <leah@libreboot.org>
this is in prep for the next change, where non-init
functions will be moved to another file, again named
include/lib.sh
Signed-off-by: Leah Rowe <leah@libreboot.org>
a lot of init code was handled outside of any function. the
coding style used in the rest of the build system has now
been introduced, with xbmk_init being the main function.
Signed-off-by: Leah Rowe <leah@libreboot.org>
this was used alongside the xgcc linking, so that coreboot
trees could specify that another tree was to be downloaded.
since this variable will no longer be used, it should be
removed, to avoid dead code bloat.
Signed-off-by: Leah Rowe <leah@libreboot.org>
the "xtree" variable is used by projects such as u-boot,
to export a CROSS_COMPILE variable specifying prefix for
gnu compilers, and for building the named coreboot tree.
for example, xtree can be "default", which is then the
coreboot tree downloaded, for use of crossgcc.
however, it is also used to symlink identical versions
of crossgcc between coreboot trees. this latter feature
was only needed for fam15h boards which were previously
split between two mostly identical coreboot trees, that
were later merged into a single tree, and this feature
is therefore no longer used.
remove this dead code, to reduce bloat in the build system.
Signed-off-by: Leah Rowe <leah@libreboot.org>
In the previous revision, I make hardcoded use of
/usr/local/bin and /usr/bin as search locations, instead
of relying on PATH, when the user has a python venv, because
in those cases, we cannot rely on PATH so we use a python
command to detect the venv and then force use of the
normal system path for python.
However, there's no guarantee that the real Python will
indeed live at these locations. For example, some distros
like Nix or Guix will use many locations for different
versions of a given package, and it's for the birds as to
what given package version the user might be running.
Therefore, this patch retains that current hardcoded
assumption of /usr/local/bin and /usr/bin but *only* as
a fallback solution, instead checking realpath first.
The "realpath" command isn't technically POSIX standard,
but in practise it is available on GNU coreutils, Busybox,
and the various BSD userlands.
I could perhaps *import* a realpath utility, and use that,
but this should be fine.
Signed-off-by: Leah Rowe <leah@libreboot.org>
If the user has a virtual environment, the current logic
will cause lbmk to hang. A useful workaround is to force
use of the direct path to the system binary of python.
This works by detecting a virtual environment first, and
deferring to the old behaviour if no venv is found. If one
is found, then it will not rely on PATH, but instead only
search the standard locations /usr/local/bin and /usr/bin.
Signed-off-by: Leah Rowe <leah@libreboot.org>
alper made a fix to this file a few hours ago, but
forgot to update the copyright header
i'm doing it for alper, as a courtesy
Signed-off-by: Leah Rowe <leah@libreboot.org>
Properly set $pyver to "3" when we detect we can use python3. In the
following version checks, use the $python we detected instead of a
'python' from PATH because the latter might be a python2 while still
co-existing with a python3.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Alper Nebi Yasak