We were previously not handling picotool at all, and
pico-sdk would download picotool itself, at build time.
This means that the source archive, if created, would
not contain picotool. While not strictly required, for
complete corresponding source, since it's a toolchain
and not the actual pico-serprog firmware, it is my policy
that releases must include full corresponding source code,
when it is feasible to do so.
I must say, I intensely dislike cmake, with such burning
passion; I am thoroughly displeased by how hacky this is,
but it works and now nothing is in my way for a Libreboot
20241206 rev8 release!
Signed-off-by: Leah Rowe <leah@libreboot.org>
See:
https://docs.python.org/3/library/sys.html#sys.version_info
The sys.version_info tuple is a more reliable way to
get the version. Our previous logic assumed that Python
would always output "Python versionnumber", but this may
not always be how it works. We've seen this for example
where Debian modifies some GNU toolchains to include Debian
something in the output.
Python has a standard method built in for outputting exact
the information we need. In my system, what I got was this:
(3, 11, 2, 'final', 0)
That output was from running this command:
python -c 'import sys; print(sys.version_info[:])'
This is much more robust, so use this instead.
Signed-off-by: Leah Rowe <leah@libreboot.org>
we already check the python version, and set a variable
for it, so that we can reliably use python3, even if
python in PATH doesn't correspond to python3. for
example if a system has python as python2 and python3
as python3
well, we use that when running deguard for example, but
various upstream projects that we use may need python,
and all of them use python3, not 2
so, re-use the python variable set up by lbmk, and
set it up in PATH accordingly. this now makes the note
about python3 obsolete, on docs/build.md in lbwww.git
Signed-off-by: Leah Rowe <leah@libreboot.org>
They may not actually always be binary blobs, at least not
software. I started referring to these as "vendor files" some
time ago, for this reason.
With this terminology, it applies properly to any sort of file
from the vendor. For example, it may be that in the future, we
start inserting the MFS section of an an Intel ME image, into
the Intel ME.
We already do that with deguard for example (set MFS config),
on MEv11 based setup. That is a vendor *file*, and though it
may still actually be a binary blob, it's not software, but
configuration.
The term "blob" normally means compiled software, in most people's
minds, but the term blob is technically accurate for any blob,
not just software; however, we have to keep people's perception
in mind.
Whereas, "vendor file" is also understood by most people to
include code supplied by the vendor.
We haven't done any releases yet with this ROM image file name
prefix, so it's perfectly OK to handle it now, without handling
the old one for backwards compatibility.
Signed-off-by: Leah Rowe <leah@libreboot.org>
Users running setmac on an X200 tarball for example, will
now see it being modified, if they didn't specify
setmac keep, so they might think vendor files are being
inserted, which they are not.
Therefore, a confirmation is provided at the end of the output.
Signed-off-by: Leah Rowe <leah@libreboot.org>
./mk inject libreboot-YYYYMMDD_board.tar.xz setmac restore
This does the same thing as a normal setmac command, except
that it does not alter the MAC address; it is also not the
same as "keep", which skips *writing* the GbE region in-ROM.
The *restore* argument writes the default, unmodified GbE file
kept by lbmk, unmodified because nvmutil is skipped when the
user specifies this argument.
This option is useful for debugging purposes, because it can
be used to verify whether anything else is being wrongly
modified by the script; the "nuke" command can be executed
afterward, and the hash file inspected versus release.
Signed-off-by: Leah Rowe <leah@libreboot.org>
MAC addresses are generic, inside Libreboot images where
an Intel GbE region is specified.
We commonly get users flashing multiple systems for their
own use, and sometimes they complain that they networking
broke, because they don't know that the MAC address is
identical on each machine.
This still doesn't work around the case where the same machine
is used, e.g. multiple T440p thinkpads, but if they have one
of each model, it can work nicely, because we do in fact
change it for various platforms.
This change will also reduce the number of people at conferences
in the future, where there are multiple Libreboot users, having
MAC address conflicts.
Changing the MAC address is a good practise, so we enforce good
practise. The user can still retain the old behaviour by
using this command:
./mk inject libreboot-YYYYMMDD_boardname.tar.xz setmac keep
The "keep" argument clears new_mac, which will then skip
changing the MAC address. They can also still set an arbitrary
MAC address as an argument for setmac, e.g.:
./mk inject libreboot-YYYYMMDD_boardname.tar.xz setmac 00🇩🇪ad:c0:ff:ee
This change will be covered in the documentation.
Signed-off-by: Leah Rowe <leah@libreboot.org>
if the user ran this on an x60 tarball, the no-gbe
warning seems confusing since that one has intel gbe,
but pre-ifd, so no gbe region in the flash; on pre-ifd
systems e.g. ich7 southbridge, the mac address was baked
into a separate gbe nvm on mask rom, inaccessible to users
Signed-off-by: Leah Rowe <leah@libreboot.org>
We already have code to handle this, but it's possible
that I might break it in the future, due to the complex
logic of this script.
So, I've implemented this catch-all check at the end of
the process. It still relies on the actual setting of
the variables, upon which this check is based, to be set
correctly.
This condition will most certainly never be met, unless
I break some other part of the code in the future. That
is precisely what this overly pedantic check is for.
Example scenarios:
I forget to set xchanged=y, on a new modification.
I set has_hashes erroneously.
The variables are re-used between runs, and not properly
reset; at present, a given run of ./mk inject only
operates on a single target, but this latter fact could
change in the future.
need_files is set erroneously; vendorfiles detected as
being required, when they aren't.
These are just a few examples. As such, this is a preventative
bug fix, because it's preventing a bug.
The main reason I want this i n here is because I need to ensure
that vendor files are properly deleted, for a given release.
If I accidentally includes ones that I'm not supposed to,
inside ROM images, that could be a big problem.
Signed-off-by: Leah Rowe <leah@libreboot.org>
where the nuke command is used, we need the files to be
there; if they're not, it will try to nuke them, which will result
in an error in most cases, but there may be some cases where that
isn't true, for instance if only the Intel ME is needed; it'll be
writing zeroes over zeroes.
we want to only allow technically correct behaviour, because
technically correct is the best kind of correct.
it is theoretically possible that a double-nuke might affect
certain behaviours unpredictably. for example, if vendor.sh
later integrates another tool that works whereby the same command
inserts or nukes depending on a certain condition, but with the
same command, and where that command would return zero in both
cases.
this is a preventative bug fix, because it fixes an issue that
does not yet actually occur in practise.
Signed-off-by: Leah Rowe <leah@libreboot.org>
the user must be well-informed as to the next step, which
this script directly influences
guide the user accordingly
Signed-off-by: Leah Rowe <leah@libreboot.org>
The message at the end that states a file was
not modified, is not currently printed when vendor
files are not needed, and setmac is not used.
This patch fixes that, so the user now sees a
confirmation of such change, or lack thereof.
Signed-off-by: Leah Rowe <leah@libreboot.org>
This is because the user may have specified setmac.
I tried without this change, on a fresh lbmk, setting
the MAC address on an X200 tarball, and it produced an
error that ifdtool was unavailable.
Signed-off-by: Leah Rowe <leah@libreboot.org>
Observe the following prior patch:
commit 818f3d630c
Author: Leah Rowe <leah@libreboot.org>
Date: Fri Jan 3 17:06:14 2025 +0000
vendor.sh: Don't error if vcfg is unset
Now:
This patch made vendor inject more robust, and speeds
up the processing of images where no vendor files are
needed, but it broke setmac on such tar archives.
This new patch works around it. For example, I was
able to run ./mk inject on an X200 tarball to change
the MAC address; no vendorfiles are inserted, because
it's not needed.
The further check for whether a board uses Intel GbE
still protects against accidental modification.
Signed-off-by: Leah Rowe <leah@libreboot.org>
It should return 1 instead, in readcfg(), because this
is not an error condition; vcfg not being set means
that the board doesn't use vendor files, which is
perfectly normal and should not yield an error.
This fixes a build error under certain conditions,
found during release-build testing.
This bug was exposed when I fixed double quoting issues
as per shellcheck tests.
Signed-off-by: Leah Rowe <leah@libreboot.org>
it should fix more build errors that might have appeared
in the aforementioned revision, mentioned in the previous
commit message
Signed-off-by: Leah Rowe <leah@libreboot.org>
the bug was actually caused by chkvars
add an escape for the quotes and bam. fixed.
without this, i got the following e.g.
For command: ./mk dependencies debian
Output:
./mk: 1: [: apt-get: unexpected operator
ERROR ./mk: pkg_add unset
Someone reported a similar issue with the Arch one,
which is also now fixed. This regression was caused
by the previous commit:
commit 0cf58c2273
Author: Leah Rowe <leah@libreboot.org>
Date: Thu Jan 2 23:52:45 2025 +0000
fix lbmk shellcheck errors
I forgot to escape the double quotes in an eval.
Signed-off-by: Leah Rowe <leah@libreboot.org>
This check is a good idea, but not viable here,
because the modules naturally aren't set in all
circumstances, so it just causes a build error.
Signed-off-by: Leah Rowe <leah@libreboot.org>
i also removed that printf, because the path it prints is
actually wrong sometimes; in the recent re-write of vendor.sh,
it prints the correct path instead
Signed-off-by: Leah Rowe <leah@libreboot.org>
There was also a condition in run_make_command that is now
an OR, where it was an AND, on script/trees, to fix the use
of mixed (and erroneous) OR/AND operators.
I'm planning a much more invasive audit than this. These are
light fixes, intended for Libreboot 20241206 rev8.
Signed-off-by: Leah Rowe <leah@libreboot.org>
Don't extract to bin/release/
Modify the tarball instead. Previously, the tarball would
not be modified, but a lot of users thought the tarball was
being modified and ignored bin/release/, where the injected
images were actually being saved to.
Don't copy the tarball either. Just modify it in-place.
Don't allow single-rom injection either; only allow the
tarball-based method.
The command syntax has changed, but:
./mk inject tarball.tar.xz
This is the same. What has changed is nuke, and MAC address
modification. Observe:
./mk inject tarball.tar.xz nuke
./mk inject tarball.tar.xz setmac
./mk inject tarball.tar.xz setmac ??:??:??:??:??:??
./mk inject tarball.tar.xz setmac 00:1f:16:??:22:aa
These are just a few examples. The MAC address syntax is
the same as used for nvmutil, which means you can set it
randomly. Also:
./mk inject tarball.tar.xz setmac
You can use the *setmac* command *repeatedly*, even if
you've already injected a given archive. It'll just
update the archive, but skip injecting other files
that were already injected.
If you use setmac without a MAC address, it will randomise
the MAC address. This is therefore very similar to the
command structure used in nvmutil.
The code for injection is generally more robust, with
stronger error checks. This design change was done, so
that the user doesn't accidentally brick their machine.
The non-injected images have a prefix in the file name
saying "DO_NOT_FLASH", and those non-injected images are
padded by 1 byte. That way, the user knows not to flash it
and if they try, flashprog will throw an error.
The prefix and padding is removed on injection. Old images
without the padding/prefix can still be injected, via
tarballs; this new code is backwards-compatible with tarballs
from older Libreboot releases.
A common thing I see sometimes is a user will say they have
a black screen or something, and I say: did you insert vendor
files? And they say yes. And they did. But they extracted and
flashed from the tarball, which wasn't injected, because
they didn't release about bin/release/
No amount of RTFM is justified. The previous design flaw
is a bug. We must always observe user safety first, no matter
what, so that has now been done.
Signed-off-by: Leah Rowe <leah@libreboot.org>
I believed that the compressed nature of refcode was the only
non-reproducible thing, but turns out you also need to run
rmodtool on the refcode to make the binary relocatable in
cbfs. This is based on my reading of the coreboot Makefile.
With this change, I can now provide release binaries for
the HP EliteBook 820 G2.
Signed-off-by: Leah Rowe <leah@libreboot.org>
./mk dependencies debian --reinstall
Add --reinstall and it'll do:
apt-get install --reinstall
This can be useful when updating from a stable release
to a testing release. The variable, "reinstall" can be
configured for other distros, but it's currently only
configured for Debian-based distros.
Also, it can be anything. For example, you could add -y;
however, a 4th argument will not be accepted. For example,
you cannot do:
./mk dependencies debian --reinstall -y
If you do this, it'll only see --reinstall; similarly, if
you did this command:
./mk dependencies debian -y --reinstall
then -y would be passed, but not --reinstall. This is an
intentional design decision, in case you accidentally pasted
or subshelled something that outputted something undesirable,
to prevent possible abuse.
Signed-off-by: Leah Rowe <leah@libreboot.org>
When doing ./mk release, the build system would create
symlinks inside xbmkpath/ relative to the current work tree,
which will differ from what's in PATH.
Since XBMK_CACHE is already set globally, from the main work
tree and the release-build work tree, that means we can know
reliably that PATH is always correct if we put xbmkpath/
inside XBMK_CACHE.
Signed-off-by: Leah Rowe <leah@libreboot.org>
Remove all symlinks each time, to ensure that no
stragglers are left behind, since they are being
re-generated each time anyway.
The code for determining version numbers has now
been unified under gnu_setver()
Signed-off-by: Leah Rowe <leah@libreboot.org>
We were checking the shorthand version number, but
the precise version numbers need to match.
Also: when we searched $PATH/gnat-$gccver, we assumed
that the full version would then match, without checking
it, so now it is checked precisely.
Signed-off-by: Leah Rowe <leah@libreboot.org>
When doing e.g. $@ we should use double quotes to prevent globbing.
Thanks go to XRevan86 for pointing this out.
Signed-off-by: Leah Rowe <leah@libreboot.org>
When I tested Debian Trixie, and Debian Sid, I saw that
GCC in PATH pointed to gcc-14, but gnat in path pointed
to GNAT-13, even if you manually install gnat-14.
GNAT 14 was marked experimental, but GCC 14 was marked
for use, in the apt repositories.
So this patch doesn't address the mismatch when doing e.g.
apt-get install gcc gnat
I will address the actual package dependency in a follow-up
patch, on the Debian dependencies config.
Signed-off-by: Leah Rowe <leah@libreboot.org>
Previously serprog_rp2040, but we now also support
the RP2530 boards.
Therefore, serprog_pico is a nice generic name. The
directory on release archives will now be serprog_pico
instead of serprog_rp2040; it will contain serprog images
for both RP2040 and RP2530 devices.
Signed-off-by: Leah Rowe <leah@libreboot.org>
rp2040 and rp2530 platforms can't share a cmake build directory. we
could just delete the build directory after every compilation, but that
would be really wasteful (every tool would need to be recomiled every
time. instead create new build directories as new plaforms are found
and symlink them to the point where the build directory used to be.
to find out which platform we're compiling for, we crudely parse the
board headers file.
there surely would be better ways to do this, but this hack works
with all the boards in pico-sdk 2.1.0.
Signed-off-by: Riku Viitanen <riku.viitanen@protonmail.com>
set this variable in the tmpclone function. otherwise,
certain submodules might always download every time,
when handling multiple projects.
Signed-off-by: Leah Rowe <leah@libreboot.org>
Leah Rowe