Linux on ARM supports FDPIC binaries intended for use on no-MMU
systems. This patch enables support for building a toolchain that
produces FDPIC binaries, only for ARMv7-M platforms, for which FDPIC
binaries are relevant.
The target name for a FDPIC toolchain must be
arm-<vendor>-uclinuxfdpiceabi, which doesn't follow the standard
format and requires a special case.
Signed-off-by: Ben Wolsieffer <Ben.Wolsieffer@hefring.com>
Signed-off-by: Waldemar Brodkorb <wbx@openadk.org>
Signed-off-by: Dario Binacchi <dario.binacchi@amarulasolutions.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Some buildsystems (or their use of it by packages) will cause flags in
LDFLAGS to be re-orderded, or even dropped, causing some
incomprehensible mayhem... For example, gpsd [0] has this in its
SConscript [1](typoes not mines, for once):
591 # scons uses gcc, or clang, to link. Thus LDFLAGS does not serve its
592 # traditional function of providing arguments to ln. LDFLAGS set in the
593 # environment before running scons get moved into CCFLAGS by scons.
594 # LDFLAGS set while running scons get ignored.
[--SNIP--]
611 for i in ["ARFLAGS",
612 "CCFLAGS",
613 "CFLAGS",
614 "CPPFLAGS",
615 "CXXFLAGS",
616 "LDFLAGS",
617 "LINKFLAGS",
618 "SHLINKFLAGS",
619 ]:
620 if i in os.environ:
621 # MergeFlags() puts the options where scons wants them, not
622 # where you asked them to go.
623 env.MergeFlags(Split(os.getenv(i)))
So, when LDFLAGS (our TARGET_LDFLAGS) contains "-z text" (without the
quotes), that gets turned into a command line like (line-splitted for
readability):
[...]/buildroot/output/host/bin/aarch64_be-buildroot-linux-musl-gcc \
-o gpsd-3.25/drivers/driver_rtcm2.os \
-c \
--sysroot=[...]/buildroot/output/host/aarch64_be-buildroot-linux-musl/sysroot \
-O3 -g0 \
-z \
-fPIC -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 \
gpsd-3.25/drivers/driver_rtcm2.c
Notice how there is a lone "-z" without any following keyword.
This then causes a build failure that looks totally unrelated [2]:
In file included from gpsd-3.25/drivers/../include/gpsd.h:36,
from gpsd-3.25/drivers/driver_rtcm2.c:65:
gpsd-3.25/drivers/../include/os_compat.h:40:8: error: redefinition of ‘struct timespec’
40 | struct timespec {
| ^~~~~~~~
In file included from [...]/buildroot/output/host/aarch64_be-buildroot-linux-musl/sysroot/usr/include/sys/select.h:16,
from gpsd-3.25/drivers/../include/gpsd.h:31:
[...]/buildroot/output/host/aarch64_be-buildroot-linux-musl/sysroot/usr/include/bits/alltypes.h:237:8: note: originally defined here
237 | struct timespec { time_t tv_sec; int :8*(sizeof(time_t)-sizeof(long))*(__BYTE_ORDER==4321); long tv_nsec; int :8*(sizeof(time_t)-sizeof(long))*(__BYTE_ORDER!=4321); };
| ^~~~~~~~
gpsd-3.25/drivers/../include/os_compat.h:48:5: error: conflicting types for ‘clock_gettime’; have ‘int(clockid_t, struct timespec *)’ {aka ‘int(int, struct timespec *)’}
48 | int clock_gettime(clockid_t, struct timespec *);
| ^~~~~~~~~~~~~
In file included from gpsd-3.25/drivers/../include/gpsd.h:33:
[...]/buildroot/output/host/aarch64_be-buildroot-linux-musl/sysroot/usr/include/time.h:104:5: note: previous declaration of ‘clock_gettime’ with type ‘int(clockid_t, struct timespec *)’ {aka ‘int(int, struct timespec *)’}
104 | int clock_gettime (clockid_t, struct timespec *);
| ^~~~~~~~~~~~~
scons: *** [gpsd-3.25/drivers/driver_rtcm2.os] Error 1
scons: building terminated because of errors.
make[1]: *** [package/pkg-generic.mk:289: [...]/buildroot/output/build/gpsd-3.25/.stamp_built] Error 2
make: *** [Makefile:83: _all] Error 2
Although undocumented, neither in gcc not ld (clang unchecked by lack of
a clang toolchain here, and by lack of clang knowledge), -z accepts the
keyword to be snatch-glued onto it, like -zkeyword, rather than be
spearated with a space. So, use that to pass -ztext.
Fixes:
http://autobuild.buildroot.org/results/c03/c039989947b960ac6af17c87090366abc26dcb6d/http://autobuild.buildroot.net/results/bc35d3e7b0e8c59c776652070650af3c749250ee/
[0] Our other scons-based package, mongodb, does not build for another
reason that probably hides the same issue as seen with gpsd.
[1] As explained in gpsd's SConscript (se above), Scons does play tricks
with variables:
https://scons.org/doc/production/HTML/scons-man.html#f-MergeFlagshttps://scons.org/doc/production/HTML/scons-man.html#f-ParseFlags
Quoting:
Flag values are translated according to the prefix found, and added
to the following construction variables:
[...]
-Wl, LINKFLAGS
[...]
- CCFLAGS
[...]
Any other strings not associated with options are assumed to be the
names of libraries and added to the $LIBS construction variable.
So in our case, it finds that -z is an unknown option that matches the
'-' prefix, so it is added to CFLAGS, while 'text' is a string on its
own, so added to LIBS, and thus it would try to link with -ltext
(supposedly, because we do not even go that far). Funnily enough, we can
se that "-Wl," is a known option prefix, that is added to LINKFLAGS (to
properly be used with gcc or clang, not ld).
As a consequence, gpsd's buildsystem does drop -ztext from the link
flags, and only passes it to the compile flags, which brings us back to
before we banned textrels in a1a2f498d7 (package/Makefile.in: ban
textrels on musl toolchains). Fixing gpsd is a task for another,
separate patch...
[2] I spent quite some time to look at recent, time-related changes in
Buildroot, especially due to the infamous time64_t issues... Alas, that
was not related, and only a git-bisect pinpointed the actual issue. Poor
polar bears...
Signed-off-by: Yann E. MORIN <yann.morin@orange.com>
Cc: J. Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-by: J. Neuschäfer <j.neuschaefer@gmx.net>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
After the recent addition to -Wl,-z,text to TARGET_LDFLAGS in case of
musl-libc and dynamic linking, it was found that some packages pass
TARGET_LDFLAGS directly to ld, but the -Wl syntax only works with
compiler drivers (gcc/clang). This commit changes the flag to -z text,
which the gcc and clang also understand and pass to the linker.
Reported-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: J. Neuschäfer <j.neuschaefer@gmx.net>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
musl-libc doesn't support TEXTRELs[1] and programs with TEXTRELs will
crash on start-up under musl.
This patch forbids the use of TEXTRELs on musl toolchains with dynamic
linking.
To verify this patch:
- Delete package/micropython/0001-py-nlrthumb-Make-non-Thumb2-long-jump-workaround-opt.patch
- Build micropython (before v1.23) with a musl toolchain and
BR2_SHARED_LIBS. The build should abort while linking micropython.
[1]: https://www.openwall.com/lists/musl/2020/09/25/4
Signed-off-by: J. Neuschäfer <j.neuschaefer@gmx.net>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Post-build, post-image, and other build scripts may run some commands in
parallel, for example to parallelize xargs, Makefiles, etc. Export
PARALLEL_JOBS to these scripts so they can enforce the same job limits
that other Buildroot packages use.
Signed-off-by: Brandon Maier <brandon.maier@collins.com>
Signed-off-by: Arnout Vandecappelle <arnout@mind.be>
Note that here, it is in fact not really relevant. We only extract a
tarball, and we don't use any "modern" or GNU-only options like
--strip-components. However, for consistency it's better to use the same
tar everywhere.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: "Maier, Brandon L Collins" <Brandon.Maier@collins.com>
Reviewed-by: brandon.maier@collins.com
[Arnout: quote TAR="..."]
Signed-off-by: Arnout Vandecappelle <arnout@mind.be>
A number of packages try to detect if they are running in a git repo and run
git describe at build time instead of using the hard coded version number if
it succeed, leading to odd version numbers as they end up picking up the
Buildroot git version if building inside a Buildroot git checkout, E.G.:
rauc --version
rauc 2023.11-562-g9c954953b4+
This is because rauc builds with meson and uses vcs_tag:
https://github.com/rauc/rauc/blob/v1.11/meson.build#L168-L171https://mesonbuild.com/Reference-manual_functions.html#vcs_tag
Another example is micropython, where we already work around it by passing
GIT_DIR=.
In the context of Buildroot the packages are never built in their own git
checkout, so pass GIT_DIR=. to ensure git doesn't walk back up the
directory tree and finds the Buildroot git repo, which fixes the rauc (and
similar) issues.
>>> rauc 1.11 Building
..
ninja: Entering directory `/home/peko/source/buildroot/output-rauc/build/rauc-1.11//build'
[1/29] Generating version.h with a custom command
fatal: not a git repository: '.'
cat output-rauc/build/rauc-1.11/build/version.h
#define PACKAGE_STRING "rauc 1.11"
#define PACKAGE_VERSION "1.11"
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Y2038 is now almost only 15 years away, and embedded systems built
today are potentially going to still be operational in 15 years, and
even though they are supposed to receive updates by then, we all know
how things go, and potentially some of these embedded systems will not
receive any update.
In 2038, the signed 32-bit representation of time_t used on 32-bit
architectures will overflow, causing all time-related functions to go
back in time in a surprising way.
The Linux kernel has already been modified to support a 64-bit
representation of time_t on 32-bit architectures, but from a C library
perspective, the situation varies:
- glibc uses this 64-bit time_t representation on 32-bit systems
since glibc 2.34, but only if -D_TIME_BITS=64 is
specified. Therefore, this commit adds an option to add this flag
globally to the build, when glibc is the C library and the
architecture is not 64-bit.
- musl uses unconditionally a 64-bit time_t representation on 32-bit
systems since musl 1.2.0. So there is nothing to do here since
Buildroot has been using a musl >= 1.2.0, used since Buildroot
2020.05. No Buildroot option is needed here.
- uClibc-ng does not support a 64-bit time_t representation on 32-bit
systems, so systems using uClibc-ng will not be Y2038 compliant, at
least for now. No Buildroot option is needed here.
It should be noted that being Y2038-compliant will only work if all
application/library code is correct. For example if an
application/library stores a timestamp in an "int" instead of using
the proper time_t type, then the mechanisms described above will not
fix this, and the application/library will continue to be broken in
terms of Y2038 support.
Possible discussions points about this patch:
- Should we have an option at all, or should we unconditionally pass
-D_TIME_BITS=64, like we have been doing for _FILE_OFFSET_BITS=64
for quite some time. The reasoning for having an option is that
the mechanism is itself opt-in in glibc, and generally relatively
new, so it seemed logical for now to make it optional as well in
Buildroot.
- Should we show something (a Config.in comment?) in the musl and
uClibc-ng case to let the user know that the code is Y2038
compliant (musl) or not Y2038 compliant (uClibc-ng). Or should this
discussion be part of the Buildroot documentation?
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Make 4.4 introduces a shuffle mode which randomizes prerequisites
in order to better flush out issues with parallel builds. On the other
hand, we use MAKE1 to build packages that are known to be broken with
parallel build. For these, passing the shuffle option would be
counter-productive and lead to spurious build failures.
The --shuffle=none option exists to turn off shuffling again. We can't
add this option unconditionally, however, because Make < 4.4 doesn't
know it. Therefore, conditionally pass --shuffle=none only if there is a
shuffle option in MAKEFLAGS.
Signed-off-by: James Hilliard <james.hilliard1@gmail.com>
Signed-off-by: Arnout Vandecappelle <arnout@mind.be>
Since it was introduced in 5a6087d62e (toolchain: add powerpc SPE ABI
support), the CLASSIC vs. SPE choice for the ABI was never really a
choice: CPU without SPE could only use the CLASSIC ABI, while CPUs with
SPE could only use the SPE ABI.
Commit b4c824562b (powerpc: add BR2_POWERPC_CPU_HAS_SPE to replace
adhoc deps/checks) added a blind option that CPUs with SPE would select
rather than duplicate the ad-hoc dependencies in both CLASSIC and SPE
ABI options. Since then, it was even more obvious that the ABI choice
was really not a choice, as the two options have mutually exclusive
conditions.
Drop the useless choice, and directly use the blind option as selected
by the specific CPUs.
We don't need legacy handling, because the situation fixes itself.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Joel Stanley <joel@jms.id.au>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
When we introduced support for the paranoid check of unsafe libraries
and headers path with commit 4ac8f78d37 (Add option for paranoid
unsafe path checking) back in 2014, we made it optional, as we expected
that would break quite a few packages.
Now, almost 8 years later, we only have three packages that explicitly
reference the option (dillo, gnuradio, and libtalloc), either in a patch
or in their .mk.
The option has been enabled by default since 2016, with 61c8854cef
(toolchain: enable paranoid unsafe path check by default), and that has
not triggered many build failures in a while.
The minimal defconfig used by test-pkg has also had it enabled as of
b6c98b3549 (minimal.config: add BR2_COMPILER_PARANOID_UNSAFE_PATH=y)
in 2017.
It is time to make that globally unconditional now.
There is still a remnant, in our binutils patches. As our toolchain may
get used outside of Buildroot, people may got the expectation that path
poisoning is only a warning, so we keep the current behaviour.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Romain Naour <romain.naour@gmail.com>
Acked-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Commit fda53f0791 ("package/Makefile.in:
add detection for the lack of C library") added an $(error ...)
message when no C library is available for the currently selected
architecture.
However, this error message pops up not just when building, so for
example, the command:
make BR2_HAVE_DOT_CONFIG=y VARS=%_LICENSE printvars
no longer works (this command is used by the pkg-stats script).
We restore a functional behavior by doing the check only when
BR_BUILDING=y.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
We recently had several cases of architecture configurations for which
no C library was available, leading to a build failure during the gcc
build. In order to more easily detect those bogus configurations,
let's bail out very early by detecting the lack of C library
selection.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
[yann.morin.1998@free.fr:
- move as final else clause in existing conditional block
]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
This reverts commit 6e33e59080.
This unfortunately breaks a number of packages, as glibc errors out if 64bit
time_t is used without 64bit file offsets, and some packages undefine
_FILE_OFFSET_BITS leading to build breakage:
# if ! defined (_FILE_OFFSET_BITS) || _FILE_OFFSET_BITS != 64
# error "_TIME_BITS=64 is allowed only with _FILE_OFFSET_BITS=64"
..
So revert it for 2022.02.
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
This will use gcc-ar, gcc-nm and gcc-ranlib instead of the normal
binutils tools. The difference is that with the wrappers, gcc plugins
will be automatically picked up.
gcc 4.7 introduced these wrappers, to detect the prefix and keep gcc
specifics out of Makefile.in, a new variable BR2_TOOLCHAIN_BUTILS_PREFIX
will be used to carry the prefix on supported versions.
Note that binutils added some automatic loading with the 'bfd-plugins'
directory (somewhere around 2.28), but the first implementation had
issues, and generally depends on correctly setup symlinks (often broken,
may point to some other gcc's library). The wrappers always work
painless.
The original motivation (now ~2 years in use) was to add "-flto
-ffat-lto-objects" to both BR2_TARGET_OPTIMIZATION and
BR2_TARGET_LDFLAGS, and have target binaries lto optimized.
Not all packages will compile with this option, further work could
white/blacklist packages (adding -fno-lto to the options).
Signed-off-by: Norbert Lange <nolange79@gmail.com>
[yann.morin.1998@free.fr: don't introduce a Config.in blind option]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
In order to be package agnostic, the install phase is now using cargo
instead of install. TARGET_CONFIGURE_OPTS is now also set when running
cargo in order to support cross compiling C code within cargo.
This commit also adds support/download/cargo-post-process to perform
the vendoring on Cargo packages.
The <pkg>_LICENSE variable of cargo packages is expanded with ",
vendored dependencies licenses probably not listed" as currently for
all packages, the licenses of the vendored dependencies are not taken
into account.
Signed-off-by: Patrick Havelange <patrick.havelange@essensium.com>
[Thomas: add support for host-cargo-package and vendoring]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Using *-uclinux-* seems like an only partially followed convention.
And at least RISC-V 64-bits gcc does not know about uclinux tuples.
So switch back to the normal "linux" one for now.
Signed-off-by: Christoph Hellwig <hch@lst.de>
[Damien]
* Make the change conditional on BR2_RISCV_64 being "y".
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Enable selecting elf2flt for RISC-V 64-bits no MMU builds and add
support for it in the form of an additional patch (upstream pull request
https://github.com/uclinux-dev/elf2flt/pull/22).
Also modify the package Makefile.in file to add the -fPIC option to
the target CFLAGS for RISC-V 64-bits no MMU builds.
This patch is based on initial work by Christoph Hellwig <hch@lst.de>.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
When BR2_BINFMT_FLAT_ONE is selected, elf2flt must be invoked with the
"-r" option to ensure that a single contiguous binary image is created,
regardless of the architecture default image format implemented by
elf2flt.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Enabling -DNEBUG, although correct on the paper, causes a lot of
packages to fail to build because they explicitly require not building
with NDEBUG; they use assert() to check actual runtime errors and expect
it to not be elidded away (sometimes with side effects in the arguments
passed to assert().
This reverts commit 5a8c50fe05, as
discussed on the list:
http://lists.busybox.net/pipermail/buildroot/2021-July/313646.html
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
The 'assert' statement in glibc honors the 'NDEBUG' preprocessor macro: if
it is set, then the assert statement is compiled away.
Define this 'NDEBUG' macro when BR2_ENABLE_RUNTIME_DEBUG is disabled (the
default case).
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Reviewed-by: Yann E. MORIN <yann.morin.1998@free.fr>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
If BR2_ENABLE_DEBUG is not set, Buildroot did not pass any flag
to control debug level. This means that the build system of the package
itself would control it.
Instead, provide an explicit '-g0' (no debugging symbols) to get consistent
behavior across packages.
Suggested-by: Arnout Vandecappelle <arnout@mind.be>
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Sometimes, post-build or post-image scripts need to reinvoke
Buildroot's make, for example to execute "make printvars".
However, so far post-build/image/fakeroot can't trivially run printvars
in a way that worked for both in-tree and out-of-tree builds. Indeed:
* "make printvars" would work for in-tree builds, but not out of tree
builds
* "make -C ${O} printvars" would work for out-of-tree builds, but not
in-tree builds
* "make -C ${BR2_CONFIG%/*} printvars" works in both cases, but it is
a bit cryptic, and two maintainers did not even immediately think of
it
In order to solve this, this commit exposes $(CONFIG_DIR) to
post-build/image/fakeroot scripts, through the EXTRA_ENV variable.
The documentation is updated accordingly.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
[yann.morin.1998@free.fr:
- reference BR2_CONFIG as an exemple
- slightly reword the commit log accordingly
- move the doc for CONFIG_DIR next to that of BR2_CONFIG
]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
This provides generic functions for Qt5 qmake based packages. It will
make it possible to remove lots of redefinition of
QT5_xxx_{CONFIGURE|BUILD|INSTALL_STAGING}_CMDS. Additionally it
provides a generic target install method which will make most of the
package specific commands obsolete.
This is done by re-running the install step of the qmake generated
Makefile with the package build directory prepended (to the
staging/host path). Even though this does create lengthy pathes it
allows for easy separation of the staging files from the host destined
files by just omitting the resulting BUILD_DIR+HOST_DIR path from the
following rsync call to the real target folder. The cleanup of many
files we dont want in target is deferred to the target-finalize
step. In addition to what's being removed already, we also have to
cleanup some Qt5 specific files (prl) and the documentation directory.
This approach was chosen over copying all files recorded in the pkg-files-list
after some discussion which Thomas Petazzoni summed up:
"We don't yet use pkg-files-list really as part of the build
process anywhere, I feel a bit more comfortable at this point with what
Andreas is proposing."
Thanks to this infrastructure, it will be possible to get rid of the
many conditional install commands because qmake already takes care of
this when generating the Makefile install targets with the given or
autodetected configure options of each package.
However, custom install steps may have to remain in cases where a
particular Buildroot option has no corresponding setting in the
packages configuration options.
Signed-off-by: Andreas Naumann <anaumann@ultratronik.de>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
In preparation for adding support for the C-SKY architecture in the
internal toolchain backend, we need to make sure that GNU_TARGET_NAME
will contain the appropriate ABI, i.e abiv1 or abiv2 depending on the
selected C-SKY core.
Signed-off-by: Guo Ren <ren_guo@c-sky.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
A note is added to tie off the discussion on why moving _FORTIFY_SOURCE
related flags into the toolchain wrapper doesn't currently work.
- Currently -D_FORTIFY_SOURCE and optimizations are passed through
CFLAGS
- Packages like linux-tools ignore CFLAGS entirely and some
autotools toolchain testing cases dependent on not using
CFLAGS.
- If FORTIFY_SOURCE is passed through the wrapper, then linux-tools
will no longer be able to ignore it, because it's enforced at a
lower-level and since the optimization -Os/g/1/2/3 are via CFLAGS,
there is no optimization flag set. Therefore linux-tools will do
all its configuration tests with FORTIFY_SOURCE forcefully enabled
at the wrapper level, but no optimization enabled, and consequently
tests will fail.
Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The RELRO/PIE flags are currently passed via CFLAGS/LDFLAGS and this patch
proposes moving them to the toolchain wrapper.
(1) The flags should _always_ be passed, without leaving the possibility
for any package to ignore them. I.e, when BR2_RELRO_FULL=y is used
in a build, all executables should be built PIE. Passing those
options through the wrapper ensures they are used during the build
of all packages.
(2) Some options are incompatible with -fPIE. For example, when
building object files for a shared libraries, -fPIC is used, and
-fPIE shouldn't be used in combination with -fPIE. Similarly, -r
or -static are directly incompatible as they are different link
time behaviors then the intent of PIE. Passing those options
through the wrapper allows to add some "smart" logic to only pass
-fPIE/-pie when relevant.
(3) Some toolchain, kernel and bootloader packages may want to
explicitly disable PIE in a build where the rest of the userspace
has intentionally enabled it. The wrapper provides an option
to key on the -fno-pie/-no-pie and bypass the appending of RELRO
flags.
The current Kernel and U-boot source trees include this option.
https://github.com/torvalds/linux/commit/8438ee76b004ef66d125ade64c91fc128047d244https://github.com/u-boot/u-boot/commit/6ace36e19a8cfdd16ce7c02625edf36864897bf5
If using PIE with a older Kernel and/or U-boot version, a backport of these
changes might be required. However this patchset also uses the
__KERNEL__ and __UBOOT__ defines as a way to disable PIE.
NOTE: The current implementation via CFLAGS/LDFLAGS has caused some
build time failures as the conditional logic doesn't yet exist in
Buildroot:
https://bugs.busybox.net/show_bug.cgi?id=11206https://bugs.busybox.net/show_bug.cgi?id=11321
Good summary of the most common build failures related to
enabling pie: https://wiki.ubuntu.com/SecurityTeam/PIE
[Peter: minor cleanups]
Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
They were added back in 5432f26f0 (Adding Central config.cache options),
supposedly to be able to cache the result of configure tests, but they
were never, ever referenced anywhere in our code... Besides, we dropped
the idea of getting a configure cache long ago now (it does not work)...
They are causing spurious error messages on some distros (e.g. Fedora)
which use GNU's which (whatever package that comes from), while it is
silent on other distros (e.g. Ubuntu) which use debianutils' which.
Drop them.
Reported-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The hardening options are compiler flags, not pure pre-processor flags, so
put them in CFLAGS, not CPPFLAGS.
This fixes build errors where -D_FORTIFY_SOURCE=2 whas put in CPPFLAGS and
then applied to configure tests which could fail since the required -O2 is
only in CFLAGS.
Originally submitted as
http://patchwork.ozlabs.org/patch/904057/
Signed-off-by: Stefan Sørensen <stefan.sorensen@spectralink.com>
Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Add a new infrastructure to ease the development of packages that use Meson as
their build system.
Signed-off-by: Eric Le Bihan <eric.le.bihan.dev@free.fr>
[Thomas:
- move global variables definition outside of the inner-meson-package
macro
- for consistency, remove double quote around value passed to meson
in the host configure step.
- minor formatting fixes.]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
This patch adds a new infrastructure for golang based packages.
Signed-off-by: Angelo Compagnucci <angelo@amarulasolutions.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Reviewed-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
[Arnout:
- Rewrap comments to 80 columns.
- Create a global definition of GO_TARGET_ENV.
- <PKG>_GO_ENV is appended to the default env instead of replacing it.
- Add a note to inner-golang-package that only target is supported.
]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Some packages that use libtool really need some love to be able to
disable C++ support.
This is because libtool will want to call AC_PROG_CXXCPP as soon as CXX
is set non-empty to something different from 'no'. Then, AC_PROG_CXXCPP
will want a C++ preprocessor that works on valid input *and* fail on
invalid input.
So, providing 'false' as the C++ compiler will then require that we do
have a working C++ preprocessor. Which is totally counter-productive
since we do not have a C++ compiler to start with...
bd39d11d2e (core/infra: fix build on toolchain without C++) was a
previous attempt at fixing this, by using the host's C++ preprocessor.
However, that is very incorrect (that's my code, I can say so!) because
the set of defines will most probably be different for the host and the
target, thus causing all sorts of trouble. For example, on ARM we'd have
to include different headers for soft-float vs hard-float, which is
decided based on a macro, which is not defined for x86, and thus may
redirect to the wrong (and missing) header.
Instead, we notice that libtool uses the magic value 'no' to decide that
a C++ compiler is not available, in which case it skips the call to
AC_PROG_CXXCPP.
Given that 'no' is not provided by any package in Debian and
derivatives, as well as in Fedora, we can assume that no system will
have an executable called 'no'. Hence, we use that as a magic value to
disable C++ detection altogether.
Fixes: #10846 (again)
Reported-by: Damien Riegel <damien.riegel@savoirfairelinux.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Damien Riegel <damien.riegel@savoirfairelinux.com>
Cc: Peter Seiderer <ps.report@gmx.net>
Cc: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Cc: Peter Korsgaard <peter@korsgaard.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Tested-by: Peter Seiderer <ps.report@gmx.net>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
-Ofast (introduced in GCC 4.6) It combines the existing optimization level -O3
with options that can affect standards compliance but result in better optimized
code. For example, -Ofast enables -ffast-math.
Signed-off-by: Joshua Henderson <joshua.henderson@microchip.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Autotools-based packages that do not need C++ but check for it, and use
libtool, will fail to configure on distros that lack /lib/cpp.
This is the case for example on Arch Linux, where expat fails to build
with:
configure: error: in `/home/dkc/src/buildroot/build/build/expat-2.2.4':
configure: error: C++ preprocessor "/lib/cpp" fails sanity check
This is because libtool uses AC_PROC_CXXCPP, which can not be avoided,
and does require a cpp that passes some "sanity" checks (does not choke
on valid input, but does choke on invalid input). So we can use neither
/bin/false nor /bin/true...
We instead need something that can digest some basic C++ preprocessor
input. We can't use the target preprocessor: that does not work, because
it obviously has no C++ cupport:
arm-linux-cpp.br_real: error: conftest.cpp: C++ compiler not
installed on this system
We can however consider that the host machine does have a C++ compiler,
so we use the host' cpp, which is gcc's compiler wrapper that ends up
calling the host's C++ preprocessor.
That would give us a valid C++ preprocessor when we don't have one, in
fact. But autotools will then correctly fail anyway, because there is
indeed no C++ compiler at all, as we can see in this excerpt of a
configure log from expat:
checking whether we are using the GNU C++ compiler... no
checking whether false accepts -g... no
checking dependency style of false... none
checking how to run the C++ preprocessor... cpp
checking whether the false linker (/home/ymorin/dev/buildroot/O/host/bin/arm-linux-ld) supports shared libraries... yes
libtool.m4: error: problem compiling CXX test program
checking for false option to produce PIC... -DPIC
checking if false PIC flag -DPIC works... no
checking if false static flag works... no
checking if false supports -c -o file.o... no
checking if false supports -c -o file.o... (cached) no
checking whether the false linker (/home/ymorin/dev/buildroot/O/host/bin/arm-linux-ld) supports shared libraries... yes
So, using the host's C++ preprocessor (by way of gcc's wrapper) leads to
a working situation, where the end result is as expected.
Reported-by: Damien Riegel <damien.riegel@savoirfairelinux.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Damien Riegel <damien.riegel@savoirfairelinux.com>
Cc: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Cc: Peter Korsgaard <peter@korsgaard.com>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Ben Wolsieffer