This reverts commit 8e91385a2c.
This commit is incorrect, as it is perfectly valid for
BR2_TOOLCHAIN_EXTERNAL_PATH to be empty. The help text of
BR2_TOOLCHAIN_EXTERNAL_PATH even documents it as a supported case:
If empty, the compiler will be searched in $PATH.
Commit 392b0a26f5 ("toolchain-external:
default BR2_TOOLCHAIN_EXTERNAL_PATH to empty") even made that the
default saying "In addition, it in fact works correctly when it is
empty. In that case, the toolchain will be searched in PATH."
A user has reported that commit
8945ba4948 (the backport of 8e91385a2c to
the 2022.02.x LTS branch) breaks his use-case:
https://lore.kernel.org/buildroot/CADBnMvhgaozAgZgy3njckjL1i0U6bZ0fLrq-kdFF-qpGhFWgmw@mail.gmail.com/
Reported-by: Kristof Havasi <havasiefr@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
[yann.morin.1998@free.fr: reference 8e91385a2c on master]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Convert BR2_TOOLCHAIN_EXTERNAL_PATH to an absolute path when used.
Otherwise the symbolic links to the external toolchain binaries are
not installed in host/bin when BR2_TOOLCHAIN_EXTERNAL_PATH is relative.
This happens because TOOLCHAIN_EXTERNAL_INSTALL_WRAPPER
changes directory into host/bin to create the symbolic links.
From there the tools are no longer found via the relative path and
a single symbolic link host/bin/$(prefix)-* is created instead.
Although relative paths sounds like something less than ideal to put in
a Buildroot configuration, it's actually rather typical to put the
buildroot sources as a submodule (or subdirectory) of custom sources
(either in a BR2_EXTERNAL or not), in which case the relative path is
well-defined.
Signed-off-by: Juergen Stuber <juergen@jstuber.net>
Reviewed-by: Romain Naour <romain.naour@smile.fr>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
`which' has been discontinued after 2.21 release in 2015 due this (git
repository is empty [1]) and version shipped in Debian produces warning
[2]:
/usr/bin/which: this version of `which' is deprecated; use `command -v' in scripts instead.
`command is POSIX [3] and supported on all common shells (bash, zsh,
dash, busybox sh, mksh).
Patch tested on dash as the default shell.
[1] https://git.savannah.gnu.org/cgit/which.git
[2] 3a8dd10b45
[3] https://pubs.opengroup.org/onlinepubs/9699919799/utilities/command.html
Signed-off-by: Petr Vorel <petr.vorel@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
gcc installs a libstdcxx-...so-gdb.py file that gdb will load automatically
when it loads libstdcxx.so, via the mechanism described at [1].
However, the auto-load file installed by gcc contains hardcoded paths
referring to the location where the (external) toolchain was built, which
are normally not available.
Fix up the paths in the load file so that the pretty printers can be loaded
automatically.
Note that gdb will only auto-load the file if its location is marked as
'safe'. A subsequent commit will take care of that.
Technically, there could be more than one load file, e.g. in lib and
usr/lib, so fix them all. This was for example observed in
BR2_TOOLCHAIN_EXTERNAL_ARM_AARCH64.
In a very specific case with a local custom toolchain, there were actually
two 'python' directories, which would break the sed command, so arbitrarily
limit to the first one encountered.
[1] https://sourceware.org/gdb/onlinedocs/gdb/objfile_002dgdbdotext-file.html
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
From: Julien Boibessot <julien.boibessot@armadeus.com>
It could be usefull to have ldd on the target so install it.
Signed-off-by: Julien Boibessot <julien.boibessot@armadeus.com>
[Sébastien: add commit message]
Signed-off-by: Sébastien Szymanski <sebastien.szymanski@armadeus.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
The external toolchain configure step calls the
check_kernel_headers_version make function to compare the kernel
headers version declared in the configuration with the actual kernel
headers of the toolchain.
This function takes 4 arguments, but due to a missing comma what
should be the first two arguments are both passed into the first
argument. Due to this, when check_kernel_headers_version does:
if ! support/scripts/check-kernel-headers.sh $(1) $(2) $(3) \
$(if $(BR2_TOOLCHAIN_HEADERS_LATEST),$(4),strict); \
Then:
$(1) contains "$(BUILD_DIR) $$(call toolchain_find_sysroot,$$(TOOLCHAIN_EXTERNAL_CC))"
$(2) contains "$$(call qstrip,$$(BR2_TOOLCHAIN_HEADERS_AT_LEAST))"
$(3) contains "$$(if $$(BR2_TOOLCHAIN_EXTERNAL_CUSTOM),loose,strict))"
So from the point of view of check-kernel-headers.sh, it already has
four arguments, and therefore the additional argument passed by:
$(if $(BR2_TOOLCHAIN_HEADERS_LATEST),$(4),strict); \
is ignored, defeating the $(BR2_TOOLCHAIN_HEADERS_LATEST) test.
The practical consequence is that a toolchain that has 5.4 kernel
headers but declared as using 5.3 kernel headers does not abort the
build, because the check is considered "loose" while it should be
"strict".
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
When Buildroot is released, it knows up to a certain kernel header
version, and no later. However, it is possible that an external
toolchain will be used, that uses headers newer than the latest version
Buildroot knows about.
This may also happen when testing a development, an rc-class, or a newly
released kernel, either in an external toolchain, or with an internal
toolchain with custom headers (same-as-kernel, custom version, custom
git, custom tarball).
In the current state, Buildroot would refuse to use such toolchains,
because the test is for strict equality.
We'd like to make that situation possible, but we also want the user not
to be lenient at the same time, and select the right headers version
when it is known.
So, we add a new Kconfig blind option that the latest kernel headers
version selects. This options is then used to decide whether we do a
strict or loose check of the kernel headers.
Suggested-by: Aaron Sierra <asierra@xes-inc.com>
Signed-off-by: Vincent Fazio <vfazio@xes-inc.com>
[yann.morin.1998@free.fr:
- only do a loose check for the latest version
- expand commit log
]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Tested-by: Vincent Fazio <vfazio@xes-inc.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
This patch extends the "copy extra GCC libraries to target" feature to
also work for internal toolchains. The variable has been renamed to be
BR2_TOOLCHAIN_EXTRA_LIBS and the configuration option moved under the
generic toolchain package. For external toolchains, the step that does
the copy is still in the copy_toolchain_lib_root() helper which copies
from the sysroot to the target. For the internal toolchain, the host
gcc-final package does a post install hook to copy the libraries from
the toolchain build folders to both the sysroot and target(!static).
Examples where this can be useful is for adding debug libraries to the
target like the GCC libsanitizer (libasan/liblsan/...).
Cc: Markus Mayer <mmayer@broadcom.com>
Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Commit 32bec8ee2f
("toolchain-external: copy ld*.so* for all C libraries") changed (among
other things) the glob pattern to catch the dynamic loader from
ld*.so.*
to
ld*.so*
thus now matching files like 'ld-2.20.so' in addition to files like
'ld.so.1'.
However, there is no apparent reason why that change was made. It is
not explicitly mentioned in the commit message as to why that would be
needed, nor is clear based on the rest of the changes in that
commit. But it turns out that it causes too many files to be copied
with some toolchains.
In most toolchains, the structure looks like this:
-rwxr-xr-x 1 tdescham tdescham 834364 Feb 16 21:23 output/target/lib/ld-2.16.so
lrwxrwxrwx 1 tdescham tdescham 10 Feb 16 21:23 output/target/lib/ld.so.1 -> ld-2.16.so
So, a symlink 'ld.so.1' which points to another file. Applications
would have 'ld.so.1' (the link) encoded as program interpreter
(readelf -l <program>, see INTERP entry)
The patterns like 'ld*.so*' are passed as argument to
copy_toolchain_lib_root which is defined in toolchain/helpers.mk.
This macro copy_toolchain_lib_root will find all files/links matching
the pattern. If a match is a regular file, it is simply copied. If it
is a symbolic link, the link is copied and then the logic is
recursively repeated on the link destination. That destination could
either again be a link or a regular file. In the first case we recurse
again, in the latter we stop and continue with the next match of the
pattern.
The problem this patch is solving is when a toolchain does not have
this structure with a link and a real file, but rather two actual
files:
-rwxr-xr-x 1 tdescham tdescham 170892 Feb 16 21:55 output/target/lib/ld-2.20.so
-rwxr-xr-x 1 tdescham tdescham 170892 Feb 16 21:55 output/target/lib/ld.so.1
In this case the pattern 'ld*.so*' would find two regular file matches
and copy both. On the other hand, the pattern 'ld*.so.*' would only
find the 'ld.so.1' file and copy just that. This saves about 170K in
rootfs size.
Closer inspection reveals that this particular toolchain has more such
dedoubled symbolic links, e.g. the standard pattern of
'usr/lib/libfoo.so -> libfoo.so.1 -> libfoo.so.1.0.2' is not present,
and each of these three components are real files. In any case, it is
obvious that the toolchain itself is 'broken'.
That being said, because we have the logic that recursively resolves
symbolic links, TOOLCHAIN_EXTERNAL_LIBS really only needs to contain
the "initial" name of the library to be copied.
Therefore, revert the glob pattern back to what it was.
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
[Thomas: improve the commit log with the additional details from Thomas]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
In Buildroot, the internal toolchain backend uses the SSP support from
the C library, not that of gcc.
Some external toolchains come with SSP suport in gcc, which is
implemented in libssp.so, rather than in the C library.
When a toolchain even has both, it is up to the compiler to decide
whether it will link to libssp or use the support from the C library.
However, in the latter case, a (incorrectly written) package may decide
to explicitly link with libssp.so when it is available (even though the
compiler may have decided otherwise if left by itself). This is the case
for example with sox, which results in runtime failures, such as:
$ sox
sox: error while loading shared libraries: libssp.so.0: cannot open
shared object file: No such file or directory
Even if sox is wrong in doing so, the case for libssp-only toolchains is
still valid, and we must copy it as we copy other libs.
Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Some installations mount /tmp with the 'noexec' option, which prevents
running the program generated there to check the kernel headers.
Avoid the problem by generating the program under $(BUILD_DIR), passed
as the first argument to check-kernel-headers.sh.
We could globally export a TMPDIR environment variable with some path
under $(BUILD_DIR) but such solution would be too intrusive, depriving
the user from the freedom to set TMPDIR at his will (or needs).
Fixes: https://bugs.busybox.net/show_bug.cgi?id=12241
Signed-off-by: Carlos Santos <unixmania@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Some toolchain vendors may have backported those options to older gcc
versions, and we have no way to know, so we have to check that the
user's selection is acceptable.
Extend the macro that currently checks for SSP in the toolchain, with
a new test that the actual SSP option is recognised and accepted.
Note that the SSP option is either totaly empty, or an already-quoted
string, so we can safely and easily assign it to a shell variable to
test and use it.
Note that we do not introduce BR2_TOOLCHAIN_HAS_SSP_STRONG, because:
- our internal toolchain infra only supports gcc >= 4.9, so it has
SSP strong;
- of the external pre-built toolchains, only the codesourcery-arm
one has a gcc-4.8 which lacks SSP strong, all the others have a
gcc >= 4.9;
- we'd still have to do the actual check for custom external
toolchains anyway.
So, we're not adding BR2_TOOLCHAIN_HAS_SSP_STRONG just for a single
case.
Signed-off-by: "Yann E. MORIN" <yann.morin@orange.com>
Cc: Matt Weber <matthew.weber@rockwellcollins.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Commit 23c0e97b29 (toolchain-external: anchor sysroot regex with /)
tried to make the find-sysroot work more consistently, especially for
toolchains where the C library is located in a sub-directory, like the
"Realtek mips toolchain".
After that patch, the '/' that was trailing in the returned path got
removed now. This in turn breaks the Codesourcery toolchain.
We fix that by appending the now-missing trailing '/'.
Fixes:
http://autobuild.buildroot.net/results/9284d571668148febce23d96a9c0a97a6b2b43dc
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: 陈小 刚 <shawn_chen@realsil.com.cn>
Cc: Arnout Vandecappelle <arnout@mind.be>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Anchor the regex in toolchain_find_sysroot macro with a / to avoid
unexpected substitution for Realtek mips toolchain, for which the libc.a
path ends with 'mips-linux-uclibc/lib/libc.a'.
Signed-off-by: 陈小 刚 <shawn_chen@realsil.com.cn>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Add new BR2_TOOLCHAIN_HAS_OPENMP option for toolchains with OpenMP
support.
Signed-off-by: Ed Blake <ed.blake@sondrel.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
commit e0d14fb21b (toolchain-external: drop no longer needed
CC_TARGET_<foo>_ variables) dropped the CC_TARGET_* variables, but missed
one. Fix that.
Reported-by: Mark Corbin <mark.corbin@embecosm.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Since the introduction of the GCC_TARGET_<foo> variables in
arch/arch.mk in commit bd0640a213
("arch: allow GCC target options to be optionally overwritten") and
the removal of the BR2_GCC_TARGET_CPU_REVISION, the CC_TARGET_<foo>_
variables in pkg-toolchain-external.mk map 1:1 with the corresponding
GCC_TARGET_<foo> variables.
So let's drop the CC_TARGET_<foo>_ variables, and use directly the
GCC_TARGET_<foo> ones.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
In commit 325bb37942, support for the
Blackfin architecture was removed. This was our only use of
BR2_GCC_TARGET_CPU_REVISION, and since this config option somewhat
complicates the calculation of the --with-cpu/-mcpu option values,
let's drop it.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The BR2_GCC_TARGET_* configuration variables are copied to
corresponding GCC_TARGET_* variables which may then be optionally
modified or overwritten by architecture specific makefiles.
All makefiles must use the new GCC_TARGET_* variables instead
of the BR2_GCC_TARGET_* versions.
Signed-off-by: Mark Corbin <mark.corbin@embecosm.com>
[Thomas: simplify include of arch/arch.mk]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
The check_musl function currently builds a program and verifies if the
program interpreter starts with /lib/ld-musl. While this works fine
for dynamically linked programs, this obviously doesn't work for a
purely static musl toolchain such as [1].
There is no easy way to identify a toolchain as using the musl C
library. For glibc, dynamic linking is always supported, so we look at
the dynamic linker name. For uClibc, there is a distinctive
uClibc_config.h header file. There is no such distinctive feature in
musl.
We end up resorting to looking for the string MUSL_LOCPATH, which is
used by musl locale_map.c source file. This string has been present in
musl since 2014. It certainly isn't a very stable or convincing
solution to identify the C library as being musl, but it's the best we
could find.
Note that we are sure there is a libc.a file, because the
check_unusable_toolchain function checks that there is a such a file.
[1] http://autobuild.buildroot.net/toolchains/tarballs/br-arm-musl-static-2018.05.tar.bz2
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
There are cases where a downloaded toolchain doesn't have its binaries
placed directly in a "bin" subfolder (where BuildRoot currently looks
for them).
A common example is the official Raspberry Pi Toolchain
(https://github.com/raspberrypi/tools), which has its binaries in
"arm-bcm2708/arm-linux-gnueabihf/bin".
This commit introduces BR2_TOOLCHAIN_EXTERNAL_REL_BIN_PATH that defaults
to "bin" and can be changed as needed.
Signed-off-by: Calin Crisan <ccrisan@gmail.com>
Reviewed-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
[Thomas: rework a bit how TOOLCHAIN_EXTERNAL_REL_BIN_PATH is defined.]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
libatomic, like libgcc_s, is provided by gcc, so there is no reason to
copy it over only for the glibc and uclibc cases, it should also be
copied for the musl case. Without this, a program linked with
libatomic on a musl system will fail to run due to the missing
library.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Reviewed-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The upcoming per-package SDK functionality is heavily based on the
fact that HOST_DIR, STAGING_DIR and TARGET_DIR are evaluated during
the configure/build/install steps of the packages. Therefore, any
evaluation-during-assignment using := is going to cause problems, and
need to be turned into evaluation-during-use using =.
This patch fix up one such instance in the external toolchain code.
This change is independent from the per-package SDK functionality, and
could be applied separately.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Reviewed-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Since 392b0a26f5 (toolchain-external: default BR2_TOOLCHAIN_EXTERNAL_PATH
to empty), calling 'make clean' or similar can yield a spurious stderr
message:
dirname: missing operand
Try 'dirname --help' for more information.
Which is definitely baffling and unsettling...
It turns out that it is pretty trivial to reproduce, and this defconfig
is just enough:
$ cat my-defconfig
BR2_TOOLCHAIN_EXTERNAL=y
$ make BR2_DEFCONFIG=$(pwd)/my-defconfig defconfig
$ make clean
dirname: missing operand
Try 'dirname --help' for more information.
[--snip--]
This is because the cross-compiler is not found in the PATH (and for
good reasons, I don't have it in the PATH, not even at all).
So, when the cross-compiler is not found in the path, we simply
continue as if all was good, and postpone the check to much later,
when we try to copy the toolchain libs...
So, use a make construct rather than calling to the shell: $(dir ...)
does not whine if passed nothing.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Arnout Vandecappelle <arnout@mind.be>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Thomas De Schampheleire <patrickdepinguin@gmail.com>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
-mnan option was added in gcc-4.9.0 so make sure that users cannot
select the NaN mode when using toolchains that have a gcc older
than 4.9.0, and also make sure that the -mnan option is not passed at
all to the toolchain-wrapper and target cflags.
Signed-off-by: Vicente Olivert Riera <Vincent.Riera@imgtec.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
MIPS32 support different FP modes (32,xx,64), so give the user the
opportunity to choose between them. That will cause host-gcc to be built
using the --with-fp-32=[32|xx|64] configure option. Also the
-mfp[32|xx|64] gcc option will be added to TARGET_CFLAGS and to the
toolchain wrapper.
FP mode option shouldn't be used for soft-float, so we add logic in the
toolchain wrapper if -msoft-float is among the arguments in order to not
append the -fp[[32|xx|64] option, otherwise the compilation may fail.
Information about FP modes here:
- https://sourceware.org/binutils/docs/as/MIPS-Options.html
- https://dmz-portal.imgtec.com/wiki/MIPS_O32_ABI_-_FR0_and_FR1_Interlinking#5._Generating_modeless_code
Signed-off-by: Vicente Olivert Riera <Vincent.Riera@imgtec.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
MIPS supports two different NaN encodings, legacy and 2008. Information
about MIPS NaN encodings can be found here:
https://sourceware.org/binutils/docs/as/MIPS-NaN-Encodings.html
NaN legacy is the only option available for R2 cores and older.
NaN 2008 is the only option available for R6 cores.
R5 cores can have either NaN legacy or NaN 2008, depending on the
implementation. So, if the user selects a generic R5 target architecture
variant, we show a choice menu with both options available. For well
known R5 cores we directly select the NaN enconding they use.
Signed-off-by: Vicente Olivert Riera <Vincent.Riera@imgtec.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Commit 32bec8ee2f (toolchain-external: copy ld*.so* for all C libraries)
removed the definition of TOOLCHAIN_EXTERNAL_MUSL_LD_LINK. Remove also the
reference to it.
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Since things are no longer installed in $(HOST_DIR)/usr, the callers
should also not refer to it.
This is a mechanical change with
git grep -l '$(HOST_DIR)/usr/bin' | xargs sed -i 's%$(HOST_DIR)/usr/bin%$(HOST_DIR)/bin%g'
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
libgcc_s.so is now added to TOOLCHAIN_EXTERNAL_LIBS for glibc/uclibc
in one place, and for musl in another place. Bottom line: it should be
in TOOLCHAIN_EXTERNAL_LIBS unconditionally.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Currently, for the dynamic loader, we're copying ld*.so* for glibc and
uClibc, except for glibc/EABIhf where we are explicitly copying
ld-linux-armhf.so.*. For musl, we're not copying the dynamic linker
because it's simply a symbolic link to libc.so. However, the name of
the musl dynamic linker changes from one architecture to the other,
and we don't handle all cases.
Since handling the musl dynamic linker symlink creation is becoming
more and more annoying to maintain, this commit makes musl use the
same mechanism as glibc/uClibc: put the dynamic linker in
TOOLCHAIN_EXTERNAL_LIBS.
In addition, the special condition on glibc/EABIhf was added in
11ec38b695 ("toolchain-external: fix
Linaro ARM toolchain support") because an old Linaro toolchain had two
dynamic loaders, and we wanted to copy only one. But 1/ this is old
and 2/ having the two dynamic linkers doesn't really matter.
So this commit simply unconditionally adds "ld*.so*" to
TOOLCHAIN_EXTERNAL_LIBS, regardless of the C library being chosen. It
re-uses the musl dynamic linker symlink from the sysroot, which makes
it always correct, and allows us to remove the
TOOLCHAIN_EXTERNAL_MUSL_LD_LINK hook, and all the related logic.
This commit therefore solves two problems with the musl dynamic linker
symbolic link creation logic:
1 We support all architectures, without having to hardcode in
Buildroot the mapping between the CPU architecture and the
corresponding dynamic linker name. For example, our current logic
was not handling the mips64+n32 ABI case, where the dynamic linker
is named ld-musl-mipsn32el.so.1.
2 We support Crosstool-NG musl toolchains, where the dynamic linker
is in /lib, but libc.so is in /usr/lib.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
---
This commit therefore replaces:
- https://patchwork.ozlabs.org/patch/780411/ (was another solution
for solving problem 1 above)
- https://patchwork.ozlabs.org/patch/763977/ and
https://patchwork.ozlabs.org/patch/748974/ (was another solution
for solving problem 2 above)
We currently use $(HOST_DIR)/usr as the prefix for host packages. That
has a few disadvantages:
- There are some things installed in $(HOST_DIR)/etc and
$(HOST_DIR)/sbin, which is inconsistent.
- To pack a buildroot-built toolchain into a tarball for use as an
external toolchain, you have to pack output/host/usr instead of the
more obvious output/host.
- Because of the above, the internal toolchain wrapper breaks which
forces us to work around it (call the actual toolchain executable
directly). This is OK for us, but when used in another build system,
that's a problem.
- Paths are four characters longer.
To allow us to gradually eliminate $(HOST_DIR)/usr while building
packages, replace it with a symlink to .
The symlinks from $(HOST_DIR)/usr/$(GNU_TARGET_NAME) and
$(HOST_DIR)/usr/lib that were added previously are removed again.
Note that the symlink creation will break when $(HOST_DIR)/usr
already exists as a directory, i.e. when rebuilding in an existing
output directory. This is necessary: if we don't break it now, the
following commits (which remove the usr part from various variables)
_will_ break it.
At the same time as creating this symlink, we have to update the
external toolchain wrapper and the external toolchain symlinks to go
one directory less up. Indeed, $(HOST_DIR) is one level less up than
it was before.
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Reviewed-by: Romain Naour <romain.naour@smile.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Static build with external musl toolchain leaves a dangling symlink to
libc.so. Don't create that symlink on static build.
Cc: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Remove mention of toolchains the we don't have.
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Reviewed-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The external toolchain code has some logic to calculate the correct name
for the dynamic linker symbolic link that needs to be created when the
musl C library is being used. There was already some handling for the
mipsel+soft-float case, but not for the mips+soft-float case. Due to
this, the symbolic link was incorrectly named, and programs were
referencing an non-existing file.
Reported-by: Florent Jacquet <florent.jacquet@free-electrons.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The input to copy_toolchain_lib_root is not one library, not a list of
libraries, but a library name pattern with glob wildcards.
This pattern is then passed to 'find' to get the actual list of libraries
matching the pattern. Reflect this using an appropriate variable name.
Note: if the root of the buildroot tree contains a file matching one of
these library patterns, the copying of libraries from staging to target will
not be correct. It is not impossible to fix that, e.g. using 'set -f', but
maybe it's not worth it.
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The toolchain from the Cavium Octeon SDK has a sysroot layout as follows:
./lib32
./lib32/octeon2
./lib32-fp
./lib64
./lib64/octeon2
./lib64-fp
./usr
./usr/lib
./usr/lib32
./usr/lib32/octeon2
./usr/lib32-fp
./usr/lib64
./usr/lib64/octeon2
./usr/lib64-fp
./usr/bin
./usr/bin32
./usr/bin32-fp
./usr/bin64-fp
./usr/libexec
./usr/libexec32
./usr/libexec32-fp
./usr/libexec64-fp
./usr/sbin
./usr/sbin32
./usr/sbin32-fp
./usr/sbin64-fp
./usr/include
./usr/share
./sbin
./sbin32
./sbin32-fp
./sbin64-fp
./etc
./var
with the following selections:
- lib64 : default
- lib64/octeon2 : -march=octeon2
- lib64-fp : -march=octeon3
- lib32 : -mabi=n32
- lib32/octeon2 : -mabi=n32 -march=octeon2
- lib32-fp : -mabi=n32 -march=octeon3
In case of '-mabi=n32 -march=octeon2' (but same is true for n64+octeon2)the
original Buildroot toolchain logic would copy both the libraries in
lib32 as the subdirectory lib32/octeon2, which means that every library is
installed twice (but only one of each is really needed).
While ARCH_LIB_DIR is determined by the location of libc.a, which in this
case is effectively:
<sysroot>/usr/lib32/octeon2/libc.a
the variable only retains 'lib32' and not 'lib32/octeon2' as expected.
To make Buildroot cope with this style of toolchain layout, we need to adapt
the calculation of ARCH_LIB_DIR to also include the second part.
This, in turn, means that ARCH_LIB_DIR is no longer guaranteed to be a
singular path component, resulting in some additional changes.
Certain older Linaro toolchains actually had the same layout. Libraries were
located in lib/<tuple> rather than lib directly. Previously, this was
handled by adding a toolchain-specific fixup that creates a symlink
lib/<tuple> -> lib, but with this patch this would no longer be needed.
Note that one difference with the Octeon case is that these Linaro
toolchains are not actually multilib, i.e. there is just one location with
the libraries and thus there is no problem with duplicated libraries.
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The check_arm_abi function takes as second argument the path to the
cross-readelf, but does not use it. Therefore, this commit gets rid of
this unnecessary argument.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The current test to verify if the toolchain uses musl or not is based on
checking if /lib/libc.so or /lib/libm.so exist in the sysroot. However,
some toolchains (notably Crosstool-NG ones) put these libraries in
/usr/lib/.
To fix this, build a minimal C program and check if the program
interpreter contains /lib/ld-musl.
Signed-off-by: Ilya Kuzmich <ilya.kuzmich@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
TOOLCHAIN_EXTERNAL_READELF is defined to
$(TOOLCHAIN_EXTERNAL_CROSS)readelf$(TOOLCHAIN_EXTERNAL_SUFFIX), where
TOOLCHAIN_EXTERNAL_SUFFIX is .br_real for Buildroot
toolchains. However, this is bogus, because readelf is not wrapped by
the Buildroot toolchain wrapper, so "<arch>-readelf.br_real" never
exists.
Therefore, it should simply be defined as
$(TOOLCHAIN_EXTERNAL_CROSS)readelf. Currently,
TOOLCHAIN_EXTERNAL_READELF is not used anywhere, so it wasn't visible,
but a follow-up commit will make use of it.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
libanl.so is needed for asynchronous network address and service
translation, declared in netdb.h
Signed-off-by: Jesper Bækdahl <jbb@gamblify.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The variables LIBC_A_LOCATION and ARCH_LIBC_A_LOCATION were killed in commit
646bd86908 but the corresponding descriptions
were never removed.
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
This toolchain has many problems which are fixed in contemporary gcc
and uClibc-ng. In addition, several hacks are needed to be able to
work with this toolchain. All these hacks are removed as well. Also
the package exceptions for this toolchain are removed.
The BR2_BFIN_INSTALL_FDPIC_SHARED and BR2_BFIN_INSTALL_FLAT_SHARED
options don't get a legacy entry. For the ADI toolchain, there already
is a legacy entry, so it doesn't make sense to add it twice. For other
external toolchains, these options didn't actually work, because they
rely on the specific layout of the ADI toolchain.
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The toolchain-external-package infrastructure is just a copy of the
toolchain-external commands, replacing TOOLCHAIN_EXTERNAL by $(2)
and adding double-dollars everywhere.
toolchain-external itself is converted to a virtual package, but it
is faked a little to make sue the toolchains that haven't been
converted to toolchain-external-package yet keep on working.
The TOOLCHAIN_EXTERNAL_MOVE commands don't have to be redefined
for every toolchain-external-package instance, so that is moved
out into the common part of pkg-toolchain-external.mk.
The musl-compat-headers dependency stays in the toolchain-external
package itself.
The musl ld link is duplicated in the legacy toolchain-external and
the toolchain-external-package, because they have separate hooks.
The handling of TOOLCHAIN_EXTERNAL_BIN deserves some special attention,
because its value will be different for different
toolchain-external-package instances. However, the value only depends
on variables that are set by Kconfig (BR2_TOOLCHAIN_EXTERNAL_PREFIX
and BR2_TOOLCHAIN_EXTERNAL_DOWNLOAD) so it can easily be used in
the generic part. So we don't have to do anything specific for this
variable after all.
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Romain Naour <romain.naour@gmail.com>
Reviewed-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
