The linux-headers -> linux dependency causes a circular dependency, breaking
the source/legal-info/graph-depends/.. targets:
make graph-depends
Getting targets
Getting dependencies for ['toolchain-buildroot', 'toolchain', 'busybox',
'glibc', 'initscripts', 'linux-headers', 'skeleton', 'linux',
'host-fakeroot', 'host-makedevs', 'rootfs-cpio', 'rootfs-initramfs']
Getting dependencies for ['host-kmod', 'host-gcc-final',
'host-gcc-initial', 'host-gawk']
Getting dependencies for ['host-gmp', 'host-binutils', 'host-pkgconf',
'host-mpfr', 'host-mpc']
Getting dependencies for ['host-m4']
Recursion detected for : toolchain
which is a dependency of: linux
which is a dependency of: linux-headers
which is a dependency of: glibc
which is a dependency of: host-gcc-final
which is a dependency of: toolchain-buildroot
which is a dependency of: toolchain
Makefile:721: recipe for target 'graph-depends' failed
make: *** [graph-depends] Error 1
Fix it by instead duplicating in linux-headers the 10-20 lines of linux.mk
logic that infer the _SOURCE/_SITE/_VERSION from the BR2_LINUX_KERNEL_*
variables.
This does mean that we extract the kernel sources twice though.
[Peter: use same git/hg tarball as linux kernel to not clone twice, minor fixes]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Some heavily (and most often improperly) modified Linux kernels may export
new APIs to userland, so as to speak to custom hardware or custom kernel
facilities.
However, we currently have no easy way to use such kernels as a source
for the linux-headers package, which precludes having those userland
headers intalled for userland applications to use them.
We do have a way for the kernel to use the same version as for the
headers, but that is definitely not enough, as the linux-headers package
has a version choice that is far less versatile and capable than that of
the linux package.
Add a new option for the linux-headers package, for the user to specify
that the version (really, the sources) of the kernel be used to install
the headers from.
We do that by making linux-headers patch-depend on the linux package.
We can't have linux-header simply depend on linux, because the simple
dependency means the the dependee will be configured, built and installed
before the dependent is configured. And since linux is a target package,
it depends on the toolchain, which internally dependes on linux-headers,
which would depend on linux, and we'd get a circular dependency.
Using patch-depend will ensure that linux is extracted and patched
before linux-headers is extracted, which is really all we need.
Then, we install the headers from the linux source tree, rather than
from linux-headers' source tree (as there's nothing in there!).
Since we need to install a private set for uClibc (see cde947f, uclibc:
prevent rebuilding after installation to staging), we explicitly set
INSTALL_HDR_PATH when calling the kernel' install-headers rule in
LINUX_HEADERS_CONFIGURE_CMDS, so that the headers are installed in
linux-headers' $(@D) instead of linux' $(@D).
Finally, as there is no way to know the kernel version in this case, we
must still prompt the user for the kernel series the headers are from
(like we do for a custom version) and check for consistency at build
time.
Note however that this still leaves users that want to built their
such-kernel outside of Buildroot out in the cold.
[Peter: drop comment as suggested by Thomas]
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Karoly Kasza <kaszak@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Peter Korsgaard <jacmet@uclibc.org>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Since the trailing slash is stripped from $($(PKG)_SITE) by pkg-generic.mk:
$(call DOWNLOAD,$($(PKG)_SITE:/=)/$($(PKG)_SOURCE))
so it is redundant.
This patch removes it from $(PKG)_SITE variable for BR consistency.
Signed-off-by: Jerzy Grzegorek <jerzy.grzegorek@trzebnica.net>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Currently we configure uClibc to use kernel headers from "staging" folder with
KERNEL_HEADERS="$(STAGING_DIR)/usr/include". This path is added to include
search path of uClibc build system in Rules.mak "CFLAGS += -I$(KERNEL_HEADERS)".
At the same time on uClibc installation to "staging" we point to the same
location "$(STAGING_DIR)/usr" (headers effectively go in "usr/include").
So after every installation to "staging" dependences get touched (even though we
copy the same headers every time) and so we may see lots of sources in uClibc
get rebuilt.
This has 2 consequences:
1. Longer build time - becase even on ordinary buildroot build uClibc is built
twice. On "uclibc building" and on "uclibc installation to target".
2. Symbols in libuClibc built initially (that is later installed in
"staging/sysroot") are situated with different offset compared to second build
(later copied in "target"). This happens because as described above only part
of sources get rebuilt and then on final linkage object files are linked in
different order.
And (2) leads to problems on remote rebugging: gdbserver reports offsets that
correspond to pointless assembly in libuClibc on host.
Here's how it looks like.
Before this patch:
$ cd ~/br2_output/i586/target/lib
$ i586-buildroot-linux-uclibc-readelf -s libuClibc-0.9.33.2.so | grep kill
423: 0000c42c 54 FUNC GLOBAL DEFAULT 7 kill
$ cd ~/br2_output/i586/staging/lib
$ i586-buildroot-linux-uclibc-readelf -s libuClibc-0.9.33.2.so | grep kill
423: 0000b518 54 FUNC GLOBAL DEFAULT 7 kill
After this patch:
$ cd ~/br2_output/i586/target/lib
$ i586-buildroot-linux-uclibc-readelf -s libuClibc-0.9.33.2.so | grep kill
423: 0000b518 54 FUNC GLOBAL DEFAULT 7 kill
$ cd ~/br2_output/i586/staging/lib
$ i586-buildroot-linux-uclibc-readelf -s libuClibc-0.9.33.2.so | grep kill
423: 0000b518 54 FUNC GLOBAL DEFAULT 7 kill
Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Peter Korsgaard <peter@korsgaard.com>
Cc: Noam Camus <noamc@ezchip.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Old toolchains, with old gcc that do not support -print-sysroot, break the
kernel-headers version check script: it fails to find the sysroot of the
toolchain, and thus ends up including the host's linux/version.h.
Most of the time, this will break early, since the host's kernel headers
will not match the toolchain settings.
But it can happen that the check is succesful, although the configuration
of the toolchain is wrong:
- the custom toolchain has kernel headers vX.Y
- the user selected vX.Z (Z!=Y)
- the host has headers vX.Y
In this case, the check passes OK, but the build of some packages later on
will break (which is exactly what those _AT_LEAST_XXX options were added to
avoid).
Fix that by passing the sysroot to the check script, instead of the cross
compiler.
We get the sysroot as thus:
- for custom toolchains, we use the macro toolchain_find_sysroot. We can
do that, because we already have a complete sysroot with libc.a at that
time.
- for internal toolchain using a custom kernel headers version, we just
use $(STAGING_DIR). We can't use the macro as for custom toolchains
above, because at the time we install the kernel headers, we do not yet
have a complete sysroot with a libc.a. But we can just use
$(STAGING_DIR), since we're only interested in the kernel headers.
For all other types of toolchains, we already have the _AT_LEAST_XXX options
properly set, so we need not add a check in this case.
Fixes:
http://autobuild.buildroot.net/results/f33/f331a6eff0b0b93c73af52db3a6b43e4e598577e/http://autobuild.buildroot.net/results/a57/a5797c025bec50c10efdcff74945aab4021d05e4/
[...]
[Thanks to Thomas for pointing out the toolchain_find_sysroot macro!]
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Ensure the kernel headers version matches exactly the one manually
specified by the user.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Cc: Thomas De Schampheleire <patrickdepinguin@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
For some architectures (eg. Arc, Cris, Hexagon, ia64, Parisc, Score and
Xtensa), the Linux buildsystem tries to call the cross-compiler when
installing the headers.
This is a spurious call, since a cross-compiler is not needed at all to
install the headers.
As some users have reported the issue, just add a comment in linux-headers.mk
directing the user to ignore those errors.
Reported-by: Noam Camus <noamc@ezchip.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
This commit makes the dependency from the target toolchain explicit.
This way we can buid from command line a package that use
inner-generic-package right after the configuration phase, example:
make clean <package-name>
Also remove TARGETS_ALL because the only purpose was to add toolchain
dependency so it's superseded by this commit.
To prevent circular dependency add the new variable
<pkgname>_ADD_TOOLCHAIN_DEPENDENCY to avoid adding the toolchain
dependency for toolchain packages.
This is also a step forward supporting top-level parallel make.
Signed-off-by: Fabio Porcedda <fabio.porcedda@gmail.com>
Acked-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The headers and kernels where changed to XZ format on commit
98b5cc3eb4, but the headers reverted back
to bz2 on the packaging of the toolchain.
This causes double kernel downloads when the versions match, so switch
back the headers to XZ.
Signed-off-by: Gustavo Zacarias <gustavo@zacarias.com.ar>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
