| Age | Commit message (Collapse) | Author |
|---|
|
Added checks for libgmp in configure scripts are not really portable, at
least 'gdb' subdir configure used host's libgmp and consequently failed.
At least there's 'host_configargs' to define additional arguments; use
it to point sub-configure to the right location.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Fix git-backed package patching and patch updating scripts:
* patch_git.sh was entirely broken (since forever?!) due to a missing
'cd' call
* update-patches-git formatted the marker commit as patch instead of
ignoring it
The major improvement of this review is the elimination of the mandatory
final marker commit. Putting it before the series being applied works
well and allows to add a patch to the series without demanding a
git-rebase call to get the marker back on top.
While being at it:
* Call 'git am --abort' only if there was a git repo already
* Call git-commit and git-format-patch with extra options to avoid
side-effects from a user's .gitconfig
* Explicitly specify output format in git-log to avoid surprises, also
split output properly with 'read'
* Eliminate fake_hdr variable
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
When using git to apply kernel patches, the extracted kernel sources
become a git repository with HEAD not being at a tag. Therefore
$(LINUX_DIR)/scripts/setlocalversion appends '+' to the kernel version
string which ADK in turn does not expect. In consequence, 'modprobe' on
target won't find any modules. Avoid this by creating an empty
'$(LINUX_DIR)/.scmversion' file - setlocalversion recognizes it and uses
its content instead of doing its own detection.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Without this, a call to 'make package=foo package' won't have the
desired effect since the directory named 'package' fulfills the
requirement.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
These are used by cryptsetup.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
This is an alternative to using MD_RAID, basically RAID managed via
lvm2 instead of mdadm.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Hard to run a NAS without it nowadays.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
This combines the typical kernel make call idiom into a single variable
reference. A side-effect is that parallel make is enabled (via added
'-j' flag) in many cases, but that shouldn't cause harm.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Introduce ADK_TARGET_KERNEL_LOADADDR symbol holding the right kernel
loadaddr depending on chosen target system instead of hard-coding the
value into kernel-vars.mk.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
It exists, so use it instead of manually stating $(TOPDIR)/scripts.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
No need to change the initramfs 'init' binary location in kernels
anymore, image.mk adds a symlink from /init to /sbin/init to cover for
that.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
OpenADK's kernel hack to call /sbin/init in initramfs is ugly. Instead,
ship /init as a symlink to /sbin/init. This is safe, since initramfs is
always a filesystem with symlink support.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Recent gnutls update introduced "gnulib module fopen" (gl/fopen.c) which
tries to get fopen() declaration from stdio.h by defining __need_FILE
before including the header. While this seems unnecessary for the
purpose (alegedly __need_FILE is used by headers including stdio.h to
get the FILE declaration while avoiding namespace pollution),
uClibc-ng's stdio.h behaves unexpected if __need_FILE is defined:
1) features.h is not included, therefore __BEGIN_NAMESPACE_STD (and
likely more) is not known later on
2) it declares *only* FILE, nothing else
To avoid these problems, don't define __need_FILE in gnutls and pray
this works with other libc's as well. Actually one should fix uClibc-ng
instead.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Starting with version 1.33, libgpg-error dropped the first two parts of
a four part "triplet" while canonicalizing CPU-VENDOR-KERNEL-SYSTEM.
Therefore the resulting header name for arm-openadk-linux-uclibceabi
becomes lock-obj-pub.linux-uclibceabi.h - make sure it exists by
creating an appropriate symlink. Leave the old symlink name in place to
avoid breaking any other triplet values.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Configure searches libexpat in host's rootfs and consequently fails.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Have to pass HOST_CC variable so _makenames helper is correctly compiled
for host arch.
Also PAM detection checks host's headers, override this based on package
choice.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Have to link against libuuid.
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
Otherwise the library builds, but programs trying to use it have to link
against libtinfo themselves (which they don't expect).
Signed-off-by: Phil Sutter <phil@nwl.cc>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
