The .init and .fini sections are built by concatenating code
fragments. Putting mips16 code in the middle of a mips32 code block
doesn't work. Make gcc built the magic crt stuff in no-mips16 mode.
This is specific to 4.6-linaro but is probably portable to other gcc
flavors. Adding this to the t-libgcc-mips16 makefile fragment is a
hack not suitable for pushing upstream, but there is no mips/t-linux
or mips/t-uclibc and I am not going to touch gcc/configure for two
lines.
Signed-off-by: Jay Carlson <nop@nop.com>
Signed-off-by: Florian Fainelli <florian@openwrt.org>
SVN-Revision: 36200
For now, build uclibc without -mips16. There's no mips16 syscall
support AFAIK and uclibc uses inline assembly syscalls a lot.
In addition, touching errno means touching a TLS model symbol, and
that's not supported in gcc 4.6.
The __set_errno macro can be put back to calling
errno_location(). This allows much of the library to be built in
mips16 mode. I don't understand the implications to the thread library
of doing this.
A list of "build as -mno-mips16" C source files can be placed in the
mips architecture-dependent build files. Maintaining the list would be
no fun.
Signed-off-by: Jay Carlson <nop@nop.com>
Signed-off-by: Florian Fainelli <florian@openwrt.org>
SVN-Revision: 36199
this patch is a copy of
trunk/toolchain/eglibc/patches/2.16/200-add-dl-search-paths.patch in
trunk/toolchain/eglibc/patches/2.17/200-add-ld-search-paths.patch
please use svn copy instead of applying it (like in
https://dev.openwrt.org/changeset/34290/)
the patch add /usr/lib in the ld search path (for eglibc 2.17),
thus helping libs like libnl-tiny.so (for iw), libjson.so.0 (for ubus,
jshn) to be found
Signed-off-by: Etienne CHAMPETIER <etienne.champetier@free.fr>
Signed-off-by: Florian Fainelli <florian@openwrt.org>
SVN-Revision: 36141
It turns out that the symlink may exists already by the time the toolchain
Makefile gets an opportunity to run. While we asked to replace the target, ln
doesn't do so until after dereferencing the existing symlink. This results in
an unintended symlink that refers to itself. Instead, create the link without
dereferencing any symlinks by using the -n option.
Signed-off-by: John Szakmeister <john@szakmeister.net>
SVN-Revision: 36140
There are no good reasons for keeping this version around when 4.6.3 and
4.6-linaro are there.
Signed-off-by: Luka Perkov <luka@openwrt.org>
SVN-Revision: 35805
Fixes warnings exposed by netifd while using IN6_ARE_ADDR_EQUAL turned
into errors.
Signed-off-by: Florian Fainelli <florian@openwrt.org>
SVN-Revision: 35404
Some of these were copy/pasted from uClibc, but they are not necessary
at all for musl-libc.
Signed-off-by: Florian Fainelli <florian@openwrt.org>
SVN-Revision: 34598
Musl is an alternative C-library, see http://www.musl-libc.org for more infos.
Signed-off-by: Florian Fainelli <florian@openwrt.org>
SVN-Revision: 34314
There are no good reasons for keeping this version around when 4.7.2 and
4.7-linaro are there.
Signed-off-by: Florian Fainelli <florian@openwrt.org>
SVN-Revision: 34296
Turns on -O2 optimization when building GCC thus making the cross-compiler a
little faster (about 25%).
Signed-off-by: Florian Fainelli <florian@openwrt.org>
SVN-Revision: 34266
There is no point in keeping around 2.21 when 2.21.1 which is a bugfix release
is there.
Signed-off-by: Florian Fainelli <florian@openwrt.org>
SVN-Revision: 34265
uClibc declares eventfd() as taking two arguments but doesn't properly
pass the second argument to the kernel.
The problem is discussed at:
http://lists.uclibc.org/pipermail/uclibc/2012-May/046873.html
This patch is taken from uclibc 0.9.33 git, so will presumably
be integrated in any future releases.
Signed-off-by: Andy Leiserson <andy@leiserson.org>
SVN-Revision: 33478
testing uclibc changes is tricky because the final gcc tends to miscompile
uclibc code or barf up internal compiler errors.
install binutils into $(TOOLCHAIN_DIR)/initial (without changing the configure
prefix) and copy it from there to $(TOOLCHAIN_DIR)/ so that the initial gcc
can be put into $(PATH) for the uclibc build, even if the final gcc
is already installed.
SVN-Revision: 32553
insight GDB version 6.8-1 sources have apparently changed.
The original file is no longer available upstream.
Signed-off-by: Daniel Golle <dgolle@allnet.de>
SVN-Revision: 32438
When selecting a specific eglibc version, it comes with a specific SVN
revision that should not be modified as it (more or less) correspond to
a tagged release. This patch disable the possibility to select a specific
SVN revision on known eglib versions.
This patch also disables the possibility to select the trunk branch of
eglibc. There are multiple reasons for that:
* trunk/HEAD may not even compile
* the OpenWrt built system makes using trunk/HEAD a difficult thing, as
OpenWRT fetches the source tree and store it in a compressed tar archive.
Subsequent build get the source from the tar archive - not from SVN,
making the use of trunk/HEAD largelly innefective.
* we cannot know the corresponding version of trunk/HEAD, meaning that
we'll face compiling issues when we'll try to copy the libc files -
unless the build system is fixed with this specific issue in mind.
Signed-off-by: Emmanuel Deloget <logout@free.fr>
SVN-Revision: 31502
