libc: silence warnings about undefined cpp tokens

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

fix race condition when generating linker scripts

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

arm/nptl: Sync unwinding with glibc.

Signed-off-by: Khem Raj <raj.khem@gmail.com>

arm/nptl: sysdep-cancel.h needs syscall handing for EABI

* Synced with eabi version of sysdep-cancel.h from glibc.
  we have been using OABI version which does not work

Signed-off-by: Khem Raj <raj.khem@gmail.com>

MIPS: Fix build error with linuxthreads.old

This patch avoid the following build error:
  AS libc/sysdeps/linux/mips/vfork.os
In file included from libc/sysdeps/linux/mips/vfork.S:22:
./libc/sysdeps/linux/mips/sysdep.h:41:20: error: regdef.h: No such file
or directory
libc/sysdeps/linux/mips/vfork.S: Assembler messages:
libc/sysdeps/linux/mips/vfork.S:40: Warning: Unrecognized register name `sp'
libc/sysdeps/linux/mips/vfork.S:40: Warning: Bad .frame directive
libc/sysdeps/linux/mips/vfork.S:40: Error: junk at end of line, first
unrecognized character is `s'
libc/sysdeps/linux/mips/vfork.S:44: Error: illegal operands `dsubu
sp,FRAMESZ'
libc/sysdeps/linux/mips/vfork.S:45: Error: bad or irreducible absolute
expression
libc/sysdeps/linux/mips/vfork.S:65: Error: illegal operands `daddu
sp,FRAMESZ'
libc/sysdeps/linux/mips/vfork.S:69: Error: illegal operands `li a0,0x4112'
libc/sysdeps/linux/mips/vfork.S:70: Error: illegal operands `move a1,sp'
libc/sysdeps/linux/mips/vfork.S:73: Error: illegal operands `li
v0,(5000+55)'
libc/sysdeps/linux/mips/vfork.S:78: Error: illegal operands `bnez
a3,.Lerror'
libc/sysdeps/linux/mips/vfork.S:88: Error: illegal operands `dla
t9,__syscall_error'
libc/sysdeps/linux/mips/vfork.S:90: Error: illegal operands `jr t9'

Signed-off-by: Abdoulaye Walsimou Gaye <awg@embtoolkit.org>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

avoid using c99 syntax

Signed-off-by: Mirko Vogt <dev@nanl.de>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

get rid of needless c99 construct

Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: i386 needs atomic.h for atomic_exchange_acq

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

nptl: fix order of includes for out-of-tree builds

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

libc: define cfi_personality and cfi_lsda

required for proper, current subarch handling

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

librt: fix linking WRT pthreads

Linking order matters. Always did and still does today.
See how this makes the ugly and wrong hack to put pthread symbols into
librt _and_ linking librt against pthread moot?

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

libc: Fix non-NPTL threads build

This patch fixes some issues building uclibc with linuxthreads, that seem to have
been introduced by the NPTL changes.

Signed-off-by: Andrew Stubbs <ams@codesourcery.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>

librt: simplify handling LDFLAGS

plus a few cosmetic touch-ups

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

libm: typo fix NO_LONG_DOUBLE variant of gamma_r

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

sh4: Fixes for SH-4 without an FPU

This patch disables SH-4 optimizations that rely on the FPU when
building for variants that don't have an FPU, such as SH-4AL.

Signed-off-by: Andrew Stubbs <ams@codesourcery.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>

handle i386 TARGET_SUBARCH

so we can pick the correct nptl locking impls

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

nptl: fix symlinking headers

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

nptl: avoid spurious rebuilds of libpthread*.a

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

nptl: handle PTHREADS_DEBUG_SUPPORT properly

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

nptl: fix race condition when generating libpthread.so

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

hcreate_r: silence warning in helper func

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

wordexp: silence warning

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

nptl: remove relocation in pthread_rwlock_init()

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

silence some warnings about missing prototypes

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

testsuite: disable some tests for !LFS

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

nptl: fix wrong path to nptl_db

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

lxdialog: fix ncursesw include detection

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

buildsys: fix clean vs. headerclean

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

buildsys: conceal install_headers

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

buildsys: fix lib/ directory prerequisite

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

libm: unbreak scalbn for 64-bit processors

Obvious typo.

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

testsuite: .gitignore symlink

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

testsuite: do not run inet tests if !ipv4 && !ipv6

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

lgamma: fix sign handling

add testcase from
http://sourceware.org/bugzilla/show_bug.cgi?id=4407

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

libm: unhide signgam

Fixes signgam checks in math tests on x86_64. It's R_X86_64_COPY for program,
so the program didn't see any changes from libm.

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

do not generate garbled interp

for parallel make

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

nptl: fix for !ipv4 && !ipv6

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

mips/socket.h: add SOCK_CLOEXEC and SOCK_NONBLOCK definitions for mips

* __ss_aligntype is defined to unsigned long int

Signed-off-by: Khem Raj <raj.khem@gmail.com>

strverscmp: I forgot to export it

Result was:

strverscmp.o:
000000ec T __GI_strverscmp

i.e. no plain "strverscmp"!

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ldso/mips: Fix relocation parsing so that they work on both mips32/mips64

Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Rob Landley <rob@landley.net>

inotify: add inotify_init1 system call support

This patch introduces support for inotify_init1 system call, found
since Linux 2.6.27.

Signed-off-by: Vladimir Zapolskiy <vzapolskiy@gmail.com>
Signed-off-by: Khem Raj <raj.khem@gmail.com>

bits/socket.h: add SOCK_CLOEXEC and SOCK_NONBLOCK support

This patch adds support for SOCK_CLOEXEC and SOCK_NONBLOCK socket
descriptor flags, which are introduced since Linux 2.6.27

Signed-off-by: Vladimir Zapolskiy <vzapolskiy@gmail.com>
Signed-off-by: Khem Raj <raj.khem@gmail.com>

utils: Fix ldd segfault

Fixes ldd segfault on this testcase:

extern void _dl_getenv(void);
void foo(void)
{
    printf("foo: %x\n", &_dl_getenv);
}

linked as -shared

It fixes bug #1507

Signed-off-by: Natanael Copa <ncopa@alpinelinux.org>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>

mips: Add vfork to libc

* When using NPTL get clone.o from nptl.
* Only use vfork if syscall is there.
* Add libc_a_SSRC to LINUX_LIBC_ARCH_OBJ.
* Use CLEAN_* for clean target.

Signed-off-by: Khem Raj <raj.khem@gmail.com>

mips/bits/pthreadtypes.h: Re-include sgidefs.h

* This was dropped accidently in last merge
  removal meant a lot of warnings about undefined
  _ABI64.

Signed-off-by: Khem Raj <raj.khem@gmail.com>

static build: define pthreads wrappers only for threaded builds

Otherwise we get compiler errors due to undefined prototypes
as noticed by Denys.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

some (all?) archs don't define asm, so use builtin __asm__

Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

Fix fcntl64 for 64 bit targets.

64 bit targets often don't have a separate fcntl64() system call, because they don't need one.

Signed-off-by: Rob Landley <rob@landley.net>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

workaround GCC PR32219

we ended up calling 0
Fixes bug #1033

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

resolv.h: fix wring indentation of #ifdef/#endif pairs

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

utils/Makefile.in: Define GETCONF_DIR for host builds of getconf.

Signed-off-by: Khem Raj <raj.khem@gmail.com>

powerpc/Makefile.arch: Add objclean and headers_clean targets for NPTL.

Signed-off-by: Khem Raj <raj.khem@gmail.com>

powerpc: Add TLS and NPTL support

Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

mips/Makefile.arch: Remove -f option to $(do_awk)

Signed-off-by: Khem Raj <raj.khem@gmail.com>

resolv: various memory corruption and off by one fixes

Fixes resolution of names with AAAA entries and gethostbyaddr issues.

Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

static build: fix internal locking weaks to get pulled in always

Linker is smart and does not pull in weaks.os, ever. This happens
because that compilation unit does not get strong references and
ld eliminates dead code. We really need the weaks for static build
in a compilation unit that is always there, otherwise it won't work.

Signed-off-by: Timo Teras <timo.teras@iki.fi>
Acked-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

ldso: fix x86_64 R_X86_64_TPOFF64 and R_X86_64_DTPOFF64 relocations

R_X86_64_TPOFF64 revealed by trivial testcase:
===================================================================
 #include <stdio.h>
 #include <errno.h>

int main() {
        void *a = &errno;

        printf("errno addr: %llx\n", a);
        __asm__("movq    errno@gottpoff(%%rip), %0;\n"
                "add    %%fs:0x0,%0;" : "=r"(a) );
        printf("got errno addr: %llx\n", a);

        return 0;
}
===================================================================

The addresses application got with R_X86_64_TPOFF64 was different than the once libc
internal __errno_location returned.

R_X86_64_DTPOFF64 testcase is even simpler than that:
===================================================================
 #include <stdio.h>
 #include <errno.h>
 #include <netdb.h>
 #undef h_errno

extern __thread int h_errno;

int main() {
        printf("h_errno addr: %llx\n", &h_errno);
        printf("__h_errno_location addr: %llx\n", __h_errno_location());
        return 0;
}
===================================================================

but needs to be linked with "-lpthread". This way we've got h_errno relocation via
R_X86_64_TPOFF64 in application and h_errno relocation via R_X86_64_DTPOFF64 in
libpthread which has its own __h_errno_location() (probably we can kill it later?).
And addresses were different again.

The problem is that both relocations resolve symbols in external modules and thus
should use symbol_addr instead of sym->st_value.

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

Fix gen_wc8bit support for finding a UTF8 locale on host.

gen_wc8bit was previously patched to support UTF8 locales other than the
default en_US.UTF-8 by patch id 0e14b8f425c562749d3821a42414b1ce6b3b0923.

This was incomplete in its implementation and still failed where item(s)
in the list returned by 'locale -a' already had 'UTF8' appended to the
name. This patch rectifies this situation, plus adds an extra fix to
avoid trimming '\n' from an empty line which could cause a segfault.

Signed-off-by: Andy Gibbs <andyg1001@hotmail.co.uk>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

nptl: use arch-specific pthread_cond_timedwait and pthread_cond_wait on x86_64

Generic ones were broken on x86_64, pthread/ex2 test failed while pthread/ex5
worked fine. So use arch-specific functions and fix linkage failure with
__pthread_mutex_cond_lock_adjust.

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: use arch-specific cancellation routines for x86_64

They were present in sources, but not compiled in. Generic routines are
broken badly on x86_64, even basic threading test like pthread/ex1 didn't
work with them.

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: fix sem_open with O_CREAT

Temporary file name template passed to __gen_tempname had no "XXXXXX" in it,
so __gen_tempname returned EINVAL which led to sem_open failure.

Fixes NPTL tests tst-sem4, tst-sem7, tst-sem8, tst-sem9.

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

time.c: fix unreadable code style: assignments within if(), misplaced comments

No logic changes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

optional /etc/localtime support

This patch teaches uclibc to use /etc/localtime.
This is possible since recent TZif2 file format contains
TZ-like ASCII string at the end.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

include dlfcn.h in ldso.h to get RTLD_NODELETE definition

Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

fix a few build errors for the no threads case

Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

Fix typo (OPTIMIZATIONS -> OPTIMIZATION)

Signed-off-by: Tony Wu <tung7970@gmail.com>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

MIPS: restore INLINE_SYSCALL macro

The MIPS specific INLINE_SYSCALL macro has been renamed to
INLINE_SYSCALL_NCS with:

  763bbf9e9a27426c9be8322dca5ddf2cb4dbc464

  syscall: unify part 2: NCS variety

  Declare common NCS (non-constant syscall) variants and convert the
  existing ports over to this.

This change breaks system calls. The code generated with using of the
new macro does not obey the restartable syscall convention used by the
linux kernel. When it tries to restart the syscall the errno value is
not being replaced by the syscall number.

This causes weird behaviour of the 'ping' command in busybox for
example:

root@OpenWrt:/# ping 192.168.1.254
PING 192.168.1.254 (192.168.1.254): 56 data bytes
64 bytes from 192.168.1.254: seq=0 ttl=128 time=6.292 ms
ping: recvfrom: Function not implemented
64 bytes from 192.168.1.254: seq=1 ttl=128 time=0.719 ms
ping: recvfrom: Function not implemented
64 bytes from 192.168.1.254: seq=2 ttl=128 time=0.489 ms
ping: recvfrom: Function not implemented
64 bytes from 192.168.1.254: seq=3 ttl=128 time=0.486 ms
ping: recvfrom: Function not implemented
64 bytes from 192.168.1.254: seq=4 ttl=128 time=0.487 ms
ping: recvfrom: Function not implemented
64 bytes from 192.168.1.254: seq=5 ttl=128 time=0.939 ms
ping: recvfrom: Function not implemented
64 bytes from 192.168.1.254: seq=6 ttl=128 time=0.971 ms
ping: recvfrom: Function not implemented
64 bytes from 192.168.1.254: seq=7 ttl=128 time=0.488 ms
ping: recvfrom: Funct^C

Signed-off-by: Gabor Juhos <juhosg@openwrt.org>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: mips needs updated clone() implementation

Signed-off-by: Andreas Schultz <andreas.schultz@gmail.com>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

libc/x86_64: use ELF-compatible definitions for assembler

Fixes open/read/write/close breakage with NPTL on x86_64 (due to missing
'.type' declaration).

Done this way because there are already ELF-style definitions in assemebler
code for x86_64 and looks like HAVE_ELF is not defined and used this days.
Or the other way around, it'd be a bit strange for x86_64 not to have ELF.

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

libc/regex: fix internal_function declarations

Build failed with

| In file included from libc/misc/regex/regex.c:59:
| libc/misc/regex/regexec.c:315: error: conflicting types for 're_search_2_stub'
| libc/misc/regex/regexec.c:44: note: previous declaration of 're_search_2_stub' was here
| libc/misc/regex/regexec.c:358: error: conflicting types for 're_search_stub'
| libc/misc/regex/regexec.c:49: note: previous declaration of 're_search_stub' was here
.... etc

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

libpthread/x86_64: fix clean

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

enable building __fcntl_nocancel, as it used unconditionally now

also sync up not-cancel.h headers between all three thread libraries

Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

Makefile.in: install sgtty.h

Proper and useful header, for example used by pacemaker.

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: Harmonize objects and headers makefile clean targets

Update all the nptl clean targets (both objects and headers) to
use the new syntax using CLEAN_<full_path> and HEADER_<full_path>.

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>

build: Fix infinite loop when no threading support is enabled

The build loops forever due to when invoking make $(pregen-headers-y)
and $(pregen-headers-y) is empty (no thread support).
Parallel build should continue to work as well.
It fixes Bugzilla #1633.

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>

libpthread/nptl_db: define NOT_IN_libc

NPTL build for x86_64 failed with

  LD libthread_db-0.9.32-git.so
  .../ld: libpthread/nptl_db/libthread_db_so.a(td_ta_event_getmsg.oS): relocation R_X86_64_PC32 against undefined symbol `__GI_memset' can not be used when making a shared object; recompile with -fPIC

Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl_db: Remove compiler warning due to old-style function definition

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>

nptl: Remove compiler warning due to old-style function definition

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>

libc: Remove compiler warning due to old-style function definition

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>

change ABI_VERSION back to just include the major number

By default, GCC hard codes the name for ld.so to /lib/ld-uClibc.so.0
so our default should match that. If distribution makers want to change
it, they need to change the default gcc configuration as well.

Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

Merge commit 'origin/master' into nptl

Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

utils: fix install prefix for getconf

So we don't get /usr/usr/bin/getconf when DEVEL_PREFIX="/usr"

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

nptl: fix dynamic initialization of libpthread

If libpthread get's pulled in via dlopen(), we need to do libpthread
initialization in dlopen(). Achieve this by making initializer function
out of __pthread_initialize_minimal_internal. Add the proper linker
flags and make it callable multiple times.

Add also nodelete flag which ensures that libpthread will not get
unmapped after it's been loading. Though, ld.so does not yet
support this.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: fix libc internal, dynamically enabled locking

Final iteration to fix libc internal locking if libpthread is pulled
in by dlopen call (directly or indirectly).

We cannot really use the weak symbol trick for shared build, since
the symbols won't get refreshed if libpthread is pulled in dynamically.
In glibc, they have #ifdef SHARED magic to either use pthread_functions
table, or weaks. But as we shared object files with both builds, this
does not sounds good either.

The reintroduces the libc weaks.c, but uses them now only with static
build. For dynamic build, we still use the symbols with same name, but
provide weaks in forward.c so they end up dereferencing the
pthread_functions table indirectly if we are not linked to libpthread.
Mutex initialization is hard coded as inline, as it needs to happen even
if libpthread is not initially loaded.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

ldso: support RTLD_NODELETE and DF_1_NODELETE

Honor the nodelete flags so we don't delete shared library if it's
sticky. This is useful for libpthread if it gets pulled in by a
dlopen'ed library.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: fix warnings of shadowing __self

Stdio locking macroes do:
  void *__self = THREAD_SELF;

But THREAD_SELF uses __self also internally which causes shadowing
warnings. Just rename the outer variable for now. Might be an idea
to convert the macroes to static inline functions.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: fix malloc library locking

Update malloc library to use internal uclibc locking primitives
to get the libpthread calls correct.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: proper soname handling

Since sublevel releases are not ABI compatible we need to adjust
the soname to include the sublevel version.

This makes it possible to install ABI incompatible versions of the
library side by side so clean upgrades are possible.

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: remove unneeded libc-lock.h include

I semi-accidentally added include for libc-lock.h because the intention
was to use locking macroes from there. However, we ended up using using
the weak alias stuff. This is additionally good since now this can result
in compile errors like:

In file included from /toolchain/include/bits/uClibc_mutex.h:16,
                 from /toolchain/include/bits/uClibc_stdio.h:107,
                 from /toolchain/include/stdio.h:72,
                 from /work/freetype-2.3.11/include/freetype/config/ftstdlib.h:100,
                 from ./builds/unix/ftconfig.h:43,
                 from builds/unix/ftsystem.c:21:
/toolchain/include/bits/libc-lock.h:309: error: expected '=', ',',
';', 'asm' or '__attribute__' before 'void'

Remove the unneeded include causing breakage.

Reported-by: Kevin Day <thekevinday@gmail.com>
Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: mark forwarded functions protected

The libc stubs should never override the libpthread provided
functions inside libpthread, otherwise the struct pthread_functions
that is used by the forwarder stubs goes bad.

Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

libc: remove libc weak __pthreads_* wrappers

It is not possible to override for libpthread to override
the weak libc definitions. This has never worked in uclibc, and
does no longer work in glibc either (unless you use LD_DYNAMIC_WEAK).

The proper thing to do is have weak prototypes in libc, and
definitions in libpthread only. This way libc runs even if
those functions are not defined, but just needs to protect
against the NULL values (done by implementing __uclibc_maybe_call).

This fix the problems if libc is linked before libpthread or
if libpthread is pulled by a dependency library.

Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

libc-symbols: add attribute_protected

Definition to use protected visibility.

Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

ldso/i386: support protected symbols

Make sure protected symbols are not overridden by other modules.

Signed-off-by: Joakim Tjernlund <joakim.tjernlund@transmode.se>
Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

Revert "nptl: mark symbols with libc forwarder hidden"

This reverts most of commit 1610762362e651f86ca284ac59a1d7ec88034e4e.

Marking the symbols hidden is wrong, because libc accesses
the pthreads internal symbols for internal locking.

Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: fix libc sigtimedwait

It seems that 57e8823548ad6e65d33b2153edeb18fb0edc20e6 removed completely
sigtimedwait symbol from libc which is wrong. I hope there is not too
many other things like this.

Apparently the libc_hidden_* macros actually make previously hidden
symbols visible globally (creates alias from __GI_* to *). This is
probably ancient confusion from times when gcc did not support
visibility attribute and hiding symbols was done using hacks like
this.

This also adds attribute_hidden to the internal __sigtimedwait for
nptl case.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: fix libc sigaction signal checking

We should not check for SIGCANCEL in __libc_sigaction because nptl calls this
function to setup this signal. Nptl provides it's own override for sigaction
that checks that the user cannot override signals nptl uses internally.
Linuxthreads does not use SIGCANCEL at all so this affects nptl only.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

PTR_DEMANGLE: add documentation

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

Merge commit 'origin/master' into nptl

Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

nptl: turn pointer demangling off

Currently the stub calls would
demangle the pointer that is never mangled. Alternatively, we could sync
nptl/sysdeps/unix/sysv/linux/libc_pthread_init.c with new glibc version
that properly mangles the pointers, but that would need more validation
to make sure that all struct pthread_function readers do demangling.
uclibc currently does not do pointer mangling everywhere so it's more
consistent to just disable it here too.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>

linuxthreads.new: initialize stdio locking

uClibc requires the threading library to enable locking for
stdio, or the locking is not done at all.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>