| Age | Commit message (Collapse) | Author |
|---|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
No logic changes
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
|
|
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>
|
|
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
|
|
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
|
|
Signed-off-by: Tony Wu <tung7970@gmail.com>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
|
|
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>
|
|
Signed-off-by: Andreas Schultz <andreas.schultz@gmail.com>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
|
|
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>
|
|
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>
|
|
Signed-off-by: Roman I Khimov <khimov@altell.ru>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
|
|
also sync up not-cancel.h headers between all three thread libraries
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
|
|
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
|
|
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
|
|
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>
|
|
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
Definition to use protected visibility.
Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
|
|
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
|
|
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>
|
|
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>
|
|
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>
|
|
Add attribute_hidden to all symbols having libc forwarder. This prevents
recursive self calls which would happen if libc is before libpthread in
linking order: the forwarder functions would call itself via the function
table, since the libpthread symbols would get overwritten with libc ones.
This has not been a problem in glibc since there these symbols are marked
hidden with linker version-script. Since we don't use one, we need to mark
these explicitly.
Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
|
|
allows 'make realclean && make utils' to behave properly
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
|
|
On hardened system it could be useful to disable the use
of LD_PRELOAD environment variable for preloading shared objects
before the system libraries. So this patch add a config option,
LDSO_PRELOAD_ENV_SUPPORT, to control this behaviour.
It is enabled by default.
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
|
|
part1; Needs improvement:
Think about a sensible way to map feature sets to the mandated syms.
Our LEGACY stuff maps to OB, OBXSI for example.
A "normal" config should satisfy the BASE (i.e. $foo.SUSv4.syms)
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
|
|
i386 was missing a prototype for __tls_get_addr()
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
|
