OSDN Git Service

uclinux-h8/uClibc.git
13 years agolibm_sh: Move fenv functions into sh4 sub-folder
Carmelo Amoroso [Fri, 17 Dec 2010 07:11:12 +0000 (08:11 +0100)]
libm_sh: Move fenv functions into sh4 sub-folder

Indeed fenv funxtions are SH4 specific, so move them into
an SH4 specific sub-folder.

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agosh_fpu: Do not allow inclusion of fpu_control on not SH4 core
Carmelo Amoroso [Wed, 15 Dec 2010 17:00:16 +0000 (18:00 +0100)]
sh_fpu: Do not allow inclusion of fpu_control on not SH4 core

Instead of emitting a warning even when compiling for SH4, stop
compilation with an #error directive when not on SH4 core.

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agosh: fix memcpy saving/restoring FR12-FR15 registers
Giuseppe Cavallaro [Mon, 13 Dec 2010 10:39:36 +0000 (11:39 +0100)]
sh: fix memcpy saving/restoring FR12-FR15 registers

This patch fixes a bug in the memcpy that doesn't save/restore
the FR12-FR15 registers (callee save registers in ST40 ABI) while
copying many cache lines with FPU in single paired precision mode
and by using all FPU registers (DR and XD).

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Reviewed-by: Carmelo Amoroso <carmelo.amoroso@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agonptl: fix Unwind_Resume PLT calls
Timo Teräs [Tue, 7 Dec 2010 09:27:29 +0000 (11:27 +0200)]
nptl: fix Unwind_Resume PLT calls

My change a49b3a18e463cbe8c94c41501e386e7f4c61609e fixed two
Unwind_Resume calls to go via PLT to avoid text relocations for PIC
builds. However, it looks the reason for upstream not using PLT calls
is that ebx gets clobbered. So we need to reload it.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
13 years agogetconf: only use specs that are available
Thomas Petazzoni [Sat, 4 Dec 2010 09:29:42 +0000 (10:29 +0100)]
getconf: only use specs that are available

On older glibc (like 2.7), the _SC_V7_* symbols are not defined,
making the build of current uClibc fail on Debian Lenny systems and
other systems using a fairly old glibc. So we make sure to only use
the specs names that are defined by the host C library.

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agocreate DEVEL_PREFIX/MULTILIB_DIR dir rather than DEVEL_PREFIX/lib
Natanael Copa [Tue, 24 Aug 2010 08:19:27 +0000 (08:19 +0000)]
create DEVEL_PREFIX/MULTILIB_DIR dir rather than DEVEL_PREFIX/lib

This fixes issue with GNU Make 3.82 when running:

 make install DESTDIR=$someplace

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Khem Raj <raj.khem@gmail.com>
13 years agoConfig.sparc: remove warning about shared libs on v7
Austin Foxley [Wed, 1 Dec 2010 23:32:41 +0000 (15:32 -0800)]
Config.sparc: remove warning about shared libs on v7

Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
13 years agolibc: fix strtoq
Natanael Copa [Fri, 26 Nov 2010 14:49:56 +0000 (14:49 +0000)]
libc: fix strtoq

strtoq should always return a quad_t and be an alias of strtol on
64 bit and strtoll on 32 bit.

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agolibc_sh: add __tls_get_addr to static libc (SH)
Filippo Arcidiacono [Mon, 29 Nov 2010 22:31:49 +0000 (23:31 +0100)]
libc_sh: add __tls_get_addr to static libc (SH)

Add __tls_get_addr function to static libc allowing to handle
local dynamic TLS access model.
On SH, linker optimizations are not required, so __tls_get_addr
can be called even in statically linked binaries.

Signed-off-by: Filippo Arcidiacono <filippo.arcidiacono@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agoRevert "libc_tls: Do not include generic libc-tls from arch specific implementation"
Carmelo Amoroso [Wed, 1 Dec 2010 16:04:05 +0000 (17:04 +0100)]
Revert "libc_tls: Do not include generic libc-tls from arch specific implementation"

This reverts commit 0c468129356f5ce6780e7369c271d74631ae03ba.
Indeed recent updates into buildsys managed to filter-out generic implementation
when arch specific one is defined, so duplicated symbols into the libc.a

In the meanwhile, do a minor fix into mips version not using relative path
to include <generic/libc-tls.c>

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agolibc_tls: Do not include generic libc-tls from arch specific implementation
Carmelo Amoroso [Mon, 29 Nov 2010 21:21:52 +0000 (22:21 +0100)]
libc_tls: Do not include generic libc-tls from arch specific implementation

libpthread/nptl/sysdeps/generic/libc-tls.c is built as well, so it does not need
that arch specific version of libc-tls.c (MIPS and ALPHA) includes it.
The arch libc-tls.s is aimed to provide the implementation of __tls_get_addr
for the static libc.a, because on these archs the linker relaxations are not
required and it could be possible to have local-dynamic access models in static
libraries as well.

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agolibc_utmp: Fix getutmp and getutmpx for x86_64
Carmelo Amoroso [Sun, 28 Nov 2010 19:44:21 +0000 (20:44 +0100)]
libc_utmp: Fix getutmp and getutmpx for x86_64

On x86_64, when WORDSIZE_COMPAT32 is enabled, the ut_tv field
of 'struct utmp' and 'struct utmpx' are defined as two nested structs
instead of being defined as 'struct timeval', so it is not possible to
directly assign the two ut_tv values.
This patch split the assignment by setting each fields (tv_sec, tv_usec)
of the ut_tv filed separately
It is als compatible with 'struct timeval' usage.

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agoRevert "ldso: silence warning about unused tls var if !tls"
Bernhard Reutner-Fischer [Wed, 24 Nov 2010 22:03:03 +0000 (23:03 +0100)]
Revert "ldso: silence warning about unused tls var if !tls"

This reverts commit 23fa805150d573a913cad69d34c06f3b2ce54270.

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agoConfig: default ia32 to i486
Bernhard Reutner-Fischer [Wed, 24 Nov 2010 22:02:11 +0000 (23:02 +0100)]
Config: default ia32 to i486

NPTL requires i486 on ia32 since i386 doesn't provide required lll

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agolibc: silence warning
Bernhard Reutner-Fischer [Wed, 24 Nov 2010 21:46:02 +0000 (22:46 +0100)]
libc: silence warning

about implicit declaration of memset in system()

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agoldso: silence warning about unused tls var if !tls
Bernhard Reutner-Fischer [Wed, 24 Nov 2010 21:33:01 +0000 (22:33 +0100)]
ldso: silence warning about unused tls var if !tls

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agoconfig_parser: hide symbols
Bernhard Reutner-Fischer [Wed, 24 Nov 2010 20:06:38 +0000 (21:06 +0100)]
config_parser: hide symbols

Our impl has diverged from busybox' so make sure nobody else uses it.

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agotest: ignore some more binaries
Bernhard Reutner-Fischer [Wed, 24 Nov 2010 19:54:51 +0000 (20:54 +0100)]
test: ignore some more binaries

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agowchar: bug #1471: fix cornercase in mbrtowc
Bernhard Reutner-Fischer [Wed, 24 Nov 2010 19:51:46 +0000 (20:51 +0100)]
wchar: bug #1471: fix cornercase in mbrtowc

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agobuildsys: add make {,install_}startfiles
Bernhard Reutner-Fischer [Wed, 24 Nov 2010 19:21:58 +0000 (20:21 +0100)]
buildsys: add make {,install_}startfiles

This builds and installs the crt files.

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agotest/math: bessel depend on XSI_MATH
Bernhard Reutner-Fischer [Wed, 24 Nov 2010 19:20:47 +0000 (20:20 +0100)]
test/math: bessel depend on XSI_MATH

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agoresolv.c fails for /etc/hosts lookups
Philip Nye [Fri, 19 Nov 2010 16:50:35 +0000 (16:50 +0000)]
resolv.c fails for /etc/hosts lookups

Patch attached:

Fix a bug in offset calculations when parsing /etc/hosts in resolv.c.

Formerly a miscalculation meant that having found the correct line, the code
was trashing its own result data.

Signed-off-by: Philip Nye <philipn@engarts.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agoDisables the automatic disable of shared library for sparc-v7.
Konrad Eisele [Thu, 18 Nov 2010 18:32:37 +0000 (10:32 -0800)]
Disables the automatic disable of shared library for sparc-v7.

I dont see a dependency between shared library and v7.
The difference between v7 and v8 is that in v8 you have
the added math asm-insns: smul/umul/udiv/sdiv/urem.
It shouldnt affect shared libraries creation.

Signed-off-by: Konrad Eisele <konrad@gaisler.com>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
13 years agoAdding quad float gcc intrinsic implementation for SPARC
Austin Foxley [Thu, 18 Nov 2010 18:20:43 +0000 (10:20 -0800)]
Adding quad float gcc intrinsic implementation for SPARC

Sparc-Gcc generates q_xxx intrinsic calls for quad float ("long double")
code. The routines from glibc's glibc-2.9/sysdeps/sparc/sparc32/soft-fp/*
where taken and ported to uclibc.

Signed-off-by: Konrad Eisele <konrad@gaisler.com>
Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
13 years agontpl: fix static linking by not leaking SSP_ALL_CFLAGS
Natanael Copa [Thu, 11 Nov 2010 15:54:19 +0000 (15:54 +0000)]
ntpl: fix static linking by not leaking SSP_ALL_CFLAGS

The SSP_ALL_CFLAGS in nptl arch CFLAGS leaks out and forces things
like dl-support.c, brk.c, sbrk.c memcpy, etc to be built with
-fstack-protector-all. This is bad when linking statically since
initializing TLS will call those functions before SSP is initialized.

The libpthread itself will still be built with -fstack-protector-all
due to CFLAGS-nptl has SSP_ALL_CFLAGS in libpthread/nptl/Makefile.in

Thanks to Timo Teras for helping with this.

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agonptl: use correct unwind header for unwind-forcedunwind
Bernhard Reutner-Fischer [Wed, 10 Nov 2010 19:10:46 +0000 (20:10 +0100)]
nptl: use correct unwind header for unwind-forcedunwind

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomicroblaze: build infrastructure
Steven J. Magnani [Wed, 10 Nov 2010 18:41:45 +0000 (19:41 +0100)]
microblaze: build infrastructure

Fix Microblaze config and makefiles.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomicroblaze: thread support
Steven J. Magnani [Wed, 10 Nov 2010 18:37:26 +0000 (19:37 +0100)]
microblaze: thread support

Header files needed to build linuxthreads.old for microblaze.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomicroblaze: optimized memcpy/memmove
Steven J. Magnani [Wed, 10 Nov 2010 18:33:52 +0000 (19:33 +0100)]
microblaze: optimized memcpy/memmove

Port optimized memcpy/memmove from the kernel.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomicroblaze: misc headers
Steven J. Magnani [Wed, 10 Nov 2010 18:32:27 +0000 (19:32 +0100)]
microblaze: misc headers

Other header files for uClibc compilation.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomicroblaze: kernel headers
Steven J. Magnani [Wed, 10 Nov 2010 18:32:13 +0000 (19:32 +0100)]
microblaze: kernel headers

Sync kernel header definitions with those used in recent mainline kernels.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomicroblaze: Restore soft-float
Steven J. Magnani [Wed, 10 Nov 2010 18:31:11 +0000 (19:31 +0100)]
microblaze: Restore soft-float

Software floating point for microblaze.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomicroblaze: vfork/clone interface
Steven J. Magnani [Wed, 10 Nov 2010 18:30:02 +0000 (19:30 +0100)]
microblaze: vfork/clone interface

Fix the microblaze vfork() and clone() implementations.
Add support for clone2().

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomicroblaze: syscall interface
Steven J. Magnani [Wed, 10 Nov 2010 18:27:57 +0000 (19:27 +0100)]
microblaze: syscall interface

Fix the microblaze syscall interface.
Recent mainline kernels no longer carry userland code to invoke syscalls
and microblaze can use the uClibc generic code.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomicroblaze: Fix crt bootstrap
Steven J. Magnani [Wed, 10 Nov 2010 18:25:30 +0000 (19:25 +0100)]
microblaze: Fix crt bootstrap

Rework crt bootstrap to work with the new __uClibc_main(),
and hardwire crt init/fini code since the awk approach to generating it on
the fly doesn't work for microblaze. The output from the gcc 4.1.2 compiler
is scrambled so that the tags expected by the awk script to bracket the
init/fini entry and exit code no longer do.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomicroblaze: Fix setjmp/longjmp
Steven J. Magnani [Wed, 10 Nov 2010 18:23:45 +0000 (19:23 +0100)]
microblaze: Fix setjmp/longjmp

Some state and registers are missing from setjmp/longjmp handling.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agoldso/include/dl-hash.h: suppress compiler warning. no code changes
Denys Vlasenko [Tue, 2 Nov 2010 06:44:35 +0000 (07:44 +0100)]
ldso/include/dl-hash.h: suppress compiler warning. no code changes

The warning was repeated for every .c file which is including
this header:

./ldso/include/dl-hash.h: In function '_dl_find_hash':
./ldso/include/dl-hash.h:150: warning: unused parameter 'tpntp'

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agoldouble_wrappers: make code less verbose via WRAPPER2 macro
Denys Vlasenko [Tue, 2 Nov 2010 06:35:01 +0000 (07:35 +0100)]
ldouble_wrappers: make code less verbose via WRAPPER2 macro

Tested: ran testsuite

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agoldouble_wrappers: remove i386-specific optimization
Denys Vlasenko [Tue, 2 Nov 2010 06:24:50 +0000 (07:24 +0100)]
ldouble_wrappers: remove i386-specific optimization

It is deemed too unsafe. Quoting Timo:

If I'm building with "-fPIC -pg" it instruments all C functions with
profiler stuff which is called via PLT and causes EBX reloads
 --> crash

-fno-omit-frame-pointer is sometimes useful for profiling too
 --> crash

Also the upcoming -fsplit-stack will be broken by this too (that might
need additional uclibc support though).

And I'm pretty sure there's also other similar compiler features.
There's no predefined #defines in gcc for any of these.

What I'm trying to say that there are *numerous* situations when the
compiler can create stack frame for you without you ever knowing it. And
if you want to do a tail jump, you really should be doing it from .S
file where you control fully the prologue/epilogue code. (GCC naked
attribute does not seem to work on x86.)

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: disable tail jump trick if __SSP_ALL__ (was using wrong conditional)
Denys Vlasenko [Sun, 31 Oct 2010 14:58:18 +0000 (15:58 +0100)]
libm: disable tail jump trick if __SSP_ALL__ (was using wrong conditional)

Thanks, Timo!

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: fix tgamma to actually do return true gamma function
Denys Vlasenko [Sun, 31 Oct 2010 03:36:02 +0000 (04:36 +0100)]
libm: fix tgamma to actually do return true gamma function

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: improve readability of math.h; expand comments and docs
Denys Vlasenko [Sun, 31 Oct 2010 02:03:38 +0000 (03:03 +0100)]
libm: improve readability of math.h; expand comments and docs

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: remove int_WRAPPER_C99 macro, add test which checks that I did not break it
Denys Vlasenko [Sun, 31 Oct 2010 02:02:00 +0000 (03:02 +0100)]
libm: remove int_WRAPPER_C99 macro, add test which checks that I did not break it

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: revert wrong ldouble wrapper hack "fix"
Denys Vlasenko [Sat, 30 Oct 2010 23:42:08 +0000 (01:42 +0200)]
libm: revert wrong ldouble wrapper hack "fix"

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: fix false positives in double tests
Denys Vlasenko [Sat, 30 Oct 2010 23:23:34 +0000 (01:23 +0200)]
libm: fix false positives in double tests

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: fix remaining failures in test-float
Denys Vlasenko [Sat, 30 Oct 2010 22:45:06 +0000 (00:45 +0200)]
libm: fix remaining failures in test-float

These two failures are fixed:
Failure: Test: ilogb (0.0) == FP_ILOGB0 plus exceptions allowed
Failure: Test: ilogb (NaN) == FP_ILOGBNAN plus exceptions allowed

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: fix rint/scalb testcase failures
Denys Vlasenko [Sat, 30 Oct 2010 21:45:41 +0000 (23:45 +0200)]
libm: fix rint/scalb testcase failures

These failures no longer happen:

Failure: Test: scalb (2.0, 0.5) == NaN plus invalid exception
Failure: Test: scalb (3.0, -2.5) == NaN plus invalid exception
Failure: Test: rint (0.5) == 0.0
Failure: Test: rint (1.5) == 2.0
Failure: Test: rint (2.5) == 2.0
Failure: Test: rint (3.5) == 4.0
Failure: Test: rint (4.5) == 4.0
Failure: Test: rint (-0.5) == -0.0
Failure: Test: rint (-1.5) == -2.0
Failure: Test: rint (-2.5) == -2.0
Failure: Test: rint (-3.5) == -4.0
Failure: Test: rint (-4.5) == -4.0

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: fix powf testcase failures
Denys Vlasenko [Sat, 30 Oct 2010 19:25:35 +0000 (21:25 +0200)]
libm: fix powf testcase failures

Fixed failures:
Failure: Test: pow (1, NaN) == 1
Failure: Test: pow (1, inf) == 1
Failure: Test: pow (-1, inf) == 1
Failure: Test: pow (1, -inf) == 1
Failure: Test: pow (-1, -inf) == 1

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: fix scalb testsuite failures
Denys Vlasenko [Sat, 30 Oct 2010 18:30:14 +0000 (20:30 +0200)]
libm: fix scalb testsuite failures

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agolibm: fix testuite failure in modf
Denys Vlasenko [Sat, 30 Oct 2010 18:19:20 +0000 (20:19 +0200)]
libm: fix testuite failure in modf

This error is gone:
Failure: Test: modf (NaN, &x) == NaN
Result:
 is:         -0.00000000000000000000e+00  -0
 should be:   nan   nan

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agotest/math/*: unbreak
Denys Vlasenko [Sat, 30 Oct 2010 17:40:35 +0000 (19:40 +0200)]
test/math/*: unbreak

test/math was not running at all, presume it has bit rotted:

It was trying to run libm-test.c as if it is a binary.

It was looking for libm-test-ulps (file with allowed
errors in lower bits) in arch-specific dirs which do not exist
in uclibc, as a result wrong file (libm-test.inc) was used instead.

Test failure was not showing the error result, user had to
fish it out from some .out files.

I added libm-test-ulps-ARCH files from recent glibc,
this filtered out a lot of false positives.
For example, cosf(M_PI_6l * 4.0) ideally should be -0.5,
we are getting -0.50000005047356477217, and this isn't
a failure (the difference is one lowest bit of mantissa).

"make check UCLIBC_ONLY=1 VERBOSE=1" still fails,
but not as catastrophically as before.

For the record, the failure occurs on the stage where we check
32-bit float functions, these tests fail:

Failure: Test: modf (NaN, &x) == NaN
Failure: Test: ilogb (NaN) == FP_ILOGBNAN plus exceptions allowed
Failure: Test: scalb (2.0, 0.5) == NaN plus invalid exception
Failure: Test: scalb (3.0, -2.5) == NaN plus invalid exception
Failure: Test: scalb (0, NaN) == NaN
Failure: Test: scalb (1, NaN) == NaN
Failure: Test: scalb (0, inf) == NaN plus invalid exception
Failure: Test: scalb (-0, inf) == NaN plus invalid exception
Failure: Test: scalb (1, inf) == inf
Failure: Test: scalb (-1, inf) == -inf
Failure: Test: scalb (inf, -inf) == NaN plus invalid exception
Failure: Test: scalb (-inf, -inf) == NaN plus invalid exception
Failure: Test: scalb (1, NaN) == NaN
Failure: Test: scalb (0, NaN) == NaN
Failure: Test: scalb (inf, NaN) == NaN
Failure: Test: pow (1, NaN) == 1
Failure: Test: pow (1, inf) == 1
Failure: Test: pow (-1, inf) == 1
Failure: Test: pow (1, -inf) == 1
Failure: Test: pow (-1, -inf) == 1
Failure: Test: rint (0.5) == 0.0
Failure: Test: rint (1.5) == 2.0
Failure: Test: rint (2.5) == 2.0
Failure: Test: rint (3.5) == 4.0
Failure: Test: rint (4.5) == 4.0
Failure: Test: rint (-0.5) == -0.0
Failure: Test: rint (-1.5) == -2.0
Failure: Test: rint (-2.5) == -2.0
Failure: Test: rint (-3.5) == -4.0
Failure: Test: rint (-4.5) == -4.0

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agosleep: include Linus' email in the comment
Denys Vlasenko [Sat, 30 Oct 2010 13:06:20 +0000 (15:06 +0200)]
sleep: include Linus' email in the comment

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agosleep: document testing result on 2.4.x kernels
Denys Vlasenko [Fri, 29 Oct 2010 02:01:27 +0000 (04:01 +0200)]
sleep: document testing result on 2.4.x kernels

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agosleep: add comment with test program for SIG_IGNed SIGCHLD
Denys Vlasenko [Fri, 29 Oct 2010 01:44:57 +0000 (03:44 +0200)]
sleep: add comment with test program for SIG_IGNed SIGCHLD

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years agoFix msync() redefinition for noMMU build
Bernhard Reutner-Fischer [Wed, 27 Oct 2010 20:18:23 +0000 (22:18 +0200)]
Fix msync() redefinition for noMMU build

An architecture that _may_ have a MMU (and thus implements
the msync syscall), but for which the kernel and uClibc are built as
noMMU, we have a stub for msync.

Thanks to Steven J. Magnani for noticing.

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agomips: fix errno setting after syscall
Alexander Gordeev [Thu, 7 Oct 2010 14:54:24 +0000 (18:54 +0400)]
mips: fix errno setting after syscall

If there was an error during syscall then after it's completion a3
register holds a non-zero value and v0 holds an actual error code which
should be saved in errno. This can be achieved by calling
__syscall_error with the value from v0 as a parameter. So this value
should be stored in a0, but the appropriate assembly instructions are
missing. Fixed this now by adding "move a0, v0".

I think it was once fixed by 58c5f8ba4cdf62342d05a546d15404cbbb3c4e07
but then something went wrong. Again...

Signed-off-by: Alexander Gordeev <lasaine@lvk.cs.msu.su>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years ago*: inline constant __sig{add,del}set and __sigismember
Denys Vlasenko [Fri, 22 Oct 2010 15:01:05 +0000 (17:01 +0200)]
*: inline constant __sig{add,del}set and __sigismember

    text           data     bss     dec     hex filename
-    318              4       0     322     142 libc/pwd_grp/lckpwdf.o
+    312              4       0     316     13c libc/pwd_grp/lckpwdf.o
-    166              0       1     167      a7 libc/stdlib/abort.o
+    157              0       1     158      9e libc/stdlib/abort.o
-     42              0       0      42      2a libc/sysdeps/linux/common/pause.o
+     27              0       0      27      1b libc/sysdeps/linux/common/pause.o

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agoremove superfluous libc_hidden_proto(memcpy) and #include
Denys Vlasenko [Fri, 22 Oct 2010 14:52:41 +0000 (16:52 +0200)]
remove superfluous libc_hidden_proto(memcpy) and #include

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agosleep: tiny code shrink
Denys Vlasenko [Fri, 22 Oct 2010 13:46:04 +0000 (15:46 +0200)]
sleep: tiny code shrink

...or rather, it WILL BE code shrink when gcc become clever enough
to not emit a second, useless XORing of ebx:

31 db                  xor    %ebx,%ebx
85 c0                  test   %eax,%eax
74 11                  je     73 <__GI_sleep+0x73>
31 db                  xor    %ebx,%ebx    <=== ?!

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agosleep: employ __USE_EXTERN_INLINES (with necessary fixes)
Denys Vlasenko [Fri, 22 Oct 2010 13:24:13 +0000 (15:24 +0200)]
sleep: employ __USE_EXTERN_INLINES (with necessary fixes)

__USE_EXTERN_INLINES was unused and had bit-rotted, had to fix it
when it didn't work as intended at first.

    text    data     bss     dec     hex filename
-    168       0       0     168      a8 libc/unistd/sleep.o
+    146       0       0     146      92 libc/unistd/sleep.o

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agosleep: check "SIGCHLD is SIG_IGN'ed" first. Saves two syscalls in common case
Denys Vlasenko [Fri, 22 Oct 2010 13:22:40 +0000 (15:22 +0200)]
sleep: check "SIGCHLD is SIG_IGN'ed" first. Saves two syscalls in common case

    text    data     bss     dec     hex filename
-    197       0       0     197      c5 libc/unistd/sleep.o
+    168       0       0     168      a8 libc/unistd/sleep.o

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agolibm/x86: use call instead of jump for wrappers
Timo Teräs [Fri, 22 Oct 2010 11:58:13 +0000 (13:58 +0200)]
libm/x86: use call instead of jump for wrappers

GCC can emit prologue/epilogue code for the functions in various
different cases:
 - frame pointers
 - PIC build (to load ebx for indirect calls/jumps)
 - forced stack smashing protection

If we used jump in such cases, we'd corrupt the call stack and
crash.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agosleep: remove commented-out code. no code changes
Denys Vlasenko [Fri, 22 Oct 2010 11:48:57 +0000 (13:48 +0200)]
sleep: remove commented-out code. no code changes

It can be easily reconstructed, since it's obvious

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agosleep: code shrink
Denys Vlasenko [Thu, 21 Oct 2010 16:19:35 +0000 (18:19 +0200)]
sleep: code shrink

Use less stack by using same "sigset_t set" object for new and saved
signal set, remove redundant clearing of set, and do not save/restore
errno around sigprocmask(SIG_SETMASK) - it never changes it.

While at it, improve comments and make code style consistent
across sleep.c file.

    text           data     bss     dec     hex filename
-    242              0       0     242      f2 libc/unistd/sleep.o
+    197              0       0     197      c5 libc/unistd/sleep.o

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agovfprintf.c: reduce a chunk of #ifdef forest and remove one goto inside it
Denys Vlasenko [Tue, 19 Oct 2010 13:07:00 +0000 (15:07 +0200)]
vfprintf.c: reduce a chunk of #ifdef forest and remove one goto inside it

No code changes according to objdump.

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years ago_vfprintf.c: de-obfuscate badly twisted fragment. no code changes.
Denys Vlasenko [Tue, 19 Oct 2010 12:46:05 +0000 (14:46 +0200)]
_vfprintf.c: de-obfuscate badly twisted fragment. no code changes.

objdump confirms that I did not mess it up.

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agovfprintf.c: remove endif comments which clog up the source. no code changes
Denys Vlasenko [Tue, 19 Oct 2010 12:37:40 +0000 (14:37 +0200)]
vfprintf.c: remove endif comments which clog up the source. no code changes

Example:

        --ppfs->maxposarg;

Verified with objdump that no code is changed

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agovfprintf.c: de-obfuscate if(with nested assignments). no logic changes
Denys Vlasenko [Tue, 19 Oct 2010 12:07:51 +0000 (14:07 +0200)]
vfprintf.c: de-obfuscate if(with nested assignments). no logic changes

God knows this file is hard to read as-is, some readability improvement
is in order.

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agogetproto: increase line buffer size, simlify and fix alias handling
Natanael Copa [Thu, 14 Oct 2010 06:35:10 +0000 (06:35 +0000)]
getproto: increase line buffer size, simlify and fix alias handling

We increase line buffer size, reduce MAXALIASES and make sure we don't
segfault when there are too manuy aliases in /etc/protocols.

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agogetnet: simplify alias handling and reduce MAXALIASES
Natanael Copa [Thu, 14 Oct 2010 06:35:09 +0000 (06:35 +0000)]
getnet: simplify alias handling and reduce MAXALIASES

Reduce MAXALIASES to something lower. There will probably never be
need for more than 1 alias but we allow a few extra.

While here we alos fix segfault when there are too many aliases.

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agogetservice: fix handling of long lines
Natanael Copa [Thu, 14 Oct 2010 06:35:08 +0000 (06:35 +0000)]
getservice: fix handling of long lines

Don't try to be smart by dynamically realloc buffersize as it doesn't
work. Instead, be simple and allocate a buffer big enough.

This fixes a memory leak when calling getserv{ent,byname,byport}
multiple times.

To save memory we reduce number of max aliases. We seldomly will need
more than 1 anyways. While here, fix segfault that happened if there
were too many aliases.

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agoparse_config: discard rest of incomplete line
Natanael Copa [Thu, 14 Oct 2010 06:35:07 +0000 (06:35 +0000)]
parse_config: discard rest of incomplete line

If line is longer then size of given buffer and buffer is not allocated by
the config parser itself, then discard rest of line.

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agoconfig parser: always initialize line pointer
Natanael Copa [Thu, 14 Oct 2010 06:35:06 +0000 (06:35 +0000)]
config parser: always initialize line pointer

We must always initialize line pointer since data pointer might
have changed due to a realloc (in getserv.c for example).

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agogetservice: getservent_r must return ERANGE when buffer is too small
Natanael Copa [Thu, 14 Oct 2010 06:35:05 +0000 (06:35 +0000)]
getservice: getservent_r must return ERANGE when buffer is too small

This fixes issue introduced by 72e1a1ce186c39f07282398e2af9eb0253e60f15

This should also fix the following testcase to exit with error rather
than cause an endless loop.

int main(void) {
if (getservbyname("non-existing", "udp") == NULL)
err(1, "getservbyname");
return 0;
}

Reported by Pirmin Walthert
http://lists.uclibc.org/pipermail/uclibc/2010-August/044277.html

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agoconfig parser: do not assume that realloc return same pointer
Natanael Copa [Thu, 14 Oct 2010 06:35:04 +0000 (06:35 +0000)]
config parser: do not assume that realloc return same pointer

We need to update the parser->line pointer on realloc and do not
initialize the token array til after the potensial realloc in
bb_get_chunk_with_continuation().

While here, also replace a realloc() with malloc() where pointer always
is NULL.

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agosparc: pipe.S
Austin Foxley [Fri, 1 Oct 2010 23:06:23 +0000 (16:06 -0700)]
sparc: pipe.S

return value of pipe on success is supposed to be 0

Signed-off-by: Austin Foxley <austinf@cetoncorp.com>
13 years agoldso_sh: add support for protected symbols to SH
Salvatore Cro [Thu, 23 Sep 2010 10:01:20 +0000 (12:01 +0200)]
ldso_sh: add support for protected symbols to SH

Protected symbols should not be overridden by
symbols from other modules. Such symbols are exported
i.e. globally visible, but references from whithin
defining modules are satisfied locally.

Signed-off-by: Salvatore Cro <salvatore.cro@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agolibc: Sort HEADERS_RM alphabetically
Carmelo Amoroso [Wed, 22 Sep 2010 10:13:21 +0000 (12:13 +0200)]
libc: Sort HEADERS_RM alphabetically

No functional changes, just a code tidy-up.
Please keep them sorted in future changes.

Signed-off-by: Filippo Arcidiacono <filippo.arcidiacono@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agonptl: Fix libpthread build when UCLIBC_LINUX_SPECIFIC is disabled
Carmelo Amoroso [Mon, 20 Sep 2010 13:03:17 +0000 (15:03 +0200)]
nptl: Fix libpthread build when UCLIBC_LINUX_SPECIFIC is disabled

NPTL library needs both madvise and statfs symbols, that are guarded
by UCLIBC_LINUX_SPECIFIC option. This fix provides these symbols too
when NPTL is used, indipendently by UCLIBC_LINUX_SPECIFIC choice.
Otherwise libpthread link fails as below:

LD libpthread-0.9.32-git.so
libpthread/nptl/libpthread_so.a(pthread_create.oS): In function `__free_tcb':
pthread_create.c:(.text+0x1184): undefined reference to `madvise'
libpthread/nptl/libpthread_so.a(sem_open.oS): In function `__where_is_shmfs':
sem_open.c:(.text+0x764): undefined reference to `statfs'
collect2: ld returned 1 exit status
make: *** [lib/libpthread.so] Error 1

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agolibc: Handle cancellation in non multiplexed socket calls
Carmelo Amoroso [Wed, 30 Jun 2010 06:32:15 +0000 (08:32 +0200)]
libc: Handle cancellation in non multiplexed socket calls

For those archs that provide non multiplexed socket calls
it possible to implement the lib C wrappers without calling the
multi-purpose __socketcall. For a subset of these functions that
are cancellation point, it needs to correctly handle cancellation.

Signed-off-by: Francesco Rundo <francesco.rundo@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agonptl: Remove nptl/sysdeps/generic directory from include list
Carmelo Amoroso [Thu, 16 Sep 2010 12:52:11 +0000 (14:52 +0200)]
nptl: Remove nptl/sysdeps/generic directory from include list

All archs have now moved their generic sysdep.h header from
sysdeps/generic to sysdeps/common directory, so it is possible
to remove the sysdeps/generic directory from the include list.

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agonptl_sh: Fix compiler warning due to shadowed variable
Carmelo Amoroso [Tue, 20 Jul 2010 05:45:19 +0000 (07:45 +0200)]
nptl_sh: Fix compiler warning due to shadowed variable

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agolibc: Add canonicalize_file_name function
Carmelo Amoroso [Thu, 16 Sep 2010 10:22:34 +0000 (12:22 +0200)]
libc: Add canonicalize_file_name function

Add canonicalize_file_name function and its related tests.
Required by elfutils and coreutils (readlink).

Signed-off-by: Salvatore Cro <salvatore.cro@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agotest: Update gitignore to take into account new tests added.
Carmelo Amoroso [Thu, 16 Sep 2010 09:49:20 +0000 (11:49 +0200)]
test: Update gitignore to take into account new tests added.

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agoldso: Do not adjust dynamic section entries unnecessarily
Filippo Arcidiacono [Tue, 27 Jul 2010 15:45:26 +0000 (17:45 +0200)]
ldso: Do not adjust dynamic section entries unnecessarily

Signed-off-by: Filippo Arcidiacono <filippo.arcidiacono@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agotests: Added new nptl tests
Salvatore Cro [Fri, 10 Sep 2010 08:55:43 +0000 (10:55 +0200)]
tests: Added new nptl tests

tests: Added several nptl tests from glibc 2.10.1

Signed-off-by: Salvatore Cro <salvatore.cro@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agosh: update the memcpy adding a new loop with aggressive prefetching
Salvatore Cro [Thu, 9 Sep 2010 14:10:21 +0000 (16:10 +0200)]
sh: update the memcpy adding a new loop with aggressive prefetching

After exploring different prefetch distance-degree combinations
in this new update of the memcpy function, a new loop has been added
for moving many cache lines with an aggressive prefetching schema.
Prefetch has been removed when move few cache line aligned blocks.
As final result, this memcpy gives us the same performances for small
sizes (we already had!) and better numbers for big copies.
In case of SH4-300 CPU Series, benchmarks show a gain of ~20% for sizes
from 4KiB to 256KiB.
In case of the SH4-200, there is a gain of ~40% for sizes bigger than
32KiB.

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agosh: move data without fetching cache block within the memset
Salvatore Cro [Thu, 9 Sep 2010 14:08:54 +0000 (16:08 +0200)]
sh: move data without fetching cache block within the memset

With this patch the movca.l instruction is used within the memset.
The current memset implementation only uses the FPU and there is
an real gain for all the sizes.
Adding the movca.l instruction numbers always are better than the generic code.
There is a big gain for size greater than 64 KiB but number are worst for 4-32KiB
sizes compared with the implementation without movca.l.

Time Memory Bandwidth (Mbytes)
-------------------------------------------------
    Generic         SH4          SH4
                   (FPU)     (FPU+movca.l)
-------------------------------------------------
512         1143  1998          1596
1 KiB       1273  2567          1915
2 KiB       1350  2993          2128
4-32KiB     1391  3262          2252
64KiB-16MiB 170  186       *830*

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agolibc: Fix cancellation handling in some C functions
Salvatore Cro [Thu, 9 Sep 2010 14:01:04 +0000 (16:01 +0200)]
libc: Fix cancellation handling in some C functions

According to POSIX.1-2008 standard, the following syscalls shall be
cancellation points : waitid, sleep, fdatasync, ppoll.
Further, if generic syscall is not available and stubs are
configured, provide the stub implementation for function.

Signed-off-by: Salvatore Cro <salvatore.cro@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agolibubacktrace: Provide uClibc with backtrace functions
Salvatore Cro [Thu, 9 Sep 2010 13:45:44 +0000 (15:45 +0200)]
libubacktrace: Provide uClibc with backtrace functions

A new shared object, libubacktrace.so.0 is added to uClibc
to provide backtrace functions to support application self-debugging.
This set of functions requires to dynamically load libgcc_s.so so they
need to call dlopen/dlsym that are provided by libdl. For this reason
they cannot be included into libc.so.0 but are provided by a new library.

User application that wants to use backtrace needs to be compiled with
-fexceptions option and -rdynamic to get full symbols printed and must be
linked against libubacktrace.so

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agomisc: utmpx based logging support
Salvatore Cro [Thu, 9 Sep 2010 13:38:15 +0000 (15:38 +0200)]
misc: utmpx based logging support

misc: Added support for accessing user accounting database based
on utmpx structure.

Signed-off-by: Salvatore Cro <salvatore.cro@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agonptl: fix static linking
Natanael Copa [Thu, 9 Sep 2010 12:03:10 +0000 (14:03 +0200)]
nptl: fix static linking

We need dl-tls.c for static libc or we will get missing symbols when
linking thinkgs like openssl static.

Signed-off-by: Natanael Copa <natanael.copa@gmail.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agotests: nptl tests dependencies cleanup
Salvatore Cro [Thu, 9 Sep 2010 06:20:44 +0000 (08:20 +0200)]
tests: nptl tests dependencies cleanup

tests: some cleanings on nptl tests' dependencies.

Signed-off-by: Salvatore Cro <salvatore.cro@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agoldso: Fix compilation for x86_64 without TLS support.
Will Newton [Mon, 6 Sep 2010 14:59:21 +0000 (15:59 +0100)]
ldso: Fix compilation for x86_64 without TLS support.

Add an #ifdef to avoid using TLS structures when TLS is not enabled.

Signed-off-by: Will Newton <will.newton@imgtec.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agomisc: Fix build error about missing PAGE_SIZE
Abdoulaye Walsimou Gaye [Sun, 5 Sep 2010 22:45:10 +0000 (00:45 +0200)]
misc: Fix build error about missing PAGE_SIZE

This patch fixes the following build error:

  CC libc/misc/internals/__h_errno_location.os
  CC libc/misc/internals/parse_config.os
libc/misc/internals/parse_config.c: In function
'bb_get_chunk_with_continuation':
libc/misc/internals/parse_config.c:77: error: 'PAGE_SIZE' undeclared (first use in this function)
libc/misc/internals/parse_config.c:77: error: (Each undeclared identifier is reported only once
libc/misc/internals/parse_config.c:77: error: for each function it appears in.)
make[2]: *** [libc/misc/internals/parse_config.os] Error 1

Signed-off-by: Abdoulaye Walsimou Gaye <awg@embtoolkit.org>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agolibc: Provide both getpgid and __getpgid symbols
Carmelo Amoroso [Mon, 23 Aug 2010 22:14:00 +0000 (00:14 +0200)]
libc: Provide both getpgid and __getpgid symbols

Indeed unistd.h header does export '__getpgid' by default, while
uClibc provide only 'getpgid' implementation. The 'getpgid' symbol
is exported by standard header if __USE_XOPEN_EXTENDED is defined.
This patch alignes uClibc implementation with standard header (matching
with glibc behaviour).

Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
13 years agoFix resolver broken in NPTL build
Timo Teräs [Wed, 12 May 2010 07:32:48 +0000 (10:32 +0300)]
Fix resolver broken in NPTL build

The combination of

commit aab4df0fb51660300559f5f29290709db2f7bfee
  resolv.c: add support for per thread res_state

commit cca45baf8353d1e338d232f5bdb2d1d6b357f1da
  /etc/resolv.conf: support "timeout:n" and "attempts:n" options

.. and NPTL results in broken resolver in very annoying ways.

Now, it seems that most of the uclibc code does not work well
if res_state is TLS variable. Technically, this is the correct thing
to do since this gives proper per-thread resolving behavior, and
it also makes the config options overridable per thread. This
probably what apps expect as glibc does it too.

But alas, most places use _res to sync up static global variables
which results in breakage. It gets more or less randomly selected
which threads options get applied. Also in case of multiple servers
it looks like the retry logic is shared between all threads, e.g.
two concurrent resolutions can make other resolvers skip nameservers
due to shared "last_ns_num".

And finally the timeout/attempts commit breaks the accumulated stuff
horribly. What happens is:
 1. multithreaded application startups, initializes resolver,
    resolves things just fine
 2. resolv.conf gets changed, application calls res_init
    after res_init uclibc will call res_sync on all resolver functions
    to refresh globals from the TLS variable _res
 3. res_init was called only in one thread, so other thread's
    _res contains all zeroes (yes, this is correct app usage:
    res_init should be called only from one thread)
 4. threads not calling res_init get broken resolver due to timeout
    being set to zero

Now, one proper solution would be to:
 1. make __open_nameservers return the configuration options
 2. pass the config options struct to res_sync_func so it can do
    the proper overrides from per-thread _res
 3. remove the related globals and use locally config options from
    __open_nameservers

But technically, the correct thing (as in glibc does this) is:
 - res_init increases global "res_init timestamp"
 - use _res (or pointer within there) for resolver retries etc.
   get proper per-thread behavior
 - resolvers functions call "maybe_init" which reinitialize the
   TLS'ed resolver state (e.g. reload resolv.conf) if their res_init
   timestamp is out dated

As an immediate emergency kludge fix, the following might do:

resolv: fix options handling for TLS _res

If _res is in TLS (NPTL), it might not get initialized. Assume
zeroes mean default values.

Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
13 years ago_uintmaxtostr: fix indentation (spaces->tabs), no code changes
Denys Vlasenko [Thu, 26 Aug 2010 10:40:28 +0000 (12:40 +0200)]
_uintmaxtostr: fix indentation (spaces->tabs), no code changes

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
13 years agoresolv: simplify MAXALIAS handling
Bernhard Reutner-Fischer [Thu, 19 Aug 2010 17:07:43 +0000 (19:07 +0200)]
resolv: simplify MAXALIAS handling

Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
13 years agogetserv: fix reading services lines w > 80 chars
Bernhard Reutner-Fischer [Thu, 19 Aug 2010 17:06:26 +0000 (19:06 +0200)]
getserv: fix reading services lines w > 80 chars

e.g. getservbyname()

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