Joseph Myers [Fri, 25 Aug 2017 17:02:33 +0000 (17:02 +0000)]
Use __HAVE_DISTINCT_FLOAT128 in tgmath.h.
When _Float128 is ABI-equivalent to long double, there is no need for
tgmath.h to have any special _Float128 handling: it's always OK to
call the long double versions of functions for _Float128 arguments in
that case, and the logic to determine return types is generic. Thus,
this patch changes the use of __HAVE_FLOAT128 to
__HAVE_DISTINCT_FLOAT128, as a minor optimization to reduce the size
of the macro expansions in the ABI-equivalent case.
Tested for x86_64.
* math/tgmath.h [__HAVE_FLOAT128]: Change conditional to
[__HAVE_DISTINCT_FLOAT128].
Joseph Myers [Fri, 25 Aug 2017 16:11:13 +0000 (16:11 +0000)]
Clean up bits/math-finite.h for aliasing types.
This patch cleans up how bits/math-finite.h handles types that are
ABI-aliases of other types.
For such types, no __*_finite functions exist; instead,
bits/math-finite.h must redirect calls to a the functions for a
canonical choice of type for each floating-point format. (For the
actual public interfaces, symbols need exporting for each type, even
those that are ABI-aliases, because of standard requirements that
programs can declare the functions themselves without including
<math.h>, but that does not apply to __*_finite.)
At present, there is a special-case conditional in bits/math-finite.h
on __MATH_DECLARING_LDOUBLE && defined __NO_LONG_DOUBLE_MATH to handle
redirecting long double function calls to double __*_finite. This
patch replaces this by a more general mechanism. math.h, before each
inclusion of bits/math-finite.h, defines _MSUFTO_ as the suffix to use
on the target of redirection, in addition to the existing _MSUF_.
This way, __MATH_DECLARING_LDOUBLE can go away, as can the special
conditional in bits/math-finite.h. With this patch, math.h is now
prepared for the case of supporting float128 functions as aliases of
long double ones on platforms where long double is binary128, with
_MSUFTO_ appropriately defined for that case, and appropriate _MSUFTO_
definitions can easily be included when supporting _Float32 / _Float64
/ _Float32x / _Float64x (which will always be ABI-aliases of another
type when supported).
Tested for x86_64, and did a compilation test for ARM with
build-many-glibcs.py to cover the long double = double case.
* math/math.h (_MSUFTO_): Define and undefine for each inclusion
of <bits/math-finite.h>.
(__MATH_DECLARING_LDOUBLE): Do not define and undefine for each
inclusion of <bits/math-finite.h>.
* math/bits/math-finite.h (__REDIRTO_X): Do not define
conditionally on [__MATH_DECLARING_LDOUBLE && defined
__NO_LONG_DOUBLE_MATH].
(__MATH_REDIRCALL): Use _MSUFTO_ in __REDIRTO call.
(__MATH_REDIRCALL_2): Likewise.
(__MATH_REDIRCALL_INTERNAL): Likewise.
(__REDIRFROM (lgamma, , _MSUF_)): Likewise.
(__REDIRFROM (gamma, , _MSUF_)): Likewise.
(__REDIRFROM (tgamma, , _MSUF_)): Likewise.
Joseph Myers [Fri, 25 Aug 2017 11:33:04 +0000 (11:33 +0000)]
Remove powerpc32 sqrt wrappers.
This patch removes the powerpc32-specific wrappers for sqrt and sqrtf.
These wrappers, by adding architecture-specific uses of _LIB_VERSION
and __kernel_standard, unnecessarily complicate cleanups of libm error
handling. They also do not serve a useful optimization purpose. GCC
knows about sqrt as a built-in function, and can generate direct calls
to a hardware square root instruction, either on its own, in the
-fno-math-errno case, or together with an inline check for the
argument being negative and a call to the out-of-line sqrt function
for error handling only in that case (and has been able to do so for a
long time). Thus in practice the wrapper will only be called only in
the case of negative arguments, which is not a case it is useful to
optimize for.
Tested with build-many-glibcs.py for powerpc-linux-gnu-power4.
Fix the C++ version of issignaling when __NO_LONG_DOUBLE_MATH is defined
When __NO_LONG_DOUBLE_MATH is defined, __issignalingl is not available,
thus issignaling with long double argument should call __issignaling,
instead.
Tested for powerpc64le.
* math/math.h [defined __cplusplus] (issignaling): In the long
double case, call __issignalingl only if __NO_LONG_DOUBLE_MATH
is not defined. Call __issignaling, otherwise.
Remove cancellation support for syscall generation
This patch removes the cancellation mark from the auto-generation syscall
script. Now all the cancellable syscalls are done throught C code using
the SYSCALL_CANCEL macro. It simplifies the assembly required to each
architecture port, since the SYSCALL_CANCEL uses the already defined
INLINE_SYSCALL macros, and allows a more straigh fix on cancellation
machanism (since no more specific assembly fixes will be required).
Checked on i686-linux-gnu, x86_64-linux-gnu, x86_64-linux-gnux32,
aarch64-linux-gnu, arm-linux-gnueabihf, and powerpc64le-linux-gnu.A
Also with build-many-glibc.py with remaning touched architectures.
Szabolcs Nagy [Thu, 24 Aug 2017 14:56:11 +0000 (15:56 +0100)]
Fix CFLAGS override in sysdeps/ieee754/dbl-64
Use += instead of = to avoid overriding target specific CFLAGS settings.
Ideally the settings in target Makefiles would have precedence, but the
Makefile inclusion order does not allow that, with this fix at least the
target settings are not dropped.
H.J. Lu [Wed, 23 Aug 2017 15:22:52 +0000 (08:22 -0700)]
string/stratcliff.c: Replace int with size_t [BZ #21982]
Fix GCC 7 errors when string/stratcliff.c is compiled with -O3:
stratcliff.c: In function ‘do_test’:
cc1: error: assuming signed overflow does not occur when assuming that (X - c) <= X is always true [-Werror=strict-overflow]
[BZ #21982]
* string/stratcliff.c (do_test): Declare size, nchars, inner,
middle and outer with size_t instead of int. Repleace %d and
%Zd with %zu in printf. Update "MAX (0, nchars - 128)" and
"MAX (outer, nchars - 64)" to support unsigned outer and
nchars. Also exit loop when outer == 0.
This patch consolidate the remaning non cancellable syscall definitions
on not-cancel.h header. They are:
* __fcntl_nocancel: Moved from fcntl.h to not-cancel.h.
* __sigsuspend_nocancel: Removed since 988f991b50 it is not used or
defined anymore.
* __nanosleep_nocancel: Removed since 6f33fd046b it is defined on
not-cancel.h.
Now all non-cancellable syscall definition are defined on not-cancel
(the only exceptions is the stdio symbol __fxprintf_nocancel which
uses non cancellable open and it is used on getopt implementation).
Checked on x86_64-linux-gnu and with build-many-glibc.py.
* include/fcntl.h (__fcntl_nocancel): Remove definition.
* include/signal.h (__sigsuspend_nocancel): Likewise.
* include/time.h (__nanosleep_nocancel): Likewise.
* sysdeps/generic/not-cancel.h (__fcntl_nocancel): New macro.
* login/utmp_file.c: Include non cancellable syscall header.
* sysdeps/unix/sysv/linux/not-cancel.h (__fcntl_nocancel): New
prototype.
Declare ifunc resolver to return a pointer to the same type as the target
function to help GCC detect incompatibilities between the two when it's
enhanced to do so.
builds for powerpc64le fail in the declaration of some ifunc resolvers,
because the ifunc is declared with unmatching return types. One of the
declarations comes from the __ifunc_resolver macro, which was patched by
the aforementioned commit:
/* Helper / base macros for indirect function symbols. */
#define __ifunc_resolver(type_name, name, expr, arg, init, classifier) \
classifier inhibit_stack_protector \
__typeof (type_name) *name##_ifunc (arg) \
whereas the other comes from the unpatched __ifunc macro when
HAVE_GCC_IFUNC is not defined:
This patch changes the return type of the ifunc resolver in the __ifunc
macro, so that it matches the return type of the target function,
similarly to what the aforementioned commit does.
Tested for powerpc64le and s390x with unpatched GCC.
* include/libc-symbols.h: [!defined HAVE_GCC_IFUNC] (__ifunc):
Change the return type of the ifunc resolver to match the return
type of the target function.
The problem for x32 is the {INTERNAL,INLINE}_SYSCALL C macros explicit
cast the arguments to 'long int', thus passing as 32 bits arguments
that should be passed to 64 bits.
Previous x32 implementation uses the auto-generated syscalls from
assembly macros (syscalls.list), so the {INTERNAL,INLINE}_SYSCALL
macros are never used with 64 bit argument in x32 (which are
internally broken for this ILP).
To fix it I used a strategy similar to MIPS64n32 (although both
ABI differs for some syscalls on how top pass 64-bits arguments)
where argument types for kernel call are defined using GCC extension
'typeof' with a arithmetic operation. This allows 64-bits arguments
to be defined while 32-bits argument will still passed as 32-bits.
I also cleanup the {INLINE,INTERNAL}_SYSCALL definition by defining
'inline_syscallX' instead of constructing the argument passing using
macros (it adds some readability) and removed the ununsed
INTERNAL_SYSCALL_NCS_TYPES define (since the patch idea is exactly to
avoid requiric explicit types passing).
Remove duplicate inclusion of header math-svid-compat.h
The header math-svid-compat.h has been unintentionally included twice in
the wrappers for the remainder functions. This patch removes the
duplicate inclusions.
During the development of float128 on powerpc64le, the ulps have been
generated since early versions of the patches. On those versions, the
functions gamma and pow10 were mistakenly thought to be part of the API.
After review, the functions were removed, however the ulps for them were
carried in the ulps file.
This patch removes such entries from the ulps file, as well as it
shrinks the ulps for the cpow and lgamma functions.
Joseph Myers [Tue, 22 Aug 2017 17:55:42 +0000 (17:55 +0000)]
Fix tgmath.h handling of complex integers (bug 21684).
The tgmath.h macros return a real type not a complex type when an
argument is of complex integer type (a GNU extension) and there are no
arguments of complex floating type. It seems clear that just as real
integers are mapped to double for tgmath.h, so complex integers should
be mapped to _Complex double.
This patch implements such a mapping. The main complication in fixing
this bug is that the tgmath.h macros expand their arguments a large
number of times, resulting in exponential blowup of the size of the
expansion when calls to tgmath.h macros are used in the arguments of
such macros; it would be unfortunate for fixing a bug with a fairly
obscure extension to make the macros expand their arguments even more
times. Thus, this patch optimizes the definitions of the relevant
macros. __tgmath_real_type previously expanded its argument 7 times
and now expands it 3 times. __tgmath_complex_type, used in place of
__tgmath_real_type only for functions that might return either real or
complex types, not for complex functions that always return real types
or always return complex types, expands its argument 5 times. So the
sizes of the macro expansions from nested macro calls are
correspondingly reduced (remembering that each tgmath.h macro expands
__tgmath_real_type, or sometimes now __tgmath_complex_type, several
times).
Sometimes the real return type resulted from calling a complex
function and converting the result to a real type; sometimes it
resulted from calling a real function, because the logic for
determining whether arguments were real or complex, based on sizeof,
was confused by integer promotions applying to e.g. short int but not
_Complex short int. The relevant tests are converted to use a new
macro __expr_is_real, which, by calling __builtin_classify_type rather
than comparing the results of two calls to sizeof, also reduces the
number of times macros expand their arguments.
Although there are reductions in the number of times macros expand
their arguments, I do not consider this to fix bug 21660, since a
proper fix means each macro expanding its arguments only once (via
using new compiler features designed for that purpose).
Tested for x86_64.
[BZ #21684]
* math/tgmath.h (__floating_type): Simplify definitions.
(__real_integer_type): New macro.
(__complex_integer_type): Likewise.
(__expr_is_real): Likewise.
(__tgmath_real_type_sub): Update comment to describe handling of
complex types.
(__tgmath_complex_type_sub): New macro.
(__tgmath_complex_type): Likewise.
[__HAVE_FLOAT128 && __GLIBC_USE (IEC_60559_TYPES_EXT)]
(__TGMATH_CF128): Use __expr_is_real.
(__TGMATH_UNARY_REAL_IMAG): Use __tgmath_complex_type and
__expr_is_real.
(__TGMATH_BINARY_REAL_IMAG): Likewise.
(__TGMATH_UNARY_REAL_IMAG_RET_REAL): Use __expr_is_real.
* math/gen-tgmath-tests.py (Type.create_type): Create complex
integer types.
This patch consolidates all the non cancellable nanosleep calls to use
the __nanosleep_nocancel identifier. For non cancellable targets it will
be just a macro to call the default respective symbol while on Linux
will be a internal one.
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
* nptl/pthread_mutex_timedlock.c (__pthread_mutex_timedlock): Replace
nanosleep_not_cancel with __nanosleep_nocancel.
* sysdeps/generic/not-cancel.h (nanosleep_not_cancel): Remove macro.
(__nanosleep_nocancel): New macro.
* sysdeps/unix/sysv/linux/nanosleep.c (__nanosleep_nocancel): New
function.
* sysdeps/unix/sysv/linux/not-cancel.h (nanosleep_not_cancel): Remove
macro.
(__nanosleep_nocancel): New prototype.
This patch consolidates all the non cancellable pause calls to use
the __pause_nocancel identifier. For non cancellable targets it will
be just a macro to call the default respective symbol while on Linux
will be a internal one.
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
* nptl/pthread_mutex_lock.c (__pthread_mutex_lock_full): Replace
pause_not_cancel with __pause_nocancel.
* sysdeps/generic/not-cancel.h (pause_not_cancel): Remove macro.
(__pause_nocancel): New macro.
* sysdeps/unix/sysv/linux/not-cancel.h (pause_not_cancel): Remove
macro.
(__pause_nocancel): New prototype.
* sysdeps/unix/sysv/linux/pause.c (__pause_nocancel): New function.
Martin Sebor [Tue, 22 Aug 2017 15:35:23 +0000 (09:35 -0600)]
Declare ifunc resolver to return a pointer to the same type as the target
function to help GCC detect incompatibilities between the two when it's
enhanced to do so.
Provide a C++ version of issignaling that does not use __MATH_TG
The macro __MATH_TG contains the logic to select between long double and
_Float128, when these types are ABI-distinct. This logic relies on
__builtin_types_compatible_p, which is not available in C++ mode.
On the other hand, C++ function overloading provides the means to
distinguish between the floating-point types. The overloading
resolution will match the correct parameter regardless of type
qualifiers, i.e.: const and volatile.
Tested for powerpc64le, s390x, and x86_64.
* math/math.h [defined __cplusplus] (issignaling): Provide a C++
definition for issignaling that does not rely on __MATH_TG,
since __MATH_TG uses __builtin_types_compatible_p, which is only
available in C mode.
(CFLAGS-test-math-issignaling.cc): New variable.
* math/Makefile [CXX] (tests): Add test-math-issignaling.
* math/test-math-issignaling.cc: New test for C++ implementation
of type-generic issignaling.
* sysdeps/powerpc/powerpc64le/Makefile [subdir == math]
(CXXFLAGS-test-math-issignaling.cc): Add -mfloat128 to the build
options of test-math-issignaling on powerpc64le.
H.J. Lu [Tue, 22 Aug 2017 11:49:14 +0000 (13:49 +0200)]
Add hidden visibility to internal function prototypes
Add hidden visibility to internal function prototypes to allow direct
access to internal functions within libc.a without using GOT when the
compiler defaults to -fPIE.
Size comparison of elf/ldconfig when the compiler defaults to -fPIE:
On x86-64:
text data bss dec hex
Before: 619646 20132 5488 645266 9d892
After : 619502 20132 5488 645122 9d802
On i686:
text data bss dec hex
Before: 550333 10748 3060 564141 89bad
After : 546453 10732 3060 560245 88c75
* include/libc-symbols.h (__hidden_proto_hiddenattr): New for
building libc.a.
(hidden_proto): Likewise.
(hidden_tls_proto): Likewise.
(__hidden_proto): Likewise.
Refactor long double information into bits/long-double.h.
resulted in sparc32 configurations installing the ldbl-opt version of
bits/long-double.h instead of the intended
sysdeps/unix/sysv/linux/sparc version.
For sparc32 by itself, this is not a problem, since the ldbl-opt
version is correct for sparc32. However, both sparc32 and sparc64 are
supposed to install sets of headers that work for both of them, so
that a single sysroot, whichever order the libraries are built and
installed in, works for both. The effect of having the wrong version
installed is that you end up with a miscompiled sparc64 libstdc++
which fails glibc's configure tests for the C++ compiler.
This patch moves the header from sysdeps/unix/sysv/linux/sparc to
separate copies of the same file for sparc32 and sparc64, to ensure it
comes before ldbl-opt in the sysdeps directory ordering.
Tested with build-many-glibcs.py for sparc64-linux-gnu and
sparcv9-linux-gnu.
[BZ #21987]
* sysdeps/unix/sysv/linux/sparc/bits/long-double.h: Remove file
and copy to ...
* sysdeps/unix/sysv/linux/sparc/sparc32/bits/long-double.h:
... here.
* sysdeps/unix/sysv/linux/sparc/sparc64/bits/long-double.h:
... and here.
Joseph Myers [Tue, 22 Aug 2017 00:30:51 +0000 (00:30 +0000)]
Fix position of tests-unsupported definition in assert/Makefile.
tests-unsupported has to be defined before the inclusion of Rules in a
subdirectory Makefile; otherwise it is ineffective. This patch fixes
the ordering in assert/Makefile, where a recent test addition put
tests-unsupported too late (resulting in build failures when the C++
compiler was missing or broken, and thereby showing up the unrelated
bug 21987).
Incidentally, I don't see why these tests depend on
$(have-cxx-thread_local) rather than just a working C++ compiler.
Tested in such a configuration (broken compiler/libstdc++) with
build-many-glibcs.py.
Joseph Myers [Mon, 21 Aug 2017 21:43:32 +0000 (21:43 +0000)]
Fix GCC 7 build of k_standard.c.
This patch adds a default case to k_standard.c that calls
__builtin_unreachable, to avoid an uninitialized variable error from
GCC 7 (reported in
<https://sourceware.org/ml/libc-alpha/2017-08/msg01012.html>).
Tested for x86_64 (with GCC 7).
* sysdeps/ieee754/k_standard.c (__kernel_standard): Add default
case calling __builtin_unreachable.
This patch consolidates all the non cancellable waitpid calls to use
the __waitpid_nocancel identifier. For non cancellable targets it will
be just a macro to call the default respective symbol while on Linux
will be a internal one.
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
* libio/ioopen.c (_IO_waitpid): Replace waitpid_not_cancel with
__waitpid_nocancel.
* sysdeps/generic/not-cancel.h (waitpid_not_cancel): Remove macro.
(__waitpid_nocancel): New macro.
* sysdeps/unix/sysv/linux/not-cancel.h (waitpid_not_cancel): Remove
macro.
(__waitpid_nocancel): Replace macro with a function.
* sysdeps/unix/sysv/linux/waitpid.c (__waitpid_nocancel): New
function.
This patch consolidates all the non cancellable fcntl calls to use
the __fcntl_nocancel identifier. For non cancellable targets it will
be just a macro to call the default respective symbol while on Linux
will be a internal one.
Since its prototype is already defined at internal fcntl.h header, it
is removed from not-cancel.h one.
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
This patch consolidates all the non cancellable writev calls to use
the __writev_nocancel identifier. For non cancellable targets it will
be just a macro to call the default respective symbol while on Linux
will be a internal one.
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
The errors are of the following form, building math/atest-exp:
[...]
/scratch/jmyers/glibc-bot/build/glibcs/aarch64-linux-gnu/glibc/stdlib/rshift.o: In function `__mpn_rshift':
/scratch/jmyers/glibc-bot/build/glibcs/aarch64-linux-gnu/glibc-src/stdlib/rshift.c:45: undefined reference to `__assert_fail'
/scratch/jmyers/glibc-bot/install/compilers/aarch64-linux-gnu/lib/gcc/aarch64-glibc-linux-gnu/7.2.1/../../../../aarch64-glibc-linux-gnu/bin/ld: /scratch/jmyers/glibc-bot/build/glibcs/aarch64-linux-gnu/glibc/math/atest-exp: hidden symbol `__assert_fail' isn't defined
/scratch/jmyers/glibc-bot/install/compilers/aarch64-linux-gnu/lib/gcc/aarch64-glibc-linux-gnu/7.2.1/../../../../aarch64-glibc-linux-gnu/bin/ld: final link failed: Bad value
This test is using various objects from the build of libc. Some of
those objects contain references to __assert_fail. Because those
references are hidden, they cannot refer to __assert_fail from
libc.so. Given the tests using internal objects those symbols in
libc.a can't safely be made hidden, so this patch reverts the problem
commit until any alternative approach that doesn't break the build can
be found.
Joseph Myers [Mon, 21 Aug 2017 17:46:34 +0000 (17:46 +0000)]
Remove SPARC sqrt wrappers (bug 21973).
This patch removes the SPARC-specific wrappers for sqrt and sqrtf.
These wrappers, by adding architecture-specific uses of _LIB_VERSION
and __kernel_standard, unnecessarily complicate cleanups of libm error
handling. They also do not serve a useful optimization purpose. GCC
knows about sqrt as a built-in function, and can generate direct calls
to a hardware square root instruction, either on its own, in the
-fno-math-errno case, or together with an inline check for the
argument being negative and a call to the out-of-line sqrt function
for error handling only in that case (and has been able to do so for a
long time). Thus in practice the wrapper will only be called only in
the case of negative arguments, which is not a case it is useful to
optimize for.
The removal of the wrappers also uncovers, and fixes, an old bug.
32-bit SPARC libm used (checked with glibc 2.8 binaries) to have a
sqrtl compat symbol, version GLIBC_2.0, for old binaries when sqrtl
was an alias of sqrt (I don't have pre-glibc-2.4 binaries for SPARC to
hand to check for the sqrtl symbol in those). This disappeared,
probably with:
Removing the wrappers brings back the generic ldbl-opt logic for
creating such compat symbols, and so restores the compat symbol that
should be there. This could of course also be fixed in the wrappers -
but as noted above, the wrappers are optimizing a case it's not useful
to optimize, so the bug of the missing compat symbol serves to
illustrate the risks involved with the extra complexity of
architecture-specific function versions where not needed.
Joseph Myers [Mon, 21 Aug 2017 17:45:10 +0000 (17:45 +0000)]
Obsolete matherr, _LIB_VERSION, libieee.a.
This patch obsoletes support for SVID libm error handling (the system
where a user-defined function matherr is called on a libm function
error; only enabled if you also set _LIB_VERSION = _SVID_ or
_LIB_VERSION = _XOPEN_) and the use of the _LIB_VERSION global
variable to control libm error handling. matherr and _LIB_VERSION are
made into compat symbols, not supported for new ports or for static
linking. The libieee.a object file (which sets _LIB_VERSION = _IEEE_,
so disabling errno setting for some functions) is also removed, and
all the related definitions are removed from math.h.
The manual already recommends against using matherr, and it's already
not supported for _Float128 functions (those use new wrappers that
don't support matherr, only errno) - this patch means that it becomes
possible to e.g. add sinf32 as an alias to sinf without that resulting
in undesired matherr support in sinf32 for existing glibc ports.
matherr support is not part of any standard supported by glibc (it was
removed in XPG4).
Because matherr is a function to be defined by the user, of course
user programs defining such a function will still continue to link; it
just quietly won't be used. If they try to write to the library's
copy of _LIB_VERSION to enable SVID error handling, however, they will
get a link error (but if they define their own _LIB_VERSION variable,
they won't).
I expect the most likely case of build failures from this patch to be
programs with unconditional cargo-culted uses of -lieee (based on a
notion of "I want IEEE floating point", not any actual requirement for
that library).
Ideally, the new-port-or-static-linking case would use the new
wrappers used for _Float128. This is not implemented in this patch,
because of the complication of architecture-specific (powerpc32 and
sparc) sqrt wrappers that use _LIB_VERSION and __kernel_standard
directly. Thus, the old wrappers and __kernel_standard are still
built unconditionally, and _LIB_VERSION still exists in static libm.
But when the old wrappers and __kernel_standard are built in the
non-compat case, _LIB_VERSION and matherr are defined as macros so
code to support those features isn't actually built into static libm
or new ports' shared libm after this patch.
I intend to move to the new wrappers for static libm and new ports in
followup patches. I believe the sqrt wrappers for powerpc32 and sparc
can reasonably be removed. GCC already optimizes the normal case of
sqrt by generating code that uses a hardware instruction and only
calls the sqrt function if the argument was negative (if
-fno-math-errno, of course, it just uses the hardware instruction
without any check for negative argument being needed). Thus those
wrappers will only actually get called in the case of negative
arguments, which is not a case it makes sense to optimize for. But
even without removing the powerpc32 and sparc wrappers it should still
be possible to move to the new wrappers for static libm and new ports,
just without having those dubious architecture-specific optimizations
in static libm.
Everything said about matherr equally applies to matherrf and matherrl
(IA64-specific, undocumented), except that the structure of IA64 libm
means it won't be converted to using the new wrappers (it doesn't use
the old ones either, but its own error-handling code instead).
As with other tests of compat symbols, I expect test-matherr and
test-matherr-2 to need to become appropriately conditional once we
have a system for disabling such tests for ports too new to have the
relevant symbols.
Tested for x86_64 and x86, and with build-many-glibcs.py.
H.J. Lu [Mon, 21 Aug 2017 12:50:38 +0000 (05:50 -0700)]
Add hidden visibility to internal function prototypes
Add hidden visibility to internal function prototypes to allow direct
access to internal functions within libc.a without using GOT when the
compiler defaults to -fPIE.
Size comparison of elf/ldconfig when the compiler defaults to -fPIE:
On x86-64:
text data bss dec hex
Before: 619646 20132 5488 645266 9d892
After : 619502 20132 5488 645122 9d802
On i686:
text data bss dec hex
Before: 550333 10748 3060 564141 89bad
After : 546453 10732 3060 560245 88c75
* include/libc-symbols.h (__hidden_proto_hiddenattr): New for
building libc.a.
(hidden_proto): Likewise.
(hidden_tls_proto): Likewise.
(__hidden_proto): Likewise.
H.J. Lu [Mon, 21 Aug 2017 12:47:10 +0000 (05:47 -0700)]
Enable hidden visibility in libc.a compiled with PIE
When building libc.a with PIE, enable hidden visibility to allow direct
access to definitions within libc.a without using GOT.
Size comparison of elf/ldconfig when the compiler defaults to -fPIE:
On x86-64:
text data bss dec hex
Before: 619206 20132 5488 644826 9d6da
After : 619062 20132 5488 644682 9d64a
On i686:
text data bss dec hex
Before: 556305 10816 3056 570177 8b341
After : 553688 10756 3056 567500 8a8cc
* include/libc-symbols.h (attribute_hidden): Enable hidden
visibility in libc.a compiled with PIE.
H.J. Lu [Sun, 30 Jul 2017 04:04:09 +0000 (21:04 -0700)]
Don't compile non-lib modules as lib modules [BZ #21864]
Some programs have more than one source files. These non-lib modules
should not be compiled with -DMODULE_NAME=libc. This patch puts these
non-lib modules in $(others-extras) and adds $(others-extras) to
all-nonlib.
which also recommended that they should also be removed from glibc. These
functions exist only in libc.a and are used for gcov from versions of GCC
older than GCC 3.3.
This patch consolidates all the non cancellable close calls to use
the __close_nocancel{_nostatus} identifier. For non cancellable targets
it will be just a macro to call the default respective symbol while on Linux
will be a internal one.
Also, since it is used on libcrypto it is also exported in GLIBC_PRIVATE
namespace.
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
This patch consolidates all the non cancellable openat{64} calls to use
the __openat{64}_nocancel identifier. For non cancellable targets it will
be just a macro to call the default respective symbol while on Linux
will be a internal one.
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
H.J. Lu [Fri, 18 Aug 2017 16:31:19 +0000 (09:31 -0700)]
Mark internal argz functions with attribute_hidden [BZ #18822]
Move internal argz function prototypes to include/argz.h and mark them
with attribute_hidden to allow direct access within libc.so and libc.a
without using GOT nor PLT. This also brings string/argz.h closer to the
gnulib version.
Do not use __builtin_types_compatible_p in C++ mode (bug 21930)
The logic to define isinf for float128 depends on the availability of
__builtin_types_compatible_p, which is only available in C mode,
however, the conditionals do not check for C or C++ mode. This lead to
an error in libstdc++ configure, as reported by bug 21930.
This patch adds a conditional for C mode in the definition of isinf for
float128. No definition is provided in C++ mode, since libstdc++
headers undefine isinf.
Tested for powerpc64le (glibc test suite and libstdc++-v3 configure).
[BZ #21930]
* math/math.h (isinf): Check if in C or C++ mode before using
__builtin_types_compatible_p, since this is a C mode feature.
This patch consolidates all the non cancellable write calls to use
the __write_nocancel identifier. For non cancellable targets it will
be just a macro to call the default respective symbol while on Linux
will be a internal one.
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
This patch consolidates all the non cancellable read calls to use
the __read_nocancel identifier. For non cancellable targets it will
be just a macro to call the default respective symbol while on Linux
will be a internal one.
Also, since it is used on libcrypto it is also exported in GLIBC_PRIVATE
namespace.
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
This patch consolidates all the non cancellable open calls to use
the __open_nocancel identifier. For non cancellable targets it will
be just a macro to call the default respective symbol while on Linux
will be a internal one.
To be consistent with the following non cancellable openat call, a
new __open64_nocancel is also added (although not currently used).
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
Wilco Dijkstra [Thu, 17 Aug 2017 15:27:20 +0000 (16:27 +0100)]
Add math benchmark latency test
This patch further improves math function benchmarking by adding a latency
test in addition to throughput. This enables more accurate comparisons of the
math functions. The latency test works by creating a dependency on the previous
iteration: func_res = F (func_res * zero + input[i]). The multiply by zero
avoids changing the input.
It reports reciprocal throughput and latency in nanoseconds (depending on the
timing header used) and max/min throughput in iterations per second:
Resolve some historically special cases of ambiguous width
[BZ #21750]
* unicode-gen/utf8_gen.py (U+00AD): Set width to 1.
* unicode-gen/utf8_gen.py (U+1160..U+11FF): Set width to 0.
* unicode-gen/utf8_gen.py (U+3248..U+324F): Set width to 2.
* unicode-gen/utf8_gen.py (U+4DC0..U+4DFF): Likewise.
[BZ #19852]
[BZ #21750]
* unicode-gen/utf8_gen.py: Process EastAsianWidth lines before
UnicodeData lines so the latter have precedence; remove hack
to group output by EastAsianWidth ranges.
Joseph Myers [Wed, 16 Aug 2017 20:33:59 +0000 (20:33 +0000)]
Fix sigval namespace (bug 21944).
XPG4.2 defines the siginfo_t type, but not union sigval or its
contents (which were added in the 1993 edition of POSIX.1), resulting
in namespace violations for sigval, sival_int and sival_ptr for
signal.h and sys/wait.h for that standard because those headers
incorrectly expose those names in that case.
This patch fixes this problem. The public type in this case is union
sigval, but various places in the headers use the sigval_t name for
it; direct uses of union sigval are already properly guarded or in
headers not in XPG4.2. Now, sigval_t, although not a standard name,
does seem to be widely used outside glibc. The approach taken by this
patch is to make installed headers use the name __sigval_t instead.
__sigval_t is then defined to either union sigval or union __sigval
(where union __sigval has __-prefixed member names as well), depending
on whether there are any namespace issues with the union sigval name
and its members. In the case where union __sigval is used, sigval_t
is not defined at all, to avoid the problem of sigval_t having a C++
mangled name that depends on feature test macros. sigval_t is still
defined by signal.h if __USE_MISC (reflecting the nonstandard nature
of that name).
Tested for x86_64.
[BZ #21944]
* signal/bits/types/__sigval_t.h: New file.
* signal/Makefile (headers): Add bits/types/__sigval_t.h.
* signal/bits/types/sigval_t.h: Include <bits/types/__sigval_t.h>
and define sigval_t using __sigval_t.
* include/bits/types/__sigval_t.h: New file.
* bits/types/sigevent_t.h: Include <bits/types/__sigval_t.h>
instead of <bits/types/__sigval_t.h>.
(struct sigevent): Use __sigval_t instead of sigval_t.
* bits/types/siginfo_t.h: Include <bits/types/__sigval_t.h>
instead of <bits/types/__sigval_t.h>.
(siginfo_t): Use __sigval_t instead of sigval_t.
* sysdeps/unix/sysv/linux/bits/types/sigevent_t.h: Include
<bits/types/__sigval_t.h> instead of <bits/types/__sigval_t.h>.
(struct sigevent): Use __sigval_t instead of sigval_t.
* sysdeps/unix/sysv/linux/bits/types/siginfo_t.h: Include
<bits/types/__sigval_t.h> instead of <bits/types/__sigval_t.h>.
(siginfo_t): Use __sigval_t instead of sigval_t.
* signal/signal.h [__USE_MISC]: Include <bits/types/sigval_t.h>.
Joseph Myers [Wed, 16 Aug 2017 17:01:27 +0000 (17:01 +0000)]
Allow abort PLT references in libc.so for SH.
Given my patch
<https://gcc.gnu.org/ml/gcc-patches/2017-08/msg00965.html> for the
ICEs building a glibc string function test for SH, the testsuite can
build completely for SH with GCC 7 and mainline. However, there is a
test failure that does not appear for GCC 6: check-localplt fails
because of an abort PLT reference in libc.so.
Given the lack of a trap insn pattern for SH, it seems unavoidable
that the compiler might sometimes generate abort calls, and such abort
calls (generated from __builtin_trap when there is no trap insn
pattern) will be unaffected by the normal mapping to __GI_abort for
calls within glibc. Thus, this patch allows (but does not require) an
abort PLT reference in libc.so for SH.
Tested for sh4-linux-gnu with build-many-glibcs.py (GCC 7, with my
patch applied).
* sysdeps/unix/sysv/linux/sh/localplt.data: Allow abort in
libc.so.
Florian Weimer [Tue, 15 Aug 2017 12:46:19 +0000 (14:46 +0200)]
i386: Replace internal_function attribute for __mcount_internal
__mcount_internal is called from assembler code. Use an explicit
regparm attribute to pass both arguments in registers, to match what
used to happen with internal_function before commit fbdc1e3e8de7f49e439b6e274d3e7e07da78416e (i386: Do not set
internal_function).
Mike FABIAN [Mon, 14 Aug 2017 17:48:27 +0000 (19:48 +0200)]
Change language name in LC_IDENTIFICATION of bn_BD and bn_IN from “Bengali” to “Bangla”
[BZ #14925]
* locales/bn_BD (LC_IDENTIFICATION): Change language name in
“title” and “language” from Bengali to Bangla.
* locales/bn_IN (LC_IDENTIFICATION): Likewise.
Mike FABIAN [Mon, 14 Aug 2017 17:05:17 +0000 (19:05 +0200)]
Use “copy "i18n"” in km_KH locale
The custom stuff which was in LC_CTYPE of the km_KH locale seems
to be a very incomplete subset of what one gets by using
“copy "i18n"”. I cannot find anything special there which is not
in “copy "i18n"”, only lots of stuff which is missing.
[BZ #20008]
* locales/km_KH (LC_CTYPE): Use “copy "i18n"”.
This commit replaces the functions with always-failing stubs,
with true compat symbols. Due to a happy accident, the calling
convention of the stub is identical for the internal_function
and non-internal_function case on i386.
In addition, this commit auto-generates the __nss_*_lookup2
function declarations as part of <nsswitch.h>.
Mike FABIAN [Mon, 14 Aug 2017 15:12:37 +0000 (17:12 +0200)]
Use two letter abbreviations in abday in all German locales.
[BZ #20482]
* locales/de_AT (LC_TIME): Use 2 letter abbreviations in abday.
* locales/de_BE (LC_TIME): Use 2 letter abbreviations in abday.
* locales/de_CH (LC_TIME): Use 2 letter abbreviations in abday.
* locales/de_DE (LC_TIME): Use readable ASCII in abday.
* locales/de_IT (LC_TIME): Use readable ASCII in abday.
* locales/de_LU (LC_TIME): Use 2 letter abbreviations in abday.
Default semantic for mmap2 syscall is to take the offset in 4096-byte
units. However m68k and ia64 mmap2 implementation take in the
configured pageunit units and for both architecture it can be
different values.
This patch fixes the m68k runtime discover of mmap2 offset unit
and adds the ia64 definition to find it at runtime.
Checked the basic tst-mmap and tst-mmap-offset on m68k (the system
is configured with 4k, so current code is already passing on this
system) and a sanity check on x86_64-linux-gnu (which should not be
affected by this change). Sergei also states that ia64 loader now
work correctly with this change.
Adhemerval Zanella <adhemerval.zanella@linaro.org>
Sergei Trofimovich <slyfox@inbox.ru>
* sysdeps/unix/sysv/linux/m68k/mmap_internal.h (MMAP2_PAGE_SHIFT):
Rename to MMAP2_PAGE_UNIT.
* sysdeps/unix/sysv/linux/mmap.c: Include mmap_internal iff
__OFF_T_MATCHES_OFF64_T is not defined.
* sysdeps/unix/sysv/linux/mmap_internal.h (page_unit): Declare as
uint64_t.
(MMAP2_PAGE_UNIT) [MMAP2_PAGE_UNIT == -1]: Redefine to page_unit.
(page_unit) [MMAP2_PAGE_UNIT != -1]: Remove definition.
Florian Weimer [Mon, 14 Aug 2017 11:03:34 +0000 (13:03 +0200)]
i386: Do not set internal_function
All calls to functions with the internal_function attribute
have been removed from assembler implementations, which means that
the definition of internal_function can be changed at the C level
without causing ABI issues with assembler code.
_dl_fixup still uses a regparm calling convention on i386, but this
is controlled through ARCH_FIXUP_ATTRIBUTE, not internal_function.
Florian Weimer [Mon, 14 Aug 2017 09:44:24 +0000 (11:44 +0200)]
_dl_fini: Remove internal_function attribute
Assembler code passes the address of _dl_fini to __libc_start_main,
whose function pointer argument lacks the attribute. This means
that calls could use the wrong ABI. Fortunately, for zero-parameter
void-returning functions, internal_function does not change ABI
on i386 (the only architecture which uses internal_function), so
this inconsistency was harmless (which is why it had not been
noticed so far).