This patch series cleans up the many uses of __ieee754_sqrt(f/l) in GLIBC.
The goal is to enable GCC to do the inlining, and if this fails call the
__ieee754_sqrt function. This is done by internally declaring sqrt with asm
redirects. The compat symbols and sqrt wrappers need to disable the redirect.
The redirect is also disabled if there are already redirects defined when
using -ffinite-math-only.
All math functions (but not math tests, non-library code and libnldbl) are
built with -fno-math-errno which means GCC will typically inline sqrt as a
single instruction. This means targets are no longer forced to add a special
inline for sqrt.
* include/math.h (sqrt): Declare with asm redirect.
(sqrtf): Likewise.
(sqrtl): Likewise.
(sqrtf128): Likewise.
* Makeconfig: Add -fno-math-errno for libc/libm, but build testsuite,
nonlib and libnldbl with -fmath-errno.
* math/w_sqrt_compat.c: Define NO_MATH_REDIRECT.
* math/w_sqrt_template.c: Likewise.
* math/w_sqrtf_compat.c: Likewise.
* math/w_sqrtl_compat.c: Likewise.
* sysdeps/i386/fpu/w_sqrt.c: Likewise.
* sysdeps/i386/fpu/w_sqrt_compat.c: Likewise.
* sysdeps/generic/math-type-macros-float128.h: Remove math.h and
complex.h.
This patch removes further parts of sysdeps/x86/fpu/bits/mathinline.h
that are only of value for optimization with older compiler versions,
in accordance with general principles of preferring the let the
compiler deal with such inlining through built-in functions.
In general, GCC supports inlining all these functions as of version
4.3 or earlier. However, some inlines in GCC may have had excessively
restrictive conditions in past GCC versions (e.g. requiring
-ffast-math when the inline is valid under broader conditions). (In
particular, GCC had, before GCC 7, unnecessarily restrictive
conditions on when it could apply floor and ceil inlines corresponding
to the ones removed here. The same was true for rint, but
bits/mathinline.h *also* was excessively restrictive there.)
The removed sincos inlines are for __sincos etc. functions (not a
public interface and not currently used in this header either; not in
a part of the header ever used for building glibc itself). Likewise,
the atan2 inlines included one for __atan2l, also not a public
interface and not used for building glibc itself (calls inside glibc
generally use __ieee754_atan2l, for which there is a separate
__LIBC_INTERNAL_MATH_INLINES case in this header).
Tested for x86_64 and x86.
* sysdeps/x86/fpu/bits/mathinline.h [__FAST_MATH__]
(__sincos_code): Remove define and undefine.
[__FAST_MATH__] (__sincos): Remove inline function.
[__FAST_MATH__] (__sincosf): Remove inline function.
[__FAST_MATH__] (__sincosl): Remove inline function.
(__atan2l): Remove inline functions.
[!__GNUC_PREREQ (3, 4)] (__atan2_code): Remove macro.
[!__GNUC_PREREQ (3, 4) && __FAST_MATH__] (atan2): Remove inline
function.
(floor): Remove inline function.
(ceil): Likewise.
[__FAST_MATH__] (__ldexp_code): Remove macro.
[__FAST_MATH__] (ldexp): Remove inline function.
[__FAST_MATH__ && __USE_ISOC99] (ldexpf): Likewise.
[__FAST_MATH__ && __USE_ISOC99] (ldexpl): Likewise.
[__FAST_MATH__ && __USE_ISOC99] (rint): Likewise.
[__USE_ISOC99] (__lrint_code): Remove macro.
[__USE_ISOC99] (__llrint_code): Likewise.
[__USE_ISOC99] (lrintf): Remove inline function.
[__USE_ISOC99] (lrint): Likewise.
[__USE_ISOC99] (lrintl): Likewise.
[__USE_ISOC99] (llrint): Likewise.
[__USE_ISOC99] (llrintf): Likewise.
[__USE_ISOC99] (llrintl): Likewise.
In accordance with the general principle of preferring to let the
compiler optimize function calls based on their standard semantics
rather than putting inline definitions of such functions in installed
headers, this patch removes various such inline definitions in the x86
bits/mathinline.h that were already disabled for GCC 3.5 or later and
so were only used with very old compilers (for which good optimization
is particularly unimportant); along with those inlines, a definition
of __M_SQRT2, which was only used in such inline functions, is also
removed. This is similar to an early step in removing the string.h
inlines; I intend to follow up with further removals of
bits/mathinline.h inline definitions in appropriate logical groups
(with GCC bugs filed in cases where GCC doesn't already support
corresponding optimizations).
Tested for x86_64 and x86.
* sysdeps/x86/fpu/bits/mathinline.h [!__GNUC_PREREQ (3, 4)]
(lrintf): Remove definitions used only with old GCC.
[!__GNUC_PREREQ (3, 4)] (lrint): Likewise.
[!__GNUC_PREREQ (3, 4)] (llrintf): Likewise.
[!__GNUC_PREREQ (3, 4)] (llrint): Likewise.
[!__GNUC_PREREQ (3, 4)] (fmaxf): Likewise.
[!__GNUC_PREREQ (3, 4)] (fmax): Likewise.
[!__GNUC_PREREQ (3, 4)] (fminf): Likewise.
[!__GNUC_PREREQ (3, 4)] (fmin): Likewise.
[!__GNUC_PREREQ (3, 4)] (rint): Likewise.
[!__GNUC_PREREQ (3, 4)] (rintf): Likewise.
[!__GNUC_PREREQ (3, 4)] (nearbyint): Likewise.
[!__GNUC_PREREQ (3, 4)] (nearbyintf): Likewise.
[!__GNUC_PREREQ (3, 4)] (ceil): Likewise.
[!__GNUC_PREREQ (3, 4)] (ceilf): Likewise.
[!__GNUC_PREREQ (3, 4)] (floor): Likewise.
[!__GNUC_PREREQ (3, 4)] (floorf): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 5)] (tan): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 5)] (fmod): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 4)] (sin): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 4)] (cos): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 5)] (log10): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 5)] (asin): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 5)] (acos): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 4)] (atan): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 5)] (log1p): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 5)] (logb): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 5)] (log2): Likewise.
[__FAST_MATH__ && !__GNUC_PREREQ (3, 5)] (drem): Likewise.
[__FAST_MATH__] (__M_SQRT2): Remove macro.
The mutually misaligned inputs on aarch64 are compared with a simple
byte copy, which is not very efficient. Enhance the comparison
similar to strcmp by loading a double-word at a time. The peak
performance improvement (i.e. 4k maxlen comparisons) due to this on
the strncmp microbenchmark is as follows:
falkor: 3.5x (up to 72% time reduction)
cortex-a73: 3.5x (up to 71% time reduction)
cortex-a53: 3.5x (up to 71% time reduction)
All mutually misaligned inputs from 16 bytes maxlen onwards show
upwards of 15% improvement and there is no measurable effect on the
performance of aligned/mutually aligned inputs.
* sysdeps/aarch64/strncmp.S (count): New macro.
(strncmp): Store misaligned length in SRC1 in COUNT.
(mutual_align): Adjust.
(misaligned8): Load dword at a time when it is safe.
* sysdeps/mach/hurd/reboot.c: Include <hurd/paths.h>
(reboot): Lookup _SERVERS_STARTUP instead of calling proc_getmsgport to get a
port to the startup server.
Jeff Law noticed that native PowerPC builds were broken by my having
made math_ldbl_opt.h not include math.h. nldbl-compat.c formerly got
math.h via libioP.h and math_ldbl_opt.h, *without* __NO_LONG_DOUBLE_MATH;
after my change it got it via nldbl-compat.h *with* __NO_LONG_DOUBLE_MATH,
but __NO_LONG_DOUBLE_MATH mode is forbidden on hosts that define
__HAVE_DISTINCT_FLOAT128, so the build breaks. This is the quick fix.
* sysdeps/ieee754/ldbl-opt/nldbl-compat.c: Include math.h
before nldbl-compat.h.
The sysdeps/ieee754/ldbl-opt version of math_ldbl_opt.h includes
math.h and math_private.h, despite not having any need for those
headers itself; the sysdeps/generic version doesn't. About 20 files
are relying on math_ldbl_opt.h to include math.h and/or math_private.h
for them, even though none of them necessarily used on a platform that
needs ldbl-opt support.
* sysdeps/ieee754/ldbl-opt/math_ldbl_opt.h: Don't include
math.h or math_private.h.
* sysdeps/alpha/fpu/s_isnan.c
* sysdeps/ieee754/ldbl-128ibm/s_ceill.c
* sysdeps/ieee754/ldbl-128ibm/s_floorl.c
* sysdeps/ieee754/ldbl-128ibm/s_llrintl.c
* sysdeps/ieee754/ldbl-128ibm/s_llroundl.c
* sysdeps/ieee754/ldbl-128ibm/s_lrintl.c
* sysdeps/ieee754/ldbl-128ibm/s_lroundl.c
* sysdeps/ieee754/ldbl-128ibm/s_rintl.c
* sysdeps/ieee754/ldbl-128ibm/s_roundl.c
* sysdeps/ieee754/ldbl-128ibm/s_truncl.c
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/e_hypot.c
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/e_hypotf.c:
* sysdeps/powerpc/powerpc64/fpu/multiarch/e_expf.c
* sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypot.c
* sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypotf.c:
Include math_private.h.
* sysdeps/ieee754/ldbl-64-128/s_finitel.c
* sysdeps/ieee754/ldbl-64-128/s_fpclassifyl.c
* sysdeps/ieee754/ldbl-64-128/s_isinfl.c
* sysdeps/ieee754/ldbl-64-128/s_isnanl.c
* sysdeps/ieee754/ldbl-64-128/s_signbitl.c
* sysdeps/powerpc/power7/fpu/s_logb.c:
Include math.h and math_private.h.
On Alpha, the register $at is, by default, reserved for use by the
assembler, in the expansion of pseudo-instructions. It's also used
by the special calling convention for _mcount. We get warnings from
Alpha clone.S because the code to call _mcount isn't properly marked
up to tell the assembler not to use $at itself.
* sysdeps/unix/sysv/linux/alpha/clone.s (__clone): Wrap manual
uses of $at in .set noat / .set at.
Since __libc_longjmp is a private interface for cancellation implementation
in libpthread, there is no need to provide hidden __libc_longjmp in libc.
Tested with build-many-glibcs.py.
* include/setjmp.h (__libc_longjmp): Remove libc_hidden_proto.
* setjmp/longjmp.c (__libc_longjmp): Remove libc_hidden_def.
* sysdeps/s390/longjmp.c (__libc_longjmp): Likewise.
* sysdeps/unix/sysv/linux/sparc/sparc64/longjmp.S (__libc_longjmp):
Likewise.
On sparc32 tst-makecontext fails, as backtrace called within a context
created by makecontext to yield infinite backtrace.
Fix that the same way than nios2 by adding a nop just before
__startcontext. This is needed as otherwise FDE lookup just repeatedly
finds __setcontext's FDE in an infinite loop, due to the convention of
using 'address - 1' for FDE lookup.
Changelog:
[BZ #22919]
* sysdeps/unix/sysv/linux/sparc/sparc32/setcontext.S (__startcontext):
Add nop before __startcontext, add explaining comments.
Some SPE opcodes clashes with some recent PowerISA opcodes and
until recently gas did not complain about it. However binutils
recently changed it and now VLE configured gas does not support to
assembler some instruction that might class with VLE (HTM for
instance). It also does not help that glibc build hardware lock
elision support as default (regardless of assembler support).
Although runtime will not actually enables TLE on SPE hardware
(since kernel will not advertise it), I see little advantage on
adding HTM support on SPE built glibc. SPE uses an incompatible
ABI which does not allow share the same build with default
powerpc and HTM code slows down SPE without any benefict.
This patch fixes it by only building HTM when SPE configuration
is not used.
Checked with a powerpc-linux-gnuspe build. I also did some sniff
tests on a e500 hardware without any issue.
[BZ #22926]
* sysdeps/powerpc/powerpc32/sysdep.h (ABORT_TRANSACTION_IMPL): Define
empty for __SPE__.
* sysdeps/powerpc/sysdep.h (ABORT_TRANSACTION): Likewise.
* sysdeps/unix/sysv/linux/powerpc/elision-lock.c (__lll_lock_elision):
Do not build hardware transactional code for __SPE__.
* sysdeps/unix/sysv/linux/powerpc/elision-trylock.c
(__lll_trylock_elision): Likewise.
* sysdeps/unix/sysv/linux/powerpc/elision-unlock.c
(__lll_unlock_elision): Likewise.
This patch refactors the ARCH_FORK macro and the required architecture
specific header to simplify the required architecture definitions
to provide the fork syscall semantic and proper document current
Linux clone ABI variant.
Instead of require the reimplementation of arch-fork.h header, this
patch changes the ARCH_FORK to an inline function with clone ABI
defined by kernel-features.h define. The generic kernel ABI meant
for newer ports is used as default and redefine if the architecture
requires.
Checked on x86_64-linux-gnu and i686-linux-gnu. Also with a build
for all the afected ABIs.
* sysdeps/nptl/fork.c (ARCH_FORK): Replace by auch_fork.
* sysdeps/unix/sysv/linux/alpha/arch-fork.h: Remove file.
* sysdeps/unix/sysv/linux/riscv/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/aarch64/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/arm/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/hppa/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/i386/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/ia64/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/m68k/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/microblaze/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/mips/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/nios2/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/powerpc/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/s390/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/sh/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/sparc/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/tile/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/x86_64/arch-fork.h: Likewise.
* sysdeps/unix/sysv/linux/arch-fork.h (arch_fork): New function.
* sysdeps/unix/sysv/linux/aarch64/kernel-features.h: New file.
* sysdeps/unix/sysv/linux/riscv/kernel-features.h: Likewise.
* sysdeps/unix/sysv/linux/arm/kernel-features.h
(__ASSUME_CLONE_BACKWARDS): Define.
* sysdeps/unix/sysv/linux/createthread.c (ARCH_CLONE): Define to
__clone2 if __NR_clone2 is defined.
* sysdeps/unix/sysv/linux/hppa/kernel-features.h
(__ASSUME_CLONE_BACKWARDS): Likewise.
* sysdeps/unix/sysv/linux/i386/kernel-features.h
(__ASSUME_CLONE_BACKWARDS): Likewise.
* sysdeps/unix/sysv/linux/ia64/kernel-features.h
(__ASSUME_CLONE2): Likewise.
* sysdeps/unix/sysv/linux/microblaze/kernel-features.h
(__ASSUME_CLONE_BACKWARDS3): Likewise.
* sysdeps/unix/sysv/linux/kernel-features.h: Document possible clone
variants and the define architecture can use.
(__ASSUME_CLONE_DEFAULT): Define as default.
* sysdeps/unix/sysv/linux/mips/kernel-features.h
(__ASSUME_CLONE_BACKWARDS): Likewise.
* sysdeps/unix/sysv/linux/powerpc/kernel-features.h
(__ASSUME_CLONE_BACKWARDS): Likewise.
* sysdeps/unix/sysv/linux/s390/kernel-features.h
(__ASSUME_CLONE_BACKWARDS2): Likewise.
I goofed up when changing the loop8 name to loop16 and missed on out
the branch instance. Fixed and actually build tested this time.
* sysdeps/aarch64/memcmp.S (more16): Fix branch target loop16.
This improved memcmp provides a fast path for compares up to 16 bytes
and then compares 16 bytes at a time, thus optimizing loads from both
sources. The glibc memcmp microbenchmark retains performance (with an
error of ~1ns) for smaller compare sizes and reduces up to 31% of
execution time for compares up to 4K on the APM Mustang. On Qualcomm
Falkor this improves to almost 48%, i.e. it is almost 2x improvement
for sizes of 2K and above.
* sysdeps/aarch64/memcmp.S: Widen comparison to 16 bytes at a
time.
* sysdeps/mach/hurd/bits/stat.h [__USE_ATFILE] (UTIME_NOW,
UTIME_OMIT): New macros.
* sysdeps/mach/hurd/futimens.c (__futimens): Try to use __file_utimens
before reverting to converting time spec to time value and calling
__file_utimes.
* sysdeps/mach/hurd/utime-helper.c: New file.
* sysdeps/mach/hurd/futimes.c: Include "utime-helper.c".
(__futimes): Try to use utime_ts_from_tval and __file_utimens before
reverting to utime_tvalue_from_tval and __file_utimes.
* sysdeps/mach/hurd/lutimes.c: Include "utime-helper.c".
(__lutimes): Just call hurd_futimens after lookup.
* sysdeps/mach/hurd/utimes.c: Likewise.
Building glibc for s390 with -Os (32-bit only, with GCC 7) fails with:
In file included from ../sysdeps/s390/multiarch/8bit-generic.c:370:0,
from ebcdic-at-de.c:28:
../iconv/loop.c: In function '__to_generic_vx':
../iconv/loop.c:264:22: error: 'ch' may be used uninitialized in this function [-Werror=maybe-uninitialized]
if (((Character) >> 7) == (0xe0000 >> 7)) \
^~
In file included from ebcdic-at-de.c:28:0:
../sysdeps/s390/multiarch/8bit-generic.c:340:15: note: 'ch' was declared here
uint32_t ch; \
^
../iconv/loop.c:325:7: note: in expansion of macro 'BODY'
BODY
^~~~
It's fairly easy to see, looking at the (long) expansion of the BODY
macro, that this is a false positive and the relevant variable 'ch' is
always initialized before use, in one of two possible places. As
such, disabling the warning for -Os with the DIAG_* macros is the
natural approach to fix this build failure. However, because of the
location at which the warning is reported, the disabling needs to go
in iconv/loop.c, around the definition of UNICODE_TAG_HANDLER (not
inside the definition), as that macro definition is where the
uninitialized use is reported, whereas the code that needs to be
reasoned about to see that the warning is a false positive is in the
definition of BODY elsewhere.
Thus, the patch adds such disabling in iconv/loop.c, with a comment
pointing to the s390-specific code and a comment in the s390-specific
code pointing to the generic file to alert people to the possible need
to update one place when changing the other. It would be possible if
desired to use #ifdef __s390__ around the disabling, though in general
we try to avoid that sort of thing in generic files. (Or some
extremely specialized macros for "disable -Wmaybe-uninitialized in
this particular place" could be specified, defined to 0 in a lot of
different files that include iconv/loop.c and to 1 in that particular
s390 file.)
Tested that this fixed -Os compilation for s390-linux-gnu with
build-many-glibcs.py.
* iconv/loop.c (UNICODE_TAG_HANDLER): Disable
-Wmaybe-uninitialized for -Os.
* sysdeps/s390/multiarch/8bit-generic.c (BODY): Add comment about
this disabling.
This patch defines _DIRENT_MATCHES_DIRENT64 to either 0 or 1 and adjust its
usage from checking its definition to its value.
Checked on a build for major Linux abis.
* bits/dirent.h (__INO_T_MATCHES_INO64_T): Define regardless whether
__INO_T_MATCHES_INO64_T is defined.
* sysdeps/unix/sysv/linux/bits/dirent.h: Likewise.
* dirent/alphasort.c: Check _DIRENT_MATCHES_DIRENT64 value instead
of definition.
* dirent/alphasort64.c: Likewise.
* dirent/scandir.c: Likewise.
* dirent/scandir64-tail.c: Likewise.
* dirent/scandir64.c: Likewise.
* dirent/scandirat.c: Likewise.
* dirent/scandirat64.c: Likewise.
* dirent/versionsort.c: Likewise.
* dirent/versionsort64.c: Likewise.
* include/dirent.h: Likewise.
Now that send might be implemented calling sendto syscall on Linux,
I am seeing some issue in some kernel configurations where tst-cancel4
sendto do not block as expected.
The socket used to force the syscall blocking is used with default
system configuration for buffer sending size, which might not be
suffice to force blocking. This patch fixes it by explicit setting
buffer socket lower than the buffer size used. It also enables sendto
cancellation tests to work in both ways (since internally send is
implemented routing to sendto on Linux kernel).
The patch also removes unrequired make rules on some archictures
for send/recv. The generic nptl Makefile already set the compiler flags
required on some architectures for correct unwinding and libc object
are not strictly required to support unwind (since pthread_cancel
requires linking against libpthread).
Checked on aarch64-linux-gnu and x86_64-linux-gnu. I also did a
sniff test with tst-cancel{4,5} on a simulated mips64-linux-gnu.
* nptl/tst-cancel4-common.h (set_socket_buffer): New function.
* nptl/tst-cancel4-common.c (do_test): Call set_socket_buffer
for socketpair endpoint.
* nptl/tst-cancel4.c (tf_send): Call set_socket_buffer and use
WRITE_BUFFER_SIZE as buffer size for sending socket.
(tf_sendto): Use SOCK_STREAM instead of SOCK_DGRAM and fix an
issue on system where send is implemented with sendto syscall.
* sysdeps/unix/sysv/linux/mips/mips64/Makefile [$(subdir) = socket]
(CFLAGS-recv.c, CFLAGS-send.c): Remove rules.
[$(subdir) = nptl] (CFLAGS-recv.c, CFLAGS-send.c): Likewise.
* sysdeps/unix/sysv/linux/riscv/rv64/Makefile: Remove file.
This patch fixes the i386 sa_restorer field initialization for sigaction
syscall for kernel with vDSO. As described in bug report, i386 Linux
(and compat on x86_64) interprets SA_RESTORER clear with nonzero
sa_restorer as a request for stack switching if the SS segment is 'funny'.
This means that anything that tries to mix glibc's signal handling with
segmentation (for instance through modify_ldt syscall) is randomly broken
depending on what values lands in sa_restorer.
The testcase added is based on Linux test tools/testing/selftests/x86/ldt_gdt.c,
more specifically in do_multicpu_tests function. The main changes are:
- C11 atomics instead of plain access.
- Remove x86_64 support which simplifies the syscall handling and fallbacks.
- Replicate only the test required to trigger the issue.
Checked on i686-linux-gnu.
[BZ #21269]
* sysdeps/unix/sysv/linux/i386/Makefile (tests): Add tst-bz21269.
* sysdeps/unix/sysv/linux/i386/sigaction.c (SET_SA_RESTORER): Clear
sa_restorer for vDSO case.
* sysdeps/unix/sysv/linux/i386/tst-bz21269.c: New file.
so interfaces needing it can get it.
* stdlib/errno.h (error_t): Move definition to...
* bits/types/error_t.h: ... new header.
* stdlib/Makefile (headers): Add bits/types/error_t.h.
* sysdeps/mach/hurd/bits/errno.h (error_t): Move definition to...
* sysdeps/mach/hurd/bits/types/error_t.h: ... new header.
* sysdeps/mach/hurd/errnos.awk (error_t): Likewise.
* hurd/hurd.h: Include <bits/types/error_t.h>
* hurd/hurd/fd.h: Include <bits/types/error_t.h>
* hurd/hurd/id.h: Include <errno.h> and <bits/types/error_t.h>
* hurd/hurd/lookup.h: Include <errno.h> and <bits/types/error_t.h>
* hurd/hurd/resource.h: Include <bits/types/error_t.h>
* hurd/hurd/signal.h: Include <bits/types/error_t.h>
* hurd/hurd/sigpreempt.h: Include <bits/types/error_t.h>
Compiling the testsuite for powerpc (multi-arch configurations) with
-Os with GCC 7 fails with:
In file included from ifuncmod1.c:7:0,
from ifuncdep1.c:3:
../sysdeps/powerpc/ifunc-sel.h: In function 'ifunc_sel':
../sysdeps/powerpc/ifunc-sel.h:12:3: error: asm operand 2 probably doesn't match constraints [-Werror]
__asm__ ("mflr 12\n\t"
^~~~~~~
../sysdeps/powerpc/ifunc-sel.h:12:3: error: asm operand 3 probably doesn't match constraints [-Werror]
../sysdeps/powerpc/ifunc-sel.h:12:3: error: asm operand 4 probably doesn't match constraints [-Werror]
../sysdeps/powerpc/ifunc-sel.h:12:3: error: impossible constraint in 'asm'
The "i" constraints on function pointers require the function call to
be inlined so the compiler can see the constant function pointer
arguments passed to the asm. This patch marks the relevant functions
as always_inline accordingly.
Tested that this fixes the -Os testsuite build for
powerpc-linux-gnu-power4, powerpc64-linux-gnu, powerpc64le-linux-gnu
with build-many-glibcs.py.
* sysdeps/powerpc/ifunc-sel.h (ifunc_sel): Make always_inline.
(ifunc_one): Likewise.
libpthread_nonshared.a is unused after this, so remove it from the
build.
There is no ABI impact because pthread_atfork was implemented using
__register_atfork in libc even before this change.
pthread_atfork has to be a weak alias because pthread_* names are not
reserved in libc.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
As discussed in bug 22902, the i386 fenv_private.h implementation has
problems for float128 for the case of 32-bit glibc built with libgcc
from GCC configured using --with-fpmath=sse.
The optimized floating-point state handling in fenv_private.h needs to
know which floating-point state - x87 or SSE - is used for each
floating-point type, so that only one state needs updating / testing
for libm code using that state internally. On 32-bit x86, the x87
rounding mode is always used for float128, but the x87 exception flags
are only used when libgcc is built using x87 floating-point
arithmetic; if libgcc is built for SSE arithmetic, the SSE exception
flags are used.
The choice of arithmetic with which libgcc is built is independent of
that with which glibc is built. Thus, since glibc cannot tell the
choice used in libgcc, the default implementations of
libc_feholdexcept_setroundf128 and libc_feupdateenv_testf128 (which
use the <fenv.h> functions, thus using both x87 and SSE state on
processors that have both) need to be used; this patch updates the
code accordingly.
Tested for 32-bit x86; HJ reports testing in the --with-fpmath=sse
case.
[BZ #22902]
* sysdeps/i386/fpu/fenv_private.h [!__x86_64__]
(libc_feholdexcept_setroundf128): New macro.
[!__x86_64__] (libc_feupdateenv_testf128): Likewise.
On sparc, localplt test failures appear when building with -Os because
of a call to strtoumax from
sysdeps/unix/sysv/linux/sparc/sparc64/get_clockfreq.c, and strtoumax
is not inlined when building with -Os. This patch fixes those
failures by using libc_hidden_proto and libc_hidden_def for strtoumax.
Tested with build-many-glibcs.py for
sparc64-linux-gnu-disable-multi-arch, sparc64-linux-gnu,
sparcv9-linux-gnu-disable-multi-arch, sparcv9-linux-gnu that this
fixes that test failure with -Os.
[BZ #15105]
* sysdeps/wordsize-32/strtoumax.c (strtoumax): Use
libc_hidden_def.
* sysdeps/wordsize-64/strtoumax.c (strtoumax): Likewise.
* include/inttypes.h: New file.
* sysdeps/pthread/timer_routines.c: Include <timer_routines.h> instead
of <nptl/pthreadP.h>
(thread_attr_compare): Move function to...
* sysdeps/nptl/timer_routines.h: ... new header.
Linux ptrace headers define macros whose tokens conflict with the
constants of enum __ptrace_request causing build errors when
asm/ptrace.h or linux/ptrace.h are included before sys/ptrace.h.
* sysdeps/unix/sysv/linux/powerpc/sys/ptrace.h: Undefine Linux
macros used in __ptrace_request.
Signed-off-by: Tulio Magno Quites Machado Filho <tuliom@linux.vnet.ibm.com>
Glibc build generates header files to define constants from special .sym
files. If a .sym file includes the same header file which it generates,
it leads to circular dependency which may lead to build hang on a
many-core machine. Define GEN_AS_CONST_HEADERS when generating header
files to avoid circular dependency.
<tcb-offsets.h> is needed for i686 and it isn't needed for x86-64 at
least since glibc 2.23.
Tested on i686 and x86-64.
[BZ #22792]
* Makerules ($(common-objpfx)%.h): Pass -DGEN_AS_CONST_HEADERS
to $(CC).
* sysdeps/unix/sysv/linux/i386/lowlevellock.h: Include
<tcb-offsets.h> only if GEN_AS_CONST_HEADERS isn't defined.
* sysdeps/unix/sysv/linux/x86_64/lowlevellock.h: Don't include
<tcb-offsets.h>.
Current implementation (sysdeps/nptl/fork.c) replicates the atfork
handlers list backward to invoke the child handlers after fork/clone
syscall.
The internal atfork handlers is implemented as a single-linked list
so a lock-free algorithm can be used, trading fork mulithread call
performance for some code complexity and dynamic stack allocation
(since the backwards list should not fail).
This patch refactor it to use a dynarary instead of a linked list.
It simplifies the external variables need to be exported and also
the internal atfork handler member definition.
The downside is a serialization of fork call in multithread, since to
operate on the dynarray the internal lock should be used. However
as noted by Florian, it already acquires external locks for malloc
and libio so it is already hitting some lock contention. Besides,
posix_spawn should be faster and more scalable to run external programs
in multithread environments.
Checked on x86_64-linux-gnu.
* nptl/Makefile (routines): Remove unregister-atfork.
* nptl/register-atfork.c (fork_handler_pool): Remove variable.
(fork_handler_alloc): Remove function.
(fork_handlers, fork_handler_init): New variables.
(__fork_lock): Rename to atfork_lock.
(__register_atfork, __unregister_atfork, libc_freeres_fn): Rewrite
to use a dynamic array to add/remove atfork handlers.
* sysdeps/nptl/fork.c (__libc_fork): Likewise.
* sysdeps/nptl/fork.h (__fork_lock, __fork_handlers, __linkin_atfork):
Remove declaration.
(fork_handler): Remove next, refcntr, and need_signal member.
(__run_fork_handler_type): New enum.
(__run_fork_handlers): New prototype.
* sysdeps/nptl/libc-lockP.h (__libc_atfork): Remove declaration.
This patch renames the nptl-signals.h header to internal-signals.h.
On Linux the definitions and functions are not only NPTL related, but
used for other POSIX definitions as well (for instance SIGTIMER for
posix times, SIGSETXID for id functions, and signal block/restore
helpers) and since generic functions will be places and used in generic
implementation it makes more sense to decouple it from NPTL.
Checked on x86_64-linux-gnu.
* sysdeps/nptl/nptl-signals.h: Move to ...
* sysdeps/generic/internal-signals.h: ... here. Adjust internal
comments.
* sysdeps/unix/sysv/linux/internal-signals.h: Add include guards.
(__nptl_is_internal_signal): Rename to __is_internal_signal.
(__nptl_clear_internal_signals): Rename to __clear_internal_signals.
* sysdeps/unix/sysv/linux/raise.c: Adjust nptl-signal.h to
include-signals.h rename.
* nptl/pthreadP.h: Likewise.
* sysdeps/unix/sysv/linux/spawni.c (__spawni_child): Call
__is_internal_signal instead of __nptl_is_internal_signal.
I accidentally set the loop jump back label as misaligned8 instead of
do_misaligned. The typo is harmless but it's always nice to not have
to unnecessarily execute those two instructions.
* sysdeps/aarch64/strcmp.S (do_misaligned): Jump back to
do_misaligned, not misaligned8.
* sysdeps/aarch64/multiarch/Makefile (sysdep_routines):
Add memcpy_thunderx2.
* sysdeps/aarch64/multiarch/ifunc-impl-list.c (MAX_IFUNC):
Increment to 4.
(__libc_ifunc_impl_list): Add __memcpy_thunderx2.
* sysdeps/aarch64/multiarch/memcpy.c (libc_ifunc): Add IS_THUNDERX2
and IS_THUNDERX2PA checks.
* sysdeps/aarch64/multiarch/memcpy_thunderx.S (USE_THUNDERX2):
Use macro to set name appropriately.
(memcpy): Use USE_THUNDERX2 macro to modify prefetches.
* sysdeps/aarch64/multiarch/memcpy_thunderx2.S: New file.
* sysdeps/unix/sysv/linux/aarch64/cpu-features.h (IS_THUNDERX2PA):
New macro.
(IS_THUNDERX2): New macro.
After regenerating ULPs from scratch in
commit 8e7196c875, I've missed
to test it with multiple gcc versions. Hence, here is a further update.
ChangeLog:
* sysdeps/s390/fpu/libm-test-ulps: Regenerated.
