platform/upstream/glibc.git
5 years agoS390: Test that lazy binding does not clobber R0
Ilya Leoshkevich [Fri, 10 Aug 2018 07:07:43 +0000 (09:07 +0200)]
S390: Test that lazy binding does not clobber R0

The following combinations need to be tested:
* 32- (g5, esa and zarch) and 64-bit
  * linux32 glibc/configure CC='gcc -m31 -march=g5'
  * linux32 glibc/configure CC='gcc -m31'
  * linux32 glibc/configure CC='gcc -m31 -mzarch'
* With and without VX:
  * glibc/configure libc_cv_asm_s390_vx=no
* With and without profiling (using LD_PROFILE)
* With and without pltexit (using LD_AUDIT)

ChangeLog:

* sysdeps/s390/Makefile: Register the new tests.
* sysdeps/s390/tst-dl-runtime-mod.S: New file.
* sysdeps/s390/tst-dl-runtime-profile-audit.c: New file.
* sysdeps/s390/tst-dl-runtime-profile-noaudit.c: New file.
* sysdeps/s390/tst-dl-runtime-resolve-audit.c: New file.
* sysdeps/s390/tst-dl-runtime-resolve-noaudit.c: New file.
* sysdeps/s390/tst-dl-runtime.c: New file.

5 years agoS390: Do not clobber R0 in 64-bit _dl_runtime_profile
Ilya Leoshkevich [Fri, 10 Aug 2018 07:07:43 +0000 (09:07 +0200)]
S390: Do not clobber R0 in 64-bit _dl_runtime_profile

Preparation for the usage of R0 by __fentry__.

ChangeLog:

* sysdeps/s390/s390-64/dl-trampoline.h (_dl_runtime_profile):
Do not clobber R0.

5 years agoS390: Do not clobber R0 in 64-bit _dl_runtime_resolve
Ilya Leoshkevich [Fri, 10 Aug 2018 07:07:43 +0000 (09:07 +0200)]
S390: Do not clobber R0 in 64-bit _dl_runtime_resolve

Preparation for the usage of R0 by __fentry__.

ChangeLog:

* sysdeps/s390/s390-64/dl-trampoline.h (_dl_runtime_resolve):
Do not clobber R0.

5 years agoS390: Do not clobber R0 in 32-bit _dl_runtime_profile
Ilya Leoshkevich [Fri, 10 Aug 2018 07:07:43 +0000 (09:07 +0200)]
S390: Do not clobber R0 in 32-bit _dl_runtime_profile

Preparation for the usage of R0 by __fentry__.

ChangeLog:

* sysdeps/s390/s390-32/dl-trampoline.h (_dl_runtime_profile):
Do not clobber R0.

5 years agoS390: Do not clobber R0 in 32-bit _dl_runtime_resolve
Ilya Leoshkevich [Fri, 10 Aug 2018 07:07:42 +0000 (09:07 +0200)]
S390: Do not clobber R0 in 32-bit _dl_runtime_resolve

Preparation for the usage of R0 by __fentry__.

ChangeLog:

* sysdeps/s390/s390-32/dl-trampoline.h (_dl_runtime_resolve):
Do not clobber R0.

5 years agoS390: Use symbolic offsets for stack variables in 64-bit _dl_runtime_profile
Ilya Leoshkevich [Fri, 10 Aug 2018 07:07:42 +0000 (09:07 +0200)]
S390: Use symbolic offsets for stack variables in 64-bit _dl_runtime_profile

Maintainability improvement.

ChangeLog:

* sysdeps/s390/s390-64/dl-trampoline.h (_dl_runtime_profile):
Use symbolic offsets for stack variables.

5 years agoS390: Use symbolic offsets for stack variables in 64-bit _dl_runtime_resolve
Ilya Leoshkevich [Fri, 10 Aug 2018 07:07:42 +0000 (09:07 +0200)]
S390: Use symbolic offsets for stack variables in 64-bit _dl_runtime_resolve

Maintainability improvement.

ChangeLog:

* sysdeps/s390/s390-64/dl-trampoline.h (_dl_runtime_resolve):
Use symbolic offsets for stack variables.

5 years agoS390: Use symbolic offsets for stack variables in 32-bit _dl_runtime_profile
Ilya Leoshkevich [Fri, 10 Aug 2018 07:07:41 +0000 (09:07 +0200)]
S390: Use symbolic offsets for stack variables in 32-bit _dl_runtime_profile

Maintainability improvement.

ChangeLog:

* sysdeps/s390/s390-32/dl-trampoline.h (_dl_runtime_profile):
Use symbolic offsets for stack variables.

5 years agoS390: Use symbolic offsets for stack variables in 32-bit _dl_runtime_resolve
Ilya Leoshkevich [Fri, 10 Aug 2018 07:07:41 +0000 (09:07 +0200)]
S390: Use symbolic offsets for stack variables in 32-bit _dl_runtime_resolve

Maintainability improvement.

ChangeLog:

* sysdeps/s390/s390-32/dl-trampoline.h (_dl_runtime_resolve):
Use symbolic offsets for stack variables.

5 years agoReplace gen-libm-test.pl with gen-libm-test.py.
Joseph Myers [Thu, 9 Aug 2018 15:34:12 +0000 (15:34 +0000)]
Replace gen-libm-test.pl with gen-libm-test.py.

Following the recent discussion of using Python instead of Perl and
Awk for glibc build / test, this patch replaces gen-libm-test.pl with
a new gen-libm-test.py script.  This script should work with all
Python versions supported by glibc (tested by hand with Python 2.7,
tested in the build system with Python 3.5; configure prefers Python 3
if available).

This script is designed to give identical output to gen-libm-test.pl
for ease of verification of the change, except for generated comments
referring to .py instead of .pl.  (That is, identical for actual
inputs passed to the script, not necessarily for all possible input;
for example, this version more precisely follows the C standard syntax
for floating-point constants when deciding when to add LIT macro
calls.)  In one place a comment notes that the generation of
NON_FINITE flags is replicating a bug in the Perl script to assist in
such comparisons (with the expectation that this bug can then be
separately fixed in the Python script later).

Tested for x86_64, including comparison of generated files (and hand
testing of the case of generating a sorted libm-test-ulps file, which
isn't covered by normal "make check").

I'd expect to follow this up by extending the new script to produce
the ulps tables for the manual as well (replacing
manual/libm-err-tab.pl, so that then we just have one ulps file
parser) - at which point the manual build would depend on both Perl
and Python (eliminating the Perl dependency would require someone to
rewrite summary.pl in Python, and that would only eliminate the
*direct* Perl dependency; current makeinfo is written in Perl so there
would still be an indirect dependency).

I think install.texi is more or less equally out-of-date regarding
Perl and Python uses before and after this patch, so I don't think
this patch depends on my patch
<https://sourceware.org/ml/libc-alpha/2018-08/msg00133.html> to update
install.texi regarding such uses (pending review).

* math/gen-libm-test.py: New file.
* math/gen-libm-test.pl: Remove.
* math/Makefile [$(PERL) != no]: Change condition to [PYTHON].
($(objpfx)libm-test-ulps.h): Use gen-libm-test.py instead of
gen-libm-test.pl.
($(libm-test-c-noauto-obj)): Likewise.
($(libm-test-c-auto-obj)): Likewise.
($(libm-test-c-narrow-obj)): Likewise.
(regen-ulps): Likewise.
* math/README.libm-test: Update references to gen-libm-test.pl.
* math/libm-test-driver.c (struct test_fj_f_data): Update comment
referencing gen-libm-test.pl.
* math/libm-test-nexttoward.inc (nexttoward_test_data): Likewise.
* math/libm-test-support.c: Likewise.
* math/libm-test-support.h: Likewise.
* sysdeps/generic/libm-test-ulps: Likewise.

5 years agohurd: Avoid PLTs for __pthread_get/setspecific
Samuel Thibault [Wed, 8 Aug 2018 23:26:19 +0000 (01:26 +0200)]
hurd: Avoid PLTs for __pthread_get/setspecific

* sysdeps/htl/pthreadP.h [IS_IN (libpthread)] (__pthread_getspecific,
__pthread_setspecific): Add hidden proto.
* sysdeps/htl/pt-getspecific.c (__pthread_getspecific): Add hidden def.
* sysdeps/htl/pt-setspecific.c (__pthread_setspecific): Add hidden def.

5 years ago[aarch64] Fix value of MIN_PAGE_SIZE for testing
Siddhesh Poyarekar [Wed, 8 Aug 2018 15:46:07 +0000 (21:16 +0530)]
[aarch64] Fix value of MIN_PAGE_SIZE for testing

MIN_PAGE_SIZE is normally set to 4096 but for testing it can be set to
16 so that it exercises the page crossing code for every misaligned
access.  The value was set to 15, which is obviously wrong, so fixed
as obvious and tested.

* sysdeps/aarch64/strlen.S [TEST_PAGE_CROSS](MIN_PAGE_SIZE):
Fix value.

5 years agoMove comment from libm-test-nextdown.inc to libm-test-nexttoward.inc.
Joseph Myers [Wed, 8 Aug 2018 16:16:59 +0000 (16:16 +0000)]
Move comment from libm-test-nextdown.inc to libm-test-nexttoward.inc.

When libm tests were split into separate per-function .inc files, a
comment relating to the nexttoward tests ended up at the end of
libm-test-nextdown.inc (because the split was based on starting each
function's tests with the <function>_test_data definition, which
failed to allow for comments before such definitions).  This patch
moves that comment to the correct location.

Tested for x86_64.

* math/libm-test-nextdown.inc (do_test): Move comment to ....
* math/libm-test-nexttoward.inc (nexttoward_test_data): ... here.

5 years agohurd: Add missing symbols for proper libc_get/setspecific
Samuel Thibault [Tue, 7 Aug 2018 23:55:04 +0000 (01:55 +0200)]
hurd: Add missing symbols for proper libc_get/setspecific

* htl/Versions (__pthread_getspecific, __pthread_setspecific): Add
symbols.

5 years agobenchtests: Clean up the alloc_bufs
Siddhesh Poyarekar [Tue, 7 Aug 2018 19:14:56 +0000 (00:44 +0530)]
benchtests: Clean up the alloc_bufs

Drop realloc_bufs in favour of making alloc_bufs transparently
reallocate the buffers if it had allocated before.  Also consolidate
computation of buffer lengths so that they don't get repeated on every
reallocation.

* benchtests/bench-string.h (buf1_size, buf2_size): New
variables.
(init_sizes): New function.
(test_init): Use it.
(alloc_buf, exit_error): New functions.
(alloc_bufs): Use ALLOC_BUF.
(realloc_bufs): Remove.
* benchtests/bench-memcmp.c (do_test): Adjust.
* benchtests/bench-memset-large.c (do_test): Likewise.
* benchtests/bench-memset-walk.c (do_test): Likewise.
* benchtests/bench-memset.c (do_test): Likewise.
* benchtests/bench-strncmp.c (do_test): Likewise.

5 years agoRISC-V: Don't use ps_get_thread_area in libthread_db (bug 23402)
Andreas Schwab [Thu, 12 Jul 2018 07:24:36 +0000 (09:24 +0200)]
RISC-V: Don't use ps_get_thread_area in libthread_db (bug 23402)

Since RISC-V stores the thread pointer in a general register libthread_db
can just ask the debugger for the register contents instead of trying to
call ps_get_thread_area.  This enables thread debugging in gdb.

* sysdeps/riscv/nptl/tls.h (DB_THREAD_SELF): Use REGISTER instead
of CONST_THREAD_AREA.

5 years agox86: Move STATE_SAVE_OFFSET/STATE_SAVE_MASK to sysdep.h
H.J. Lu [Mon, 6 Aug 2018 13:25:28 +0000 (06:25 -0700)]
x86: Move STATE_SAVE_OFFSET/STATE_SAVE_MASK to sysdep.h

Move STATE_SAVE_OFFSET and STATE_SAVE_MASK to sysdep.h to make
sysdeps/x86/cpu-features.h a C header file.

* sysdeps/x86/cpu-features.h (STATE_SAVE_OFFSET): Removed.
(STATE_SAVE_MASK): Likewise.
Don't check __ASSEMBLER__ to include <cpu-features-offsets.h>.
* sysdeps/x86/sysdep.h (STATE_SAVE_OFFSET): New.
(STATE_SAVE_MASK): Likewise.
* sysdeps/x86_64/dl-trampoline.S: Include <cpu-features-offsets.h>
instead of <cpu-features.h>.

5 years agoImprove ChangeLog message.
DJ Delorie [Fri, 3 Aug 2018 23:24:14 +0000 (19:24 -0400)]
Improve ChangeLog message.

5 years agoRegen RISC-V rvd ULPs
DJ Delorie [Fri, 3 Aug 2018 17:52:01 +0000 (13:52 -0400)]
Regen RISC-V rvd ULPs

* sysdeps/riscv/rv64/rvd/libm-test-ulps: Update.

Note: I regen'd these from scratch, but I'm only committing the
increases, as I only tested on hardware.  There were a few 2->1
decreases that I omitted, possibly "for now".

5 years agoRISC-V: Fix rounding save/restore bug.
DJ Delorie [Fri, 3 Aug 2018 17:43:31 +0000 (13:43 -0400)]
RISC-V: Fix rounding save/restore bug.

* sysdeps/riscv/rvf/math_private.h (libc_feholdexcept_setround_riscv):
Fix rounding save-restore bug.

Fixes about a hundred off-by-ULP failures in the math testsuite.

5 years agoConsistently terminate libm-test-*.inc TEST lines with commas.
Joseph Myers [Fri, 3 Aug 2018 16:56:02 +0000 (16:56 +0000)]
Consistently terminate libm-test-*.inc TEST lines with commas.

Some TEST_* lines in libm-test-*.inc end with semicolons not commas.
This works at present because gen-libm-test.pl ignores whatever comes
after the TEST_* call, but is logically wrong (since the TEST_* calls
generate array elements, not statements) and the Python replacement
for gen-libm-test.pl that I'm working on is stricter about the syntax
here.  This patch fixes the lines in question to use commas like most
such lines already do.

Tested for x86_64.

* math/libm-test-ilogb.inc (ilogb_test_data): Use ',' not ';'
after TEST_* calls.
* math/libm-test-llogb.inc (llogb_test_data): Likewise.
* math/libm-test-logb.inc (logb_test_data): Likewise.

5 years agoSimplify and speedup strstr/strcasestr first match
Wilco Dijkstra [Fri, 3 Aug 2018 16:24:12 +0000 (17:24 +0100)]
Simplify and speedup strstr/strcasestr first match

Looking at the benchtests, both strstr and strcasestr spend a lot of time
in a slow initialization loop handling one character per iteration.
This can be simplified and use the much faster strlen/strnlen/strchr/memcmp.
Read ahead a few cachelines to reduce the number of strnlen calls, which
improves performance by ~3-4%.  This patch improves the time taken for the
full strstr benchtest by >40%.

* string/strcasestr.c (STRCASESTR): Simplify and speedup first match.
* string/strstr.c (AVAILABLE): Likewise.

5 years agox86: Don't include <init-arch.h> in assembly codes
H.J. Lu [Fri, 3 Aug 2018 15:04:49 +0000 (08:04 -0700)]
x86: Don't include <init-arch.h> in assembly codes

There is no need to include <init-arch.h> in assembly codes since all
x86 IFUNC selector functions are written in C.  Tested on i686 and
x86-64.  There is no code change in libc.so, ld.so and libmvec.so.

* sysdeps/i386/i686/multiarch/bzero-ia32.S: Don't include
<init-arch.h>.
* sysdeps/x86_64/fpu/multiarch/svml_d_sin8_core-avx2.S: Likewise.
* sysdeps/x86_64/fpu/multiarch/svml_s_expf16_core-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S: Likewise.

5 years agoUpdate be translations.
Carlos O'Donell [Fri, 3 Aug 2018 14:13:47 +0000 (10:13 -0400)]
Update be translations.

5 years agoUpdate be translations.
Carlos O'Donell [Thu, 2 Aug 2018 18:48:10 +0000 (14:48 -0400)]
Update be translations.

5 years agox86: Cleanup cpu-features-offsets.sym
H.J. Lu [Fri, 3 Aug 2018 13:40:48 +0000 (06:40 -0700)]
x86: Cleanup cpu-features-offsets.sym

Remove the unused macros.  There is no code changes in libc.so nor
ld.so on i686 and x86-64.

* sysdeps/x86/cpu-features-offsets.sym
(rtld_global_ro_offsetof): Removed.
(CPU_FEATURES_SIZE): Likewise.
(CPUID_OFFSET): Likewise.
(CPUID_SIZE): Likewise.
(CPUID_EAX_OFFSET): Likewise.
(CPUID_EBX_OFFSET): Likewise.
(CPUID_ECX_OFFSET): Likewise.
(CPUID_EDX_OFFSET): Likewise.
(FAMILY_OFFSET): Likewise.
(MODEL_OFFSET): Likewise.
(FEATURE_OFFSET): Likewise.
(FEATURE_SIZ): Likewise.
(COMMON_CPUID_INDEX_1): Likewise.
(COMMON_CPUID_INDEX_7): Likewise.
(FEATURE_INDEX_1): Likewise.
(RTLD_GLOBAL_RO_DL_X86_CPU_FEATURES_OFFSET): Updated.

5 years agoFix ChangeLog date.
Carlos O'Donell [Thu, 2 Aug 2018 19:32:23 +0000 (15:32 -0400)]
Fix ChangeLog date.

5 years agoAdd convenience target 'install-locale-files'.
Carlos O'Donell [Thu, 26 Jul 2018 14:14:55 +0000 (10:14 -0400)]
Add convenience target 'install-locale-files'.

The convenience install target 'install-locale-files' is created
to allow distributions to install all of the SUPPORTED locales as
files instead of into the locale-archive.

You invoke the new convenience target like this:
make localedata/install-locale-files DESTDIR=<prefix>

5 years ago[benchtests] Fix compare_strings.py for python2
Siddhesh Poyarekar [Thu, 2 Aug 2018 18:56:45 +0000 (00:26 +0530)]
[benchtests] Fix compare_strings.py for python2

Python 2 does not have a FileNotFoundError so drop it in favour of
simply printing out the last (and most informative) line of the
exception.

* benchtests/scripts/compare_strings.py: Import traceback.
(parse_file): Pretty-print error.

5 years agoAdd ChangeLog for the last commit
Siddhesh Poyarekar [Thu, 2 Aug 2018 18:26:18 +0000 (23:56 +0530)]
Add ChangeLog for the last commit

5 years agoRename the glibc.tune namespace to glibc.cpu
Siddhesh Poyarekar [Thu, 2 Aug 2018 18:19:19 +0000 (23:49 +0530)]
Rename the glibc.tune namespace to glibc.cpu

The glibc.tune namespace is vaguely named since it is a 'tunable', so
give it a more specific name that describes what it refers to.  Rename
the tunable namespace to 'cpu' to more accurately reflect what it
encompasses.  Also rename glibc.tune.cpu to glibc.cpu.name since
glibc.cpu.cpu is weird.

* NEWS: Mention the change.
* elf/dl-tunables.list: Rename tune namespace to cpu.
* sysdeps/powerpc/dl-tunables.list: Likewise.
* sysdeps/x86/dl-tunables.list: Likewise.
* sysdeps/aarch64/dl-tunables.list: Rename tune.cpu to
cpu.name.
* elf/dl-hwcaps.c (_dl_important_hwcaps): Adjust.
* elf/dl-hwcaps.h (GET_HWCAP_MASK): Likewise.
* manual/README.tunables: Likewise.
* manual/tunables.texi: Likewise.
* sysdeps/powerpc/cpu-features.c: Likewise.
* sysdeps/unix/sysv/linux/aarch64/cpu-features.c
(init_cpu_features): Likewise.
* sysdeps/x86/cpu-features.c: Likewise.
* sysdeps/x86/cpu-features.h: Likewise.
* sysdeps/x86/cpu-tunables.c: Likewise.
* sysdeps/x86_64/Makefile: Likewise.
* sysdeps/x86/dl-cet.c: Likewise.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
5 years agoDo not define various fenv.h macros for MIPS soft-float (bug 23479).
Joseph Myers [Thu, 2 Aug 2018 15:53:29 +0000 (15:53 +0000)]
Do not define various fenv.h macros for MIPS soft-float (bug 23479).

MIPS soft-float glibc does not support floating-point exceptions and
rounding modes, and uses a different ABI from hard-float so a
soft-float compilation cannot use a glibc that does support
floating-point exceptions and rounding modes.  Thus, bits/fenv.h
should not, when compiling for soft-float, define macros for the
unsupported features.

This patch changes it accordingly to define those macros only for
hard-float.  None of the exception macros are defined for soft-float,
with FE_ALL_EXCEPT defined to 0 in that case, and only FE_TONEAREST is
defined of the rounding-mode macros, and FE_NOMASK_ENV is not defined;
this is consistent with how architectures lacking exception and
rounding mode support generally define things in this header.  As well
as making the header more correct for this case, this also means the
generic math_private.h optimizations for this case automatically apply
(inlining libm-internal fenv.h function calls that are trivial when
exceptions and rounding modes are not supported).

The mips64 sfp-machine.h then needs similar changes to disable more of
the exception and rounding mode handling for soft-float.  (The mips32
sfp-machine.h is already used only for soft-float, has no integration
with hardware exceptions or rounding modes and so needs no changes.)

Existing binaries might use the old FE_NOMASK_ENV value as an argument
to fesetenv / feupdateenv and expect an error for it (given that it
was defined in a header that also defined FE_ALL_EXCEPT to a nonzero
value).  To preserve that error, wrappers for the fallback fesetenv
and feupdateenv are created in sysdeps/mips/nofpu/.

Tested for mips64 (hard-float and soft-float, all three ABIs).

[BZ #23479]
* sysdeps/mips/bits/fenv.h (FE_INEXACT): Define only if
[__mips_hard_float].
(FE_UNDERFLOW): Likewise.
(FE_OVERFLOW): Likewise.
(FE_DIVBYZERO): Likewise.
(FE_INVALID): Likewise.
(FE_ALL_EXCEPT): Define to 0 if [!__mips_hard_float].
(FE_TOWARDZERO): Define only if [__mips_hard_float].
(FE_UPWARD): Likewise.
(FE_DOWNWARD): Likewise.
(__FE_UNDEFINED): Define if [!__mips_hard_float]
(FE_NOMASK_ENV): Define only if [__mips_hard_float].
* sysdeps/mips/mips64/sfp-machine.h (_FP_DECL_EX): Define only if
[__mips_hard_float].
(FP_ROUNDMODE): Likewise.
(FP_RND_NEAREST): Likewise.
(FP_RND_ZERO): Likewise.
(FP_RND_PINF): Likewise.
(FP_RND_MINF): Likewise.
(FP_EX_INVALID): Likewise.
(FP_EX_OVERFLOW): Likewise.
(FP_EX_UNDERFLOW): Likewise.
(FP_EX_DIVZERO): Likewise.
(FP_EX_INEXACT): Likewise.
(FP_INIT_ROUNDMODE): Likewise.
* sysdeps/mips/nofpu/fesetenv.c: New file.
* sysdeps/mips/nofpu/feupdateenv.c: Likewise.

5 years agoFix math/test-misc.c for undefined fenv.h macros.
Joseph Myers [Wed, 1 Aug 2018 22:22:00 +0000 (22:22 +0000)]
Fix math/test-misc.c for undefined fenv.h macros.

math/test-misc.c contains some code that uses fenv.h macros
FE_UNDERFLOW, FE_OVERFLOW and FE_UPWARD without being conditional on
those macros being defined.

That would normally break the build for configurations (typically
soft-float) not defining those macros.  However, the code in question
is inside LDBL_MANT_DIG > DBL_MANT_DIG conditionals.  And, while we
have configurations lacking rounding mode and exception support where
LDBL_MANT_DIG > DBL_MANT_DIG (soft-float MIPS64 and RISC-V), those
configurations currently define the fenv.h macros in question even for
soft-float.

There may be some case for defining those macros in cases where a
soft-float compilation could use a hard-float libm (where both
soft-float and hard-float can use the same ABI, as on ARM and RISC-V,
for example).  But MIPS is not such a case - the hard-float and
soft-float ABIs are incompatible - and thus I am testing a patch to
stop defining those macros for soft-float MIPS (motivated by reducing
the extent to which architectures need their own definitions of
math-tests.h macros - if lack of rounding mode / exception support can
be determined by the lack of macros in fenv.h, that avoids the need
for math-tests.h to declare that lack as well).  Introducing a case of
LDBL_MANT_DIG > DBL_MANT_DIG without these macros defined shows up the
problem with math/test-misc.c.  This patch then fixes that problem by
adding appropriate conditionals.

Tested for MIPS64 in conjunction with changes to stop defining the
macros in question in bits/fenv.h for soft-float.

* math/test-misc.c (do_test) [LDBL_MANT_DIG > DBL_MANT_DIG]: Make
code using FE_UNDERFLOW conditional on [FE_UNDERFLOW], code using
FE_OVERFLOW conditional on [FE_OVERFLOW] and code using FE_UPWARD
conditional on [FE_UPWARD].

5 years agoregex: fix memory leak in Gnulib
Paul Eggert [Wed, 1 Aug 2018 20:22:16 +0000 (13:22 -0700)]
regex: fix memory leak in Gnulib

Problem and fix reported by Assaf Gordon in:
https://lists.gnu.org/r/bug-gnulib/2018-07/txtqLKNwBdefE.txt
* posix/regcomp.c (free_charset) [!_LIBC]: Free range_starts and
range_ends members too, as they are defined in 'struct
re_charset_t' even if not _LIBC.  This affects only Gnulib.

5 years agox86: Rename get_common_indeces to get_common_indices
H.J. Lu [Wed, 1 Aug 2018 11:57:34 +0000 (04:57 -0700)]
x86: Rename get_common_indeces to get_common_indices

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* sysdeps/x86/cpu-features.c (get_common_indeces): Renamed to
...
(get_common_indices): This.
(init_cpu_features): Updated.

5 years agoMove SNAN_TESTS_PRESERVE_PAYLOAD out of math-tests.h.
Joseph Myers [Wed, 1 Aug 2018 11:21:16 +0000 (11:21 +0000)]
Move SNAN_TESTS_PRESERVE_PAYLOAD out of math-tests.h.

Continuing moving macros out of math-tests.h to smaller headers
following typo-proof conventions instead of using #ifndef, this patch
moves SNAN_TESTS_PRESERVE_PAYLOAD out to its own sysdeps header.

Tested with build-many-glibcs.py.

* sysdeps/generic/math-tests-snan-payload.h: New file.
* sysdeps/hppa/math-tests-snan-payload.h: Likewise.
* sysdeps/mips/math-tests-snan-payload.h: Likewise.
* sysdeps/riscv/math-tests-snan-payload.h: Likewise.
* sysdeps/generic/math-tests.h: Include
<math-tests-snan-payload.h>.
(SNAN_TESTS_PRESERVE_PAYLOAD): Do not define macro here.
* sysdeps/hppa/math-tests.h: Remove file.
* sysdeps/mips/math-tests.h [!__mips_nan2008]
(SNAN_TESTS_PRESERVE_PAYLOAD): Do not define macro here.
* sysdeps/riscv/math-tests.h (SNAN_TESTS_PRESERVE_PAYLOAD):
Likewise.

5 years agoMove SNAN_TESTS_TYPE_CAST out of math-tests.h.
Joseph Myers [Wed, 1 Aug 2018 11:18:57 +0000 (11:18 +0000)]
Move SNAN_TESTS_TYPE_CAST out of math-tests.h.

The math-tests.h header has many different macros and groups of
macros, defined using #ifndef in the generic version which is included
by architecture versions with #include_next after possibly defining
non-default versions of some of those macros.

This use of #ifndef is contrary to our normal typo-proof conventions
for macro definitions.  This patch moves one of the macros,
SNAN_TESTS_TYPE_CAST, out to its own sysdeps header, to follow those
typo-proof conventions more closely.

Tested with build-many-glibcs.py.

2018-08-01  Joseph Myers  <joseph@codesourcery.com>

* sysdeps/generic/math-tests-snan-cast.h: New file.
* sysdeps/powerpc/math-tests-snan-cast.h: Likewise.
* sysdeps/generic/math-tests.h: Include <math-tests-snan-cast.h>.
(SNAN_TESTS_TYPE_CAST): Do not define macro here.
* sysdeps/powerpc/math-tests.h (SNAN_TESTS_TYPE_CAST): Likewise.

5 years agoAdd version.h, and NEWS update to ChangeLog.
Carlos O'Donell [Wed, 1 Aug 2018 05:42:50 +0000 (01:42 -0400)]
Add version.h, and NEWS update to ChangeLog.

5 years agoOpen master branch for glibc 2.29 development
Carlos O'Donell [Wed, 1 Aug 2018 05:34:25 +0000 (01:34 -0400)]
Open master branch for glibc 2.29 development

5 years agoUpdate NEWS, version.h, and features.h for glibc 2.28.
Carlos O'Donell [Wed, 1 Aug 2018 05:10:47 +0000 (01:10 -0400)]
Update NEWS, version.h, and features.h for glibc 2.28.

5 years agoSynchronize translation project PO files.
Carlos O'Donell [Wed, 1 Aug 2018 04:59:03 +0000 (00:59 -0400)]
Synchronize translation project PO files.

5 years agohurd: Fix exec usage of mach_setup_thread
Samuel Thibault [Tue, 31 Jul 2018 22:10:03 +0000 (00:10 +0200)]
hurd: Fix exec usage of mach_setup_thread

Exec needs that mach_setup_thread does *not* set up TLS since it works on
another task, so we have to split this into mach_setup_tls.

* mach/mach.h (__mach_setup_tls, mach_setup_tls): Add prototypes.
* mach/setup-thread.c (__mach_setup_thread): Move TLS setup to...
(__mach_setup_tls): ... new function.
(mach_setup_tls): New alias.
* hurd/hurdsig.c (_hurdsig_init): Call __mach_setup_tls after
__mach_setup_thread.
* sysdeps/mach/hurd/profil.c (update_waiter): Likewise.
* sysdeps/mach/hurd/setitimer.c (setitimer_locked): Likewise.
* mach/Versions [libc] (mach_setup_tls): Add symbol.
* sysdeps/mach/hurd/i386/libc.abilist (mach_setup_tls): Likewise.

5 years agohurd: Add missing ChangeLog entry
Samuel Thibault [Tue, 31 Jul 2018 22:07:29 +0000 (00:07 +0200)]
hurd: Add missing ChangeLog entry

5 years agohurd: Fix startup of static binaries linked against libpthread
Samuel Thibault [Tue, 31 Jul 2018 20:33:05 +0000 (22:33 +0200)]
hurd: Fix startup of static binaries linked against libpthread

* sysdeps/mach/hurd/i386/init-first.c (init1): Move ELF hdr and TLS
initialization...
(init): ... before initializing libpthread.

5 years agoUpdate tooling versions verified to work with glibc.
Carlos O'Donell [Tue, 31 Jul 2018 20:37:07 +0000 (16:37 -0400)]
Update tooling versions verified to work with glibc.

5 years agoUpdate contrib.texi contributions.
Carlos O'Donell [Tue, 31 Jul 2018 19:26:24 +0000 (15:26 -0400)]
Update contrib.texi contributions.

5 years agoUpdate translation for be.
Carlos O'Donell [Tue, 31 Jul 2018 19:23:23 +0000 (15:23 -0400)]
Update translation for be.

5 years agoUpdate SH libm-tests-ulps
Adhemerval Zanella [Tue, 31 Jul 2018 13:33:53 +0000 (10:33 -0300)]
Update SH libm-tests-ulps

* sysdeps/sh/libm-test-ulps: Update.

5 years agoUpdate translations for bg, de, hr, pt_BR, sv, and vi.
Carlos O'Donell [Mon, 30 Jul 2018 20:20:00 +0000 (16:20 -0400)]
Update translations for bg, de, hr, pt_BR, sv, and vi.

5 years agox86/CET: Fix property note parser [BZ #23467]
H.J. Lu [Mon, 30 Jul 2018 23:14:46 +0000 (16:14 -0700)]
x86/CET: Fix property note parser [BZ #23467]

GNU_PROPERTY_X86_FEATURE_1_AND may not be the first property item.  We
need to check each property item until we reach the end of the property
or find GNU_PROPERTY_X86_FEATURE_1_AND.

This patch adds 2 tests.  The first test checks if IBT is enabled and
the second test reads the output from the first test to check if IBT
is is enabled.  The second second test fails if IBT isn't enabled
properly.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
[BZ #23467]
* sysdeps/unix/sysv/linux/x86/Makefile (tests): Add
tst-cet-property-1 and tst-cet-property-2 if CET is enabled.
(CFLAGS-tst-cet-property-1.o): New.
(ASFLAGS-tst-cet-property-dep-2.o): Likewise.
($(objpfx)tst-cet-property-2): Likewise.
($(objpfx)tst-cet-property-2.out): Likewise.
* sysdeps/unix/sysv/linux/x86/tst-cet-property-1.c: New file.
* sysdeps/unix/sysv/linux/x86/tst-cet-property-2.c: Likewise.
* sysdeps/unix/sysv/linux/x86/tst-cet-property-dep-2.S: Likewise.
* sysdeps/x86/dl-prop.h (_dl_process_cet_property_note): Parse
each property item until GNU_PROPERTY_X86_FEATURE_1_AND is found.

5 years agox86: Add tst-get-cpu-features-static to $(tests) [BZ #23458]
H.J. Lu [Mon, 30 Jul 2018 23:11:05 +0000 (16:11 -0700)]
x86: Add tst-get-cpu-features-static to $(tests) [BZ #23458]

All tests should be added to $(tests).

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
[BZ #23458]
* sysdeps/x86/Makefile (tests): Add tst-get-cpu-features-static.

5 years agohurd: Fix some ld.so symbol override from libc
Samuel Thibault [Sat, 28 Jul 2018 13:12:42 +0000 (15:12 +0200)]
hurd: Fix some ld.so symbol override from libc

ld.so symbols to be overriden by libc need to be extern to really get
overriden.

* sysdeps/mach/hurd/not-errno.h: New file.
* sysdeps/mach/hurd/i386/localplt.data: Update accordingly.

5 years agohurd: Fix some ld.so symbol override from libc
Samuel Thibault [Sat, 28 Jul 2018 12:57:10 +0000 (14:57 +0200)]
hurd: Fix some ld.so symbol override from libc

ld.so symbols to be overriden by libc need to be extern to really get
overriden.

* sysdeps/mach/hurd/dl-unistd.h (__access, __brk, __lseek, __read,
__sbrk): Do not set attribute_hidden.
* sysdeps/mach/hurd/i386/ld.abilist: Update accordingly.
* sysdeps/mach/hurd/i386/localplt.data: Update accordingly.

5 years agoUpdate translations for cs, pl, and uk.
Carlos O'Donell [Sat, 28 Jul 2018 00:07:33 +0000 (20:07 -0400)]
Update translations for cs, pl, and uk.

5 years agox86/CET: Don't parse beyond the note end
H.J. Lu [Fri, 27 Jul 2018 20:20:51 +0000 (13:20 -0700)]
x86/CET: Don't parse beyond the note end

Simply check if "ptr < ptr_end" since "ptr" is always incremented by 8.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* sysdeps/x86/dl-prop.h (_dl_process_cet_property_note): Don't
parse beyond the note end.

5 years agoUpdate NEWS with ISO 14651 update information.
Carlos O'Donell [Fri, 27 Jul 2018 19:56:24 +0000 (15:56 -0400)]
Update NEWS with ISO 14651 update information.

5 years agoNEWS: Avoid the words "nominative" and "genitive".
Rafal Luzynski [Fri, 27 Jul 2018 09:27:48 +0000 (11:27 +0200)]
NEWS: Avoid the words "nominative" and "genitive".

Glibc supports two grammatical forms of month names and we keep adding
the locale data which actually use this feature but those two forms are
not necessarily nominative and genitive.  It is better to use a more
generic term.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
5 years agoFix Linux fcntl OFD locks on unsupported kernels
Adhemerval Zanella [Fri, 27 Jul 2018 13:07:14 +0000 (10:07 -0300)]
Fix Linux fcntl OFD locks on unsupported kernels

This patch make the OFD tests return unsupported if kernel does not
support OFD locks (it was added on 3.15).

Checked on a ia64-linux-gnu with Linux 3.14.

* sysdeps/unix/sysv/linux/tst-ofdlocks.c: Return unsupported if
kernel does not support OFD locks.
* sysdeps/unix/sysv/linux/tst-ofdlocks-compat.c: Likewise.

5 years agoAdd the missing ChangeLog entry for commit be525a69a66
H.J. Lu [Fri, 27 Jul 2018 17:28:30 +0000 (10:28 -0700)]
Add the missing ChangeLog entry for commit be525a69a66

5 years agohurd: Fix some ld.so symbol override from libc
Samuel Thibault [Fri, 27 Jul 2018 09:18:29 +0000 (11:18 +0200)]
hurd: Fix some ld.so symbol override from libc

ld.so symbols to be overriden by libc need to be extern to really get
overriden.

More fixes are needed to avoid the hidden attribute.

* sysdeps/mach/hurd/Versions (libc): Make __access and
__access_noerrno external so they can override the ld symbols.
(ld): Make __access, __read, __sbrk, __strtoul_internal, __write,
__writev, __open64, __access_noerrno extern so they can be overrided.
* sysdeps/mach/hurd/i386/libc.abilist: Update accordingly.
* sysdeps/mach/hurd/i386/ld.abilist: Update accordingly.

5 years agoUpdate libc.pot.
Carlos O'Donell [Fri, 27 Jul 2018 02:21:34 +0000 (22:21 -0400)]
Update libc.pot.

5 years agoUpdate powerpc-nofpu ulps.
Joseph Myers [Thu, 26 Jul 2018 22:51:06 +0000 (22:51 +0000)]
Update powerpc-nofpu ulps.

* sysdeps/powerpc/nofpu/libm-test-ulps: Update.

5 years agox86: Populate COMMON_CPUID_INDEX_80000001 for Intel CPUs [BZ #23459]
H.J. Lu [Thu, 26 Jul 2018 17:49:19 +0000 (10:49 -0700)]
x86: Populate COMMON_CPUID_INDEX_80000001 for Intel CPUs [BZ #23459]

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
[BZ #23459]
* sysdeps/x86/cpu-features.c (get_extended_indices): New
function.
(init_cpu_features): Call get_extended_indices for both Intel
and AMD CPUs.
* sysdeps/x86/cpu-features.h (COMMON_CPUID_INDEX_80000001):
Remove "for AMD" comment.

5 years agox86: Correct index_cpu_LZCNT [BZ # 23456]
H.J. Lu [Thu, 26 Jul 2018 18:10:08 +0000 (11:10 -0700)]
x86: Correct index_cpu_LZCNT [BZ # 23456]

cpu-features.h has

 #define bit_cpu_LZCNT (1 << 5)
 #define index_cpu_LZCNT COMMON_CPUID_INDEX_1
 #define reg_LZCNT

But the LZCNT feature bit is in COMMON_CPUID_INDEX_80000001:

Initial EAX Value: 80000001H
ECX Extended Processor Signature and Feature Bits:
Bit 05: LZCNT available

index_cpu_LZCNT should be COMMON_CPUID_INDEX_80000001, not
COMMON_CPUID_INDEX_1.  The VMX feature bit is in COMMON_CPUID_INDEX_1:

Initial EAX Value: 01H
Feature Information Returned in the ECX Register:
5 VMX

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
[BZ # 23456]
* sysdeps/x86/cpu-features.h (index_cpu_LZCNT): Set to
COMMON_CPUID_INDEX_80000001.

5 years agoFix string/tst-xbzero-opt if build with gcc head.
Stefan Liebler [Thu, 26 Jul 2018 15:09:44 +0000 (17:09 +0200)]
Fix string/tst-xbzero-opt if build with gcc head.

On s390x, the test string/tst-xbzero-opt is failing if build with gcc head:
FAIL: no clear/prepare: expected 32 got 0
FAIL: no clear/test: expected some got 0
FAIL: ordinary clear/prepare: expected 32 got 0
INFO: ordinary clear/test: found 0 patterns (memset not eliminated)
PASS: explicit clear/prepare: expected 32 got 32
PASS: explicit clear/test: expected 0 got 0

In setup_no_clear / setup_ordinary_clear, GCC is omitting the memcpy loop
in prepare_test_buffer. Thus count_test_patterns does not find any of the
test_pattern.

This patch calls use_test_buffer in order to force the compiler to really copy
the pattern to buf.

ChangeLog:

* string/tst-xbzero-opt.c (use_test_buffer): New function.
(prepare_test_buffer): Call use_test_buffer as compiler barrier.

5 years agohtl: Use weak aliases for public symbols
Florian Weimer [Thu, 26 Jul 2018 06:34:05 +0000 (08:34 +0200)]
htl: Use weak aliases for public symbols

Strong definitions of flockfile, funlockfile, ftrylockfile can conflict
with application symbols when linking statically.

5 years agoAdd missing localedata/en_US.UTF-8.in (Bug 23393).
Carlos O'Donell [Thu, 26 Jul 2018 01:58:10 +0000 (21:58 -0400)]
Add missing localedata/en_US.UTF-8.in (Bug 23393).

Commit 7cd7d36f1feb3ccacf476e909b115b45cdd46e77 failed to include
the new testing file en_US.UTF-8.in.

5 years agoKeep expected behaviour for [a-z] and [A-z] (Bug 23393).
Carlos O'Donell [Wed, 25 Jul 2018 21:00:45 +0000 (17:00 -0400)]
Keep expected behaviour for [a-z] and [A-z] (Bug 23393).

In commit 9479b6d5e08eacce06c6ab60abc9b2f4eb8b71e4 we updated all of
the collation data to harmonize with the new version of ISO 14651
which is derived from Unicode 9.0.0.  This collation update brought
with it some changes to locales which were not desirable by some
users, in particular it altered the meaning of the
locale-dependent-range regular expression, namely [a-z] and [A-Z], and
for en_US it caused uppercase letters to be matched by [a-z] for the
first time.  The matching of uppercase letters by [a-z] is something
which is already known to users of other locales which have this
property, but this change could cause significant problems to en_US
and other similar locales that had never had this change before.
Whether this behaviour is desirable or not is contentious and GNU Awk
has this to say on the topic:
https://www.gnu.org/software/gawk/manual/html_node/Ranges-and-Locales.html
While the POSIX standard also has this further to say: "RE Bracket
Expression":
http://pubs.opengroup.org/onlinepubs/9699919799/xrat/V4_xbd_chap09.html
"The current standard leaves unspecified the behavior of a range
expression outside the POSIX locale. ... As noted above, efforts were
made to resolve the differences, but no solution has been found that
would be specific enough to allow for portable software while not
invalidating existing implementations."
In glibc we implement the requirement of ISO POSIX-2:1993 and use
collation element order (CEO) to construct the range expression, the
API internally is __collseq_table_lookup().  The fact that we use CEO
and also have 4-level weights on each collation rule means that we can
in practice reorder the collation rules in iso14651_t1_common (the new
data) to provide consistent range expression resolution *and* the
weights should maintain the expected total order.  Therefore this
patch does three things:

* Reorder the collation rules for the LATIN script in
  iso14651_t1_common to deinterlace uppercase and lowercase letters in
  the collation element orders.

* Adds new test data en_US.UTF-8.in for sort-test.sh which exercises
  strcoll* and strxfrm* and ensures the ISO 14651 collation remains.

* Add back tests to tst-fnmatch.input and tst-regexloc.c which
  exercise that [a-z] does not match A or Z.

The reordering of the ISO 14651 data is done in an entirely mechanical
fashion using the following program attached to the bug:
https://sourceware.org/bugzilla/show_bug.cgi?id=23393#c28

It is up for discussion if the iso14651_t1_common data should be
refined further to have 3 very tight collation element ranges that
include only a-z, A-Z, and 0-9, which would implement the solution
sought after in:
https://sourceware.org/bugzilla/show_bug.cgi?id=23393#c12
and implemented here:
https://www.sourceware.org/ml/libc-alpha/2018-07/msg00854.html

No regressions on x86_64.
Verified that removal of the iso14651_t1_common change causes tst-fnmatch
to regress with:
422: fnmatch ("[a-z]", "A", 0) = 0 (FAIL, expected FNM_NOMATCH) ***
...
425: fnmatch ("[A-Z]", "z", 0) = 0 (FAIL, expected FNM_NOMATCH) ***

5 years agoia64: Work around incorrect type of IA64 uc_sigmask
H.J. Lu [Wed, 25 Jul 2018 20:54:08 +0000 (13:54 -0700)]
ia64: Work around incorrect type of IA64 uc_sigmask

Work around:

https://sourceware.org/bugzilla/show_bug.cgi?id=21634

* sysdeps/unix/sysv/linux/ia64/tst-setcontext4.c: New file.

5 years agoFix ISO C threads installed header and HURD assumption
Adhemerval Zanella [Wed, 25 Jul 2018 18:20:44 +0000 (15:20 -0300)]
Fix ISO C threads installed header and HURD assumption

Set installed NPTL header as the expected one (instead of an
internal one for glibc testsuite) and add a hurd specific
stdc-predef with __STDC_NO_THREADS__.

Checked on both i686-linux-gnu and i686-gnu that both threads.h
and stdc-predef.h are the expected ones.

* nptl/threads.h: Move to ...
* sysdeps/nptl/threads.h: ... here.
* sysdeps/hurd/stdc-predef.h: New file.

5 years agox86/CET: Add a setcontext test for CET
H.J. Lu [Wed, 25 Jul 2018 15:40:39 +0000 (08:40 -0700)]
x86/CET: Add a setcontext test for CET

Verify that setcontext works with gaps above and below the newly
allocated shadow stack.

* sysdeps/unix/sysv/linux/x86/Makefile (tests): Add
tst-cet-setcontext-1 if CET is enabled.
(CFLAGS-tst-cet-setcontext-1.c): Add -mshstk.
* sysdeps/unix/sysv/linux/x86/tst-cet-setcontext-1.c: New file.

5 years agoFix C11 conformance issues
Adhemerval Zanella [Wed, 25 Jul 2018 13:20:15 +0000 (13:20 +0000)]
Fix C11 conformance issues

Remove conformace assumption of NPTL implementation for ISO C threads
and revert wrong libcrypt addition on linknamespace-libs-XPG4.

The i686-gnu target now shows two new conformance failures:

FAIL: conform/ISO11/threads.h/conform
FAIL: conform/ISO11/threads.h/linknamespace

It is expected due missing HTL ISO C threads support and both conformance
.out files indicates the reason ("#error "HTL does not implement ISO C
threads").

Checked on i686-linux-gnu and i686-gnu.

* include/threads.h: Move to ...
* sysdeps/nptl/threads.h: ... here.
* sysdeps/htl/threads.h: New file.
* conform/Makefile (linknamespace-libs-ISO11): Use
static-thread-library instead of linking libpthread.
(linknamespace-libs-XPG4): Revert wrong libcrypt.a addition.

5 years agoC11 threads: Fix timeout and locking issues
Florian Weimer [Wed, 25 Jul 2018 14:31:45 +0000 (16:31 +0200)]
C11 threads: Fix timeout and locking issues

5 years agox86-64/CET: Extend ucontext_t to save shadow stack
H.J. Lu [Wed, 25 Jul 2018 13:37:14 +0000 (06:37 -0700)]
x86-64/CET: Extend ucontext_t to save shadow stack

This patch adds a field to ucontext_t to save shadow stack:

1. getcontext and swapcontext are updated to save the caller's shadow
stack pointer and return addresses.
2. setcontext and swapcontext are updated to restore shadow stack and
jump to new context directly.
3. makecontext is updated to allocate a new shadow stack and set the
caller's return address to __start_context.

Since makecontext allocates a new shadow stack when making a new
context and kernel allocates a new shadow stack for clone/fork/vfork
syscalls, we track the current shadow stack base.  In setcontext and
swapcontext, if the target shadow stack base is the same as the current
shadow stack base, we unwind the shadow stack.  Otherwise it is a stack
switch and we look for a restore token.

We enable shadow stack at run-time only if program and all used shared
objects, including dlopened ones, are shadow stack enabled, which means
that they must be compiled with GCC 8 or above and glibc 2.28 or above.
We need to save and restore shadow stack only if shadow stack is enabled.
When caller of getcontext, setcontext, swapcontext and makecontext is
compiled with smaller ucontext_t, shadow stack won't be enabled at
run-time.  We check if shadow stack is enabled before accessing the
extended field in ucontext_t.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* sysdeps/unix/sysv/linux/x86/sys/ucontext.h (ucontext_t): Add
__ssp.
* sysdeps/unix/sysv/linux/x86_64/__start_context.S: Include
<asm/prctl.h> and "ucontext_i.h" when shadow stack is enabled.
(__push___start_context): New.
* sysdeps/unix/sysv/linux/x86_64/getcontext.S: Include
<asm/prctl.h>.
(__getcontext): Record the current shadow stack base.  Save the
caller's shadow stack pointer and base.
* sysdeps/unix/sysv/linux/x86_64/makecontext.c: Include
<pthread.h>, <libc-pointer-arith.h> and <sys/prctl.h>.
(__push___start_context): New prototype.
(__makecontext): Call __push___start_context to allocate a new
shadow stack, push __start_context onto the new stack as well
as the new shadow stack.
* sysdeps/unix/sysv/linux/x86_64/setcontext.S: Include
<asm/prctl.h>.
(__setcontext): Restore the target shadow stack.
* sysdeps/unix/sysv/linux/x86_64/swapcontext.S: Include
<asm/prctl.h>.
(__swapcontext): Record the current shadow stack base.  Save
the caller's shadow stack pointer and base.  Restore the target
shadow stack.
* sysdeps/unix/sysv/linux/x86_64/sysdep.h
(STACK_SIZE_TO_SHADOW_STACK_SIZE_SHIFT): New.
* sysdeps/unix/sysv/linux/x86_64/ucontext_i.sym (oSSP): New.

5 years agoAdd tests for setcontext on the context from makecontext
H.J. Lu [Wed, 25 Jul 2018 12:12:59 +0000 (05:12 -0700)]
Add tests for setcontext on the context from makecontext

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* stdlib/Makefile ((tests): Add tst-setcontext6, tst-setcontext7,
tst-setcontext8 and tst-setcontext9.
* stdlib/tst-setcontext6.c: New file.
* stdlib/tst-setcontext7.c: Likewise.
* stdlib/tst-setcontext8.c: Likewise.
* stdlib/tst-setcontext9.c: Likewise.

5 years agoAdd a test for multiple setcontext calls
H.J. Lu [Wed, 25 Jul 2018 12:00:30 +0000 (05:00 -0700)]
Add a test for multiple setcontext calls

Check multiple setcontext calls.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>.
* stdlib/Makefile ((tests): Add tst-setcontext5.
* stdlib/tst-setcontext5.c: New file.

5 years agoAdd another test for setcontext
H.J. Lu [Wed, 25 Jul 2018 11:57:21 +0000 (04:57 -0700)]
Add another test for setcontext

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* stdlib/Makefile (tests): Add tst-setcontext4.
* stdlib/tst-setcontext4.c: New file.

5 years agoAdd a test for multiple makecontext calls
H.J. Lu [Wed, 16 May 2018 19:28:19 +0000 (12:28 -0700)]
Add a test for multiple makecontext calls

Check multiple makecontext calls.

* stdlib/Makefile (tests): Add tst-swapcontext1.
* stdlib/tst-swapcontext1.c: New test.

5 years agox86/CET: Add tests with legacy non-CET shared objects
H.J. Lu [Wed, 25 Jul 2018 11:46:53 +0000 (04:46 -0700)]
x86/CET: Add tests with legacy non-CET shared objects

Check binary compatibility of CET-enabled executables:

1. When CET-enabled executable is used with legacy non-CET shared object
at run-time, ld.so should disable SHSTK and put legacy non-CET shared
objects in legacy bitmap.
2. When IBT-enabled executable dlopens legacy non-CET shared object,
ld.so should put legacy shared object in legacy bitmap.
3. Use GLIBC_TUNABLES=glibc.tune.x86_shstk=[on|off|permissive] to
control how SHSTK is enabled.

* sysdeps/x86/Makefile (tests): Add tst-cet-legacy-1,
tst-cet-legacy-2, tst-cet-legacy-2a, tst-cet-legacy-3,
tst-cet-legacy-4, tst-cet-legacy-4a, tst-cet-legacy-4b
and tst-cet-legacy-4c.
(modules-names): Add tst-cet-legacy-mod-1, tst-cet-legacy-mod-2
and tst-cet-legacy-mod-4.
(CFLAGS-tst-cet-legacy-2.c): New.
(CFLAGS-tst-cet-legacy-mod-1.c): Likewise.
(CFLAGS-tst-cet-legacy-mod-2.c): Likewise.
(CFLAGS-tst-cet-legacy-3.c): Likewise.
(CFLAGS-tst-cet-legacy-4.c): Likewise.
(CFLAGS-tst-cet-legacy-mod-4.c): Likewise.
($(objpfx)tst-cet-legacy-1): Likewise.
($(objpfx)tst-cet-legacy-2): Likewise.
($(objpfx)tst-cet-legacy-2.out): Likewise.
($(objpfx)tst-cet-legacy-2a): Likewise.
($(objpfx)tst-cet-legacy-2a.out): Likewise.
($(objpfx)tst-cet-legacy-4): Likewise.
($(objpfx)tst-cet-legacy-4.out): Likewise.
($(objpfx)tst-cet-legacy-4a): Likewise.
($(objpfx)tst-cet-legacy-4a.out): Likewise.
(tst-cet-legacy-4a-ENV): Likewise.
($(objpfx)tst-cet-legacy-4b): Likewise.
($(objpfx)tst-cet-legacy-4b.out): Likewise.
(tst-cet-legacy-4b-ENV): Likewise.
($(objpfx)tst-cet-legacy-4c): Likewise.
($(objpfx)tst-cet-legacy-4c.out): Likewise.
(tst-cet-legacy-4c-ENV): Likewise.
* sysdeps/x86/tst-cet-legacy-1.c: New file.
* sysdeps/x86/tst-cet-legacy-2.c: Likewise.
* sysdeps/x86/tst-cet-legacy-2a.c: Likewise.
* sysdeps/x86/tst-cet-legacy-3.c: Likewise.
* sysdeps/x86/tst-cet-legacy-4.c: Likewise.
* sysdeps/x86/tst-cet-legacy-4a.c: Likewise.
* sysdeps/x86/tst-cet-legacy-4b.c: Likewise.
* sysdeps/x86/tst-cet-legacy-4c.c: Likewise.
* sysdeps/x86/tst-cet-legacy-mod-1.c: Likewise.
* sysdeps/x86/tst-cet-legacy-mod-2.c: Likewise.
* sysdeps/x86/tst-cet-legacy-mod-4.c: Likewise.

5 years agox86: Rename __glibc_reserved2 to ssp_base in tcbhead_t
H.J. Lu [Wed, 25 Jul 2018 11:39:26 +0000 (04:39 -0700)]
x86: Rename __glibc_reserved2 to ssp_base in tcbhead_t

This will be used to record the current shadow stack base for shadow
stack switching by getcontext, makecontext, setcontext and swapcontext.
If the target shadow stack base is the same as the current shadow stack
base, we unwind the shadow stack.  Otherwise it is a stack switch and
we look for a restore token to restore the target shadow stack.

* sysdeps/i386/nptl/tcb-offsets.sym (SSP_BASE_OFFSET): New.
* sysdeps/i386/nptl/tls.h (tcbhead_t): Replace __glibc_reserved2
with ssp_base.
* sysdeps/x86_64/nptl/tcb-offsets.sym (SSP_BASE_OFFSET): New.
* sysdeps/x86_64/nptl/tls.h (tcbhead_t): Replace __glibc_reserved2
with ssp_base.

5 years agoFix out of bounds access in findidxwc (bug 23442)
Andreas Schwab [Tue, 24 Jul 2018 16:02:28 +0000 (18:02 +0200)]
Fix out of bounds access in findidxwc (bug 23442)

If usrc is a prefix of cp but one character shorter an out of bounds
access to usrc was done.

5 years agoMention ISO C threads addition
Adhemerval Zanella [Tue, 24 Jul 2018 21:10:32 +0000 (18:10 -0300)]
Mention ISO C threads addition

* NEWS: Add ISO C threads addition.

5 years agoalpha: mlock2, copy_file_range syscalls were introduced in kernel 4.13
Florian Weimer [Tue, 24 Jul 2018 19:03:04 +0000 (21:03 +0200)]
alpha: mlock2, copy_file_range syscalls were introduced in kernel 4.13

5 years agox86/CET: Extend arch_prctl syscall for CET control
H.J. Lu [Tue, 24 Jul 2018 19:23:03 +0000 (12:23 -0700)]
x86/CET: Extend arch_prctl syscall for CET control

CET arch_prctl bits should be defined in <asm/prctl.h> from Linux kernel
header files.  Add x86 <include/asm/prctl.h> for pre-CET kernel header
files.

Note: sysdeps/unix/sysv/linux/x86/include/asm/prctl.h should be removed
if <asm/prctl.h> from the required kernel header files contains CET
arch_prctl bits.

 /* CET features:
    IBT:   GNU_PROPERTY_X86_FEATURE_1_IBT
    SHSTK: GNU_PROPERTY_X86_FEATURE_1_SHSTK
  */

 /* Return CET features in unsigned long long *addr:
      features: addr[0].
      shadow stack base address: addr[1].
      shadow stack size: addr[2].
  */
 # define ARCH_CET_STATUS 0x3001
 /* Disable CET features in unsigned int features.  */
 # define ARCH_CET_DISABLE 0x3002
 /* Lock all CET features.  */
 # define ARCH_CET_LOCK 0x3003
 /* Allocate a new shadow stack with unsigned long long *addr:
      IN: requested shadow stack size: *addr.
      OUT: allocated shadow stack address: *addr.
  */
 # define ARCH_CET_ALLOC_SHSTK 0x3004
 /* Return legacy region bitmap info in unsigned long long *addr:
     address: addr[0].
     size: addr[1].
  */
 # define ARCH_CET_LEGACY_BITMAP 0x3005

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* sysdeps/unix/sysv/linux/x86/include/asm/prctl.h: New file.
* sysdeps/unix/sysv/linux/x86/cpu-features.c: Include
<sys/prctl.h> and <asm/prctl.h>.
(get_cet_status): Call arch_prctl with ARCH_CET_STATUS.
* sysdeps/unix/sysv/linux/x86/dl-cet.h: Include <sys/prctl.h>
and <asm/prctl.h>.
(dl_cet_allocate_legacy_bitmap): Call arch_prctl with
ARCH_CET_LEGACY_BITMAP.
(dl_cet_disable_cet): Call arch_prctl with ARCH_CET_DISABLE.
(dl_cet_lock_cet): Call arch_prctl with ARCH_CET_LOCK.
* sysdeps/x86/libc-start.c: Include <startup.h>.

5 years agosh: Do not define __ASSUME_STATX
Florian Weimer [Tue, 24 Jul 2018 17:06:27 +0000 (19:06 +0200)]
sh: Do not define __ASSUME_STATX

5 years agoAdd manual documentation for threads.h
Rical Jasan [Tue, 6 Dec 2016 22:47:02 +0000 (20:47 -0200)]
Add manual documentation for threads.h

This patch updates the manual and adds a new chapter to the manual,
explaining types macros, constants and functions defined by ISO C11
threads.h standard.

[BZ# 14092]
* manual/debug.texi: Update adjacent chapter name.
* manual/probes.texi: Likewise.
* manual/threads.texi (ISO C Threads): New section.
(POSIX Threads): Convert to a section.

5 years agonptl: Add test cases for ISO C11 threads
Adhemerval Zanella [Mon, 5 Dec 2016 20:32:11 +0000 (18:32 -0200)]
nptl: Add test cases for ISO C11 threads

This patch adds to testsuite new test cases to test all new introduced
C11 threads functions, types and macros are tested.

Checked with a build for all major ABI (aarch64-linux-gnu, alpha-linux-gnu,
arm-linux-gnueabi, i386-linux-gnu, ia64-linux-gnu, m68k-linux-gnu,
microblaze-linux-gnu [1], mips{64}-linux-gnu, nios2-linux-gnu,
powerpc{64le}-linux-gnu, s390{x}-linux-gnu, sparc{64}-linux-gnu,
and x86_64-linux-gnu).

Also ran a full check on aarch64-linux-gnu, x86_64-linux-gnu, i686-linux-gnu,
arm-linux-gnueabhf, and powerpc64le-linux-gnu.

Adhemerval Zanella  <adhemerval.zanella@linaro.org>
Juan Manuel Torres Palma  <jmtorrespalma@gmail.com>

[BZ #14092]
* nptl/Makefile (tests): Add new test files.
* nptl/tst-call-once.c : New file. Tests C11 functions and types.
* nptl/tst-cnd-basic.c: Likewise.
* nptl/tst-cnd-broadcast.c: Likewise.
* nptl/tst-cnd-timedwait.c: Likewise.
* nptl/tst-mtx-basic.c: Likewise.
* nptl/tst-mtx-recursive.c: Likewise.
* nptl/tst-mtx-timedlock.c: Likewise.
* nptl/tst-mtx-trylock.c: Likewise.
* nptl/tst-thrd-basic.c: Likewise.
* nptl/tst-thrd-detach.c: Likewise.
* nptl/tst-thrd-sleep.c: Likewise.
* nptl/tst-tss-basic.c: Likewise.

5 years agonptl: Add abilist symbols for C11 threads
Adhemerval Zanella [Tue, 27 Jun 2017 14:02:16 +0000 (11:02 -0300)]
nptl: Add abilist symbols for C11 threads

This patch adds the required abilist symbols for C11 thread support.

Checked with a build for all major ABI (aarch64-linux-gnu, alpha-linux-gnu,
arm-linux-gnueabi, i386-linux-gnu, ia64-linux-gnu, m68k-linux-gnu,
microblaze-linux-gnu [1], mips{64}-linux-gnu, nios2-linux-gnu,
powerpc{64le}-linux-gnu, s390{x}-linux-gnu, sparc{64}-linux-gnu,
and x86_64-linux-gnu).

Also ran a full check on aarch64-linux-gnu, x86_64-linux-gnu, i686-linux-gnu,
arm-linux-gnueabhf, and powerpc64le-linux-gnu.

[BZ #14092]
* sysdeps/unix/sysv/linux/aarch64/libc.abilist [GLIBC_2.28]
(thrd_current, thrd_equal, thrd_sleep, thrd_yield): Add C11 thread
symbols.
* sysdeps/unix/sysv/linux/alpha/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/arm/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/hppa/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/i386/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/ia64/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/m68k/coldfire/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/m68k/m680x0/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/microblaze/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/mips/mips32/fpu/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/mips/mips64/n32/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/mips/mips64/n64/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/nios2/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/libc.abilist:
Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/libc.abilist:
Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc64/libc-le.abilist:
Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc64/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/riscv/rv64/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/s390/s390-32/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/s390/s390-64/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/sh/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/sparc/sparc32/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/sparc/sparc64/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/x86_64/64/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/x86_64/x32/libc.abilist: Likewise.
* sysdeps/unix/sysv/linux/aarch64/libpthread.abilist [[GLIBC_2.28]
(call_once, cnd_broadcast, cnd_destroy, cnd_init, cnd_signal,
cnd_timedwait, cnd_wait, mtx_destroy, mtx_init, mtx_lock,
mtx_timedlock, mtx_trylock, mtx_unlokc, thrd_create, thrd_detach,
thrd_exit, thrd_join, tss_create, tss_delete, tss_get, tss_set):
Likewise.
* sysdeps/unix/sysv/linux/aarch64/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/alpha/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/arm/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/hppa/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/i386/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/ia64/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/m68k/coldfire/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/m68k/m680x0/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/microblaze/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/mips/mips32/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/mips/mips64/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/nios2/libpthread.abilist: Likewise.
ikewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc32/libpthread.abilist:
* sysdeps/unix/sysv/linux/powerpc/powerpc64/libpthread-le.abilist:
Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc64/libpthread.abilist:
Likewise.
* sysdeps/unix/sysv/linux/riscv/rv64/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/s390/s390-32/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/s390/s390-64/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/sh/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/sparc/sparc32/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/sparc/sparc64/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/x86_64/64/libpthread.abilist: Likewise.
* sysdeps/unix/sysv/linux/x86_64/x32/libpthread.abilist: Likewise.

5 years agonptl: Add C11 threads tss_* functions
Adhemerval Zanella [Tue, 27 Jun 2017 14:00:05 +0000 (11:00 -0300)]
nptl: Add C11 threads tss_* functions

This patch adds the tss_* definitions from C11 threads (ISO/IEC 9899:2011),
more specifically tss_create, tss_delete, tss_get, tss_set, and required
types.

Mostly of the definitions are composed based on POSIX conterparts, including
tss_t (pthread_key_t).

Checked with a build for all major ABI (aarch64-linux-gnu, alpha-linux-gnu,
arm-linux-gnueabi, i386-linux-gnu, ia64-linux-gnu, m68k-linux-gnu,
microblaze-linux-gnu [1], mips{64}-linux-gnu, nios2-linux-gnu,
powerpc{64le}-linux-gnu, s390{x}-linux-gnu, sparc{64}-linux-gnu,
and x86_64-linux-gnu).

Also ran a full check on aarch64-linux-gnu, x86_64-linux-gnu, i686-linux-gnu,
arm-linux-gnueabhf, and powerpc64le-linux-gnu.

[BZ #14092]
* conform/data/threads.h-data (thread_local): New macro.
(TSS_DTOR_ITERATIONS): Likewise.
(tss_t): New type.
(tss_dtor_t): Likewise.
(tss_create): New function.
(tss_get): Likewise.
(tss_set): Likewise.
(tss_delete): Likewise.
* nptl/Makefile (libpthread-routines): Add tss_create, tss_delete,
tss_get, and tss_set objects.
* nptl/Versions (libpthread) [GLIBC_2.28]: Likewise.
* nptl/tss_create.c: New file.
* nptl/tss_delete.c: Likewise.
* nptl/tss_get.c: Likewise.
* nptl/tss_set.c: Likewise.
* sysdeps/nptl/threads.h (thread_local): New define.
(TSS_DTOR_ITERATIONS): Likewise.
(tss_t): New typedef.
(tss_dtor_t): Likewise.
(tss_create): New prototype.
(tss_get): Likewise.
(tss_set): Likewise.
(tss_delete): Likewise.

5 years agonptl: Add C11 threads cnd_* functions
Adhemerval Zanella [Tue, 27 Jun 2017 13:55:10 +0000 (10:55 -0300)]
nptl: Add C11 threads cnd_* functions

This patch adds the cnd_* definitions from C11 threads (ISO/IEC 9899:2011),
more specifically cnd_broadcast, cnd_destroy, cnd_init, cnd_signal,
cnd_timedwait, cnd_wait, and required types.

Mostly of the definitions are composed based on POSIX conterparts, and
cnd_t is also based on internal pthreads fields, but with distinct internal
layout to avoid possible issues with code interchange (such as trying to pass
POSIX structure on C11 functions and to avoid inclusion of pthread.h).  The
idea is to make it possible to share POSIX internal implementation for mostly
of the code making adjust where only required.

Checked with a build for all major ABI (aarch64-linux-gnu, alpha-linux-gnu,
arm-linux-gnueabi, i386-linux-gnu, ia64-linux-gnu, m68k-linux-gnu,
microblaze-linux-gnu [1], mips{64}-linux-gnu, nios2-linux-gnu,
powerpc{64le}-linux-gnu, s390{x}-linux-gnu, sparc{64}-linux-gnu,
and x86_64-linux-gnu).

Also ran a full check on aarch64-linux-gnu, x86_64-linux-gnu, i686-linux-gnu,
arm-linux-gnueabhf, and powerpc64le-linux-gnu.

[BZ #14092]
* conform/data/threads.h-data (cnd_t): New type.
(cnd_init): New function.
(cnd_signal): Likewise.
(cnd_broadcast): Likewise.
(cnd_wait): Likewise.
(cnd_timedwait): Likewise.
(cnd_destroy): Likewise.
* nptl/Makefile (libpthread-routines): Add cnd_broadcast,
cnd_destroy, cnd_init, cnd_signal, cnd_timedwait, and cnd_wait
object.
* nptl/Versions (libpthread) [GLIBC_2.28]: Likewise.
* nptl/cnd_broadcast.c: New file.
* nptl/cnd_destroy.c: Likewise.
* nptl/cnd_init.c: Likewise.
* nptl/cnd_signal.c: Likewise.
* nptl/cnd_timedwait.c: Likewise.
* nptl/cnd_wait.c: Likewise.
* sysdeps/nptl/threads.h (cnd_t): New type.
(cnd_init): New prototype.
(cnd_signa): Likewise.
(cnd_broadcast): Likewise.
(cnd_wait): Likewise.
(cnd_timedwait): Likewise.
(cnd_destroy): Likewise.

5 years agonptl: Add C11 threads call_once functions
Adhemerval Zanella [Tue, 27 Jun 2017 13:49:51 +0000 (10:49 -0300)]
nptl: Add C11 threads call_once functions

This patch adds the call_* definitions from C11 threads (ISO/IEC 9899:2011),
more specifically call_once and required types.

Mostly of the definitions are composed based on POSIX conterparts,including
once_flag (pthread_once_t).  The idea is to make possible to share POSIX
internal implementations for mostly of the code (and making adjustment only
when required).

Checked with a build for all major ABI (aarch64-linux-gnu, alpha-linux-gnu,
arm-linux-gnueabi, i386-linux-gnu, ia64-linux-gnu, m68k-linux-gnu,
microblaze-linux-gnu [1], mips{64}-linux-gnu, nios2-linux-gnu,
powerpc{64le}-linux-gnu, s390{x}-linux-gnu, sparc{64}-linux-gnu,
and x86_64-linux-gnu).

Also ran a full check on aarch64-linux-gnu, x86_64-linux-gnu, i686-linux-gnu,
arm-linux-gnueabhf, and powerpc64le-linux-gnu.

[BZ #14092]
* conform/data/threads.h-data (ONCE_FLAG_INIT): New macro.
(once_flag): New type.
(call_once): New function.
* nptl/Makefile (libpthread-routines): Add call_once object.
* nptl/Versions (libphread) [GLIBC_2.28]: Add call_once symbol.
* nptl/call_once.c: New file.
* sysdeps/nptl/threads.h (ONCE_FLAG_INIT): New define.
(once_flag): New type.
(call_once): New prototype.

5 years agonptl: Add C11 threads mtx_* functions
Adhemerval Zanella [Tue, 27 Jun 2017 13:45:04 +0000 (10:45 -0300)]
nptl: Add C11 threads mtx_* functions

This patch adds the mtx_* definitions from C11 threads (ISO/IEC 9899:2011),
more specifically mtx_init, mtx_destroy, mtx_lock, mtx_timedlock, mtx_trylock,
mtx_unlock, and required types.

Mostly of the definitions are composed based on POSIX conterparts, and mtx_t
is also based on internal pthread fields, but with a distinct internal layout
to avoid possible issues with code interchange (such as trying to pass POSIX
structure on C11 functions and to avoid inclusion of pthread.h).  The idea
is to make possible to share POSIX internal implementations for mostly of
the code (and making adjustment only when required).

Checked with a build for all major ABI (aarch64-linux-gnu, alpha-linux-gnu,
arm-linux-gnueabi, i386-linux-gnu, ia64-linux-gnu, m68k-linux-gnu,
microblaze-linux-gnu [1], mips{64}-linux-gnu, nios2-linux-gnu,
powerpc{64le}-linux-gnu, s390{x}-linux-gnu, sparc{64}-linux-gnu,
and x86_64-linux-gnu).

Also ran a full check on aarch64-linux-gnu, x86_64-linux-gnu, i686-linux-gnu,
arm-linux-gnueabhf, and powerpc64le-linux-gnu.

[BZ #14092]
* conform/data/threads.h-data (mtx_plain): New constant.
(mtx_recursive): Likewise.
(mtx_timed): Likewise.
(mtx_t): New type.
(mtx_init): New function.
(mtx_lock): Likewise.
(mtx_timedlock): Likewise.
(mtx_trylock): Likewise.
(mtx_unlock): Likewise.
(mtx_destroy): Likewise.
* nptl/Makefile (libpthread-routines): Add mtx_destroy, mtx_init,
mtx_lock, mtx_timedlock, mtx_trylock, and mtx_unlock object.
* nptl/Versions (libpthread) [GLIBC_2.28]): Add mtx_init, mtx_lock,
mtx_timedlock, mtx_trylock, mtx_unlock, and mtx_destroy.
* nptl/mtx_destroy.c: New file.
* nptl/mtx_init.c: Likewise.
* nptl/mtx_lock.c: Likewise.
* nptl/mtx_timedlock.c: Likewise.
* nptl/mtx_trylock.c: Likewise.
* nptl/mtx_unlock.c: Likewise.
* sysdeps/nptl/threads.h (mtx_plain): New enumeration.
(mtx_recursive): Likewise.
(mtx_timed): Likewise.
(mtx_t): New type.
(mtx_init): New prototype.
(mtx_lock): Likewise.
(mtx_timedlock): Likewise.
(mtx_trylock): Likewise.
(mtx_unlock): Likewise.
(mtx_destroy): Likewise.

5 years agonptl: Add C11 threads thrd_* functions
Adhemerval Zanella [Tue, 27 Jun 2017 13:29:29 +0000 (10:29 -0300)]
nptl: Add C11 threads thrd_* functions

This patch adds the thrd_* definitions from C11 threads (ISO/IEC 9899:2011),
more specifically thrd_create, thrd_curent, rhd_detach, thrd_equal,
thrd_exit, thrd_join, thrd_sleep, thrd_yield, and required types.

Mostly of the definitions are composed based on POSIX conterparts, such as
thrd_t (using pthread_t).  For thrd_* function internally direct
POSIX pthread call are used with the exceptions:

  1. thrd_start uses pthread_create internal implementation, but changes
     how to actually calls the start routine.  This is due the difference
     in signature between POSIX and C11, where former return a 'void *'
     and latter 'int'.
     To avoid calling convention issues due 'void *' to int cast, routines
     from C11 threads are started slight different than default pthread one.
     Explicit cast to expected return are used internally on pthread_create
     and the result is stored back to void also with an explicit cast.

  2. thrd_sleep uses nanosleep internal direct syscall to avoid clobbering
     errno and to handle expected standard return codes.  It is a
     cancellation entrypoint to be consistent with both thrd_join and
     cnd_{timed}wait.

  3. thrd_yield also uses internal direct syscall to avoid errno clobbering.

Checked with a build for all major ABI (aarch64-linux-gnu, alpha-linux-gnu,
arm-linux-gnueabi, i386-linux-gnu, ia64-linux-gnu, m68k-linux-gnu,
microblaze-linux-gnu [1], mips{64}-linux-gnu, nios2-linux-gnu,
powerpc{64le}-linux-gnu, s390{x}-linux-gnu, sparc{64}-linux-gnu,
and x86_64-linux-gnu).

Also ran a full check on aarch64-linux-gnu, x86_64-linux-gnu, i686-linux-gnu,
arm-linux-gnueabhf, and powerpc64le-linux-gnu.

[BZ #14092]
* conform/Makefile (conformtest-headers-ISO11): Add threads.h.
(linknamespace-libs-ISO11): Add libpthread.a.
* conform/data/threads.h-data: New file: add C11 thrd_* types and
functions.
* include/stdc-predef.h (__STDC_NO_THREADS__): Remove definition.
* nptl/Makefile (headers): Add threads.h.
(libpthread-routines): Add new C11 thread thrd_create, thrd_current,
thrd_detach, thrd_equal, thrd_exit, thrd_join, thrd_sleep, and
thrd_yield.
* nptl/Versions (libpthread) [GLIBC_2.28]): Add new C11 thread
thrd_create, thrd_current, thrd_detach, thrd_equal, thrd_exit,
thrd_join, thrd_sleep, and thrd_yield symbols.
* nptl/descr.h (struct pthread): Add c11 field.
* nptl/pthreadP.h (ATTR_C11_THREAD): New define.
* nptl/pthread_create.c (START_THREAD_DEFN): Call C11 thread start
routine with expected function prototype.
(__pthread_create_2_1): Add C11 threads check based on attribute
value.
* sysdeps/unix/sysdep.h (INTERNAL_SYSCALL_CANCEL): New macro.
* nptl/thrd_create.c: New file.
* nptl/thrd_current.c: Likewise.
* nptl/thrd_detach.c: Likewise.
* nptl/thrd_equal.c: Likewise.
* nptl/thrd_exit.c: Likewise.
* nptl/thrd_join.c: Likewise.
* nptl/thrd_priv.h: Likewise.
* nptl/thrd_sleep.c: Likewise.
* nptl/thrd_yield.c: Likewise.
* include/threads.h: Likewise.

5 years agoAdd <bits/indirect-return.h>
H.J. Lu [Tue, 24 Jul 2018 14:55:35 +0000 (07:55 -0700)]
Add <bits/indirect-return.h>

Add <bits/indirect-return.h> and include it in <ucontext.h>.
__INDIRECT_RETURN defined in <bits/indirect-return.h> indicates if
swapcontext requires special compiler treatment.  The default
__INDIRECT_RETURN is empty.

On x86, when shadow stack is enabled, __INDIRECT_RETURN is defined
with indirect_return attribute, which has been added to GCC 9, to
indicate that swapcontext returns via indirect branch.  Otherwise
__INDIRECT_RETURN is defined with returns_twice attribute.

When shadow stack is enabled, remove always_inline attribute from
prepare_test_buffer in string/tst-xbzero-opt.c to avoid:

tst-xbzero-opt.c: In function ‘prepare_test_buffer’:
tst-xbzero-opt.c:105:1: error: function ‘prepare_test_buffer’ can never be inlined because it uses setjmp
 prepare_test_buffer (unsigned char *buf)

when indirect_return attribute isn't available.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* bits/indirect-return.h: New file.
* misc/sys/cdefs.h (__glibc_has_attribute): New.
* sysdeps/x86/bits/indirect-return.h: Likewise.
* stdlib/Makefile (headers): Add bits/indirect-return.h.
* stdlib/ucontext.h: Include <bits/indirect-return.h>.
(swapcontext): Add __INDIRECT_RETURN.
* string/tst-xbzero-opt.c (ALWAYS_INLINE): New.
(prepare_test_buffer): Use it.

5 years agoFix out-of-bounds access in IBM-1390 converter (bug 23448)
Andreas Schwab [Tue, 24 Jul 2018 12:08:34 +0000 (14:08 +0200)]
Fix out-of-bounds access in IBM-1390 converter (bug 23448)

The IBM-1390 converter can consume/produce two UCS4 characters in each
loop.

5 years agox86: Update vfork to pop shadow stack
H.J. Lu [Tue, 24 Jul 2018 12:20:48 +0000 (05:20 -0700)]
x86: Update vfork to pop shadow stack

The shadow stack prevents us from pushing the saved return PC onto
the stack and returning normally.  Instead we pop the shadow stack
and return directly.  This is the safest way to return and ensures
any stack manipulations done by the vfork'd child doesn't cause the
parent to terminate when CET is enabled.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* sysdeps/unix/sysv/linux/i386/vfork.S (SYSCALL_ERROR_HANDLER):
Redefine if shadow stack is enabled.
(SYSCALL_ERROR_LABEL): Likewise.
(__vfork): Pop shadow stack and jump back to to caller directly
when shadow stack is in use.
* sysdeps/unix/sysv/linux/x86_64/vfork.S (SYSCALL_ERROR_HANDLER):
Redefine if shadow stack is enabled.
(SYSCALL_ERROR_LABEL): Likewise.
(__vfork): Pop shadow stack and jump back to to caller directly
when shadow stack is in use.

5 years agox86-64: Add endbr64 to tst-quadmod[12].S
H.J. Lu [Tue, 24 Jul 2018 12:11:59 +0000 (05:11 -0700)]
x86-64: Add endbr64 to tst-quadmod[12].S

Add endbr64 to tst-quadmod1.S and tst-quadmod2.S so that func and foo
can be called indirectly.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* sysdeps/x86_64/tst-quadmod1.S (func): Add endbr64 if IBT is
enabled.
(foo): Likewise.
* sysdeps/x86_64/tst-quadmod2.S (func) : Likewise.
(foo): Likewise.

5 years agoUse binutils 2.31 branch in build-many-glibcs.py.
Joseph Myers [Fri, 20 Jul 2018 16:11:15 +0000 (16:11 +0000)]
Use binutils 2.31 branch in build-many-glibcs.py.

* scripts/build-many-glibcs.py (Context.checkout): Default
binutils version to 2.31 branch.

5 years agoChange URL of gcc's tarball
Zong Li [Fri, 20 Jul 2018 15:30:05 +0000 (15:30 +0000)]
Change URL of gcc's tarball

After 7.1.0 version, there is no .bz2 format of gcc's tarball.

* scripts/build-many-glibcs.py (Context.checkout_tar): Change the
URL of gcc's tarball.

5 years agoregcomp: Fix off-by-one bug in build_equiv_class [BZ #23396]
Florian Weimer [Fri, 20 Jul 2018 09:58:51 +0000 (11:58 +0200)]
regcomp: Fix off-by-one bug in build_equiv_class [BZ #23396]

This bug is very similar to bug 23036: The existing code assumed that
the length count included the length byte itself.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>