Commit 06ab719d30 ("Fix Linux fcntl OFD
locks for non-LFS architectures (BZ#20251)") introduced the use of
fcntl64 into the utmp implementation. However, the lock file
structure was not updated to struct flock64 at that point.
If the file processing takes a long time for some reason, SIGALRM can
arrive while the file is still being processed. At that point, file
access will fail with EINTR. Disarming the timer after lock
acquisition avoids that. (If there was a previous alarm, it is the
responsibility of the caller to deal with the EINTR error.)
Commit 5a3afa9738 (login: Replace
macro-based control flow with function calls in utmp) introduced
a regression because after it, __libc_updwtmp attempts to unlock
the wrong file descriptor.
* sysdeps/aarch64/multiarch/memset_base64.S (DC_ZVA_THRESHOLD):
Disable DC ZVA code if this macro is defined as zero.
* sysdeps/aarch64/multiarch/memset_emag.S (DC_ZVA_THRESHOLD):
Change to zero to disable using DC ZVA.
C2X (current version in git, postdating the most recent public draft
available as a PDF on the WG14 website) adds the interfaces from TS
18661-4, other than the reduction functions, as unconditionally
required and visible in <math.h> for hosted implementations; the
__STDC_WANT_IEC_60559_FUNCS_EXT__ macro is not included in C2X at all
(as the reduction functions aren't included at all, expected to end up
in a version of the TS updated to be relative to C2X).
This patch updates the glibc headers accordingly, following the same
pattern used for 18661-1 functions. As the only 18661-4 functions
currently supported are the exp10 functions that have been in glibc
for a very long time, this means that in fact __GLIBC_USE
(IEC_60559_FUNCS_EXT) ends up being used only to determine the
definition of __GLIBC_USE_IEC_60559_FUNCS_EXT_C2X, not yet in any
headers other than bits/libc-header-start.h.
(I hope to add the other 18661-4 functions to glibc at some point; the
main complication is the current lack of MPFR support for many of
these functions.)
Tested for x86_64.
* bits/libc-header-start.h (__GLIBC_USE_IEC_60559_FUNCS_EXT):
Update comment.
(__GLIBC_USE_IEC_60559_FUNCS_EXT_C2X): New macro.
* bits/math-finite.h [__GLIBC_USE (IEC_60559_FUNCS_EXT)]: Change
to [__GLIBC_USE (IEC_60559_FUNCS_EXT_C2X)].
* math/bits/mathcalls.h [__GLIBC_USE (IEC_60559_FUNCS_EXT)]:
Likewise.
C2X adds the memccpy, strdup and strndup functions. This patch duly
adds __GLIBC_USE (ISOC2X) to the conditions under which <string.h>
declares them.
Tested for x86_64.
* string/string.h (memccpy): Also declare if [__GLIBC_USE (ISOC2X)].
(strdup): Likewise.
(strndup): Likewise.
This patch starts preparation for C2X support in glibc headers by
adding a feature test macro _ISOC2X_SOURCE and corresponding
__GLIBC_USE (ISOC2X). (I chose to use the newer __GLIBC_USE style for
this rather than the older __USE_* macros tested with #ifdef.) As
with other such macros, C2X features are also enabled by compiling for
a standard newer than C17, or by using _GNU_SOURCE.
This patch does not itself enable anything new in the headers for C2X;
that is to be done in followup patches. (For example, most of the TS
18661-1 functions should be declared for C2X without any
__STDC_WANT_IEC_60559_BFP_EXT__ being needed, but the ones that
18661-1 adds to Annex F because of their close relation to IEEE 754
formats do still need the WANT macro in C2X.)
Once C2X becomes an actual standard we'll presumably move to using the
actual year in the feature test macro and __GLIBC_USE, with some
period when both macro spellings are accepted, as was done with
_ISOC9X_SOURCE.
Tested for x86_64.
* include/features.h (_ISOC2X_SOURCE): New feature test macro.
Undefine and define to 1 if [_GNU_SOURCE].
(__GLIBC_USE_ISOC2X): New macro. Undefine and redefine depending
on [_ISOC2X_SOURCE] and [__STDC_VERSION__ > 201710L].
(__USE_ISOC11): Also define to 1 if [_ISOC2X_SOURCE].
(__USE_ISOC99): Likewise.
(__USE_ISOC95): Likewise.
* manual/creature.texi (_ISOC2X_SOURCE): Document.
dlopen can no longer open PIE binaries, so it is not necessary
to link the executable as non-PIE to trigger a dlopen failure.
If we hard-code the path to the real executable, we can run the test
with and without hard-coded paths because the dlopen path will not
be recognized as the main program in both cases. (With an explict
loader invocation, the loader currently adds argv[0] to l_libname
for the main map and the dlopen call suceeds as a result; it does
not do that in standard mode.)
Fixes `<total type="rest" size="..."> incorrectly showing as 0 most
of the time.
The rest value being wrong is significant because to compute the
actual amount of memory handed out via malloc, the user must subtract
it from <system type="current" size="...">. That result being wrong
makes investigating memory fragmentation issues like
<https://bugzilla.redhat.com/show_bug.cgi?id=843478> close to
impossible.
This patch adds the SYNC_FILE_RANGE_WRITE_AND_WAIT constant from Linux
5.2 (a new name for a combination of existing bits, not actually a new
kernel interface) to bits/fcntl-linux.h.
Tested for x86_64.
* sysdeps/unix/sysv/linux/bits/fcntl-linux.h [__USE_GNU]
(SYNC_FILE_RANGE_WRITE_AND_WAIT): New macro.
This patch adds the AT_RECURSIVE constant from Linux 5.2 to glibc's
fcntl.h. (Existing practice in glibc is that the AT_* constants go
directly in io/fcntl.h rather than somewhere OS-specific.)
Tested for x86_64.
* io/fcntl.h [__USE_GNU] (AT_RECURSIVE): New macro.
The commit 5e855c8954
"s390: Enable VDSO for static linking" removed the definition of VDSO_SETUP
which leads to not setup the vdso symbols.
Instead it jumps to false addresses.
This patch just re adds the removed VDSO_SETUP macro definition.
ChangeLog:
* sysdeps/unix/sysv/linux/s390/init-first.c (VDSO_SETUP): New define.
This patch adds the CLONE_PIDFD constant from Linux 5.2 to glibc's
bits/sched.h.
Tested for x86_64.
* sysdeps/unix/sysv/linux/bits/sched.h [__USE_GNU] (CLONE_PIDFD):
New macro.
This patch assumes static vDSO is supported as default, it is now supported
on all current architectures that support vDSO. It allows removing both
ALWAYS_USE_VSYSCALL define, which an architecture requires to explicit define
and USE_VSYSCALL (which defines vDSO only for shared or if architecture defines
ALWAYS_USE_VSYSCALL).
Checked with a build against all affected ABIs.
[BZ #19767]
* sysdeps/unix/sysv/linux/aarch64/sysdep.h (ALWAYS_USE_VSYSCALL):
Remove definition.
* sysdeps/unix/sysv/linux/arm/sysdep.h (ALWAYS_USE_VSYSCALL):
Likewise.
* sysdeps/unix/sysv/linux/i386/sysdep.h (ALWAYS_USE_VSYSCALL):
Likewise.
* sysdeps/unix/sysv/linux/mips/mips32/sysdep.h (ALWAYS_USE_VSYSCALL):
Likewise.
* sysdeps/unix/sysv/linux/mips/mips64/n32/sysdep.h
(ALWAYS_USE_VSYSCALL): Likewise.
* sysdeps/unix/sysv/linux/mips/mips64/n64/sysdep.h
(ALWAYS_USE_VSYSCALL): Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep.h
(ALWAYS_USE_VSYSCALL): Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc64/sysdep.h
(ALWAYS_USE_VSYSCALL): Likewise.
* sysdeps/unix/sysv/linux/riscv/sysdep.h (ALWAYS_USE_VSYSCALL):
Likewise.
* sysdeps/unix/sysv/linux/s390/s390-32/sysdep.h
(ALWAYS_USE_VSYSCALL): Likewise.
* sysdeps/unix/sysv/linux/s390/s390-64/sysdep.h
(ALWAYS_USE_VSYSCALL): Likewise.
* sysdeps/unix/sysv/linux/sparc/sysdep.h (ALWAYS_USE_VSYSCALL):
Likewise.
* sysdeps/unix/sysv/linux/x86_64/sysdep.h (ALWAYS_USE_VSYSCALL):
Likewise.
* sysdeps/unix/sysv/linux/x86/libc-vdso.h: Remove #if USE_VSYSCALL.
* sysdeps/unix/sysv/linux/sysdep-vdso.h: Likewise.
* sysdeps/unix/sysv/linux/sysdep.h (ALWAYS_USE_VSYSCALL,
USE_VSYSCALL): Remove defitions.
Although s390 only enables vDSO for dynamically linked elf binaries
(arch/s390/kernel/vdso.c:217), there is no indication in the code or
associated commit message for why not enable it for statically linked
binaries as well. To double check, I rebuilt a kernel with the
check removed and the vDSO does work for static build for supplied
symbols.
Checked on s390x-linux-gnu and s390-linux-gnu.
[BZ #19767]
* sysdeps/unix/sysv/linux/s390/init-first.c: Remove #ifdef SHARED.
* sysdeps/unix/sysv/linux/s390/libc-vdso.h: Likewise.
* sysdeps/unix/sysv/linux/s390/s390-32/sysdep.h
(ALWAYS_USE_VSYSCALL): Define.
* sysdeps/unix/sysv/linux/s390/s390-64/sysdep.h
(ALWAYS_USE_VSYSCALL): Likewise.
There is just one file-based implementation, so this dispatch
mechanism is unnecessary. Instead of the vtable pointer
__libc_utmp_jump_table, use a non-negative file_fd as the indicator
that the backend is initialized.
This patch updates the Linux kernel version in a comment in
syscall-names.list to agree with the following "kernel" line.
* sysdeps/unix/sysv/linux/syscall-names.list: Update comment.
The tst-mman-consts.py test includes a kernel version number, to avoid
failures because of newly added constants in the kernel (if kernel
headers are newer than this version of glibc) or missing constants in
the kernel (if kernel headers are older than this version of glibc).
This patch updates it to 5.2 to reflect that the MAP_* constants in
glibc are still current as of that kernel version.
Tested with build-many-glibcs.py.
* sysdeps/unix/sysv/linux/tst-mman-consts.py (main): Update Linux
kernel version number to 5.2.
Some implementations in sysdeps/powerpc/powerpc64/power8/*.S still had
pre power8 compatible binutils hardcoded macros and were not using
.machine power8.
This patch should not have semantic changes, in fact it should have the
same exact code generated.
Tested that generated stripped shared objects are identical when
using "strip --remove-section=.note.gnu.build-id".
Checked on:
- powerpc64le, power9, build-many-glibcs.py, gcc 6.4.1 20180104, binutils 2.26.2.20160726
- powerpc64le, power8, debian 9, gcc 6.3.0 20170516, binutils 2.28
- powerpc64le, power9, ubuntu 19.04, gcc 8.3.0, binutils 2.32
- powerpc64le, power9, opensuse tumbleweed, gcc 9.1.1 20190527, binutils 2.32
- powerpc64, power9, debian 10, gcc 8.3.0, binutils 2.31.1
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
This is missing bit for fully fix BZ#15813 (the other two were fixed
by 359653aaac).
Checked on x86_64-linux-gnu.
[BZ #15813]
sysdeps/posix/tempname.c (__gen_tempname): get entrypy on each
attempt.
The changes introduce a memory leak for gconv steps arrays whose
first element is an internal conversion, which has a fixed
reference count which is not decremented. As a result, after the
change in commit 50ce3eae5b, the steps
array is never freed, resulting in an unbounded memory leak.
This reverts commit 50ce3eae5b
("gconv: Check reference count in __gconv_release_cache
[BZ #24677]") and commit 7e740ab2e7
("libio: Fix gconv-related memory leak [BZ #24583]"). It
reintroduces bug 24583. (Bug 24677 was just a regression caused by
the second commit.)
Commit ffe8a9a831, "powerpc: Remove
rt_sigreturn usage on context function", removed from powerpc32
swapcontext a setting of r31 that is relied upon in subsequent code.
I'm not sure why this didn't produce test failures in Adhemerval's
32-bit testing; in my (soft-float) testing in preparation for 2.30
release, I see several context-related failures
FAIL: stdlib/tst-makecontext2
FAIL: stdlib/tst-makecontext3
FAIL: stdlib/tst-setcontext
FAIL: stdlib/tst-setcontext2
FAIL: stdlib/tst-setcontext4
FAIL: stdlib/tst-setcontext7
FAIL: stdlib/tst-setcontext9
FAIL: stdlib/tst-swapcontext1
that did not appear in 2.29 testing. This patch restores the removed
register setting in question, and thus fixes those failures.
Tested for powerpc (soft-float).
* sysdeps/unix/sysv/linux/powerpc/powerpc32/swapcontext-common.S
(__CONTEXT_FUNC_NAME): Restore setting of r31.
Recent GCC versions warn about the attempt to return the address of a
local variable:
tst-pthread-getattr.c: In function ‘allocate_and_test’:
tst-pthread-getattr.c:54:10: error: function returns address of local variable [-Werror=return-local-addr]
54 | return mem;
| ^~~
In file included from ../include/alloca.h:3,
from tst-pthread-getattr.c:26:
../stdlib/alloca.h:35:23: note: declared here
35 | # define alloca(size) __builtin_alloca (size)
| ^~~~~~~~~~~~~~~~~~~~~~~
tst-pthread-getattr.c:51:9: note: in expansion of macro ‘alloca’
51 | mem = alloca ((size_t) (mem - target));
| ^~~~~~
The address itself is used in a check in the caller, so using
uintptr_t instead is reasonable.
Commit 35e038c1d2 started to use an
incomplete list of subdirs based on $(all-subdirs) causing
testroot.pristine to miss files from nss.
Tested if the list of files in testroot.pristine remains the same.
[BZ #24794]
* Makeconfig (all-subdirs): Improved source comments.
* Makefile (testroot.pristine/install.stamp): Pass
subdirs='$(sorted-subdirs)' to make install.
This fixes a regression introduced in commit
7e740ab2e7 ("libio: Fix gconv-related
memory leak [BZ #24583]").
__gconv_release_cache is only ever called with heap-allocated
arrays which contain at least one member. The statically allocated
ASCII steps are filtered out by __wcsmbs_close_conv.
When compiled with -O3 and AVX, GCC 8 and 9 optimize some loops in
sysdeps/ieee754/dbl-64/branred.c with 256-bit vector instructions,
which leads to store forward stall:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90579
There is no easy fix in compiler. This patch limits vector width to
128 bits to work around this issue. It improves performance of sin
and cos by more than 40% on Skylake compiled with -O3 -march=skylake.
Tested with GCC 7/8/9 on x86-64.
[BZ #24603]
* sysdeps/x86_64/configure.ac: Check if -mprefer-vector-width=128
works.
* sysdeps/x86_64/configure: Regenerated.
* sysdeps/x86_64/fpu/Makefile (CFLAGS-branred.c): New. Set
to -mprefer-vector-width=128 if supported.
The kernel changes for a 64-bit time_t on 32-bit architectures
resulted in <asm/socket.h> indirectly including <linux/posix_types.h>.
The latter is not namespace-clean for the POSIX version of
<sys/socket.h>.
This issue has persisted across several Linux releases, so this commit
creates our own copy of the SO_* definitions for !__USE_MISC mode.
The new test socket/tst-socket-consts ensures that the copy is
consistent with the kernel definitions (which vary across
architectures). The test is tricky to get right because CPPFLAGS
includes include/libc-symbols.h, which in turn defines _GNU_SOURCE
unconditionally.
Tested with build-many-glibcs.py. I verified that a discrepancy in
the definitions actually results in a failure of the
socket/tst-socket-consts test.
Whenever a sub-make is created, it inherits the variable subdirs from its
parent. This is also true when make check is called with a restricted
list of subdirs. In this scenario, make install is executed "partially"
and testroot.pristine ends up with an incomplete installation.
[BZ #24794]
* Makefile (testroot.pristine/install.stamp): Pass
subdirs='$(all-subdirs)' to make install.
Reviewed-by: DJ Delorie <dj@redhat.com>
