This patch avoid probing the __NR_clock_getttime64 syscall each time
__clock_gettime64 is issued on a kernel without 64 bit time support.
Once ENOSYS is obtained, only 32-bit clock_gettime are used.
The following snippet:
clock_gettime (CLOCK_REALTIME, &(struct timespec) { 0 });
clock_gettime (CLOCK_MONOTONIC, &(struct timespec) { 0 });
clock_gettime (CLOCK_BOOTTIME, &(struct timespec) { 0 });
clock_gettime (20, &(struct timespec) { 0 });
On a kernel without 64 bit time support and with vDSO support results
on the following syscalls:
syscall_0x193(0, 0xff87ba30, [...]) = -1 ENOSYS (Function not implemented)
clock_gettime(CLOCK_BOOTTIME, {tv_sec=927082, tv_nsec=474382032}) = 0
clock_gettime(0x14 /* CLOCK_??? */, 0xff87b9f8) = -1 EINVAL (Invalid argument)
While on a kernel without vDSO support:
syscall_0x193(0, 0xbec95550, 0xb6ed2000, 0x1, 0xbec95550, 0) = -1 (errno 38)
clock_gettime(CLOCK_REALTIME, {tv_sec=1576615930, tv_nsec=638250162}) = 0
clock_gettime(CLOCK_MONOTONIC, {tv_sec=1665478, tv_nsec=638779620}) = 0
clock_gettime(CLOCK_BOOTTIME, {tv_sec=1675418, tv_nsec=292932704}) = 0
clock_gettime(0x14 /* CLOCK_??? */, 0xbec95530) = -1 EINVAL (Invalid argument)
Checked on i686-linux-gnu on 4.15 kernel and on a 5.3 kernel.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
No architecture currently defines the vDSO symbol. On architectures
with 64-bit time_t the HAVE_CLOCK_GETTIME_VSYSCALL is renamed to
HAVE_CLOCK_GETTIME64_VSYSCALL, it simplifies clock_gettime code.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
This patch moves the vDSO setup from libc to loader code, just after
the vDSO link_map setup. For static case the initialization
is moved to _dl_non_dynamic_init instead.
Instead of using the mangled pointer, the vDSO data is set as
attribute_relro (on _rtld_global_ro for shared or _dl_vdso_* for
static). It is read-only even with partial relro.
It fixes BZ#24967 now that the vDSO pointer is setup earlier than
malloc interposition is called.
Also, vDSO calls should not be a problem for static dlopen as
indicated by BZ#20802. The vDSO pointer would be zero-initialized
and the syscall will be issued instead.
Checked on x86_64-linux-gnu, i686-linux-gnu, aarch64-linux-gnu,
arm-linux-gnueabihf, powerpc64le-linux-gnu, powerpc64-linux-gnu,
powerpc-linux-gnu, s390x-linux-gnu, sparc64-linux-gnu, and
sparcv9-linux-gnu. I also run some tests on mips.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
The code is similar to the one at elf/dl-reloc.c, where it checks for
the l_relro_size from the link_map (obtained from PT_GNU_RELRO header
from program headers) and calls_dl_protected_relro.
For testing I will use the ones proposed by Florian's patch
'elf: Add tests for working RELRO protection' [1].
Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
aarch64-linux-gnu, s390x-linux-gnu, and sparc64-linux-gnu. I also
check with --enable-static pie on x86_64-linux-gnu, i686-linux-gnu,
and aarch64-linux-gnu which seems the only architectures where
static PIE is actually working (as per 9d7a3741c9, on
arm-linux-gnueabihf, powerpc64{le}-linux-gnu, and s390x-linux-gnu
I am seeing runtime issues not related to my patch).
[1] https://sourceware.org/ml/libc-alpha/2019-10/msg00059.html
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
The IFUNC bypass to vDSO is used when USE_IFUNC_TIME is set.
Currently powerpc and x86 defines it. Otherwise the generic
implementation is used, which calls clock_gettime.
Checked on powerpc64le-linux-gnu, powerpc64-linux-gnu,
powerpc-linux-gnu-power4, x86_64-linux-gnu, and i686-linux-gnu.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
The IFUNC bypass to vDSO is used when USE_IFUNC_GETTIMEOFDAY is set.
Currently aarch64, powerpc*, and x86 defines it. Otherwise the
generic implementation is used, which calls clock_gettime.
Checked on aarch64-linux-gnu, powerpc64le-linux-gnu,
powerpc64-linux-gnu, powerpc-linux-gnu-power4, x86_64-linux-gnu,
and i686-linux-gnu.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
The clock_getres is a new implementation added on Linux 5.4
(abed3d826f2f).
Checked with a build against mips-linux-gnu and mips64-linux-gnu.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Add the missing time and clock_getres vDSO symbol names on x86.
For time, the iFUNC already uses expected name so it affects only
the static build.
The clock_getres is a new implementation added on Linux 5.3
(f66501dc53e72).
Checked on x86-linux-gnu and i686-linux-gnu.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
The auto-generated vDSO call shows some issues:
- It requires sync the auto-generated C file with current glibc
implementation;
- It still uses symbol redirections hacks where libc-symbols.h
provide macros that uses compiler builtins
(libc_ifunc_redirected for instance);
- It does not handle all required compiler handling
(inhibit_stack_protector on iFUNC resolver).
- No architecure uses it.
Checked with a build against all major ABIs.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
This is the only use of auto-generation syscall which uses a vDSO
plus IFUNC and the current x86 generic implementation already covers
the expected semantic.
Checked on x86_64-linux-gnu-x32.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
As indicated on libc-help [1] the ec138c67cb commit broke 32-bit
builds when configured with --enable-kernel=5.1 or higher. The
scenario 10 from [2] might also occur in this configuration and
INLINE_VSYSCALL will try to use the vDSO symbol and
HAVE_CLOCK_GETTIME64_VSYSCALL does not set HAVE_VSYSCALL prior its
usage.
Also, there is no easy way to just enable the code to use one
vDSO symbol since the macro INLINE_VSYSCALL is redefined if
HAVE_VSYSCALL is set.
Instead of adding more pre-processor handling and making the code
even more convoluted, this patch removes the requirement of defining
HAVE_VSYSCALL before including sysdep-vdso.h to enable vDSO usage.
The INLINE_VSYSCALL is now expected to be issued inside a
HAVE_*_VSYSCALL check, since it will try to use the internal vDSO
pointers.
Both clock_getres and clock_gettime vDSO code for time64_t were
removed since there is no vDSO setup code for the symbol (an
architecture can not set HAVE_CLOCK_GETTIME64_VSYSCALL).
Checked on i686-linux-gnu (default and with --enable-kernel=5.1),
x86_64-linux-gnu, aarch64-linux-gnu, and powerpc64le-linux-gnu.
I also checked against a build to mips64-linux-gnu and
sparc64-linux-gnu.
[1] https://sourceware.org/ml/libc-help/2019-12/msg00014.html
[2] https://sourceware.org/ml/libc-alpha/2019-12/msg00142.html
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
The result of INTERNAL_SYSCALL_CANCEL should be checked with
macros INTERNAL_SYSCALL_ERROR_P and INTERNAL_SYSCALL_ERRNO instead
of comparing the result directly.
Checked on powerpc-linux-gnu.
This patch adds a new macro, libm_alias_finite, to define all _finite
symbol. It sets all _finite symbol as compat symbol based on its first
version (obtained from the definition at built generated first-versions.h).
The <fn>f128_finite symbols were introduced in GLIBC 2.26 and so need
special treatment in code that is shared between long double and float128.
It is done by adding a list, similar to internal symbol redifinition,
on sysdeps/ieee754/float128/float128_private.h.
Alpha also needs some tricky changes to ensure we still emit 2 compat
symbols for sqrt(f).
Passes buildmanyglibc.
Co-authored-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
It is not specified what should be the content of d_t_fmt and date_fmt
but in the built-in C locale those fields have only one difference:
date_fmt contains "%Z" (the current time zone) while d_t_fmt does not.
For most of the locales this commit does the following operation:
copy d_t_fmt to date_fmt, and then remove "%Z" from d_t_fmt.
If "%Z" was originally missing from d_t_fmt add it to date_fmt.
It also corrects comments where necessary.
Exceptions:
* In bo_CN, dz_BT, and km_KH "%Z" has not been added to date_fmt because
it was too difficult. In these locales date_fmt has been set to the
copy of d_t_fmt.
* In en_DK "%Z" has not been removed from d_t_fmt in order to preserve
the conformance with the standard mentioned in the comment.
The command to identify and initially edit the locales that need the
update was:
for i in `grep -lw d_t_fmt *`
do
if ! grep -qw date_fmt $i ; then
awk '/d_t_fmt/ { print $0; gsub("d_t_fmt", "date_fmt"); } //{ print $0 }' < $i > $i.next
mv $i.next $i
fi
done
and then each file was further edited manually.
Hurd uses an empty prefix, so the linker scripts end up in /lib, the
find command picked them up, and stripping them failed because they
are not ELF files.
Since the switch away from auto-generated wrappers for these system
calls, the kludge is already included in the C source file of the
system call wrapper.
This command uses pre-built compilers to re-install the Linux headers
from the current sources into a temporary location and runs glibc's
“make update-syscalls-lists” against that. This updates the glibc
source tree with the current system call numbers.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
The new classes GlibcPolicyForCompiler and GlibcPolicyForBuild allow
customization of the Glibc.build_glibc method, replacing the existing
for_compiler flag.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Use <arch-syscall.h> instead of <asm/unistd.h> to obtain the system
call numbers. A few direct includes of <asm/unistd.h> need to be
removed (if the system call numbers are already provided indirectly
by <sysdep.h>) or replaced with <sys/syscall.h>.
Current Linux headers for alpha define the required system call names,
so most of the _NR_* hacks are no longer needed. For the 32-bit arm
architecture, eliminate the INTERNAL_SYSCALL_ARM macro, now that we
have regular system call names for cacheflush and set_tls. There are
more such cleanup opportunities for other architectures, but these
cleanups are required to avoid macro redefinition errors during the
build.
For ia64, it is desirable to use <asm/break.h> directly to obtain
the break number for system calls (which is not a system call number
itself). This requires replacing __BREAK_SYSCALL with
__IA64_BREAK_SYSCALL because the former is defined as an alias in
<asm/unistd.h>, but not in <asm/break.h>.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
The new tables are currently only used for consistency checks
with the installed kernel headers and the architecture-independent
system call names table. They are based on Linux 5.4.
The goal is to use these architecture-specific tables to ensure
that system call wrappers are available irrespective of the version
of the installed kernel headers.
The tables are formatted in the form of C header files so that they
can be used directly in an #include directive, without external
preprocessing. (External preprocessing of a plain table file
would introduce cross-subdirectory dependency issues.) However,
the intent is that they can still be treated as tables and can be
processed by simple tools.
The irregular system call names on 32-bit arm add a complication.
The <fixup-asm-unistd.h> header is introduced to work around that,
and the system calls are listed under regular names in the
<arch-syscall.h> file.
A make target, update-syscalls-list, is added to patch the glibc
sources with data from the current kernel headers.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
I've updated copyright dates in glibc for 2020. This is the patch for
the changes not generated by scripts/update-copyrights and subsequent
build / regeneration of generated files. As well as the usual annual
updates, mainly dates in --version output (minus libc.texinfo which
previously had to be handled manually but is now successfully updated
by update-copyrights), there is a fix to
sysdeps/unix/sysv/linux/powerpc/bits/termios-c_lflag.h where a typo in
the copyright notice meant it failed to be updated automatically.
Please remember to include 2020 in the dates for any new files added
in future (which means updating any existing uncommitted patches you
have that add new files to use the new copyright dates in them).
Currently d_t_fmt formats time as "plkst. %H un %M". A quick Google
search says that "plkst." means "o’clock" and "un" means "and".
Also this format does not display seconds.
CLDR does not mention anything like that. We have no reason to use
anything different than "%H:%M:%S".
This adds _hurd_sigstate_set_global_rcv used by libpthread to enable
POSIX-confirming behavior of signals on a per-thread basis.
This also provides a sigstate destructor _hurd_sigstate_delete, and a
global process signal state, which needs to be locked and check when
global disposition is enabled, thus the addition of _hurd_sigstate_lock
_hurd_sigstate_actions _hurd_sigstate_pending _hurd_sigstate_unlock helpers.
This also updates all the glibc code accordingly.
This also drops support for get_int(INIT_SIGMASK), which did not make sense
any more since we do not have a single signal thread any more.
During fork/spawn, this also reinitializes the child global sigstate's
lock. That cures an issue that would very rarely cause a deadlock in the
child in fork, tries to unlock ss' critical section lock at the end of
fork. This will typically (always?) be observed in /bin/sh, which is not
surprising as that is the foremost caller of fork.
To reproduce an intermediate state, add an endless loop if
_hurd_global_sigstate is locked after __proc_dostop (cast through
volatile); that is, while still being in the fork's parent process.
When that triggers (use the libtool testsuite), the signal thread has
already locked ss (which is _hurd_global_sigstate), and is stuck at
hurdsig.c:685 in post_signal, trying to lock _hurd_siglock (which the
main thread already has locked and keeps locked until after
__task_create). This is the case that ss->thread == MACH_PORT_NULL, that
is, a global signal. In the main thread, between __proc_dostop and
__task_create is the __thread_abort call on the signal thread which would
abort any current kernel operation (but leave ss locked). Later in fork,
in the parent, when _hurd_siglock is unlocked in fork, the parent's
signal thread can proceed and will unlock eventually the global sigstate.
In the client, _hurd_siglock will likewise be unlocked, but the global
sigstate never will be, as the client's signal thread has been configured
to restart execution from _hurd_msgport_receive. Thus, when the child
tries to unlock ss' critical section lock at the end of fork, it will
first lock the global sigstate, will spin trying to lock it, which can
never be successful, and we get our deadlock.
Options seem to be:
* Move the locking of _hurd_siglock earlier in post_signal -- but that
may generally impact performance, if this locking isn't generally
needed anyway?
On the other hand, would it actually make sense to wait here until we
are not any longer in a critical section (which is meant to disable
signal delivery anyway (but not for preempted signals?))?
* Clear the global sigstate in the fork's child with the rationale that
we're anyway restarting the signal thread from a clean state. This
has now been implemented.
Why has this problem not been observed before Jérémie's patches? (Or has
it? Perhaps even more rarely?) In _S_msg_sig_post, the signal is now
posted to a *global receiver thread*, whereas previously it was posted to
the *designated signal-receiving thread*. The latter one was in a
critical section in fork, so didn't try to handle the signal until after
leaving the critical section? (Not completely analyzed and verified.)
Another question is what the signal is that is being received
during/around the time __proc_dostop executes.
This should not change the current behavior, although this fixes a few
minor bugs which were made apparent in the process of global signal
disposition work:
- Split into more functions
- Scope variables more restrictively
- Split out inner functions
- refactor check_pending_signals
- make sigsuspend POSIX-conformant.
- fix uninitialized act value.
Adapted from the Linux x86 functions.
Not thoroughly tested, but manual testing as well as glibc tests look fine, and
manual -lpthread testing also looks fine (within the given bounds for a new
stack to be used with makecontext).
This has also been in use in Debian since 2013.
Some compiler versions, e.g. GCC 7, complain when -mlong-double-128 is
used together with -mabi=ibmlongdouble or -mabi=ieeelongdouble,
producing the following error message:
cc1: error: ‘-mabi=ibmlongdouble’ requires ‘-mlong-double-128’
This patch removes -mlong-double-128 from the compilation lines that
explicitly request -mabi=*longdouble.
Tested for powerpc64le.
Reviewed-by: Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com>
Some of the files that provide stdio.h and wchar.h functions have a
filename prefixed with 'io', such as 'iovsprintf.c'. On platforms that
imply ldbl-128ibm-compat, these files must be compiled with the flag
-mabi=ibmlongdouble. This patch adds this flag to their compilation.
Notice that this is not required for the other files that provide
similar functions, because filenames that are not prefixed with 'io'
have ldbl-128ibm-compat counterparts in the Makefile, which already adds
-mabi=ibmlongdouble to them.
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
On platforms where long double has IEEE binary128 format as a third
option (initially, only powerpc64le), many exported functions are
redirected to their __*ieee128 equivalents. This redirection is
provided by installed headers such as stdio-ldbl.h, and is supposed to
work correctly with user code.
However, during the build of glibc, similar redirections are employed,
in internal headers, such as include/stdio.h, in order to avoid extra
PLT entries. These redirections conflict with the redirections to
__*ieee128, and must be avoided during the build. This patch protects
the second redirections with a test for __LONG_DOUBLE_USES_FLOAT128, a
new macro that is defined to 1 when functions that deal with long double
typed values reuses the _Float128 implementation (this is currently only
true for powerpc64le).
Tested for powerpc64le, x86_64, and with build-many-glibcs.py.
Co-authored-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
Reviewed-by: Florian Weimer <fweimer@redhat.com>
All architectures now uses the Linux generic implementation which
uses __NR_rt_sigprocmask.
Checked on x86_64-linux-gnu, sparc64-linux-gnu, ia64-linux-gnu,
s390x-linux-gnu, and alpha-linux-gnu.
The functions do not fail regardless of the argument value. Also, for
Linux the return value is not correct on some platforms due the missing
usage of INTERNAL_SYSCALL_ERROR_P / INTERNAL_SYSCALL_ERRNO macros.
Checked on x86_64-linux-gnu, i686-linux-gnu, and sparc64-linux-gnu.
On powerpc64le, the libm_alias_float128_other_r_ldbl macro is
used to create an alias between totalorderf128 and __totalorderlieee128,
as well as between the totalordermagf128 and __totalordermaglieee128.
However, the totalorder* and totalordermag* functions changed their
parameter type since commit ID 42760d7646 and got compat symbols for
their old versions. With this change, the aforementioned macro would
create two conflicting aliases for __totalorderlieee128 and
__totalordermaglieee128.
This patch avoids the creation of the alias between the IEEE long double
symbols (__totalorderl*ieee128) and the compat symbols, because the IEEE
long double functions have never been exported thus don't need such
compat symbol.
Tested for powerpc64le.
Reviewed-by: Joseph Myers <joseph@codesourcery.com>
This patch adds IEEE long double versions of q*cvt* functions for
powerpc64le. Unlike all other long double to/from string conversion
functions, these do not rely on internal functions that can take
floating-point numbers with different formats and act on them
accordingly, instead, the related files are rebuilt with the
-mabi=ieeelongdouble compiler flag set.
Having -mabi=ieeelongdouble passed to the compiler causes the object
files to be marked with a .gnu_attribute that is incompatible with the
.gnu_attribute in files built with -mabi=ibmlongdouble (the default).
The difference causes error messages similar to the following:
ld: libc_pic.a(s_isinfl.os) uses IBM long double,
libc_pic.a(ieee128-qefgcvt_r.os) uses IEEE long double.
collect2: error: ld returned 1 exit status
make[2]: *** [../Makerules:649: libc_pic.os] Error 1
Although this warning is useful in other situations, the library
actually needs to have functions with different long double formats, so
.gnu_attribute generation is explicitly disabled for these files with
the use of -mno-gnu-attribute.
Tested for powerpc64le on the branch that actually enables the
sysdeps/ieee754/ldbl-128ibm-compat for powerpc64le.
Reviewed-by: Paul E. Murphy <murphyp@linux.ibm.com>
This patch refactors the *cvt functions implementation in a way that
makes it easier to re-use them for implementing the IEEE long double on
powerpc64le. By removing the macros that generate the function names
(APPEND combined with FUNC_PREFIX), the new code makes it easier to
define new function names, such as __qecvtieee128.
Tested that installed stripped binaries for all build-many-glibcs
targets remain identical before and after this patch. Also tested for
powerpc64le and x86_64.
Reviewed-by: Paul E. Murphy <murphyp@linux.ibm.com>
This patch refactors the *cvt functions implementation in a way that
makes it easier to re-use them for implementing the IEEE long double on
powerpc64le. By splitting the implementation per se in one file
(efgcvt-template.c) and the alias definitions in others (e.g. efgcvt.c),
the new code makes it easier to define new function names, such as
__qecvtieee128.
Tested that installed stripped binaries for all build-many-glibcs
targets remain identical before and after this patch. Also tested for
powerpc64le and x86_64.
Reviewed-by: Paul E. Murphy <murphyp@linux.ibm.com>
Due to the branch prediction issue of Kunpeng processor, we found
memset_generic has poor performance on middle sizes setting, and so
we reconstructed the logic, expanded the loop by 4 times in set_long
to solve the problem, even when setting below 1K sizes have benefit.
Another change is that DZ_ZVA seems no work when setting zero, so we
discarded it and used set_long to set zero instead. Fewer branches and
predictions also make the zero case have slightly improvement.
Checked on aarch64-linux-gnu.
Reviewed-by: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
Optimize the strlen implementation by using vector operations and
loop unrolling in main loop.Compared to __strlen_generic,it reduces
latency of cases in bench-strlen by 7%~18% when the length of src
is greater than 128 bytes, with gains throughout the benchmark.
Checked on aarch64-linux-gnu.
Reviewed-by: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
Kunpeng processer is a 64-bit Arm-compatible CPU released by Huawei,
and we have already signed a copyright assignement with the FSF.
This patch adds its to cpu list, and related macro for IFUNC.
Checked on aarch64-linux-gnu.
Reviewed-by: Szabolcs Nagy <Szabolcs.Nagy@arm.com>
