Now that both pthread_mutex_t and pthread_rwlock_t static initializer
are parametrized in their own headers HPPA pthread.h is identical to
generic nptl one.
This patch adds a default pthread-offsets.h based on default
thread definitions from struct_mutex.h and struct_rwlock.h.
The idea is to simplify new ports inclusion.
This patch adds a default pthreadtypes-arch.h, the idea is to simpify
new ports inclusion and an override is required only if the architecture
adds some arch-specific extensions or requirement.
The default values on the new generic header are based on current
architecture define value and they are not optimal compared to current
code requirements as below.
- On 64 bits __SIZEOF_PTHREAD_BARRIER_T is defined as 32 while is
sizeof (struct pthread_barrier) is 20 bytes.
- On 32 bits __SIZEOF_PTHREAD_ATTR_T is defined as 36 while
sizeof (struct pthread_attr) is 32.
The default values are not changed so the generic header could be
used by some architectures.
This patch adds a new generic __pthread_rwlock_arch_t definition meant
to be used by new ports. Its layout mimics the current usage on some
64 bits ports and it allows some ports to use the generic definition.
The arch __pthread_rwlock_arch_t definition is moved from
pthreadtypes-arch.h to another arch-specific header (struct_rwlock.h).
Also the static intialization macro for pthread_rwlock_t is set to use
an arch defined on (__PTHREAD_RWLOCK_INITIALIZER) which simplifies its
implementation.
The default pthread_rwlock_t layout differs from current ports with:
1. Internal layout is the same for 32 bits and 64 bits.
2. Internal flag is an unsigned short so it should not required
additional padding to align for word boundary (if it is the case
for the ABI).
The current way of defining the common mutex definition for POSIX and
C11 on pthreadtypes-arch.h (added by commit 06be6368da16104be5) is
not really the best options for newer ports. It requires define some
misleading flags that should be always defined as 0
(__PTHREAD_COMPAT_PADDING_MID and __PTHREAD_COMPAT_PADDING_END), it
exposes options used solely for linuxthreads compat mode
(__PTHREAD_MUTEX_USE_UNION and __PTHREAD_MUTEX_NUSERS_AFTER_KIND), and
requires newer ports to explicit define them (adding more boilerplate
code).
This patch adds a new default __pthread_mutex_s definition meant to
be used by newer ports. Its layout mimics the current usage on both
32 and 64 bits ports and it allows most ports to use the generic
definition. Only ports that use some arch-specific definition (such
as hardware lock-elision or linuxthreads compat) requires specific
headers.
For 32 bit, the generic definitions mimic the other 32-bit ports
of using an union to define the fields uses on adaptive and robust
mutexes (thus not allowing both usage at same time) and by using a
single linked-list for robust mutexes. Both decisions seemed to
follow what recent ports have done and make the resulting
pthread_mutex_t/mtx_t object smaller.
Also the static intialization macro for pthread_mutex_t is set to use
a macro __PTHREAD_MUTEX_INITIALIZER where the architecture can redefine
in its struct_mutex.h if it requires additional fields to be
initialized.
The new rwlock implementation added by cc25c8b4c1196 (2.25) removed
support for lock-elision. This patch removes remaining the
arch-specific unused definitions.
nptl: Add tests for internal pthread_rwlock_t offsets
This patch new build tests to check for internal fields offsets for
internal pthread_rwlock_t definition. Althoug the '__data.__flags'
field layout should be preserved due static initializators, the patch
also adds tests for the futexes that may be used in a shared memory
(although using different libc version in such scenario is not really
supported).
The offsets of pthread_mutex_t __data.__nusers, __data.__spins,
__data.elision, __data.list are not required to be constant over
the releases. Only the __data.__kind is used for static
initializers.
This patch also adds an additional size check for __data.__kind.
Rafał Lużyński [Tue, 1 Oct 2019 20:11:47 +0000 (22:11 +0200)]
ru_UA locale: use copy "ru_RU" in LC_TIME (bug 25044)
Replacing incorrect abbreviated weekday names "Пнд", "Вто", "Срд"...
with correct ones "Пн", "Вт", "Ср"... makes the LC_TIME sections in
those two locales almost identical. The only remaining difference
was that ab_alt_mon elements in ru_UA were lowercase while in ru_RU
they had the first letter uppercase, the latter was pointed as
a better choice by a native speaker. This commit unifies LC_TIME
between ru_RU and ru_UA.
arm: Fix armv7 selection after 'Split BE/LE abilist'
It adds the missing Implies for armv7, armv6, armv6t2 after the
commit 1673ba87fefe019c. Without the Implies a build with the
compiler targeting the aforementioned architecture does not select
the arch-specific optimization including the ifunc selectors.
I checked with a build against armv5, armv6, armv6t2, armv7, and
armv7-neon for both LE and BE. For armv6 and armv7 I also checked
that both sysdeps selection and the resulting implementation built
is the expected ones.
ldbl-128ibm-compat: Add wide character scanning functions
Similarly to what was done for regular character scanning functions,
this patch uses the new mode mask, SCANF_LDBL_USES_FLOAT128, in the
'mode' argument of the wide characters scanning function,
__vfwscanf_internal (which is also extended to support scanning
floating-point values with IEEE binary128, by redirecting calls to
__wcstold_internal to __wcstof128_internal).
Tested for powerpc64le.
Reviewed-By: Paul E. Murphy <murphyp@linux.ibm.com>
ldbl-128ibm-compat: Add regular character scanning functions
The 'mode' argument to __vfscanf_internal allows the selection of the
long double format for all long double arguments requested by the format
string. Currently, there are two possibilities: long double with the
same format as double or long double as something else. The 'something
else' format varies between architectures, and on powerpc64le, it means
IBM Extended Precision format.
In preparation for the third option of long double format on
powerpc64le, this patch uses the new mode mask,
SCANF_LDBL_USES_FLOAT128, which tells __vfscanf_internal to call
__strtof128_internal, instead of __strtold_internal, and save the output
into a _Float128 variable.
Tested for powerpc64le.
Reviewed-By: Paul E. Murphy <murphyp@linux.ibm.com>
The format string can request positional parameters, instead of relying
on the order in which they appear as arguments. Since this has an
effect on how the type of each argument is determined, this patch
extends the test cases to use positional parameters with mixed double
and long double types, to verify that the IEEE long double
implementations of *printf work correctly in this scenario.
Tested for powerpc64le.
Reviewed-By: Paul E. Murphy <murphyp@linux.ibm.com>
A single format string can take double and long double parameters at the
same time. Internally, these parameters are routed to the same
function, which correctly reads them and calls the underlying functions
responsible for the actual conversion to string. This patch adds a new
case to test this scenario.
Tested for powerpc64le.
Reviewed-By: Paul E. Murphy <murphyp@linux.ibm.com>
Similarly to what was done for the regular character, fortified printing
functions, this patch combines the mode masks PRINTF_LDBL_USES_FLOAT128
and PRINTF_FORTIFY to provide wide character versions of fortified
printf functions. It also adds two flavors of test cases: one that
explicitly calls the fortified functions, and another that reuses the
non-fortified test, but defining _FORTIFY_SOURCE as 2. The first
guarantees that the implementations are actually being tested
(independently of what's in bits/wchar2.h), whereas the second
guarantees that the redirections calls the correct function in the IBM
and IEEE long double cases.
Tested for powerpc64le.
Reviewed-By: Paul E. Murphy <murphyp@linux.ibm.com>
Since the introduction of internal functions with explicit flags for the
printf family of functions, the 'mode' parameter can be used to select
which format long double parameters have (with the mode flags:
PRINTF_LDBL_IS_DBL and PRINTF_LDBL_USES_FLOAT128), as well as to select
whether to check for overflows (mode flag: PRINTF_FORTIFY).
This patch combines PRINTF_LDBL_USES_FLOAT128 and PRINTF_FORTIFY to
provide the IEEE binary128 version of printf-like function for platforms
where long double can take this format, in addition to the double format
and to some non-ieee format (currently, this means powerpc64le).
There are two flavors of test cases provided with this patch: one that
explicitly calls the fortified functions, for instance __asprintf_chk,
and another that reuses the non-fortified test, but defining
_FORTIFY_SOURCE as 2. The first guarantees that the implementations are
actually being tested (in bits/stdio2.h, vprintf gets redirected to
__vfprintf_chk, which would leave __vprintf_chk untested), whereas the
second guarantees that the redirections calls the correct function in
the IBM and IEEE long double cases.
Tested for powerpc64le.
Reviewed-By: Paul E. Murphy <murphyp@linux.ibm.com>
ldbl-128ibm-compat: Add wide character printing functions
Similarly to what was done for regular character printing functions,
this patch uses the new mode mask, PRINTF_LDBL_USES_FLOAT128, in the
'mode' argument of the wide characters printing function,
__vfwprintf_internal (which is also extended to support printing
floating-point values with IEEE binary128, by saving floating-point
values into variables of type __float128 and adjusting the parameters to
__printf_fp and __printf_fphex as if it was a call from a wide-character
version of strfromf128 (even though such version does not exist)).
Tested for powerpc64le.
Reviewed-By: Paul E. Murphy <murphyp@linux.ibm.com>
ldbl-128ibm-compat: Add regular character printing functions
The 'mode' argument to __vfprintf_internal allows the selection of the
long double format for all long double arguments requested by the format
string. Currently, there are two possibilities: long double with the
same format as double or long double as something else. The 'something
else' format varies between architectures, and on powerpc64le, it means
IBM Extended Precision format.
In preparation for the third option of long double format on
powerpc64le, this patch uses the new mode mask,
PRINTF_LDBL_USES_FLOAT128, which tells __vfprintf_internal to save the
floating-point values into variables of type __float128 and adjusts the
parameters to __printf_fp and __printf_fphex as if it was a call from
strfromf128.
Many files from the stdio-common, wcsmbs, argp, misc, and libio
directories will have IEEE binary128 counterparts. Setting the correct
compiler options to these files (original and counterparts) would
produce a large amount of repetitive Makefile rules. To avoid this
repetition, this patch adds a Makefile routine that iterates over the
files adding or removing the appropriate flags.
Tested for powerpc64le.
Reviewed-By: Florian Weimer <fweimer@redhat.com> Reviewed-By: Joseph Myers <joseph@codesourcery.com> Reviewed-By: Paul E. Murphy <murphyp@linux.ibm.com>
Use C99-compliant scanf under _GNU_SOURCE with modern compilers.
added the DEPRECATED_SCANF macro to select when redirections of *scanf
functions to their ISO C99 compliant versions should happen, it
accidentally missed doing it for vfwscanf, vwscanf, and vswscanf.
Tested for powerpc64le and with build-many-glibcs (i686-linux-gnu and
nios2-linux-gnu are failing with current master, and with this patch,
but I didn't see a regression).
The generic pselect implementation has the very specific race condition
that motived the creation of the pselect syscall (no atomicity in
signal mask set/reset). Using it as generic implementation is
counterproductive Also currently only microblaze uses it as fallback
when used on kernel prior 3.15.
This patch moves the generic implementation to a microblaze specific
one, sets the generic internal as a ENOSYS, and cleanups the Linux
generic implementation.
The microblaze implementation mimics the previous Linux generic one,
where it either uses pselect6 directly if __ASSUME_PSELECT or a
first try pselect6 then the fallback otherwise.
Checked on x86_64-linux-gnu and microblaze-linux-gnu.
Paul A. Clarke [Thu, 21 Nov 2019 17:57:41 +0000 (11:57 -0600)]
Remove duplicate inline implementation of issignalingf
Very recent commit 854e91bf6b4221f424ffa13b9ef50f35623b7b74 enabled
inline of issignalingf() in general (__issignalingf in include/math.h).
There is another implementation for an inline use of issignalingf
(issignalingf_inline in sysdeps/ieee754/flt-32/math_config.h)
which could instead make use of the new enablement.
Replace the use of issignalingf_inline with __issignaling. Using
issignaling (instead of __issignalingf) will allow future enhancements
to the type-generic implementation, issignaling, to be automatically
adopted.
The implementations are slightly different, and compile to slightly
different code, but I measured no significant performance difference.
The second implementation was brought to my attention by: Suggested-by: Joseph Myers <joseph@codesourcery.com> Reviewed-by: Joseph Myers <joseph@codesourcery.com>
Don't use a custom wrapper macro around __has_include (bug 25189).
This causes issues when using clang with -frewrite-includes to e.g.,
submit the translation unit to a distributed compiler.
In my case, I was building Firefox using sccache.
See [1] for a reduced test-case since I initially thought this was a
clang bug, and [2] for more context.
Apparently doing this is invalid C++ per [cpp.cond], which mentions [3]:
> The #ifdef and #ifndef directives, and the defined conditional
> inclusion operator, shall treat __has_include and __has_cpp_attribute
> as if they were the names of defined macros. The identifiers
> __has_include and __has_cpp_attribute shall not appear in any context
> not mentioned in this subclause.
Florian Weimer [Thu, 31 Oct 2019 12:28:49 +0000 (13:28 +0100)]
Introduce DL_LOOKUP_FOR_RELOCATE flag for _dl_lookup_symbol_x
This will allow changes in dependency processing during non-lazy
binding, for more precise processing of NODELETE objects: During
initial relocation in dlopen, the fate of NODELETE objects is still
unclear, so objects which are depended upon by NODELETE objects
cannot immediately be marked as NODELETE.
rtld: Check __libc_enable_secure before honoring LD_PREFER_MAP_32BIT_EXEC (CVE-2019-19126) [BZ #25204]
The problem was introduced in glibc 2.23, in commit b9eb92ab05204df772eb4929eccd018637c9f3e9
("Add Prefer_MAP_32BIT_EXEC to map executable pages with MAP_32BIT").
Florian Weimer [Thu, 31 Oct 2019 12:23:06 +0000 (13:23 +0100)]
Enhance _dl_catch_exception to allow disabling exception handling
In some cases, it is necessary to introduce noexcept regions
where raised dynamic loader exceptions (e.g., from lazy binding)
are fatal, despite being nested in a code region with an active
exception handler. This change enhances _dl_catch_exception with
to provide such a capability. The existing function is reused,
so that it is not necessary to introduce yet another function with
a similar purpose.
Florian Weimer [Sun, 3 Nov 2019 10:20:23 +0000 (11:20 +0100)]
Avoid zero-length array at the end of struct link_map [BZ #25097]
l_audit ends up as an internal array with _rtld_global, and GCC 10
warns about this.
This commit does not change the layout of _rtld_global, so it is
suitable for backporting. Future changes could allocate more of the
audit state dynamically and remove it from always-allocated data
structures, to optimize the common case of inactive auditing.
Florian Weimer [Sat, 2 Nov 2019 19:04:02 +0000 (20:04 +0100)]
Introduce link_map_audit_state accessor function
To improve GCC 10 compatibility, it is necessary to remove the l_audit
zero-length array from the end of struct link_map. In preparation of
that, this commit introduces an accessor function for the audit state,
so that it is possible to change the representation of the audit state
without adjusting the code that accesses it.
Florian Weimer [Sun, 3 Nov 2019 10:39:56 +0000 (11:39 +0100)]
Redefine _IO_iconv_t to store a single gconv step pointer [BZ #25097]
libio can only deal with gconv conversions which consist of a single
step. Not using __gconv_info simplifies the data structures somewhat.
This eliminates a new GCC 10 warning about subscribing an inner
zero-length array.
Tested on x86_64-linux-gnu with mainline GCC. Built with
build-many-glibcs.py, also with mainline GCC. Due to GCC PR 92039,
there are failures left on 32-bit architectures with float128 support.
Krzysztof Koch [Wed, 13 Nov 2019 11:57:17 +0000 (11:57 +0000)]
Add new script for plotting string benchmark JSON output
Add a script for visualizing the JSON output generated by existing
glibc string microbenchmarks.
Overview:
plot_strings.py is capable of plotting benchmark results in the
following formats, which are controlled with the -p or --plot argument:
1. absolute timings (-p time): plot the timings as they are in the
input benchmark results file.
2. relative timings (-p rel): plot relative timing difference with
respect to a chosen ifunc (controlled with -b argument).
3. performance relative to max (-p max): for each varied parameter
value, plot 1/timing as the percentage of the maximum value out of
the plotted ifuncs.
4. throughput (-p thru): plot varied parameter value over timing
For all types of graphs, there is an option to explicitly specify
the subset of ifuncs to plot using the --ifuncs parameter.
For plot types 1. and 4. one can hide/expose exact benchmark figures
using the --values flag.
When plotting relative timing differences between ifuncs, the first
ifunc listed in the input JSON file is the baseline, unless the
baseline implementation is explicitly chosen with the --baseline
parameter. For the ease of reading, the script marks the statistically
insignificant range on the graphs. The default is +-5% but this
value can be controlled with the --threshold parameter.
To accommodate for the heterogeneity in benchmark results files,
one can control i.e the x-axis scale, the resolution (dpi) of the
generated figures or the key to access the varied parameter value
in the JSON file. The corresponding options are --logarithmic,
--resolution or --key. The --key parameter ensures that plot_strings.py
works with all files which pass JSON schema validation. The schema
can be chosen with the --schema parameter.
If a window manager is available, one can enable interactive
figure display using the --display flag.
Finally, one can use the --grid flag to enable grid lines in the
generated figures.
Implementation:
plot_strings.py traverses the JSON tree until a 'results' array
is found and generates a separate figure for each such array.
The figure is then saved to a file in one of the available formats
(controlled with the --extension parameter).
As the tree is traversed, the recursive function tracks the metadata
about the test being run, so that each figure has a unique and
meaningful title and filename.
While plot_strings.py works with existing benchmarks, provisions
have been made to allow adding more structure and metadata to these
benchmarks. Currently, many benchmarks produce multiple timing values
for the same value of the varied parameter (typically 'length').
Mutiple data points for the same parameter usually mean that some other
parameter was varied as well, for example, if memmove's src and dst
buffers overlap or not (see bench-memmove-walk.c and
bench-memmove-walk.out).
Unfortunately, this information is not exposed in the benchmark output
file, so plot_strings.py has to resort to computing the geometric mean
of these multiple values. In the process, useful information about the
benchmark configuration is lost. Also, averaging the timings for
different alignments can hide useful characterstics of the benchmarked
ifuncs.
Testing:
plot_strings.py has been tested on all existing string microbenchmarks
which produce results in JSON format. The script was tested on both
Windows 10 and Ubuntu 16.04.2 LTS. It runs on both python 2 and 3
(2.7.12 and 3.5.12 tested).
Useful commands:
1. Plot timings for all ifuncs in bench-strlen.out:
$ ./plot_strings.py bench-strlen.out
2. Display help:
$ ./plot_strings.py -h
3. Plot throughput for __memset_avx512_unaligned_erms and
__memset_avx512_unaligned. Save the generated figure in pdf format to
'results/'. Use logarithmic x-axis scale, show grid lines and expose
the performance numbers:
$ ./plot_strings.py bench.out -o results/ -lgv -e pdf -p thru \
-i __memset_avx512_unaligned_erms __memset_avx512_unaligned
4. Plot relative timings for all ifuncs in bench.out with __generic_memset
as baseline. Display percentage difference threshold of +-10%:
$ ./plot_strings.py bench.out -p rel -b __generic_memset -t 10
Discussion:
1. I would like to propose relaxing the benchout_strings.schema.json
to allow specifying either a 'results' array with 'timings' (as before)
or a 'variants' array. See below example:
'variants' array consists of objects such that each object has a 'name'
attribute to describe the configuration of a particular test in the
benchmark. This can be a description, for example, of how the parameter
was varied or what was the buffer alignment tested. The 'name' attribute
is then followed by another 'variants' array or a 'results' array.
The nesting of variants allows arbitrary grouping of benchmark timings,
while allowing description of these groups. Using recusion, it is
possible to proceduraly create titles and filenames for the figures being
generated.
Florian Weimer [Tue, 12 Nov 2019 11:25:49 +0000 (12:25 +0100)]
login: Use pread64 in utmp implementation
This reduces the possible error scenarios considerably because
no longer can file seek fail, leaving the file descriptor in an
inconsistent state and out of sync with the cache.
As a result, it is possible to avoid setting file_offset to -1
to make an error persistent. Instead, subsequent calls will retry
the operation and report any errors returned by the kernel.
This change also avoids reading the file from the start if pututline
is called multiple times, to work around lock acquisition failures
due to timeouts.
Florian Weimer [Thu, 31 Oct 2019 12:28:26 +0000 (13:28 +0100)]
Clarify purpose of assert in _dl_lookup_symbol_x
Only one of the currently defined flags is incompatible with versioned
symbol lookups, so it makes sense to check for that flag and not its
complement.
Reviewed-by: Carlos O'Donell <carlos@redhat.com> Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
Change-Id: I3384349cef90cfd91862ebc34a4053f0c0a99404
Krzysztof Koch [Tue, 5 Nov 2019 17:35:18 +0000 (17:35 +0000)]
aarch64: Increase small and medium cases for __memcpy_generic
Increase the upper bound on medium cases from 96 to 128 bytes.
Now, up to 128 bytes are copied unrolled.
Increase the upper bound on small cases from 16 to 32 bytes so that
copies of 17-32 bytes are not impacted by the larger medium case.
Benchmarking:
The attached figures show relative timing difference with respect
to 'memcpy_generic', which is the existing implementation.
'memcpy_med_128' denotes the the version of memcpy_generic with
only the medium case enlarged. The 'memcpy_med_128_small_32' numbers
are for the version of memcpy_generic submitted in this patch, which
has both medium and small cases enlarged. The figures were generated
using the script from:
https://www.sourceware.org/ml/libc-alpha/2019-10/msg00563.html
Depending on the platform, the performance improvement in the
bench-memcpy-random.c benchmark ranges from 6% to 20% between
the original and final version of memcpy.S
Tested against GLIBC testsuite and randomized tests.
Florian Weimer [Tue, 12 Nov 2019 11:02:57 +0000 (12:02 +0100)]
login: Introduce matches_last_entry to utmp processing
This simplifies internal_getut_nolock and fixes a regression,
introduced in commit be6b16d975683e6cca57852cd4cfe715b2a9d8b1
("login: Acquire write lock early in pututline [BZ #24882]")
in pututxline because __utmp_equal can only compare process-related
utmp entries.
Florian Weimer [Tue, 12 Nov 2019 11:41:34 +0000 (12:41 +0100)]
slotinfo in struct dtv_slotinfo_list should be flexible array [BZ #25097]
GCC 10 will warn about subscribing inner length zero arrays. Use a GCC
extension in csu/libc-tls.c to allocate space for the static_slotinfo
variable. Adjust nptl_db so that the type description machinery does
not attempt to determine the size of the flexible array member slotinfo.
Joseph Myers [Mon, 11 Nov 2019 15:04:48 +0000 (15:04 +0000)]
Declare asctime_r, ctime_r, gmtime_r, localtime_r for C2X.
C2X adds the asctime_r, ctime_r, gmtime_r and localtime_r functions.
This patch duly adds __GLIBC_USE (ISOC2X) to the conditions under
which <time.h> declares them.
Lukasz Majewski [Tue, 29 Oct 2019 14:08:00 +0000 (15:08 +0100)]
y2038: linux: Provide __ppoll64 implementation
This patch provides new __ppoll64 explicit 64 bit function for handling polling
events (with struct timespec specified timeout) for a set of file descriptors.
Moreover, a 32 bit version - __ppoll has been refactored to internally use
__ppoll64.
The __ppoll is now supposed to be used on systems still supporting 32 bit time
(__TIMESIZE != 64) - hence the necessary conversion to 64 bit struct
__timespec64.
The new ppoll_time64 syscall available from Linux 5.1+ has been used, when
applicable.
The Linux kernel checks if passed tv_nsec value overflows, so there is no need
to repeat it in the glibc.
When ppoll syscall on systems supporting 32 bit time ABI is used, the check is
performed if passed data (which may have 64 bit tv_sec) fits into 32 bit range.
Build tests:
- The code has been tested on x86_64/x86 (native compilation):
make PARALLELMFLAGS="-j8" && make check PARALLELMFLAGS="-j8" && \\
make xcheck PARALLELMFLAGS="-j8"
- The glibc has been build tested (make PARALLELMFLAGS="-j8") for
x86 (i386), x86_64-x32, and armv7
Run-time tests:
- Run specific tests on ARM/x86 32bit systems (qemu):
https://github.com/lmajewski/meta-y2038 and run tests:
https://github.com/lmajewski/y2038-tests/commits/master
- Use of cross-test-ssh.sh for ARM (armv7):
make PARALLELMFLAGS="-j8" test-wrapper='./cross-test-ssh.sh root@192.168.7.2' xcheck
Linux kernel, headers and minimal kernel version for glibc build test
matrix:
- Linux v5.1 (with ppoll_time64) and glibc build with v5.1 as
minimal kernel version (--enable-kernel="5.1.0")
The __ASSUME_TIME64_SYSCALLS flag defined.
- Linux v5.1 and default minimal kernel version
The __ASSUME_TIME64_SYSCALLS not defined, but kernel supports ppoll_time64
syscall.
- Linux v4.19 (no ppoll_time64 support) with default minimal kernel version for
contemporary glibc
This kernel doesn't support ppoll_time64 syscall, so the fallback to ppoll is
tested.
Above tests were performed with Y2038 redirection applied as well as without
(so the __TIMESIZE != 64 execution path is checked as well).
Andreas Schwab [Wed, 30 Oct 2019 09:38:36 +0000 (10:38 +0100)]
Fix array bounds violation in regex matcher (bug 25149)
If the regex has more subexpressions than the number of elements allocated
in the regmatch_t array passed to regexec then proceed_next_node may
access the regmatch_t array outside its bounds.
No testcase added because even without this bug it would then crash in
pop_fail_stack which is bug 11053.
Alistair Francis [Fri, 21 Jun 2019 20:00:23 +0000 (13:00 -0700)]
sysdeps/clock_nanosleep: Use clock_nanosleep_time64 if avaliable
The clock_nanosleep syscall is not supported on newer 32-bit platforms (such
as RV32). To fix this issue let's use clock_nanosleep_time64 if it is
avaliable.
Florian Weimer [Thu, 7 Nov 2019 17:15:18 +0000 (18:15 +0100)]
login: Acquire write lock early in pututline [BZ #24882]
It has been reported that due to lack of fairness in POSIX file
locking, the current reader-to-writer lock upgrade can result in
lack of forward progress. Acquiring the write lock directly
hopefully avoids this issue if there are only writers.
This also fixes bug 24882 due to the cache revalidation in
__libc_pututline.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Change-Id: I57e31ae30719e609a53505a0924dda101d46372e
nptl: Add missing placeholder abi symbol from nanosleep move
Adds the __libpthread_version_placeholder symbol with the same version
of nanosleep/__nanosleep that was removed by 79a547b162657b3f and that
is not provided by other symbols.
Florian Weimer [Thu, 7 Nov 2019 08:53:41 +0000 (09:53 +0100)]
login: Remove double-assignment of fl.l_whence in try_file_lock
Since l_whence is the second member of struct flock, it is written
twice. The double-assignment is technically undefined behavior due to
the lack of a sequence point.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Change-Id: I2baf9e70690e723c61051b25ccbd510aec15976c
liqingqing [Thu, 7 Nov 2019 00:26:54 +0000 (00:26 +0000)]
math: enhance the endloop condition of function handle_input_flag
In the function handle_input_flag, the end-loop condition is not
correct, because when the loop variable i equals 16
(num_input_flag_types), then input_flags[16] will be out of bounds.
(This issue is only relevant with invalid input files to
gen-auto-libm-tests.)
The generic version is straightforward. For Hurd, its nanosleep
implementation is moved to clock_nanosleep with adjustments from
generic unix implementation.
The generic clock_nanosleep unix version is also removed since
it calls nanosleep.
Checked on x86_64-linux-gnu and powerpc64le-linux-gnu.
Checked on x86_64-linux-gnu and powerpc64le-linux-gnu. I also checked
the libpthread.so .gnu.version_d entries for every ABI affected and
all of them contains the required versions (including for architectures
which exports __nanosleep with a different version).
Stefan Liebler [Wed, 6 Nov 2019 07:07:40 +0000 (08:07 +0100)]
S390: Fp comparison are now raising FE_INVALID with gcc 10.
The s390 gcc bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77918
"S390: Floating point comparisons don't raise invalid for unordered operands."
is fixed with gcc 10. Thus we conditionally set FIX_COMPARE_INVALID
to 0 or 1.
Arjun Shankar [Tue, 5 Nov 2019 15:41:25 +0000 (16:41 +0100)]
Fix run-one-test so that it runs elf tests
The `test' make target passes a trailing slash in the subdir argument. This
does not play well with elf/rtld-Rules which looks for `elf' without any
trailing slash, and therefore doesn't find a match when running an elf test
individually. This commit removes the trailing slash from the invocation.
Mike Crowe [Mon, 4 Nov 2019 19:36:21 +0000 (16:36 -0300)]
nptl: Fix niggles with pthread_clockjoin_np
Joseph Myers spotted[1] that 69ca4b54c151cec42ccca5e05790efc1a8206b47 added
pthread_clockjoin_np to sysdeps/nptl/pthread.h but not to its hppa-specific
equivalent sysdeps/unix/sysv/linux/hppa/pthread.h.
Rafal Luzynski spotted[2] typos in the NEWS entry and manual updates too.
Florian Weimer spotted[3] that the clockid parameter was not using a
reserved identifier in pthread.h.
hppa: Align __clone stack argument to 8 bytes (Bug 25066)
The hppa architecture requires strict alignment for loads and stores.
As a result, the minimum stack alignment that will work is 8 bytes.
This patch adjusts __clone() to align the stack argument passed to it.
It also adjusts slightly some formatting.
Lukasz Majewski [Thu, 24 Oct 2019 12:34:46 +0000 (14:34 +0200)]
y2038: linux: Provide __futimens64 implementation
This patch provides new __futimens64 explicit 64 bit function for
setting access and modification time of file (by using its file descriptor).
Moreover, a 32 bit version - __futimens has been refactored to internally use
__futimens64.
The __futimens is now supposed to be used on systems still supporting
32 bit time (__TIMESIZE != 64) - hence the necessary conversions to 64 bit
struct __timespec64.
When pointer to struct __timespec64 is NULL - the file access and modification
time is set to the current one (by the kernel) and no conversions from struct
timespec to __timespec64 are performed.
The __futimens64 reuses __utimensat64_helper defined for __utimensat64.
The test procedure for __futimens64 is the same as for __utimensat64 conversion
patch.
Lukasz Majewski [Thu, 24 Oct 2019 10:12:16 +0000 (12:12 +0200)]
y2038: linux: Provide __utimensat64 implementation
This patch provides new __utimensat64 explicit 64 bit function for
setting access and modification time of a file. Moreover, a 32 bit version
- __utimensat has been refactored to internally use __utimensat64.
The __utimensat is now supposed to be used on systems still supporting
32 bit time (__TIMESIZE != 64) - hence the necessary conversions to 64 bit
struct __timespec64.
When pointer to struct __timespec64 is NULL - the file access and modification
time is set to the current one and no conversions from struct timespec to
__timespec64 are performed.
The new utimensat_time64 syscall available from Linux 5.1+ has been used,
when applicable.
The new helper function - __utimensat64_helper - has been introduced to
facilitate code re-usage on function providing futimens syscall handling.
The Linux kernel checks if passed tv_nsec value overflows, so there is no
need to repeat it in glibc.
When utimensat syscall on systems supporting 32 bit time ABI is used,
the check is performed if passed data (which may have 64 bit tv_sec) fits
into 32 bit range.
Build tests:
- The code has been tested on x86_64/x86 (native compilation):
make PARALLELMFLAGS="-j8" && make xcheck PARALLELMFLAGS="-j8"
- The glibc has been build tested (make PARALLELMFLAGS="-j8") for
x86 (i386), x86_64-x32, and armv7
Run-time tests:
- Run specific tests on ARM/x86 32bit systems (qemu):
https://github.com/lmajewski/meta-y2038 and run tests:
https://github.com/lmajewski/y2038-tests/commits/master
- Use of cross-test-ssh.sh for ARM (armv7):
make PARALLELMFLAGS="-j8" test-wrapper='./cross-test-ssh.sh root@192.168.7.2' xcheck
Linux kernel, headers and minimal kernel version for glibc build test
matrix:
- Linux v5.1 (with utimensat_time64) and glibc build with v5.1 as
minimal kernel version (--enable-kernel="5.1.0")
The __ASSUME_TIME64_SYSCALLS flag defined.
- Linux v5.1 and default minimal kernel version
The __ASSUME_TIME64_SYSCALLS not defined, but kernel supports utimensat_time64
syscall.
- Linux v4.19 (no utimensat_time64 support) with default minimal kernel
version for contemporary glibc
This kernel doesn't support utimensat_time64 syscall, so the fallback
to utimensat is tested.
The above tests were performed with Y2038 redirection applied as well as
without (so the __TIMESIZE != 64 execution path is checked as well).
Mike Crowe [Thu, 31 Oct 2019 13:05:17 +0000 (10:05 -0300)]
nptl: Add pthread_timedjoin_np, pthread_clockjoin_np NULL timeout test
Passing NULL as the timeout parameter to pthread_timedjoin_np has resulted
in it behaving like pthread_join for a long time. Since that is now the
documented behaviour, we ought to test that both it and the new
pthread_clockjoin_np support it.
Mike Crowe [Thu, 31 Oct 2019 13:03:21 +0000 (10:03 -0300)]
nptl: Add pthread_clockjoin_np
Introduce pthread_clockjoin_np as a version of pthread_timedjoin_np that
accepts a clockid_t parameter to indicate which clock the timeout should be
measured against. This mirrors the recently-added POSIX-proposed "clock"
wait functions.
This patch adds the generic futex_lock_pi and futex_unlock_pi to wrap
around the syscall machinery required to issue the syscall calls. It
simplifies a bit the futex code required to implement PI mutexes.
nptl: Replace non cancellable pause/nanosleep with futex
To help y2038 work avoid duplicate all the logic of nanosleep on
non cancellable version, the patch replace it with a new futex
operation, lll_timedwait. The changes are:
- Add a expected value for __lll_clocklock_wait, so it can be used
to wait for generic values.
- Remove its internal atomic operation and move the logic to
__lll_clocklock. It makes __lll_clocklock_wait even more generic
and __lll_clocklock slight faster on fast-path (since it won't
require a function call anymore).
- Add lll_timedwait, which uses __lll_clocklock_wait, to replace both
__pause_nocancel and __nanosleep_nocancel.
It also allows remove the sparc32 __lll_clocklock_wait implementation
(since it is similar to the generic one).
Checked on x86_64-linux-gnu, sparcv9-linux-gnu, and i686-linux-gnu.
NPTL is already Linux specific, there is no need to parametrize low
level lock futex operations and add a sysdep Linux specific
implementation. This patch moves the relevant Linux code to nptl one.
NPTL is already Linux specific, there is no need to parametrize futex
operations and add a sysdep Linux specific implementation. This patch
moves the relevant Linux code to nptl one.
DJ Delorie [Wed, 30 Oct 2019 22:03:14 +0000 (18:03 -0400)]
Base max_fast on alignment, not width, of bins (Bug 24903)
set_max_fast sets the "impossibly small" value based on,
eventually, MALLOC_ALIGNMENT. The comparisons for the smallest
chunk used is, eventually, MIN_CHUNK_SIZE. Note that i386
is the only platform where these are the same, so a smallest
chunk *would* be put in a no-fastbins fastbin.
This change calculates the "impossibly small" value
based on MIN_CHUNK_SIZE instead, so that we can know it will
always be impossibly small.
Zack Weinberg [Mon, 19 Aug 2019 14:31:07 +0000 (10:31 -0400)]
Revise the documentation of simple calendar time.
This is a thorough revision of all the material relating to the
functions time, stime, gettimeofday, settimeofday, clock_gettime,
clock_getres, clock_settime, and difftime, spilling over into the
discussion of time-related data types (which now get their own
section) and touching the adjtime family as well (which deserves its
own thorough revision, but I'd have to do a bunch of research first).
Substantive changes are:
* Document clock_gettime, clock_getres, and clock_settime. (Only
CLOCK_REALTIME and CLOCK_MONOTONIC are documented; the others are
either a bit too Linux-specific, or have more to do with measuring
CPU/processor time. That section _also_ deserves its own thorough
revision but again I'd have to do a bunch of research first.)
* Present gettimeofday, settimeofday, and struct timeval as obsolete
relative to clock_*.
* Remove the documentation of struct timezone. Matching POSIX,
say that the type of the second argument to gettimeofday and
settimeofday is [const] void *.
* Clarify ISO C and POSIX's requirements on time_t. Clarify the
circumstances under which difftime is equivalent to simple
subtraction.
* Consolidate documentation of most of the time-related data types
into a new section "Time Types," right after "Time Basics." (The
exceptions are struct tm, which stays in "Broken-down Time," and
struct times, which stays in "Processor And CPU Time."
* The "Elapsed Time" section is now called "Calculating Elapsed Time"
and includes only difftime and the discussion of how to compute
timeval differences by hand.
* Fold the "Simple Calendar Time," "High Resolution Calendar," and
"High Accuracy Clock" sections together into two new sections titled
"Getting the Time" and "Setting and Adjusting the Time."
Consolidate generic gettimeofday implementation to use clock_gettime.
Linux ports that still provide gettimeofday through vDSO are not
changed.
Remove sysdeps/unix/clock_gettime.c, which implemented clock_gettime
using gettimeofday; new OS ports must provide a real implementation of
clock_gettime.
Rename sysdeps/mach/gettimeofday.c to sysdeps/mach/clock_gettime.c and
convert into an implementation of clock_gettime. It only supports
CLOCK_REALTIME; Mach does not appear to have any support for monotonic
clocks. It uses __host_get_time, which provides at best microsecond
resolution. Hurd is currently using sysdeps/posix/clock_getres.c for
clock_getres; its output for CLOCK_REALTIME is based on
sysconf (_SC_CLK_TCK), and I do not know whether that gives the
correct result.
Unlike settimeofday, there are no known uses of gettimeofday's
vestigial "get time zone" feature that are not bugs. (The per-process
timezone support in localtime and friends is unrelated, and the
programs that set the kernel's offset between the hardware clock and
UTC do not need to read it back.) Therefore, this feature is dummied
out. Henceforth, if gettimeofday's "struct timezone" argument is not
NULL, it will write zeroes to both fields. Any program that is
actually looking at this data will thus think it is running in UTC,
which is probably more correct than whatever it was doing before.
[__]gettimeofday no longer has any internal callers, so we can now
remove its internal prototype and PLT bypass aliases. The
__gettimeofday@GLIBC_2.0 export remains, in case it is used by any
third-party code.
It also allows to simplify the arch-specific implementation on x86 and
powerpc to remove the hack to disable the internal route to non iFUNC
variant for internal symbol.
This patch also fixes a missing optimization on aarch64, powerpc, and
x86 where the code used on static build do not use the vDSO.
Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
powerpc64-linux-gnu, powerpc-linux-gnu, and aarch64-linux-gnu.
Zack Weinberg [Thu, 22 Aug 2019 21:10:35 +0000 (17:10 -0400)]
Use clock_gettime to implement timespec_get.
timespec_get is the same function as clock_gettime, with an obnoxious
coating of NIH painted on it by the ISO C committee. In addition to
the rename, it takes its arguments in a different order, it returns 0
on *failure* or a positive number on *success*, and it requires that
all of its TIME_* constants be positive. This last means we cannot
directly reuse the existing CLOCK_* constants for it, because
those have been allocated starting with CLOCK_REALTIME = 0 on all
existing platforms.
This patch simply promotes the sysdeps/posix implementation to
universal, and removes the Linux-specific implementation, whose
apparent reason for existing was to cut out one function call's worth
of overhead.
Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
powerpc64-linux-gnu, powerpc-linux-gnu, and aarch64-linux-gnu.
ftime is an obsolete variation on gettimeofday, offering only
millisecond time resolution; it was probably a system call in ooold
versions of BSD Unix. For historic reasons, we had three
implementations of it. These are all consolidated into time/ftime.c,
and then the function is deprecated.
For some reason, the implementation of ftime in terms of gettimeofday
was rounding rather than truncating microseconds to milliseconds. In
all the other places where we use a higher-resolution time function to
implement a lower-resolution one, we truncate. ftime is changed to
match, just for tidiness' sake.
Like gettimeofday, ftime tries to report the time zone, and using that
information is always a bug. This patch dummies out the reported
timezone information; the timezone and dstflag fields of the
returned "struct timeb" will always be zero.
Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
powerpc64-linux-gnu, and powerpc-linux-gnu.
Change most internal uses of time to __clock_gettime.
As for gettimeofday, time will be implemented based on clock_gettime
on all platforms and internal code should use clock_gettime
directly. In addition to removing a layer of indirection, this will
allow us to remove the PLT-bypass gunk for gettimeofday.
The changed code always assumes __clock_gettime (CLOCK_REALTIME)
or __clock_gettime (CLOCK_REALTIME_COARSE) (for Linux case) cannot
fail, using the same rationale for gettimeofday change. And internal
helper was added (time_now).
Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
powerpc64-linux-gnu, and powerpc-linux-gnu.
Change the default implementation of time to call clock_gettime,
to align with new Linux ports that are expected to only implement
__NR_clock_gettime. Arch-specific implementation that either call
the time vDSO or route to gettimeofday vDSO are not removed.
Also for Linux, CLOCK_REALTIME_COARSE is used instead of generic
CLOCK_REALTIME clockid. This takes less CPU time and its behavior
better matches what the current glibc does.
Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
powerpc64-linux-gnu, powerpc-linux-gnu, and aarch64-linux-gnu.
Zack Weinberg [Wed, 28 Aug 2019 12:25:49 +0000 (08:25 -0400)]
Use clock_settime to implement settimeofday.
Unconditionally, on all ports, use clock_settime to implement
settimeofday. Remove sysdeps/unix/clock_settime.c, which implemented
clock_settime by calling settimeofday; new OS ports must henceforth
provide a real implementation of clock_settime.
Hurd had a real implementation of settimeofday but not of
clock_settime; this patch converts it into an implementation of
clock_settime. It only supports CLOCK_REALTIME and microsecond
resolution; Hurd/Mach does not appear to have any support for
finer-resolution clocks.
The vestigial "set time zone" feature of settimeofday complicates the
generic settimeofday implementation a little. The only remaining uses
of this feature that aren't just bugs, are using it to inform the
Linux kernel of the offset between the hardware clock and UTC, on
systems where the hardware clock doesn't run in UTC (usually because
of dual-booting with Windows). There currently isn't any other way to
do this. However, the callers that do this call settimeofday with
_only_ the timezone argument non-NULL. Therefore, glibc's new
behavior is: callers of settimeofday must supply one and only one of
the two arguments. If both arguments are non-NULL, or both arguments
are NULL, the call fails and sets errno to EINVAL.
When only the timeval argument is supplied, settimeofday calls
__clock_settime(CLOCK_REALTIME), same as stime.
When only the timezone argument is supplied, settimeofday calls a new
internal function called __settimezone. On Linux, only, this function
will pass the timezone structure to the settimeofday system call. On
all other operating systems, and on Linux architectures that don't
define __NR_settimeofday, __settimezone is a stub that always sets
errno to ENOSYS and returns -1.
The settimeoday syscall is enabled on Linux by the flag
COMPAT_32BIT_TIME, which is an option to either 32-bits ABIs or COMPAT
builds (defined usually by 64-bit kernels that want to support 32-bit
ABIs, such as x86). The idea to future 64-bit time_t only ABIs
is to not provide settimeofday syscall.
The same semantics are implemented for Linux/Alpha's GLIBC_2.0 compat
symbol for settimeofday.
There are no longer any internal callers of __settimeofday, so the
internal prototype is removed.
Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
powerpc64-linux-gnu, powerpc-linux-gnu, and aarch64-linux-gnu.
Zack Weinberg [Fri, 16 Aug 2019 19:03:16 +0000 (15:03 -0400)]
Use clock_settime to implement stime; withdraw stime.
Unconditionally, on all ports, use clock_settime to implement stime,
not settimeofday or a direct syscall. Then convert stime into a
compatibility symbol and remove its prototype from time.h.
Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
powerpc64-linux-gnu, powerpc-linux-gnu, and aarch64-linux-gnu.
Zack Weinberg [Sat, 17 Aug 2019 00:38:22 +0000 (20:38 -0400)]
Change most internal uses of __gettimeofday to __clock_gettime.
Since gettimeofday will shortly be implemented in terms of
clock_gettime on all platforms, internal code should use clock_gettime
directly; in addition to removing a layer of indirection, this will
allow us to remove the PLT-bypass gunk for gettimeofday. (We can't
quite do that yet, but it'll be coming later in this patch series.)
In many cases, the changed code does fewer conversions.
The changed code always assumes __clock_gettime (CLOCK_REALTIME)
cannot fail. Most of the call sites were assuming gettimeofday could
not fail, but a few places were checking for errors. POSIX says
clock_gettime can only fail if the clock constant is invalid or
unsupported, and CLOCK_REALTIME is the one and only clock constant
that's required to be supported. For consistency I grepped the entire
source tree for any other places that checked for errors from
__clock_gettime (CLOCK_REALTIME), found one, and changed it too.
(For the record, POSIX also says gettimeofday can never fail.)
(It would be nice if we could declare that GNU systems will always
support CLOCK_MONOTONIC as well as CLOCK_REALTIME; there are several
places where we are using CLOCK_REALTIME where _MONOTONIC would be
more appropriate, and/or trying to use _MONOTONIC and then falling
back to _REALTIME. But the Hurd doesn't support CLOCK_MONOTONIC yet,
and it looks like adding it would involve substantial changes to
gnumach's internals and API. Oh well.)
A few Hurd-specific files were changed to use __host_get_time instead
of __clock_gettime, as this seemed tidier. We also assume this cannot
fail. Skimming the code in gnumach leads me to believe the only way
it could fail is if __mach_host_self also failed, and our
Hurd-specific code consistently assumes that can't happen, so I'm
going with that.
With the exception of support/support_test_main.c, test cases are not
modified, mainly because I didn't want to have to figure out which
test cases were testing gettimeofday specifically.
The definition of GETTIME in sysdeps/generic/memusage.h had a typo and
was not reading tv_sec at all. I fixed this. It appears nobody has been
generating malloc traces on a machine that doesn't have a superseding
definition.
There are a whole bunch of places where the code could be simplified
by factoring out timespec subtraction and/or comparison logic, but I
want to keep this patch as mechanical as possible.
Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
powerpc64-linux-gnu, powerpc-linux-gnu, and aarch64-linux-gnu.
Zack Weinberg [Mon, 19 Aug 2019 18:18:08 +0000 (14:18 -0400)]
Linux/Alpha: don't use timeval32 system calls.
Linux/Alpha has two versions of several system call wrappers that take
or return data of type "struct timeval" (possibly nested inside a
larger structure). The GLIBC_2.0 version is a compat symbol that
calls __NR_osf_foo or __NR_old_foo and uses a struct timeval with a
32-bit tv_sec field. The GLIBC_2.1 version is used for current code,
calls __NR_foo, and uses a struct timeval with a 64-bit tv_sec field.
This patch changes all of the compat symbols of this type to be
wrappers around their GLIBC_2.1 counterparts; the compatibility system
calls will no longer be used. It serves as a proposal for part of how
we do the transition to 64-bit time_t on systems that currently use
32-bit time_t:
* The patched glibc will NOT use system calls that involve 32-bit
time_t to implement its compatibility symbols. This will make both
our lives and the kernel maintainers' lives easier. The primary
argument I've seen against it is that the kernel could warn about
uses of the old system calls, helping people find old binaries that
need to be recompiled. I think there are several other ways we
could accomplish this, e.g. scripts to scan the filesystem for
binaries with references to the old symbol versions, or issuing
diagnostics ourselves.
* The compat symbols do NOT report failure after the Y2038 deadline.
An earlier revision of this patch had them return -1 and set errno
to EOVERFLOW, but Adhemerval pointed out that many of them have
already performed side effects at the point where we discover the
overflow, so that would break more than it fixes. Also, we don't
want people to be _checking_ for EOVERFLOW from these functions; we
want them to recompile with 64-bit time_t. So it's not actually
useful for them to report failure to the calling code.
* What they do do, when they encounter overflow, is saturate the
overflowed "struct timeval"(s): tv_sec is set to INT32_MAX and
tv_nsec is set to 999999. That means time stops advancing for
programs with 32-bit time_t when they reach the deadline. That's
obviously going to break stuff, but I think wrapping around is
probably going to break _more_ stuff. I'd be interested to hear
arguments against, if anyone has one.
The new header file tv32-compat.h is currently Alpha-specific but I
mean for it to be reused to aid in writing wrappers for all affected
architectures. I only put it in sysdeps/unix/sysv/linux/alpha for now
because I haven't checked whether the various "foo32" structures it
defines agree with the ABI for ports other than Linux/Alpha.