Florian Weimer [Fri, 25 Oct 2024 14:50:10 +0000 (16:50 +0200)]
elf: Run constructors on cyclic recursive dlopen (bug 31986)
This is conceptually similar to the reported bug, but does not
depend on auditing. The fix is simple: just complete execution
of the constructors. This exposed the fact that the link map
for statically linked executables does not have l_init_called
set, even though constructors have run.
Joseph Myers [Wed, 23 Oct 2024 16:45:15 +0000 (16:45 +0000)]
Add more tests of pthread_mutexattr_gettype and pthread_mutexattr_settype
Add basic tests of pthread_mutexattr_gettype and
pthread_mutexattr_settype with each valid mutex kind, plus test for
EINVAL with an invalid mutex kind.
Arjun Shankar [Fri, 18 Oct 2024 14:03:25 +0000 (16:03 +0200)]
libio: Fix a deadlock after fork in popen
popen modifies its file handler book-keeping under a lock that wasn't
being taken during fork. This meant that a concurrent popen and fork
could end up copying the lock in a "locked" state into the fork child,
where subsequently calling popen would lead to a deadlock due to the
already (spuriously) held lock.
This commit fixes the deadlock by appropriately taking the lock before
fork, and releasing/resetting it in the parent/child after the fork.
A new test for concurrent popen and fork is also added. It consistently
hangs (and therefore fails via timeout) without the fix applied. Reviewed-by: Florian Weimer <fweimer@redhat.com>
DJ Delorie [Thu, 10 Oct 2024 21:16:35 +0000 (17:16 -0400)]
manual: Document stdio.h functions that may be macros
Glibc has two gnu-extension functions that are implemented as
macros but not documented as such: fread_unlocked and
fwrite_unlocked. Document them as such.
Additionally, putc_unlocked and getc_unlocked are documented in
POSIX as possibly being macros. Update the manual to add a warning
about those also, depite glibc not implementing them as macros.
Joseph Myers [Mon, 21 Oct 2024 20:56:48 +0000 (20:56 +0000)]
Check time arguments to pthread_timedjoin_np and pthread_clockjoin_np
The pthread_timedjoin_np and pthread_clockjoin_np functions do not
check that a valid time has been specified. The documentation for
these functions in the glibc manual isn't sufficiently detailed to say
if they should, but consistency with POSIX functions such as
pthread_mutex_timedlock and pthread_cond_timedwait strongly indicates
that an EINVAL error is appropriate (even if there might be some
ambiguity about exactly where such a check should go in relation to
other checks for whether the thread exists, whether it's immediately
joinable, etc.). Copy the logic for such a check used in
pthread_rwlock_common.c.
pthread_join_common had some logic calling valid_nanoseconds before
commit 9e92278ffad441daf588ff1ff5bd8094aa33fbfd, "nptl: Remove
clockwait_tid"; I haven't checked exactly what cases that detected.
Jonathan Wakely [Thu, 17 Oct 2024 15:49:45 +0000 (16:49 +0100)]
Add .b4-config file
This makes b4 use inbox.sourceware.org instead of the default host
lore.kernel.org, so that every b4 user doesn't have to configure this
themselves for the glibc repo.
The commit 'sparc: Use Linux kABI for syscall return'
(86c5d2cf0ce046279baddc7faa27da71f1a89fde) did not take into account
a subtle sparc syscall kABI constraint. For syscalls that might block
indefinitely, on an interrupt (like SIGCONT) the kernel will set the
instruction pointer to just before the syscall:
arch/sparc/kernel/signal_64.c
476 static void do_signal(struct pt_regs *regs, unsigned long orig_i0)
477 {
[...]
525 if (restart_syscall) {
526 switch (regs->u_regs[UREG_I0]) {
527 case ERESTARTNOHAND:
528 case ERESTARTSYS:
529 case ERESTARTNOINTR:
530 /* replay the system call when we are done */
531 regs->u_regs[UREG_I0] = orig_i0;
532 regs->tpc -= 4;
533 regs->tnpc -= 4;
534 pt_regs_clear_syscall(regs);
535 fallthrough;
536 case ERESTART_RESTARTBLOCK:
537 regs->u_regs[UREG_G1] = __NR_restart_syscall;
538 regs->tpc -= 4;
539 regs->tnpc -= 4;
540 pt_regs_clear_syscall(regs);
541 }
However, on a SIGCONT it seems that 'g1' register is being clobbered after the
syscall returns. Before 86c5d2cf0ce046279, the 'g1' was always placed jus
before the 'ta' instruction which then reloads the syscall number and restarts
the syscall.
On master, where 'g1' might be placed before 'ta':
$ cat test.c
#include <unistd.h>
int main ()
{
pause ();
}
$ gcc test.c -o test
$ strace -f ./t
[...]
ppoll(NULL, 0, NULL, NULL, 0
Just moving the 'g1' setting near the syscall asm is not suffice,
the compiler might optimize it away (as I saw on cancellation.c by
trying this fix). Instead, I have change the inline asm to put the
'g1' setup in ithe asm block. This would require to change the asm
constraint for INTERNAL_SYSCALL_NCS, since the syscall number is not
constant.
Checked on sparc64-linux-gnu.
Reported-by: René Rebe <rene@exactcode.de> Tested-by: Sam James <sam@gentoo.org> Reviewed-by: Sam James <sam@gentoo.org>
Carlos O'Donell [Wed, 9 Oct 2024 22:32:26 +0000 (18:32 -0400)]
manual: Fix and test @deftypef* function formatting
The manual contained several instances of incorrect formatting
that were correct texinfo but produced incorrectly rendered manuals
or incorrect behaviour from the tooling.
The most important was incorrect quoting of function returns
by failing to use {} to quote the return. The impact of this
mistake means that 'info libc func' does not jump to the function
in question but instead to the introductory page under the assumption
that func doesn't exist. The function returns are now correctly
quoted.
The second issue was the use of a category specifier with
@deftypefun which doesn't accept a category specifier. If a category
specifier is required then @deftypefn needs to be used. This is
corrected by changing the command to @deftypefn for such functions
that used {Deprecated function} as a category.
The last issue is a missing space between the function name and the
arguments which results in odd function names like "epoll_wait(int"
instead of "epoll_wait". This also impacts the use of 'info libc'
and is corrected.
We additionally remove ';' from the end of function arguments and
add an 'int' return type for dprintf.
Lastly we add a new test check-deftype.sh which verifies the expected
formatting of @deftypefun, @deftypefunx, @deftypefn, and
@deftypefnx. The new test is also run as the summary file is
generated to ensure we don't generate incorrect results.
The existing check-safety.sh is also run directly as a test to increase
coverage since the existing tests only ran on manual install.
The new tests now run as part of the standard "make check" that
pre-commit CI runs and developers should run.
No regressions on x86_64.
HTML and PDF rendering reviewed and looks correct for all changes.
Paul Zimmermann [Tue, 8 Oct 2024 09:28:04 +0000 (11:28 +0200)]
replace tgammaf by the CORE-MATH implementation
The CORE-MATH implementation is correctly rounded (for any rounding mode).
This can be checked by exhaustive tests in a few minutes since there are
less than 2^32 values to check against for example GNU MPFR.
This patch also adds some bench values for tgammaf.
Tested on x86_64 and x86 (cfarm26).
With the initial GNU libc code it gave on an Intel(R) Core(TM) i7-8700:
Signed-off-by: Alexei Sibidanov <sibid@uvic.ca> Signed-off-by: Paul Zimmermann <Paul.Zimmermann@inria.fr>
Changes in v2:
- include <math.h> (fix the linknamespace failures)
- restored original benchtests/strcoll-inputs/filelist#en_US.UTF-8 file
- restored original wrapper code (math/w_tgammaf_compat.c),
except for the dealing with the sign
- removed the tgammaf/float entries in all libm-test-ulps files
- address other comments from Joseph Myers
(https://sourceware.org/pipermail/libc-alpha/2024-July/158736.html)
Changes in v3:
- pass NULL argument for signgam from w_tgammaf_compat.c
- use of math_narrow_eval
- added more comments
Changes in v4:
- initialize local_signgam to 0 in math/w_tgamma_template.c
- replace sysdeps/ieee754/dbl-64/gamma_productf.c by dummy file
Changes in v5:
- do not mention local_signgam any more in math/w_tgammaf_compat.c
- initialize local_signgam to 1 instead of 0 in w_tgamma_template.c
and added comment
Changes in v6:
- pass NULL as 2nd argument of __ieee754_gammaf_r in
w_tgammaf_compat.c, and check for NULL in e_gammaf_r.c
Changes in v7:
- added Signed-off-by line for Alexei Sibidanov (author of the code)
Changes in v8:
- added Signed-off-by line for Paul Zimmermann (submitted of the patch)
Changes in v9:
- address comments from review by Adhemerval Zanella Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
This request the page to be never written out to swap, it will be zeroed
under memory pressure (so kernel can just drop the page), it is inherited
by fork, it is not counted against @code{mlock} budget, and if there is
no enough memory to service a page faults there is no fatal error (so not
signal is sent).
* fstat/newfstatat for loongarch (it should be safe to add since 255dc1e4ed8 that undefine them).
* clone3 for nios2, which only adds the entry point but defined
__ARCH_BROKEN_SYS_CLONE3 (the syscall will always return ENOSYS).
* uretprobe for x86_64 and x32.
Update syscall-names.list and regenerate the arch-syscall.h headers
with build-many-glibcs.py update-syscalls.
The recursive lock used on abort does not synchronize with a new process
creation (either by fork-like interfaces or posix_spawn ones), nor it
is reinitialized after fork().
Also, the SIGABRT unblock before raise() shows another race condition,
where a fork or posix_spawn() call by another thread, just after the
recursive lock release and before the SIGABRT signal, might create
programs with a non-expected signal mask. With the default option
(without POSIX_SPAWN_SETSIGDEF), the process can see SIG_DFL for
SIGABRT, where it should be SIG_IGN.
To fix the AS-safe, raise() does not change the process signal mask,
and an AS-safe lock is used if a SIGABRT is installed or the process
is blocked or ignored. With the signal mask change removal,
there is no need to use a recursive loc. The lock is also taken on
both _Fork() and posix_spawn(), to avoid the spawn process to see the
abort handler as SIG_DFL.
A read-write lock is used to avoid serialize _Fork and posix_spawn
execution. Both sigaction (SIGABRT) and abort() requires to lock
as writer (since both change the disposition).
The fallback is also simplified: there is no need to use a loop of
ABORT_INSTRUCTION after _exit() (if the syscall does not terminate the
process, the system is broken).
The proposed fix changes how setjmp works on a SIGABRT handler, where
glibc does not save the signal mask. So usage like the below will now
always abort.
static volatile int chk_fail_ok;
static jmp_buf chk_fail_buf;
static void
handler (int sig)
{
if (chk_fail_ok)
{
chk_fail_ok = 0;
longjmp (chk_fail_buf, 1);
}
else
_exit (127);
}
[...]
signal (SIGABRT, handler);
[....]
chk_fail_ok = 1;
if (! setjmp (chk_fail_buf))
{
// Something that can calls abort, like a failed fortify function.
chk_fail_ok = 0;
printf ("FAIL\n");
}
Such cases will need to use sigsetjmp instead.
The _dl_start_profile calls sigaction through _profil, and to avoid
pulling abort() on loader the call is replaced with __libc_sigaction.
Checked on x86_64-linux-gnu and aarch64-linux-gnu.
The BZ#24967 fix (1bdda52fe92fd01b424c) missed the time for
architectures that define USE_IFUNC_TIME. Although it is not
an issue, since there is no pointer mangling, there is also no need
to call dl_vdso_vsym since the vDSO setup was already done by the
loader.
linux: Use GLRO(dl_vdso_gettimeofday) on gettimeofday
The BZ#24967 fix (1bdda52fe92fd01b424c) missed the gettimeofday for
architectures that define USE_IFUNC_GETTIMEOFDAY. Although it is not
an issue, since there is no pointer mangling, there is also no need
to call dl_vdso_vsym since the vDSO setup was already done by the
loader.
Stefan Liebler [Fri, 27 Sep 2024 08:26:29 +0000 (10:26 +0200)]
S390: Don't use r11 for cu-instructions as used as frame-pointer. [BZ# 32192]
Building the s390 specific iconv modules - utf16-utf32-z9.c, utf8-utf32-z9.c
and utf8-utf16-z9.c - with -fno-omit-frame-pointer leads to a build error
"error: %r11 cannot be used in 'asm' here" as r11 is needed as frame-pointer.
The cuXY-instructions need two even-odd register pairs. Therefore the register
pinning is used. This patch just uses a different register pair. Reviewed-by: Florian Weimer <fweimer@redhat.com>
Carlos O'Donell [Mon, 7 Oct 2024 22:04:22 +0000 (18:04 -0400)]
Fix whitespace related license issues.
Several copies of the licenses in files contained whitespace related
problems. Two cases are addressed here, the first is two spaces
after a period which appears between "PURPOSE." and "See". The other
is a space after the last forward slash in the URL. Both issues are
corrected and the licenses now match the official textual description
of the license (and the other license in the sources).
Since these whitespaces changes do not alter the paragraph structure of
the license, nor create new sentences, they do not change the license.
Add tests of freopen adding or removing "c" (non-cancelling I/O) from
the mode string (so completing my planned tests of freopen with
different features used in the mode strings). Note that it's in the
nature of the uncertain time at which cancellation might act (possibly
during freopen, possibly during subsequent reads) that these can leak
memory or file descriptors, so these do not include leak tests.
linux: sparc: Fix clone for LEON/sparcv8 (BZ 31394)
The sparc clone mitigation (faeaa3bc9f76030) added the use of
flushw, which is not support by LEON/sparcv8. As discussed on
the libc-alpha, 'ta 3' is a working alternative [1].
GCC aligns global data to 16 bytes if their size is >= 16 bytes. This patch
changes the exp2f_data struct slightly so that the fields are better aligned.
As a result on targets that support them, load-pair instructions accessing
poly_scaled and invln2_scaled are now 16-byte aligned.
Even though building glibc with 64 bit time_t flags is not supported,
and the usual way is to patch the build system to avoid it; some
systems do enable it by default, and it increases the requirements
to build glibc in such cases (it also does not help newcomers when
trying to build glibc).
The conform namespace and linknamespace tests also do not expect
that flag to be set by default, so disable it as well.
Checked with a build/check for major ABI and some (i386, arm,
mipsel, hppa) with a toolchain that has LFS flags by default. Reviewed-by: DJ Delorie <dj@redhat.com>
Disable _FILE_OFFSET_BITS if the compiler defaults to it
Even though building glibc with LFS flags is not supported, and the
the usual way is to patch the build system to avoid it [1]; some system
do enable it by default, and it increases the requirements to build
glibc in such cases (it also does not help newcomers when trying
to build glibc).
The conform namespace and linknamespace tests also do not expect
that flag to be set by default, so disable it as well.
Checked with a build/check for major ABI and some (i386, arm,
mipsel, hppa) with a toolchain that has LFS flags by default.
[1] https://sourceware.org/bugzilla/show_bug.cgi?id=31624 Reviewed-by: DJ Delorie <dj@redhat.com>
The -Wp does not work properly if the compiler is configured to enable
fortify by default, since it bypasses the compiler driver (which defines
the fortify flags in this case).
This patch is similar to the one used on Ubuntu [1].
I checked with a build for x86_64-linux-gnu, i686-linux-gnu,
aarch64-linux-gnu, s390x-linux-gnu, and riscv64-linux-gnu with
gcc-13 that enables the fortify by default.
Co-authored-by: Matthias Klose <matthias.klose@canonical.com>
[1] https://git.launchpad.net/~ubuntu-core-dev/ubuntu/+source/glibc/tree/debian/patches/ubuntu/fix-fortify-source.patch Reviewed-by: DJ Delorie <dj@redhat.com>
H.J. Lu [Sat, 7 Sep 2024 15:32:32 +0000 (08:32 -0700)]
libio: Set _vtable_offset before calling _IO_link_in [BZ #32148]
Since _IO_vtable_offset is used to detect the old binaries, set it
in _IO_old_file_init_internal before calling _IO_link_in which checks
_IO_vtable_offset. Add a glibc 2.0 test with copy relocation on
_IO_stderr_@GLIBC_2.0 to verify that fopen won't cause memory corruption.
This fixes BZ #32148.
Signed-off-by: H.J. Lu <hjl.tools@gmail.com> Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
Add a new fwrite test that exercises buffer overflow
Exercises fwrite's internal buffer when doing a file operation.
The new test, exercises 2 overflow behaviors:
1. Call fwrite multiple times making usage of fwrite's internal buffer.
The total number of bytes written is larger than fwrite's internal
buffer, forcing an automatic flush.
2. Call fwrite a single time with an amount of data that is larger than
fwrite's internal buffer.
x86/string: Fixup alignment of main loop in str{n}cmp-evex [BZ #32212]
The loop should be aligned to 32-bytes so that it can ideally run out
the DSB. This is particularly important on Skylake-Server where
deficiencies in it's DSB implementation make it prone to not being
able to run loops out of the DSB.
This results in a 25% performance degradation for the non-aligned
version.
The fix is to just ensure the code layout is such that the loop is
aligned. (Which was previously the case but was accidentally dropped
in 84e7c46df).
NB: The fix was actually 64-byte alignment. This is because 64-byte
alignment generally produces more stable performance than 32-byte
aligned code (cache line crosses can affect perf), so if we are going
past 16-byte alignmnent, might as well go to 64. 64-byte alignment
also matches most other functions we over-align, so it creates a
common point of optimization.
Times are reported as ratio of Time_With_Patch /
Time_Without_Patch. Lower is better.
The values being reported is the geometric mean of the ratio across
all tests in bench-strcmp and bench-strncmp.
Note this patch is only attempting to improve the Skylake-Server
strcmp for long strings. The rest of the numbers are only to test for
regressions.
The 2.6% regression on TGL-strcmp is due to slowdowns caused by
changes in alignment of code handling small sizes (most on the
page-cross logic). These should be safe to ignore because 1) We
previously only 16-byte aligned the function so this behavior is not
new and was essentially up to chance before this patch and 2) this
type of alignment related regression on small sizes really only comes
up in tight micro-benchmark loops and is unlikely to have any affect
on realworld performance.
Mike FABIAN [Thu, 12 Sep 2024 13:02:55 +0000 (15:02 +0200)]
Update to Unicode 16.0.0 [BZ #32168]
Unicode 16.0.0 Support: Character encoding, character type info, and
transliteration tables are all updated to Unicode 16.0.0, using
the generator scripts contributed by Mike FABIAN (Red Hat).
Changes in CHARMAP and WIDTH:
Total added characters in newly generated CHARMAP: 5185
Total removed characters in newly generated WIDTH: 1
Total added characters in newly generated WIDTH: 170
The removed character from WIDTH is U+1171E AHOM CONSONANT SIGN MEDIAL RA.
It changed like this:
EastAsianWidth.txt 15.1.0: 1171D..1171F ; N # Mn [3] AHOM CONSONANT SIGN MEDIAL LA..AHOM CONSONANT SIGN MEDIAL LIGATING RA
EastAsianWidth.txt 16.0.0: 1171E ; N # Mc AHOM CONSONANT SIGN MEDIAL RA
I.e it changed from Mn (Mark Nonspacing) to Mc (Mark Spacing
combining). So it should now have width 1 instead of 0, therefore it
is OK that it was removed from WIDTH, characters not in WIDTH get
width 1 by default.
Nothing suspicious when browsing the list of the 170 added characters.
Changes in ctype:
alpha: Added 4452 characters in new ctype which were not in old ctype
combining: Added 51 characters in new ctype which were not in old ctype
combining_level3: Added 43 characters in new ctype which were not in old ctype
graph: Added 5185 characters in new ctype which were not in old ctype
lower: Added 25 characters in new ctype which were not in old ctype
print: Added 5185 characters in new ctype which were not in old ctype
punct: Missing 33 characters of old ctype in new ctype
punct: Added 766 characters in new ctype which were not in old ctype
tolower: Added 27 characters in new ctype which were not in old ctype
totitle: Added 27 characters in new ctype which were not in old ctype
toupper: Added 27 characters in new ctype which were not in old ctype
upper: Added 27 characters in new ctype which were not in old ctype
Nothing suspicous in the additions.
About the 33 characters removed from `punct`:
U+0363 - U+036F are identical in UnicodeData.txt. Difference in DerivedCoreProperties.txt:
DerivedCoreProperties.txt 15.1.0: not there.
DerivedCoreProperties.txt 16.0.0: 0363..036F ; Alphabetic # Mn [13] COMBINING LATIN SMALL LETTER A..COMBINING LATIN SMALL LETTER X
So that’s the reason why they are added to `alpha` and removed from `punct`.
Same for U+1DD3 - U+1DE6, they are identical in UnicodeData.txt but there is a difference in DerivedCoreProperties.txt:
DerivedCoreProperties.txt 15.1.0: 1DE7..1DF4 ; Alphabetic # Mn [14] COMBINING LATIN SMALL LETTER ALPHA..COMBINING LATIN SMALL LETTER U WITH DIAERESIS
DerivedCoreProperties.txt 16.0.0: 1DD3..1DF4 ; Alphabetic # Mn [34] COMBINING LATIN SMALL LETTER FLATTENED OPEN A ABOVE..COMBINING LATIN SMALL LETTER U WITH DIAERESIS
So they became `Alphabetic` and were thus added to `alpha` and removed from `punct`.
scripts: Remove arceb-linux-gnu from build-many-glibcs.py
This was discussed on the hallway track at GNU Tools Cauldron
2024. There are concerns about stability of the big-endian
GCC backend, and Linux removed support for the only big-endian
ARC platform in commit dd7c7ab01a04d645b7e7baa8530bfd81e31a2202
("ARC: [plat-eznps]: Drop support for EZChip NPS platform").
In Linux 6.11, fstat and newfstatat are added back. To avoid the messy
usage of the fstat, newfstatat, and statx system calls, we will continue
using statx only in glibc, maintaining consistency with previous versions of
the LoongArch-specific glibc implementation.
Use the setresuid32 system call if it is available, prefering
it over setresuid. If both system calls exist, setresuid
is the 16-bit variant. This fixes a build failure on
sparcv9-linux-gnu.
elf: Move __rtld_malloc_init_stubs call into _dl_start_final
Calling an extern function in a different translation unit before
self-relocation is brittle. The compiler may load the address
at an earlier point in _dl_start, before self-relocation. In
_dl_start_final, the call is behind a compiler barrier, so this
cannot happen.
elf: Remove version assert in check_match in elf/dl-lookup.c
This case is detected early in the elf/dl-version.c consistency
checks. (These checks could be disabled in the future to allow
the removal of symbol versioning from objects.)
Commit f0b2132b35 ("ld.so: Support moving versioned symbols between
sonames [BZ #24741]) removed another call to _dl_name_match_p. The
_dl_check_caller function no longer exists, and the remaining calls
to _dl_name_match_p happen under the loader lock. This means that
atomic accesses are no longer required for the l_libname list. This
supersedes commit 395be7c218 ("elf: Fix data race in _dl_name_match_p
[BZ #21349]").
iconv: Use $(run-program-prefix) for running iconv (bug 32197)
With --enable-hardcoded-path-in-tests, $(test-program-prefix)
does not redirect to the built glibc, but we need to run
iconv (the program) against the built glibc even with
--enable-hardcoded-path-in-tests, as it is using the ABI
path for the dynamic linker (as an installed program).
Use $(run-program-prefix) instead.
Joe Ramsay [Mon, 23 Sep 2024 14:32:53 +0000 (15:32 +0100)]
AArch64: Improve codegen in users of ADVSIMD expm1f helper
Rearrange operations so MOV is not necessary in reduction or around
the special-case handler. Reduce memory access by using more indexed
MLAs in polynomial.
Joe Ramsay [Mon, 23 Sep 2024 14:32:14 +0000 (15:32 +0100)]
AArch64: Improve codegen in users of AdvSIMD log1pf helper
log1pf is quite register-intensive - use fewer registers for the
polynomial, and make various changes to shorten dependency chains in
parent routines. There is now no spilling with GCC 14. Accuracy moves
around a little - comments adjusted accordingly but does not require
regen-ulps.
Use the helper in log1pf as well, instead of having separate
implementations. The more accurate polynomial means special-casing can
be simplified, and the shorter dependency chain avoids the usual dance
around v0, which is otherwise difficult.
There is a small duplication of vectors containing 1.0f (or 0x3f800000) -
GCC is not currently able to efficiently handle values which fit in FMOV
but not MOVI, and are reinterpreted to integer. There may be potential
for more optimisation if this is fixed.
Joe Ramsay [Mon, 23 Sep 2024 14:30:20 +0000 (15:30 +0100)]
AArch64: Improve codegen in SVE F32 logs
Reduce MOVPRFXs by using unpredicated (non-destructive) instructions
where possible. Similar to the recent change to AdvSIMD F32 logs,
adjust special-case arguments and bounds to allow for more optimal
register usage. For all 3 routines one MOVPRFX remains in the
reduction, which cannot be avoided as immediate AND and ASR are both
destructive.
Joe Ramsay [Mon, 23 Sep 2024 14:26:12 +0000 (15:26 +0100)]
AArch64: Improve codegen in SVE expf & related routines
Reduce MOV and MOVPRFX by improving special-case handling. Use inline
helper to duplicate the entire computation between the special- and
non-special case branches, removing the contention for z0 between x
and the return value.
Also rearrange some MLAs and MLSs - by making the multiplicand the
destination we can avoid a MOVPRFX in several cases. Also change which
constants go in the vector used for lanewise ops - the last lane is no
longer wasted.
Spotted that shift was incorrect in exp2f and exp10f, w.r.t. to the
comment that explains it. Fixed - worst-case ULP for exp2f moves
around but it doesn't change significantly for either routine.
Worst-case error for coshf increases due to passing x to exp rather
than abs(x) - updated the comment, but does not require regen-ulps.
Linux: Use readdir64_r for compat __old_readdir64_r (bug 32128)
It is not necessary to do the conversion at the getdents64
layer for readdir64_r. Doing it piecewise for readdir64
is slightly simpler and allows deleting __old_getdents64.
This fixes bug 32128 because readdir64_r handles the length
check correctly.
The tests check that O_EXCL is used properly, that 0600 is used
as the mode, that the characters used are as expected, and that
the distribution of names generated is reasonably random.
The tests run very slowly on some kernel versions, so make them
xtests.
Add tests of special cases for freopen that were omitted from the more
general tests of different modes and similar issues. The special
cases in the three tests here are logically unconnected, it was simply
convenient to put these tests in one patch.
* Test freopen with a NULL path to the new file, in a chroot. Rather
than asserting that this fails (logically, failure in this case is
an implementation detail; it's not required for freopen to rely on
/proc), verify that either it fails (without memory leaks) or that
it succeeds and behaves as expected on success. There is no check
for file descriptor leaks because the machinery for that also
depends on /proc, so can't be used in a chroot.
* Test that freopen and freopen64 are genuinely different in
configurations with 32-bit off_t by checking for an EFBIG trying to
write past 2GB in a file opened with freopen in such a configuration
but no error with 64-bit off_t or when opening with freopen64.
Joseph Myers [Fri, 20 Sep 2024 23:25:32 +0000 (23:25 +0000)]
Make tst-strtod-underflow type-generic
The test tst-strtod-underflow covers various edge cases close to the
underflow threshold for strtod (especially cases where underflow on
architectures with after-rounding tininess detection depends on the
rounding mode). Make it use the type-generic machinery, with
corresponding test inputs for each supported floating-point format, so
that other functions in the strtod family are tested for underflow
edge cases as well.
Joseph Myers [Fri, 20 Sep 2024 23:24:45 +0000 (23:24 +0000)]
Add tests of more strtod special cases
There is very little test coverage of inputs to strtod-family
functions that don't contain anything that can be parsed as a number
(one test of ".y" in tst-strtod2), and none that I can see of skipping
initial whitespace. Add some tests of these things to tst-strtod2.
Joseph Myers [Fri, 20 Sep 2024 23:24:02 +0000 (23:24 +0000)]
Add more tests of strtod end pointer
Although there are some tests in tst-strtod2 and tst-strtod3 for the
end pointer provided by strtod when it doesn't parse the whole string,
they aren't very thorough. Add tests of more such cases to
tst-strtod2.
Joseph Myers [Fri, 20 Sep 2024 23:23:13 +0000 (23:23 +0000)]
Make tst-strtod2 and tst-strtod5 type-generic
Some of the strtod tests use type-generic machinery in tst-strtod.h to
test the strto* functions for all floating types, while others only
test double even when the tests are in fact meaningful for all
floating types.
Convert tst-strtod2 and tst-strtod5 to use the type-generic machinery
so they test all floating types. I haven't tried to convert them to
use newer test interfaces in other ways, just made the changes
necessary to use the type-generic machinery.
Implement run-built-tests=no for make xcheck, always build xtests
Previously, the second occurrence of the xtests target
expected all xtests to run (as the result of specifying
$(xtests)), but these tests have not been run due to
the the first xtests target is set up for run-built-tests=no:
it only runs tests in $(xtests-special). Consequently,
xtests are reported as UNSUPPORTED with “make xcheck
run-built-tests=no”. The xtests were not built, either.
After this change always, xtests are built regardless
of the $(run-built-tests) variable (except for xtests listed
in $(tests-unsupported)). To fix the UNSUPPORTED issue,
introduce xtests-expected and use that manage test
expectations in the second xtests target.
Add new testcase elf/tst-startup-errno.c which tests that errno is set
to 0 at first ELF constructor execution and at the start of the
program's main function.
Add new file libio/tst-fclose-unopened2.c that tests whether fclose on an
unopened file returns EOF.
This test differs from tst-fclose-unopened.c by ensuring the file's buffer
is allocated prior to double-fclose. A comment in tst-fclose-unopened.c
now clarifies that it is testing a file with an unallocated buffer.
Calling fclose on unopened files normally causes a use-after-free bug,
however the standard streams are an exception since they are not
deallocated by fclose.
iconv: Support in-place conversions (bug 10460, bug 32033)
Check if any of the input files overlaps with the output file, and use
a temporary file in this case, so that the input is no clobbered
before it is read. This fixes bug 10460. It allows to use iconv
more easily as a functional replacement for GNU recode.
The updated output buffer management truncates the output file
if there is no input, fixing bug 32033.
In several converters, a __GCONV_ILLEGAL_INPUT result gets overwritten
with __GCONV_FULL_OUTPUT. As a result, iconv (the function) returns
E2BIG instead of EILSEQ. The iconv program does not see the original
EILSEQ failure, does not recognize the invalid input, and may
incorrectly exit successfully.
To address this, a new __flags bit is used to indicate a sticky input
error state. All __GCONV_ILLEGAL_INPUT results are replaced with a
function call that sets this new __GCONV_ENCOUNTERED_ILLEGAL_INPUT and
returns __GCONV_ILLEGAL_INPUT. The iconv program checks for
__GCONV_ENCOUNTERED_ILLEGAL_INPUT and overrides the exit status.
The converter changes introducing __gconv_mark_illegal_input are
mostly mechanical, except for the res variable initialization in
iconvdata/iso-2022-jp.c: this error gets overwritten with __GCONV_OK
and other results in the following code. If res ==
__GCONV_ILLEGAL_INPUT afterwards, STANDARD_TO_LOOP_ERR_HANDLER below
will handle it.
The __gconv_mark_illegal_input changes do not alter the errno value
set by the iconv function. This is simpler to implement than
reviewing each __GCONV_FULL_OUTPUT result and adjust it not to
override a previous __GCONV_ILLEGAL_INPUT result. Doing it that way
would also change some E2BIG errors in to EILSEQ errors, so it had to
be done conditionally (under a flag set by the iconv program only), to
avoid confusing buffer management in other applications.
iconv: Do not use mmap in iconv (the program) (bug 17703)
On current systems, very large files are needed before
mmap becomes beneficial. Simplify the implementation.
This exposed that inptr was not initialized correctly in
process_fd. Handling multiple input files resulted in
EFAULT in read because a null pointer was passed. This
could be observed previously if an input file was not
mappable and was reported as bug 17703.
Joe Ramsay [Thu, 19 Sep 2024 16:34:02 +0000 (17:34 +0100)]
AArch64: Add vector logp1 alias for log1p
This enables vectorisation of C23 logp1, which is an alias for log1p.
There are no new tests or ulp entries because the new symbols are simply
aliases.
hurd: Avoid file_check_access () RPC for access (F_OK)
A common use case of access () / faccessat () is checking for file
existence, not any specific access permissions. In that case, we can
avoid doing the file_check_access () RPC; whether the given path had
been successfully resolved to a file is all we need to know to answer.
This is prompted by GLib switching to use faccessat (F_OK) to implement
g_file_query_exists () for local files.
https://gitlab.gnome.org/GNOME/glib/-/merge_requests/4272
tst: Extend cross-test-ssh.sh to support passing glibc tunables
This patch adds new flag --glibctunables to the cross-test-ssh.sh script
to pass Glibc tunables to the system on which tests are executed.
The value to pass can be also provided via the GLIBC_TUNABLES environment
variable.
This works similar to the TIMEOUTFACTOR variable.
Sometimes it is useful to cross test glibc with some non-default tunable,
and a global environment variable is the easiest way to inject some
tunable value into most tests. With this patch using cross-test-ssh.sh
script becomes very similar to running a test natively on the local host
when using non-default tunable is important. Reviewed-by: Arjun Shankar <arjun@redhat.com>
Linux: Add the sched_setattr and sched_getattr functions
And struct sched_attr.
In sysdeps/unix/sysv/linux/bits/sched.h, the hack that defines
sched_param around the inclusion of <linux/sched/types.h> is quite
ugly, but the definition of struct sched_param has already been
dropped by the kernel, so there is nothing else we can do and maintain
compatibility of <sched.h> with a wide range of kernel header
versions. (An alternative would involve introducing a separate header
for this functionality, but this seems unnecessary.)
The existing sched_* functions that change scheduler parameters
are already incompatible with PTHREAD_PRIO_PROTECT mutexes, so
there is no harm in adding more functionality in this area.
The documentation mostly defers to the Linux manual pages.
git://sourceware.org / glibc.git / log