Linux 6.11 has getrandom() in vDSO. It operates on a thread-local opaque
state allocated with mmap using flags specified by the vDSO.
Multiple states are allocated at once, as many as fit into a page, and
these are held in an array of available states to be doled out to each
thread upon first use, and recycled when a thread terminates. As these
states run low, more are allocated.
To make this procedure async-signal-safe, a simple guard is used in the
LSB of the opaque state address, falling back to the syscall if there's
reentrancy contention.
Also, _Fork() is handled by blocking signals on opaque state allocation
(so _Fork() always sees a consistent state even if it interrupts a
getrandom() call) and by iterating over the thread stack cache on
reclaim_stack. Each opaque state will be in the free states list
(grnd_alloc.states) or allocated to a running thread.
The cancellation is handled by always using GRND_NONBLOCK flags while
calling the vDSO, and falling back to the cancellable syscall if the
kernel returns EAGAIN (would block). Since getrandom is not defined by
POSIX and cancellation is supported as an extension, the cancellation is
handled as 'may occur' instead of 'shall occur' [1], meaning that if
vDSO does not block (the expected behavior) getrandom will not act as a
cancellation entrypoint. It avoids a pthread_testcancel call on the fast
path (different than 'shall occur' functions, like sem_wait()).
It is currently enabled for x86_64, which is available in Linux 6.11,
and aarch64, powerpc32, powerpc64, loongarch64, and s390x, which are
available in Linux 6.12.
Link: https://pubs.opengroup.org/onlinepubs/9799919799/nframe.html [1]
Co-developed-by: Jason A. Donenfeld <Jason@zx2c4.com>
Tested-by: Jason A. Donenfeld <Jason@zx2c4.com> # x86_64
Tested-by: Adhemerval Zanella <adhemerval.zanella@linaro.org> # x86_64, aarch64
Tested-by: Xi Ruoyao <xry111@xry111.site> # x86_64, aarch64, loongarch64
Tested-by: Stefan Liebler <stli@linux.ibm.com> # s390x
Add a new test tst-faccessat-setuid that iterates through real and
effective UID/GID combination and tests the faccessat() interface for
default and AT_EACCESS flags.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Use libsupport convenience functions and macros instead of the old
test-skeleton. Also add a new xdup() convenience wrapper function.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Add xdup as the error-checking version of dup for test cases.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
When building with e.g. -std=c99 and _ATFILE_SOURCE, stat.h was missing
including bits/types/struct_timespec.h to get the struct timespec
declaration for utimensat.
For instance, 1073741906 leads to system 16, subsystem 0 and code 82,
which is in range (max_code is 122), but not defined. Return EINVAL in
that case, like
A static analyzer apparently reported an uninitialized use of the
variable result in sem_open in the case where the file is required to
exist but does not exist.
The report appears to be correct; set result to SEM_FAILED in that
case, and add a test for it.
Note: the test passes for me even without the sem_open fix, I guess
because result happens to get value SEM_FAILED (i.e. 0) when
uninitialized.
Tested for x86_64.
Per the rseq syscall documentation, 3 fields are required to be
initialized by userspace prior to registration, they are 'cpu_id',
'rseq_cs' and 'flags'. Since we have no guarantee that 'struct pthread'
is cleared on all architectures, explicitly set those 3 fields prior to
registration.
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Reviewed-by: Florian Weimer <fweimer@redhat.com>
The declaration of found_other_class could be jumped
over via the goto just above it, but the code jumped
to uses found_other_class. Move the declaration
up a bit to ensure it's properly declared and initialized.
The commit 9247f53219 triggered some regressions on loongarch and
riscv:
math/test-float-log10
math/test-float32-log10
And it is due a wrong sync with CORE-MATH for special 0.0/-0.0
inputs.
Checked on aarch64-linux-gnu and loongarch64-linux-gnu-lp64d.
Wire vasprintf into test infrastructure for formatted printf output
specifiers.
Owing to mtrace logging these tests take amounts of time to complete
similar to those of corresponding asprintf tests, so set timeouts for
the tests accordingly, with a global default for all the vasprintf
tests, and then individual higher settings for double and long double
tests each.
Reviewed-by: DJ Delorie <dj@redhat.com>
Wire asprintf into test infrastructure for formatted printf output
specifiers.
Owing to mtrace logging of lots of memory allocation calls these tests
take a considerable amount of time to complete, except for the character
conversion, taking from 00m20s for 'tst-printf-format-as-s --direct s',
through 01m10s and 03m53s for 'tst-printf-format-as-char --direct i' and
'tst-printf-format-as-double --direct f' respectively, to 19m24s for
'tst-printf-format-as-ldouble --direct f', all in standalone execution
from NFS on a RISC-V FU740@1.2GHz system and with output redirected over
100Mbps network via SSH. It is with the skeleton's stub implementation
of dladdr(3); execution times with regular dladdr(3) are up to over
twice longer.
Set timeouts for the tests accordingly then, with a global default for
all the asprintf tests, and then individual higher settings for double
and long double tests each.
Reviewed-by: DJ Delorie <dj@redhat.com>
This is a collection of tests for formatted printf output specifiers
covering the d, i, o, u, x, and X integer conversions, the e, E, f, F,
g, and G floating-point conversions, the c character conversion, and the
s string conversion. Also the hh, h, l, and ll length modifiers are
covered with the integer conversions as is the L length modifier with
the floating-point conversions.
The -, +, space, #, and 0 flags are iterated over, as permitted by the
conversion handled, in tuples of 1..5, including tuples with repetitions
of 2, and combined with field width and/or precision, again as permitted
by the conversion. The resulting format string is then used to produce
output from respective sets of input data corresponding to the specific
conversion under test. POSIX extensions beyond ISO C are not used.
Output is produced in the form of records which include both the format
string (and width and/or precision where given in the form of separate
arguments) and the conversion result, and is verified with GNU AWK using
the format obtained from each such record against the reference value
also supplied, relying on the fact that GNU AWK has its own independent
implementation of format processing, striving to be ISO C compatible.
In the course of implementation I have determined that in the non-bignum
mode GNU AWK uses system sprintf(3) for the floating-point conversions,
defeating the objective of doing the verification against an independent
implementation. Additionally the bignum mode (using MPFR) is required
to correctly output wider integer and floating-point data. Therefore
for the conversions affected the relevant shell scripts sanity-check AWK
and terminate with unsupported status if the bignum mode is unavailable
for floating-point data or where data is output incorrectly.
The f and F floating-point conversions are build-time options for GNU
AWK, depending on the environment, so they are probed for before being
used. Similarly the a and A floating-point conversions, however they
are currently not used, see below. Also GNU AWK does not handle the b
or B integer conversions at all at the moment, as at 5.3.0. Support for
the a, A, b, and B conversions can however be easily added following the
approach taken for the f and F conversions.
Output produced by gawk for the a and A floating-point conversions does
not match one produced by us: insufficient precision is used where one
hasn't been explicitly given, e.g. for the negated maximum finite IEEE
754 64-bit value of -1.79769313486231570814527423731704357e+308 and "%a"
format we produce -0x1.fffffffffffffp+1023 vs gawk's -0x1.000000p+1024
and a different exponent is chosen otherwise, such as with "%.a" where
we output -0x2p+1023 vs gawk's -0x1p+1024 for the same value, or "%.20a"
where -0x1.fffffffffffff0000000p+1023 is our output, but gawk produces
-0xf.ffffffffffff80000000p+1020 instead. Consequently I chose not to
include a and A conversions in testing at this time.
And last but not least there are numerous corner cases that GNU AWK does
not handle correctly, which are worked around by explicit handling in
the AWK script. These are in particular:
- extraneous leading 0 produced for the alternative form with the o
conversion, e.g. { printf "%#.2o", 1 } produces "001" rather than
"01",
- unexpected 0 produced where no characters are expected for the input
of 0 and the alternative form with the precision of 0 and the integer
hexadecimal conversions, e.g. { printf "%#.x", 0 } produces "0" rather
than "",
- missing + character in the non-bignum mode only for the input of 0
with the + flag, precision of 0 and the signed integer conversions,
e.g. { printf "%+.i", 0 } produces "" rather than "+",
- missing space character in the non-bignum mode only for the input of 0
with the space flag, precision of 0 and the signed integer
conversions, e.g. { printf "% .i", 0 } produces "" rather than " ",
- for released gawk versions of up to 4.2.1 missing - character for the
input of -NaN with the floating-point conversions, e.g. { printf "%e",
"-nan" }' produces "nan" rather than "-nan",
- for released gawk versions from 5.0.0 onwards + character output for
the input of -NaN with the floating-point conversions, e.g. { printf
"%e", "-nan" }' produces "+nan" rather than "-nan",
- for released gawk versions from 5.0.0 onwards + character output for
the input of Inf or NaN in the absence of the + or space flags with
the floating-point conversions, e.g. { printf "%e", "inf" }' produces
"+inf" rather than "inf",
- for released gawk versions of up to 4.2.1 missing + character for the
input of Inf or NaN with the + flag and the floating-point
conversions, e.g. { printf "%+e", "inf" }' produces "inf" rather than
"+inf",
- for released gawk versions of up to 4.2.1 missing space character for
the input of Inf or NaN with the space flag and the floating-point
conversions, e.g. { printf "% e", "nan" }' produces "nan" rather than
" nan",
- for released gawk versions from 5.0.0 onwards + character output for
the input of Inf or NaN with the space flag and the floating-point
conversions, e.g. { printf "% e", "inf" }' produces "+inf" rather than
" inf",
- for released gawk versions from 5.0.0 onwards the field width is
ignored for the input of Inf or NaN and the floating-point
conversions, e.g. { printf "%20e", "-inf" }' produces "-inf" rather
than " -inf",
NB for released gawk versions of up to 4.2.1 floating-point conversion
issues apply to the bignum mode only, as in the non-bignum mode system
sprintf(3) is used. As from version 5.0.0 specialized handling has been
added for [-]Inf and [-]NaN inputs and the issues listed apply to both
modes. The '--posix' flag makes gawk versions from 5.0.0 onwards avoid
the issue with field width and the + character unconditionally output
for the input of Inf or NaN, however not the remaining issues and then
the 'gensub' function is not supported in the POSIX mode, so to go this
path I deemed not worth it.
Each test completes within single seconds except for the long double
one. There the F/f formats produce a large number of digits, which
appears to be computationally intensive and CPU-bound. Standalone
execution time for 'tst-printf-format-p-ldouble --direct f' is in the
range of 00m36s for POWER9@2.166GHz and 09m52s for FU740@1.2GHz and
output redirected locally to /dev/null, and 10m11s for FU740 and output
redirected over 100Mbps network via SSH to /dev/null, so the throughput
of the network adds very little (~3.2% in this case) to the processing
time. This is with IEEE 754 quad.
So I have scaled the timeout for 'tst-printf-format-skeleton-ldouble'
accordingly. Regardless, following recent practice the test has been
added to the standard rather than extended set. However, unlike most
of the remaining tests it has been split by the conversion specifier,
so as to allow better parallelization of this long-running test. As
a side effect this lets the test report the unsupported status for the
F/f conversions where applicable, so 'tst-printf-format-p-double' has
been split for consistency as well.
Only printf itself is handled at the moment, but the infrastructure
provides for all the printf family functions to be verified, changes
for which to be supplied separately. The complication around having
some tests iterating over all the relevant conversion specifiers and
other verifying conversion specifiers individually combined with
iterating over printf family functions has hit a peculiarity in GNU
make where the use of multiple targets with a pattern rule is handled
differently from such use with an ordinary rule. Consequently it
seems impossible to bulk-define a pattern rule using '$(foreach ...)',
where each target would simply trigger the recipe according to the
pattern and matching dependencies individually (such a rule does work,
but implies all targets to be updated with a single recipe execution).
Therefore as a compromise a single single-target pattern rule has been
defined that has listed all the conversion-specific scripts and all the
test executables as dependencies. Consequently tests will be rerun in
the absence of changes to their actual sources or scripts whenever an
unrelated file has changed that has been listed. Also all the formatted
printf output tests will always be built whenever any single one is to
be run. This only affects test development and not test runs in the
field, though it does change the order of execution of the individual
steps and also acts as a Makefile barrier in parallel runs. As the
execution time dominates the compilation time for these tests it is not
seen as a serious shortcoming.
As pointed out by Florian Weimer <fweimer@redhat.com> the malloc tracing
facility can take a substantial amount of time in calling dladdr(3) to
determine the caller's location. This is not needed by the verification
made with these tests, so I chose to interpose the symbol with a stub
implementation that always fails in the shared skeleton. We have total
control over the test environment, so I think it is a safe and minimal
impact approach. If there's ever anything else added to the tests that
would actually rely on dladdr(3) returning usable results, only then we
can think of a different approach.
Reviewed-by: DJ Delorie <dj@redhat.com>
On GCC 11 (x86-64), the previous code produced test failures like
this one:
Failure: Test: exp10m1_towardzero (-0x1.1p+4)
Result:
is: -1.00000000e+00 -0x1.000000p+0
should be: -9.99999940e-01 -0x1.fffffep-1
difference: 5.96046447e-08 0x1.000000p-24
ulp : 1.0000
max.ulp : 0.0000
Apply a similar fix to exp2m1f.
Co-authored-by: Paul Zimmermann <Paul.Zimmermann@inria.fr>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Change name of the access_rights argument to access_restrictions
of the following functions:
- pkey_alloc()
- pkey_set()
as this argument refers to access restrictions rather than access
rights and previous name might have been misleading.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Update the name of the argument in several pkey_*() functions that refers
to access restrictions rather than access rights: change access "rights"
to access "restrictions".
Specify that the result of the pkey_get() should be checked using bitwise
operations rather than plain equals comparison.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Before commit ee1ada1bdb
("elf: Rework exception handling in the dynamic loader
[BZ #25486]"), the previous order called the main calloc
to allocate a shadow GOT/PLT array for auditing support.
This happened before libc.so.6 ELF constructors were run, so
a user malloc could run without libc.so.6 having been
initialized fully. One observable effect was that
environ was NULL at this point.
It does not seem to be possible at present to trigger such
an allocation, but it seems more robust to delay switching
to main malloc after ld.so self-relocation is complete.
The elf/tst-rtld-no-malloc-audit test case fails with a
2.34-era glibc that does not have this fix.
Reviewed-by: DJ Delorie <dj@redhat.com>
For a long time, libc.so.6 has dependend on ld.so, which
means that there is a reference to ld.so in all processes,
and rtld_multiple_ref is always true. In fact, if
rtld_multiple_ref were false, some of the ld.so setup code
would not run.
Reviewed-by: DJ Delorie <dj@redhat.com>
Linux 3.15 and 6.2 added HWCAP2_* values for Arm. These bits have
already been added to dl-procinfo.{c,h} in commits 9aea0cb842 and
8ebe9c0b38. Also add them to <bits/hwcap.h> so that they can be used
in user code. For example, for checking bits in the value returned by
getauxval(AT_HWCAP2).
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Reviewed-by: Yury Khrustalev <yury.khrustalev@arm.com>
This patch starts preparation for C2Y support in glibc headers by
adding a feature test macro _ISOC2Y_SOURCE and corresponding
__GLIBC_USE (ISOC2Y). (I mostly copied the work of Joseph Myers
for C2X). As with other such macros, C2Y features are also
enabled by compiling for a standard newer than C23, or by using
_GNU_SOURCE.
This patch does not itself enable anything new in the headers for C2Y;
that is to be done in followup patches. (For example an implementation
of WG14 N3349.)
Once C2Y becomes an actual standard we'll presumably move to using the
actual year in the feature test macro and __GLIBC_USE, with some
period when both macro spellings are accepted, as was done with
_ISOC2X_SOURCE.
Tested for x86_64.
Signed-off-by: Lenard Mollenkopf <glibc@lenardmollenkopf.de>
By using a combination of mask-and-add instead of the shift-based
index calculation the routines can share the same table as other
variants with no performance degradation.
The tables change name because of other changes in downstream AOR.
Reviewed-by: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
The CORE-MATH exp2m1f implementation showed slight worse latency
when using x86_64 baseline ABI. This patch adds a ifunc variant
with similar performance for x86_64-v3.
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: DJ Delorie <dj@redhat.com>
The CORE-MATH exp10m1f implementation showed slight worse latency
when using x86_64 baseline ABI. This patch adds a ifunc variant
with similar performance for x86_64-v3.
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: DJ Delorie <dj@redhat.com>
The CORE-MATH implementation is correctly rounded (for any rounding mode)
and shows better performance compared to the generic exp2m1f.
The code was adapted to glibc style and to use the definition of
math_config.h (to handle errno, overflow, and underflow). The
only change is to handle FLT_MAX_EXP for FE_DOWNWARD or FE_TOWARDZERO.
The benchmark inputs are based on exp2f ones.
Benchtest on x64_64 (Ryzen 9 5900X, gcc 14.2.1), aarch64 (Neoverse-N1,
gcc 13.3.1), and powerpc (POWER10, gcc 13.2.1):
Latency master patched improvement
x86_64 40.6042 48.7104 -19.96%
x86_64v2 40.7506 35.9032 11.90%
x86_64v3 35.2301 31.7956 9.75%
i686 102.094 94.6657 7.28%
aarch64 18.2704 15.1387 17.14%
power10 11.9444 8.2402 31.01%
reciprocal-throughput master patched improvement
x86_64 20.8683 16.1428 22.64%
x86_64v2 19.5076 10.4474 46.44%
x86_64v3 19.2106 10.4014 45.86%
i686 56.4054 59.3004 -5.13%
aarch64 12.0781 7.3953 38.77%
power10 6.5306 5.9388 9.06%
The generic implementation calls __ieee754_exp2f and x86_64 provides
an optimized ifunc version (built with -mfma -mavx2, not correctly
rounded). This explains the performance difference for x86_64.
Same for i686, where the ABI provides an optimized __ieee754_exp2f
version built with '-msse2 -mfpmath=sse'. When built wth same
flags, the new algorithm shows a better performance:
master patched improvement
latency 102.094 91.2823 10.59%
reciprocal-throughput 56.4054 52.7984 6.39%
Signed-off-by: Alexei Sibidanov <sibid@uvic.ca>
Signed-off-by: Paul Zimmermann <Paul.Zimmermann@inria.fr>
Signed-off-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
The CORE-MATH implementation is correctly rounded (for any rounding mode)
and shows better performance compared to the generic exp10m1f.
The code was adapted to glibc style and to use the definition of
math_config.h (to handle errno, overflow, and underflow). I mostly
fixed some small issues in corner cases (sNaN handling, -INFINITY,
a specific overflow check).
Benchtest on x64_64 (Ryzen 9 5900X, gcc 14.2.1), aarch64 (Neoverse-N1,
gcc 13.3.1), and powerpc (POWER10, gcc 13.2.1):
Latency master patched improvement
x86_64 45.4690 49.5845 -9.05%
x86_64v2 46.1604 36.2665 21.43%
x86_64v3 37.8442 31.0359 17.99%
i686 121.367 93.0079 23.37%
aarch64 21.1126 15.0165 28.87%
power10 12.7426 8.4929 33.35%
reciprocal-throughput master patched improvement
x86_64 19.6005 17.4005 11.22%
x86_64v2 19.6008 11.1977 42.87%
x86_64v3 17.5427 10.2898 41.34%
i686 59.4215 60.9675 -2.60%
aarch64 13.9814 7.9173 43.37%
power10 6.7814 6.4258 5.24%
The generic implementation calls __ieee754_exp10f which has an
optimized version, although it is not correctly rounded, which is
the main culprit of the the latency difference for x86_64 and
throughp for i686.
Signed-off-by: Alexei Sibidanov <sibid@uvic.ca>
Signed-off-by: Paul Zimmermann <Paul.Zimmermann@inria.fr>
Signed-off-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
