Florian Weimer [Thu, 27 Jun 2024 14:26:56 +0000 (16:26 +0200)]
Enhanced test coverage for strncmp, wcsncmp
Add string/test-strncmp-nonarray and
wcsmbs/test-wcsncmp-nonarray.
This is the test that uncovered bug 31934. Test run time
is more than one minute on a fairly current system, so turn
these into xtests that do not run automatically.
Thought about doing `exit` as well since its only called once per
process, but since its easy to imagine a hot path leading into
`exit(0)`, its less clear if its profitable.
Even if C/POSIX standard states that exit is not formally thread-unsafe,
calling it more than once is UB. The glibc already supports
it for the single-thread, and both elf/nodelete2.c and tst-rseq-disable.c
call exit from a DSO destructor (which is called by _dl_fini, registered
at program startup with __cxa_atexit).
However, there are still race issues when it is called more than once
concurrently by multiple threads. A recent Rust PR triggered this
issue [1], which resulted in an Austin Group ask for clarification [2].
Besides it, there is a discussion to make concurrent calling not UB [3],
wtih a defined semantic where any remaining callers block until the first
call to exit has finished (reentrant calls, leaving through longjmp, and
exceptions are still undefined).
For glibc, at least reentrant calls are required to be supported to avoid
changing the current behaviour. This requires locking using a recursive
lock, where any exit called by atexit() handlers resumes at the point of
the current handler (thus avoiding calling the current handle multiple
times).
Checked on x86_64-linux-gnu and aarch64-linux-gnu.
[1] https://github.com/rust-lang/rust/issues/126600
[2] https://austingroupbugs.net/view.php?id=1845
[3] https://www.openwall.com/lists/libc-coord/2024/07/24/4 Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* mseal for all architectures.
* map_shadow_stack for x32.
* Replace sync_file_range with sync_file_range2 for csky (which
fixes a broken sync_file_range usage).
Update syscall-names.list and regenerate the arch-syscall.h headers
with build-many-glibcs.py update-syscalls.
Tested with build-many-glibcs.py. Reviewed-by: Florian Weimer <fweimer@redhat.com>
Michael Karcher [Sun, 28 Jul 2024 13:30:57 +0000 (15:30 +0200)]
Mitigation for "clone on sparc might fail with -EFAULT for no valid reason" (bz 31394)
It seems the kernel can not deal with uncommitted stack space in the area intended
for the register window when executing the clone() system call. So create a nested
frame (proxy for the kernel frame) and flush it from the processor to memory to
force committing pages to the stack before invoking the system call.
Bug: https://www.mail-archive.com/debian-glibc@lists.debian.org/msg62592.html
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31394
See-also: https://lore.kernel.org/sparclinux/62f9be9d-a086-4134-9a9f-5df8822708af@mkarcher.dialup.fu-berlin.de/ Signed-off-by: Michael Karcher <sourceware-bugzilla@mkarcher.dialup.fu-berlin.de> Reviewed-by: DJ Delorie <dj@redhat.com>
manual: make setrlimit() description less ambiguous
The existing description for setrlimit() has some ambiguity. It could be
understood to have the semantics of getrlimit(), i.e., the limits from the
process are stored in the provided rlp pointer.
Make the description more explicit that rlp are the input values, and that
the limits of the process is changed with this function.
The manual entry for `putc' described what "most systems" do instead of
describing the glibc implementation and its guarantees. This commit
fixes that by warning that putc may be implemented as a macro that
double-evaluates `stream', and removing the performance claim.
Even though the current `putc' implementation does not double-evaluate
`stream', offering this obscure guarantee as an extension to what
POSIX allows does not seem very useful.
The entry for `putwc' is also edited to bring it in line with `putc'. Reviewed-by: Florian Weimer <fweimer@redhat.com>
This helps compilers split the codegen for setting up the arguments
(`__expression`, `__filename`, etc...) from the potentially hot cold
where the `assert` is to a presumably cold region on the assertion
failure path.
Reviewed-by: Florian Weimer <fweimer@redhat.com> Reviewed-by: Sam James <sam@gentoo.org>
stdio-common: Add test for vfscanf with matches longer than INT_MAX [BZ #27650]
Complement commit b03e4d7bd25b ("stdio: fix vfscanf with matches longer
than INT_MAX (bug 27650)") and add a test case for the issue, inspired
by the reproducer provided with the bug report.
This has been verified to succeed as from the commit referred and fail
beforehand.
As the test requires 2GiB of data to be passed around its performance
has been evaluated using a choice of systems and the execution time
determined to be respectively in the range of 9s for POWER9@2.166GHz,
24s for FU740@1.2GHz, and 40s for 74Kf@950MHz. As this is on the verge
of and beyond the default timeout it has been increased by the factor of
8. Regardless, following recent practice the test has been added to the
standard rather than extended set.
Add a FAIL test failure helper analogous to FAIL_RET, that does not
cause the current function to return, providing a standardized way to
report a test failure with a message supplied while permitting the
caller to continue executing, for further reporting, cleaning up, etc.
Update existing test cases that provide a conflicting definition of FAIL
by removing the local FAIL definition and then as follows:
- tst-fortify-syslog: provide a meaningful message in addition to the
file name already added by <support/check.h>; 'support_record_failure'
is already called by 'support_print_failure_impl' invoked by the new
FAIL test failure helper.
- tst-ctype: no update to FAIL calls required, with the name of the file
and the line number within of the failure site additionally included
by the new FAIL test failure helper, and error counting plus count
reporting upon test program termination also already provided by
'support_record_failure' and 'support_report_failure' respectively,
called by 'support_print_failure_impl' and 'adjust_exit_status' also
respectively. However in a number of places 'printf' is called and
the error count adjusted by hand, so update these places to make use
of FAIL instead. And last but not least adjust the final summary just
to report completion, with any error count following as reported by
the test driver.
- test-tgmath2: no update to FAIL calls required, with the name of the
file of the failure site additionally included by the new FAIL test
failure helper. Also there is no need to track the return status by
hand as any call to FAIL will eventually cause the test case to return
an unsuccesful exit status regardless of the return status from the
test function, via a call to 'adjust_exit_status' made by the test
driver.
posix: Use <support/check.h> facilities in tst-truncate and tst-truncate64
Remove local FAIL macro in favor to FAIL_RET from <support/check.h>,
which provides equivalent reporting, with the name of the file of the
failure site additionally included, for the tst-truncate-common core
shared between the tst-truncate and tst-truncate64 tests.
nptl: Use <support/check.h> facilities in tst-setuid3
Remove local FAIL macro in favor to FAIL_EXIT1 from <support/check.h>,
which provides equivalent reporting, with the name of the file and the
line number within of the failure site additionally included. Remove
FAIL_ERR altogether and include ": %m" explicitly with the format string
supplied to FAIL_EXIT1 as there seems little value to have a separate
macro just for this.
1. Add shadow stack support to x32 signal.
2. Use the 64-bit map_shadow_stack syscall for x32.
3. Set up shadow stack for x32.
Add the map_shadow_stack system call to <fixup-asm-unistd.h> and regenerate
arch-syscall.h. Tested on Intel Tiger Lake with CET enabled x32. There
are no regressions with CET enabled x86-64. There are no changes in CET
enabled x86-64 _dl_start_user.
Signed-off-by: H.J. Lu <hjl.tools@gmail.com> Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
H.J. Lu [Tue, 23 Jul 2024 00:47:21 +0000 (17:47 -0700)]
x86-64: Remove sysdeps/x86_64/x32/dl-machine.h
Remove sysdeps/x86_64/x32/dl-machine.h by folding x32 ARCH_LA_PLTENTER,
ARCH_LA_PLTEXIT and RTLD_START into sysdeps/x86_64/dl-machine.h. There
are no regressions on x86-64 nor x32. There are no changes in x86-64
_dl_start_user. On x32, _dl_start_user changes are
resolv: Do not wait for non-existing second DNS response after error (bug 30081)
In single-request mode, there is no second response after an error
because the second query has not been sent yet. Waiting for it
introduces an unnecessary timeout.
Miguel Martín [Tue, 16 Jul 2024 15:14:57 +0000 (17:14 +0200)]
malloc: add multi-threaded tests for aligned_alloc/calloc/malloc
Improve aligned_alloc/calloc/malloc test coverage by adding
multi-threaded tests with random memory allocations and with/without
cross-thread memory deallocations.
Perform a number of memory allocation calls with random sizes limited
to 0xffff.
Use the existing DSO ('malloc/tst-aligned_alloc-lib.c') to randomize
allocator selection.
The multi-threaded allocation/deallocation is staged as described below:
- Stage 1: Half of the threads will be allocating memory and the
other half will be waiting for them to finish the allocation.
- Stage 2: Half of the threads will be allocating memory and the
other half will be deallocating memory.
- Stage 3: Half of the threads will be deallocating memory and the
second half waiting on them to finish.
Add 'malloc/tst-aligned-alloc-random-thread.c' where each thread will
deallocate only the memory that was previously allocated by itself.
Add 'malloc/tst-aligned-alloc-random-thread-cross.c' where each thread
will deallocate memory that was previously allocated by another thread.
The intention is to be able to utilize existing malloc testing to ensure
that similar allocation APIs are also exposed to the same rigors. Reviewed-by: Arjun Shankar <arjun@redhat.com>
Miguel Martín [Tue, 16 Jul 2024 15:14:56 +0000 (17:14 +0200)]
malloc: avoid global locks in tst-aligned_alloc-lib.c
Make sure the DSO used by aligned_alloc/calloc/malloc tests does not get
a global lock on multithreaded tests. Reviewed-by: Arjun Shankar <arjun@redhat.com>
elf: Fix localplt.awk for DT_RELR-enabled builds (BZ 31978)
For each input readelf output, localplt.awk parses each 'Relocation
section' entry, checks its offset against the dynamic section entry, and
saves each DT_JMPREL, DT_RELA, and DT_REL offset value it finds. After
all lines are read, the script checks if any segment offset differed
from 0, meaning at least one 'Relocation section' was matched.
However, if the shared object was built with RELR support and the static
linker could place all the relocation on DT_RELR, there would be no
DT_JMPREL, DT_RELA, and DT_REL entries; only a DT_RELR.
For the current three ABIs that support (aarch64, x86, and powerpc64),
the powerpc64 ld.so shows the behavior above. Both x86_64 and aarch64
show extra relocations on '.rela.dyn', which makes the script check to
succeed.
This patch fixes by handling DT_RELR, where the offset is checked
against the dynamic section entries and if the shared object contains an
entry it means that there are no extra PLT entries (since all
relocations are relative).
It fixes the elf/check-localplt failure on powerpc.
Checked with a build/check for aarch64-linux-gnu, x86_64-linux-gnu,
i686-linux-gnu, arm-linux-gnueabihf, s390x-linux-gnu, powerpc-linux-gnu,
powerpc64-linux-gnu, and powerpc64le-linux-gnu. Reviewed-by: Carlos O'Donell <carlos@redhat.com>
linux: Also check pkey_get for ENOSYS on tst-pkey (BZ 31996)
The powerpc pkey_get/pkey_set support was only added for 64-bit [1],
and tst-pkey only checks if the support was present with pkey_alloc
(which does not fail on powerpc32, at least running a 64-bit kernel).
Samuel Thibault [Wed, 17 Jul 2024 12:56:14 +0000 (14:56 +0200)]
htl: Fix __pthread_init_thread declaration and definition
0e75c4a4634f ("hurd: Fix pthread_self() without libpthread") added a
declaration for ___pthread_init_thread instead of __pthread_init_thread,
and missed defining the external hidden symbol.
Samuel Thibault [Wed, 17 Jul 2024 12:06:25 +0000 (14:06 +0200)]
hurd: Fix pthread_self() without libpthread
5476f8cd2e68 ("htl: move pthread_self info libc.") moved the htl
pthread_self() function from libpthread to libc, replacing the previous libc
stub that just returns 0. And 53da64d1cf36 ("htl: Initialize ___pthread_self
early") added initialization code which is needed before being able to
call pthread_self. It is currently in libpthread, and thus never called
before programs can call pthread_self from libc, which then segfaults
when accessing _pthread_self()->thread.
This moves the initialization to libc itself, as initialized variables, so
pthread_self can always be called fine.
LoongArch: Add cfi instructions for _dl_tlsdesc_dynamic
In _dl_tlsdesc_dynamic, there are three 'addi.d sp, sp, -size'
instructions to allocate stack size for Float/LSX/LASX registers.
Every 'addi.d sp, sp, -size' needs a cfi_adjust_cfa_offset because
of sp is used to compute CFA. But only one 'addi.d sp, sp, -size'
will be run according to HWCAP value. And all cfi_adjust_cfa_offset
will be executed in stack unwinding, it result in incorrect CFA.
Change _dl_tlsdesc_dynamic to _dl_tlsdesc_dynamic,
_dl_tlsdesc_dynamic_lsx and _dl_tlsdesc_dynamic_lasx.
Conflicting cfi instructions can be distributed to the three functions.
And cfi instructions can correspond to stack down instructions.
Add pthread_getname_np and pthread_setname_np for Hurd
We use thread_get_name and thread_set_name to get and set the thread
name, so nothing is stored in the thread structure since these functions
are supposed to be called sparingly.
One notable difference with Linux is that the thread name is up to 32
chars, whereas Linux's is 16.
Also added a mach_RPC_CHECK to check for the existing of gnumach RPCs.
Samuel Thibault [Sat, 13 Jul 2024 15:00:55 +0000 (17:00 +0200)]
hurd: Fix restoring message to be retried
save_data stores the start of the original message to be retried,
overwritten by the EINTR reply. In 64b builds the overwrite is however
rounded up to the 64b pointer size, so we have to save more than just
the 32b err.
nptl: Convert tst-sem11 and tst-sem12 tests to use the test driver
Fix an issue with commit 2af4e3e5668f ("Test of semaphores.") by making
the tst-sem11 and tst-sem12 tests use the test driver, preventing them
from ever causing testing to hang forever and never complete, such as
currently happening with the 'mips-linux-gnu' (o32 ABI) target. Adjust
the name of the PREPARE macro, which clashes with the interpretation of
its presence by the test driver, by using a TF_ prefix in reference to
the name of the 'tf' function. Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
nptl: Add copyright notice tst-sem11 and tst-sem12 tests
Add a copyright notice to the tst-sem11 and tst-sem12 tests, observing
that they have been originally contributed back in 2007, with commit 2af4e3e5668f ("Test of semaphores."). Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Stefan Liebler [Thu, 11 Jul 2024 09:28:53 +0000 (11:28 +0200)]
s390x: Fix segfault in wcsncmp [BZ #31934]
The z13/vector-optimized wcsncmp implementation segfaults if n=1
and there is only one character (equal on both strings) before
the page end. Then it loads and compares one character and misses
to check n again. The following load fails.
This patch removes the extra load and compare of the first character
and just start with the loop which uses vector-load-to-block-boundary.
This code-path also checks n.
With this patch both tests are passing:
- the simplified one mentioned in the bugzilla 31934
- the full one in Florian Weimer's patch:
"manual: Document a GNU extension for strncmp/wcsncmp"
(https://patchwork.sourceware.org/project/glibc/patch/874j9eml6y.fsf@oldenburg.str.redhat.com/):
On s390x-linux-gnu (z16), the new wcsncmp test fails due to bug 31934. Reviewed-by: Carlos O'Donell <carlos@redhat.com>
DJ Delorie [Mon, 8 Jul 2024 21:52:15 +0000 (17:52 -0400)]
manual: add syscalls
The purpose of this patch is to add some system calls that (1) aren't
otherwise documented, and (2) are merely redirected to the kernel, so
can refer to their documentation; and define a standard way of doing
so in the future. A more detailed explaination of how system calls
are wrapped is added along with reference to the Linux Man-Pages
project.
Default version of man-pages is in configure.ac but can be overridden
by --with-man-pages=X.Y
Adam Sampson [Mon, 6 May 2024 17:16:33 +0000 (18:16 +0100)]
ldconfig: Ignore all GDB extension files
ldconfig already ignores files with the -gdb.py suffix, but GDB also
looks for -gdb.gdb and -gdb.scm files. These aren't as widely used, but
libguile at least comes with a -gdb.scm file.
Rename is_gdb_python_file to is_gdb_extension_file, and make it
recognise all three types of GDB extension.
Signed-off-by: Adam Sampson <ats@offog.org> Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
This hasn't been looked at for a loong time (already guessing from
the number of missing entries), and it ain't pretty.
There are some 9-ulps results for float.
- ZaZaZebra (qemu-system-m68k clone of PowerBook 190 system)
- GCC 13.3.1 20240614 (Gentoo 13.3.1_p20240614 p17)
- ld GNU ld (Gentoo 2.42 p6) 2.42.0
- Linux ZaZaZebra 4.19.0-5-m68k #1 Gentoo 4.19.37-5 (2019-06-19) m68k 68040 68040 GNU/Linux
- manual build
- ../glibc/configure --enable-fortify-source --prefix=/usr
- Tested by Immolo (via Andreas K. Hüttel)
Signed-off-by: Andreas K. Hüttel <dilfridge@gentoo.org>
stdlib: fix arc4random fallback to /dev/urandom (BZ 31612)
The __getrandom_nocancel used by __arc4random_buf uses
INLINE_SYSCALL_CALL (which returns -1/errno) and the loop checks for
the return value instead of errno to fallback to /dev/urandom.
The malloc code now uses __getrandom_nocancel_nostatus, which uses
INTERNAL_SYSCALL_CALL, so there is no need to use the variant that does
not set errno (BZ#29624).
Michael Jeanson [Wed, 3 Jul 2024 16:35:34 +0000 (12:35 -0400)]
nptl: fix potential merge of __rseq_* relro symbols
While working on a patch to add support for the extensible rseq ABI, we
came across an issue where a new 'const' variable would be merged with
the existing '__rseq_size' variable. We tracked this to the use of
'-fmerge-all-constants' which allows the compiler to merge identical
constant variables. This means that all 'const' variables in a compile
unit that are of the same size and are initialized to the same value can
be merged.
In this specific case, on 32 bit systems 'unsigned int' and 'ptrdiff_t'
are both 4 bytes and initialized to 0 which should trigger the merge.
However for reasons we haven't delved into when the attribute 'section
(".data.rel.ro")' is added to the mix, only variables of the same exact
types are merged. As far as we know this behavior is not specified
anywhere and could change with a new compiler version, hence this patch.
Move the definitions of these variables into an assembler file and add
hidden writable aliases for internal use. This has the added bonus of
removing the asm workaround to set the values on rseq registration.
Florian Weimer [Thu, 2 May 2024 14:26:36 +0000 (16:26 +0200)]
manual: Recommendations for dynamic linker hardening
This new section in the manual provides recommendations for
use of glibc in environments with higher integrity requirements.
It's reflecting both current implementation shortcomings, and
challenges we inherit from ELF and psABI requirements.
The new testcase elf/tst-tunables-enable_secure-env segfaults on s390 (31bit).
There _start parses the auxiliary vector for some additional checks.
Therefore it skips over the zeros after the environment variables ...
0x7fffac20: 0x7fffbd17 0x7fffbd32 0x7fffbd69 0x00000000
------------------------------------------------^^^last environment variable
... and then it parses the auxiliary vector and stops at AT_NULL.
0x7fffac30: 0x00000000 0x00000021 0x00000000 0x00000000
--------------------------------^^^AT_SYSINFO_EHDR--------------^^^AT_NULL
----------------^^^newp-----------------------------------------^^^oldp
Afterwards it tries to access AT_PHDR which points to somewhere and segfaults.
Due to not incorporating the skip_env variable in the computation of oldp
when shuffling down the auxv in rtld.c, it just copies one entry with AT_NULL
and value 0x00000021 and stops the loop. In reality we have skipped
GLIBC_TUNABLES environment variable (=> skip_env=1). Thus we should copy from
here:
0x7fffac40: 0x00000021 0x7ffff000 0x00000010 0x007fffff
----------------^^^fixed-oldp
This patch fixes the computation of oldp when shuffling down auxiliary vector.
It also adds some checks in the testcase. Those checks also fail on
s390x (64bit) and x86_64 without the fix.
hppa/vdso: Provide 64-bit clock_gettime() vDSO only
Adhemerval noticed that the gettimeofday() and 32-bit clock_gettime()
vDSO calls won't be used by glibc on hppa, so there is no need to
declare them. Both syscalls will be emulated by utilizing return values
of the 64-bit clock_gettime() vDSO instead.
The clang open fortify wrapper from 4228baef1a94e8bde84ad74 added
a restriction where open with 3 arguments where flags do not
contain O_CREAT or O_TMPFILE are handled as invalid. They are
not invalid, since the third argument is ignored, and the gcc
wrapper also allows it.
Checked x86_64-linux-gnu and with a yocto build for some affected
packages. Tested-by: “Khem Raj <raj.khen@gmail.com>”
H.J. Lu [Thu, 20 Jun 2024 22:28:13 +0000 (15:28 -0700)]
Add --disable-static-c++-tests option [BZ #31797]
By default, if the C++ toolchain lacks support for static linking,
configure fails to find the C++ header files and the glibc build fails.
The --disable-static-c++-link-check option allows the glibc build to
finish, but static C++ tests will fail if the C++ toolchain doesn't
have the necessary static C++ libraries which may not be easily installed.
Add --disable-static-c++-tests option to skip the static C++ link check
and tests. This fixes BZ #31797.
The current minimum GCC version of glibc build is GCC 6.2 or newer. But
building i686 glibc with GCC 6.4 on Fedora 40 failed since the C++ header
files couldn't be found which was caused by the static C++ link check
failure due to missing __divmoddi4 which was referenced in i686 libc.a
and added to GCC 7. Add --disable-static-c++-link-check configure option
to disable the static C++ link test. The newly built i686 libc.a can be
used by GCC 6.4 to create static C++ tests. This fixes BZ #31412.
elf: Support recursive use of dynamic TLS in interposed malloc
It turns out that quite a few applications use bundled mallocs that
have been built to use global-dynamic TLS (instead of the recommended
initial-exec TLS). The previous workaround from
commit afe42e935b3ee97bac9a7064157587777259c60e ("elf: Avoid some
free (NULL) calls in _dl_update_slotinfo") does not fix all
encountered cases unfortunatelly.
This change avoids the TLS generation update for recursive use
of TLS from a malloc that was called during a TLS update. This
is possible because an interposed malloc has a fixed module ID and
TLS slot. (It cannot be unloaded.) If an initially-loaded module ID
is encountered in __tls_get_addr and the dynamic linker is already
in the middle of a TLS update, use the outdated DTV, thus avoiding
another call into malloc. It's still necessary to update the
DTV to the most recent generation, to get out of the slow path,
which is why the check for recursion is needed.
The bookkeeping is done using a global counter instead of per-thread
flag because TLS access in the dynamic linker is tricky.
All this will go away once the dynamic linker stops using malloc
for TLS, likely as part of a change that pre-allocates all TLS
during pthread_create/dlopen.
Carlos O'Donell [Wed, 19 Jun 2024 15:48:05 +0000 (11:48 -0400)]
Fix conditionals on mtrace-based tests (bug 31892)
The conditionals for several mtrace-based tests in catgets, elf, libio,
malloc, misc, nptl, posix, and stdio-common were incorrect leading to
test failures when bootstrapping glibc without perl.
The correct conditional for mtrace-based tests requires three checks:
first checking for run-built-tests, then build-shared, and lastly that
PERL is not equal to "no" (missing perl). Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
MayShao-oc [Sat, 29 Jun 2024 03:58:28 +0000 (11:58 +0800)]
x86: Set default non_temporal_threshold for Zhaoxin processors
Current 'non_temporal_threshold' set to 'non_temporal_threshold_lowbound'
on Zhaoxin processors without ERMS. The default
'non_temporal_threshold_lowbound' is too small for the KH-40000 and KX-7000
Zhaoxin processors, this patch updates the value to
'shared / cachesize_non_temporal_divisor'. Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
MayShao-oc [Sat, 29 Jun 2024 03:58:27 +0000 (11:58 +0800)]
x86_64: Optimize large size copy in memmove-ssse3
This patch optimizes large size copy using normal store when src > dst
and overlap. Make it the same as the logic in memmove-vec-unaligned-erms.S.
Current memmove-ssse3 use '__x86_shared_cache_size_half' as the non-
temporal threshold, this patch updates that value to
'__x86_shared_non_temporal_threshold'. Currently, the
__x86_shared_non_temporal_threshold is cpu-specific, and different CPUs
will have different values based on the related nt-benchmark results.
However, in memmove-ssse3, the nontemporal threshold uses
'__x86_shared_cache_size_half', which sounds unreasonable.
The performance is not changed drastically although shows overall
improvements without any major regressions or gains.
Results on Zhaoxin KX-7000:
bench-memcpy geometric_mean(N=20) New / Original: 0.999
bench-memcpy-random geometric_mean(N=20) New / Original: 0.999
bench-memcpy-large geometric_mean(N=20) New / Original: 0.978
bench-memmove geometric_mean(N=20) New / Original: 1.000
bench-memmmove-large geometric_mean(N=20) New / Original: 0.962
Results on Intel Core i5-6600K:
bench-memcpy geometric_mean(N=20) New / Original: 1.001
bench-memcpy-random geometric_mean(N=20) New / Original: 0.999
bench-memcpy-large geometric_mean(N=20) New / Original: 1.001
bench-memmove geometric_mean(N=20) New / Original: 0.995
bench-memmmove-large geometric_mean(N=20) New / Original: 0.936 Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
MayShao-oc [Sat, 29 Jun 2024 03:58:26 +0000 (11:58 +0800)]
x86: Set preferred CPU features on the KH-40000 and KX-7000 Zhaoxin processors
Fix code formatting under the Zhaoxin branch and add comments for
different Zhaoxin models.
Unaligned AVX load are slower on KH-40000 and KX-7000, so disable
the AVX_Fast_Unaligned_Load.
Enable Prefer_No_VZEROUPPER and Fast_Unaligned_Load features to
use sse2_unaligned version of memset,strcpy and strcat. Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
git://sourceware.org / glibc.git / log