The latest implementations of memcpy are actually faster than the Falkor
implementations [1], so remove the falkor/phecda ifuncs for memcpy and
the now unused IS_FALKOR/IS_PHECDA defines.
[1] https://sourceware.org/pipermail/libc-alpha/2022-December/144227.html
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
(cherry picked from commit 2f5524cc53)
Add a specialized memset for the common ZVA size of 64 to avoid the
overhead of reading the ZVA size. Since the code is identical to
__memset_falkor, remove the latter.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
(cherry picked from commit 3d7090f14b)
Cleanup emag memset - merge the memset_base64.S file, remove
the unused ZVA code (since it is disabled on emag).
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
(cherry picked from commit 9627ab99b5)
Cleanup ifuncs. Remove uses of libc_hidden_builtin_def, use ENTRY rather than
ENTRY_ALIGN, remove unnecessary defines and conditional compilation. Rename
strlen_mte to strlen_generic. Remove rtld-memset.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 9fd3409842)
Add support for MOPS in cpu_features and INIT_ARCH. Add ifuncs using MOPS for
memcpy, memmove and memset (use .inst for now so it works with all binutils
versions without needing complex configure and conditional compilation).
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 2bd0017988)
Linux 6.5 adds a new AArch64 HWCAP2 value, HWCAP2_MOPS. Add it to
glibc's bits/hwcap.h.
Tested with build-many-glibcs.py for aarch64-linux-gnu.
(cherry picked from commit ff5d2abd18)
Improve SVE memcpy by copying 2 vectors if the size is small enough.
This improves performance of random memcpy by ~9% on Neoverse V1, and
33-64 byte copies are ~16% faster.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit d2d3f3720c)
Use shrn for narrowing the mask which simplifies code and speeds up small
strings. Unroll the first search loop to improve performance on large
strings.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 55599d4804)
Optimize strnlen using the shrn instruction and improve the main loop.
Small strings are around 10% faster, large strings are 40% faster on
modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit ad098893ba)
Optimize strlen by unrolling the main loop. Large strings are 64% faster on
modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 03c8ce5000)
Unroll the main loop. Large strings are around 20% faster on modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 349e48c01e)
Simplify calculation of the mask using shrn. Unroll the main loop.
Small strings are 20% faster on modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 51541a2297)
Use shrn for the mask, merge tst+bne into cbnz, and tweak code alignment.
Performance improves slightly as a result.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 1bbb1a2022)
Optimize the main loop - large strings are 43% faster on modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 0077624177)
Optimize the main loop - large strings are 40% faster on modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit ce758d4f06)
Since __memcpy_simd is the fastest memcpy on almost all cores, replace
the generic memcpy with it. If SVE is available, a SVE memcpy will be
used by default (including for Neoverse N2).
(cherry picked from commit e6f3fe362f)
We found that string functions were using AND+ADDP
to find the nibble/syndrome mask but there is an easier
opportunity through `SHRN dst.8b, src.8h, 4` (shift
right every 2 bytes by 4 and narrow to 1 byte) and has
same latency on all SIMD ARMv8 targets as ADDP. There
are also possible gaps for memcmp but that's for
another patch.
We see 10-20% savings for small-mid size cases (<=128)
which are primary cases for general workloads.
(cherry picked from commit 3c99806989)
Add an initial SVE memcpy implementation. Copies up to 32 bytes use SVE
vectors which improves the random memcpy benchmark significantly.
Cleanup the memcpy and memmove ifunc selectors.
(cherry picked from commit 9f298bfe1f)
Rewrite memcmp to improve performance. On small and medium inputs performance
is 10-20% better. Large inputs use a SIMD loop processing 64 bytes per
iteration, which is 30-50% faster depending on the size.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit b51eb35c57)
This restore the 2.33 semantic for arena_get2. It was changed by
11a02b035b to avoid arena_get2 call malloc (back when __get_nproc
was refactored to use an scratch_buffer - 903bc7dcc2). The
__get_nproc was refactored over then and now it also avoid to call
malloc.
The 11a02b035b did not take in consideration any performance
implication, which should have been discussed properly. The
__get_nprocs_sched is still used as a fallback mechanism if procfs
and sysfs is not acessible.
Checked on x86_64-linux-gnu.
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit 472894d2cf)
Ffsll function randomly regress by ~20%, depending on how code gets
aligned in memory. Ffsll function code size is 17 bytes. Since default
function alignment is 16 bytes, it can load on 16, 32, 48 or 64 bytes
aligned memory. When ffsll function load at 16, 32 or 64 bytes aligned
memory, entire code fits in single 64 bytes cache line. When ffsll
function load at 48 bytes aligned memory, it splits in two cache line,
hence random regression.
Ffsll function size reduction from 17 bytes to 12 bytes ensures that it
will always fit in single 64 bytes cache line.
This patch fixes ffsll function random performance regression.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
(cherry picked from commit 9d94997b5f)
_dl_tlsdesc_undefweak and _dl_tlsdesc_dynamic access the thread pointer
via the tcb field in TCB:
_dl_tlsdesc_undefweak:
_CET_ENDBR
movq 8(%rax), %rax
subq %fs:0, %rax
ret
_dl_tlsdesc_dynamic:
...
subq %fs:0, %rax
movq -8(%rsp), %rdi
ret
Since the tcb field in TCB is a pointer, %fs:0 is a 32-bit location,
not 64-bit. It should use "sub %fs:0, %RAX_LP" instead. Since
_dl_tlsdesc_undefweak returns ptrdiff_t and _dl_make_tlsdesc_dynamic
returns void *, RAX_LP is appropriate here for x32 and x86-64. This
fixes BZ #31185.
(cherry picked from commit 81be2a61da)
On x32, I got
FAIL: elf/tst-tlsgap
$ gdb elf/tst-tlsgap
...
open tst-tlsgap-mod1.so
Thread 2 "tst-tlsgap" received signal SIGSEGV, Segmentation fault.
[Switching to LWP 2268754]
_dl_tlsdesc_dynamic () at ../sysdeps/x86_64/dl-tlsdesc.S:108
108 movq (%rsi), %rax
(gdb) p/x $rsi
$4 = 0xf7dbf9005655fb18
(gdb)
This is caused by
_dl_tlsdesc_dynamic:
_CET_ENDBR
/* Preserve call-clobbered registers that we modify.
We need two scratch regs anyway. */
movq %rsi, -16(%rsp)
movq %fs:DTV_OFFSET, %rsi
Since the dtv field in TCB is a pointer, %fs:DTV_OFFSET is a 32-bit
location, not 64-bit. Load the dtv field to RSI_LP instead of rsi.
This fixes BZ #31184.
(cherry picked from commit 3502440397)
This patch fixes a very recently added leak in getaddrinfo.
This was assigned CVE-2023-5156.
Resolves: BZ #30884
Related: BZ #30842
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
(cherry picked from commit ec6b95c330)
When an NSS plugin only implements the _gethostbyname2_r and
_getcanonname_r callbacks, getaddrinfo could use memory that was freed
during tmpbuf resizing, through h_name in a previous query response.
The backing store for res->at->name when doing a query with
gethostbyname3_r or gethostbyname2_r is tmpbuf, which is reallocated in
gethosts during the query. For AF_INET6 lookup with AI_ALL |
AI_V4MAPPED, gethosts gets called twice, once for a v6 lookup and second
for a v4 lookup. In this case, if the first call reallocates tmpbuf
enough number of times, resulting in a malloc, th->h_name (that
res->at->name refers to) ends up on a heap allocated storage in tmpbuf.
Now if the second call to gethosts also causes the plugin callback to
return NSS_STATUS_TRYAGAIN, tmpbuf will get freed, resulting in a UAF
reference in res->at->name. This then gets dereferenced in the
getcanonname_r plugin call, resulting in the use after free.
Fix this by copying h_name over and freeing it at the end. This
resolves BZ #30843, which is assigned CVE-2023-4806.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
(cherry picked from commit 973fe93a56)
All other cases of failures due to lack of memory return EAI_MEMORY, so
it seems wrong to return EAI_SYSTEM here. The only reason
convert_hostent_to_gaih_addrtuple could fail is on calloc failure.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit b587456c0e)
Simplify the loop a wee bit and clean up variable names too.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit ac4653ef50)
Flatten the condition nesting and replace the alloca for RET.AT/ATR with
a single array LOCAL_AT[2]. This gets rid of alloca and alloca
accounting.
`git diff -b` is probably the best way to view this change since much of
the diff is whitespace changes.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit 657472b2a5)
The macro is quite a pain to debug, so make gethosts into a function to
make it easier to maintain.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit cfa3bd48cb)
Add a new member got_ipv6 to indicate if the results have an IPv6
result and use it instead of the local got_ipv6.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit e7e5315b7f)
Add a free_at flag in gaih_result to indicate if res.at needs to be
freed by the caller.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit b44389cb7f)
Introduce the gaih_result structure and general paradigm for cleanups
that follow to process the lookup request and return a result. A lookup
function (like text_to_binary_address), should return an integer error
code and set members of gaih_result based on what it finds. If the
function does not have a result and no errors have occurred during the
lookup, it should return 0 and res.at should be set to NULL, allowing a
subsequent function to do the lookup until we run out of options.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit 26dea46119)
Refactor the code to split out the service resolution code into a
separate function. Allocate the service tuples array just once to the
size of the typeproto array, thus avoiding the unnecessary pointer
chasing and stack allocations.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit 8d6cf99f2f)
Use realloc in convert_hostent_to_gaih_addrtuple and fix up pointers in
the result list so that a single block is maintained for
hostbyname3_r/hostbyname2_r and freed in gaih_inet. This result is
never merged with any other results, since the hosts database does not
permit merging.
Resolves BZ #28852.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit 3004604607)
Simplify logic for allocation of canon to remove the canonbuf variable;
canon now always points to an allocated block. Also pull the canon name
set into a separate function.
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit d01411f6bc)
Allocations for address tuples is currently a bit confusing because of
the pointer chasing through PAT, making it hard to observe the sequence
in which allocations have been made. Narrow scope of the pointer
chasing through PAT so that it is only used where necessary.
This also tightens actions behaviour with the hosts database in
getaddrinfo to comply with the manual text. The "continue" action
discards previous results and the "merge" action results in an immedate
lookup failure. Consequently, chaining of allocations across modules is
no longer necessary, thus opening up cleanup opportunities.
A test has been added that checks some combinations to ensure that they
work correctly.
Resolves: BZ #28931
Signed-off-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit 1c37b8022e)
The:
```
if (shared_per_thread > 0 && threads > 0)
shared_per_thread /= threads;
```
Code was accidentally moved to inside the else scope. This doesn't
match how it was previously (before af992e7abd).
This patch fixes that by putting the division after the `else` block.
(cherry picked from commit 084fb31bc2)
On some machines we end up with incomplete cache information. This can
make the new calculation of `sizeof(total-L3)/custom-divisor` end up
lower than intended (and lower than the prior value). So reintroduce
the old bound as a lower bound to avoid potentially regressing code
where we don't have complete information to make the decision.
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit 8b9a0af8ca)
After:
```
commit af992e7abd
Author: Noah Goldstein <goldstein.w.n@gmail.com>
Date: Wed Jun 7 13:18:01 2023 -0500
x86: Increase `non_temporal_threshold` to roughly `sizeof_L3 / 4`
```
Split `shared` (cumulative cache size) from `shared_per_thread` (cache
size per socket), the `shared_per_thread` *can* be slightly off from
the previous calculation.
Previously we added `core` even if `threads_l2` was invalid, and only
used `threads_l2` to divide `core` if it was present. The changed
version only included `core` if `threads_l2` was valid.
This change restores the old behavior if `threads_l2` is invalid by
adding the entire value of `core`.
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit 47f7472178)
Current `non_temporal_threshold` set to roughly '3/4 * sizeof_L3 /
ncores_per_socket'. This patch updates that value to roughly
'sizeof_L3 / 4`
The original value (specifically dividing the `ncores_per_socket`) was
done to limit the amount of other threads' data a `memcpy`/`memset`
could evict.
Dividing by 'ncores_per_socket', however leads to exceedingly low
non-temporal thresholds and leads to using non-temporal stores in
cases where REP MOVSB is multiple times faster.
Furthermore, non-temporal stores are written directly to main memory
so using it at a size much smaller than L3 can place soon to be
accessed data much further away than it otherwise could be. As well,
modern machines are able to detect streaming patterns (especially if
REP MOVSB is used) and provide LRU hints to the memory subsystem. This
in affect caps the total amount of eviction at 1/cache_associativity,
far below meaningfully thrashing the entire cache.
As best I can tell, the benchmarks that lead this small threshold
where done comparing non-temporal stores versus standard cacheable
stores. A better comparison (linked below) is to be REP MOVSB which,
on the measure systems, is nearly 2x faster than non-temporal stores
at the low-end of the previous threshold, and within 10% for over
100MB copies (well past even the current threshold). In cases with a
low number of threads competing for bandwidth, REP MOVSB is ~2x faster
up to `sizeof_L3`.
The divisor of `4` is a somewhat arbitrary value. From benchmarks it
seems Skylake and Icelake both prefer a divisor of `2`, but older CPUs
such as Broadwell prefer something closer to `8`. This patch is meant
to be followed up by another one to make the divisor cpu-specific, but
in the meantime (and for easier backporting), this patch settles on
`4` as a middle-ground.
Benchmarks comparing non-temporal stores, REP MOVSB, and cacheable
stores where done using:
https://github.com/goldsteinn/memcpy-nt-benchmarks
Sheets results (also available in pdf on the github):
https://docs.google.com/spreadsheets/d/e/2PACX-1vS183r0rW_jRX6tG_E90m9qVuFiMbRIJvi5VAE8yYOvEOIEEc3aSNuEsrFbuXw5c3nGboxMmrupZD7K/pubhtml
Reviewed-by: DJ Delorie <dj@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
(cherry picked from commit af992e7abd)
There are reports for hang in __check_pf:
https://github.com/JoeDog/siege/issues/4
It is reproducible only under specific configurations:
1. Large number of cores (>= 64) and large number of threads (> 3X of
the number of cores) with long lived socket connection.
2. Low power (frequency) mode.
3. Power management is enabled.
While holding lock, __check_pf calls make_request which calls __sendto
and __recvmsg. Since __sendto and __recvmsg are cancellation points,
lock held by __check_pf won't be released and can cause deadlock when
thread cancellation happens in __sendto or __recvmsg. Add a cancellation
cleanup handler for __check_pf to unlock the lock when cancelled by
another thread. This fixes BZ #20975 and the siege hang issue.
(cherry picked from commit a443bd3fb2)
Fix bug that SIGCHLD is erroneously blocked forever in the following
scenario:
1. Thread A calls system but hasn't returned yet
2. Thread B calls another system but returns
SIGCHLD would be blocked forever in thread B after its system() returns,
even after the system() in thread A returns.
Although POSIX does not require, glibc system implementation aims to be
thread and cancellation safe. This bug was introduced in
5fb7fc9635 when we moved reverting signal
mask to happen when the last concurrently running system returns,
despite that signal mask is per thread. This commit reverts this logic
and adds a test.
Signed-off-by: Adam Yi <ayi@janestreet.com>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
(cherry picked from commit 436a604b7d)
The divss instruction clobbers its first argument, and the constraints
need to reflect that. Fortunately, with GCC 12, generated code does
not actually change, so there is no externally visible bug.
Suggested-by: Jakub Jelinek <jakub@redhat.com>
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
(cherry picked from commit 5d1ccdda7b)
The minimum non_temporal_threshold is 0x4040. non_temporal_threshold may
be set to less than the minimum value when the shared cache size isn't
available (e.g., in an emulator) or by the tunable. Add checks for
minimum and maximum of non_temporal_threshold.
This fixes BZ #29953.
(cherry picked from commit 48b74865c6)
