20 Commits

Author	SHA1	Message	Date
H.J. Lu	543efedcb3	x86: Check the lower byte of EAX of CPUID leaf 2 [BZ #30643] ... The old Intel software developer manual specified that the low byte of EAX of CPUID leaf 2 returned 1 which indicated the number of rounds of CPUDID leaf 2 was needed to retrieve the complete cache information. The newer Intel manual has been changed to that it should always return 1 and be ignored. If the lower byte isn't 1, CPUID leaf 2 can't be used. In this case, we ignore CPUID leaf 2 and use CPUID leaf 4 instead. If CPUID leaf 4 doesn't contain the cache information, cache information isn't available at all. This addresses BZ #30643. (cherry picked from commit 1493622f4f)	2025-01-10 12:09:58 -08:00
Florian Weimer	d626c31ce5	x86: Avoid integer truncation with large cache sizes (bug 32470) ... Some hypervisors report 1 TiB L3 cache size. This results in some variables incorrectly getting zeroed, causing crashes in memcpy/memmove because invariants are violated. (cherry picked from commit 61c3450db9)	2024-12-17 19:14:43 +01:00
Noah Goldstein	561e9dadc0	x86: Fix incorrect scope of setting shared_per_thread [BZ# 30745] ... 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)	2023-09-05 17:17:25 -05:00
Noah Goldstein	1c3ecf5858	x86: Use 3/4*sizeof(per-thread-L3) as low bound for NT threshold. ... 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)	2023-09-05 17:17:25 -05:00
Noah Goldstein	47c7d2eb03	x86: Fix slight bug in shared_per_thread cache size calculation. ... 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)	2023-09-05 17:17:25 -05:00
Noah Goldstein	d1b1da26ea	x86: Increase non_temporal_threshold to roughly sizeof_L3 / 4 ... 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)	2023-09-05 17:17:25 -05:00
H.J. Lu	b7cc55a24e	x86: Check minimum/maximum of non_temporal_threshold [BZ #29953] ... 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)	2023-04-20 15:49:52 +02:00
Noah Goldstein	863987a6ef	x86: Add bounds x86_non_temporal_threshold ... The lower-bound (16448) and upper-bound (SIZE_MAX / 16) are assumed by memmove-vec-unaligned-erms. The lower-bound is needed because memmove-vec-unaligned-erms unrolls the loop aggressively in the L(large_memset_4x) case. The upper-bound is needed because memmove-vec-unaligned-erms right-shifts the value of `x86_non_temporal_threshold` by LOG_4X_MEMCPY_THRESH (4) which without a bound may overflow. The lack of lower-bound can be a correctness issue. The lack of upper-bound cannot. (cherry picked from commit b446822b6a)	2022-07-18 11:43:23 -07:00
Noah Goldstein	7f7a728b71	x86: Fix misordered logic for setting rep_movsb_stop_threshold ... Move the setting of `rep_movsb_stop_threshold` to after the tunables have been collected so that the `rep_movsb_stop_threshold` (which is used to redirect control flow to the non_temporal case) will use any user value for `non_temporal_threshold` (set using glibc.cpu.x86_non_temporal_threshold) (cherry picked from commit 0355915514)	2022-07-18 11:41:48 -07:00
Paul Eggert	581c785bf3	Update copyright dates with scripts/update-copyrights ... I used these shell commands: ../glibc/scripts/update-copyrights $PWD/../gnulib/build-aux/update-copyright (cd ../glibc && git commit -am"[this commit message]") and then ignored the output, which consisted lines saying "FOO: warning: copyright statement not found" for each of 7061 files FOO. I then removed trailing white space from math/tgmath.h, support/tst-support-open-dev-null-range.c, and sysdeps/x86_64/multiarch/strlen-vec.S, to work around the following obscure pre-commit check failure diagnostics from Savannah. I don't know why I run into these diagnostics whereas others evidently do not. remote: *** 912-#endif remote: *** 913: remote: *** 914- remote: *** error: lines with trailing whitespace found ... remote: *** error: sysdeps/unix/sysv/linux/statx_cp.c: trailing lines	2022-01-01 11:40:24 -08:00
Noah Goldstein	475b63702e	x86: Double size of ERMS rep_movsb_threshold in dl-cacheinfo.h ... No bug. This patch doubles the rep_movsb_threshold when using ERMS. Based on benchmarks the vector copy loop, especially now that it handles 4k aliasing, is better for these medium ranged. On Skylake with ERMS: Size, Align1, Align2, dst>src,(rep movsb) / (vec copy) 4096, 0, 0, 0, 0.975 4096, 0, 0, 1, 0.953 4096, 12, 0, 0, 0.969 4096, 12, 0, 1, 0.872 4096, 44, 0, 0, 0.979 4096, 44, 0, 1, 0.83 4096, 0, 12, 0, 1.006 4096, 0, 12, 1, 0.989 4096, 0, 44, 0, 0.739 4096, 0, 44, 1, 0.942 4096, 12, 12, 0, 1.009 4096, 12, 12, 1, 0.973 4096, 44, 44, 0, 0.791 4096, 44, 44, 1, 0.961 4096, 2048, 0, 0, 0.978 4096, 2048, 0, 1, 0.951 4096, 2060, 0, 0, 0.986 4096, 2060, 0, 1, 0.963 4096, 2048, 12, 0, 0.971 4096, 2048, 12, 1, 0.941 4096, 2060, 12, 0, 0.977 4096, 2060, 12, 1, 0.949 8192, 0, 0, 0, 0.85 8192, 0, 0, 1, 0.845 8192, 13, 0, 0, 0.937 8192, 13, 0, 1, 0.939 8192, 45, 0, 0, 0.932 8192, 45, 0, 1, 0.927 8192, 0, 13, 0, 0.621 8192, 0, 13, 1, 0.62 8192, 0, 45, 0, 0.53 8192, 0, 45, 1, 0.516 8192, 13, 13, 0, 0.664 8192, 13, 13, 1, 0.659 8192, 45, 45, 0, 0.593 8192, 45, 45, 1, 0.575 8192, 2048, 0, 0, 0.854 8192, 2048, 0, 1, 0.834 8192, 2061, 0, 0, 0.863 8192, 2061, 0, 1, 0.857 8192, 2048, 13, 0, 0.63 8192, 2048, 13, 1, 0.629 8192, 2061, 13, 0, 0.627 8192, 2061, 13, 1, 0.62 Signed-off-by: Noah Goldstein <goldstein.w.n@gmail.com> Reviewed-by: H.J. Lu <hjl.tools@gmail.com>	2021-11-06 16:18:08 -05:00
H.J. Lu	cf2c57526b	x86: Set rep_movsb_threshold to 2112 on processors with FSRM ... The glibc memcpy benchmark on Intel Core i7-1065G7 (Ice Lake) showed that REP MOVSB became faster after 2112 bytes: Vector Move REP MOVSB length=2112, align1=0, align2=0: 24.20 24.40 length=2112, align1=1, align2=0: 26.07 23.13 length=2112, align1=0, align2=1: 27.18 28.13 length=2112, align1=1, align2=1: 26.23 25.16 length=2176, align1=0, align2=0: 23.18 22.52 length=2176, align1=2, align2=0: 25.45 22.52 length=2176, align1=0, align2=2: 27.14 27.82 length=2176, align1=2, align2=2: 22.73 25.56 length=2240, align1=0, align2=0: 24.62 24.25 length=2240, align1=3, align2=0: 29.77 27.15 length=2240, align1=0, align2=3: 35.55 29.93 length=2240, align1=3, align2=3: 34.49 25.15 length=2304, align1=0, align2=0: 34.75 26.64 length=2304, align1=4, align2=0: 32.09 22.63 length=2304, align1=0, align2=4: 28.43 31.24 Use REP MOVSB for data size > 2112 bytes in memcpy on processors with fast short REP MOVSB (FSRM). * sysdeps/x86/dl-cacheinfo.h (dl_init_cacheinfo): Set rep_movsb_threshold to 2112 on processors with fast short REP MOVSB (FSRM).	2021-05-03 05:08:22 -07:00
H.J. Lu	f53ffc9b90	x86: Handle _SC_LEVEL1_ICACHE_LINESIZE [BZ #27444] ... commit 2d651eb926 Author: H.J. Lu <hjl.tools@gmail.com> Date: Fri Sep 18 07:55:14 2020 -0700 x86: Move x86 processor cache info to cpu_features missed _SC_LEVEL1_ICACHE_LINESIZE. 1. Add level1_icache_linesize to struct cpu_features. 2. Initialize level1_icache_linesize by calling handle_intel, handle_zhaoxin and handle_amd with _SC_LEVEL1_ICACHE_LINESIZE. 3. Return level1_icache_linesize for _SC_LEVEL1_ICACHE_LINESIZE. Reviewed-by: Carlos O'Donell <carlos@redhat.com>	2021-03-15 05:43:26 -07:00
Siddhesh Poyarekar	a1b8b06a55	x86: Use SIZE_MAX instead of (long int)-1 for tunable range value ... The tunable types are SIZE_T, so set the ranges to the correct maximum value, i.e. SIZE_MAX.	2021-02-10 19:08:33 +05:30
Siddhesh Poyarekar	61117bfa1b	tunables: Simplify TUNABLE_SET interface ... The TUNABLE_SET interface took a primitive C type argument, which resulted in inconsistent type conversions internally due to incorrect dereferencing of types, especialy on 32-bit architectures. This change simplifies the TUNABLE setting logic along with the interfaces. Now all numeric tunable values are stored as signed numbers in tunable_num_t, which is intmax_t. All calls to set tunables cast the input value to its primitive type and then to tunable_num_t for storage. This relies on gcc-specific (although I suspect other compilers woul also do the same) unsigned to signed integer conversion semantics, i.e. the bit pattern is conserved. The reverse conversion is guaranteed by the standard.	2021-02-10 19:08:33 +05:30
Sajan Karumanchi	6e02b3e932	x86: Adding an upper bound for Enhanced REP MOVSB. ... In the process of optimizing memcpy for AMD machines, we have found the vector move operations are outperforming enhanced REP MOVSB for data transfers above the L2 cache size on Zen3 architectures. To handle this use case, we are adding an upper bound parameter on enhanced REP MOVSB:'__x86_rep_movsb_stop_threshold'. As per large-bench results, we are configuring this parameter to the L2 cache size for AMD machines and applicable from Zen3 architecture supporting the ERMS feature. For architectures other than AMD, it is the computed value of non-temporal threshold parameter. Reviewed-by: Premachandra Mallappa <premachandra.mallappa@amd.com>	2021-02-02 12:42:15 +01:00
H.J. Lu	ff6d62e9ed	<sys/platform/x86.h>: Remove the C preprocessor magic ... In <sys/platform/x86.h>, define CPU features as enum instead of using the C preprocessor magic to make it easier to wrap this functionality in other languages. Move the C preprocessor magic to internal header for better GCC codegen when more than one features are checked in a single expression as in x86-64 dl-hwcaps-subdirs.c. 1. Rename COMMON_CPUID_INDEX_XXX to CPUID_INDEX_XXX. 2. Move CPUID_INDEX_MAX to sysdeps/x86/include/cpu-features.h. 3. Remove struct cpu_features and __x86_get_cpu_features from <sys/platform/x86.h>. 4. Add __x86_get_cpuid_feature_leaf to <sys/platform/x86.h> and put it in libc. 5. Make __get_cpu_features() private to glibc. 6. Replace __x86_get_cpu_features(N) with __get_cpu_features(). 7. Add _dl_x86_get_cpu_features to GLIBC_PRIVATE. 8. Use a single enum index for each CPU feature detection. 9. Pass the CPUID feature leaf to __x86_get_cpuid_feature_leaf. 10. Return zero struct cpuid_feature for the older glibc binary with a smaller CPUID_INDEX_MAX [BZ #27104]. 11. Inside glibc, use the C preprocessor magic so that cpu_features data can be loaded just once leading to more compact code for glibc. 256 bits are used for each CPUID leaf. Some leaves only contain a few features. We can add exceptions to such leaves. But it will increase code sizes and it is harder to provide backward/forward compatibilities when new features are added to such leaves in the future. When new leaves are added, _rtld_global_ro offsets will change which leads to race condition during in-place updates. We may avoid in-place updates by 1. Rename the old glibc. 2. Install the new glibc. 3. Remove the old glibc. NB: A function, __x86_get_cpuid_feature_leaf , is used to avoid the copy relocation issue with IFUNC resolver as shown in IFUNC resolver tests.	2021-01-21 05:58:17 -08:00
H.J. Lu	2d651eb926	x86: Move x86 processor cache info to cpu_features ... 1. Move x86 processor cache info to _dl_x86_cpu_features in ld.so. 2. Update tunable bounds with TUNABLE_SET_WITH_BOUNDS. 3. Move x86 cache info initialization to dl-cacheinfo.h and initialize x86 cache info in init_cpu_features (). 4. Put x86 cache info for libc in cacheinfo.h, which is included in libc-start.c in libc.a and is included in cacheinfo.c in libc.so. Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>	2021-01-14 11:38:45 -08:00
Paul Eggert	2b778ceb40	Update copyright dates with scripts/update-copyrights ... I used these shell commands: ../glibc/scripts/update-copyrights $PWD/../gnulib/build-aux/update-copyright (cd ../glibc && git commit -am"[this commit message]") and then ignored the output, which consisted lines saying "FOO: warning: copyright statement not found" for each of 6694 files FOO. I then removed trailing white space from benchtests/bench-pthread-locks.c and iconvdata/tst-iconv-big5-hkscs-to-2ucs4.c, to work around this diagnostic from Savannah: remote: *** pre-commit check failed ... remote: *** error: lines with trailing whitespace found remote: error: hook declined to update refs/heads/master	2021-01-02 12:17:34 -08:00
H.J. Lu	0f09154c64	x86: Initialize CPU info via IFUNC relocation [BZ 26203] ... X86 CPU features in ld.so are initialized by init_cpu_features, which is invoked by DL_PLATFORM_INIT from _dl_sysdep_start. But when ld.so is loaded by static executable, DL_PLATFORM_INIT is never called. Also x86 cache info in libc.o and libc.a is initialized by a constructor which may be called too late. Since some fields in _rtld_global_ro in ld.so are initialized by dynamic relocation, we can also initialize x86 CPU features in _rtld_global_ro in ld.so and cache info in libc.so by initializing dummy function pointers in ld.so and libc.so via IFUNC relocation. Key points: 1. IFUNC is always supported, independent of --enable-multi-arch or --disable-multi-arch. Linker generates IFUNC relocations from input IFUNC objects and ld.so performs IFUNC relocations. 2. There are no IFUNC dependencies in ld.so before dynamic relocation have been performed, 3. The x86 CPU features in ld.so is initialized by DL_PLATFORM_INIT in dynamic executable and by IFUNC relocation in dlopen in static executable. 4. The x86 cache info in libc.o is initialized by IFUNC relocation. 5. In libc.a, both x86 CPU features and cache info are initialized from ARCH_INIT_CPU_FEATURES, not by IFUNC relocation, before __libc_early_init is called. Note: _dl_x86_init_cpu_features can be called more than once from DL_PLATFORM_INIT and during relocation in ld.so.	2020-10-16 16:17:53 -07:00