ColinIanKing/stress-ng V0.17.02

on GitHub

Stress-ng V0.17.02 "omniferous optimized overreacher"

Key features:

• many stressor related optimisations:
• ~5.4% improvement on Alderlake i9-12900
• ~5.2% improvement on 24 core ARM® Cortex-A53
• ~2.2% improvement on StarFive JH7100 64 bit RISC-V RV64GC
• ~7.2% improvement on Libre Computer AML-A311D-CC (4 x ARM Cortex-A72 cores, 2 x Cortex-A53 cores)
• ~5.2% improvement on Libre Computer AML-S905D3-CC (4 x ARMv8 Cortex-A55)
  (based on Geometric mean of bogo-ops throughput of over 300 stressors)

New features:

• sigxfsz and sigxcpu stressors for SIGXFSZ and SIGXCPU signals
• bsearch, hsearch, lsearch have new method options to specify libc or optimized non-libc search functions
• io-uring stressor has new --io-uring-entries option to specify number of entries in the io-uring

Changelog:

[Sascha Hauer]

• stress-physpage: explicitly use 64bit type for physical addresses
• stress-ng.h: define _FILE_OFFSET_BITS before including features.h

[Rulin Huang]

• Reduce the file operations if the unused pid can be found within
  several tries

[Colin Ian King]

• stress-bigheap: ensure value in *uintptr is initialized
• stress-workload: clean up some clang-scan-build warnings
• stress-workload: remove redundant assignments to variable t
• stress-workload: remove redundant initialization of t_begin
• stress-sigpipe: add wrapper to pipe_child to clear up tcc warning
• stress-reboot: add wrapper to reboot_clone_func to clear up tcc
  warning
• stress-vma: remove volatile cast, it is unnecessary
• stress-lsearch: add casts to const char * to avoid void * arithmetic
• stress-bsearch: cast base to const char * to avoid void * arithmetic
• core-time: return -1 if time not available via
  stress_time_now_timespec
• stress-msync: make pointer buf unclobberable
• stress-msg: remove another invalid msgrcv calls that work for
  OpenBSD systems
• stress-msg: remove invalid msgrcv calls that work for *BSD systems
• core-shim: Fix non-PURE shim_enosys function
• core-resources: include core-pthread.h for pthread types
• stress-get: don't call gettimeofday with NULL values
• core-shim/stress-get: use gettimeofday() for non-linux shim_time calls
• core-shim: fix typo: tlock -> tloc
• core-shim/stress-get: use time() for non-linux shim_time calls
• stress-get: fix error messages, replace gettimeval with gettimeofday
• stress-get: only exercise time(NULL) on Linux
• core-helper: move ret inside #ifdef block
• README.md: Update contributor list
• stress-*: clean up a handful of spelling mistakes
• stress-stream: unmap in correct idx order
• README.md: add another link to a research paper
• stress-prctl: munmap page_anon on error return path
• stress-seal: unmap buf on error return paths
• stress-cyclic: unmap latencies mappings on error return
• stress-zero: unmap read/write buffers on error return paths
• stress-stream: simplify error unmapping return paths
• stress-rawdev: fix memory leak on mmap'd buffer on error return path
• stress-fail: use fail error exit path for non-implemented /dev/full
• stress-sockmany: fix memory leak on signal handler error return path
• stress-signest: fix memory leak on signal handler error return path
• stress-sock: fix memory leak on fork failure error return path
• stress-sigpipe: fix memory leak on signal handler error return path
• stress-stackmmap: make stack_sig non-clobberable
• stress-sigbus: make ptr non-clobberable
• core-helper: add stress_mmap_populate to mmap with MAP_POPULATE
• stress-qsort: add missing tab
• stress-bseach: add -O3 for nonlibc bsearch function
• stress-waitcpu: add some more nops
• stress-vm: remove volaile where it's not required, add memory barrier
• stress-vm: speed up prime-gray-0 and prime-gray-1 by reducing total
  memory read/writes
• stress-vm: speed up prime-0 and prime-1 by reducing total memory
  read/writes
• stress-vma: provide different handlers for SIGBUS and SIGSEGV
• stress-vma: optimize matrics, add cacheline pad and remove volatile
• stress-vdso: add dummy function call to emulate minimal vdso function
• stress-udp: don't memset the send buffer on each send
• stress-trig: replace iterations with macro STRESS_TRIG_LOOPS
• stress-trig: unroll loops, speed up by ~4.2%
• stress-memthrash: only report lack of NUMA support with instance 0
• stress-schedmix: reduce compute overhead on prime number CPU load
• stress-stack: unroll loop in stress_stack_alloc
• stress-sockpair: optimize buffer setting and checking
• stress-sockdiag: move bogo-op count to outside parser loop
• stress-skiplist: use 8 bits for skip_list_random_level generation
• stress-skiplist: optimize ln2 operation using clzl
• stress-sigxfs: optimize counter increment, remove branch
• stress-sigxfsz: don't make async_sigs volatile
• core-helper: only declare array buf when it is required
• core-helper: voidify function argument dumpable
• core-helper: only declare array buf when it is required
• core-shim: move include <sys/uio.h> to core-shim.h
• stress-gpu: move declarations for gpu_freq_sum and gpu_freq_count
• core-numa: add stress_numa_nodes helper for non-linux builds
• stress-stream: don't print numa nodes related info for non-linux
  systems
• core-shim: add missing include of sys/uio.h required for struct iovec
• stress-mmapfixed: fix array overflow, iterate over only the array
  elements
• stress-gpu: ensure usec computations are using uint64_t type
• stress-hsearch: voidify return from stress_get_setting
• core-helper: use STRESS_MAXIMUM on maximum pid
• stress-vma: add in extra protection flags and mmap flags for more
  stress
• stress-vma: add PROT_NONE back
• stress-vma: ensure page is always mappable
• stress-ng: move ci counter struct into args and args into stat struct
• stress-signal: don't use volatile for in-signal handler counter
  increment
• stress-ring-pipe: replace % operator with compare and set
• stress-resched: remove need to check for non-mapped yields array
• README.md: add more research paper links
• stress-remap: unroll a handful of loops
• stress-regs: reference registers rather than v variable
• stress-readahead: unroll some loops
• stress-randlist: add prefetch to next pointer in list scanning
• core-mwc: use clz for shift maximization for 8 and 16 bit mwci
  functions
• core-sort: optimize of powers of 2 size for
  stress_sort_data_int32_shuffle
• stress-gpu: only report GPU frequency if > 0.0 MHz
• core-time: optimize get_time_now()
• stress-procfs: use MAP_POPULATE on mmap to reduce read-fault hit
• stress-procfs: perform less timeout checks on 1 byte reads and
  optimize timeout paths
• stress-poll: switch order of read return checks
• stress-opcode: inline and unroll loop in stress_opcode_random
• stress-null: scale down mmaps to 1 once every 500 iterations
• stress-nice: add ( ) to clarify precedence
• stress-syncload: add yield delay for ppc
• core-builtin.h: Add llabs builtin shim wrapper
• core-ops: add in missing OPT_hsearch_method
• stress-nanosleep: ensure no overflow occurs on 32 bit systems
• stress-hsearch: add hsearch method for libc and non-libc versions
• stress-mmaphuge: simplyfy the set and check memory exercising
• stress-mmapfixed: make vec array uint32 to speed up mincore checking
• stress-mmapfixed: move memset in stress_mmapfixed_is_mapped_slow
• core-helper: add missing space between ) and {
• core-helper: stress_process_dumpable: use stress_system_write
  to write data
• core-helper: stress_get_memlimits: use stress_system_read to read data
• stress-lsearch: use lfind_nonlibc in lsearch_nonlibc
• stress-lsearch: add lsearch method for libc and non-libc versions
• stress-bsearch: add --bsearch-method short help
• stress-lockofd: remove yield point, it's not helping that much
• stress-lockf: remove yield point, it's not helping that much
• stress-locka: remove yield point, it's not helping that much
• Makefile.config: echo the number of configurations in config.h
• stress-jpeg: compute gradient image using float instead of double
• stress-io-uring: add UNLIKELY hint on error return check
• stress-iomix: stress-_iomix_rd_wr_mmap: just read 1 byte from page
• core-mwc: move stress_rndstr from core-helper to core-mwc
• stress-icache: add missing space before + operator
• stress-hdd: optimize hdd_fill_buf
• core-mwc: move stress_rndbuf from core-helper to core-mwc
• stress-gpu: add GPU frequency metrics based on 10 samples per second
• stress-ng: emit plural of instance if instances > 1
• core-config-check: replace "will" with "may"
• stress-getrandom: reduce overhead of gettimeofday calls
• stress-full: fix and optimize stress_data_is_not_zero
• core-mwc: optimize modulo mwc32 and mwc64 functions
• stress-fifo: add poll replacement to select, disable it for now
• stress-far-branch: inline the branch shuffling, reduce iterations
  from 5 to 1
• stress-epoll: don't fill send buffer on each send
• core-time: make function stress_timeval_to_double PURE
• core-helper: add PURE annotation to two more helper functions
• core-helper: optimize stress_rndstr
• stress-clock: move memsets outside of loops for minor performance
  improvement
• stress-bsearch: allocate bsearch array using mmap
• stress-ng: fix short help for change-cpu, fix missing -
• stress-bsearch: add bsearch method for libc and non-libc versions
• stress-qsort: add in missing qsort-method short help
• stress-ng: don't timeout until all stressors are spawned
• core-helper: fix low memory swap check
• stress-sigxfsz: remove unused include files
• stress-sigxcpu: Add new SIGXCPU stressor
• stress-sig*: add new class "signal", add signal stressors to this
  class
• stress-cpu-online: write the string length rather than buffer size
• core-helper: make inline helper function stress_chr_munge PURE
• stress-pthread: constify variable trun
• stress-pthread: use same racy pid for tgid rather than fetching one
  3 times
• README.md: Update contributors name list
• stress-sigxfsz: Add new SIGXFSZ stressor
• README.md: add some more kernel issues found with stress-ng
• core-cpuidle: report C states with commas between each state
• stress-sysbadaddr: add missing #include <termios.h>
• Add in some missing <sched.h> includes
• fix up build errors from unchecked in changes
• stress-ng.h: move #include <sched.h> to sources that require it
• stress-af-alg: add in missing #include <sys/socket.h>
• README.md: add another citation link
• stress-ng.h: move #include <pthread.h> to core-pthread.h
• stress-ng.h: move #include <termios.h> to sources that require it
• stress-ng.h: move #include <pwd.h> to sources that require it
• stress-ng.h: remove #include <getopt.h>
• stress-ng.h: move #include <sys/socket.h> to sources that require it
• core-helper: move MEM_CACHE_SIZE from stress-ng.h to core-helper.c
• stress-stream: decouple the default guess cache size from
  MEM_CACHE_SIZE
• core-attribute: add extra checks for attribute noinline
• core-attribute: add extra checks for attribute always_inline
• core-attribute: add extra checks for attribute weak
• core-attribute: add extra checks for attribute noreturn
• core-attribute: add extra checks for attribute warn_unused_result
• core-attribute: add extra build checks for hot attribute
• core-attribute: add PURE macro for gcc pure attribute
• stress-crypt: use the correct size to strlcpy to
• stress-cpu-online: report the sysfs file when writes fail
• core-helper: take into consideration NUMA nodes for full cache size
• stress-msync: mark fd as NOCLOBBER, clean up gcc 13.2.1 warning
• Makefile: remove gcc optimization flags
• Makefile: fix gcc detection for extra gcc optimization flags
• stress-memthrash: add a buffer 8 bit reverse method
• stress-malloc: use valloc in non-thread scenarios
• stress-malloc: add valloc (or valloc emulation) for more calls to
  exercise
• stress-hsearch: add -O3 optimization
• stress-hsearch: move repeated verify flag check to a const bool
• stress-fma: improve fma buffer copying
• core-sort: minor re-ordering of load/stores and compute for minor
  speed improvement
• core-sort: minor optimization in stress_sort_data_int32_init
• stress-zlib: improve rarely0 and rarely1 performance
• stress-ng: clean up some more whitespace / tab issues
• stress-ng: replace spaces with tab
• stress-zlib: implement rdrand filler for non-x86 systems
• stress-zlib: improve performance of rdrand filling of buffer
• stress-tree: remove the UNLIKELY hint in avl_find
• stress-tree: remove the UNLIKELY hint in binary_find
• stress-tree: remove TARGET_CLONES from binary tree find/insert
• stress-stream: add TARGET_CLONES for index0 level streaming
• core-nt-store.h: use a union cast for stress_nt_store_double store
• stress-prefetch: clean up some source white space/tabs
• stress-io-uring: tweak number of io-uring entries depending on CPU
  count
• stress-io-uring: add --io-uring-entries N option to specify ring size
• Revert "stress-context: avoid using alternative stacks in a
  swapcontext"
• stress-fork: replace slow mwc8 with fair round-robin switch
  selection index
• stress-hash: don't call expensive mwc reseed for non-verify mode
• stress-branch: load next goto address in previous loop
• stress-branch: re-order branching code to compute idx before jmp
• test/test-crypt-r: check for support of GNU libcrypt data fields
• stress-ng: Fix stupid extraneous {
• Makefile.config: remove some trailing spaces
• stress-*: improve rate metrics by using harmonic mean
• kernel-coverage: remove scheduler names
• stress-cpu: replace 64 bit mwc and 7 shifts with 2 x 32 bit mwc and
  6 shifts
• stress-crypt: remove redundant prefix initialization
• stress-crypt: remove a few more args from stress_crypt_id call
• stress-crypt: optimize crypt_r usage with less memory buffer copying
• stress-memfd: store and read data in cacheline strides
• kernel-coverage: add bcachefs
• stress-metamix: remove empty line
• stress-metamix: handle EINTR for fdatasync and fsync
• stress-iomix: break out of stressor loops on EINTR for read or writes
• core-helper: add bcachefs magic