Loading...
--- libmalloc/libmalloc-521.100.59/tests/threaded_stress.c
+++ /dev/null
@@ -1,343 +0,0 @@
-#include <stdlib.h>
-#include <stdio.h>
-#include <stdatomic.h>
-#include <math.h>
-#include <unistd.h>
-#include <sys/sysctl.h>
-#include <mach/mach.h>
-#include <pthread.h>
-#include <malloc/malloc.h>
-#include <darwintest.h>
-
-#include <../src/internal.h>
-
-// These tests are based on perf_contended_malloc_free, but intended as
-// functional stress tests rather than performance tests.
-
-T_GLOBAL_META(T_META_TAG_XZONE);
-
-// move the darwintest assertion code out of the straight line execution path
-// since it is has non-trivial overhead and codegen impact even if the assertion
-// is never triggered.
-#define iferr(_e) if(__builtin_expect(!!(_e), 0))
-
-#pragma mark -
-
-uint64_t
-random_busy_counts(unsigned int *seed, uint64_t *first, uint64_t *second)
-{
- uint64_t random = rand_r(seed);
- *first = 0x4 + (random & (0x10 - 1));
- random >>= 4;
- *second = 0x4 + (random & (0x10 - 1));
- random >>= 4;
- return random;
-}
-
-// By default busy() does no cpu busy work in the malloc bench
-enum {
- busy_is_nothing = 0,
- busy_is_cpu_busy,
- busy_is_cpu_yield,
-};
-static int busy_select = busy_is_nothing;
-
-static double
-cpu_busy(uint64_t n)
-{
- double d = M_PI;
- uint64_t i;
- for (i = 0; i < n; i++) d *= M_PI;
- return d;
-}
-
-static double
-cpu_yield(uint64_t n)
-{
- uint64_t i;
- for (i = 0; i < n; i++) {
-#if defined(__arm__) || defined(__arm64__)
- asm volatile("yield");
-#elif defined(__x86_64__) || defined(__i386__)
- asm volatile("pause");
-#else
-#error Unrecognized architecture
-#endif
- }
- return 0;
-}
-
-__attribute__((noinline))
-static double
-busy(uint64_t n)
-{
- switch(busy_select) {
- case busy_is_cpu_busy:
- return cpu_busy(n);
- case busy_is_cpu_yield:
- return cpu_yield(n);
- default:
- return 0;
- }
-}
-
-static semaphore_t ready_sem, start_sem;
-static uint32_t nthreads;
-static _Atomic uint32_t active_thr;
-static _Atomic int64_t todo;
-
-static uint32_t
-ncpu(void)
-{
- static uint32_t activecpu, physicalcpu;
- if (!activecpu) {
- uint32_t n;
- size_t s = sizeof(n);
- sysctlbyname("hw.activecpu", &n, &s, NULL, 0);
- activecpu = n;
- s = sizeof(n);
- sysctlbyname("hw.physicalcpu", &n, &s, NULL, 0);
- physicalcpu = n;
- }
- return MIN(activecpu, physicalcpu);
-}
-
-static uint32_t live_allocations;
-static void **allocations;
-static size_t max_rand, min_size, incr_size;
-
-static void
-malloc_threaded_stress(bool singlethreaded, size_t from, size_t to, size_t incr,
- uint32_t live_allocations_count, uint64_t iterations,
- void *(*thread_fn)(void *))
-{
- kern_return_t kr;
- int r;
- int batch_size;
- char *e;
-
- if (singlethreaded) {
- nthreads = 1;
- } else {
- if ((e = getenv("THREADED_STRESS_NTHREADS"))) {
- nthreads = strtoul(e, NULL, 0);
- }
-
- if (nthreads < 2) {
- nthreads = ncpu();
- }
- }
- if ((e = getenv("THREADED_STRESS_CPU_BUSY"))) {
- busy_select = strtoul(e, NULL, 0);
- }
-
- atomic_init(&todo, iterations);
- atomic_init(&active_thr, nthreads);
-
- live_allocations = live_allocations_count;
- allocations = malloc(sizeof(allocations[0]) * live_allocations);
- T_QUIET; T_ASSERT_NOTNULL(allocations, "allocations array");
- incr_size = incr;
- min_size = from;
- max_rand = (to - from) / incr;
- assert((to - from) % incr == 0);
-
- kr = semaphore_create(mach_task_self(), &ready_sem, SYNC_POLICY_FIFO, 0);
- T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "semaphore_create");
- kr = semaphore_create(mach_task_self(), &start_sem, SYNC_POLICY_FIFO, 0);
- T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "semaphore_create");
-
- pthread_t threads[nthreads];
- for (int i = 0; i < nthreads; i++) {
- r = pthread_create(&threads[i], NULL, thread_fn,
- (void *)(uintptr_t)(i + 1));
- T_QUIET; T_ASSERT_POSIX_ZERO(r, "pthread_create");
- }
-
- for (int i = 0; i < nthreads; i++) {
- kr = semaphore_wait(ready_sem);
- iferr (kr) {T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "semaphore_wait");}
- }
-
- kr = semaphore_signal_all(start_sem);
- iferr (kr) {T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "semaphore_signal_all");}
-
- for (int i = 0; i < nthreads; i++) {
- r = pthread_join(threads[i], NULL);
- T_QUIET; T_ASSERT_POSIX_ZERO(r, "pthread_join");
- }
-}
-
-static void *
-malloc_size_stress_thread(void *arg)
-{
- kern_return_t kr;
- int r;
- unsigned int seed;
- volatile double dummy;
- uint64_t pos, remaining_frees;
- void *alloc;
-
- seed = (uintptr_t)arg; // each thread repeats its own sequence
- // start threads off in different positions in allocations array
- pos = (seed - 1) * (live_allocations / nthreads);
- remaining_frees = live_allocations;
- kr = semaphore_wait_signal(start_sem, ready_sem);
- T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "semaphore_wait_signal");
-
- while (1) {
- uint64_t first, second;
- uint64_t random = random_busy_counts(&seed, &first, &second);
- if (atomic_fetch_sub_explicit(&todo, 1, memory_order_relaxed) > 0) {
- dummy = busy(first);
- alloc = malloc(min_size + (random % (max_rand + 1)) * incr_size);
- iferr (!alloc) { T_ASSERT_POSIX_ZERO(errno, "malloc"); }
- } else {
- if (!remaining_frees--) break;
- alloc = NULL;
- }
- alloc = atomic_exchange(
- (_Atomic(void *) *)&allocations[(pos++)%live_allocations],
- alloc);
- if (alloc) {
- // Size once while allocated
- (void)malloc_size(alloc);
-
- dummy = busy(second);
- free(alloc);
-
- // Try again while (possibly) free
- malloc_size(alloc);
- }
- }
-
- atomic_fetch_sub_explicit(&active_thr, 1, memory_order_relaxed);
- return NULL;
-}
-
-T_DECL(threaded_stress_malloc_size_tiny,
- "multi-threaded stress test for tiny malloc_size")
-{
- uint64_t iterations = 2000000ull;
-#if TARGET_OS_TV || TARGET_OS_WATCH
- iterations = 200000ull;
-#endif // TARGET_OS_TV || TARGET_OS_WATCH
-
- malloc_threaded_stress(false, 16, 256, 16, 2048,
- iterations, malloc_size_stress_thread);
-}
-
-T_DECL(threaded_stress_malloc_size_small,
- "multi-threaded stress test for small malloc_size")
-{
- uint64_t iterations = 200000ull;
-#if TARGET_OS_TV || TARGET_OS_WATCH
- iterations = 20000ull;
-#endif // TARGET_OS_TV || TARGET_OS_WATCH
-
- malloc_threaded_stress(false, 2048, 8192, 2048, 64,
- iterations, malloc_size_stress_thread);
-}
-
-static void *
-malloc_fork_stress_thread(void *arg)
-{
- kern_return_t kr;
- int r;
- unsigned int seed;
- volatile double dummy;
- uint64_t pos, remaining_frees;
- void *alloc;
- bool parent = true;
- uint64_t children = 0;
-
- char *e;
- unsigned long fork_prob = 100000;
- if ((e = getenv("THREADED_STRESS_FORK_PROB"))) {
- unsigned long env_prob = strtoul(e, NULL, 0);
- if (env_prob) {
- fork_prob = env_prob;
- }
- }
-
- seed = (uintptr_t)arg; // each thread repeats its own sequence
- // start threads off in different positions in allocations array
- pos = (seed - 1) * (live_allocations / nthreads);
- remaining_frees = live_allocations;
- kr = semaphore_wait_signal(start_sem, ready_sem);
- T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "semaphore_wait_signal");
-
- while (1) {
- uint64_t first, second;
- uint64_t random = random_busy_counts(&seed, &first, &second);
- if (parent && (random % fork_prob) == 0) {
- pid_t pid = fork();
- if (pid == -1) {
- if (errno != EAGAIN) {
- T_ASSERT_POSIX_SUCCESS(pid, "fork()");
- }
- } else if (pid == 0) {
- parent = false;
- } else {
- children++;
- }
- }
-
- if (atomic_fetch_sub_explicit(&todo, 1, memory_order_relaxed) > 0) {
- dummy = busy(first);
- alloc = malloc(min_size + (random % (max_rand + 1)) * incr_size);
- iferr (!alloc) { T_ASSERT_POSIX_ZERO(errno, "malloc"); }
- memset(alloc, 'a', 16);
- } else {
- if (!remaining_frees--) break;
- alloc = NULL;
- }
- alloc = atomic_exchange(
- (_Atomic(void *) *)&allocations[(pos++)%live_allocations],
- alloc);
- if (alloc) {
- dummy = busy(second);
- free(alloc);
- }
- }
-
- if (parent) {
- for (uint64_t i = 0; i < children; i++) {
- int status = 0;
- pid_t child = wait(&status);
- if (child == -1) {
- T_ASSERT_POSIX_SUCCESS(child, "wait()");
- }
- T_QUIET; T_ASSERT_TRUE(WIFEXITED(status), "child exited");
- T_QUIET; T_ASSERT_EQ(WEXITSTATUS(status), 0, "child succeeded");
- }
- }
-
- atomic_fetch_sub_explicit(&active_thr, 1, memory_order_relaxed);
- return NULL;
-}
-
-T_DECL(threaded_stress_fork, "multi-threaded stress test for fork",
- T_META_ENVVAR("MallocNanoZone=0")) // rdar://118860589
-{
- uint64_t iterations = 2000000ull;
-#if TARGET_OS_TV || TARGET_OS_WATCH
- iterations = 200000ull;
-#endif // TARGET_OS_TV || TARGET_OS_WATCH
-
- malloc_threaded_stress(false, 16, 256, 16, 2048,
- iterations, malloc_fork_stress_thread);
-}
-
-T_DECL(threaded_stress_fork_small,
- "multi-threaded stress test of small for fork",
- T_META_ENVVAR("MallocNanoZone=0")) // rdar://118860589
-{
- uint64_t iterations = 200000ull;
-#if TARGET_OS_TV || TARGET_OS_WATCH
- iterations = 20000ull;
-#endif // TARGET_OS_TV || TARGET_OS_WATCH
-
- malloc_threaded_stress(false, 2048, 8192, 2048, 64,
- iterations, malloc_fork_stress_thread);
-}