]> git.saurik.com Git - apple/xnu.git/blobdiff - tests/contextswitch.c
xnu-4903.270.47.tar.gz
[apple/xnu.git] / tests / contextswitch.c
index 3969ead2ed4a516f012081d44bbb84dcb8260596..b2ec166249527a37ba7a14bfd8f6fbe7d719fe81 100644 (file)
 #include <darwintest.h>
 #include <stdatomic.h>
 
-#define MAX_THREADS    32
-#define SPIN_SECS      6
-#define THR_SPINNER_PRI        63
-#define THR_MANAGER_PRI        62
+#define MAX_THREADS     32
+#define SPIN_SECS       6
+#define THR_SPINNER_PRI 63
+#define THR_MANAGER_PRI 62
 #define WARMUP_ITERATIONS 100
 #define POWERCTRL_SUCCESS_STR "Factor1: 1.000000"
 
@@ -38,14 +38,14 @@ static _Atomic uint32_t keep_going = 1;
 static dt_stat_time_t s;
 
 static struct {
-    pthread_t thread;
-    bool measure_thread;
+       pthread_t thread;
+       bool measure_thread;
 } threads[MAX_THREADS];
 
-static uint64_t 
-nanos_to_abs(uint64_t nanos) 
-{ 
-    return nanos * timebase_info.denom / timebase_info.numer;
+static uint64_t
+nanos_to_abs(uint64_t nanos)
+{
+       return nanos * timebase_info.denom / timebase_info.numer;
 }
 
 extern char **environ;
@@ -53,137 +53,141 @@ extern char **environ;
 static void
 csw_perf_test_init(void)
 {
-    int spawn_ret, pid;
-    char *const clpcctrl_args[] = {"/usr/local/bin/clpcctrl", "-f", "5000", NULL};
-    spawn_ret = posix_spawn(&pid, clpcctrl_args[0], NULL, NULL, clpcctrl_args, environ);
-    waitpid(pid, &spawn_ret, 0);
+       int spawn_ret, pid;
+       char *const clpcctrl_args[] = {"/usr/local/bin/clpcctrl", "-f", "5000", NULL};
+       spawn_ret = posix_spawn(&pid, clpcctrl_args[0], NULL, NULL, clpcctrl_args, environ);
+       waitpid(pid, &spawn_ret, 0);
 }
 
 static void
 csw_perf_test_cleanup(void)
 {
-    int spawn_ret, pid;
-    char *const clpcctrl_args[] = {"/usr/local/bin/clpcctrl", "-d", NULL};
-    spawn_ret = posix_spawn(&pid, clpcctrl_args[0], NULL, NULL, clpcctrl_args, environ);
-    waitpid(pid, &spawn_ret, 0);
+       int spawn_ret, pid;
+       char *const clpcctrl_args[] = {"/usr/local/bin/clpcctrl", "-d", NULL};
+       spawn_ret = posix_spawn(&pid, clpcctrl_args[0], NULL, NULL, clpcctrl_args, environ);
+       waitpid(pid, &spawn_ret, 0);
 }
 
 static pthread_t
-create_thread(uint32_t thread_id, uint32_t priority, bool fixpri, 
-        void *(*start_routine)(void *))
+create_thread(uint32_t thread_id, uint32_t priority, bool fixpri,
+    void *(*start_routine)(void *))
 {
-    int rv;
-    pthread_t new_thread;
-    struct sched_param param = { .sched_priority = (int)priority };
-    pthread_attr_t attr;
+       int rv;
+       pthread_t new_thread;
+       struct sched_param param = { .sched_priority = (int)priority };
+       pthread_attr_t attr;
 
-    T_ASSERT_POSIX_ZERO(pthread_attr_init(&attr), "pthread_attr_init");
+       T_ASSERT_POSIX_ZERO(pthread_attr_init(&attr), "pthread_attr_init");
 
-    T_ASSERT_POSIX_ZERO(pthread_attr_setschedparam(&attr, &param),
-            "pthread_attr_setschedparam");
+       T_ASSERT_POSIX_ZERO(pthread_attr_setschedparam(&attr, &param),
+           "pthread_attr_setschedparam");
 
-    if (fixpri) {
-        T_ASSERT_POSIX_ZERO(pthread_attr_setschedpolicy(&attr, SCHED_RR),
-                "pthread_attr_setschedpolicy");
-    }
+       if (fixpri) {
+               T_ASSERT_POSIX_ZERO(pthread_attr_setschedpolicy(&attr, SCHED_RR),
+                   "pthread_attr_setschedpolicy");
+       }
 
-    T_ASSERT_POSIX_ZERO(pthread_create(&new_thread, &attr, start_routine,
-            (void*)(uintptr_t)thread_id), "pthread_create");
+       T_ASSERT_POSIX_ZERO(pthread_create(&new_thread, &attr, start_routine,
+           (void*)(uintptr_t)thread_id), "pthread_create");
 
-    T_ASSERT_POSIX_ZERO(pthread_attr_destroy(&attr), "pthread_attr_destroy");
+       T_ASSERT_POSIX_ZERO(pthread_attr_destroy(&attr), "pthread_attr_destroy");
 
-    threads[thread_id].thread = new_thread;
+       threads[thread_id].thread = new_thread;
 
-    return new_thread;
+       return new_thread;
 }
 
 /* Spin until a specified number of seconds elapses */
 static void
 spin_for_duration(uint32_t seconds)
 {
-    uint64_t duration       = nanos_to_abs((uint64_t)seconds * NSEC_PER_SEC);
-    uint64_t current_time   = mach_absolute_time();
-    uint64_t timeout        = duration + current_time;
+       uint64_t duration       = nanos_to_abs((uint64_t)seconds * NSEC_PER_SEC);
+       uint64_t current_time   = mach_absolute_time();
+       uint64_t timeout        = duration + current_time;
 
-    uint64_t spin_count = 0;
+       uint64_t spin_count = 0;
 
-    while (mach_absolute_time() < timeout && atomic_load_explicit(&keep_going,
-               memory_order_relaxed)) {
-        spin_count++;
-    }
+       while (mach_absolute_time() < timeout && atomic_load_explicit(&keep_going,
+           memory_order_relaxed)) {
+               spin_count++;
+       }
 }
 
 static void *
 spin_thread(void *arg)
 {
-    uint32_t thread_id = (uint32_t) arg;
-    char name[30] = "";
+       uint32_t thread_id = (uint32_t) arg;
+       char name[30] = "";
 
-    snprintf(name, sizeof(name), "spin thread %2d", thread_id);
-    pthread_setname_np(name);
-    T_ASSERT_MACH_SUCCESS(semaphore_wait_signal(semaphore, worker_sem),
+       snprintf(name, sizeof(name), "spin thread %2d", thread_id);
+       pthread_setname_np(name);
+       T_ASSERT_MACH_SUCCESS(semaphore_wait_signal(semaphore, worker_sem),
            "semaphore_wait_signal");
-    spin_for_duration(SPIN_SECS);
-    return NULL;
+       spin_for_duration(SPIN_SECS);
+       return NULL;
 }
 
 static void *
 thread(void *arg)
 {
-    uint32_t thread_id = (uint32_t) arg;
-    char name[30] = "";
-
-    snprintf(name, sizeof(name), "thread %2d", thread_id);
-    pthread_setname_np(name);
-    T_ASSERT_MACH_SUCCESS(semaphore_wait_signal(semaphore, worker_sem), "semaphore_wait");
-
-    if (threads[thread_id].measure_thread) {
-        for (int i = 0; i < WARMUP_ITERATIONS; i++) {
-            thread_switch(THREAD_NULL, SWITCH_OPTION_NONE, 0);
-        }
-        T_STAT_MEASURE_LOOP(s) {
-            if(thread_switch(THREAD_NULL, SWITCH_OPTION_NONE, 0))
-                T_ASSERT_FAIL("thread_switch");
-        }
-        atomic_store_explicit(&keep_going, 0, memory_order_relaxed);
-    } else {
-        while (atomic_load_explicit(&keep_going, memory_order_relaxed)) {
-            if (thread_switch(THREAD_NULL, SWITCH_OPTION_NONE, 0))
-                T_ASSERT_FAIL("thread_switch");
-        }
-    }
-    return NULL;
+       uint32_t thread_id = (uint32_t) arg;
+       char name[30] = "";
+
+       snprintf(name, sizeof(name), "thread %2d", thread_id);
+       pthread_setname_np(name);
+       T_ASSERT_MACH_SUCCESS(semaphore_wait_signal(semaphore, worker_sem), "semaphore_wait");
+
+       if (threads[thread_id].measure_thread) {
+               for (int i = 0; i < WARMUP_ITERATIONS; i++) {
+                       thread_switch(THREAD_NULL, SWITCH_OPTION_NONE, 0);
+               }
+               T_STAT_MEASURE_LOOP(s) {
+                       if (thread_switch(THREAD_NULL, SWITCH_OPTION_NONE, 0)) {
+                               T_ASSERT_FAIL("thread_switch");
+                       }
+               }
+               atomic_store_explicit(&keep_going, 0, memory_order_relaxed);
+       } else {
+               while (atomic_load_explicit(&keep_going, memory_order_relaxed)) {
+                       if (thread_switch(THREAD_NULL, SWITCH_OPTION_NONE, 0)) {
+                               T_ASSERT_FAIL("thread_switch");
+                       }
+               }
+       }
+       return NULL;
 }
 
-void check_device_temperature(void)
+void
+check_device_temperature(void)
 {
-    char buffer[256];
-    FILE *pipe = popen("powerctrl Factor1", "r");
-    
-    if (pipe == NULL) {
-        T_FAIL("Failed to check device temperature");
-        T_END;
-    }
-
-    fgets(buffer, sizeof(buffer), pipe);
-    
-    if (strncmp(POWERCTRL_SUCCESS_STR, buffer, strlen(POWERCTRL_SUCCESS_STR))) {
-        T_PERF("temperature", 0.0, "factor", "device temperature");
-    } else {
-        T_PASS("Device temperature check pass");
-        T_PERF("temperature", 1.0, "factor", "device temperature");
-    }
-    pclose(pipe);
+       char buffer[256];
+       FILE *pipe = popen("powerctrl Factor1", "r");
+
+       if (pipe == NULL) {
+               T_FAIL("Failed to check device temperature");
+               T_END;
+       }
+
+       fgets(buffer, sizeof(buffer), pipe);
+
+       if (strncmp(POWERCTRL_SUCCESS_STR, buffer, strlen(POWERCTRL_SUCCESS_STR))) {
+               T_PERF("temperature", 0.0, "factor", "device temperature");
+       } else {
+               T_PASS("Device temperature check pass");
+               T_PERF("temperature", 1.0, "factor", "device temperature");
+       }
+       pclose(pipe);
 }
 
-void record_perfcontrol_stats(const char *sysctlname, const char *units, const char *info)
+void
+record_perfcontrol_stats(const char *sysctlname, const char *units, const char *info)
 {
-    int data = 0;
-    size_t data_size = sizeof(data);
-    T_ASSERT_POSIX_ZERO(sysctlbyname(sysctlname,
-           &data, &data_size, NULL, 0), 
+       int data = 0;
+       size_t data_size = sizeof(data);
+       T_ASSERT_POSIX_ZERO(sysctlbyname(sysctlname,
+           &data, &data_size, NULL, 0),
            "%s", sysctlname);
-    T_PERF(info, data, units, info);
+       T_PERF(info, data, units, info);
 }
 
 
@@ -192,94 +196,93 @@ T_GLOBAL_META(T_META_NAMESPACE("xnu.scheduler"));
 /* Disable the test on MacOS for now */
 T_DECL(perf_csw, "context switch performance", T_META_TAG_PERF, T_META_CHECK_LEAKS(false), T_META_ASROOT(true))
 {
-
 #if !CONFIG_EMBEDDED
-    T_SKIP("Not supported on MacOS");
-    return;
+       T_SKIP("Not supported on MacOS");
+       return;
 #endif /* CONFIG_EMBEDDED */
-    check_device_temperature();
-
-    T_ATEND(csw_perf_test_cleanup);
-
-    csw_perf_test_init();
-    pthread_setname_np("main thread");
-
-    T_ASSERT_MACH_SUCCESS(mach_timebase_info(&timebase_info), "mach_timebase_info");
-
-    struct sched_param param = {.sched_priority = 48};
-
-    T_ASSERT_POSIX_ZERO(pthread_setschedparam(pthread_self(), SCHED_FIFO, &param),
-            "pthread_setschedparam");
-
-    T_ASSERT_MACH_SUCCESS(semaphore_create(mach_task_self(), &semaphore,
-            SYNC_POLICY_FIFO, 0), "semaphore_create");
-
-    T_ASSERT_MACH_SUCCESS(semaphore_create(mach_task_self(), &worker_sem,
-            SYNC_POLICY_FIFO, 0), "semaphore_create");
-    
-    size_t ncpu_size = sizeof(g_numcpus);
-    T_ASSERT_POSIX_ZERO(sysctlbyname("hw.ncpu", &g_numcpus, &ncpu_size, NULL, 0),
-            "sysctlbyname hw.ncpu");
-
-    printf("hw.ncpu: %d\n", g_numcpus);
-    uint32_t n_spinners = g_numcpus - 1;
-
-    int mt_supported = 0;
-    size_t mt_supported_size = sizeof(mt_supported);
-    T_ASSERT_POSIX_ZERO(sysctlbyname("kern.monotonic.supported", &mt_supported,
-            &mt_supported_size, NULL, 0), "sysctlbyname kern.monotonic.supported");
-
-    for (uint32_t thread_id = 0; thread_id < n_spinners; thread_id++) {
-        threads[thread_id].thread = create_thread(thread_id, THR_SPINNER_PRI,
-                true, &spin_thread);
-    }
-
-    s = dt_stat_time_create("context switch time");
-
-    create_thread(n_spinners, THR_MANAGER_PRI, true, &thread);
-    threads[n_spinners].measure_thread = true;
-    create_thread(n_spinners + 1, THR_MANAGER_PRI, true, &thread);
-
-    /* Allow the context switch threads to get into sem_wait() */
-    for (uint32_t thread_id = 0; thread_id < n_spinners + 2; thread_id++) {
-        T_ASSERT_MACH_SUCCESS(semaphore_wait(worker_sem), "semaphore_wait");
-    }
-    
-    int enable_callout_stats = 1;
-    size_t enable_size = sizeof(enable_callout_stats);
-
-    if (mt_supported) {
-        /* Enable callout stat collection */
-        T_ASSERT_POSIX_ZERO(sysctlbyname("kern.perfcontrol_callout.stats_enabled",
-                NULL, 0, &enable_callout_stats, enable_size),
-                "sysctlbyname kern.perfcontrol_callout.stats_enabled");
-    }
-    
-    T_ASSERT_MACH_SUCCESS(semaphore_signal_all(semaphore), "semaphore_signal");
-
-
-    for (uint32_t thread_id = 0; thread_id < n_spinners + 2; thread_id++) {
-        T_ASSERT_POSIX_ZERO(pthread_join(threads[thread_id].thread, NULL),
-                "pthread_join %d", thread_id);
-    }
-
-    if (mt_supported) {
-        record_perfcontrol_stats("kern.perfcontrol_callout.oncore_instr",
-                "instructions", "oncore.instructions");
-        record_perfcontrol_stats("kern.perfcontrol_callout.offcore_instr",
-                "instructions", "offcore.instructions");
-        record_perfcontrol_stats("kern.perfcontrol_callout.oncore_cycles",
-                "cycles", "oncore.cycles");
-        record_perfcontrol_stats("kern.perfcontrol_callout.offcore_cycles",
-                "cycles", "offcore.cycles");
-
-        /* Disable callout stat collection */
-        enable_callout_stats = 0;
-        T_ASSERT_POSIX_ZERO(sysctlbyname("kern.perfcontrol_callout.stats_enabled",
-                NULL, 0, &enable_callout_stats, enable_size),
-                "sysctlbyname kern.perfcontrol_callout.stats_enabled");
-    }
-
-    check_device_temperature();
-    dt_stat_finalize(s);
+       check_device_temperature();
+
+       T_ATEND(csw_perf_test_cleanup);
+
+       csw_perf_test_init();
+       pthread_setname_np("main thread");
+
+       T_ASSERT_MACH_SUCCESS(mach_timebase_info(&timebase_info), "mach_timebase_info");
+
+       struct sched_param param = {.sched_priority = 48};
+
+       T_ASSERT_POSIX_ZERO(pthread_setschedparam(pthread_self(), SCHED_FIFO, &param),
+           "pthread_setschedparam");
+
+       T_ASSERT_MACH_SUCCESS(semaphore_create(mach_task_self(), &semaphore,
+           SYNC_POLICY_FIFO, 0), "semaphore_create");
+
+       T_ASSERT_MACH_SUCCESS(semaphore_create(mach_task_self(), &worker_sem,
+           SYNC_POLICY_FIFO, 0), "semaphore_create");
+
+       size_t ncpu_size = sizeof(g_numcpus);
+       T_ASSERT_POSIX_ZERO(sysctlbyname("hw.ncpu", &g_numcpus, &ncpu_size, NULL, 0),
+           "sysctlbyname hw.ncpu");
+
+       printf("hw.ncpu: %d\n", g_numcpus);
+       uint32_t n_spinners = g_numcpus - 1;
+
+       int mt_supported = 0;
+       size_t mt_supported_size = sizeof(mt_supported);
+       T_ASSERT_POSIX_ZERO(sysctlbyname("kern.monotonic.supported", &mt_supported,
+           &mt_supported_size, NULL, 0), "sysctlbyname kern.monotonic.supported");
+
+       for (uint32_t thread_id = 0; thread_id < n_spinners; thread_id++) {
+               threads[thread_id].thread = create_thread(thread_id, THR_SPINNER_PRI,
+                   true, &spin_thread);
+       }
+
+       s = dt_stat_time_create("context switch time");
+
+       create_thread(n_spinners, THR_MANAGER_PRI, true, &thread);
+       threads[n_spinners].measure_thread = true;
+       create_thread(n_spinners + 1, THR_MANAGER_PRI, true, &thread);
+
+       /* Allow the context switch threads to get into sem_wait() */
+       for (uint32_t thread_id = 0; thread_id < n_spinners + 2; thread_id++) {
+               T_ASSERT_MACH_SUCCESS(semaphore_wait(worker_sem), "semaphore_wait");
+       }
+
+       int enable_callout_stats = 1;
+       size_t enable_size = sizeof(enable_callout_stats);
+
+       if (mt_supported) {
+               /* Enable callout stat collection */
+               T_ASSERT_POSIX_ZERO(sysctlbyname("kern.perfcontrol_callout.stats_enabled",
+                   NULL, 0, &enable_callout_stats, enable_size),
+                   "sysctlbyname kern.perfcontrol_callout.stats_enabled");
+       }
+
+       T_ASSERT_MACH_SUCCESS(semaphore_signal_all(semaphore), "semaphore_signal");
+
+
+       for (uint32_t thread_id = 0; thread_id < n_spinners + 2; thread_id++) {
+               T_ASSERT_POSIX_ZERO(pthread_join(threads[thread_id].thread, NULL),
+                   "pthread_join %d", thread_id);
+       }
+
+       if (mt_supported) {
+               record_perfcontrol_stats("kern.perfcontrol_callout.oncore_instr",
+                   "instructions", "oncore.instructions");
+               record_perfcontrol_stats("kern.perfcontrol_callout.offcore_instr",
+                   "instructions", "offcore.instructions");
+               record_perfcontrol_stats("kern.perfcontrol_callout.oncore_cycles",
+                   "cycles", "oncore.cycles");
+               record_perfcontrol_stats("kern.perfcontrol_callout.offcore_cycles",
+                   "cycles", "offcore.cycles");
+
+               /* Disable callout stat collection */
+               enable_callout_stats = 0;
+               T_ASSERT_POSIX_ZERO(sysctlbyname("kern.perfcontrol_callout.stats_enabled",
+                   NULL, 0, &enable_callout_stats, enable_size),
+                   "sysctlbyname kern.perfcontrol_callout.stats_enabled");
+       }
+
+       check_device_temperature();
+       dt_stat_finalize(s);
 }