new version commit

deps/jemalloc/src/atomic.c

@@ -0,0 +1,2 @@
+#define	JEMALLOC_ATOMIC_C_
+#include "jemalloc/internal/jemalloc_internal.h"

deps/jemalloc/src/background_thread.c

@@ -0,0 +1,939 @@
+#define JEMALLOC_BACKGROUND_THREAD_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+
+JEMALLOC_DIAGNOSTIC_DISABLE_SPURIOUS
+
+/******************************************************************************/
+/* Data. */
+
+/* This option should be opt-in only. */
+#define BACKGROUND_THREAD_DEFAULT false
+/* Read-only after initialization. */
+bool opt_background_thread = BACKGROUND_THREAD_DEFAULT;
+size_t opt_max_background_threads = MAX_BACKGROUND_THREAD_LIMIT + 1;
+
+/* Used for thread creation, termination and stats. */
+malloc_mutex_t background_thread_lock;
+/* Indicates global state.  Atomic because decay reads this w/o locking. */
+atomic_b_t background_thread_enabled_state;
+size_t n_background_threads;
+size_t max_background_threads;
+/* Thread info per-index. */
+background_thread_info_t *background_thread_info;
+
+/******************************************************************************/
+
+#ifdef JEMALLOC_PTHREAD_CREATE_WRAPPER
+
+static int (*pthread_create_fptr)(pthread_t *__restrict, const pthread_attr_t *,
+    void *(*)(void *), void *__restrict);
+
+static void
+pthread_create_wrapper_init(void) {
+#ifdef JEMALLOC_LAZY_LOCK
+	if (!isthreaded) {
+		isthreaded = true;
+	}
+#endif
+}
+
+int
+pthread_create_wrapper(pthread_t *__restrict thread, const pthread_attr_t *attr,
+    void *(*start_routine)(void *), void *__restrict arg) {
+	pthread_create_wrapper_init();
+
+	return pthread_create_fptr(thread, attr, start_routine, arg);
+}
+#endif /* JEMALLOC_PTHREAD_CREATE_WRAPPER */
+
+#ifndef JEMALLOC_BACKGROUND_THREAD
+#define NOT_REACHED { not_reached(); }
+bool background_thread_create(tsd_t *tsd, unsigned arena_ind) NOT_REACHED
+bool background_threads_enable(tsd_t *tsd) NOT_REACHED
+bool background_threads_disable(tsd_t *tsd) NOT_REACHED
+void background_thread_interval_check(tsdn_t *tsdn, arena_t *arena,
+    arena_decay_t *decay, size_t npages_new) NOT_REACHED
+void background_thread_prefork0(tsdn_t *tsdn) NOT_REACHED
+void background_thread_prefork1(tsdn_t *tsdn) NOT_REACHED
+void background_thread_postfork_parent(tsdn_t *tsdn) NOT_REACHED
+void background_thread_postfork_child(tsdn_t *tsdn) NOT_REACHED
+bool background_thread_stats_read(tsdn_t *tsdn,
+    background_thread_stats_t *stats) NOT_REACHED
+void background_thread_ctl_init(tsdn_t *tsdn) NOT_REACHED
+#undef NOT_REACHED
+#else
+
+static bool background_thread_enabled_at_fork;
+
+static void
+background_thread_info_init(tsdn_t *tsdn, background_thread_info_t *info) {
+	background_thread_wakeup_time_set(tsdn, info, 0);
+	info->npages_to_purge_new = 0;
+	if (config_stats) {
+		info->tot_n_runs = 0;
+		nstime_init(&info->tot_sleep_time, 0);
+	}
+}
+
+static inline bool
+set_current_thread_affinity(int cpu) {
+#if defined(JEMALLOC_HAVE_SCHED_SETAFFINITY)
+	cpu_set_t cpuset;
+	CPU_ZERO(&cpuset);
+	CPU_SET(cpu, &cpuset);
+	int ret = sched_setaffinity(0, sizeof(cpu_set_t), &cpuset);
+
+	return (ret != 0);
+#else
+	return false;
+#endif
+}
+
+/* Threshold for determining when to wake up the background thread. */
+#define BACKGROUND_THREAD_NPAGES_THRESHOLD UINT64_C(1024)
+#define BILLION UINT64_C(1000000000)
+/* Minimal sleep interval 100 ms. */
+#define BACKGROUND_THREAD_MIN_INTERVAL_NS (BILLION / 10)
+
+static inline size_t
+decay_npurge_after_interval(arena_decay_t *decay, size_t interval) {
+	size_t i;
+	uint64_t sum = 0;
+	for (i = 0; i < interval; i++) {
+		sum += decay->backlog[i] * h_steps[i];
+	}
+	for (; i < SMOOTHSTEP_NSTEPS; i++) {
+		sum += decay->backlog[i] * (h_steps[i] - h_steps[i - interval]);
+	}
+
+	return (size_t)(sum >> SMOOTHSTEP_BFP);
+}
+
+static uint64_t
+arena_decay_compute_purge_interval_impl(tsdn_t *tsdn, arena_decay_t *decay,
+    extents_t *extents) {
+	if (malloc_mutex_trylock(tsdn, &decay->mtx)) {
+		/* Use minimal interval if decay is contended. */
+		return BACKGROUND_THREAD_MIN_INTERVAL_NS;
+	}
+
+	uint64_t interval;
+	ssize_t decay_time = atomic_load_zd(&decay->time_ms, ATOMIC_RELAXED);
+	if (decay_time <= 0) {
+		/* Purging is eagerly done or disabled currently. */
+		interval = BACKGROUND_THREAD_INDEFINITE_SLEEP;
+		goto label_done;
+	}
+
+	uint64_t decay_interval_ns = nstime_ns(&decay->interval);
+	assert(decay_interval_ns > 0);
+	size_t npages = extents_npages_get(extents);
+	if (npages == 0) {
+		unsigned i;
+		for (i = 0; i < SMOOTHSTEP_NSTEPS; i++) {
+			if (decay->backlog[i] > 0) {
+				break;
+			}
+		}
+		if (i == SMOOTHSTEP_NSTEPS) {
+			/* No dirty pages recorded.  Sleep indefinitely. */
+			interval = BACKGROUND_THREAD_INDEFINITE_SLEEP;
+			goto label_done;
+		}
+	}
+	if (npages <= BACKGROUND_THREAD_NPAGES_THRESHOLD) {
+		/* Use max interval. */
+		interval = decay_interval_ns * SMOOTHSTEP_NSTEPS;
+		goto label_done;
+	}
+
+	size_t lb = BACKGROUND_THREAD_MIN_INTERVAL_NS / decay_interval_ns;
+	size_t ub = SMOOTHSTEP_NSTEPS;
+	/* Minimal 2 intervals to ensure reaching next epoch deadline. */
+	lb = (lb < 2) ? 2 : lb;
+	if ((decay_interval_ns * ub <= BACKGROUND_THREAD_MIN_INTERVAL_NS) ||
+	    (lb + 2 > ub)) {
+		interval = BACKGROUND_THREAD_MIN_INTERVAL_NS;
+		goto label_done;
+	}
+
+	assert(lb + 2 <= ub);
+	size_t npurge_lb, npurge_ub;
+	npurge_lb = decay_npurge_after_interval(decay, lb);
+	if (npurge_lb > BACKGROUND_THREAD_NPAGES_THRESHOLD) {
+		interval = decay_interval_ns * lb;
+		goto label_done;
+	}
+	npurge_ub = decay_npurge_after_interval(decay, ub);
+	if (npurge_ub < BACKGROUND_THREAD_NPAGES_THRESHOLD) {
+		interval = decay_interval_ns * ub;
+		goto label_done;
+	}
+
+	unsigned n_search = 0;
+	size_t target, npurge;
+	while ((npurge_lb + BACKGROUND_THREAD_NPAGES_THRESHOLD < npurge_ub)
+	    && (lb + 2 < ub)) {
+		target = (lb + ub) / 2;
+		npurge = decay_npurge_after_interval(decay, target);
+		if (npurge > BACKGROUND_THREAD_NPAGES_THRESHOLD) {
+			ub = target;
+			npurge_ub = npurge;
+		} else {
+			lb = target;
+			npurge_lb = npurge;
+		}
+		assert(n_search++ < lg_floor(SMOOTHSTEP_NSTEPS) + 1);
+	}
+	interval = decay_interval_ns * (ub + lb) / 2;
+label_done:
+	interval = (interval < BACKGROUND_THREAD_MIN_INTERVAL_NS) ?
+	    BACKGROUND_THREAD_MIN_INTERVAL_NS : interval;
+	malloc_mutex_unlock(tsdn, &decay->mtx);
+
+	return interval;
+}
+
+/* Compute purge interval for background threads. */
+static uint64_t
+arena_decay_compute_purge_interval(tsdn_t *tsdn, arena_t *arena) {
+	uint64_t i1, i2;
+	i1 = arena_decay_compute_purge_interval_impl(tsdn, &arena->decay_dirty,
+	    &arena->extents_dirty);
+	if (i1 == BACKGROUND_THREAD_MIN_INTERVAL_NS) {
+		return i1;
+	}
+	i2 = arena_decay_compute_purge_interval_impl(tsdn, &arena->decay_muzzy,
+	    &arena->extents_muzzy);
+
+	return i1 < i2 ? i1 : i2;
+}
+
+static void
+background_thread_sleep(tsdn_t *tsdn, background_thread_info_t *info,
+    uint64_t interval) {
+	if (config_stats) {
+		info->tot_n_runs++;
+	}
+	info->npages_to_purge_new = 0;
+
+	struct timeval tv;
+	/* Specific clock required by timedwait. */
+	gettimeofday(&tv, NULL);
+	nstime_t before_sleep;
+	nstime_init2(&before_sleep, tv.tv_sec, tv.tv_usec * 1000);
+
+	int ret;
+	if (interval == BACKGROUND_THREAD_INDEFINITE_SLEEP) {
+		assert(background_thread_indefinite_sleep(info));
+		ret = pthread_cond_wait(&info->cond, &info->mtx.lock);
+		assert(ret == 0);
+	} else {
+		assert(interval >= BACKGROUND_THREAD_MIN_INTERVAL_NS &&
+		    interval <= BACKGROUND_THREAD_INDEFINITE_SLEEP);
+		/* We need malloc clock (can be different from tv). */
+		nstime_t next_wakeup;
+		nstime_init(&next_wakeup, 0);
+		nstime_update(&next_wakeup);
+		nstime_iadd(&next_wakeup, interval);
+		assert(nstime_ns(&next_wakeup) <
+		    BACKGROUND_THREAD_INDEFINITE_SLEEP);
+		background_thread_wakeup_time_set(tsdn, info,
+		    nstime_ns(&next_wakeup));
+
+		nstime_t ts_wakeup;
+		nstime_copy(&ts_wakeup, &before_sleep);
+		nstime_iadd(&ts_wakeup, interval);
+		struct timespec ts;
+		ts.tv_sec = (size_t)nstime_sec(&ts_wakeup);
+		ts.tv_nsec = (size_t)nstime_nsec(&ts_wakeup);
+
+		assert(!background_thread_indefinite_sleep(info));
+		ret = pthread_cond_timedwait(&info->cond, &info->mtx.lock, &ts);
+		assert(ret == ETIMEDOUT || ret == 0);
+		background_thread_wakeup_time_set(tsdn, info,
+		    BACKGROUND_THREAD_INDEFINITE_SLEEP);
+	}
+	if (config_stats) {
+		gettimeofday(&tv, NULL);
+		nstime_t after_sleep;
+		nstime_init2(&after_sleep, tv.tv_sec, tv.tv_usec * 1000);
+		if (nstime_compare(&after_sleep, &before_sleep) > 0) {
+			nstime_subtract(&after_sleep, &before_sleep);
+			nstime_add(&info->tot_sleep_time, &after_sleep);
+		}
+	}
+}
+
+static bool
+background_thread_pause_check(tsdn_t *tsdn, background_thread_info_t *info) {
+	if (unlikely(info->state == background_thread_paused)) {
+		malloc_mutex_unlock(tsdn, &info->mtx);
+		/* Wait on global lock to update status. */
+		malloc_mutex_lock(tsdn, &background_thread_lock);
+		malloc_mutex_unlock(tsdn, &background_thread_lock);
+		malloc_mutex_lock(tsdn, &info->mtx);
+		return true;
+	}
+
+	return false;
+}
+
+static inline void
+background_work_sleep_once(tsdn_t *tsdn, background_thread_info_t *info, unsigned ind) {
+	uint64_t min_interval = BACKGROUND_THREAD_INDEFINITE_SLEEP;
+	unsigned narenas = narenas_total_get();
+
+	for (unsigned i = ind; i < narenas; i += max_background_threads) {
+		arena_t *arena = arena_get(tsdn, i, false);
+		if (!arena) {
+			continue;
+		}
+		arena_decay(tsdn, arena, true, false);
+		if (min_interval == BACKGROUND_THREAD_MIN_INTERVAL_NS) {
+			/* Min interval will be used. */
+			continue;
+		}
+		uint64_t interval = arena_decay_compute_purge_interval(tsdn,
+		    arena);
+		assert(interval >= BACKGROUND_THREAD_MIN_INTERVAL_NS);
+		if (min_interval > interval) {
+			min_interval = interval;
+		}
+	}
+	background_thread_sleep(tsdn, info, min_interval);
+}
+
+static bool
+background_threads_disable_single(tsd_t *tsd, background_thread_info_t *info) {
+	if (info == &background_thread_info[0]) {
+		malloc_mutex_assert_owner(tsd_tsdn(tsd),
+		    &background_thread_lock);
+	} else {
+		malloc_mutex_assert_not_owner(tsd_tsdn(tsd),
+		    &background_thread_lock);
+	}
+
+	pre_reentrancy(tsd, NULL);
+	malloc_mutex_lock(tsd_tsdn(tsd), &info->mtx);
+	bool has_thread;
+	assert(info->state != background_thread_paused);
+	if (info->state == background_thread_started) {
+		has_thread = true;
+		info->state = background_thread_stopped;
+		pthread_cond_signal(&info->cond);
+	} else {
+		has_thread = false;
+	}
+	malloc_mutex_unlock(tsd_tsdn(tsd), &info->mtx);
+
+	if (!has_thread) {
+		post_reentrancy(tsd);
+		return false;
+	}
+	void *ret;
+	if (pthread_join(info->thread, &ret)) {
+		post_reentrancy(tsd);
+		return true;
+	}
+	assert(ret == NULL);
+	n_background_threads--;
+	post_reentrancy(tsd);
+
+	return false;
+}
+
+static void *background_thread_entry(void *ind_arg);
+
+static int
+background_thread_create_signals_masked(pthread_t *thread,
+    const pthread_attr_t *attr, void *(*start_routine)(void *), void *arg) {
+	/*
+	 * Mask signals during thread creation so that the thread inherits
+	 * an empty signal set.
+	 */
+	sigset_t set;
+	sigfillset(&set);
+	sigset_t oldset;
+	int mask_err = pthread_sigmask(SIG_SETMASK, &set, &oldset);
+	if (mask_err != 0) {
+		return mask_err;
+	}
+	int create_err = pthread_create_wrapper(thread, attr, start_routine,
+	    arg);
+	/*
+	 * Restore the signal mask.  Failure to restore the signal mask here
+	 * changes program behavior.
+	 */
+	int restore_err = pthread_sigmask(SIG_SETMASK, &oldset, NULL);
+	if (restore_err != 0) {
+		malloc_printf("<jemalloc>: background thread creation "
+		    "failed (%d), and signal mask restoration failed "
+		    "(%d)\n", create_err, restore_err);
+		if (opt_abort) {
+			abort();
+		}
+	}
+	return create_err;
+}
+
+static bool
+check_background_thread_creation(tsd_t *tsd, unsigned *n_created,
+    bool *created_threads) {
+	bool ret = false;
+	if (likely(*n_created == n_background_threads)) {
+		return ret;
+	}
+
+	tsdn_t *tsdn = tsd_tsdn(tsd);
+	malloc_mutex_unlock(tsdn, &background_thread_info[0].mtx);
+	for (unsigned i = 1; i < max_background_threads; i++) {
+		if (created_threads[i]) {
+			continue;
+		}
+		background_thread_info_t *info = &background_thread_info[i];
+		malloc_mutex_lock(tsdn, &info->mtx);
+		/*
+		 * In case of the background_thread_paused state because of
+		 * arena reset, delay the creation.
+		 */
+		bool create = (info->state == background_thread_started);
+		malloc_mutex_unlock(tsdn, &info->mtx);
+		if (!create) {
+			continue;
+		}
+
+		pre_reentrancy(tsd, NULL);
+		int err = background_thread_create_signals_masked(&info->thread,
+		    NULL, background_thread_entry, (void *)(uintptr_t)i);
+		post_reentrancy(tsd);
+
+		if (err == 0) {
+			(*n_created)++;
+			created_threads[i] = true;
+		} else {
+			malloc_printf("<jemalloc>: background thread "
+			    "creation failed (%d)\n", err);
+			if (opt_abort) {
+				abort();
+			}
+		}
+		/* Return to restart the loop since we unlocked. */
+		ret = true;
+		break;
+	}
+	malloc_mutex_lock(tsdn, &background_thread_info[0].mtx);
+
+	return ret;
+}
+
+static void
+background_thread0_work(tsd_t *tsd) {
+	/* Thread0 is also responsible for launching / terminating threads. */
+	VARIABLE_ARRAY(bool, created_threads, max_background_threads);
+	unsigned i;
+	for (i = 1; i < max_background_threads; i++) {
+		created_threads[i] = false;
+	}
+	/* Start working, and create more threads when asked. */
+	unsigned n_created = 1;
+	while (background_thread_info[0].state != background_thread_stopped) {
+		if (background_thread_pause_check(tsd_tsdn(tsd),
+		    &background_thread_info[0])) {
+			continue;
+		}
+		if (check_background_thread_creation(tsd, &n_created,
+		    (bool *)&created_threads)) {
+			continue;
+		}
+		background_work_sleep_once(tsd_tsdn(tsd),
+		    &background_thread_info[0], 0);
+	}
+
+	/*
+	 * Shut down other threads at exit.  Note that the ctl thread is holding
+	 * the global background_thread mutex (and is waiting) for us.
+	 */
+	assert(!background_thread_enabled());
+	for (i = 1; i < max_background_threads; i++) {
+		background_thread_info_t *info = &background_thread_info[i];
+		assert(info->state != background_thread_paused);
+		if (created_threads[i]) {
+			background_threads_disable_single(tsd, info);
+		} else {
+			malloc_mutex_lock(tsd_tsdn(tsd), &info->mtx);
+			if (info->state != background_thread_stopped) {
+				/* The thread was not created. */
+				assert(info->state ==
+				    background_thread_started);
+				n_background_threads--;
+				info->state = background_thread_stopped;
+			}
+			malloc_mutex_unlock(tsd_tsdn(tsd), &info->mtx);
+		}
+	}
+	background_thread_info[0].state = background_thread_stopped;
+	assert(n_background_threads == 1);
+}
+
+static void
+background_work(tsd_t *tsd, unsigned ind) {
+	background_thread_info_t *info = &background_thread_info[ind];
+
+	malloc_mutex_lock(tsd_tsdn(tsd), &info->mtx);
+	background_thread_wakeup_time_set(tsd_tsdn(tsd), info,
+	    BACKGROUND_THREAD_INDEFINITE_SLEEP);
+	if (ind == 0) {
+		background_thread0_work(tsd);
+	} else {
+		while (info->state != background_thread_stopped) {
+			if (background_thread_pause_check(tsd_tsdn(tsd),
+			    info)) {
+				continue;
+			}
+			background_work_sleep_once(tsd_tsdn(tsd), info, ind);
+		}
+	}
+	assert(info->state == background_thread_stopped);
+	background_thread_wakeup_time_set(tsd_tsdn(tsd), info, 0);
+	malloc_mutex_unlock(tsd_tsdn(tsd), &info->mtx);
+}
+
+static void *
+background_thread_entry(void *ind_arg) {
+	unsigned thread_ind = (unsigned)(uintptr_t)ind_arg;
+	assert(thread_ind < max_background_threads);
+#ifdef JEMALLOC_HAVE_PTHREAD_SETNAME_NP
+	pthread_setname_np(pthread_self(), "jemalloc_bg_thd");
+#elif defined(__FreeBSD__)
+	pthread_set_name_np(pthread_self(), "jemalloc_bg_thd");
+#endif
+	if (opt_percpu_arena != percpu_arena_disabled) {
+		set_current_thread_affinity((int)thread_ind);
+	}
+	/*
+	 * Start periodic background work.  We use internal tsd which avoids
+	 * side effects, for example triggering new arena creation (which in
+	 * turn triggers another background thread creation).
+	 */
+	background_work(tsd_internal_fetch(), thread_ind);
+	assert(pthread_equal(pthread_self(),
+	    background_thread_info[thread_ind].thread));
+
+	return NULL;
+}
+
+static void
+background_thread_init(tsd_t *tsd, background_thread_info_t *info) {
+	malloc_mutex_assert_owner(tsd_tsdn(tsd), &background_thread_lock);
+	info->state = background_thread_started;
+	background_thread_info_init(tsd_tsdn(tsd), info);
+	n_background_threads++;
+}
+
+static bool
+background_thread_create_locked(tsd_t *tsd, unsigned arena_ind) {
+	assert(have_background_thread);
+	malloc_mutex_assert_owner(tsd_tsdn(tsd), &background_thread_lock);
+
+	/* We create at most NCPUs threads. */
+	size_t thread_ind = arena_ind % max_background_threads;
+	background_thread_info_t *info = &background_thread_info[thread_ind];
+
+	bool need_new_thread;
+	malloc_mutex_lock(tsd_tsdn(tsd), &info->mtx);
+	need_new_thread = background_thread_enabled() &&
+	    (info->state == background_thread_stopped);
+	if (need_new_thread) {
+		background_thread_init(tsd, info);
+	}
+	malloc_mutex_unlock(tsd_tsdn(tsd), &info->mtx);
+	if (!need_new_thread) {
+		return false;
+	}
+	if (arena_ind != 0) {
+		/* Threads are created asynchronously by Thread 0. */
+		background_thread_info_t *t0 = &background_thread_info[0];
+		malloc_mutex_lock(tsd_tsdn(tsd), &t0->mtx);
+		assert(t0->state == background_thread_started);
+		pthread_cond_signal(&t0->cond);
+		malloc_mutex_unlock(tsd_tsdn(tsd), &t0->mtx);
+
+		return false;
+	}
+
+	pre_reentrancy(tsd, NULL);
+	/*
+	 * To avoid complications (besides reentrancy), create internal
+	 * background threads with the underlying pthread_create.
+	 */
+	int err = background_thread_create_signals_masked(&info->thread, NULL,
+	    background_thread_entry, (void *)thread_ind);
+	post_reentrancy(tsd);
+
+	if (err != 0) {
+		malloc_printf("<jemalloc>: arena 0 background thread creation "
+		    "failed (%d)\n", err);
+		malloc_mutex_lock(tsd_tsdn(tsd), &info->mtx);
+		info->state = background_thread_stopped;
+		n_background_threads--;
+		malloc_mutex_unlock(tsd_tsdn(tsd), &info->mtx);
+
+		return true;
+	}
+
+	return false;
+}
+
+/* Create a new background thread if needed. */
+bool
+background_thread_create(tsd_t *tsd, unsigned arena_ind) {
+	assert(have_background_thread);
+
+	bool ret;
+	malloc_mutex_lock(tsd_tsdn(tsd), &background_thread_lock);
+	ret = background_thread_create_locked(tsd, arena_ind);
+	malloc_mutex_unlock(tsd_tsdn(tsd), &background_thread_lock);
+
+	return ret;
+}
+
+bool
+background_threads_enable(tsd_t *tsd) {
+	assert(n_background_threads == 0);
+	assert(background_thread_enabled());
+	malloc_mutex_assert_owner(tsd_tsdn(tsd), &background_thread_lock);
+
+	VARIABLE_ARRAY(bool, marked, max_background_threads);
+	unsigned i, nmarked;
+	for (i = 0; i < max_background_threads; i++) {
+		marked[i] = false;
+	}
+	nmarked = 0;
+	/* Thread 0 is required and created at the end. */
+	marked[0] = true;
+	/* Mark the threads we need to create for thread 0. */
+	unsigned n = narenas_total_get();
+	for (i = 1; i < n; i++) {
+		if (marked[i % max_background_threads] ||
+		    arena_get(tsd_tsdn(tsd), i, false) == NULL) {
+			continue;
+		}
+		background_thread_info_t *info = &background_thread_info[
+		    i % max_background_threads];
+		malloc_mutex_lock(tsd_tsdn(tsd), &info->mtx);
+		assert(info->state == background_thread_stopped);
+		background_thread_init(tsd, info);
+		malloc_mutex_unlock(tsd_tsdn(tsd), &info->mtx);
+		marked[i % max_background_threads] = true;
+		if (++nmarked == max_background_threads) {
+			break;
+		}
+	}
+
+	return background_thread_create_locked(tsd, 0);
+}
+
+bool
+background_threads_disable(tsd_t *tsd) {
+	assert(!background_thread_enabled());
+	malloc_mutex_assert_owner(tsd_tsdn(tsd), &background_thread_lock);
+
+	/* Thread 0 will be responsible for terminating other threads. */
+	if (background_threads_disable_single(tsd,
+	    &background_thread_info[0])) {
+		return true;
+	}
+	assert(n_background_threads == 0);
+
+	return false;
+}
+
+/* Check if we need to signal the background thread early. */
+void
+background_thread_interval_check(tsdn_t *tsdn, arena_t *arena,
+    arena_decay_t *decay, size_t npages_new) {
+	background_thread_info_t *info = arena_background_thread_info_get(
+	    arena);
+	if (malloc_mutex_trylock(tsdn, &info->mtx)) {
+		/*
+		 * Background thread may hold the mutex for a long period of
+		 * time.  We'd like to avoid the variance on application
+		 * threads.  So keep this non-blocking, and leave the work to a
+		 * future epoch.
+		 */
+		return;
+	}
+
+	if (info->state != background_thread_started) {
+		goto label_done;
+	}
+	if (malloc_mutex_trylock(tsdn, &decay->mtx)) {
+		goto label_done;
+	}
+
+	ssize_t decay_time = atomic_load_zd(&decay->time_ms, ATOMIC_RELAXED);
+	if (decay_time <= 0) {
+		/* Purging is eagerly done or disabled currently. */
+		goto label_done_unlock2;
+	}
+	uint64_t decay_interval_ns = nstime_ns(&decay->interval);
+	assert(decay_interval_ns > 0);
+
+	nstime_t diff;
+	nstime_init(&diff, background_thread_wakeup_time_get(info));
+	if (nstime_compare(&diff, &decay->epoch) <= 0) {
+		goto label_done_unlock2;
+	}
+	nstime_subtract(&diff, &decay->epoch);
+	if (nstime_ns(&diff) < BACKGROUND_THREAD_MIN_INTERVAL_NS) {
+		goto label_done_unlock2;
+	}
+
+	if (npages_new > 0) {
+		size_t n_epoch = (size_t)(nstime_ns(&diff) / decay_interval_ns);
+		/*
+		 * Compute how many new pages we would need to purge by the next
+		 * wakeup, which is used to determine if we should signal the
+		 * background thread.
+		 */
+		uint64_t npurge_new;
+		if (n_epoch >= SMOOTHSTEP_NSTEPS) {
+			npurge_new = npages_new;
+		} else {
+			uint64_t h_steps_max = h_steps[SMOOTHSTEP_NSTEPS - 1];
+			assert(h_steps_max >=
+			    h_steps[SMOOTHSTEP_NSTEPS - 1 - n_epoch]);
+			npurge_new = npages_new * (h_steps_max -
+			    h_steps[SMOOTHSTEP_NSTEPS - 1 - n_epoch]);
+			npurge_new >>= SMOOTHSTEP_BFP;
+		}
+		info->npages_to_purge_new += npurge_new;
+	}
+
+	bool should_signal;
+	if (info->npages_to_purge_new > BACKGROUND_THREAD_NPAGES_THRESHOLD) {
+		should_signal = true;
+	} else if (unlikely(background_thread_indefinite_sleep(info)) &&
+	    (extents_npages_get(&arena->extents_dirty) > 0 ||
+	    extents_npages_get(&arena->extents_muzzy) > 0 ||
+	    info->npages_to_purge_new > 0)) {
+		should_signal = true;
+	} else {
+		should_signal = false;
+	}
+
+	if (should_signal) {
+		info->npages_to_purge_new = 0;
+		pthread_cond_signal(&info->cond);
+	}
+label_done_unlock2:
+	malloc_mutex_unlock(tsdn, &decay->mtx);
+label_done:
+	malloc_mutex_unlock(tsdn, &info->mtx);
+}
+
+void
+background_thread_prefork0(tsdn_t *tsdn) {
+	malloc_mutex_prefork(tsdn, &background_thread_lock);
+	background_thread_enabled_at_fork = background_thread_enabled();
+}
+
+void
+background_thread_prefork1(tsdn_t *tsdn) {
+	for (unsigned i = 0; i < max_background_threads; i++) {
+		malloc_mutex_prefork(tsdn, &background_thread_info[i].mtx);
+	}
+}
+
+void
+background_thread_postfork_parent(tsdn_t *tsdn) {
+	for (unsigned i = 0; i < max_background_threads; i++) {
+		malloc_mutex_postfork_parent(tsdn,
+		    &background_thread_info[i].mtx);
+	}
+	malloc_mutex_postfork_parent(tsdn, &background_thread_lock);
+}
+
+void
+background_thread_postfork_child(tsdn_t *tsdn) {
+	for (unsigned i = 0; i < max_background_threads; i++) {
+		malloc_mutex_postfork_child(tsdn,
+		    &background_thread_info[i].mtx);
+	}
+	malloc_mutex_postfork_child(tsdn, &background_thread_lock);
+	if (!background_thread_enabled_at_fork) {
+		return;
+	}
+
+	/* Clear background_thread state (reset to disabled for child). */
+	malloc_mutex_lock(tsdn, &background_thread_lock);
+	n_background_threads = 0;
+	background_thread_enabled_set(tsdn, false);
+	for (unsigned i = 0; i < max_background_threads; i++) {
+		background_thread_info_t *info = &background_thread_info[i];
+		malloc_mutex_lock(tsdn, &info->mtx);
+		info->state = background_thread_stopped;
+		int ret = pthread_cond_init(&info->cond, NULL);
+		assert(ret == 0);
+		background_thread_info_init(tsdn, info);
+		malloc_mutex_unlock(tsdn, &info->mtx);
+	}
+	malloc_mutex_unlock(tsdn, &background_thread_lock);
+}
+
+bool
+background_thread_stats_read(tsdn_t *tsdn, background_thread_stats_t *stats) {
+	assert(config_stats);
+	malloc_mutex_lock(tsdn, &background_thread_lock);
+	if (!background_thread_enabled()) {
+		malloc_mutex_unlock(tsdn, &background_thread_lock);
+		return true;
+	}
+
+	stats->num_threads = n_background_threads;
+	uint64_t num_runs = 0;
+	nstime_init(&stats->run_interval, 0);
+	for (unsigned i = 0; i < max_background_threads; i++) {
+		background_thread_info_t *info = &background_thread_info[i];
+		if (malloc_mutex_trylock(tsdn, &info->mtx)) {
+			/*
+			 * Each background thread run may take a long time;
+			 * avoid waiting on the stats if the thread is active.
+			 */
+			continue;
+		}
+		if (info->state != background_thread_stopped) {
+			num_runs += info->tot_n_runs;
+			nstime_add(&stats->run_interval, &info->tot_sleep_time);
+		}
+		malloc_mutex_unlock(tsdn, &info->mtx);
+	}
+	stats->num_runs = num_runs;
+	if (num_runs > 0) {
+		nstime_idivide(&stats->run_interval, num_runs);
+	}
+	malloc_mutex_unlock(tsdn, &background_thread_lock);
+
+	return false;
+}
+
+#undef BACKGROUND_THREAD_NPAGES_THRESHOLD
+#undef BILLION
+#undef BACKGROUND_THREAD_MIN_INTERVAL_NS
+
+#ifdef JEMALLOC_HAVE_DLSYM
+#include <dlfcn.h>
+#endif
+
+static bool
+pthread_create_fptr_init(void) {
+	if (pthread_create_fptr != NULL) {
+		return false;
+	}
+	/*
+	 * Try the next symbol first, because 1) when use lazy_lock we have a
+	 * wrapper for pthread_create; and 2) application may define its own
+	 * wrapper as well (and can call malloc within the wrapper).
+	 */
+#ifdef JEMALLOC_HAVE_DLSYM
+	pthread_create_fptr = dlsym(RTLD_NEXT, "pthread_create");
+#else
+	pthread_create_fptr = NULL;
+#endif
+	if (pthread_create_fptr == NULL) {
+		if (config_lazy_lock) {
+			malloc_write("<jemalloc>: Error in dlsym(RTLD_NEXT, "
+			    "\"pthread_create\")\n");
+			abort();
+		} else {
+			/* Fall back to the default symbol. */
+			pthread_create_fptr = pthread_create;
+		}
+	}
+
+	return false;
+}
+
+/*
+ * When lazy lock is enabled, we need to make sure setting isthreaded before
+ * taking any background_thread locks.  This is called early in ctl (instead of
+ * wait for the pthread_create calls to trigger) because the mutex is required
+ * before creating background threads.
+ */
+void
+background_thread_ctl_init(tsdn_t *tsdn) {
+	malloc_mutex_assert_not_owner(tsdn, &background_thread_lock);
+#ifdef JEMALLOC_PTHREAD_CREATE_WRAPPER
+	pthread_create_fptr_init();
+	pthread_create_wrapper_init();
+#endif
+}
+
+#endif /* defined(JEMALLOC_BACKGROUND_THREAD) */
+
+bool
+background_thread_boot0(void) {
+	if (!have_background_thread && opt_background_thread) {
+		malloc_printf("<jemalloc>: option background_thread currently "
+		    "supports pthread only\n");
+		return true;
+	}
+#ifdef JEMALLOC_PTHREAD_CREATE_WRAPPER
+	if ((config_lazy_lock || opt_background_thread) &&
+	    pthread_create_fptr_init()) {
+		return true;
+	}
+#endif
+	return false;
+}
+
+bool
+background_thread_boot1(tsdn_t *tsdn) {
+#ifdef JEMALLOC_BACKGROUND_THREAD
+	assert(have_background_thread);
+	assert(narenas_total_get() > 0);
+
+	if (opt_max_background_threads > MAX_BACKGROUND_THREAD_LIMIT) {
+		opt_max_background_threads = DEFAULT_NUM_BACKGROUND_THREAD;
+	}
+	max_background_threads = opt_max_background_threads;
+
+	background_thread_enabled_set(tsdn, opt_background_thread);
+	if (malloc_mutex_init(&background_thread_lock,
+	    "background_thread_global",
+	    WITNESS_RANK_BACKGROUND_THREAD_GLOBAL,
+	    malloc_mutex_rank_exclusive)) {
+		return true;
+	}
+
+	background_thread_info = (background_thread_info_t *)base_alloc(tsdn,
+	    b0get(), opt_max_background_threads *
+	    sizeof(background_thread_info_t), CACHELINE);
+	if (background_thread_info == NULL) {
+		return true;
+	}
+
+	for (unsigned i = 0; i < max_background_threads; i++) {
+		background_thread_info_t *info = &background_thread_info[i];
+		/* Thread mutex is rank_inclusive because of thread0. */
+		if (malloc_mutex_init(&info->mtx, "background_thread",
+		    WITNESS_RANK_BACKGROUND_THREAD,
+		    malloc_mutex_address_ordered)) {
+			return true;
+		}
+		if (pthread_cond_init(&info->cond, NULL)) {
+			return true;
+		}
+		malloc_mutex_lock(tsdn, &info->mtx);
+		info->state = background_thread_stopped;
+		background_thread_info_init(tsdn, info);
+		malloc_mutex_unlock(tsdn, &info->mtx);
+	}
+#endif
+
+	return false;
+}

deps/jemalloc/src/base.c

@@ -0,0 +1,514 @@
+#define JEMALLOC_BASE_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/extent_mmap.h"
+#include "jemalloc/internal/mutex.h"
+#include "jemalloc/internal/sz.h"
+
+/******************************************************************************/
+/* Data. */
+
+static base_t *b0;
+
+metadata_thp_mode_t opt_metadata_thp = METADATA_THP_DEFAULT;
+
+const char *metadata_thp_mode_names[] = {
+	"disabled",
+	"auto",
+	"always"
+};
+
+/******************************************************************************/
+
+static inline bool
+metadata_thp_madvise(void) {
+	return (metadata_thp_enabled() &&
+	    (init_system_thp_mode == thp_mode_default));
+}
+
+static void *
+base_map(tsdn_t *tsdn, extent_hooks_t *extent_hooks, unsigned ind, size_t size) {
+	void *addr;
+	bool zero = true;
+	bool commit = true;
+
+	/* Use huge page sizes and alignment regardless of opt_metadata_thp. */
+	assert(size == HUGEPAGE_CEILING(size));
+	size_t alignment = HUGEPAGE;
+	if (extent_hooks == &extent_hooks_default) {
+		addr = extent_alloc_mmap(NULL, size, alignment, &zero, &commit);
+	} else {
+		/* No arena context as we are creating new arenas. */
+		tsd_t *tsd = tsdn_null(tsdn) ? tsd_fetch() : tsdn_tsd(tsdn);
+		pre_reentrancy(tsd, NULL);
+		addr = extent_hooks->alloc(extent_hooks, NULL, size, alignment,
+		    &zero, &commit, ind);
+		post_reentrancy(tsd);
+	}
+
+	return addr;
+}
+
+static void
+base_unmap(tsdn_t *tsdn, extent_hooks_t *extent_hooks, unsigned ind, void *addr,
+    size_t size) {
+	/*
+	 * Cascade through dalloc, decommit, purge_forced, and purge_lazy,
+	 * stopping at first success.  This cascade is performed for consistency
+	 * with the cascade in extent_dalloc_wrapper() because an application's
+	 * custom hooks may not support e.g. dalloc.  This function is only ever
+	 * called as a side effect of arena destruction, so although it might
+	 * seem pointless to do anything besides dalloc here, the application
+	 * may in fact want the end state of all associated virtual memory to be
+	 * in some consistent-but-allocated state.
+	 */
+	if (extent_hooks == &extent_hooks_default) {
+		if (!extent_dalloc_mmap(addr, size)) {
+			goto label_done;
+		}
+		if (!pages_decommit(addr, size)) {
+			goto label_done;
+		}
+		if (!pages_purge_forced(addr, size)) {
+			goto label_done;
+		}
+		if (!pages_purge_lazy(addr, size)) {
+			goto label_done;
+		}
+		/* Nothing worked.  This should never happen. */
+		not_reached();
+	} else {
+		tsd_t *tsd = tsdn_null(tsdn) ? tsd_fetch() : tsdn_tsd(tsdn);
+		pre_reentrancy(tsd, NULL);
+		if (extent_hooks->dalloc != NULL &&
+		    !extent_hooks->dalloc(extent_hooks, addr, size, true,
+		    ind)) {
+			goto label_post_reentrancy;
+		}
+		if (extent_hooks->decommit != NULL &&
+		    !extent_hooks->decommit(extent_hooks, addr, size, 0, size,
+		    ind)) {
+			goto label_post_reentrancy;
+		}
+		if (extent_hooks->purge_forced != NULL &&
+		    !extent_hooks->purge_forced(extent_hooks, addr, size, 0,
+		    size, ind)) {
+			goto label_post_reentrancy;
+		}
+		if (extent_hooks->purge_lazy != NULL &&
+		    !extent_hooks->purge_lazy(extent_hooks, addr, size, 0, size,
+		    ind)) {
+			goto label_post_reentrancy;
+		}
+		/* Nothing worked.  That's the application's problem. */
+	label_post_reentrancy:
+		post_reentrancy(tsd);
+	}
+label_done:
+	if (metadata_thp_madvise()) {
+		/* Set NOHUGEPAGE after unmap to avoid kernel defrag. */
+		assert(((uintptr_t)addr & HUGEPAGE_MASK) == 0 &&
+		    (size & HUGEPAGE_MASK) == 0);
+		pages_nohuge(addr, size);
+	}
+}
+
+static void
+base_extent_init(size_t *extent_sn_next, extent_t *extent, void *addr,
+    size_t size) {
+	size_t sn;
+
+	sn = *extent_sn_next;
+	(*extent_sn_next)++;
+
+	extent_binit(extent, addr, size, sn);
+}
+
+static size_t
+base_get_num_blocks(base_t *base, bool with_new_block) {
+	base_block_t *b = base->blocks;
+	assert(b != NULL);
+
+	size_t n_blocks = with_new_block ? 2 : 1;
+	while (b->next != NULL) {
+		n_blocks++;
+		b = b->next;
+	}
+
+	return n_blocks;
+}
+
+static void
+base_auto_thp_switch(tsdn_t *tsdn, base_t *base) {
+	assert(opt_metadata_thp == metadata_thp_auto);
+	malloc_mutex_assert_owner(tsdn, &base->mtx);
+	if (base->auto_thp_switched) {
+		return;
+	}
+	/* Called when adding a new block. */
+	bool should_switch;
+	if (base_ind_get(base) != 0) {
+		should_switch = (base_get_num_blocks(base, true) ==
+		    BASE_AUTO_THP_THRESHOLD);
+	} else {
+		should_switch = (base_get_num_blocks(base, true) ==
+		    BASE_AUTO_THP_THRESHOLD_A0);
+	}
+	if (!should_switch) {
+		return;
+	}
+
+	base->auto_thp_switched = true;
+	assert(!config_stats || base->n_thp == 0);
+	/* Make the initial blocks THP lazily. */
+	base_block_t *block = base->blocks;
+	while (block != NULL) {
+		assert((block->size & HUGEPAGE_MASK) == 0);
+		pages_huge(block, block->size);
+		if (config_stats) {
+			base->n_thp += HUGEPAGE_CEILING(block->size -
+			    extent_bsize_get(&block->extent)) >> LG_HUGEPAGE;
+		}
+		block = block->next;
+		assert(block == NULL || (base_ind_get(base) == 0));
+	}
+}
+
+static void *
+base_extent_bump_alloc_helper(extent_t *extent, size_t *gap_size, size_t size,
+    size_t alignment) {
+	void *ret;
+
+	assert(alignment == ALIGNMENT_CEILING(alignment, QUANTUM));
+	assert(size == ALIGNMENT_CEILING(size, alignment));
+
+	*gap_size = ALIGNMENT_CEILING((uintptr_t)extent_addr_get(extent),
+	    alignment) - (uintptr_t)extent_addr_get(extent);
+	ret = (void *)((uintptr_t)extent_addr_get(extent) + *gap_size);
+	assert(extent_bsize_get(extent) >= *gap_size + size);
+	extent_binit(extent, (void *)((uintptr_t)extent_addr_get(extent) +
+	    *gap_size + size), extent_bsize_get(extent) - *gap_size - size,
+	    extent_sn_get(extent));
+	return ret;
+}
+
+static void
+base_extent_bump_alloc_post(base_t *base, extent_t *extent, size_t gap_size,
+    void *addr, size_t size) {
+	if (extent_bsize_get(extent) > 0) {
+		/*
+		 * Compute the index for the largest size class that does not
+		 * exceed extent's size.
+		 */
+		szind_t index_floor =
+		    sz_size2index(extent_bsize_get(extent) + 1) - 1;
+		extent_heap_insert(&base->avail[index_floor], extent);
+	}
+
+	if (config_stats) {
+		base->allocated += size;
+		/*
+		 * Add one PAGE to base_resident for every page boundary that is
+		 * crossed by the new allocation. Adjust n_thp similarly when
+		 * metadata_thp is enabled.
+		 */
+		base->resident += PAGE_CEILING((uintptr_t)addr + size) -
+		    PAGE_CEILING((uintptr_t)addr - gap_size);
+		assert(base->allocated <= base->resident);
+		assert(base->resident <= base->mapped);
+		if (metadata_thp_madvise() && (opt_metadata_thp ==
+		    metadata_thp_always || base->auto_thp_switched)) {
+			base->n_thp += (HUGEPAGE_CEILING((uintptr_t)addr + size)
+			    - HUGEPAGE_CEILING((uintptr_t)addr - gap_size)) >>
+			    LG_HUGEPAGE;
+			assert(base->mapped >= base->n_thp << LG_HUGEPAGE);
+		}
+	}
+}
+
+static void *
+base_extent_bump_alloc(base_t *base, extent_t *extent, size_t size,
+    size_t alignment) {
+	void *ret;
+	size_t gap_size;
+
+	ret = base_extent_bump_alloc_helper(extent, &gap_size, size, alignment);
+	base_extent_bump_alloc_post(base, extent, gap_size, ret, size);
+	return ret;
+}
+
+/*
+ * Allocate a block of virtual memory that is large enough to start with a
+ * base_block_t header, followed by an object of specified size and alignment.
+ * On success a pointer to the initialized base_block_t header is returned.
+ */
+static base_block_t *
+base_block_alloc(tsdn_t *tsdn, base_t *base, extent_hooks_t *extent_hooks,
+    unsigned ind, pszind_t *pind_last, size_t *extent_sn_next, size_t size,
+    size_t alignment) {
+	alignment = ALIGNMENT_CEILING(alignment, QUANTUM);
+	size_t usize = ALIGNMENT_CEILING(size, alignment);
+	size_t header_size = sizeof(base_block_t);
+	size_t gap_size = ALIGNMENT_CEILING(header_size, alignment) -
+	    header_size;
+	/*
+	 * Create increasingly larger blocks in order to limit the total number
+	 * of disjoint virtual memory ranges.  Choose the next size in the page
+	 * size class series (skipping size classes that are not a multiple of
+	 * HUGEPAGE), or a size large enough to satisfy the requested size and
+	 * alignment, whichever is larger.
+	 */
+	size_t min_block_size = HUGEPAGE_CEILING(sz_psz2u(header_size + gap_size
+	    + usize));
+	pszind_t pind_next = (*pind_last + 1 < sz_psz2ind(SC_LARGE_MAXCLASS)) ?
+	    *pind_last + 1 : *pind_last;
+	size_t next_block_size = HUGEPAGE_CEILING(sz_pind2sz(pind_next));
+	size_t block_size = (min_block_size > next_block_size) ? min_block_size
+	    : next_block_size;
+	base_block_t *block = (base_block_t *)base_map(tsdn, extent_hooks, ind,
+	    block_size);
+	if (block == NULL) {
+		return NULL;
+	}
+
+	if (metadata_thp_madvise()) {
+		void *addr = (void *)block;
+		assert(((uintptr_t)addr & HUGEPAGE_MASK) == 0 &&
+		    (block_size & HUGEPAGE_MASK) == 0);
+		if (opt_metadata_thp == metadata_thp_always) {
+			pages_huge(addr, block_size);
+		} else if (opt_metadata_thp == metadata_thp_auto &&
+		    base != NULL) {
+			/* base != NULL indicates this is not a new base. */
+			malloc_mutex_lock(tsdn, &base->mtx);
+			base_auto_thp_switch(tsdn, base);
+			if (base->auto_thp_switched) {
+				pages_huge(addr, block_size);
+			}
+			malloc_mutex_unlock(tsdn, &base->mtx);
+		}
+	}
+
+	*pind_last = sz_psz2ind(block_size);
+	block->size = block_size;
+	block->next = NULL;
+	assert(block_size >= header_size);
+	base_extent_init(extent_sn_next, &block->extent,
+	    (void *)((uintptr_t)block + header_size), block_size - header_size);
+	return block;
+}
+
+/*
+ * Allocate an extent that is at least as large as specified size, with
+ * specified alignment.
+ */
+static extent_t *
+base_extent_alloc(tsdn_t *tsdn, base_t *base, size_t size, size_t alignment) {
+	malloc_mutex_assert_owner(tsdn, &base->mtx);
+
+	extent_hooks_t *extent_hooks = base_extent_hooks_get(base);
+	/*
+	 * Drop mutex during base_block_alloc(), because an extent hook will be
+	 * called.
+	 */
+	malloc_mutex_unlock(tsdn, &base->mtx);
+	base_block_t *block = base_block_alloc(tsdn, base, extent_hooks,
+	    base_ind_get(base), &base->pind_last, &base->extent_sn_next, size,
+	    alignment);
+	malloc_mutex_lock(tsdn, &base->mtx);
+	if (block == NULL) {
+		return NULL;
+	}
+	block->next = base->blocks;
+	base->blocks = block;
+	if (config_stats) {
+		base->allocated += sizeof(base_block_t);
+		base->resident += PAGE_CEILING(sizeof(base_block_t));
+		base->mapped += block->size;
+		if (metadata_thp_madvise() &&
+		    !(opt_metadata_thp == metadata_thp_auto
+		      && !base->auto_thp_switched)) {
+			assert(base->n_thp > 0);
+			base->n_thp += HUGEPAGE_CEILING(sizeof(base_block_t)) >>
+			    LG_HUGEPAGE;
+		}
+		assert(base->allocated <= base->resident);
+		assert(base->resident <= base->mapped);
+		assert(base->n_thp << LG_HUGEPAGE <= base->mapped);
+	}
+	return &block->extent;
+}
+
+base_t *
+b0get(void) {
+	return b0;
+}
+
+base_t *
+base_new(tsdn_t *tsdn, unsigned ind, extent_hooks_t *extent_hooks) {
+	pszind_t pind_last = 0;
+	size_t extent_sn_next = 0;
+	base_block_t *block = base_block_alloc(tsdn, NULL, extent_hooks, ind,
+	    &pind_last, &extent_sn_next, sizeof(base_t), QUANTUM);
+	if (block == NULL) {
+		return NULL;
+	}
+
+	size_t gap_size;
+	size_t base_alignment = CACHELINE;
+	size_t base_size = ALIGNMENT_CEILING(sizeof(base_t), base_alignment);
+	base_t *base = (base_t *)base_extent_bump_alloc_helper(&block->extent,
+	    &gap_size, base_size, base_alignment);
+	base->ind = ind;
+	atomic_store_p(&base->extent_hooks, extent_hooks, ATOMIC_RELAXED);
+	if (malloc_mutex_init(&base->mtx, "base", WITNESS_RANK_BASE,
+	    malloc_mutex_rank_exclusive)) {
+		base_unmap(tsdn, extent_hooks, ind, block, block->size);
+		return NULL;
+	}
+	base->pind_last = pind_last;
+	base->extent_sn_next = extent_sn_next;
+	base->blocks = block;
+	base->auto_thp_switched = false;
+	for (szind_t i = 0; i < SC_NSIZES; i++) {
+		extent_heap_new(&base->avail[i]);
+	}
+	if (config_stats) {
+		base->allocated = sizeof(base_block_t);
+		base->resident = PAGE_CEILING(sizeof(base_block_t));
+		base->mapped = block->size;
+		base->n_thp = (opt_metadata_thp == metadata_thp_always) &&
+		    metadata_thp_madvise() ? HUGEPAGE_CEILING(sizeof(base_block_t))
+		    >> LG_HUGEPAGE : 0;
+		assert(base->allocated <= base->resident);
+		assert(base->resident <= base->mapped);
+		assert(base->n_thp << LG_HUGEPAGE <= base->mapped);
+	}
+	base_extent_bump_alloc_post(base, &block->extent, gap_size, base,
+	    base_size);
+
+	return base;
+}
+
+void
+base_delete(tsdn_t *tsdn, base_t *base) {
+	extent_hooks_t *extent_hooks = base_extent_hooks_get(base);
+	base_block_t *next = base->blocks;
+	do {
+		base_block_t *block = next;
+		next = block->next;
+		base_unmap(tsdn, extent_hooks, base_ind_get(base), block,
+		    block->size);
+	} while (next != NULL);
+}
+
+extent_hooks_t *
+base_extent_hooks_get(base_t *base) {
+	return (extent_hooks_t *)atomic_load_p(&base->extent_hooks,
+	    ATOMIC_ACQUIRE);
+}
+
+extent_hooks_t *
+base_extent_hooks_set(base_t *base, extent_hooks_t *extent_hooks) {
+	extent_hooks_t *old_extent_hooks = base_extent_hooks_get(base);
+	atomic_store_p(&base->extent_hooks, extent_hooks, ATOMIC_RELEASE);
+	return old_extent_hooks;
+}
+
+static void *
+base_alloc_impl(tsdn_t *tsdn, base_t *base, size_t size, size_t alignment,
+    size_t *esn) {
+	alignment = QUANTUM_CEILING(alignment);
+	size_t usize = ALIGNMENT_CEILING(size, alignment);
+	size_t asize = usize + alignment - QUANTUM;
+
+	extent_t *extent = NULL;
+	malloc_mutex_lock(tsdn, &base->mtx);
+	for (szind_t i = sz_size2index(asize); i < SC_NSIZES; i++) {
+		extent = extent_heap_remove_first(&base->avail[i]);
+		if (extent != NULL) {
+			/* Use existing space. */
+			break;
+		}
+	}
+	if (extent == NULL) {
+		/* Try to allocate more space. */
+		extent = base_extent_alloc(tsdn, base, usize, alignment);
+	}
+	void *ret;
+	if (extent == NULL) {
+		ret = NULL;
+		goto label_return;
+	}
+
+	ret = base_extent_bump_alloc(base, extent, usize, alignment);
+	if (esn != NULL) {
+		*esn = extent_sn_get(extent);
+	}
+label_return:
+	malloc_mutex_unlock(tsdn, &base->mtx);
+	return ret;
+}
+
+/*
+ * base_alloc() returns zeroed memory, which is always demand-zeroed for the
+ * auto arenas, in order to make multi-page sparse data structures such as radix
+ * tree nodes efficient with respect to physical memory usage.  Upon success a
+ * pointer to at least size bytes with specified alignment is returned.  Note
+ * that size is rounded up to the nearest multiple of alignment to avoid false
+ * sharing.
+ */
+void *
+base_alloc(tsdn_t *tsdn, base_t *base, size_t size, size_t alignment) {
+	return base_alloc_impl(tsdn, base, size, alignment, NULL);
+}
+
+extent_t *
+base_alloc_extent(tsdn_t *tsdn, base_t *base) {
+	size_t esn;
+	extent_t *extent = base_alloc_impl(tsdn, base, sizeof(extent_t),
+	    CACHELINE, &esn);
+	if (extent == NULL) {
+		return NULL;
+	}
+	extent_esn_set(extent, esn);
+	return extent;
+}
+
+void
+base_stats_get(tsdn_t *tsdn, base_t *base, size_t *allocated, size_t *resident,
+    size_t *mapped, size_t *n_thp) {
+	cassert(config_stats);
+
+	malloc_mutex_lock(tsdn, &base->mtx);
+	assert(base->allocated <= base->resident);
+	assert(base->resident <= base->mapped);
+	*allocated = base->allocated;
+	*resident = base->resident;
+	*mapped = base->mapped;
+	*n_thp = base->n_thp;
+	malloc_mutex_unlock(tsdn, &base->mtx);
+}
+
+void
+base_prefork(tsdn_t *tsdn, base_t *base) {
+	malloc_mutex_prefork(tsdn, &base->mtx);
+}
+
+void
+base_postfork_parent(tsdn_t *tsdn, base_t *base) {
+	malloc_mutex_postfork_parent(tsdn, &base->mtx);
+}
+
+void
+base_postfork_child(tsdn_t *tsdn, base_t *base) {
+	malloc_mutex_postfork_child(tsdn, &base->mtx);
+}
+
+bool
+base_boot(tsdn_t *tsdn) {
+	b0 = base_new(tsdn, 0, (extent_hooks_t *)&extent_hooks_default);
+	return (b0 == NULL);
+}

deps/jemalloc/src/bin.c

@@ -0,0 +1,95 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/bin.h"
+#include "jemalloc/internal/sc.h"
+#include "jemalloc/internal/witness.h"
+
+bin_info_t bin_infos[SC_NBINS];
+
+static void
+bin_infos_init(sc_data_t *sc_data, unsigned bin_shard_sizes[SC_NBINS],
+    bin_info_t bin_infos[SC_NBINS]) {
+	for (unsigned i = 0; i < SC_NBINS; i++) {
+		bin_info_t *bin_info = &bin_infos[i];
+		sc_t *sc = &sc_data->sc[i];
+		bin_info->reg_size = ((size_t)1U << sc->lg_base)
+		    + ((size_t)sc->ndelta << sc->lg_delta);
+		bin_info->slab_size = (sc->pgs << LG_PAGE);
+		bin_info->nregs =
+		    (uint32_t)(bin_info->slab_size / bin_info->reg_size);
+		bin_info->n_shards = bin_shard_sizes[i];
+		bitmap_info_t bitmap_info = BITMAP_INFO_INITIALIZER(
+		    bin_info->nregs);
+		bin_info->bitmap_info = bitmap_info;
+	}
+}
+
+bool
+bin_update_shard_size(unsigned bin_shard_sizes[SC_NBINS], size_t start_size,
+    size_t end_size, size_t nshards) {
+	if (nshards > BIN_SHARDS_MAX || nshards == 0) {
+		return true;
+	}
+
+	if (start_size > SC_SMALL_MAXCLASS) {
+		return false;
+	}
+	if (end_size > SC_SMALL_MAXCLASS) {
+		end_size = SC_SMALL_MAXCLASS;
+	}
+
+	/* Compute the index since this may happen before sz init. */
+	szind_t ind1 = sz_size2index_compute(start_size);
+	szind_t ind2 = sz_size2index_compute(end_size);
+	for (unsigned i = ind1; i <= ind2; i++) {
+		bin_shard_sizes[i] = (unsigned)nshards;
+	}
+
+	return false;
+}
+
+void
+bin_shard_sizes_boot(unsigned bin_shard_sizes[SC_NBINS]) {
+	/* Load the default number of shards. */
+	for (unsigned i = 0; i < SC_NBINS; i++) {
+		bin_shard_sizes[i] = N_BIN_SHARDS_DEFAULT;
+	}
+}
+
+void
+bin_boot(sc_data_t *sc_data, unsigned bin_shard_sizes[SC_NBINS]) {
+	assert(sc_data->initialized);
+	bin_infos_init(sc_data, bin_shard_sizes, bin_infos);
+}
+
+bool
+bin_init(bin_t *bin) {
+	if (malloc_mutex_init(&bin->lock, "bin", WITNESS_RANK_BIN,
+	    malloc_mutex_rank_exclusive)) {
+		return true;
+	}
+	bin->slabcur = NULL;
+	extent_heap_new(&bin->slabs_nonfull);
+	extent_list_init(&bin->slabs_full);
+	if (config_stats) {
+		memset(&bin->stats, 0, sizeof(bin_stats_t));
+	}
+	return false;
+}
+
+void
+bin_prefork(tsdn_t *tsdn, bin_t *bin) {
+	malloc_mutex_prefork(tsdn, &bin->lock);
+}
+
+void
+bin_postfork_parent(tsdn_t *tsdn, bin_t *bin) {
+	malloc_mutex_postfork_parent(tsdn, &bin->lock);
+}
+
+void
+bin_postfork_child(tsdn_t *tsdn, bin_t *bin) {
+	malloc_mutex_postfork_child(tsdn, &bin->lock);
+}

deps/jemalloc/src/bitmap.c

@@ -0,0 +1,121 @@
+#define JEMALLOC_BITMAP_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+
+/******************************************************************************/
+
+#ifdef BITMAP_USE_TREE
+
+void
+bitmap_info_init(bitmap_info_t *binfo, size_t nbits) {
+	unsigned i;
+	size_t group_count;
+
+	assert(nbits > 0);
+	assert(nbits <= (ZU(1) << LG_BITMAP_MAXBITS));
+
+	/*
+	 * Compute the number of groups necessary to store nbits bits, and
+	 * progressively work upward through the levels until reaching a level
+	 * that requires only one group.
+	 */
+	binfo->levels[0].group_offset = 0;
+	group_count = BITMAP_BITS2GROUPS(nbits);
+	for (i = 1; group_count > 1; i++) {
+		assert(i < BITMAP_MAX_LEVELS);
+		binfo->levels[i].group_offset = binfo->levels[i-1].group_offset
+		    + group_count;
+		group_count = BITMAP_BITS2GROUPS(group_count);
+	}
+	binfo->levels[i].group_offset = binfo->levels[i-1].group_offset
+	    + group_count;
+	assert(binfo->levels[i].group_offset <= BITMAP_GROUPS_MAX);
+	binfo->nlevels = i;
+	binfo->nbits = nbits;
+}
+
+static size_t
+bitmap_info_ngroups(const bitmap_info_t *binfo) {
+	return binfo->levels[binfo->nlevels].group_offset;
+}
+
+void
+bitmap_init(bitmap_t *bitmap, const bitmap_info_t *binfo, bool fill) {
+	size_t extra;
+	unsigned i;
+
+	/*
+	 * Bits are actually inverted with regard to the external bitmap
+	 * interface.
+	 */
+
+	if (fill) {
+		/* The "filled" bitmap starts out with all 0 bits. */
+		memset(bitmap, 0, bitmap_size(binfo));
+		return;
+	}
+
+	/*
+	 * The "empty" bitmap starts out with all 1 bits, except for trailing
+	 * unused bits (if any).  Note that each group uses bit 0 to correspond
+	 * to the first logical bit in the group, so extra bits are the most
+	 * significant bits of the last group.
+	 */
+	memset(bitmap, 0xffU, bitmap_size(binfo));
+	extra = (BITMAP_GROUP_NBITS - (binfo->nbits & BITMAP_GROUP_NBITS_MASK))
+	    & BITMAP_GROUP_NBITS_MASK;
+	if (extra != 0) {
+		bitmap[binfo->levels[1].group_offset - 1] >>= extra;
+	}
+	for (i = 1; i < binfo->nlevels; i++) {
+		size_t group_count = binfo->levels[i].group_offset -
+		    binfo->levels[i-1].group_offset;
+		extra = (BITMAP_GROUP_NBITS - (group_count &
+		    BITMAP_GROUP_NBITS_MASK)) & BITMAP_GROUP_NBITS_MASK;
+		if (extra != 0) {
+			bitmap[binfo->levels[i+1].group_offset - 1] >>= extra;
+		}
+	}
+}
+
+#else /* BITMAP_USE_TREE */
+
+void
+bitmap_info_init(bitmap_info_t *binfo, size_t nbits) {
+	assert(nbits > 0);
+	assert(nbits <= (ZU(1) << LG_BITMAP_MAXBITS));
+
+	binfo->ngroups = BITMAP_BITS2GROUPS(nbits);
+	binfo->nbits = nbits;
+}
+
+static size_t
+bitmap_info_ngroups(const bitmap_info_t *binfo) {
+	return binfo->ngroups;
+}
+
+void
+bitmap_init(bitmap_t *bitmap, const bitmap_info_t *binfo, bool fill) {
+	size_t extra;
+
+	if (fill) {
+		memset(bitmap, 0, bitmap_size(binfo));
+		return;
+	}
+
+	memset(bitmap, 0xffU, bitmap_size(binfo));
+	extra = (BITMAP_GROUP_NBITS - (binfo->nbits & BITMAP_GROUP_NBITS_MASK))
+	    & BITMAP_GROUP_NBITS_MASK;
+	if (extra != 0) {
+		bitmap[binfo->ngroups - 1] >>= extra;
+	}
+}
+
+#endif /* BITMAP_USE_TREE */
+
+size_t
+bitmap_size(const bitmap_info_t *binfo) {
+	return (bitmap_info_ngroups(binfo) << LG_SIZEOF_BITMAP);
+}

deps/jemalloc/src/chunk.c

@@ -0,0 +1,395 @@
+#define	JEMALLOC_CHUNK_C_
+#include "jemalloc/internal/jemalloc_internal.h"
+
+/******************************************************************************/
+/* Data. */
+
+const char	*opt_dss = DSS_DEFAULT;
+size_t		opt_lg_chunk = LG_CHUNK_DEFAULT;
+
+malloc_mutex_t	chunks_mtx;
+chunk_stats_t	stats_chunks;
+
+/*
+ * Trees of chunks that were previously allocated (trees differ only in node
+ * ordering).  These are used when allocating chunks, in an attempt to re-use
+ * address space.  Depending on function, different tree orderings are needed,
+ * which is why there are two trees with the same contents.
+ */
+static extent_tree_t	chunks_szad_mmap;
+static extent_tree_t	chunks_ad_mmap;
+static extent_tree_t	chunks_szad_dss;
+static extent_tree_t	chunks_ad_dss;
+
+rtree_t		*chunks_rtree;
+
+/* Various chunk-related settings. */
+size_t		chunksize;
+size_t		chunksize_mask; /* (chunksize - 1). */
+size_t		chunk_npages;
+size_t		map_bias;
+size_t		arena_maxclass; /* Max size class for arenas. */
+
+/******************************************************************************/
+/* Function prototypes for non-inline static functions. */
+
+static void	*chunk_recycle(extent_tree_t *chunks_szad,
+    extent_tree_t *chunks_ad, size_t size, size_t alignment, bool base,
+    bool *zero);
+static void	chunk_record(extent_tree_t *chunks_szad,
+    extent_tree_t *chunks_ad, void *chunk, size_t size);
+
+/******************************************************************************/
+
+static void *
+chunk_recycle(extent_tree_t *chunks_szad, extent_tree_t *chunks_ad, size_t size,
+    size_t alignment, bool base, bool *zero)
+{
+	void *ret;
+	extent_node_t *node;
+	extent_node_t key;
+	size_t alloc_size, leadsize, trailsize;
+	bool zeroed;
+
+	if (base) {
+		/*
+		 * This function may need to call base_node_{,de}alloc(), but
+		 * the current chunk allocation request is on behalf of the
+		 * base allocator.  Avoid deadlock (and if that weren't an
+		 * issue, potential for infinite recursion) by returning NULL.
+		 */
+		return (NULL);
+	}
+
+	alloc_size = size + alignment - chunksize;
+	/* Beware size_t wrap-around. */
+	if (alloc_size < size)
+		return (NULL);
+	key.addr = NULL;
+	key.size = alloc_size;
+	malloc_mutex_lock(&chunks_mtx);
+	node = extent_tree_szad_nsearch(chunks_szad, &key);
+	if (node == NULL) {
+		malloc_mutex_unlock(&chunks_mtx);
+		return (NULL);
+	}
+	leadsize = ALIGNMENT_CEILING((uintptr_t)node->addr, alignment) -
+	    (uintptr_t)node->addr;
+	assert(node->size >= leadsize + size);
+	trailsize = node->size - leadsize - size;
+	ret = (void *)((uintptr_t)node->addr + leadsize);
+	zeroed = node->zeroed;
+	if (zeroed)
+	    *zero = true;
+	/* Remove node from the tree. */
+	extent_tree_szad_remove(chunks_szad, node);
+	extent_tree_ad_remove(chunks_ad, node);
+	if (leadsize != 0) {
+		/* Insert the leading space as a smaller chunk. */
+		node->size = leadsize;
+		extent_tree_szad_insert(chunks_szad, node);
+		extent_tree_ad_insert(chunks_ad, node);
+		node = NULL;
+	}
+	if (trailsize != 0) {
+		/* Insert the trailing space as a smaller chunk. */
+		if (node == NULL) {
+			/*
+			 * An additional node is required, but
+			 * base_node_alloc() can cause a new base chunk to be
+			 * allocated.  Drop chunks_mtx in order to avoid
+			 * deadlock, and if node allocation fails, deallocate
+			 * the result before returning an error.
+			 */
+			malloc_mutex_unlock(&chunks_mtx);
+			node = base_node_alloc();
+			if (node == NULL) {
+				chunk_dealloc(ret, size, true);
+				return (NULL);
+			}
+			malloc_mutex_lock(&chunks_mtx);
+		}
+		node->addr = (void *)((uintptr_t)(ret) + size);
+		node->size = trailsize;
+		node->zeroed = zeroed;
+		extent_tree_szad_insert(chunks_szad, node);
+		extent_tree_ad_insert(chunks_ad, node);
+		node = NULL;
+	}
+	malloc_mutex_unlock(&chunks_mtx);
+
+	if (node != NULL)
+		base_node_dealloc(node);
+	if (*zero) {
+		if (zeroed == false)
+			memset(ret, 0, size);
+		else if (config_debug) {
+			size_t i;
+			size_t *p = (size_t *)(uintptr_t)ret;
+
+			VALGRIND_MAKE_MEM_DEFINED(ret, size);
+			for (i = 0; i < size / sizeof(size_t); i++)
+				assert(p[i] == 0);
+		}
+	}
+	return (ret);
+}
+
+/*
+ * If the caller specifies (*zero == false), it is still possible to receive
+ * zeroed memory, in which case *zero is toggled to true.  arena_chunk_alloc()
+ * takes advantage of this to avoid demanding zeroed chunks, but taking
+ * advantage of them if they are returned.
+ */
+void *
+chunk_alloc(size_t size, size_t alignment, bool base, bool *zero,
+    dss_prec_t dss_prec)
+{
+	void *ret;
+
+	assert(size != 0);
+	assert((size & chunksize_mask) == 0);
+	assert(alignment != 0);
+	assert((alignment & chunksize_mask) == 0);
+
+	/* "primary" dss. */
+	if (config_dss && dss_prec == dss_prec_primary) {
+		if ((ret = chunk_recycle(&chunks_szad_dss, &chunks_ad_dss, size,
+		    alignment, base, zero)) != NULL)
+			goto label_return;
+		if ((ret = chunk_alloc_dss(size, alignment, zero)) != NULL)
+			goto label_return;
+	}
+	/* mmap. */
+	if ((ret = chunk_recycle(&chunks_szad_mmap, &chunks_ad_mmap, size,
+	    alignment, base, zero)) != NULL)
+		goto label_return;
+	if ((ret = chunk_alloc_mmap(size, alignment, zero)) != NULL)
+		goto label_return;
+	/* "secondary" dss. */
+	if (config_dss && dss_prec == dss_prec_secondary) {
+		if ((ret = chunk_recycle(&chunks_szad_dss, &chunks_ad_dss, size,
+		    alignment, base, zero)) != NULL)
+			goto label_return;
+		if ((ret = chunk_alloc_dss(size, alignment, zero)) != NULL)
+			goto label_return;
+	}
+
+	/* All strategies for allocation failed. */
+	ret = NULL;
+label_return:
+	if (ret != NULL) {
+		if (config_ivsalloc && base == false) {
+			if (rtree_set(chunks_rtree, (uintptr_t)ret, 1)) {
+				chunk_dealloc(ret, size, true);
+				return (NULL);
+			}
+		}
+		if (config_stats || config_prof) {
+			bool gdump;
+			malloc_mutex_lock(&chunks_mtx);
+			if (config_stats)
+				stats_chunks.nchunks += (size / chunksize);
+			stats_chunks.curchunks += (size / chunksize);
+			if (stats_chunks.curchunks > stats_chunks.highchunks) {
+				stats_chunks.highchunks =
+				    stats_chunks.curchunks;
+				if (config_prof)
+					gdump = true;
+			} else if (config_prof)
+				gdump = false;
+			malloc_mutex_unlock(&chunks_mtx);
+			if (config_prof && opt_prof && opt_prof_gdump && gdump)
+				prof_gdump();
+		}
+		if (config_valgrind)
+			VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
+	}
+	assert(CHUNK_ADDR2BASE(ret) == ret);
+	return (ret);
+}
+
+static void
+chunk_record(extent_tree_t *chunks_szad, extent_tree_t *chunks_ad, void *chunk,
+    size_t size)
+{
+	bool unzeroed;
+	extent_node_t *xnode, *node, *prev, *xprev, key;
+
+	unzeroed = pages_purge(chunk, size);
+	VALGRIND_MAKE_MEM_NOACCESS(chunk, size);
+
+	/*
+	 * Allocate a node before acquiring chunks_mtx even though it might not
+	 * be needed, because base_node_alloc() may cause a new base chunk to
+	 * be allocated, which could cause deadlock if chunks_mtx were already
+	 * held.
+	 */
+	xnode = base_node_alloc();
+	/* Use xprev to implement conditional deferred deallocation of prev. */
+	xprev = NULL;
+
+	malloc_mutex_lock(&chunks_mtx);
+	key.addr = (void *)((uintptr_t)chunk + size);
+	node = extent_tree_ad_nsearch(chunks_ad, &key);
+	/* Try to coalesce forward. */
+	if (node != NULL && node->addr == key.addr) {
+		/*
+		 * Coalesce chunk with the following address range.  This does
+		 * not change the position within chunks_ad, so only
+		 * remove/insert from/into chunks_szad.
+		 */
+		extent_tree_szad_remove(chunks_szad, node);
+		node->addr = chunk;
+		node->size += size;
+		node->zeroed = (node->zeroed && (unzeroed == false));
+		extent_tree_szad_insert(chunks_szad, node);
+	} else {
+		/* Coalescing forward failed, so insert a new node. */
+		if (xnode == NULL) {
+			/*
+			 * base_node_alloc() failed, which is an exceedingly
+			 * unlikely failure.  Leak chunk; its pages have
+			 * already been purged, so this is only a virtual
+			 * memory leak.
+			 */
+			goto label_return;
+		}
+		node = xnode;
+		xnode = NULL; /* Prevent deallocation below. */
+		node->addr = chunk;
+		node->size = size;
+		node->zeroed = (unzeroed == false);
+		extent_tree_ad_insert(chunks_ad, node);
+		extent_tree_szad_insert(chunks_szad, node);
+	}
+
+	/* Try to coalesce backward. */
+	prev = extent_tree_ad_prev(chunks_ad, node);
+	if (prev != NULL && (void *)((uintptr_t)prev->addr + prev->size) ==
+	    chunk) {
+		/*
+		 * Coalesce chunk with the previous address range.  This does
+		 * not change the position within chunks_ad, so only
+		 * remove/insert node from/into chunks_szad.
+		 */
+		extent_tree_szad_remove(chunks_szad, prev);
+		extent_tree_ad_remove(chunks_ad, prev);
+
+		extent_tree_szad_remove(chunks_szad, node);
+		node->addr = prev->addr;
+		node->size += prev->size;
+		node->zeroed = (node->zeroed && prev->zeroed);
+		extent_tree_szad_insert(chunks_szad, node);
+
+		xprev = prev;
+	}
+
+label_return:
+	malloc_mutex_unlock(&chunks_mtx);
+	/*
+	 * Deallocate xnode and/or xprev after unlocking chunks_mtx in order to
+	 * avoid potential deadlock.
+	 */
+	if (xnode != NULL)
+		base_node_dealloc(xnode);
+	if (xprev != NULL)
+		base_node_dealloc(xprev);
+}
+
+void
+chunk_unmap(void *chunk, size_t size)
+{
+	assert(chunk != NULL);
+	assert(CHUNK_ADDR2BASE(chunk) == chunk);
+	assert(size != 0);
+	assert((size & chunksize_mask) == 0);
+
+	if (config_dss && chunk_in_dss(chunk))
+		chunk_record(&chunks_szad_dss, &chunks_ad_dss, chunk, size);
+	else if (chunk_dealloc_mmap(chunk, size))
+		chunk_record(&chunks_szad_mmap, &chunks_ad_mmap, chunk, size);
+}
+
+void
+chunk_dealloc(void *chunk, size_t size, bool unmap)
+{
+
+	assert(chunk != NULL);
+	assert(CHUNK_ADDR2BASE(chunk) == chunk);
+	assert(size != 0);
+	assert((size & chunksize_mask) == 0);
+
+	if (config_ivsalloc)
+		rtree_set(chunks_rtree, (uintptr_t)chunk, 0);
+	if (config_stats || config_prof) {
+		malloc_mutex_lock(&chunks_mtx);
+		assert(stats_chunks.curchunks >= (size / chunksize));
+		stats_chunks.curchunks -= (size / chunksize);
+		malloc_mutex_unlock(&chunks_mtx);
+	}
+
+	if (unmap)
+		chunk_unmap(chunk, size);
+}
+
+bool
+chunk_boot(void)
+{
+
+	/* Set variables according to the value of opt_lg_chunk. */
+	chunksize = (ZU(1) << opt_lg_chunk);
+	assert(chunksize >= PAGE);
+	chunksize_mask = chunksize - 1;
+	chunk_npages = (chunksize >> LG_PAGE);
+
+	if (config_stats || config_prof) {
+		if (malloc_mutex_init(&chunks_mtx))
+			return (true);
+		memset(&stats_chunks, 0, sizeof(chunk_stats_t));
+	}
+	if (config_dss && chunk_dss_boot())
+		return (true);
+	extent_tree_szad_new(&chunks_szad_mmap);
+	extent_tree_ad_new(&chunks_ad_mmap);
+	extent_tree_szad_new(&chunks_szad_dss);
+	extent_tree_ad_new(&chunks_ad_dss);
+	if (config_ivsalloc) {
+		chunks_rtree = rtree_new((ZU(1) << (LG_SIZEOF_PTR+3)) -
+		    opt_lg_chunk, base_alloc, NULL);
+		if (chunks_rtree == NULL)
+			return (true);
+	}
+
+	return (false);
+}
+
+void
+chunk_prefork(void)
+{
+
+	malloc_mutex_prefork(&chunks_mtx);
+	if (config_ivsalloc)
+		rtree_prefork(chunks_rtree);
+	chunk_dss_prefork();
+}
+
+void
+chunk_postfork_parent(void)
+{
+
+	chunk_dss_postfork_parent();
+	if (config_ivsalloc)
+		rtree_postfork_parent(chunks_rtree);
+	malloc_mutex_postfork_parent(&chunks_mtx);
+}
+
+void
+chunk_postfork_child(void)
+{
+
+	chunk_dss_postfork_child();
+	if (config_ivsalloc)
+		rtree_postfork_child(chunks_rtree);
+	malloc_mutex_postfork_child(&chunks_mtx);
+}

deps/jemalloc/src/chunk_dss.c

@@ -0,0 +1,198 @@
+#define	JEMALLOC_CHUNK_DSS_C_
+#include "jemalloc/internal/jemalloc_internal.h"
+/******************************************************************************/
+/* Data. */
+
+const char	*dss_prec_names[] = {
+	"disabled",
+	"primary",
+	"secondary",
+	"N/A"
+};
+
+/* Current dss precedence default, used when creating new arenas. */
+static dss_prec_t	dss_prec_default = DSS_PREC_DEFAULT;
+
+/*
+ * Protects sbrk() calls.  This avoids malloc races among threads, though it
+ * does not protect against races with threads that call sbrk() directly.
+ */
+static malloc_mutex_t	dss_mtx;
+
+/* Base address of the DSS. */
+static void		*dss_base;
+/* Current end of the DSS, or ((void *)-1) if the DSS is exhausted. */
+static void		*dss_prev;
+/* Current upper limit on DSS addresses. */
+static void		*dss_max;
+
+/******************************************************************************/
+
+static void *
+chunk_dss_sbrk(intptr_t increment)
+{
+
+#ifdef JEMALLOC_HAVE_SBRK
+	return (sbrk(increment));
+#else
+	not_implemented();
+	return (NULL);
+#endif
+}
+
+dss_prec_t
+chunk_dss_prec_get(void)
+{
+	dss_prec_t ret;
+
+	if (config_dss == false)
+		return (dss_prec_disabled);
+	malloc_mutex_lock(&dss_mtx);
+	ret = dss_prec_default;
+	malloc_mutex_unlock(&dss_mtx);
+	return (ret);
+}
+
+bool
+chunk_dss_prec_set(dss_prec_t dss_prec)
+{
+
+	if (config_dss == false)
+		return (true);
+	malloc_mutex_lock(&dss_mtx);
+	dss_prec_default = dss_prec;
+	malloc_mutex_unlock(&dss_mtx);
+	return (false);
+}
+
+void *
+chunk_alloc_dss(size_t size, size_t alignment, bool *zero)
+{
+	void *ret;
+
+	cassert(config_dss);
+	assert(size > 0 && (size & chunksize_mask) == 0);
+	assert(alignment > 0 && (alignment & chunksize_mask) == 0);
+
+	/*
+	 * sbrk() uses a signed increment argument, so take care not to
+	 * interpret a huge allocation request as a negative increment.
+	 */
+	if ((intptr_t)size < 0)
+		return (NULL);
+
+	malloc_mutex_lock(&dss_mtx);
+	if (dss_prev != (void *)-1) {
+		size_t gap_size, cpad_size;
+		void *cpad, *dss_next;
+		intptr_t incr;
+
+		/*
+		 * The loop is necessary to recover from races with other
+		 * threads that are using the DSS for something other than
+		 * malloc.
+		 */
+		do {
+			/* Get the current end of the DSS. */
+			dss_max = chunk_dss_sbrk(0);
+			/*
+			 * Calculate how much padding is necessary to
+			 * chunk-align the end of the DSS.
+			 */
+			gap_size = (chunksize - CHUNK_ADDR2OFFSET(dss_max)) &
+			    chunksize_mask;
+			/*
+			 * Compute how much chunk-aligned pad space (if any) is
+			 * necessary to satisfy alignment.  This space can be
+			 * recycled for later use.
+			 */
+			cpad = (void *)((uintptr_t)dss_max + gap_size);
+			ret = (void *)ALIGNMENT_CEILING((uintptr_t)dss_max,
+			    alignment);
+			cpad_size = (uintptr_t)ret - (uintptr_t)cpad;
+			dss_next = (void *)((uintptr_t)ret + size);
+			if ((uintptr_t)ret < (uintptr_t)dss_max ||
+			    (uintptr_t)dss_next < (uintptr_t)dss_max) {
+				/* Wrap-around. */
+				malloc_mutex_unlock(&dss_mtx);
+				return (NULL);
+			}
+			incr = gap_size + cpad_size + size;
+			dss_prev = chunk_dss_sbrk(incr);
+			if (dss_prev == dss_max) {
+				/* Success. */
+				dss_max = dss_next;
+				malloc_mutex_unlock(&dss_mtx);
+				if (cpad_size != 0)
+					chunk_unmap(cpad, cpad_size);
+				if (*zero) {
+					VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
+					memset(ret, 0, size);
+				}
+				return (ret);
+			}
+		} while (dss_prev != (void *)-1);
+	}
+	malloc_mutex_unlock(&dss_mtx);
+
+	return (NULL);
+}
+
+bool
+chunk_in_dss(void *chunk)
+{
+	bool ret;
+
+	cassert(config_dss);
+
+	malloc_mutex_lock(&dss_mtx);
+	if ((uintptr_t)chunk >= (uintptr_t)dss_base
+	    && (uintptr_t)chunk < (uintptr_t)dss_max)
+		ret = true;
+	else
+		ret = false;
+	malloc_mutex_unlock(&dss_mtx);
+
+	return (ret);
+}
+
+bool
+chunk_dss_boot(void)
+{
+
+	cassert(config_dss);
+
+	if (malloc_mutex_init(&dss_mtx))
+		return (true);
+	dss_base = chunk_dss_sbrk(0);
+	dss_prev = dss_base;
+	dss_max = dss_base;
+
+	return (false);
+}
+
+void
+chunk_dss_prefork(void)
+{
+
+	if (config_dss)
+		malloc_mutex_prefork(&dss_mtx);
+}
+
+void
+chunk_dss_postfork_parent(void)
+{
+
+	if (config_dss)
+		malloc_mutex_postfork_parent(&dss_mtx);
+}
+
+void
+chunk_dss_postfork_child(void)
+{
+
+	if (config_dss)
+		malloc_mutex_postfork_child(&dss_mtx);
+}
+
+/******************************************************************************/

deps/jemalloc/src/chunk_mmap.c

@@ -0,0 +1,210 @@
+#define	JEMALLOC_CHUNK_MMAP_C_
+#include "jemalloc/internal/jemalloc_internal.h"
+
+/******************************************************************************/
+/* Function prototypes for non-inline static functions. */
+
+static void	*pages_map(void *addr, size_t size);
+static void	pages_unmap(void *addr, size_t size);
+static void	*chunk_alloc_mmap_slow(size_t size, size_t alignment,
+    bool *zero);
+
+/******************************************************************************/
+
+static void *
+pages_map(void *addr, size_t size)
+{
+	void *ret;
+
+	assert(size != 0);
+
+#ifdef _WIN32
+	/*
+	 * If VirtualAlloc can't allocate at the given address when one is
+	 * given, it fails and returns NULL.
+	 */
+	ret = VirtualAlloc(addr, size, MEM_COMMIT | MEM_RESERVE,
+	    PAGE_READWRITE);
+#else
+	/*
+	 * We don't use MAP_FIXED here, because it can cause the *replacement*
+	 * of existing mappings, and we only want to create new mappings.
+	 */
+	ret = mmap(addr, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON,
+	    -1, 0);
+	assert(ret != NULL);
+
+	if (ret == MAP_FAILED)
+		ret = NULL;
+	else if (addr != NULL && ret != addr) {
+		/*
+		 * We succeeded in mapping memory, but not in the right place.
+		 */
+		if (munmap(ret, size) == -1) {
+			char buf[BUFERROR_BUF];
+
+			buferror(get_errno(), buf, sizeof(buf));
+			malloc_printf("<jemalloc: Error in munmap(): %s\n",
+			    buf);
+			if (opt_abort)
+				abort();
+		}
+		ret = NULL;
+	}
+#endif
+	assert(ret == NULL || (addr == NULL && ret != addr)
+	    || (addr != NULL && ret == addr));
+	return (ret);
+}
+
+static void
+pages_unmap(void *addr, size_t size)
+{
+
+#ifdef _WIN32
+	if (VirtualFree(addr, 0, MEM_RELEASE) == 0)
+#else
+	if (munmap(addr, size) == -1)
+#endif
+	{
+		char buf[BUFERROR_BUF];
+
+		buferror(get_errno(), buf, sizeof(buf));
+		malloc_printf("<jemalloc>: Error in "
+#ifdef _WIN32
+		              "VirtualFree"
+#else
+		              "munmap"
+#endif
+		              "(): %s\n", buf);
+		if (opt_abort)
+			abort();
+	}
+}
+
+static void *
+pages_trim(void *addr, size_t alloc_size, size_t leadsize, size_t size)
+{
+	void *ret = (void *)((uintptr_t)addr + leadsize);
+
+	assert(alloc_size >= leadsize + size);
+#ifdef _WIN32
+	{
+		void *new_addr;
+
+		pages_unmap(addr, alloc_size);
+		new_addr = pages_map(ret, size);
+		if (new_addr == ret)
+			return (ret);
+		if (new_addr)
+			pages_unmap(new_addr, size);
+		return (NULL);
+	}
+#else
+	{
+		size_t trailsize = alloc_size - leadsize - size;
+
+		if (leadsize != 0)
+			pages_unmap(addr, leadsize);
+		if (trailsize != 0)
+			pages_unmap((void *)((uintptr_t)ret + size), trailsize);
+		return (ret);
+	}
+#endif
+}
+
+bool
+pages_purge(void *addr, size_t length)
+{
+	bool unzeroed;
+
+#ifdef _WIN32
+	VirtualAlloc(addr, length, MEM_RESET, PAGE_READWRITE);
+	unzeroed = true;
+#else
+#  ifdef JEMALLOC_PURGE_MADVISE_DONTNEED
+#    define JEMALLOC_MADV_PURGE MADV_DONTNEED
+#    define JEMALLOC_MADV_ZEROS true
+#  elif defined(JEMALLOC_PURGE_MADVISE_FREE)
+#    define JEMALLOC_MADV_PURGE MADV_FREE
+#    define JEMALLOC_MADV_ZEROS false
+#  else
+#    error "No method defined for purging unused dirty pages."
+#  endif
+	int err = madvise(addr, length, JEMALLOC_MADV_PURGE);
+	unzeroed = (JEMALLOC_MADV_ZEROS == false || err != 0);
+#  undef JEMALLOC_MADV_PURGE
+#  undef JEMALLOC_MADV_ZEROS
+#endif
+	return (unzeroed);
+}
+
+static void *
+chunk_alloc_mmap_slow(size_t size, size_t alignment, bool *zero)
+{
+	void *ret, *pages;
+	size_t alloc_size, leadsize;
+
+	alloc_size = size + alignment - PAGE;
+	/* Beware size_t wrap-around. */
+	if (alloc_size < size)
+		return (NULL);
+	do {
+		pages = pages_map(NULL, alloc_size);
+		if (pages == NULL)
+			return (NULL);
+		leadsize = ALIGNMENT_CEILING((uintptr_t)pages, alignment) -
+		    (uintptr_t)pages;
+		ret = pages_trim(pages, alloc_size, leadsize, size);
+	} while (ret == NULL);
+
+	assert(ret != NULL);
+	*zero = true;
+	return (ret);
+}
+
+void *
+chunk_alloc_mmap(size_t size, size_t alignment, bool *zero)
+{
+	void *ret;
+	size_t offset;
+
+	/*
+	 * Ideally, there would be a way to specify alignment to mmap() (like
+	 * NetBSD has), but in the absence of such a feature, we have to work
+	 * hard to efficiently create aligned mappings.  The reliable, but
+	 * slow method is to create a mapping that is over-sized, then trim the
+	 * excess.  However, that always results in one or two calls to
+	 * pages_unmap().
+	 *
+	 * Optimistically try mapping precisely the right amount before falling
+	 * back to the slow method, with the expectation that the optimistic
+	 * approach works most of the time.
+	 */
+
+	assert(alignment != 0);
+	assert((alignment & chunksize_mask) == 0);
+
+	ret = pages_map(NULL, size);
+	if (ret == NULL)
+		return (NULL);
+	offset = ALIGNMENT_ADDR2OFFSET(ret, alignment);
+	if (offset != 0) {
+		pages_unmap(ret, size);
+		return (chunk_alloc_mmap_slow(size, alignment, zero));
+	}
+
+	assert(ret != NULL);
+	*zero = true;
+	return (ret);
+}
+
+bool
+chunk_dealloc_mmap(void *chunk, size_t size)
+{
+
+	if (config_munmap)
+		pages_unmap(chunk, size);
+
+	return (config_munmap == false);
+}

deps/jemalloc/src/ckh.c

@@ -0,0 +1,570 @@
+/*
+ *******************************************************************************
+ * Implementation of (2^1+,2) cuckoo hashing, where 2^1+ indicates that each
+ * hash bucket contains 2^n cells, for n >= 1, and 2 indicates that two hash
+ * functions are employed.  The original cuckoo hashing algorithm was described
+ * in:
+ *
+ *   Pagh, R., F.F. Rodler (2004) Cuckoo Hashing.  Journal of Algorithms
+ *     51(2):122-144.
+ *
+ * Generalization of cuckoo hashing was discussed in:
+ *
+ *   Erlingsson, U., M. Manasse, F. McSherry (2006) A cool and practical
+ *     alternative to traditional hash tables.  In Proceedings of the 7th
+ *     Workshop on Distributed Data and Structures (WDAS'06), Santa Clara, CA,
+ *     January 2006.
+ *
+ * This implementation uses precisely two hash functions because that is the
+ * fewest that can work, and supporting multiple hashes is an implementation
+ * burden.  Here is a reproduction of Figure 1 from Erlingsson et al. (2006)
+ * that shows approximate expected maximum load factors for various
+ * configurations:
+ *
+ *           |         #cells/bucket         |
+ *   #hashes |   1   |   2   |   4   |   8   |
+ *   --------+-------+-------+-------+-------+
+ *         1 | 0.006 | 0.006 | 0.03  | 0.12  |
+ *         2 | 0.49  | 0.86  |>0.93< |>0.96< |
+ *         3 | 0.91  | 0.97  | 0.98  | 0.999 |
+ *         4 | 0.97  | 0.99  | 0.999 |       |
+ *
+ * The number of cells per bucket is chosen such that a bucket fits in one cache
+ * line.  So, on 32- and 64-bit systems, we use (8,2) and (4,2) cuckoo hashing,
+ * respectively.
+ *
+ ******************************************************************************/
+#define JEMALLOC_CKH_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+
+#include "jemalloc/internal/ckh.h"
+
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/hash.h"
+#include "jemalloc/internal/malloc_io.h"
+#include "jemalloc/internal/prng.h"
+#include "jemalloc/internal/util.h"
+
+/******************************************************************************/
+/* Function prototypes for non-inline static functions. */
+
+static bool	ckh_grow(tsd_t *tsd, ckh_t *ckh);
+static void	ckh_shrink(tsd_t *tsd, ckh_t *ckh);
+
+/******************************************************************************/
+
+/*
+ * Search bucket for key and return the cell number if found; SIZE_T_MAX
+ * otherwise.
+ */
+static size_t
+ckh_bucket_search(ckh_t *ckh, size_t bucket, const void *key) {
+	ckhc_t *cell;
+	unsigned i;
+
+	for (i = 0; i < (ZU(1) << LG_CKH_BUCKET_CELLS); i++) {
+		cell = &ckh->tab[(bucket << LG_CKH_BUCKET_CELLS) + i];
+		if (cell->key != NULL && ckh->keycomp(key, cell->key)) {
+			return (bucket << LG_CKH_BUCKET_CELLS) + i;
+		}
+	}
+
+	return SIZE_T_MAX;
+}
+
+/*
+ * Search table for key and return cell number if found; SIZE_T_MAX otherwise.
+ */
+static size_t
+ckh_isearch(ckh_t *ckh, const void *key) {
+	size_t hashes[2], bucket, cell;
+
+	assert(ckh != NULL);
+
+	ckh->hash(key, hashes);
+
+	/* Search primary bucket. */
+	bucket = hashes[0] & ((ZU(1) << ckh->lg_curbuckets) - 1);
+	cell = ckh_bucket_search(ckh, bucket, key);
+	if (cell != SIZE_T_MAX) {
+		return cell;
+	}
+
+	/* Search secondary bucket. */
+	bucket = hashes[1] & ((ZU(1) << ckh->lg_curbuckets) - 1);
+	cell = ckh_bucket_search(ckh, bucket, key);
+	return cell;
+}
+
+static bool
+ckh_try_bucket_insert(ckh_t *ckh, size_t bucket, const void *key,
+    const void *data) {
+	ckhc_t *cell;
+	unsigned offset, i;
+
+	/*
+	 * Cycle through the cells in the bucket, starting at a random position.
+	 * The randomness avoids worst-case search overhead as buckets fill up.
+	 */
+	offset = (unsigned)prng_lg_range_u64(&ckh->prng_state,
+	    LG_CKH_BUCKET_CELLS);
+	for (i = 0; i < (ZU(1) << LG_CKH_BUCKET_CELLS); i++) {
+		cell = &ckh->tab[(bucket << LG_CKH_BUCKET_CELLS) +
+		    ((i + offset) & ((ZU(1) << LG_CKH_BUCKET_CELLS) - 1))];
+		if (cell->key == NULL) {
+			cell->key = key;
+			cell->data = data;
+			ckh->count++;
+			return false;
+		}
+	}
+
+	return true;
+}
+
+/*
+ * No space is available in bucket.  Randomly evict an item, then try to find an
+ * alternate location for that item.  Iteratively repeat this
+ * eviction/relocation procedure until either success or detection of an
+ * eviction/relocation bucket cycle.
+ */
+static bool
+ckh_evict_reloc_insert(ckh_t *ckh, size_t argbucket, void const **argkey,
+    void const **argdata) {
+	const void *key, *data, *tkey, *tdata;
+	ckhc_t *cell;
+	size_t hashes[2], bucket, tbucket;
+	unsigned i;
+
+	bucket = argbucket;
+	key = *argkey;
+	data = *argdata;
+	while (true) {
+		/*
+		 * Choose a random item within the bucket to evict.  This is
+		 * critical to correct function, because without (eventually)
+		 * evicting all items within a bucket during iteration, it
+		 * would be possible to get stuck in an infinite loop if there
+		 * were an item for which both hashes indicated the same
+		 * bucket.
+		 */
+		i = (unsigned)prng_lg_range_u64(&ckh->prng_state,
+		    LG_CKH_BUCKET_CELLS);
+		cell = &ckh->tab[(bucket << LG_CKH_BUCKET_CELLS) + i];
+		assert(cell->key != NULL);
+
+		/* Swap cell->{key,data} and {key,data} (evict). */
+		tkey = cell->key; tdata = cell->data;
+		cell->key = key; cell->data = data;
+		key = tkey; data = tdata;
+
+#ifdef CKH_COUNT
+		ckh->nrelocs++;
+#endif
+
+		/* Find the alternate bucket for the evicted item. */
+		ckh->hash(key, hashes);
+		tbucket = hashes[1] & ((ZU(1) << ckh->lg_curbuckets) - 1);
+		if (tbucket == bucket) {
+			tbucket = hashes[0] & ((ZU(1) << ckh->lg_curbuckets)
+			    - 1);
+			/*
+			 * It may be that (tbucket == bucket) still, if the
+			 * item's hashes both indicate this bucket.  However,
+			 * we are guaranteed to eventually escape this bucket
+			 * during iteration, assuming pseudo-random item
+			 * selection (true randomness would make infinite
+			 * looping a remote possibility).  The reason we can
+			 * never get trapped forever is that there are two
+			 * cases:
+			 *
+			 * 1) This bucket == argbucket, so we will quickly
+			 *    detect an eviction cycle and terminate.
+			 * 2) An item was evicted to this bucket from another,
+			 *    which means that at least one item in this bucket
+			 *    has hashes that indicate distinct buckets.
+			 */
+		}
+		/* Check for a cycle. */
+		if (tbucket == argbucket) {
+			*argkey = key;
+			*argdata = data;
+			return true;
+		}
+
+		bucket = tbucket;
+		if (!ckh_try_bucket_insert(ckh, bucket, key, data)) {
+			return false;
+		}
+	}
+}
+
+static bool
+ckh_try_insert(ckh_t *ckh, void const**argkey, void const**argdata) {
+	size_t hashes[2], bucket;
+	const void *key = *argkey;
+	const void *data = *argdata;
+
+	ckh->hash(key, hashes);
+
+	/* Try to insert in primary bucket. */
+	bucket = hashes[0] & ((ZU(1) << ckh->lg_curbuckets) - 1);
+	if (!ckh_try_bucket_insert(ckh, bucket, key, data)) {
+		return false;
+	}
+
+	/* Try to insert in secondary bucket. */
+	bucket = hashes[1] & ((ZU(1) << ckh->lg_curbuckets) - 1);
+	if (!ckh_try_bucket_insert(ckh, bucket, key, data)) {
+		return false;
+	}
+
+	/*
+	 * Try to find a place for this item via iterative eviction/relocation.
+	 */
+	return ckh_evict_reloc_insert(ckh, bucket, argkey, argdata);
+}
+
+/*
+ * Try to rebuild the hash table from scratch by inserting all items from the
+ * old table into the new.
+ */
+static bool
+ckh_rebuild(ckh_t *ckh, ckhc_t *aTab) {
+	size_t count, i, nins;
+	const void *key, *data;
+
+	count = ckh->count;
+	ckh->count = 0;
+	for (i = nins = 0; nins < count; i++) {
+		if (aTab[i].key != NULL) {
+			key = aTab[i].key;
+			data = aTab[i].data;
+			if (ckh_try_insert(ckh, &key, &data)) {
+				ckh->count = count;
+				return true;
+			}
+			nins++;
+		}
+	}
+
+	return false;
+}
+
+static bool
+ckh_grow(tsd_t *tsd, ckh_t *ckh) {
+	bool ret;
+	ckhc_t *tab, *ttab;
+	unsigned lg_prevbuckets, lg_curcells;
+
+#ifdef CKH_COUNT
+	ckh->ngrows++;
+#endif
+
+	/*
+	 * It is possible (though unlikely, given well behaved hashes) that the
+	 * table will have to be doubled more than once in order to create a
+	 * usable table.
+	 */
+	lg_prevbuckets = ckh->lg_curbuckets;
+	lg_curcells = ckh->lg_curbuckets + LG_CKH_BUCKET_CELLS;
+	while (true) {
+		size_t usize;
+
+		lg_curcells++;
+		usize = sz_sa2u(sizeof(ckhc_t) << lg_curcells, CACHELINE);
+		if (unlikely(usize == 0
+		    || usize > SC_LARGE_MAXCLASS)) {
+			ret = true;
+			goto label_return;
+		}
+		tab = (ckhc_t *)ipallocztm(tsd_tsdn(tsd), usize, CACHELINE,
+		    true, NULL, true, arena_ichoose(tsd, NULL));
+		if (tab == NULL) {
+			ret = true;
+			goto label_return;
+		}
+		/* Swap in new table. */
+		ttab = ckh->tab;
+		ckh->tab = tab;
+		tab = ttab;
+		ckh->lg_curbuckets = lg_curcells - LG_CKH_BUCKET_CELLS;
+
+		if (!ckh_rebuild(ckh, tab)) {
+			idalloctm(tsd_tsdn(tsd), tab, NULL, NULL, true, true);
+			break;
+		}
+
+		/* Rebuilding failed, so back out partially rebuilt table. */
+		idalloctm(tsd_tsdn(tsd), ckh->tab, NULL, NULL, true, true);
+		ckh->tab = tab;
+		ckh->lg_curbuckets = lg_prevbuckets;
+	}
+
+	ret = false;
+label_return:
+	return ret;
+}
+
+static void
+ckh_shrink(tsd_t *tsd, ckh_t *ckh) {
+	ckhc_t *tab, *ttab;
+	size_t usize;
+	unsigned lg_prevbuckets, lg_curcells;
+
+	/*
+	 * It is possible (though unlikely, given well behaved hashes) that the
+	 * table rebuild will fail.
+	 */
+	lg_prevbuckets = ckh->lg_curbuckets;
+	lg_curcells = ckh->lg_curbuckets + LG_CKH_BUCKET_CELLS - 1;
+	usize = sz_sa2u(sizeof(ckhc_t) << lg_curcells, CACHELINE);
+	if (unlikely(usize == 0 || usize > SC_LARGE_MAXCLASS)) {
+		return;
+	}
+	tab = (ckhc_t *)ipallocztm(tsd_tsdn(tsd), usize, CACHELINE, true, NULL,
+	    true, arena_ichoose(tsd, NULL));
+	if (tab == NULL) {
+		/*
+		 * An OOM error isn't worth propagating, since it doesn't
+		 * prevent this or future operations from proceeding.
+		 */
+		return;
+	}
+	/* Swap in new table. */
+	ttab = ckh->tab;
+	ckh->tab = tab;
+	tab = ttab;
+	ckh->lg_curbuckets = lg_curcells - LG_CKH_BUCKET_CELLS;
+
+	if (!ckh_rebuild(ckh, tab)) {
+		idalloctm(tsd_tsdn(tsd), tab, NULL, NULL, true, true);
+#ifdef CKH_COUNT
+		ckh->nshrinks++;
+#endif
+		return;
+	}
+
+	/* Rebuilding failed, so back out partially rebuilt table. */
+	idalloctm(tsd_tsdn(tsd), ckh->tab, NULL, NULL, true, true);
+	ckh->tab = tab;
+	ckh->lg_curbuckets = lg_prevbuckets;
+#ifdef CKH_COUNT
+	ckh->nshrinkfails++;
+#endif
+}
+
+bool
+ckh_new(tsd_t *tsd, ckh_t *ckh, size_t minitems, ckh_hash_t *hash,
+    ckh_keycomp_t *keycomp) {
+	bool ret;
+	size_t mincells, usize;
+	unsigned lg_mincells;
+
+	assert(minitems > 0);
+	assert(hash != NULL);
+	assert(keycomp != NULL);
+
+#ifdef CKH_COUNT
+	ckh->ngrows = 0;
+	ckh->nshrinks = 0;
+	ckh->nshrinkfails = 0;
+	ckh->ninserts = 0;
+	ckh->nrelocs = 0;
+#endif
+	ckh->prng_state = 42; /* Value doesn't really matter. */
+	ckh->count = 0;
+
+	/*
+	 * Find the minimum power of 2 that is large enough to fit minitems
+	 * entries.  We are using (2+,2) cuckoo hashing, which has an expected
+	 * maximum load factor of at least ~0.86, so 0.75 is a conservative load
+	 * factor that will typically allow mincells items to fit without ever
+	 * growing the table.
+	 */
+	assert(LG_CKH_BUCKET_CELLS > 0);
+	mincells = ((minitems + (3 - (minitems % 3))) / 3) << 2;
+	for (lg_mincells = LG_CKH_BUCKET_CELLS;
+	    (ZU(1) << lg_mincells) < mincells;
+	    lg_mincells++) {
+		/* Do nothing. */
+	}
+	ckh->lg_minbuckets = lg_mincells - LG_CKH_BUCKET_CELLS;
+	ckh->lg_curbuckets = lg_mincells - LG_CKH_BUCKET_CELLS;
+	ckh->hash = hash;
+	ckh->keycomp = keycomp;
+
+	usize = sz_sa2u(sizeof(ckhc_t) << lg_mincells, CACHELINE);
+	if (unlikely(usize == 0 || usize > SC_LARGE_MAXCLASS)) {
+		ret = true;
+		goto label_return;
+	}
+	ckh->tab = (ckhc_t *)ipallocztm(tsd_tsdn(tsd), usize, CACHELINE, true,
+	    NULL, true, arena_ichoose(tsd, NULL));
+	if (ckh->tab == NULL) {
+		ret = true;
+		goto label_return;
+	}
+
+	ret = false;
+label_return:
+	return ret;
+}
+
+void
+ckh_delete(tsd_t *tsd, ckh_t *ckh) {
+	assert(ckh != NULL);
+
+#ifdef CKH_VERBOSE
+	malloc_printf(
+	    "%s(%p): ngrows: %"FMTu64", nshrinks: %"FMTu64","
+	    " nshrinkfails: %"FMTu64", ninserts: %"FMTu64","
+	    " nrelocs: %"FMTu64"\n", __func__, ckh,
+	    (unsigned long long)ckh->ngrows,
+	    (unsigned long long)ckh->nshrinks,
+	    (unsigned long long)ckh->nshrinkfails,
+	    (unsigned long long)ckh->ninserts,
+	    (unsigned long long)ckh->nrelocs);
+#endif
+
+	idalloctm(tsd_tsdn(tsd), ckh->tab, NULL, NULL, true, true);
+	if (config_debug) {
+		memset(ckh, JEMALLOC_FREE_JUNK, sizeof(ckh_t));
+	}
+}
+
+size_t
+ckh_count(ckh_t *ckh) {
+	assert(ckh != NULL);
+
+	return ckh->count;
+}
+
+bool
+ckh_iter(ckh_t *ckh, size_t *tabind, void **key, void **data) {
+	size_t i, ncells;
+
+	for (i = *tabind, ncells = (ZU(1) << (ckh->lg_curbuckets +
+	    LG_CKH_BUCKET_CELLS)); i < ncells; i++) {
+		if (ckh->tab[i].key != NULL) {
+			if (key != NULL) {
+				*key = (void *)ckh->tab[i].key;
+			}
+			if (data != NULL) {
+				*data = (void *)ckh->tab[i].data;
+			}
+			*tabind = i + 1;
+			return false;
+		}
+	}
+
+	return true;
+}
+
+bool
+ckh_insert(tsd_t *tsd, ckh_t *ckh, const void *key, const void *data) {
+	bool ret;
+
+	assert(ckh != NULL);
+	assert(ckh_search(ckh, key, NULL, NULL));
+
+#ifdef CKH_COUNT
+	ckh->ninserts++;
+#endif
+
+	while (ckh_try_insert(ckh, &key, &data)) {
+		if (ckh_grow(tsd, ckh)) {
+			ret = true;
+			goto label_return;
+		}
+	}
+
+	ret = false;
+label_return:
+	return ret;
+}
+
+bool
+ckh_remove(tsd_t *tsd, ckh_t *ckh, const void *searchkey, void **key,
+    void **data) {
+	size_t cell;
+
+	assert(ckh != NULL);
+
+	cell = ckh_isearch(ckh, searchkey);
+	if (cell != SIZE_T_MAX) {
+		if (key != NULL) {
+			*key = (void *)ckh->tab[cell].key;
+		}
+		if (data != NULL) {
+			*data = (void *)ckh->tab[cell].data;
+		}
+		ckh->tab[cell].key = NULL;
+		ckh->tab[cell].data = NULL; /* Not necessary. */
+
+		ckh->count--;
+		/* Try to halve the table if it is less than 1/4 full. */
+		if (ckh->count < (ZU(1) << (ckh->lg_curbuckets
+		    + LG_CKH_BUCKET_CELLS - 2)) && ckh->lg_curbuckets
+		    > ckh->lg_minbuckets) {
+			/* Ignore error due to OOM. */
+			ckh_shrink(tsd, ckh);
+		}
+
+		return false;
+	}
+
+	return true;
+}
+
+bool
+ckh_search(ckh_t *ckh, const void *searchkey, void **key, void **data) {
+	size_t cell;
+
+	assert(ckh != NULL);
+
+	cell = ckh_isearch(ckh, searchkey);
+	if (cell != SIZE_T_MAX) {
+		if (key != NULL) {
+			*key = (void *)ckh->tab[cell].key;
+		}
+		if (data != NULL) {
+			*data = (void *)ckh->tab[cell].data;
+		}
+		return false;
+	}
+
+	return true;
+}
+
+void
+ckh_string_hash(const void *key, size_t r_hash[2]) {
+	hash(key, strlen((const char *)key), 0x94122f33U, r_hash);
+}
+
+bool
+ckh_string_keycomp(const void *k1, const void *k2) {
+	assert(k1 != NULL);
+	assert(k2 != NULL);
+
+	return !strcmp((char *)k1, (char *)k2);
+}
+
+void
+ckh_pointer_hash(const void *key, size_t r_hash[2]) {
+	union {
+		const void	*v;
+		size_t		i;
+	} u;
+
+	assert(sizeof(u.v) == sizeof(u.i));
+	u.v = key;
+	hash(&u.i, sizeof(u.i), 0xd983396eU, r_hash);
+}
+
+bool
+ckh_pointer_keycomp(const void *k1, const void *k2) {
+	return (k1 == k2);
+}

deps/jemalloc/src/div.c

@@ -0,0 +1,55 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+
+#include "jemalloc/internal/div.h"
+
+#include "jemalloc/internal/assert.h"
+
+/*
+ * Suppose we have n = q * d, all integers. We know n and d, and want q = n / d.
+ *
+ * For any k, we have (here, all division is exact; not C-style rounding):
+ * floor(ceil(2^k / d) * n / 2^k) = floor((2^k + r) / d * n / 2^k), where
+ * r = (-2^k) mod d.
+ *
+ * Expanding this out:
+ * ... = floor(2^k / d * n / 2^k + r / d * n / 2^k)
+ *     = floor(n / d + (r / d) * (n / 2^k)).
+ *
+ * The fractional part of n / d is 0 (because of the assumption that d divides n
+ * exactly), so we have:
+ * ... = n / d + floor((r / d) * (n / 2^k))
+ *
+ * So that our initial expression is equal to the quantity we seek, so long as
+ * (r / d) * (n / 2^k) < 1.
+ *
+ * r is a remainder mod d, so r < d and r / d < 1 always. We can make
+ * n / 2 ^ k < 1 by setting k = 32. This gets us a value of magic that works.
+ */
+
+void
+div_init(div_info_t *div_info, size_t d) {
+	/* Nonsensical. */
+	assert(d != 0);
+	/*
+	 * This would make the value of magic too high to fit into a uint32_t
+	 * (we would want magic = 2^32 exactly). This would mess with code gen
+	 * on 32-bit machines.
+	 */
+	assert(d != 1);
+
+	uint64_t two_to_k = ((uint64_t)1 << 32);
+	uint32_t magic = (uint32_t)(two_to_k / d);
+
+	/*
+	 * We want magic = ceil(2^k / d), but C gives us floor. We have to
+	 * increment it unless the result was exact (i.e. unless d is a power of
+	 * two).
+	 */
+	if (two_to_k % d != 0) {
+		magic++;
+	}
+	div_info->magic = magic;
+#ifdef JEMALLOC_DEBUG
+	div_info->d = d;
+#endif
+}

deps/jemalloc/src/extent_dss.c

@@ -0,0 +1,271 @@
+#define JEMALLOC_EXTENT_DSS_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/extent_dss.h"
+#include "jemalloc/internal/spin.h"
+
+/******************************************************************************/
+/* Data. */
+
+const char	*opt_dss = DSS_DEFAULT;
+
+const char	*dss_prec_names[] = {
+	"disabled",
+	"primary",
+	"secondary",
+	"N/A"
+};
+
+/*
+ * Current dss precedence default, used when creating new arenas.  NB: This is
+ * stored as unsigned rather than dss_prec_t because in principle there's no
+ * guarantee that sizeof(dss_prec_t) is the same as sizeof(unsigned), and we use
+ * atomic operations to synchronize the setting.
+ */
+static atomic_u_t	dss_prec_default = ATOMIC_INIT(
+    (unsigned)DSS_PREC_DEFAULT);
+
+/* Base address of the DSS. */
+static void		*dss_base;
+/* Atomic boolean indicating whether a thread is currently extending DSS. */
+static atomic_b_t	dss_extending;
+/* Atomic boolean indicating whether the DSS is exhausted. */
+static atomic_b_t	dss_exhausted;
+/* Atomic current upper limit on DSS addresses. */
+static atomic_p_t	dss_max;
+
+/******************************************************************************/
+
+static void *
+extent_dss_sbrk(intptr_t increment) {
+#ifdef JEMALLOC_DSS
+	return sbrk(increment);
+#else
+	not_implemented();
+	return NULL;
+#endif
+}
+
+dss_prec_t
+extent_dss_prec_get(void) {
+	dss_prec_t ret;
+
+	if (!have_dss) {
+		return dss_prec_disabled;
+	}
+	ret = (dss_prec_t)atomic_load_u(&dss_prec_default, ATOMIC_ACQUIRE);
+	return ret;
+}
+
+bool
+extent_dss_prec_set(dss_prec_t dss_prec) {
+	if (!have_dss) {
+		return (dss_prec != dss_prec_disabled);
+	}
+	atomic_store_u(&dss_prec_default, (unsigned)dss_prec, ATOMIC_RELEASE);
+	return false;
+}
+
+static void
+extent_dss_extending_start(void) {
+	spin_t spinner = SPIN_INITIALIZER;
+	while (true) {
+		bool expected = false;
+		if (atomic_compare_exchange_weak_b(&dss_extending, &expected,
+		    true, ATOMIC_ACQ_REL, ATOMIC_RELAXED)) {
+			break;
+		}
+		spin_adaptive(&spinner);
+	}
+}
+
+static void
+extent_dss_extending_finish(void) {
+	assert(atomic_load_b(&dss_extending, ATOMIC_RELAXED));
+
+	atomic_store_b(&dss_extending, false, ATOMIC_RELEASE);
+}
+
+static void *
+extent_dss_max_update(void *new_addr) {
+	/*
+	 * Get the current end of the DSS as max_cur and assure that dss_max is
+	 * up to date.
+	 */
+	void *max_cur = extent_dss_sbrk(0);
+	if (max_cur == (void *)-1) {
+		return NULL;
+	}
+	atomic_store_p(&dss_max, max_cur, ATOMIC_RELEASE);
+	/* Fixed new_addr can only be supported if it is at the edge of DSS. */
+	if (new_addr != NULL && max_cur != new_addr) {
+		return NULL;
+	}
+	return max_cur;
+}
+
+void *
+extent_alloc_dss(tsdn_t *tsdn, arena_t *arena, void *new_addr, size_t size,
+    size_t alignment, bool *zero, bool *commit) {
+	extent_t *gap;
+
+	cassert(have_dss);
+	assert(size > 0);
+	assert(alignment == ALIGNMENT_CEILING(alignment, PAGE));
+
+	/*
+	 * sbrk() uses a signed increment argument, so take care not to
+	 * interpret a large allocation request as a negative increment.
+	 */
+	if ((intptr_t)size < 0) {
+		return NULL;
+	}
+
+	gap = extent_alloc(tsdn, arena);
+	if (gap == NULL) {
+		return NULL;
+	}
+
+	extent_dss_extending_start();
+	if (!atomic_load_b(&dss_exhausted, ATOMIC_ACQUIRE)) {
+		/*
+		 * The loop is necessary to recover from races with other
+		 * threads that are using the DSS for something other than
+		 * malloc.
+		 */
+		while (true) {
+			void *max_cur = extent_dss_max_update(new_addr);
+			if (max_cur == NULL) {
+				goto label_oom;
+			}
+
+			/*
+			 * Compute how much page-aligned gap space (if any) is
+			 * necessary to satisfy alignment.  This space can be
+			 * recycled for later use.
+			 */
+			void *gap_addr_page = (void *)(PAGE_CEILING(
+			    (uintptr_t)max_cur));
+			void *ret = (void *)ALIGNMENT_CEILING(
+			    (uintptr_t)gap_addr_page, alignment);
+			size_t gap_size_page = (uintptr_t)ret -
+			    (uintptr_t)gap_addr_page;
+			if (gap_size_page != 0) {
+				extent_init(gap, arena, gap_addr_page,
+				    gap_size_page, false, SC_NSIZES,
+				    arena_extent_sn_next(arena),
+				    extent_state_active, false, true, true,
+				    EXTENT_NOT_HEAD);
+			}
+			/*
+			 * Compute the address just past the end of the desired
+			 * allocation space.
+			 */
+			void *dss_next = (void *)((uintptr_t)ret + size);
+			if ((uintptr_t)ret < (uintptr_t)max_cur ||
+			    (uintptr_t)dss_next < (uintptr_t)max_cur) {
+				goto label_oom; /* Wrap-around. */
+			}
+			/* Compute the increment, including subpage bytes. */
+			void *gap_addr_subpage = max_cur;
+			size_t gap_size_subpage = (uintptr_t)ret -
+			    (uintptr_t)gap_addr_subpage;
+			intptr_t incr = gap_size_subpage + size;
+
+			assert((uintptr_t)max_cur + incr == (uintptr_t)ret +
+			    size);
+
+			/* Try to allocate. */
+			void *dss_prev = extent_dss_sbrk(incr);
+			if (dss_prev == max_cur) {
+				/* Success. */
+				atomic_store_p(&dss_max, dss_next,
+				    ATOMIC_RELEASE);
+				extent_dss_extending_finish();
+
+				if (gap_size_page != 0) {
+					extent_dalloc_gap(tsdn, arena, gap);
+				} else {
+					extent_dalloc(tsdn, arena, gap);
+				}
+				if (!*commit) {
+					*commit = pages_decommit(ret, size);
+				}
+				if (*zero && *commit) {
+					extent_hooks_t *extent_hooks =
+					    EXTENT_HOOKS_INITIALIZER;
+					extent_t extent;
+
+					extent_init(&extent, arena, ret, size,
+					    size, false, SC_NSIZES,
+					    extent_state_active, false, true,
+					    true, EXTENT_NOT_HEAD);
+					if (extent_purge_forced_wrapper(tsdn,
+					    arena, &extent_hooks, &extent, 0,
+					    size)) {
+						memset(ret, 0, size);
+					}
+				}
+				return ret;
+			}
+			/*
+			 * Failure, whether due to OOM or a race with a raw
+			 * sbrk() call from outside the allocator.
+			 */
+			if (dss_prev == (void *)-1) {
+				/* OOM. */
+				atomic_store_b(&dss_exhausted, true,
+				    ATOMIC_RELEASE);
+				goto label_oom;
+			}
+		}
+	}
+label_oom:
+	extent_dss_extending_finish();
+	extent_dalloc(tsdn, arena, gap);
+	return NULL;
+}
+
+static bool
+extent_in_dss_helper(void *addr, void *max) {
+	return ((uintptr_t)addr >= (uintptr_t)dss_base && (uintptr_t)addr <
+	    (uintptr_t)max);
+}
+
+bool
+extent_in_dss(void *addr) {
+	cassert(have_dss);
+
+	return extent_in_dss_helper(addr, atomic_load_p(&dss_max,
+	    ATOMIC_ACQUIRE));
+}
+
+bool
+extent_dss_mergeable(void *addr_a, void *addr_b) {
+	void *max;
+
+	cassert(have_dss);
+
+	if ((uintptr_t)addr_a < (uintptr_t)dss_base && (uintptr_t)addr_b <
+	    (uintptr_t)dss_base) {
+		return true;
+	}
+
+	max = atomic_load_p(&dss_max, ATOMIC_ACQUIRE);
+	return (extent_in_dss_helper(addr_a, max) ==
+	    extent_in_dss_helper(addr_b, max));
+}
+
+void
+extent_dss_boot(void) {
+	cassert(have_dss);
+
+	dss_base = extent_dss_sbrk(0);
+	atomic_store_b(&dss_extending, false, ATOMIC_RELAXED);
+	atomic_store_b(&dss_exhausted, dss_base == (void *)-1, ATOMIC_RELAXED);
+	atomic_store_p(&dss_max, dss_base, ATOMIC_RELAXED);
+}
+
+/******************************************************************************/

deps/jemalloc/src/extent_mmap.c

@@ -0,0 +1,42 @@
+#define JEMALLOC_EXTENT_MMAP_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/extent_mmap.h"
+
+/******************************************************************************/
+/* Data. */
+
+bool	opt_retain =
+#ifdef JEMALLOC_RETAIN
+    true
+#else
+    false
+#endif
+    ;
+
+/******************************************************************************/
+
+void *
+extent_alloc_mmap(void *new_addr, size_t size, size_t alignment, bool *zero,
+    bool *commit) {
+	assert(alignment == ALIGNMENT_CEILING(alignment, PAGE));
+	void *ret = pages_map(new_addr, size, alignment, commit);
+	if (ret == NULL) {
+		return NULL;
+	}
+	assert(ret != NULL);
+	if (*commit) {
+		*zero = true;
+	}
+	return ret;
+}
+
+bool
+extent_dalloc_mmap(void *addr, size_t size) {
+	if (!opt_retain) {
+		pages_unmap(addr, size);
+	}
+	return opt_retain;
+}

deps/jemalloc/src/hash.c

@@ -0,0 +1,3 @@
+#define JEMALLOC_HASH_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"

deps/jemalloc/src/hook.c

@@ -0,0 +1,195 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+
+#include "jemalloc/internal/hook.h"
+
+#include "jemalloc/internal/atomic.h"
+#include "jemalloc/internal/mutex.h"
+#include "jemalloc/internal/seq.h"
+
+typedef struct hooks_internal_s hooks_internal_t;
+struct hooks_internal_s {
+	hooks_t hooks;
+	bool in_use;
+};
+
+seq_define(hooks_internal_t, hooks)
+
+static atomic_u_t nhooks = ATOMIC_INIT(0);
+static seq_hooks_t hooks[HOOK_MAX];
+static malloc_mutex_t hooks_mu;
+
+bool
+hook_boot() {
+	return malloc_mutex_init(&hooks_mu, "hooks", WITNESS_RANK_HOOK,
+	    malloc_mutex_rank_exclusive);
+}
+
+static void *
+hook_install_locked(hooks_t *to_install) {
+	hooks_internal_t hooks_internal;
+	for (int i = 0; i < HOOK_MAX; i++) {
+		bool success = seq_try_load_hooks(&hooks_internal, &hooks[i]);
+		/* We hold mu; no concurrent access. */
+		assert(success);
+		if (!hooks_internal.in_use) {
+			hooks_internal.hooks = *to_install;
+			hooks_internal.in_use = true;
+			seq_store_hooks(&hooks[i], &hooks_internal);
+			atomic_store_u(&nhooks,
+			    atomic_load_u(&nhooks, ATOMIC_RELAXED) + 1,
+			    ATOMIC_RELAXED);
+			return &hooks[i];
+		}
+	}
+	return NULL;
+}
+
+void *
+hook_install(tsdn_t *tsdn, hooks_t *to_install) {
+	malloc_mutex_lock(tsdn, &hooks_mu);
+	void *ret = hook_install_locked(to_install);
+	if (ret != NULL) {
+		tsd_global_slow_inc(tsdn);
+	}
+	malloc_mutex_unlock(tsdn, &hooks_mu);
+	return ret;
+}
+
+static void
+hook_remove_locked(seq_hooks_t *to_remove) {
+	hooks_internal_t hooks_internal;
+	bool success = seq_try_load_hooks(&hooks_internal, to_remove);
+	/* We hold mu; no concurrent access. */
+	assert(success);
+	/* Should only remove hooks that were added. */
+	assert(hooks_internal.in_use);
+	hooks_internal.in_use = false;
+	seq_store_hooks(to_remove, &hooks_internal);
+	atomic_store_u(&nhooks, atomic_load_u(&nhooks, ATOMIC_RELAXED) - 1,
+	    ATOMIC_RELAXED);
+}
+
+void
+hook_remove(tsdn_t *tsdn, void *opaque) {
+	if (config_debug) {
+		char *hooks_begin = (char *)&hooks[0];
+		char *hooks_end = (char *)&hooks[HOOK_MAX];
+		char *hook = (char *)opaque;
+		assert(hooks_begin <= hook && hook < hooks_end
+		    && (hook - hooks_begin) % sizeof(seq_hooks_t) == 0);
+	}
+	malloc_mutex_lock(tsdn, &hooks_mu);
+	hook_remove_locked((seq_hooks_t *)opaque);
+	tsd_global_slow_dec(tsdn);
+	malloc_mutex_unlock(tsdn, &hooks_mu);
+}
+
+#define FOR_EACH_HOOK_BEGIN(hooks_internal_ptr)				\
+for (int for_each_hook_counter = 0;					\
+    for_each_hook_counter < HOOK_MAX;					\
+    for_each_hook_counter++) {						\
+	bool for_each_hook_success = seq_try_load_hooks(		\
+	    (hooks_internal_ptr), &hooks[for_each_hook_counter]);	\
+	if (!for_each_hook_success) {					\
+		continue;						\
+	}								\
+	if (!(hooks_internal_ptr)->in_use) {				\
+		continue;						\
+	}
+#define FOR_EACH_HOOK_END						\
+}
+
+static bool *
+hook_reentrantp() {
+	/*
+	 * We prevent user reentrancy within hooks.  This is basically just a
+	 * thread-local bool that triggers an early-exit.
+	 *
+	 * We don't fold in_hook into reentrancy.  There are two reasons for
+	 * this:
+	 * - Right now, we turn on reentrancy during things like extent hook
+	 *   execution.  Allocating during extent hooks is not officially
+	 *   supported, but we don't want to break it for the time being.  These
+	 *   sorts of allocations should probably still be hooked, though.
+	 * - If a hook allocates, we may want it to be relatively fast (after
+	 *   all, it executes on every allocator operation).  Turning on
+	 *   reentrancy is a fairly heavyweight mode (disabling tcache,
+	 *   redirecting to arena 0, etc.).  It's possible we may one day want
+	 *   to turn on reentrant mode here, if it proves too difficult to keep
+	 *   this working.  But that's fairly easy for us to see; OTOH, people
+	 *   not using hooks because they're too slow is easy for us to miss.
+	 *
+	 * The tricky part is
+	 * that this code might get invoked even if we don't have access to tsd.
+	 * This function mimics getting a pointer to thread-local data, except
+	 * that it might secretly return a pointer to some global data if we
+	 * know that the caller will take the early-exit path.
+	 * If we return a bool that indicates that we are reentrant, then the
+	 * caller will go down the early exit path, leaving the global
+	 * untouched.
+	 */
+	static bool in_hook_global = true;
+	tsdn_t *tsdn = tsdn_fetch();
+	tcache_t *tcache = tsdn_tcachep_get(tsdn);
+	if (tcache != NULL) {
+		return &tcache->in_hook;
+	}
+	return &in_hook_global;
+}
+
+#define HOOK_PROLOGUE							\
+	if (likely(atomic_load_u(&nhooks, ATOMIC_RELAXED) == 0)) {	\
+		return;							\
+	}								\
+	bool *in_hook = hook_reentrantp();				\
+	if (*in_hook) {							\
+		return;							\
+	}								\
+	*in_hook = true;
+
+#define HOOK_EPILOGUE							\
+	*in_hook = false;
+
+void
+hook_invoke_alloc(hook_alloc_t type, void *result, uintptr_t result_raw,
+    uintptr_t args_raw[3]) {
+	HOOK_PROLOGUE
+
+	hooks_internal_t hook;
+	FOR_EACH_HOOK_BEGIN(&hook)
+		hook_alloc h = hook.hooks.alloc_hook;
+		if (h != NULL) {
+			h(hook.hooks.extra, type, result, result_raw, args_raw);
+		}
+	FOR_EACH_HOOK_END
+
+	HOOK_EPILOGUE
+}
+
+void
+hook_invoke_dalloc(hook_dalloc_t type, void *address, uintptr_t args_raw[3]) {
+	HOOK_PROLOGUE
+	hooks_internal_t hook;
+	FOR_EACH_HOOK_BEGIN(&hook)
+		hook_dalloc h = hook.hooks.dalloc_hook;
+		if (h != NULL) {
+			h(hook.hooks.extra, type, address, args_raw);
+		}
+	FOR_EACH_HOOK_END
+	HOOK_EPILOGUE
+}
+
+void
+hook_invoke_expand(hook_expand_t type, void *address, size_t old_usize,
+    size_t new_usize, uintptr_t result_raw, uintptr_t args_raw[4]) {
+	HOOK_PROLOGUE
+	hooks_internal_t hook;
+	FOR_EACH_HOOK_BEGIN(&hook)
+		hook_expand h = hook.hooks.expand_hook;
+		if (h != NULL) {
+			h(hook.hooks.extra, type, address, old_usize, new_usize,
+			    result_raw, args_raw);
+		}
+	FOR_EACH_HOOK_END
+	HOOK_EPILOGUE
+}

deps/jemalloc/src/hooks.c

@@ -0,0 +1,12 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+
+/*
+ * The hooks are a little bit screwy -- they're not genuinely exported in the
+ * sense that we want them available to end-users, but we do want them visible
+ * from outside the generated library, so that we can use them in test code.
+ */
+JEMALLOC_EXPORT
+void (*hooks_arena_new_hook)() = NULL;
+
+JEMALLOC_EXPORT
+void (*hooks_libc_hook)() = NULL;

deps/jemalloc/src/huge.c

@@ -0,0 +1,347 @@
+#define	JEMALLOC_HUGE_C_
+#include "jemalloc/internal/jemalloc_internal.h"
+
+/******************************************************************************/
+/* Data. */
+
+uint64_t	huge_nmalloc;
+uint64_t	huge_ndalloc;
+size_t		huge_allocated;
+
+malloc_mutex_t	huge_mtx;
+
+/******************************************************************************/
+
+/* Tree of chunks that are stand-alone huge allocations. */
+static extent_tree_t	huge;
+
+void *
+huge_malloc(size_t size, bool zero, dss_prec_t dss_prec)
+{
+
+	return (huge_palloc(size, chunksize, zero, dss_prec));
+}
+
+void *
+huge_palloc(size_t size, size_t alignment, bool zero, dss_prec_t dss_prec)
+{
+	void *ret;
+	size_t csize;
+	extent_node_t *node;
+	bool is_zeroed;
+
+	/* Allocate one or more contiguous chunks for this request. */
+
+	csize = CHUNK_CEILING(size);
+	if (csize == 0) {
+		/* size is large enough to cause size_t wrap-around. */
+		return (NULL);
+	}
+
+	/* Allocate an extent node with which to track the chunk. */
+	node = base_node_alloc();
+	if (node == NULL)
+		return (NULL);
+
+	/*
+	 * Copy zero into is_zeroed and pass the copy to chunk_alloc(), so that
+	 * it is possible to make correct junk/zero fill decisions below.
+	 */
+	is_zeroed = zero;
+	ret = chunk_alloc(csize, alignment, false, &is_zeroed, dss_prec);
+	if (ret == NULL) {
+		base_node_dealloc(node);
+		return (NULL);
+	}
+
+	/* Insert node into huge. */
+	node->addr = ret;
+	node->size = csize;
+
+	malloc_mutex_lock(&huge_mtx);
+	extent_tree_ad_insert(&huge, node);
+	if (config_stats) {
+		stats_cactive_add(csize);
+		huge_nmalloc++;
+		huge_allocated += csize;
+	}
+	malloc_mutex_unlock(&huge_mtx);
+
+	if (config_fill && zero == false) {
+		if (opt_junk)
+			memset(ret, 0xa5, csize);
+		else if (opt_zero && is_zeroed == false)
+			memset(ret, 0, csize);
+	}
+
+	return (ret);
+}
+
+bool
+huge_ralloc_no_move(void *ptr, size_t oldsize, size_t size, size_t extra)
+{
+
+	/*
+	 * Avoid moving the allocation if the size class can be left the same.
+	 */
+	if (oldsize > arena_maxclass
+	    && CHUNK_CEILING(oldsize) >= CHUNK_CEILING(size)
+	    && CHUNK_CEILING(oldsize) <= CHUNK_CEILING(size+extra)) {
+		assert(CHUNK_CEILING(oldsize) == oldsize);
+		return (false);
+	}
+
+	/* Reallocation would require a move. */
+	return (true);
+}
+
+void *
+huge_ralloc(void *ptr, size_t oldsize, size_t size, size_t extra,
+    size_t alignment, bool zero, bool try_tcache_dalloc, dss_prec_t dss_prec)
+{
+	void *ret;
+	size_t copysize;
+
+	/* Try to avoid moving the allocation. */
+	if (huge_ralloc_no_move(ptr, oldsize, size, extra) == false)
+		return (ptr);
+
+	/*
+	 * size and oldsize are different enough that we need to use a
+	 * different size class.  In that case, fall back to allocating new
+	 * space and copying.
+	 */
+	if (alignment > chunksize)
+		ret = huge_palloc(size + extra, alignment, zero, dss_prec);
+	else
+		ret = huge_malloc(size + extra, zero, dss_prec);
+
+	if (ret == NULL) {
+		if (extra == 0)
+			return (NULL);
+		/* Try again, this time without extra. */
+		if (alignment > chunksize)
+			ret = huge_palloc(size, alignment, zero, dss_prec);
+		else
+			ret = huge_malloc(size, zero, dss_prec);
+
+		if (ret == NULL)
+			return (NULL);
+	}
+
+	/*
+	 * Copy at most size bytes (not size+extra), since the caller has no
+	 * expectation that the extra bytes will be reliably preserved.
+	 */
+	copysize = (size < oldsize) ? size : oldsize;
+
+#ifdef JEMALLOC_MREMAP
+	/*
+	 * Use mremap(2) if this is a huge-->huge reallocation, and neither the
+	 * source nor the destination are in dss.
+	 */
+	if (oldsize >= chunksize && (config_dss == false || (chunk_in_dss(ptr)
+	    == false && chunk_in_dss(ret) == false))) {
+		size_t newsize = huge_salloc(ret);
+
+		/*
+		 * Remove ptr from the tree of huge allocations before
+		 * performing the remap operation, in order to avoid the
+		 * possibility of another thread acquiring that mapping before
+		 * this one removes it from the tree.
+		 */
+		huge_dalloc(ptr, false);
+		if (mremap(ptr, oldsize, newsize, MREMAP_MAYMOVE|MREMAP_FIXED,
+		    ret) == MAP_FAILED) {
+			/*
+			 * Assuming no chunk management bugs in the allocator,
+			 * the only documented way an error can occur here is
+			 * if the application changed the map type for a
+			 * portion of the old allocation.  This is firmly in
+			 * undefined behavior territory, so write a diagnostic
+			 * message, and optionally abort.
+			 */
+			char buf[BUFERROR_BUF];
+
+			buferror(get_errno(), buf, sizeof(buf));
+			malloc_printf("<jemalloc>: Error in mremap(): %s\n",
+			    buf);
+			if (opt_abort)
+				abort();
+			memcpy(ret, ptr, copysize);
+			chunk_dealloc_mmap(ptr, oldsize);
+		} else if (config_fill && zero == false && opt_junk && oldsize
+		    < newsize) {
+			/*
+			 * mremap(2) clobbers the original mapping, so
+			 * junk/zero filling is not preserved.  There is no
+			 * need to zero fill here, since any trailing
+			 * uninititialized memory is demand-zeroed by the
+			 * kernel, but junk filling must be redone.
+			 */
+			memset(ret + oldsize, 0xa5, newsize - oldsize);
+		}
+	} else
+#endif
+	{
+		memcpy(ret, ptr, copysize);
+		iqalloct(ptr, try_tcache_dalloc);
+	}
+	return (ret);
+}
+
+#ifdef JEMALLOC_JET
+#undef huge_dalloc_junk
+#define	huge_dalloc_junk JEMALLOC_N(huge_dalloc_junk_impl)
+#endif
+static void
+huge_dalloc_junk(void *ptr, size_t usize)
+{
+
+	if (config_fill && config_dss && opt_junk) {
+		/*
+		 * Only bother junk filling if the chunk isn't about to be
+		 * unmapped.
+		 */
+		if (config_munmap == false || (config_dss && chunk_in_dss(ptr)))
+			memset(ptr, 0x5a, usize);
+	}
+}
+#ifdef JEMALLOC_JET
+#undef huge_dalloc_junk
+#define	huge_dalloc_junk JEMALLOC_N(huge_dalloc_junk)
+huge_dalloc_junk_t *huge_dalloc_junk = JEMALLOC_N(huge_dalloc_junk_impl);
+#endif
+
+void
+huge_dalloc(void *ptr, bool unmap)
+{
+	extent_node_t *node, key;
+
+	malloc_mutex_lock(&huge_mtx);
+
+	/* Extract from tree of huge allocations. */
+	key.addr = ptr;
+	node = extent_tree_ad_search(&huge, &key);
+	assert(node != NULL);
+	assert(node->addr == ptr);
+	extent_tree_ad_remove(&huge, node);
+
+	if (config_stats) {
+		stats_cactive_sub(node->size);
+		huge_ndalloc++;
+		huge_allocated -= node->size;
+	}
+
+	malloc_mutex_unlock(&huge_mtx);
+
+	if (unmap)
+		huge_dalloc_junk(node->addr, node->size);
+
+	chunk_dealloc(node->addr, node->size, unmap);
+
+	base_node_dealloc(node);
+}
+
+size_t
+huge_salloc(const void *ptr)
+{
+	size_t ret;
+	extent_node_t *node, key;
+
+	malloc_mutex_lock(&huge_mtx);
+
+	/* Extract from tree of huge allocations. */
+	key.addr = __DECONST(void *, ptr);
+	node = extent_tree_ad_search(&huge, &key);
+	assert(node != NULL);
+
+	ret = node->size;
+
+	malloc_mutex_unlock(&huge_mtx);
+
+	return (ret);
+}
+
+dss_prec_t
+huge_dss_prec_get(arena_t *arena)
+{
+
+	return (arena_dss_prec_get(choose_arena(arena)));
+}
+
+prof_ctx_t *
+huge_prof_ctx_get(const void *ptr)
+{
+	prof_ctx_t *ret;
+	extent_node_t *node, key;
+
+	malloc_mutex_lock(&huge_mtx);
+
+	/* Extract from tree of huge allocations. */
+	key.addr = __DECONST(void *, ptr);
+	node = extent_tree_ad_search(&huge, &key);
+	assert(node != NULL);
+
+	ret = node->prof_ctx;
+
+	malloc_mutex_unlock(&huge_mtx);
+
+	return (ret);
+}
+
+void
+huge_prof_ctx_set(const void *ptr, prof_ctx_t *ctx)
+{
+	extent_node_t *node, key;
+
+	malloc_mutex_lock(&huge_mtx);
+
+	/* Extract from tree of huge allocations. */
+	key.addr = __DECONST(void *, ptr);
+	node = extent_tree_ad_search(&huge, &key);
+	assert(node != NULL);
+
+	node->prof_ctx = ctx;
+
+	malloc_mutex_unlock(&huge_mtx);
+}
+
+bool
+huge_boot(void)
+{
+
+	/* Initialize chunks data. */
+	if (malloc_mutex_init(&huge_mtx))
+		return (true);
+	extent_tree_ad_new(&huge);
+
+	if (config_stats) {
+		huge_nmalloc = 0;
+		huge_ndalloc = 0;
+		huge_allocated = 0;
+	}
+
+	return (false);
+}
+
+void
+huge_prefork(void)
+{
+
+	malloc_mutex_prefork(&huge_mtx);
+}
+
+void
+huge_postfork_parent(void)
+{
+
+	malloc_mutex_postfork_parent(&huge_mtx);
+}
+
+void
+huge_postfork_child(void)
+{
+
+	malloc_mutex_postfork_child(&huge_mtx);
+}

deps/jemalloc/src/jemalloc_cpp.cpp

@@ -0,0 +1,141 @@
+#include <mutex>
+#include <new>
+
+#define JEMALLOC_CPP_CPP_
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#ifdef __cplusplus
+}
+#endif
+
+// All operators in this file are exported.
+
+// Possibly alias hidden versions of malloc and sdallocx to avoid an extra plt
+// thunk?
+//
+// extern __typeof (sdallocx) sdallocx_int
+//  __attribute ((alias ("sdallocx"),
+//		visibility ("hidden")));
+//
+// ... but it needs to work with jemalloc namespaces.
+
+void	*operator new(std::size_t size);
+void	*operator new[](std::size_t size);
+void	*operator new(std::size_t size, const std::nothrow_t &) noexcept;
+void	*operator new[](std::size_t size, const std::nothrow_t &) noexcept;
+void	operator delete(void *ptr) noexcept;
+void	operator delete[](void *ptr) noexcept;
+void	operator delete(void *ptr, const std::nothrow_t &) noexcept;
+void	operator delete[](void *ptr, const std::nothrow_t &) noexcept;
+
+#if __cpp_sized_deallocation >= 201309
+/* C++14's sized-delete operators. */
+void	operator delete(void *ptr, std::size_t size) noexcept;
+void	operator delete[](void *ptr, std::size_t size) noexcept;
+#endif
+
+JEMALLOC_NOINLINE
+static void *
+handleOOM(std::size_t size, bool nothrow) {
+	void *ptr = nullptr;
+
+	while (ptr == nullptr) {
+		std::new_handler handler;
+		// GCC-4.8 and clang 4.0 do not have std::get_new_handler.
+		{
+			static std::mutex mtx;
+			std::lock_guard<std::mutex> lock(mtx);
+
+			handler = std::set_new_handler(nullptr);
+			std::set_new_handler(handler);
+		}
+		if (handler == nullptr)
+			break;
+
+		try {
+			handler();
+		} catch (const std::bad_alloc &) {
+			break;
+		}
+
+		ptr = je_malloc(size);
+	}
+
+	if (ptr == nullptr && !nothrow)
+		std::__throw_bad_alloc();
+	return ptr;
+}
+
+template <bool IsNoExcept>
+JEMALLOC_ALWAYS_INLINE
+void *
+newImpl(std::size_t size) noexcept(IsNoExcept) {
+	void *ptr = je_malloc(size);
+	if (likely(ptr != nullptr))
+		return ptr;
+
+	return handleOOM(size, IsNoExcept);
+}
+
+void *
+operator new(std::size_t size) {
+	return newImpl<false>(size);
+}
+
+void *
+operator new[](std::size_t size) {
+	return newImpl<false>(size);
+}
+
+void *
+operator new(std::size_t size, const std::nothrow_t &) noexcept {
+	return newImpl<true>(size);
+}
+
+void *
+operator new[](std::size_t size, const std::nothrow_t &) noexcept {
+	return newImpl<true>(size);
+}
+
+void
+operator delete(void *ptr) noexcept {
+	je_free(ptr);
+}
+
+void
+operator delete[](void *ptr) noexcept {
+	je_free(ptr);
+}
+
+void
+operator delete(void *ptr, const std::nothrow_t &) noexcept {
+	je_free(ptr);
+}
+
+void operator delete[](void *ptr, const std::nothrow_t &) noexcept {
+	je_free(ptr);
+}
+
+#if __cpp_sized_deallocation >= 201309
+
+void
+operator delete(void *ptr, std::size_t size) noexcept {
+	if (unlikely(ptr == nullptr)) {
+		return;
+	}
+	je_sdallocx_noflags(ptr, size);
+}
+
+void operator delete[](void *ptr, std::size_t size) noexcept {
+	if (unlikely(ptr == nullptr)) {
+		return;
+	}
+	je_sdallocx_noflags(ptr, size);
+}
+
+#endif  // __cpp_sized_deallocation

deps/jemalloc/src/large.c

@@ -0,0 +1,395 @@
+#define JEMALLOC_LARGE_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/extent_mmap.h"
+#include "jemalloc/internal/mutex.h"
+#include "jemalloc/internal/rtree.h"
+#include "jemalloc/internal/util.h"
+
+/******************************************************************************/
+
+void *
+large_malloc(tsdn_t *tsdn, arena_t *arena, size_t usize, bool zero) {
+	assert(usize == sz_s2u(usize));
+
+	return large_palloc(tsdn, arena, usize, CACHELINE, zero);
+}
+
+void *
+large_palloc(tsdn_t *tsdn, arena_t *arena, size_t usize, size_t alignment,
+    bool zero) {
+	size_t ausize;
+	extent_t *extent;
+	bool is_zeroed;
+	UNUSED bool idump JEMALLOC_CC_SILENCE_INIT(false);
+
+	assert(!tsdn_null(tsdn) || arena != NULL);
+
+	ausize = sz_sa2u(usize, alignment);
+	if (unlikely(ausize == 0 || ausize > SC_LARGE_MAXCLASS)) {
+		return NULL;
+	}
+
+	if (config_fill && unlikely(opt_zero)) {
+		zero = true;
+	}
+	/*
+	 * Copy zero into is_zeroed and pass the copy when allocating the
+	 * extent, so that it is possible to make correct junk/zero fill
+	 * decisions below, even if is_zeroed ends up true when zero is false.
+	 */
+	is_zeroed = zero;
+	if (likely(!tsdn_null(tsdn))) {
+		arena = arena_choose_maybe_huge(tsdn_tsd(tsdn), arena, usize);
+	}
+	if (unlikely(arena == NULL) || (extent = arena_extent_alloc_large(tsdn,
+	    arena, usize, alignment, &is_zeroed)) == NULL) {
+		return NULL;
+	}
+
+	/* See comments in arena_bin_slabs_full_insert(). */
+	if (!arena_is_auto(arena)) {
+		/* Insert extent into large. */
+		malloc_mutex_lock(tsdn, &arena->large_mtx);
+		extent_list_append(&arena->large, extent);
+		malloc_mutex_unlock(tsdn, &arena->large_mtx);
+	}
+	if (config_prof && arena_prof_accum(tsdn, arena, usize)) {
+		prof_idump(tsdn);
+	}
+
+	if (zero) {
+		assert(is_zeroed);
+	} else if (config_fill && unlikely(opt_junk_alloc)) {
+		memset(extent_addr_get(extent), JEMALLOC_ALLOC_JUNK,
+		    extent_usize_get(extent));
+	}
+
+	arena_decay_tick(tsdn, arena);
+	return extent_addr_get(extent);
+}
+
+static void
+large_dalloc_junk_impl(void *ptr, size_t size) {
+	memset(ptr, JEMALLOC_FREE_JUNK, size);
+}
+large_dalloc_junk_t *JET_MUTABLE large_dalloc_junk = large_dalloc_junk_impl;
+
+static void
+large_dalloc_maybe_junk_impl(void *ptr, size_t size) {
+	if (config_fill && have_dss && unlikely(opt_junk_free)) {
+		/*
+		 * Only bother junk filling if the extent isn't about to be
+		 * unmapped.
+		 */
+		if (opt_retain || (have_dss && extent_in_dss(ptr))) {
+			large_dalloc_junk(ptr, size);
+		}
+	}
+}
+large_dalloc_maybe_junk_t *JET_MUTABLE large_dalloc_maybe_junk =
+    large_dalloc_maybe_junk_impl;
+
+static bool
+large_ralloc_no_move_shrink(tsdn_t *tsdn, extent_t *extent, size_t usize) {
+	arena_t *arena = extent_arena_get(extent);
+	size_t oldusize = extent_usize_get(extent);
+	extent_hooks_t *extent_hooks = extent_hooks_get(arena);
+	size_t diff = extent_size_get(extent) - (usize + sz_large_pad);
+
+	assert(oldusize > usize);
+
+	if (extent_hooks->split == NULL) {
+		return true;
+	}
+
+	/* Split excess pages. */
+	if (diff != 0) {
+		extent_t *trail = extent_split_wrapper(tsdn, arena,
+		    &extent_hooks, extent, usize + sz_large_pad,
+		    sz_size2index(usize), false, diff, SC_NSIZES, false);
+		if (trail == NULL) {
+			return true;
+		}
+
+		if (config_fill && unlikely(opt_junk_free)) {
+			large_dalloc_maybe_junk(extent_addr_get(trail),
+			    extent_size_get(trail));
+		}
+
+		arena_extents_dirty_dalloc(tsdn, arena, &extent_hooks, trail);
+	}
+
+	arena_extent_ralloc_large_shrink(tsdn, arena, extent, oldusize);
+
+	return false;
+}
+
+static bool
+large_ralloc_no_move_expand(tsdn_t *tsdn, extent_t *extent, size_t usize,
+    bool zero) {
+	arena_t *arena = extent_arena_get(extent);
+	size_t oldusize = extent_usize_get(extent);
+	extent_hooks_t *extent_hooks = extent_hooks_get(arena);
+	size_t trailsize = usize - oldusize;
+
+	if (extent_hooks->merge == NULL) {
+		return true;
+	}
+
+	if (config_fill && unlikely(opt_zero)) {
+		zero = true;
+	}
+	/*
+	 * Copy zero into is_zeroed_trail and pass the copy when allocating the
+	 * extent, so that it is possible to make correct junk/zero fill
+	 * decisions below, even if is_zeroed_trail ends up true when zero is
+	 * false.
+	 */
+	bool is_zeroed_trail = zero;
+	bool commit = true;
+	extent_t *trail;
+	bool new_mapping;
+	if ((trail = extents_alloc(tsdn, arena, &extent_hooks,
+	    &arena->extents_dirty, extent_past_get(extent), trailsize, 0,
+	    CACHELINE, false, SC_NSIZES, &is_zeroed_trail, &commit)) != NULL
+	    || (trail = extents_alloc(tsdn, arena, &extent_hooks,
+	    &arena->extents_muzzy, extent_past_get(extent), trailsize, 0,
+	    CACHELINE, false, SC_NSIZES, &is_zeroed_trail, &commit)) != NULL) {
+		if (config_stats) {
+			new_mapping = false;
+		}
+	} else {
+		if ((trail = extent_alloc_wrapper(tsdn, arena, &extent_hooks,
+		    extent_past_get(extent), trailsize, 0, CACHELINE, false,
+		    SC_NSIZES, &is_zeroed_trail, &commit)) == NULL) {
+			return true;
+		}
+		if (config_stats) {
+			new_mapping = true;
+		}
+	}
+
+	if (extent_merge_wrapper(tsdn, arena, &extent_hooks, extent, trail)) {
+		extent_dalloc_wrapper(tsdn, arena, &extent_hooks, trail);
+		return true;
+	}
+	rtree_ctx_t rtree_ctx_fallback;
+	rtree_ctx_t *rtree_ctx = tsdn_rtree_ctx(tsdn, &rtree_ctx_fallback);
+	szind_t szind = sz_size2index(usize);
+	extent_szind_set(extent, szind);
+	rtree_szind_slab_update(tsdn, &extents_rtree, rtree_ctx,
+	    (uintptr_t)extent_addr_get(extent), szind, false);
+
+	if (config_stats && new_mapping) {
+		arena_stats_mapped_add(tsdn, &arena->stats, trailsize);
+	}
+
+	if (zero) {
+		if (config_cache_oblivious) {
+			/*
+			 * Zero the trailing bytes of the original allocation's
+			 * last page, since they are in an indeterminate state.
+			 * There will always be trailing bytes, because ptr's
+			 * offset from the beginning of the extent is a multiple
+			 * of CACHELINE in [0 .. PAGE).
+			 */
+			void *zbase = (void *)
+			    ((uintptr_t)extent_addr_get(extent) + oldusize);
+			void *zpast = PAGE_ADDR2BASE((void *)((uintptr_t)zbase +
+			    PAGE));
+			size_t nzero = (uintptr_t)zpast - (uintptr_t)zbase;
+			assert(nzero > 0);
+			memset(zbase, 0, nzero);
+		}
+		assert(is_zeroed_trail);
+	} else if (config_fill && unlikely(opt_junk_alloc)) {
+		memset((void *)((uintptr_t)extent_addr_get(extent) + oldusize),
+		    JEMALLOC_ALLOC_JUNK, usize - oldusize);
+	}
+
+	arena_extent_ralloc_large_expand(tsdn, arena, extent, oldusize);
+
+	return false;
+}
+
+bool
+large_ralloc_no_move(tsdn_t *tsdn, extent_t *extent, size_t usize_min,
+    size_t usize_max, bool zero) {
+	size_t oldusize = extent_usize_get(extent);
+
+	/* The following should have been caught by callers. */
+	assert(usize_min > 0 && usize_max <= SC_LARGE_MAXCLASS);
+	/* Both allocation sizes must be large to avoid a move. */
+	assert(oldusize >= SC_LARGE_MINCLASS
+	    && usize_max >= SC_LARGE_MINCLASS);
+
+	if (usize_max > oldusize) {
+		/* Attempt to expand the allocation in-place. */
+		if (!large_ralloc_no_move_expand(tsdn, extent, usize_max,
+		    zero)) {
+			arena_decay_tick(tsdn, extent_arena_get(extent));
+			return false;
+		}
+		/* Try again, this time with usize_min. */
+		if (usize_min < usize_max && usize_min > oldusize &&
+		    large_ralloc_no_move_expand(tsdn, extent, usize_min,
+		    zero)) {
+			arena_decay_tick(tsdn, extent_arena_get(extent));
+			return false;
+		}
+	}
+
+	/*
+	 * Avoid moving the allocation if the existing extent size accommodates
+	 * the new size.
+	 */
+	if (oldusize >= usize_min && oldusize <= usize_max) {
+		arena_decay_tick(tsdn, extent_arena_get(extent));
+		return false;
+	}
+
+	/* Attempt to shrink the allocation in-place. */
+	if (oldusize > usize_max) {
+		if (!large_ralloc_no_move_shrink(tsdn, extent, usize_max)) {
+			arena_decay_tick(tsdn, extent_arena_get(extent));
+			return false;
+		}
+	}
+	return true;
+}
+
+static void *
+large_ralloc_move_helper(tsdn_t *tsdn, arena_t *arena, size_t usize,
+    size_t alignment, bool zero) {
+	if (alignment <= CACHELINE) {
+		return large_malloc(tsdn, arena, usize, zero);
+	}
+	return large_palloc(tsdn, arena, usize, alignment, zero);
+}
+
+void *
+large_ralloc(tsdn_t *tsdn, arena_t *arena, void *ptr, size_t usize,
+    size_t alignment, bool zero, tcache_t *tcache,
+    hook_ralloc_args_t *hook_args) {
+	extent_t *extent = iealloc(tsdn, ptr);
+
+	size_t oldusize = extent_usize_get(extent);
+	/* The following should have been caught by callers. */
+	assert(usize > 0 && usize <= SC_LARGE_MAXCLASS);
+	/* Both allocation sizes must be large to avoid a move. */
+	assert(oldusize >= SC_LARGE_MINCLASS
+	    && usize >= SC_LARGE_MINCLASS);
+
+	/* Try to avoid moving the allocation. */
+	if (!large_ralloc_no_move(tsdn, extent, usize, usize, zero)) {
+		hook_invoke_expand(hook_args->is_realloc
+		    ? hook_expand_realloc : hook_expand_rallocx, ptr, oldusize,
+		    usize, (uintptr_t)ptr, hook_args->args);
+		return extent_addr_get(extent);
+	}
+
+	/*
+	 * usize and old size are different enough that we need to use a
+	 * different size class.  In that case, fall back to allocating new
+	 * space and copying.
+	 */
+	void *ret = large_ralloc_move_helper(tsdn, arena, usize, alignment,
+	    zero);
+	if (ret == NULL) {
+		return NULL;
+	}
+
+	hook_invoke_alloc(hook_args->is_realloc
+	    ? hook_alloc_realloc : hook_alloc_rallocx, ret, (uintptr_t)ret,
+	    hook_args->args);
+	hook_invoke_dalloc(hook_args->is_realloc
+	    ? hook_dalloc_realloc : hook_dalloc_rallocx, ptr, hook_args->args);
+
+	size_t copysize = (usize < oldusize) ? usize : oldusize;
+	memcpy(ret, extent_addr_get(extent), copysize);
+	isdalloct(tsdn, extent_addr_get(extent), oldusize, tcache, NULL, true);
+	return ret;
+}
+
+/*
+ * junked_locked indicates whether the extent's data have been junk-filled, and
+ * whether the arena's large_mtx is currently held.
+ */
+static void
+large_dalloc_prep_impl(tsdn_t *tsdn, arena_t *arena, extent_t *extent,
+    bool junked_locked) {
+	if (!junked_locked) {
+		/* See comments in arena_bin_slabs_full_insert(). */
+		if (!arena_is_auto(arena)) {
+			malloc_mutex_lock(tsdn, &arena->large_mtx);
+			extent_list_remove(&arena->large, extent);
+			malloc_mutex_unlock(tsdn, &arena->large_mtx);
+		}
+		large_dalloc_maybe_junk(extent_addr_get(extent),
+		    extent_usize_get(extent));
+	} else {
+		/* Only hold the large_mtx if necessary. */
+		if (!arena_is_auto(arena)) {
+			malloc_mutex_assert_owner(tsdn, &arena->large_mtx);
+			extent_list_remove(&arena->large, extent);
+		}
+	}
+	arena_extent_dalloc_large_prep(tsdn, arena, extent);
+}
+
+static void
+large_dalloc_finish_impl(tsdn_t *tsdn, arena_t *arena, extent_t *extent) {
+	extent_hooks_t *extent_hooks = EXTENT_HOOKS_INITIALIZER;
+	arena_extents_dirty_dalloc(tsdn, arena, &extent_hooks, extent);
+}
+
+void
+large_dalloc_prep_junked_locked(tsdn_t *tsdn, extent_t *extent) {
+	large_dalloc_prep_impl(tsdn, extent_arena_get(extent), extent, true);
+}
+
+void
+large_dalloc_finish(tsdn_t *tsdn, extent_t *extent) {
+	large_dalloc_finish_impl(tsdn, extent_arena_get(extent), extent);
+}
+
+void
+large_dalloc(tsdn_t *tsdn, extent_t *extent) {
+	arena_t *arena = extent_arena_get(extent);
+	large_dalloc_prep_impl(tsdn, arena, extent, false);
+	large_dalloc_finish_impl(tsdn, arena, extent);
+	arena_decay_tick(tsdn, arena);
+}
+
+size_t
+large_salloc(tsdn_t *tsdn, const extent_t *extent) {
+	return extent_usize_get(extent);
+}
+
+prof_tctx_t *
+large_prof_tctx_get(tsdn_t *tsdn, const extent_t *extent) {
+	return extent_prof_tctx_get(extent);
+}
+
+void
+large_prof_tctx_set(tsdn_t *tsdn, extent_t *extent, prof_tctx_t *tctx) {
+	extent_prof_tctx_set(extent, tctx);
+}
+
+void
+large_prof_tctx_reset(tsdn_t *tsdn, extent_t *extent) {
+	large_prof_tctx_set(tsdn, extent, (prof_tctx_t *)(uintptr_t)1U);
+}
+
+nstime_t
+large_prof_alloc_time_get(const extent_t *extent) {
+	return extent_prof_alloc_time_get(extent);
+}
+
+void
+large_prof_alloc_time_set(extent_t *extent, nstime_t t) {
+	extent_prof_alloc_time_set(extent, t);
+}

deps/jemalloc/src/log.c

@@ -0,0 +1,78 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/log.h"
+
+char log_var_names[JEMALLOC_LOG_VAR_BUFSIZE];
+atomic_b_t log_init_done = ATOMIC_INIT(false);
+
+/*
+ * Returns true if we were able to pick out a segment.  Fills in r_segment_end
+ * with a pointer to the first character after the end of the string.
+ */
+static const char *
+log_var_extract_segment(const char* segment_begin) {
+	const char *end;
+	for (end = segment_begin; *end != '\0' && *end != '|'; end++) {
+	}
+	return end;
+}
+
+static bool
+log_var_matches_segment(const char *segment_begin, const char *segment_end,
+    const char *log_var_begin, const char *log_var_end) {
+	assert(segment_begin <= segment_end);
+	assert(log_var_begin < log_var_end);
+
+	ptrdiff_t segment_len = segment_end - segment_begin;
+	ptrdiff_t log_var_len = log_var_end - log_var_begin;
+	/* The special '.' segment matches everything. */
+	if (segment_len == 1 && *segment_begin == '.') {
+		return true;
+	}
+        if (segment_len == log_var_len) {
+		return strncmp(segment_begin, log_var_begin, segment_len) == 0;
+	} else if (segment_len < log_var_len) {
+		return strncmp(segment_begin, log_var_begin, segment_len) == 0
+		    && log_var_begin[segment_len] == '.';
+        } else {
+		return false;
+	}
+}
+
+unsigned
+log_var_update_state(log_var_t *log_var) {
+	const char *log_var_begin = log_var->name;
+	const char *log_var_end = log_var->name + strlen(log_var->name);
+
+	/* Pointer to one before the beginning of the current segment. */
+	const char *segment_begin = log_var_names;
+
+	/*
+	 * If log_init done is false, we haven't parsed the malloc conf yet.  To
+	 * avoid log-spew, we default to not displaying anything.
+	 */
+	if (!atomic_load_b(&log_init_done, ATOMIC_ACQUIRE)) {
+		return LOG_INITIALIZED_NOT_ENABLED;
+	}
+
+	while (true) {
+		const char *segment_end = log_var_extract_segment(
+		    segment_begin);
+		assert(segment_end < log_var_names + JEMALLOC_LOG_VAR_BUFSIZE);
+		if (log_var_matches_segment(segment_begin, segment_end,
+		    log_var_begin, log_var_end)) {
+			atomic_store_u(&log_var->state, LOG_ENABLED,
+			    ATOMIC_RELAXED);
+			return LOG_ENABLED;
+		}
+		if (*segment_end == '\0') {
+			/* Hit the end of the segment string with no match. */
+			atomic_store_u(&log_var->state,
+			    LOG_INITIALIZED_NOT_ENABLED, ATOMIC_RELAXED);
+			return LOG_INITIALIZED_NOT_ENABLED;
+		}
+		/* Otherwise, skip the delimiter and continue. */
+		segment_begin = segment_end + 1;
+	}
+}

deps/jemalloc/src/malloc_io.c

@@ -0,0 +1,675 @@
+#define JEMALLOC_MALLOC_IO_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/malloc_io.h"
+#include "jemalloc/internal/util.h"
+
+#ifdef assert
+#  undef assert
+#endif
+#ifdef not_reached
+#  undef not_reached
+#endif
+#ifdef not_implemented
+#  undef not_implemented
+#endif
+#ifdef assert_not_implemented
+#  undef assert_not_implemented
+#endif
+
+/*
+ * Define simple versions of assertion macros that won't recurse in case
+ * of assertion failures in malloc_*printf().
+ */
+#define assert(e) do {							\
+	if (config_debug && !(e)) {					\
+		malloc_write("<jemalloc>: Failed assertion\n");		\
+		abort();						\
+	}								\
+} while (0)
+
+#define not_reached() do {						\
+	if (config_debug) {						\
+		malloc_write("<jemalloc>: Unreachable code reached\n");	\
+		abort();						\
+	}								\
+	unreachable();							\
+} while (0)
+
+#define not_implemented() do {						\
+	if (config_debug) {						\
+		malloc_write("<jemalloc>: Not implemented\n");		\
+		abort();						\
+	}								\
+} while (0)
+
+#define assert_not_implemented(e) do {					\
+	if (unlikely(config_debug && !(e))) {				\
+		not_implemented();					\
+	}								\
+} while (0)
+
+/******************************************************************************/
+/* Function prototypes for non-inline static functions. */
+
+static void wrtmessage(void *cbopaque, const char *s);
+#define U2S_BUFSIZE ((1U << (LG_SIZEOF_INTMAX_T + 3)) + 1)
+static char *u2s(uintmax_t x, unsigned base, bool uppercase, char *s,
+    size_t *slen_p);
+#define D2S_BUFSIZE (1 + U2S_BUFSIZE)
+static char *d2s(intmax_t x, char sign, char *s, size_t *slen_p);
+#define O2S_BUFSIZE (1 + U2S_BUFSIZE)
+static char *o2s(uintmax_t x, bool alt_form, char *s, size_t *slen_p);
+#define X2S_BUFSIZE (2 + U2S_BUFSIZE)
+static char *x2s(uintmax_t x, bool alt_form, bool uppercase, char *s,
+    size_t *slen_p);
+
+/******************************************************************************/
+
+/* malloc_message() setup. */
+static void
+wrtmessage(void *cbopaque, const char *s) {
+	malloc_write_fd(STDERR_FILENO, s, strlen(s));
+}
+
+JEMALLOC_EXPORT void	(*je_malloc_message)(void *, const char *s);
+
+/*
+ * Wrapper around malloc_message() that avoids the need for
+ * je_malloc_message(...) throughout the code.
+ */
+void
+malloc_write(const char *s) {
+	if (je_malloc_message != NULL) {
+		je_malloc_message(NULL, s);
+	} else {
+		wrtmessage(NULL, s);
+	}
+}
+
+/*
+ * glibc provides a non-standard strerror_r() when _GNU_SOURCE is defined, so
+ * provide a wrapper.
+ */
+int
+buferror(int err, char *buf, size_t buflen) {
+#ifdef _WIN32
+	FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM, NULL, err, 0,
+	    (LPSTR)buf, (DWORD)buflen, NULL);
+	return 0;
+#elif defined(JEMALLOC_STRERROR_R_RETURNS_CHAR_WITH_GNU_SOURCE) && defined(_GNU_SOURCE)
+	char *b = strerror_r(err, buf, buflen);
+	if (b != buf) {
+		strncpy(buf, b, buflen);
+		buf[buflen-1] = '\0';
+	}
+	return 0;
+#else
+	return strerror_r(err, buf, buflen);
+#endif
+}
+
+uintmax_t
+malloc_strtoumax(const char *restrict nptr, char **restrict endptr, int base) {
+	uintmax_t ret, digit;
+	unsigned b;
+	bool neg;
+	const char *p, *ns;
+
+	p = nptr;
+	if (base < 0 || base == 1 || base > 36) {
+		ns = p;
+		set_errno(EINVAL);
+		ret = UINTMAX_MAX;
+		goto label_return;
+	}
+	b = base;
+
+	/* Swallow leading whitespace and get sign, if any. */
+	neg = false;
+	while (true) {
+		switch (*p) {
+		case '\t': case '\n': case '\v': case '\f': case '\r': case ' ':
+			p++;
+			break;
+		case '-':
+			neg = true;
+			/* Fall through. */
+		case '+':
+			p++;
+			/* Fall through. */
+		default:
+			goto label_prefix;
+		}
+	}
+
+	/* Get prefix, if any. */
+	label_prefix:
+	/*
+	 * Note where the first non-whitespace/sign character is so that it is
+	 * possible to tell whether any digits are consumed (e.g., "  0" vs.
+	 * "  -x").
+	 */
+	ns = p;
+	if (*p == '0') {
+		switch (p[1]) {
+		case '0': case '1': case '2': case '3': case '4': case '5':
+		case '6': case '7':
+			if (b == 0) {
+				b = 8;
+			}
+			if (b == 8) {
+				p++;
+			}
+			break;
+		case 'X': case 'x':
+			switch (p[2]) {
+			case '0': case '1': case '2': case '3': case '4':
+			case '5': case '6': case '7': case '8': case '9':
+			case 'A': case 'B': case 'C': case 'D': case 'E':
+			case 'F':
+			case 'a': case 'b': case 'c': case 'd': case 'e':
+			case 'f':
+				if (b == 0) {
+					b = 16;
+				}
+				if (b == 16) {
+					p += 2;
+				}
+				break;
+			default:
+				break;
+			}
+			break;
+		default:
+			p++;
+			ret = 0;
+			goto label_return;
+		}
+	}
+	if (b == 0) {
+		b = 10;
+	}
+
+	/* Convert. */
+	ret = 0;
+	while ((*p >= '0' && *p <= '9' && (digit = *p - '0') < b)
+	    || (*p >= 'A' && *p <= 'Z' && (digit = 10 + *p - 'A') < b)
+	    || (*p >= 'a' && *p <= 'z' && (digit = 10 + *p - 'a') < b)) {
+		uintmax_t pret = ret;
+		ret *= b;
+		ret += digit;
+		if (ret < pret) {
+			/* Overflow. */
+			set_errno(ERANGE);
+			ret = UINTMAX_MAX;
+			goto label_return;
+		}
+		p++;
+	}
+	if (neg) {
+		ret = (uintmax_t)(-((intmax_t)ret));
+	}
+
+	if (p == ns) {
+		/* No conversion performed. */
+		set_errno(EINVAL);
+		ret = UINTMAX_MAX;
+		goto label_return;
+	}
+
+label_return:
+	if (endptr != NULL) {
+		if (p == ns) {
+			/* No characters were converted. */
+			*endptr = (char *)nptr;
+		} else {
+			*endptr = (char *)p;
+		}
+	}
+	return ret;
+}
+
+static char *
+u2s(uintmax_t x, unsigned base, bool uppercase, char *s, size_t *slen_p) {
+	unsigned i;
+
+	i = U2S_BUFSIZE - 1;
+	s[i] = '\0';
+	switch (base) {
+	case 10:
+		do {
+			i--;
+			s[i] = "0123456789"[x % (uint64_t)10];
+			x /= (uint64_t)10;
+		} while (x > 0);
+		break;
+	case 16: {
+		const char *digits = (uppercase)
+		    ? "0123456789ABCDEF"
+		    : "0123456789abcdef";
+
+		do {
+			i--;
+			s[i] = digits[x & 0xf];
+			x >>= 4;
+		} while (x > 0);
+		break;
+	} default: {
+		const char *digits = (uppercase)
+		    ? "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"
+		    : "0123456789abcdefghijklmnopqrstuvwxyz";
+
+		assert(base >= 2 && base <= 36);
+		do {
+			i--;
+			s[i] = digits[x % (uint64_t)base];
+			x /= (uint64_t)base;
+		} while (x > 0);
+	}}
+
+	*slen_p = U2S_BUFSIZE - 1 - i;
+	return &s[i];
+}
+
+static char *
+d2s(intmax_t x, char sign, char *s, size_t *slen_p) {
+	bool neg;
+
+	if ((neg = (x < 0))) {
+		x = -x;
+	}
+	s = u2s(x, 10, false, s, slen_p);
+	if (neg) {
+		sign = '-';
+	}
+	switch (sign) {
+	case '-':
+		if (!neg) {
+			break;
+		}
+		/* Fall through. */
+	case ' ':
+	case '+':
+		s--;
+		(*slen_p)++;
+		*s = sign;
+		break;
+	default: not_reached();
+	}
+	return s;
+}
+
+static char *
+o2s(uintmax_t x, bool alt_form, char *s, size_t *slen_p) {
+	s = u2s(x, 8, false, s, slen_p);
+	if (alt_form && *s != '0') {
+		s--;
+		(*slen_p)++;
+		*s = '0';
+	}
+	return s;
+}
+
+static char *
+x2s(uintmax_t x, bool alt_form, bool uppercase, char *s, size_t *slen_p) {
+	s = u2s(x, 16, uppercase, s, slen_p);
+	if (alt_form) {
+		s -= 2;
+		(*slen_p) += 2;
+		memcpy(s, uppercase ? "0X" : "0x", 2);
+	}
+	return s;
+}
+
+size_t
+malloc_vsnprintf(char *str, size_t size, const char *format, va_list ap) {
+	size_t i;
+	const char *f;
+
+#define APPEND_C(c) do {						\
+	if (i < size) {							\
+		str[i] = (c);						\
+	}								\
+	i++;								\
+} while (0)
+#define APPEND_S(s, slen) do {						\
+	if (i < size) {							\
+		size_t cpylen = (slen <= size - i) ? slen : size - i;	\
+		memcpy(&str[i], s, cpylen);				\
+	}								\
+	i += slen;							\
+} while (0)
+#define APPEND_PADDED_S(s, slen, width, left_justify) do {		\
+	/* Left padding. */						\
+	size_t pad_len = (width == -1) ? 0 : ((slen < (size_t)width) ?	\
+	    (size_t)width - slen : 0);					\
+	if (!left_justify && pad_len != 0) {				\
+		size_t j;						\
+		for (j = 0; j < pad_len; j++) {				\
+			APPEND_C(' ');					\
+		}							\
+	}								\
+	/* Value. */							\
+	APPEND_S(s, slen);						\
+	/* Right padding. */						\
+	if (left_justify && pad_len != 0) {				\
+		size_t j;						\
+		for (j = 0; j < pad_len; j++) {				\
+			APPEND_C(' ');					\
+		}							\
+	}								\
+} while (0)
+#define GET_ARG_NUMERIC(val, len) do {					\
+	switch ((unsigned char)len) {					\
+	case '?':							\
+		val = va_arg(ap, int);					\
+		break;							\
+	case '?' | 0x80:						\
+		val = va_arg(ap, unsigned int);				\
+		break;							\
+	case 'l':							\
+		val = va_arg(ap, long);					\
+		break;							\
+	case 'l' | 0x80:						\
+		val = va_arg(ap, unsigned long);			\
+		break;							\
+	case 'q':							\
+		val = va_arg(ap, long long);				\
+		break;							\
+	case 'q' | 0x80:						\
+		val = va_arg(ap, unsigned long long);			\
+		break;							\
+	case 'j':							\
+		val = va_arg(ap, intmax_t);				\
+		break;							\
+	case 'j' | 0x80:						\
+		val = va_arg(ap, uintmax_t);				\
+		break;							\
+	case 't':							\
+		val = va_arg(ap, ptrdiff_t);				\
+		break;							\
+	case 'z':							\
+		val = va_arg(ap, ssize_t);				\
+		break;							\
+	case 'z' | 0x80:						\
+		val = va_arg(ap, size_t);				\
+		break;							\
+	case 'p': /* Synthetic; used for %p. */				\
+		val = va_arg(ap, uintptr_t);				\
+		break;							\
+	default:							\
+		not_reached();						\
+		val = 0;						\
+	}								\
+} while (0)
+
+	i = 0;
+	f = format;
+	while (true) {
+		switch (*f) {
+		case '\0': goto label_out;
+		case '%': {
+			bool alt_form = false;
+			bool left_justify = false;
+			bool plus_space = false;
+			bool plus_plus = false;
+			int prec = -1;
+			int width = -1;
+			unsigned char len = '?';
+			char *s;
+			size_t slen;
+
+			f++;
+			/* Flags. */
+			while (true) {
+				switch (*f) {
+				case '#':
+					assert(!alt_form);
+					alt_form = true;
+					break;
+				case '-':
+					assert(!left_justify);
+					left_justify = true;
+					break;
+				case ' ':
+					assert(!plus_space);
+					plus_space = true;
+					break;
+				case '+':
+					assert(!plus_plus);
+					plus_plus = true;
+					break;
+				default: goto label_width;
+				}
+				f++;
+			}
+			/* Width. */
+			label_width:
+			switch (*f) {
+			case '*':
+				width = va_arg(ap, int);
+				f++;
+				if (width < 0) {
+					left_justify = true;
+					width = -width;
+				}
+				break;
+			case '0': case '1': case '2': case '3': case '4':
+			case '5': case '6': case '7': case '8': case '9': {
+				uintmax_t uwidth;
+				set_errno(0);
+				uwidth = malloc_strtoumax(f, (char **)&f, 10);
+				assert(uwidth != UINTMAX_MAX || get_errno() !=
+				    ERANGE);
+				width = (int)uwidth;
+				break;
+			} default:
+				break;
+			}
+			/* Width/precision separator. */
+			if (*f == '.') {
+				f++;
+			} else {
+				goto label_length;
+			}
+			/* Precision. */
+			switch (*f) {
+			case '*':
+				prec = va_arg(ap, int);
+				f++;
+				break;
+			case '0': case '1': case '2': case '3': case '4':
+			case '5': case '6': case '7': case '8': case '9': {
+				uintmax_t uprec;
+				set_errno(0);
+				uprec = malloc_strtoumax(f, (char **)&f, 10);
+				assert(uprec != UINTMAX_MAX || get_errno() !=
+				    ERANGE);
+				prec = (int)uprec;
+				break;
+			}
+			default: break;
+			}
+			/* Length. */
+			label_length:
+			switch (*f) {
+			case 'l':
+				f++;
+				if (*f == 'l') {
+					len = 'q';
+					f++;
+				} else {
+					len = 'l';
+				}
+				break;
+			case 'q': case 'j': case 't': case 'z':
+				len = *f;
+				f++;
+				break;
+			default: break;
+			}
+			/* Conversion specifier. */
+			switch (*f) {
+			case '%':
+				/* %% */
+				APPEND_C(*f);
+				f++;
+				break;
+			case 'd': case 'i': {
+				intmax_t val JEMALLOC_CC_SILENCE_INIT(0);
+				char buf[D2S_BUFSIZE];
+
+				GET_ARG_NUMERIC(val, len);
+				s = d2s(val, (plus_plus ? '+' : (plus_space ?
+				    ' ' : '-')), buf, &slen);
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			} case 'o': {
+				uintmax_t val JEMALLOC_CC_SILENCE_INIT(0);
+				char buf[O2S_BUFSIZE];
+
+				GET_ARG_NUMERIC(val, len | 0x80);
+				s = o2s(val, alt_form, buf, &slen);
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			} case 'u': {
+				uintmax_t val JEMALLOC_CC_SILENCE_INIT(0);
+				char buf[U2S_BUFSIZE];
+
+				GET_ARG_NUMERIC(val, len | 0x80);
+				s = u2s(val, 10, false, buf, &slen);
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			} case 'x': case 'X': {
+				uintmax_t val JEMALLOC_CC_SILENCE_INIT(0);
+				char buf[X2S_BUFSIZE];
+
+				GET_ARG_NUMERIC(val, len | 0x80);
+				s = x2s(val, alt_form, *f == 'X', buf, &slen);
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			} case 'c': {
+				unsigned char val;
+				char buf[2];
+
+				assert(len == '?' || len == 'l');
+				assert_not_implemented(len != 'l');
+				val = va_arg(ap, int);
+				buf[0] = val;
+				buf[1] = '\0';
+				APPEND_PADDED_S(buf, 1, width, left_justify);
+				f++;
+				break;
+			} case 's':
+				assert(len == '?' || len == 'l');
+				assert_not_implemented(len != 'l');
+				s = va_arg(ap, char *);
+				slen = (prec < 0) ? strlen(s) : (size_t)prec;
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			case 'p': {
+				uintmax_t val;
+				char buf[X2S_BUFSIZE];
+
+				GET_ARG_NUMERIC(val, 'p');
+				s = x2s(val, true, false, buf, &slen);
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			} default: not_reached();
+			}
+			break;
+		} default: {
+			APPEND_C(*f);
+			f++;
+			break;
+		}}
+	}
+	label_out:
+	if (i < size) {
+		str[i] = '\0';
+	} else {
+		str[size - 1] = '\0';
+	}
+
+#undef APPEND_C
+#undef APPEND_S
+#undef APPEND_PADDED_S
+#undef GET_ARG_NUMERIC
+	return i;
+}
+
+JEMALLOC_FORMAT_PRINTF(3, 4)
+size_t
+malloc_snprintf(char *str, size_t size, const char *format, ...) {
+	size_t ret;
+	va_list ap;
+
+	va_start(ap, format);
+	ret = malloc_vsnprintf(str, size, format, ap);
+	va_end(ap);
+
+	return ret;
+}
+
+void
+malloc_vcprintf(void (*write_cb)(void *, const char *), void *cbopaque,
+    const char *format, va_list ap) {
+	char buf[MALLOC_PRINTF_BUFSIZE];
+
+	if (write_cb == NULL) {
+		/*
+		 * The caller did not provide an alternate write_cb callback
+		 * function, so use the default one.  malloc_write() is an
+		 * inline function, so use malloc_message() directly here.
+		 */
+		write_cb = (je_malloc_message != NULL) ? je_malloc_message :
+		    wrtmessage;
+	}
+
+	malloc_vsnprintf(buf, sizeof(buf), format, ap);
+	write_cb(cbopaque, buf);
+}
+
+/*
+ * Print to a callback function in such a way as to (hopefully) avoid memory
+ * allocation.
+ */
+JEMALLOC_FORMAT_PRINTF(3, 4)
+void
+malloc_cprintf(void (*write_cb)(void *, const char *), void *cbopaque,
+    const char *format, ...) {
+	va_list ap;
+
+	va_start(ap, format);
+	malloc_vcprintf(write_cb, cbopaque, format, ap);
+	va_end(ap);
+}
+
+/* Print to stderr in such a way as to avoid memory allocation. */
+JEMALLOC_FORMAT_PRINTF(1, 2)
+void
+malloc_printf(const char *format, ...) {
+	va_list ap;
+
+	va_start(ap, format);
+	malloc_vcprintf(NULL, NULL, format, ap);
+	va_end(ap);
+}
+
+/*
+ * Restore normal assertion macros, in order to make it possible to compile all
+ * C files as a single concatenation.
+ */
+#undef assert
+#undef not_reached
+#undef not_implemented
+#undef assert_not_implemented
+#include "jemalloc/internal/assert.h"

deps/jemalloc/src/mb.c

@@ -0,0 +1,2 @@
+#define	JEMALLOC_MB_C_
+#include "jemalloc/internal/jemalloc_internal.h"

deps/jemalloc/src/mutex.c

@@ -0,0 +1,223 @@
+#define JEMALLOC_MUTEX_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/malloc_io.h"
+#include "jemalloc/internal/spin.h"
+
+#ifndef _CRT_SPINCOUNT
+#define _CRT_SPINCOUNT 4000
+#endif
+
+/******************************************************************************/
+/* Data. */
+
+#ifdef JEMALLOC_LAZY_LOCK
+bool isthreaded = false;
+#endif
+#ifdef JEMALLOC_MUTEX_INIT_CB
+static bool		postpone_init = true;
+static malloc_mutex_t	*postponed_mutexes = NULL;
+#endif
+
+/******************************************************************************/
+/*
+ * We intercept pthread_create() calls in order to toggle isthreaded if the
+ * process goes multi-threaded.
+ */
+
+#if defined(JEMALLOC_LAZY_LOCK) && !defined(_WIN32)
+JEMALLOC_EXPORT int
+pthread_create(pthread_t *__restrict thread,
+    const pthread_attr_t *__restrict attr, void *(*start_routine)(void *),
+    void *__restrict arg) {
+	return pthread_create_wrapper(thread, attr, start_routine, arg);
+}
+#endif
+
+/******************************************************************************/
+
+#ifdef JEMALLOC_MUTEX_INIT_CB
+JEMALLOC_EXPORT int	_pthread_mutex_init_calloc_cb(pthread_mutex_t *mutex,
+    void *(calloc_cb)(size_t, size_t));
+#endif
+
+void
+malloc_mutex_lock_slow(malloc_mutex_t *mutex) {
+	mutex_prof_data_t *data = &mutex->prof_data;
+	nstime_t before = NSTIME_ZERO_INITIALIZER;
+
+	if (ncpus == 1) {
+		goto label_spin_done;
+	}
+
+	int cnt = 0, max_cnt = MALLOC_MUTEX_MAX_SPIN;
+	do {
+		spin_cpu_spinwait();
+		if (!atomic_load_b(&mutex->locked, ATOMIC_RELAXED)
+                    && !malloc_mutex_trylock_final(mutex)) {
+			data->n_spin_acquired++;
+			return;
+		}
+	} while (cnt++ < max_cnt);
+
+	if (!config_stats) {
+		/* Only spin is useful when stats is off. */
+		malloc_mutex_lock_final(mutex);
+		return;
+	}
+label_spin_done:
+	nstime_update(&before);
+	/* Copy before to after to avoid clock skews. */
+	nstime_t after;
+	nstime_copy(&after, &before);
+	uint32_t n_thds = atomic_fetch_add_u32(&data->n_waiting_thds, 1,
+	    ATOMIC_RELAXED) + 1;
+	/* One last try as above two calls may take quite some cycles. */
+	if (!malloc_mutex_trylock_final(mutex)) {
+		atomic_fetch_sub_u32(&data->n_waiting_thds, 1, ATOMIC_RELAXED);
+		data->n_spin_acquired++;
+		return;
+	}
+
+	/* True slow path. */
+	malloc_mutex_lock_final(mutex);
+	/* Update more slow-path only counters. */
+	atomic_fetch_sub_u32(&data->n_waiting_thds, 1, ATOMIC_RELAXED);
+	nstime_update(&after);
+
+	nstime_t delta;
+	nstime_copy(&delta, &after);
+	nstime_subtract(&delta, &before);
+
+	data->n_wait_times++;
+	nstime_add(&data->tot_wait_time, &delta);
+	if (nstime_compare(&data->max_wait_time, &delta) < 0) {
+		nstime_copy(&data->max_wait_time, &delta);
+	}
+	if (n_thds > data->max_n_thds) {
+		data->max_n_thds = n_thds;
+	}
+}
+
+static void
+mutex_prof_data_init(mutex_prof_data_t *data) {
+	memset(data, 0, sizeof(mutex_prof_data_t));
+	nstime_init(&data->max_wait_time, 0);
+	nstime_init(&data->tot_wait_time, 0);
+	data->prev_owner = NULL;
+}
+
+void
+malloc_mutex_prof_data_reset(tsdn_t *tsdn, malloc_mutex_t *mutex) {
+	malloc_mutex_assert_owner(tsdn, mutex);
+	mutex_prof_data_init(&mutex->prof_data);
+}
+
+static int
+mutex_addr_comp(const witness_t *witness1, void *mutex1,
+    const witness_t *witness2, void *mutex2) {
+	assert(mutex1 != NULL);
+	assert(mutex2 != NULL);
+	uintptr_t mu1int = (uintptr_t)mutex1;
+	uintptr_t mu2int = (uintptr_t)mutex2;
+	if (mu1int < mu2int) {
+		return -1;
+	} else if (mu1int == mu2int) {
+		return 0;
+	} else {
+		return 1;
+	}
+}
+
+bool
+malloc_mutex_init(malloc_mutex_t *mutex, const char *name,
+    witness_rank_t rank, malloc_mutex_lock_order_t lock_order) {
+	mutex_prof_data_init(&mutex->prof_data);
+#ifdef _WIN32
+#  if _WIN32_WINNT >= 0x0600
+	InitializeSRWLock(&mutex->lock);
+#  else
+	if (!InitializeCriticalSectionAndSpinCount(&mutex->lock,
+	    _CRT_SPINCOUNT)) {
+		return true;
+	}
+#  endif
+#elif (defined(JEMALLOC_OS_UNFAIR_LOCK))
+       mutex->lock = OS_UNFAIR_LOCK_INIT;
+#elif (defined(JEMALLOC_MUTEX_INIT_CB))
+	if (postpone_init) {
+		mutex->postponed_next = postponed_mutexes;
+		postponed_mutexes = mutex;
+	} else {
+		if (_pthread_mutex_init_calloc_cb(&mutex->lock,
+		    bootstrap_calloc) != 0) {
+			return true;
+		}
+	}
+#else
+	pthread_mutexattr_t attr;
+
+	if (pthread_mutexattr_init(&attr) != 0) {
+		return true;
+	}
+	pthread_mutexattr_settype(&attr, MALLOC_MUTEX_TYPE);
+	if (pthread_mutex_init(&mutex->lock, &attr) != 0) {
+		pthread_mutexattr_destroy(&attr);
+		return true;
+	}
+	pthread_mutexattr_destroy(&attr);
+#endif
+	if (config_debug) {
+		mutex->lock_order = lock_order;
+		if (lock_order == malloc_mutex_address_ordered) {
+			witness_init(&mutex->witness, name, rank,
+			    mutex_addr_comp, mutex);
+		} else {
+			witness_init(&mutex->witness, name, rank, NULL, NULL);
+		}
+	}
+	return false;
+}
+
+void
+malloc_mutex_prefork(tsdn_t *tsdn, malloc_mutex_t *mutex) {
+	malloc_mutex_lock(tsdn, mutex);
+}
+
+void
+malloc_mutex_postfork_parent(tsdn_t *tsdn, malloc_mutex_t *mutex) {
+	malloc_mutex_unlock(tsdn, mutex);
+}
+
+void
+malloc_mutex_postfork_child(tsdn_t *tsdn, malloc_mutex_t *mutex) {
+#ifdef JEMALLOC_MUTEX_INIT_CB
+	malloc_mutex_unlock(tsdn, mutex);
+#else
+	if (malloc_mutex_init(mutex, mutex->witness.name,
+	    mutex->witness.rank, mutex->lock_order)) {
+		malloc_printf("<jemalloc>: Error re-initializing mutex in "
+		    "child\n");
+		if (opt_abort) {
+			abort();
+		}
+	}
+#endif
+}
+
+bool
+malloc_mutex_boot(void) {
+#ifdef JEMALLOC_MUTEX_INIT_CB
+	postpone_init = false;
+	while (postponed_mutexes != NULL) {
+		if (_pthread_mutex_init_calloc_cb(&postponed_mutexes->lock,
+		    bootstrap_calloc) != 0) {
+			return true;
+		}
+		postponed_mutexes = postponed_mutexes->postponed_next;
+	}
+#endif
+	return false;
+}

deps/jemalloc/src/mutex_pool.c

@@ -0,0 +1,18 @@
+#define JEMALLOC_MUTEX_POOL_C_
+
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/mutex.h"
+#include "jemalloc/internal/mutex_pool.h"
+
+bool
+mutex_pool_init(mutex_pool_t *pool, const char *name, witness_rank_t rank) {
+	for (int i = 0; i < MUTEX_POOL_SIZE; ++i) {
+		if (malloc_mutex_init(&pool->mutexes[i], name, rank,
+		    malloc_mutex_address_ordered)) {
+			return true;
+		}
+	}
+	return false;
+}

deps/jemalloc/src/nstime.c

@@ -0,0 +1,170 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/nstime.h"
+
+#include "jemalloc/internal/assert.h"
+
+#define BILLION	UINT64_C(1000000000)
+#define MILLION	UINT64_C(1000000)
+
+void
+nstime_init(nstime_t *time, uint64_t ns) {
+	time->ns = ns;
+}
+
+void
+nstime_init2(nstime_t *time, uint64_t sec, uint64_t nsec) {
+	time->ns = sec * BILLION + nsec;
+}
+
+uint64_t
+nstime_ns(const nstime_t *time) {
+	return time->ns;
+}
+
+uint64_t
+nstime_msec(const nstime_t *time) {
+	return time->ns / MILLION;
+}
+
+uint64_t
+nstime_sec(const nstime_t *time) {
+	return time->ns / BILLION;
+}
+
+uint64_t
+nstime_nsec(const nstime_t *time) {
+	return time->ns % BILLION;
+}
+
+void
+nstime_copy(nstime_t *time, const nstime_t *source) {
+	*time = *source;
+}
+
+int
+nstime_compare(const nstime_t *a, const nstime_t *b) {
+	return (a->ns > b->ns) - (a->ns < b->ns);
+}
+
+void
+nstime_add(nstime_t *time, const nstime_t *addend) {
+	assert(UINT64_MAX - time->ns >= addend->ns);
+
+	time->ns += addend->ns;
+}
+
+void
+nstime_iadd(nstime_t *time, uint64_t addend) {
+	assert(UINT64_MAX - time->ns >= addend);
+
+	time->ns += addend;
+}
+
+void
+nstime_subtract(nstime_t *time, const nstime_t *subtrahend) {
+	assert(nstime_compare(time, subtrahend) >= 0);
+
+	time->ns -= subtrahend->ns;
+}
+
+void
+nstime_isubtract(nstime_t *time, uint64_t subtrahend) {
+	assert(time->ns >= subtrahend);
+
+	time->ns -= subtrahend;
+}
+
+void
+nstime_imultiply(nstime_t *time, uint64_t multiplier) {
+	assert((((time->ns | multiplier) & (UINT64_MAX << (sizeof(uint64_t) <<
+	    2))) == 0) || ((time->ns * multiplier) / multiplier == time->ns));
+
+	time->ns *= multiplier;
+}
+
+void
+nstime_idivide(nstime_t *time, uint64_t divisor) {
+	assert(divisor != 0);
+
+	time->ns /= divisor;
+}
+
+uint64_t
+nstime_divide(const nstime_t *time, const nstime_t *divisor) {
+	assert(divisor->ns != 0);
+
+	return time->ns / divisor->ns;
+}
+
+#ifdef _WIN32
+#  define NSTIME_MONOTONIC true
+static void
+nstime_get(nstime_t *time) {
+	FILETIME ft;
+	uint64_t ticks_100ns;
+
+	GetSystemTimeAsFileTime(&ft);
+	ticks_100ns = (((uint64_t)ft.dwHighDateTime) << 32) | ft.dwLowDateTime;
+
+	nstime_init(time, ticks_100ns * 100);
+}
+#elif defined(JEMALLOC_HAVE_CLOCK_MONOTONIC_COARSE)
+#  define NSTIME_MONOTONIC true
+static void
+nstime_get(nstime_t *time) {
+	struct timespec ts;
+
+	clock_gettime(CLOCK_MONOTONIC_COARSE, &ts);
+	nstime_init2(time, ts.tv_sec, ts.tv_nsec);
+}
+#elif defined(JEMALLOC_HAVE_CLOCK_MONOTONIC)
+#  define NSTIME_MONOTONIC true
+static void
+nstime_get(nstime_t *time) {
+	struct timespec ts;
+
+	clock_gettime(CLOCK_MONOTONIC, &ts);
+	nstime_init2(time, ts.tv_sec, ts.tv_nsec);
+}
+#elif defined(JEMALLOC_HAVE_MACH_ABSOLUTE_TIME)
+#  define NSTIME_MONOTONIC true
+static void
+nstime_get(nstime_t *time) {
+	nstime_init(time, mach_absolute_time());
+}
+#else
+#  define NSTIME_MONOTONIC false
+static void
+nstime_get(nstime_t *time) {
+	struct timeval tv;
+
+	gettimeofday(&tv, NULL);
+	nstime_init2(time, tv.tv_sec, tv.tv_usec * 1000);
+}
+#endif
+
+static bool
+nstime_monotonic_impl(void) {
+	return NSTIME_MONOTONIC;
+#undef NSTIME_MONOTONIC
+}
+nstime_monotonic_t *JET_MUTABLE nstime_monotonic = nstime_monotonic_impl;
+
+static bool
+nstime_update_impl(nstime_t *time) {
+	nstime_t old_time;
+
+	nstime_copy(&old_time, time);
+	nstime_get(time);
+
+	/* Handle non-monotonic clocks. */
+	if (unlikely(nstime_compare(&old_time, time) > 0)) {
+		nstime_copy(time, &old_time);
+		return true;
+	}
+
+	return false;
+}
+nstime_update_t *JET_MUTABLE nstime_update = nstime_update_impl;

deps/jemalloc/src/pages.c

@@ -0,0 +1,649 @@
+#define JEMALLOC_PAGES_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+
+#include "jemalloc/internal/pages.h"
+
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/malloc_io.h"
+
+#ifdef JEMALLOC_SYSCTL_VM_OVERCOMMIT
+#include <sys/sysctl.h>
+#ifdef __FreeBSD__
+#include <vm/vm_param.h>
+#endif
+#endif
+
+/******************************************************************************/
+/* Data. */
+
+/* Actual operating system page size, detected during bootstrap, <= PAGE. */
+static size_t	os_page;
+
+#ifndef _WIN32
+#  define PAGES_PROT_COMMIT (PROT_READ | PROT_WRITE)
+#  define PAGES_PROT_DECOMMIT (PROT_NONE)
+static int	mmap_flags;
+#endif
+static bool	os_overcommits;
+
+const char *thp_mode_names[] = {
+	"default",
+	"always",
+	"never",
+	"not supported"
+};
+thp_mode_t opt_thp = THP_MODE_DEFAULT;
+thp_mode_t init_system_thp_mode;
+
+/* Runtime support for lazy purge. Irrelevant when !pages_can_purge_lazy. */
+static bool pages_can_purge_lazy_runtime = true;
+
+/******************************************************************************/
+/*
+ * Function prototypes for static functions that are referenced prior to
+ * definition.
+ */
+
+static void os_pages_unmap(void *addr, size_t size);
+
+/******************************************************************************/
+
+static void *
+os_pages_map(void *addr, size_t size, size_t alignment, bool *commit) {
+	assert(ALIGNMENT_ADDR2BASE(addr, os_page) == addr);
+	assert(ALIGNMENT_CEILING(size, os_page) == size);
+	assert(size != 0);
+
+	if (os_overcommits) {
+		*commit = true;
+	}
+
+	void *ret;
+#ifdef _WIN32
+	/*
+	 * If VirtualAlloc can't allocate at the given address when one is
+	 * given, it fails and returns NULL.
+	 */
+	ret = VirtualAlloc(addr, size, MEM_RESERVE | (*commit ? MEM_COMMIT : 0),
+	    PAGE_READWRITE);
+#else
+	/*
+	 * We don't use MAP_FIXED here, because it can cause the *replacement*
+	 * of existing mappings, and we only want to create new mappings.
+	 */
+	{
+		int prot = *commit ? PAGES_PROT_COMMIT : PAGES_PROT_DECOMMIT;
+
+		ret = mmap(addr, size, prot, mmap_flags, -1, 0);
+	}
+	assert(ret != NULL);
+
+	if (ret == MAP_FAILED) {
+		ret = NULL;
+	} else if (addr != NULL && ret != addr) {
+		/*
+		 * We succeeded in mapping memory, but not in the right place.
+		 */
+		os_pages_unmap(ret, size);
+		ret = NULL;
+	}
+#endif
+	assert(ret == NULL || (addr == NULL && ret != addr) || (addr != NULL &&
+	    ret == addr));
+	return ret;
+}
+
+static void *
+os_pages_trim(void *addr, size_t alloc_size, size_t leadsize, size_t size,
+    bool *commit) {
+	void *ret = (void *)((uintptr_t)addr + leadsize);
+
+	assert(alloc_size >= leadsize + size);
+#ifdef _WIN32
+	os_pages_unmap(addr, alloc_size);
+	void *new_addr = os_pages_map(ret, size, PAGE, commit);
+	if (new_addr == ret) {
+		return ret;
+	}
+	if (new_addr != NULL) {
+		os_pages_unmap(new_addr, size);
+	}
+	return NULL;
+#else
+	size_t trailsize = alloc_size - leadsize - size;
+
+	if (leadsize != 0) {
+		os_pages_unmap(addr, leadsize);
+	}
+	if (trailsize != 0) {
+		os_pages_unmap((void *)((uintptr_t)ret + size), trailsize);
+	}
+	return ret;
+#endif
+}
+
+static void
+os_pages_unmap(void *addr, size_t size) {
+	assert(ALIGNMENT_ADDR2BASE(addr, os_page) == addr);
+	assert(ALIGNMENT_CEILING(size, os_page) == size);
+
+#ifdef _WIN32
+	if (VirtualFree(addr, 0, MEM_RELEASE) == 0)
+#else
+	if (munmap(addr, size) == -1)
+#endif
+	{
+		char buf[BUFERROR_BUF];
+
+		buferror(get_errno(), buf, sizeof(buf));
+		malloc_printf("<jemalloc>: Error in "
+#ifdef _WIN32
+		    "VirtualFree"
+#else
+		    "munmap"
+#endif
+		    "(): %s\n", buf);
+		if (opt_abort) {
+			abort();
+		}
+	}
+}
+
+static void *
+pages_map_slow(size_t size, size_t alignment, bool *commit) {
+	size_t alloc_size = size + alignment - os_page;
+	/* Beware size_t wrap-around. */
+	if (alloc_size < size) {
+		return NULL;
+	}
+
+	void *ret;
+	do {
+		void *pages = os_pages_map(NULL, alloc_size, alignment, commit);
+		if (pages == NULL) {
+			return NULL;
+		}
+		size_t leadsize = ALIGNMENT_CEILING((uintptr_t)pages, alignment)
+		    - (uintptr_t)pages;
+		ret = os_pages_trim(pages, alloc_size, leadsize, size, commit);
+	} while (ret == NULL);
+
+	assert(ret != NULL);
+	assert(PAGE_ADDR2BASE(ret) == ret);
+	return ret;
+}
+
+void *
+pages_map(void *addr, size_t size, size_t alignment, bool *commit) {
+	assert(alignment >= PAGE);
+	assert(ALIGNMENT_ADDR2BASE(addr, alignment) == addr);
+
+#if defined(__FreeBSD__) && defined(MAP_EXCL)
+	/*
+	 * FreeBSD has mechanisms both to mmap at specific address without
+	 * touching existing mappings, and to mmap with specific alignment.
+	 */
+	{
+		if (os_overcommits) {
+			*commit = true;
+		}
+
+		int prot = *commit ? PAGES_PROT_COMMIT : PAGES_PROT_DECOMMIT;
+		int flags = mmap_flags;
+
+		if (addr != NULL) {
+			flags |= MAP_FIXED | MAP_EXCL;
+		} else {
+			unsigned alignment_bits = ffs_zu(alignment);
+			assert(alignment_bits > 1);
+			flags |= MAP_ALIGNED(alignment_bits - 1);
+		}
+
+		void *ret = mmap(addr, size, prot, flags, -1, 0);
+		if (ret == MAP_FAILED) {
+			ret = NULL;
+		}
+
+		return ret;
+	}
+#endif
+	/*
+	 * Ideally, there would be a way to specify alignment to mmap() (like
+	 * NetBSD has), but in the absence of such a feature, we have to work
+	 * hard to efficiently create aligned mappings.  The reliable, but
+	 * slow method is to create a mapping that is over-sized, then trim the
+	 * excess.  However, that always results in one or two calls to
+	 * os_pages_unmap(), and it can leave holes in the process's virtual
+	 * memory map if memory grows downward.
+	 *
+	 * Optimistically try mapping precisely the right amount before falling
+	 * back to the slow method, with the expectation that the optimistic
+	 * approach works most of the time.
+	 */
+
+	void *ret = os_pages_map(addr, size, os_page, commit);
+	if (ret == NULL || ret == addr) {
+		return ret;
+	}
+	assert(addr == NULL);
+	if (ALIGNMENT_ADDR2OFFSET(ret, alignment) != 0) {
+		os_pages_unmap(ret, size);
+		return pages_map_slow(size, alignment, commit);
+	}
+
+	assert(PAGE_ADDR2BASE(ret) == ret);
+	return ret;
+}
+
+void
+pages_unmap(void *addr, size_t size) {
+	assert(PAGE_ADDR2BASE(addr) == addr);
+	assert(PAGE_CEILING(size) == size);
+
+	os_pages_unmap(addr, size);
+}
+
+static bool
+pages_commit_impl(void *addr, size_t size, bool commit) {
+	assert(PAGE_ADDR2BASE(addr) == addr);
+	assert(PAGE_CEILING(size) == size);
+
+	if (os_overcommits) {
+		return true;
+	}
+
+#ifdef _WIN32
+	return (commit ? (addr != VirtualAlloc(addr, size, MEM_COMMIT,
+	    PAGE_READWRITE)) : (!VirtualFree(addr, size, MEM_DECOMMIT)));
+#else
+	{
+		int prot = commit ? PAGES_PROT_COMMIT : PAGES_PROT_DECOMMIT;
+		void *result = mmap(addr, size, prot, mmap_flags | MAP_FIXED,
+		    -1, 0);
+		if (result == MAP_FAILED) {
+			return true;
+		}
+		if (result != addr) {
+			/*
+			 * We succeeded in mapping memory, but not in the right
+			 * place.
+			 */
+			os_pages_unmap(result, size);
+			return true;
+		}
+		return false;
+	}
+#endif
+}
+
+bool
+pages_commit(void *addr, size_t size) {
+	return pages_commit_impl(addr, size, true);
+}
+
+bool
+pages_decommit(void *addr, size_t size) {
+	return pages_commit_impl(addr, size, false);
+}
+
+bool
+pages_purge_lazy(void *addr, size_t size) {
+	assert(ALIGNMENT_ADDR2BASE(addr, os_page) == addr);
+	assert(PAGE_CEILING(size) == size);
+
+	if (!pages_can_purge_lazy) {
+		return true;
+	}
+	if (!pages_can_purge_lazy_runtime) {
+		/*
+		 * Built with lazy purge enabled, but detected it was not
+		 * supported on the current system.
+		 */
+		return true;
+	}
+
+#ifdef _WIN32
+	VirtualAlloc(addr, size, MEM_RESET, PAGE_READWRITE);
+	return false;
+#elif defined(JEMALLOC_PURGE_MADVISE_FREE)
+	return (madvise(addr, size,
+#  ifdef MADV_FREE
+	    MADV_FREE
+#  else
+	    JEMALLOC_MADV_FREE
+#  endif
+	    ) != 0);
+#elif defined(JEMALLOC_PURGE_MADVISE_DONTNEED) && \
+    !defined(JEMALLOC_PURGE_MADVISE_DONTNEED_ZEROS)
+	return (madvise(addr, size, MADV_DONTNEED) != 0);
+#else
+	not_reached();
+#endif
+}
+
+bool
+pages_purge_forced(void *addr, size_t size) {
+	assert(PAGE_ADDR2BASE(addr) == addr);
+	assert(PAGE_CEILING(size) == size);
+
+	if (!pages_can_purge_forced) {
+		return true;
+	}
+
+#if defined(JEMALLOC_PURGE_MADVISE_DONTNEED) && \
+    defined(JEMALLOC_PURGE_MADVISE_DONTNEED_ZEROS)
+	return (madvise(addr, size, MADV_DONTNEED) != 0);
+#elif defined(JEMALLOC_MAPS_COALESCE)
+	/* Try to overlay a new demand-zeroed mapping. */
+	return pages_commit(addr, size);
+#else
+	not_reached();
+#endif
+}
+
+static bool
+pages_huge_impl(void *addr, size_t size, bool aligned) {
+	if (aligned) {
+		assert(HUGEPAGE_ADDR2BASE(addr) == addr);
+		assert(HUGEPAGE_CEILING(size) == size);
+	}
+#ifdef JEMALLOC_HAVE_MADVISE_HUGE
+	return (madvise(addr, size, MADV_HUGEPAGE) != 0);
+#else
+	return true;
+#endif
+}
+
+bool
+pages_huge(void *addr, size_t size) {
+	return pages_huge_impl(addr, size, true);
+}
+
+static bool
+pages_huge_unaligned(void *addr, size_t size) {
+	return pages_huge_impl(addr, size, false);
+}
+
+static bool
+pages_nohuge_impl(void *addr, size_t size, bool aligned) {
+	if (aligned) {
+		assert(HUGEPAGE_ADDR2BASE(addr) == addr);
+		assert(HUGEPAGE_CEILING(size) == size);
+	}
+
+#ifdef JEMALLOC_HAVE_MADVISE_HUGE
+	return (madvise(addr, size, MADV_NOHUGEPAGE) != 0);
+#else
+	return false;
+#endif
+}
+
+bool
+pages_nohuge(void *addr, size_t size) {
+	return pages_nohuge_impl(addr, size, true);
+}
+
+static bool
+pages_nohuge_unaligned(void *addr, size_t size) {
+	return pages_nohuge_impl(addr, size, false);
+}
+
+bool
+pages_dontdump(void *addr, size_t size) {
+	assert(PAGE_ADDR2BASE(addr) == addr);
+	assert(PAGE_CEILING(size) == size);
+#ifdef JEMALLOC_MADVISE_DONTDUMP
+	return madvise(addr, size, MADV_DONTDUMP) != 0;
+#else
+	return false;
+#endif
+}
+
+bool
+pages_dodump(void *addr, size_t size) {
+	assert(PAGE_ADDR2BASE(addr) == addr);
+	assert(PAGE_CEILING(size) == size);
+#ifdef JEMALLOC_MADVISE_DONTDUMP
+	return madvise(addr, size, MADV_DODUMP) != 0;
+#else
+	return false;
+#endif
+}
+
+
+static size_t
+os_page_detect(void) {
+#ifdef _WIN32
+	SYSTEM_INFO si;
+	GetSystemInfo(&si);
+	return si.dwPageSize;
+#elif defined(__FreeBSD__)
+	/*
+	 * This returns the value obtained from
+	 * the auxv vector, avoiding a syscall.
+	 */
+	return getpagesize();
+#else
+	long result = sysconf(_SC_PAGESIZE);
+	if (result == -1) {
+		return LG_PAGE;
+	}
+	return (size_t)result;
+#endif
+}
+
+#ifdef JEMALLOC_SYSCTL_VM_OVERCOMMIT
+static bool
+os_overcommits_sysctl(void) {
+	int vm_overcommit;
+	size_t sz;
+
+	sz = sizeof(vm_overcommit);
+#if defined(__FreeBSD__) && defined(VM_OVERCOMMIT)
+	int mib[2];
+
+	mib[0] = CTL_VM;
+	mib[1] = VM_OVERCOMMIT;
+	if (sysctl(mib, 2, &vm_overcommit, &sz, NULL, 0) != 0) {
+		return false; /* Error. */
+	}
+#else
+	if (sysctlbyname("vm.overcommit", &vm_overcommit, &sz, NULL, 0) != 0) {
+		return false; /* Error. */
+	}
+#endif
+
+	return ((vm_overcommit & 0x3) == 0);
+}
+#endif
+
+#ifdef JEMALLOC_PROC_SYS_VM_OVERCOMMIT_MEMORY
+/*
+ * Use syscall(2) rather than {open,read,close}(2) when possible to avoid
+ * reentry during bootstrapping if another library has interposed system call
+ * wrappers.
+ */
+static bool
+os_overcommits_proc(void) {
+	int fd;
+	char buf[1];
+
+#if defined(JEMALLOC_USE_SYSCALL) && defined(SYS_open)
+	#if defined(O_CLOEXEC)
+		fd = (int)syscall(SYS_open, "/proc/sys/vm/overcommit_memory", O_RDONLY |
+			O_CLOEXEC);
+	#else
+		fd = (int)syscall(SYS_open, "/proc/sys/vm/overcommit_memory", O_RDONLY);
+		if (fd != -1) {
+			fcntl(fd, F_SETFD, fcntl(fd, F_GETFD) | FD_CLOEXEC);
+		}
+	#endif
+#elif defined(JEMALLOC_USE_SYSCALL) && defined(SYS_openat)
+	#if defined(O_CLOEXEC)
+		fd = (int)syscall(SYS_openat,
+			AT_FDCWD, "/proc/sys/vm/overcommit_memory", O_RDONLY | O_CLOEXEC);
+	#else
+		fd = (int)syscall(SYS_openat,
+			AT_FDCWD, "/proc/sys/vm/overcommit_memory", O_RDONLY);
+		if (fd != -1) {
+			fcntl(fd, F_SETFD, fcntl(fd, F_GETFD) | FD_CLOEXEC);
+		}
+	#endif
+#else
+	#if defined(O_CLOEXEC)
+		fd = open("/proc/sys/vm/overcommit_memory", O_RDONLY | O_CLOEXEC);
+	#else
+		fd = open("/proc/sys/vm/overcommit_memory", O_RDONLY);
+		if (fd != -1) {
+			fcntl(fd, F_SETFD, fcntl(fd, F_GETFD) | FD_CLOEXEC);
+		}
+	#endif
+#endif
+
+	if (fd == -1) {
+		return false; /* Error. */
+	}
+
+	ssize_t nread = malloc_read_fd(fd, &buf, sizeof(buf));
+#if defined(JEMALLOC_USE_SYSCALL) && defined(SYS_close)
+	syscall(SYS_close, fd);
+#else
+	close(fd);
+#endif
+
+	if (nread < 1) {
+		return false; /* Error. */
+	}
+	/*
+	 * /proc/sys/vm/overcommit_memory meanings:
+	 * 0: Heuristic overcommit.
+	 * 1: Always overcommit.
+	 * 2: Never overcommit.
+	 */
+	return (buf[0] == '0' || buf[0] == '1');
+}
+#endif
+
+void
+pages_set_thp_state (void *ptr, size_t size) {
+	if (opt_thp == thp_mode_default || opt_thp == init_system_thp_mode) {
+		return;
+	}
+	assert(opt_thp != thp_mode_not_supported &&
+	    init_system_thp_mode != thp_mode_not_supported);
+
+	if (opt_thp == thp_mode_always
+	    && init_system_thp_mode != thp_mode_never) {
+		assert(init_system_thp_mode == thp_mode_default);
+		pages_huge_unaligned(ptr, size);
+	} else if (opt_thp == thp_mode_never) {
+		assert(init_system_thp_mode == thp_mode_default ||
+		    init_system_thp_mode == thp_mode_always);
+		pages_nohuge_unaligned(ptr, size);
+	}
+}
+
+static void
+init_thp_state(void) {
+	if (!have_madvise_huge) {
+		if (metadata_thp_enabled() && opt_abort) {
+			malloc_write("<jemalloc>: no MADV_HUGEPAGE support\n");
+			abort();
+		}
+		goto label_error;
+	}
+
+	static const char sys_state_madvise[] = "always [madvise] never\n";
+	static const char sys_state_always[] = "[always] madvise never\n";
+	static const char sys_state_never[] = "always madvise [never]\n";
+	char buf[sizeof(sys_state_madvise)];
+
+#if defined(JEMALLOC_USE_SYSCALL) && defined(SYS_open)
+	int fd = (int)syscall(SYS_open,
+	    "/sys/kernel/mm/transparent_hugepage/enabled", O_RDONLY);
+#else
+	int fd = open("/sys/kernel/mm/transparent_hugepage/enabled", O_RDONLY);
+#endif
+	if (fd == -1) {
+		goto label_error;
+	}
+
+	ssize_t nread = malloc_read_fd(fd, &buf, sizeof(buf));
+#if defined(JEMALLOC_USE_SYSCALL) && defined(SYS_close)
+	syscall(SYS_close, fd);
+#else
+	close(fd);
+#endif
+
+        if (nread < 0) {
+		goto label_error; 
+        }
+
+	if (strncmp(buf, sys_state_madvise, (size_t)nread) == 0) {
+		init_system_thp_mode = thp_mode_default;
+	} else if (strncmp(buf, sys_state_always, (size_t)nread) == 0) {
+		init_system_thp_mode = thp_mode_always;
+	} else if (strncmp(buf, sys_state_never, (size_t)nread) == 0) {
+		init_system_thp_mode = thp_mode_never;
+	} else {
+		goto label_error;
+	}
+	return;
+label_error:
+	opt_thp = init_system_thp_mode = thp_mode_not_supported;
+}
+
+bool
+pages_boot(void) {
+	os_page = os_page_detect();
+	if (os_page > PAGE) {
+		malloc_write("<jemalloc>: Unsupported system page size\n");
+		if (opt_abort) {
+			abort();
+		}
+		return true;
+	}
+
+#ifndef _WIN32
+	mmap_flags = MAP_PRIVATE | MAP_ANON;
+#endif
+
+#ifdef JEMALLOC_SYSCTL_VM_OVERCOMMIT
+	os_overcommits = os_overcommits_sysctl();
+#elif defined(JEMALLOC_PROC_SYS_VM_OVERCOMMIT_MEMORY)
+	os_overcommits = os_overcommits_proc();
+#  ifdef MAP_NORESERVE
+	if (os_overcommits) {
+		mmap_flags |= MAP_NORESERVE;
+	}
+#  endif
+#else
+	os_overcommits = false;
+#endif
+
+	init_thp_state();
+
+#ifdef __FreeBSD__
+	/*
+	 * FreeBSD doesn't need the check; madvise(2) is known to work.
+	 */
+#else
+	/* Detect lazy purge runtime support. */
+	if (pages_can_purge_lazy) {
+		bool committed = false;
+		void *madv_free_page = os_pages_map(NULL, PAGE, PAGE, &committed);
+		if (madv_free_page == NULL) {
+			return true;
+		}
+		assert(pages_can_purge_lazy_runtime);
+		if (pages_purge_lazy(madv_free_page, PAGE)) {
+			pages_can_purge_lazy_runtime = false;
+		}
+		os_pages_unmap(madv_free_page, PAGE);
+	}
+#endif
+
+	return false;
+}

deps/jemalloc/src/prng.c

@@ -0,0 +1,3 @@
+#define JEMALLOC_PRNG_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"

deps/jemalloc/src/quarantine.c

@@ -0,0 +1,199 @@
+#define	JEMALLOC_QUARANTINE_C_
+#include "jemalloc/internal/jemalloc_internal.h"
+
+/*
+ * quarantine pointers close to NULL are used to encode state information that
+ * is used for cleaning up during thread shutdown.
+ */
+#define	QUARANTINE_STATE_REINCARNATED	((quarantine_t *)(uintptr_t)1)
+#define	QUARANTINE_STATE_PURGATORY	((quarantine_t *)(uintptr_t)2)
+#define	QUARANTINE_STATE_MAX		QUARANTINE_STATE_PURGATORY
+
+/******************************************************************************/
+/* Data. */
+
+malloc_tsd_data(, quarantine, quarantine_t *, NULL)
+
+/******************************************************************************/
+/* Function prototypes for non-inline static functions. */
+
+static quarantine_t	*quarantine_grow(quarantine_t *quarantine);
+static void	quarantine_drain_one(quarantine_t *quarantine);
+static void	quarantine_drain(quarantine_t *quarantine, size_t upper_bound);
+
+/******************************************************************************/
+
+quarantine_t *
+quarantine_init(size_t lg_maxobjs)
+{
+	quarantine_t *quarantine;
+
+	quarantine = (quarantine_t *)imalloc(offsetof(quarantine_t, objs) +
+	    ((ZU(1) << lg_maxobjs) * sizeof(quarantine_obj_t)));
+	if (quarantine == NULL)
+		return (NULL);
+	quarantine->curbytes = 0;
+	quarantine->curobjs = 0;
+	quarantine->first = 0;
+	quarantine->lg_maxobjs = lg_maxobjs;
+
+	quarantine_tsd_set(&quarantine);
+
+	return (quarantine);
+}
+
+static quarantine_t *
+quarantine_grow(quarantine_t *quarantine)
+{
+	quarantine_t *ret;
+
+	ret = quarantine_init(quarantine->lg_maxobjs + 1);
+	if (ret == NULL) {
+		quarantine_drain_one(quarantine);
+		return (quarantine);
+	}
+
+	ret->curbytes = quarantine->curbytes;
+	ret->curobjs = quarantine->curobjs;
+	if (quarantine->first + quarantine->curobjs <= (ZU(1) <<
+	    quarantine->lg_maxobjs)) {
+		/* objs ring buffer data are contiguous. */
+		memcpy(ret->objs, &quarantine->objs[quarantine->first],
+		    quarantine->curobjs * sizeof(quarantine_obj_t));
+	} else {
+		/* objs ring buffer data wrap around. */
+		size_t ncopy_a = (ZU(1) << quarantine->lg_maxobjs) -
+		    quarantine->first;
+		size_t ncopy_b = quarantine->curobjs - ncopy_a;
+
+		memcpy(ret->objs, &quarantine->objs[quarantine->first], ncopy_a
+		    * sizeof(quarantine_obj_t));
+		memcpy(&ret->objs[ncopy_a], quarantine->objs, ncopy_b *
+		    sizeof(quarantine_obj_t));
+	}
+	idalloc(quarantine);
+
+	return (ret);
+}
+
+static void
+quarantine_drain_one(quarantine_t *quarantine)
+{
+	quarantine_obj_t *obj = &quarantine->objs[quarantine->first];
+	assert(obj->usize == isalloc(obj->ptr, config_prof));
+	idalloc(obj->ptr);
+	quarantine->curbytes -= obj->usize;
+	quarantine->curobjs--;
+	quarantine->first = (quarantine->first + 1) & ((ZU(1) <<
+	    quarantine->lg_maxobjs) - 1);
+}
+
+static void
+quarantine_drain(quarantine_t *quarantine, size_t upper_bound)
+{
+
+	while (quarantine->curbytes > upper_bound && quarantine->curobjs > 0)
+		quarantine_drain_one(quarantine);
+}
+
+void
+quarantine(void *ptr)
+{
+	quarantine_t *quarantine;
+	size_t usize = isalloc(ptr, config_prof);
+
+	cassert(config_fill);
+	assert(opt_quarantine);
+
+	quarantine = *quarantine_tsd_get();
+	if ((uintptr_t)quarantine <= (uintptr_t)QUARANTINE_STATE_MAX) {
+		if (quarantine == QUARANTINE_STATE_PURGATORY) {
+			/*
+			 * Make a note that quarantine() was called after
+			 * quarantine_cleanup() was called.
+			 */
+			quarantine = QUARANTINE_STATE_REINCARNATED;
+			quarantine_tsd_set(&quarantine);
+		}
+		idalloc(ptr);
+		return;
+	}
+	/*
+	 * Drain one or more objects if the quarantine size limit would be
+	 * exceeded by appending ptr.
+	 */
+	if (quarantine->curbytes + usize > opt_quarantine) {
+		size_t upper_bound = (opt_quarantine >= usize) ? opt_quarantine
+		    - usize : 0;
+		quarantine_drain(quarantine, upper_bound);
+	}
+	/* Grow the quarantine ring buffer if it's full. */
+	if (quarantine->curobjs == (ZU(1) << quarantine->lg_maxobjs))
+		quarantine = quarantine_grow(quarantine);
+	/* quarantine_grow() must free a slot if it fails to grow. */
+	assert(quarantine->curobjs < (ZU(1) << quarantine->lg_maxobjs));
+	/* Append ptr if its size doesn't exceed the quarantine size. */
+	if (quarantine->curbytes + usize <= opt_quarantine) {
+		size_t offset = (quarantine->first + quarantine->curobjs) &
+		    ((ZU(1) << quarantine->lg_maxobjs) - 1);
+		quarantine_obj_t *obj = &quarantine->objs[offset];
+		obj->ptr = ptr;
+		obj->usize = usize;
+		quarantine->curbytes += usize;
+		quarantine->curobjs++;
+		if (config_fill && opt_junk) {
+			/*
+			 * Only do redzone validation if Valgrind isn't in
+			 * operation.
+			 */
+			if ((config_valgrind == false || opt_valgrind == false)
+			    && usize <= SMALL_MAXCLASS)
+				arena_quarantine_junk_small(ptr, usize);
+			else
+				memset(ptr, 0x5a, usize);
+		}
+	} else {
+		assert(quarantine->curbytes == 0);
+		idalloc(ptr);
+	}
+}
+
+void
+quarantine_cleanup(void *arg)
+{
+	quarantine_t *quarantine = *(quarantine_t **)arg;
+
+	if (quarantine == QUARANTINE_STATE_REINCARNATED) {
+		/*
+		 * Another destructor deallocated memory after this destructor
+		 * was called.  Reset quarantine to QUARANTINE_STATE_PURGATORY
+		 * in order to receive another callback.
+		 */
+		quarantine = QUARANTINE_STATE_PURGATORY;
+		quarantine_tsd_set(&quarantine);
+	} else if (quarantine == QUARANTINE_STATE_PURGATORY) {
+		/*
+		 * The previous time this destructor was called, we set the key
+		 * to QUARANTINE_STATE_PURGATORY so that other destructors
+		 * wouldn't cause re-creation of the quarantine.  This time, do
+		 * nothing, so that the destructor will not be called again.
+		 */
+	} else if (quarantine != NULL) {
+		quarantine_drain(quarantine, 0);
+		idalloc(quarantine);
+		quarantine = QUARANTINE_STATE_PURGATORY;
+		quarantine_tsd_set(&quarantine);
+	}
+}
+
+bool
+quarantine_boot(void)
+{
+
+	cassert(config_fill);
+
+	if (quarantine_tsd_boot())
+		return (true);
+
+	return (false);
+}

deps/jemalloc/src/rtree.c

@@ -0,0 +1,320 @@
+#define JEMALLOC_RTREE_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/mutex.h"
+
+/*
+ * Only the most significant bits of keys passed to rtree_{read,write}() are
+ * used.
+ */
+bool
+rtree_new(rtree_t *rtree, bool zeroed) {
+#ifdef JEMALLOC_JET
+	if (!zeroed) {
+		memset(rtree, 0, sizeof(rtree_t)); /* Clear root. */
+	}
+#else
+	assert(zeroed);
+#endif
+
+	if (malloc_mutex_init(&rtree->init_lock, "rtree", WITNESS_RANK_RTREE,
+	    malloc_mutex_rank_exclusive)) {
+		return true;
+	}
+
+	return false;
+}
+
+static rtree_node_elm_t *
+rtree_node_alloc_impl(tsdn_t *tsdn, rtree_t *rtree, size_t nelms) {
+	return (rtree_node_elm_t *)base_alloc(tsdn, b0get(), nelms *
+	    sizeof(rtree_node_elm_t), CACHELINE);
+}
+rtree_node_alloc_t *JET_MUTABLE rtree_node_alloc = rtree_node_alloc_impl;
+
+static void
+rtree_node_dalloc_impl(tsdn_t *tsdn, rtree_t *rtree, rtree_node_elm_t *node) {
+	/* Nodes are never deleted during normal operation. */
+	not_reached();
+}
+rtree_node_dalloc_t *JET_MUTABLE rtree_node_dalloc =
+    rtree_node_dalloc_impl;
+
+static rtree_leaf_elm_t *
+rtree_leaf_alloc_impl(tsdn_t *tsdn, rtree_t *rtree, size_t nelms) {
+	return (rtree_leaf_elm_t *)base_alloc(tsdn, b0get(), nelms *
+	    sizeof(rtree_leaf_elm_t), CACHELINE);
+}
+rtree_leaf_alloc_t *JET_MUTABLE rtree_leaf_alloc = rtree_leaf_alloc_impl;
+
+static void
+rtree_leaf_dalloc_impl(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *leaf) {
+	/* Leaves are never deleted during normal operation. */
+	not_reached();
+}
+rtree_leaf_dalloc_t *JET_MUTABLE rtree_leaf_dalloc =
+    rtree_leaf_dalloc_impl;
+
+#ifdef JEMALLOC_JET
+#  if RTREE_HEIGHT > 1
+static void
+rtree_delete_subtree(tsdn_t *tsdn, rtree_t *rtree, rtree_node_elm_t *subtree,
+    unsigned level) {
+	size_t nchildren = ZU(1) << rtree_levels[level].bits;
+	if (level + 2 < RTREE_HEIGHT) {
+		for (size_t i = 0; i < nchildren; i++) {
+			rtree_node_elm_t *node =
+			    (rtree_node_elm_t *)atomic_load_p(&subtree[i].child,
+			    ATOMIC_RELAXED);
+			if (node != NULL) {
+				rtree_delete_subtree(tsdn, rtree, node, level +
+				    1);
+			}
+		}
+	} else {
+		for (size_t i = 0; i < nchildren; i++) {
+			rtree_leaf_elm_t *leaf =
+			    (rtree_leaf_elm_t *)atomic_load_p(&subtree[i].child,
+			    ATOMIC_RELAXED);
+			if (leaf != NULL) {
+				rtree_leaf_dalloc(tsdn, rtree, leaf);
+			}
+		}
+	}
+
+	if (subtree != rtree->root) {
+		rtree_node_dalloc(tsdn, rtree, subtree);
+	}
+}
+#  endif
+
+void
+rtree_delete(tsdn_t *tsdn, rtree_t *rtree) {
+#  if RTREE_HEIGHT > 1
+	rtree_delete_subtree(tsdn, rtree, rtree->root, 0);
+#  endif
+}
+#endif
+
+static rtree_node_elm_t *
+rtree_node_init(tsdn_t *tsdn, rtree_t *rtree, unsigned level,
+    atomic_p_t *elmp) {
+	malloc_mutex_lock(tsdn, &rtree->init_lock);
+	/*
+	 * If *elmp is non-null, then it was initialized with the init lock
+	 * held, so we can get by with 'relaxed' here.
+	 */
+	rtree_node_elm_t *node = atomic_load_p(elmp, ATOMIC_RELAXED);
+	if (node == NULL) {
+		node = rtree_node_alloc(tsdn, rtree, ZU(1) <<
+		    rtree_levels[level].bits);
+		if (node == NULL) {
+			malloc_mutex_unlock(tsdn, &rtree->init_lock);
+			return NULL;
+		}
+		/*
+		 * Even though we hold the lock, a later reader might not; we
+		 * need release semantics.
+		 */
+		atomic_store_p(elmp, node, ATOMIC_RELEASE);
+	}
+	malloc_mutex_unlock(tsdn, &rtree->init_lock);
+
+	return node;
+}
+
+static rtree_leaf_elm_t *
+rtree_leaf_init(tsdn_t *tsdn, rtree_t *rtree, atomic_p_t *elmp) {
+	malloc_mutex_lock(tsdn, &rtree->init_lock);
+	/*
+	 * If *elmp is non-null, then it was initialized with the init lock
+	 * held, so we can get by with 'relaxed' here.
+	 */
+	rtree_leaf_elm_t *leaf = atomic_load_p(elmp, ATOMIC_RELAXED);
+	if (leaf == NULL) {
+		leaf = rtree_leaf_alloc(tsdn, rtree, ZU(1) <<
+		    rtree_levels[RTREE_HEIGHT-1].bits);
+		if (leaf == NULL) {
+			malloc_mutex_unlock(tsdn, &rtree->init_lock);
+			return NULL;
+		}
+		/*
+		 * Even though we hold the lock, a later reader might not; we
+		 * need release semantics.
+		 */
+		atomic_store_p(elmp, leaf, ATOMIC_RELEASE);
+	}
+	malloc_mutex_unlock(tsdn, &rtree->init_lock);
+
+	return leaf;
+}
+
+static bool
+rtree_node_valid(rtree_node_elm_t *node) {
+	return ((uintptr_t)node != (uintptr_t)0);
+}
+
+static bool
+rtree_leaf_valid(rtree_leaf_elm_t *leaf) {
+	return ((uintptr_t)leaf != (uintptr_t)0);
+}
+
+static rtree_node_elm_t *
+rtree_child_node_tryread(rtree_node_elm_t *elm, bool dependent) {
+	rtree_node_elm_t *node;
+
+	if (dependent) {
+		node = (rtree_node_elm_t *)atomic_load_p(&elm->child,
+		    ATOMIC_RELAXED);
+	} else {
+		node = (rtree_node_elm_t *)atomic_load_p(&elm->child,
+		    ATOMIC_ACQUIRE);
+	}
+
+	assert(!dependent || node != NULL);
+	return node;
+}
+
+static rtree_node_elm_t *
+rtree_child_node_read(tsdn_t *tsdn, rtree_t *rtree, rtree_node_elm_t *elm,
+    unsigned level, bool dependent) {
+	rtree_node_elm_t *node;
+
+	node = rtree_child_node_tryread(elm, dependent);
+	if (!dependent && unlikely(!rtree_node_valid(node))) {
+		node = rtree_node_init(tsdn, rtree, level + 1, &elm->child);
+	}
+	assert(!dependent || node != NULL);
+	return node;
+}
+
+static rtree_leaf_elm_t *
+rtree_child_leaf_tryread(rtree_node_elm_t *elm, bool dependent) {
+	rtree_leaf_elm_t *leaf;
+
+	if (dependent) {
+		leaf = (rtree_leaf_elm_t *)atomic_load_p(&elm->child,
+		    ATOMIC_RELAXED);
+	} else {
+		leaf = (rtree_leaf_elm_t *)atomic_load_p(&elm->child,
+		    ATOMIC_ACQUIRE);
+	}
+
+	assert(!dependent || leaf != NULL);
+	return leaf;
+}
+
+static rtree_leaf_elm_t *
+rtree_child_leaf_read(tsdn_t *tsdn, rtree_t *rtree, rtree_node_elm_t *elm,
+    unsigned level, bool dependent) {
+	rtree_leaf_elm_t *leaf;
+
+	leaf = rtree_child_leaf_tryread(elm, dependent);
+	if (!dependent && unlikely(!rtree_leaf_valid(leaf))) {
+		leaf = rtree_leaf_init(tsdn, rtree, &elm->child);
+	}
+	assert(!dependent || leaf != NULL);
+	return leaf;
+}
+
+rtree_leaf_elm_t *
+rtree_leaf_elm_lookup_hard(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
+    uintptr_t key, bool dependent, bool init_missing) {
+	rtree_node_elm_t *node;
+	rtree_leaf_elm_t *leaf;
+#if RTREE_HEIGHT > 1
+	node = rtree->root;
+#else
+	leaf = rtree->root;
+#endif
+
+	if (config_debug) {
+		uintptr_t leafkey = rtree_leafkey(key);
+		for (unsigned i = 0; i < RTREE_CTX_NCACHE; i++) {
+			assert(rtree_ctx->cache[i].leafkey != leafkey);
+		}
+		for (unsigned i = 0; i < RTREE_CTX_NCACHE_L2; i++) {
+			assert(rtree_ctx->l2_cache[i].leafkey != leafkey);
+		}
+	}
+
+#define RTREE_GET_CHILD(level) {					\
+		assert(level < RTREE_HEIGHT-1);				\
+		if (level != 0 && !dependent &&				\
+		    unlikely(!rtree_node_valid(node))) {		\
+			return NULL;					\
+		}							\
+		uintptr_t subkey = rtree_subkey(key, level);		\
+		if (level + 2 < RTREE_HEIGHT) {				\
+			node = init_missing ?				\
+			    rtree_child_node_read(tsdn, rtree,		\
+			    &node[subkey], level, dependent) :		\
+			    rtree_child_node_tryread(&node[subkey],	\
+			    dependent);					\
+		} else {						\
+			leaf = init_missing ?				\
+			    rtree_child_leaf_read(tsdn, rtree,		\
+			    &node[subkey], level, dependent) :		\
+			    rtree_child_leaf_tryread(&node[subkey],	\
+			    dependent);					\
+		}							\
+	}
+	/*
+	 * Cache replacement upon hard lookup (i.e. L1 & L2 rtree cache miss):
+	 * (1) evict last entry in L2 cache; (2) move the collision slot from L1
+	 * cache down to L2; and 3) fill L1.
+	 */
+#define RTREE_GET_LEAF(level) {						\
+		assert(level == RTREE_HEIGHT-1);			\
+		if (!dependent && unlikely(!rtree_leaf_valid(leaf))) {	\
+			return NULL;					\
+		}							\
+		if (RTREE_CTX_NCACHE_L2 > 1) {				\
+			memmove(&rtree_ctx->l2_cache[1],		\
+			    &rtree_ctx->l2_cache[0],			\
+			    sizeof(rtree_ctx_cache_elm_t) *		\
+			    (RTREE_CTX_NCACHE_L2 - 1));			\
+		}							\
+		size_t slot = rtree_cache_direct_map(key);		\
+		rtree_ctx->l2_cache[0].leafkey =			\
+		    rtree_ctx->cache[slot].leafkey;			\
+		rtree_ctx->l2_cache[0].leaf =				\
+		    rtree_ctx->cache[slot].leaf;			\
+		uintptr_t leafkey = rtree_leafkey(key);			\
+		rtree_ctx->cache[slot].leafkey = leafkey;		\
+		rtree_ctx->cache[slot].leaf = leaf;			\
+		uintptr_t subkey = rtree_subkey(key, level);		\
+		return &leaf[subkey];					\
+	}
+	if (RTREE_HEIGHT > 1) {
+		RTREE_GET_CHILD(0)
+	}
+	if (RTREE_HEIGHT > 2) {
+		RTREE_GET_CHILD(1)
+	}
+	if (RTREE_HEIGHT > 3) {
+		for (unsigned i = 2; i < RTREE_HEIGHT-1; i++) {
+			RTREE_GET_CHILD(i)
+		}
+	}
+	RTREE_GET_LEAF(RTREE_HEIGHT-1)
+#undef RTREE_GET_CHILD
+#undef RTREE_GET_LEAF
+	not_reached();
+}
+
+void
+rtree_ctx_data_init(rtree_ctx_t *ctx) {
+	for (unsigned i = 0; i < RTREE_CTX_NCACHE; i++) {
+		rtree_ctx_cache_elm_t *cache = &ctx->cache[i];
+		cache->leafkey = RTREE_LEAFKEY_INVALID;
+		cache->leaf = NULL;
+	}
+	for (unsigned i = 0; i < RTREE_CTX_NCACHE_L2; i++) {
+		rtree_ctx_cache_elm_t *cache = &ctx->l2_cache[i];
+		cache->leafkey = RTREE_LEAFKEY_INVALID;
+		cache->leaf = NULL;
+	}
+}

deps/jemalloc/src/safety_check.c

@@ -0,0 +1,24 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+static void (*safety_check_abort)(const char *message);
+
+void safety_check_set_abort(void (*abort_fn)(const char *)) {
+	safety_check_abort = abort_fn;
+}
+
+void safety_check_fail(const char *format, ...) {
+	char buf[MALLOC_PRINTF_BUFSIZE];
+
+	va_list ap;
+	va_start(ap, format);
+	malloc_vsnprintf(buf, MALLOC_PRINTF_BUFSIZE, format, ap);
+	va_end(ap);
+
+	if (safety_check_abort == NULL) {
+		malloc_write(buf);
+		abort();
+	} else {
+		safety_check_abort(buf);
+	}
+}

deps/jemalloc/src/sc.c

@@ -0,0 +1,313 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/bit_util.h"
+#include "jemalloc/internal/bitmap.h"
+#include "jemalloc/internal/pages.h"
+#include "jemalloc/internal/sc.h"
+
+/*
+ * This module computes the size classes used to satisfy allocations.  The logic
+ * here was ported more or less line-by-line from a shell script, and because of
+ * that is not the most idiomatic C.  Eventually we should fix this, but for now
+ * at least the damage is compartmentalized to this file.
+ */
+
+sc_data_t sc_data_global;
+
+static size_t
+reg_size_compute(int lg_base, int lg_delta, int ndelta) {
+	return (ZU(1) << lg_base) + (ZU(ndelta) << lg_delta);
+}
+
+/* Returns the number of pages in the slab. */
+static int
+slab_size(int lg_page, int lg_base, int lg_delta, int ndelta) {
+	size_t page = (ZU(1) << lg_page);
+	size_t reg_size = reg_size_compute(lg_base, lg_delta, ndelta);
+
+	size_t try_slab_size = page;
+	size_t try_nregs = try_slab_size / reg_size;
+	size_t perfect_slab_size = 0;
+	bool perfect = false;
+	/*
+	 * This loop continues until we find the least common multiple of the
+	 * page size and size class size.  Size classes are all of the form
+	 * base + ndelta * delta == (ndelta + base/ndelta) * delta, which is
+	 * (ndelta + ngroup) * delta.  The way we choose slabbing strategies
+	 * means that delta is at most the page size and ndelta < ngroup.  So
+	 * the loop executes for at most 2 * ngroup - 1 iterations, which is
+	 * also the bound on the number of pages in a slab chosen by default.
+	 * With the current default settings, this is at most 7.
+	 */
+	while (!perfect) {
+		perfect_slab_size = try_slab_size;
+		size_t perfect_nregs = try_nregs;
+		try_slab_size += page;
+		try_nregs = try_slab_size / reg_size;
+		if (perfect_slab_size == perfect_nregs * reg_size) {
+			perfect = true;
+		}
+	}
+	return (int)(perfect_slab_size / page);
+}
+
+static void
+size_class(
+    /* Output. */
+    sc_t *sc,
+    /* Configuration decisions. */
+    int lg_max_lookup, int lg_page, int lg_ngroup,
+    /* Inputs specific to the size class. */
+    int index, int lg_base, int lg_delta, int ndelta) {
+	sc->index = index;
+	sc->lg_base = lg_base;
+	sc->lg_delta = lg_delta;
+	sc->ndelta = ndelta;
+	sc->psz = (reg_size_compute(lg_base, lg_delta, ndelta)
+	    % (ZU(1) << lg_page) == 0);
+	size_t size = (ZU(1) << lg_base) + (ZU(ndelta) << lg_delta);
+	if (index == 0) {
+		assert(!sc->psz);
+	}
+	if (size < (ZU(1) << (lg_page + lg_ngroup))) {
+		sc->bin = true;
+		sc->pgs = slab_size(lg_page, lg_base, lg_delta, ndelta);
+	} else {
+		sc->bin = false;
+		sc->pgs = 0;
+	}
+	if (size <= (ZU(1) << lg_max_lookup)) {
+		sc->lg_delta_lookup = lg_delta;
+	} else {
+		sc->lg_delta_lookup = 0;
+	}
+}
+
+static void
+size_classes(
+    /* Output. */
+    sc_data_t *sc_data,
+    /* Determined by the system. */
+    size_t lg_ptr_size, int lg_quantum,
+    /* Configuration decisions. */
+    int lg_tiny_min, int lg_max_lookup, int lg_page, int lg_ngroup) {
+	int ptr_bits = (1 << lg_ptr_size) * 8;
+	int ngroup = (1 << lg_ngroup);
+	int ntiny = 0;
+	int nlbins = 0;
+	int lg_tiny_maxclass = (unsigned)-1;
+	int nbins = 0;
+	int npsizes = 0;
+
+	int index = 0;
+
+	int ndelta = 0;
+	int lg_base = lg_tiny_min;
+	int lg_delta = lg_base;
+
+	/* Outputs that we update as we go. */
+	size_t lookup_maxclass = 0;
+	size_t small_maxclass = 0;
+	int lg_large_minclass = 0;
+	size_t large_maxclass = 0;
+
+	/* Tiny size classes. */
+	while (lg_base < lg_quantum) {
+		sc_t *sc = &sc_data->sc[index];
+		size_class(sc, lg_max_lookup, lg_page, lg_ngroup, index,
+		    lg_base, lg_delta, ndelta);
+		if (sc->lg_delta_lookup != 0) {
+			nlbins = index + 1;
+		}
+		if (sc->psz) {
+			npsizes++;
+		}
+		if (sc->bin) {
+			nbins++;
+		}
+		ntiny++;
+		/* Final written value is correct. */
+		lg_tiny_maxclass = lg_base;
+		index++;
+		lg_delta = lg_base;
+		lg_base++;
+	}
+
+	/* First non-tiny (pseudo) group. */
+	if (ntiny != 0) {
+		sc_t *sc = &sc_data->sc[index];
+		/*
+		 * See the note in sc.h; the first non-tiny size class has an
+		 * unusual encoding.
+		 */
+		lg_base--;
+		ndelta = 1;
+		size_class(sc, lg_max_lookup, lg_page, lg_ngroup, index,
+		    lg_base, lg_delta, ndelta);
+		index++;
+		lg_base++;
+		lg_delta++;
+		if (sc->psz) {
+			npsizes++;
+		}
+		if (sc->bin) {
+			nbins++;
+		}
+	}
+	while (ndelta < ngroup) {
+		sc_t *sc = &sc_data->sc[index];
+		size_class(sc, lg_max_lookup, lg_page, lg_ngroup, index,
+		    lg_base, lg_delta, ndelta);
+		index++;
+		ndelta++;
+		if (sc->psz) {
+			npsizes++;
+		}
+		if (sc->bin) {
+			nbins++;
+		}
+	}
+
+	/* All remaining groups. */
+	lg_base = lg_base + lg_ngroup;
+	while (lg_base < ptr_bits - 1) {
+		ndelta = 1;
+		int ndelta_limit;
+		if (lg_base == ptr_bits - 2) {
+			ndelta_limit = ngroup - 1;
+		} else {
+			ndelta_limit = ngroup;
+		}
+		while (ndelta <= ndelta_limit) {
+			sc_t *sc = &sc_data->sc[index];
+			size_class(sc, lg_max_lookup, lg_page, lg_ngroup, index,
+			    lg_base, lg_delta, ndelta);
+			if (sc->lg_delta_lookup != 0) {
+				nlbins = index + 1;
+				/* Final written value is correct. */
+				lookup_maxclass = (ZU(1) << lg_base)
+				    + (ZU(ndelta) << lg_delta);
+			}
+			if (sc->psz) {
+				npsizes++;
+			}
+			if (sc->bin) {
+				nbins++;
+				/* Final written value is correct. */
+				small_maxclass = (ZU(1) << lg_base)
+				    + (ZU(ndelta) << lg_delta);
+				if (lg_ngroup > 0) {
+					lg_large_minclass = lg_base + 1;
+				} else {
+					lg_large_minclass = lg_base + 2;
+				}
+			}
+			large_maxclass = (ZU(1) << lg_base)
+			    + (ZU(ndelta) << lg_delta);
+			index++;
+			ndelta++;
+		}
+		lg_base++;
+		lg_delta++;
+	}
+	/* Additional outputs. */
+	int nsizes = index;
+	unsigned lg_ceil_nsizes = lg_ceil(nsizes);
+
+	/* Fill in the output data. */
+	sc_data->ntiny = ntiny;
+	sc_data->nlbins = nlbins;
+	sc_data->nbins = nbins;
+	sc_data->nsizes = nsizes;
+	sc_data->lg_ceil_nsizes = lg_ceil_nsizes;
+	sc_data->npsizes = npsizes;
+	sc_data->lg_tiny_maxclass = lg_tiny_maxclass;
+	sc_data->lookup_maxclass = lookup_maxclass;
+	sc_data->small_maxclass = small_maxclass;
+	sc_data->lg_large_minclass = lg_large_minclass;
+	sc_data->large_minclass = (ZU(1) << lg_large_minclass);
+	sc_data->large_maxclass = large_maxclass;
+
+	/*
+	 * We compute these values in two ways:
+	 *   - Incrementally, as above.
+	 *   - In macros, in sc.h.
+	 * The computation is easier when done incrementally, but putting it in
+	 * a constant makes it available to the fast paths without having to
+	 * touch the extra global cacheline.  We assert, however, that the two
+	 * computations are equivalent.
+	 */
+	assert(sc_data->npsizes == SC_NPSIZES);
+	assert(sc_data->lg_tiny_maxclass == SC_LG_TINY_MAXCLASS);
+	assert(sc_data->small_maxclass == SC_SMALL_MAXCLASS);
+	assert(sc_data->large_minclass == SC_LARGE_MINCLASS);
+	assert(sc_data->lg_large_minclass == SC_LG_LARGE_MINCLASS);
+	assert(sc_data->large_maxclass == SC_LARGE_MAXCLASS);
+
+	/* 
+	 * In the allocation fastpath, we want to assume that we can
+	 * unconditionally subtract the requested allocation size from
+	 * a ssize_t, and detect passing through 0 correctly.  This
+	 * results in optimal generated code.  For this to work, the
+	 * maximum allocation size must be less than SSIZE_MAX.
+	 */
+	assert(SC_LARGE_MAXCLASS < SSIZE_MAX);
+}
+
+void
+sc_data_init(sc_data_t *sc_data) {
+	assert(!sc_data->initialized);
+
+	int lg_max_lookup = 12;
+
+	size_classes(sc_data, LG_SIZEOF_PTR, LG_QUANTUM, SC_LG_TINY_MIN,
+	    lg_max_lookup, LG_PAGE, 2);
+
+	sc_data->initialized = true;
+}
+
+static void
+sc_data_update_sc_slab_size(sc_t *sc, size_t reg_size, size_t pgs_guess) {
+	size_t min_pgs = reg_size / PAGE;
+	if (reg_size % PAGE != 0) {
+		min_pgs++;
+	}
+	/*
+	 * BITMAP_MAXBITS is actually determined by putting the smallest
+	 * possible size-class on one page, so this can never be 0.
+	 */
+	size_t max_pgs = BITMAP_MAXBITS * reg_size / PAGE;
+
+	assert(min_pgs <= max_pgs);
+	assert(min_pgs > 0);
+	assert(max_pgs >= 1);
+	if (pgs_guess < min_pgs) {
+		sc->pgs = (int)min_pgs;
+	} else if (pgs_guess > max_pgs) {
+		sc->pgs = (int)max_pgs;
+	} else {
+		sc->pgs = (int)pgs_guess;
+	}
+}
+
+void
+sc_data_update_slab_size(sc_data_t *data, size_t begin, size_t end, int pgs) {
+	assert(data->initialized);
+	for (int i = 0; i < data->nsizes; i++) {
+		sc_t *sc = &data->sc[i];
+		if (!sc->bin) {
+			break;
+		}
+		size_t reg_size = reg_size_compute(sc->lg_base, sc->lg_delta,
+		    sc->ndelta);
+		if (begin <= reg_size && reg_size <= end) {
+			sc_data_update_sc_slab_size(sc, reg_size, pgs);
+		}
+	}
+}
+
+void
+sc_boot(sc_data_t *data) {
+	sc_data_init(data);
+}

deps/jemalloc/src/spin.c

@@ -0,0 +1,4 @@
+#define JEMALLOC_SPIN_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+
+#include "jemalloc/internal/spin.h"

deps/jemalloc/src/sz.c

@@ -0,0 +1,64 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/sz.h"
+
+JEMALLOC_ALIGNED(CACHELINE)
+size_t sz_pind2sz_tab[SC_NPSIZES+1];
+
+static void
+sz_boot_pind2sz_tab(const sc_data_t *sc_data) {
+	int pind = 0;
+	for (unsigned i = 0; i < SC_NSIZES; i++) {
+		const sc_t *sc = &sc_data->sc[i];
+		if (sc->psz) {
+			sz_pind2sz_tab[pind] = (ZU(1) << sc->lg_base)
+			    + (ZU(sc->ndelta) << sc->lg_delta);
+			pind++;
+		}
+	}
+	for (int i = pind; i <= (int)SC_NPSIZES; i++) {
+		sz_pind2sz_tab[pind] = sc_data->large_maxclass + PAGE;
+	}
+}
+
+JEMALLOC_ALIGNED(CACHELINE)
+size_t sz_index2size_tab[SC_NSIZES];
+
+static void
+sz_boot_index2size_tab(const sc_data_t *sc_data) {
+	for (unsigned i = 0; i < SC_NSIZES; i++) {
+		const sc_t *sc = &sc_data->sc[i];
+		sz_index2size_tab[i] = (ZU(1) << sc->lg_base)
+		    + (ZU(sc->ndelta) << (sc->lg_delta));
+	}
+}
+
+/*
+ * To keep this table small, we divide sizes by the tiny min size, which gives
+ * the smallest interval for which the result can change.
+ */
+JEMALLOC_ALIGNED(CACHELINE)
+uint8_t sz_size2index_tab[(SC_LOOKUP_MAXCLASS >> SC_LG_TINY_MIN) + 1];
+
+static void
+sz_boot_size2index_tab(const sc_data_t *sc_data) {
+	size_t dst_max = (SC_LOOKUP_MAXCLASS >> SC_LG_TINY_MIN) + 1;
+	size_t dst_ind = 0;
+	for (unsigned sc_ind = 0; sc_ind < SC_NSIZES && dst_ind < dst_max;
+	    sc_ind++) {
+		const sc_t *sc = &sc_data->sc[sc_ind];
+		size_t sz = (ZU(1) << sc->lg_base)
+		    + (ZU(sc->ndelta) << sc->lg_delta);
+		size_t max_ind = ((sz + (ZU(1) << SC_LG_TINY_MIN) - 1)
+				   >> SC_LG_TINY_MIN);
+		for (; dst_ind <= max_ind && dst_ind < dst_max; dst_ind++) {
+			sz_size2index_tab[dst_ind] = sc_ind;
+		}
+	}
+}
+
+void
+sz_boot(const sc_data_t *sc_data) {
+	sz_boot_pind2sz_tab(sc_data);
+	sz_boot_index2size_tab(sc_data);
+	sz_boot_size2index_tab(sc_data);
+}

deps/jemalloc/src/tcache.c

@@ -0,0 +1,798 @@
+#define JEMALLOC_TCACHE_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/mutex.h"
+#include "jemalloc/internal/safety_check.h"
+#include "jemalloc/internal/sc.h"
+
+/******************************************************************************/
+/* Data. */
+
+bool	opt_tcache = true;
+ssize_t	opt_lg_tcache_max = LG_TCACHE_MAXCLASS_DEFAULT;
+
+cache_bin_info_t	*tcache_bin_info;
+static unsigned		stack_nelms; /* Total stack elms per tcache. */
+
+unsigned		nhbins;
+size_t			tcache_maxclass;
+
+tcaches_t		*tcaches;
+
+/* Index of first element within tcaches that has never been used. */
+static unsigned		tcaches_past;
+
+/* Head of singly linked list tracking available tcaches elements. */
+static tcaches_t	*tcaches_avail;
+
+/* Protects tcaches{,_past,_avail}. */
+static malloc_mutex_t	tcaches_mtx;
+
+/******************************************************************************/
+
+size_t
+tcache_salloc(tsdn_t *tsdn, const void *ptr) {
+	return arena_salloc(tsdn, ptr);
+}
+
+void
+tcache_event_hard(tsd_t *tsd, tcache_t *tcache) {
+	szind_t binind = tcache->next_gc_bin;
+
+	cache_bin_t *tbin;
+	if (binind < SC_NBINS) {
+		tbin = tcache_small_bin_get(tcache, binind);
+	} else {
+		tbin = tcache_large_bin_get(tcache, binind);
+	}
+	if (tbin->low_water > 0) {
+		/*
+		 * Flush (ceiling) 3/4 of the objects below the low water mark.
+		 */
+		if (binind < SC_NBINS) {
+			tcache_bin_flush_small(tsd, tcache, tbin, binind,
+			    tbin->ncached - tbin->low_water + (tbin->low_water
+			    >> 2));
+			/*
+			 * Reduce fill count by 2X.  Limit lg_fill_div such that
+			 * the fill count is always at least 1.
+			 */
+			cache_bin_info_t *tbin_info = &tcache_bin_info[binind];
+			if ((tbin_info->ncached_max >>
+			     (tcache->lg_fill_div[binind] + 1)) >= 1) {
+				tcache->lg_fill_div[binind]++;
+			}
+		} else {
+			tcache_bin_flush_large(tsd, tbin, binind, tbin->ncached
+			    - tbin->low_water + (tbin->low_water >> 2), tcache);
+		}
+	} else if (tbin->low_water < 0) {
+		/*
+		 * Increase fill count by 2X for small bins.  Make sure
+		 * lg_fill_div stays greater than 0.
+		 */
+		if (binind < SC_NBINS && tcache->lg_fill_div[binind] > 1) {
+			tcache->lg_fill_div[binind]--;
+		}
+	}
+	tbin->low_water = tbin->ncached;
+
+	tcache->next_gc_bin++;
+	if (tcache->next_gc_bin == nhbins) {
+		tcache->next_gc_bin = 0;
+	}
+}
+
+void *
+tcache_alloc_small_hard(tsdn_t *tsdn, arena_t *arena, tcache_t *tcache,
+    cache_bin_t *tbin, szind_t binind, bool *tcache_success) {
+	void *ret;
+
+	assert(tcache->arena != NULL);
+	arena_tcache_fill_small(tsdn, arena, tcache, tbin, binind,
+	    config_prof ? tcache->prof_accumbytes : 0);
+	if (config_prof) {
+		tcache->prof_accumbytes = 0;
+	}
+	ret = cache_bin_alloc_easy(tbin, tcache_success);
+
+	return ret;
+}
+
+/* Enabled with --enable-extra-size-check. */
+static void
+tbin_extents_lookup_size_check(tsdn_t *tsdn, cache_bin_t *tbin, szind_t binind,
+    size_t nflush, extent_t **extents){
+	rtree_ctx_t rtree_ctx_fallback;
+	rtree_ctx_t *rtree_ctx = tsdn_rtree_ctx(tsdn, &rtree_ctx_fallback);
+
+	/*
+	 * Verify that the items in the tcache all have the correct size; this
+	 * is useful for catching sized deallocation bugs, also to fail early
+	 * instead of corrupting metadata.  Since this can be turned on for opt
+	 * builds, avoid the branch in the loop.
+	 */
+	szind_t szind;
+	size_t sz_sum = binind * nflush;
+	for (unsigned i = 0 ; i < nflush; i++) {
+		rtree_extent_szind_read(tsdn, &extents_rtree,
+		    rtree_ctx, (uintptr_t)*(tbin->avail - 1 - i), true,
+		    &extents[i], &szind);
+		sz_sum -= szind;
+	}
+	if (sz_sum != 0) {
+		safety_check_fail("<jemalloc>: size mismatch in thread cache "
+		    "detected, likely caused by sized deallocation bugs by "
+		    "application. Abort.\n");
+		abort();
+	}
+}
+
+void
+tcache_bin_flush_small(tsd_t *tsd, tcache_t *tcache, cache_bin_t *tbin,
+    szind_t binind, unsigned rem) {
+	bool merged_stats = false;
+
+	assert(binind < SC_NBINS);
+	assert((cache_bin_sz_t)rem <= tbin->ncached);
+
+	arena_t *arena = tcache->arena;
+	assert(arena != NULL);
+	unsigned nflush = tbin->ncached - rem;
+	VARIABLE_ARRAY(extent_t *, item_extent, nflush);
+
+	/* Look up extent once per item. */
+	if (config_opt_safety_checks) {
+		tbin_extents_lookup_size_check(tsd_tsdn(tsd), tbin, binind,
+		    nflush, item_extent);
+	} else {
+		for (unsigned i = 0 ; i < nflush; i++) {
+			item_extent[i] = iealloc(tsd_tsdn(tsd),
+			    *(tbin->avail - 1 - i));
+		}
+	}
+	while (nflush > 0) {
+		/* Lock the arena bin associated with the first object. */
+		extent_t *extent = item_extent[0];
+		unsigned bin_arena_ind = extent_arena_ind_get(extent);
+		arena_t *bin_arena = arena_get(tsd_tsdn(tsd), bin_arena_ind,
+		    false);
+		unsigned binshard = extent_binshard_get(extent);
+		assert(binshard < bin_infos[binind].n_shards);
+		bin_t *bin = &bin_arena->bins[binind].bin_shards[binshard];
+
+		if (config_prof && bin_arena == arena) {
+			if (arena_prof_accum(tsd_tsdn(tsd), arena,
+			    tcache->prof_accumbytes)) {
+				prof_idump(tsd_tsdn(tsd));
+			}
+			tcache->prof_accumbytes = 0;
+		}
+
+		malloc_mutex_lock(tsd_tsdn(tsd), &bin->lock);
+		if (config_stats && bin_arena == arena && !merged_stats) {
+			merged_stats = true;
+			bin->stats.nflushes++;
+			bin->stats.nrequests += tbin->tstats.nrequests;
+			tbin->tstats.nrequests = 0;
+		}
+		unsigned ndeferred = 0;
+		for (unsigned i = 0; i < nflush; i++) {
+			void *ptr = *(tbin->avail - 1 - i);
+			extent = item_extent[i];
+			assert(ptr != NULL && extent != NULL);
+
+			if (extent_arena_ind_get(extent) == bin_arena_ind
+			    && extent_binshard_get(extent) == binshard) {
+				arena_dalloc_bin_junked_locked(tsd_tsdn(tsd),
+				    bin_arena, bin, binind, extent, ptr);
+			} else {
+				/*
+				 * This object was allocated via a different
+				 * arena bin than the one that is currently
+				 * locked.  Stash the object, so that it can be
+				 * handled in a future pass.
+				 */
+				*(tbin->avail - 1 - ndeferred) = ptr;
+				item_extent[ndeferred] = extent;
+				ndeferred++;
+			}
+		}
+		malloc_mutex_unlock(tsd_tsdn(tsd), &bin->lock);
+		arena_decay_ticks(tsd_tsdn(tsd), bin_arena, nflush - ndeferred);
+		nflush = ndeferred;
+	}
+	if (config_stats && !merged_stats) {
+		/*
+		 * The flush loop didn't happen to flush to this thread's
+		 * arena, so the stats didn't get merged.  Manually do so now.
+		 */
+		unsigned binshard;
+		bin_t *bin = arena_bin_choose_lock(tsd_tsdn(tsd), arena, binind,
+		    &binshard);
+		bin->stats.nflushes++;
+		bin->stats.nrequests += tbin->tstats.nrequests;
+		tbin->tstats.nrequests = 0;
+		malloc_mutex_unlock(tsd_tsdn(tsd), &bin->lock);
+	}
+
+	memmove(tbin->avail - rem, tbin->avail - tbin->ncached, rem *
+	    sizeof(void *));
+	tbin->ncached = rem;
+	if (tbin->ncached < tbin->low_water) {
+		tbin->low_water = tbin->ncached;
+	}
+}
+
+void
+tcache_bin_flush_large(tsd_t *tsd, cache_bin_t *tbin, szind_t binind,
+    unsigned rem, tcache_t *tcache) {
+	bool merged_stats = false;
+
+	assert(binind < nhbins);
+	assert((cache_bin_sz_t)rem <= tbin->ncached);
+
+	arena_t *tcache_arena = tcache->arena;
+	assert(tcache_arena != NULL);
+	unsigned nflush = tbin->ncached - rem;
+	VARIABLE_ARRAY(extent_t *, item_extent, nflush);
+
+#ifndef JEMALLOC_EXTRA_SIZE_CHECK
+	/* Look up extent once per item. */
+	for (unsigned i = 0 ; i < nflush; i++) {
+		item_extent[i] = iealloc(tsd_tsdn(tsd), *(tbin->avail - 1 - i));
+	}
+#else
+	tbin_extents_lookup_size_check(tsd_tsdn(tsd), tbin, binind, nflush,
+	    item_extent);
+#endif
+	while (nflush > 0) {
+		/* Lock the arena associated with the first object. */
+		extent_t *extent = item_extent[0];
+		unsigned locked_arena_ind = extent_arena_ind_get(extent);
+		arena_t *locked_arena = arena_get(tsd_tsdn(tsd),
+		    locked_arena_ind, false);
+		bool idump;
+
+		if (config_prof) {
+			idump = false;
+		}
+
+		bool lock_large = !arena_is_auto(locked_arena);
+		if (lock_large) {
+			malloc_mutex_lock(tsd_tsdn(tsd), &locked_arena->large_mtx);
+		}
+		for (unsigned i = 0; i < nflush; i++) {
+			void *ptr = *(tbin->avail - 1 - i);
+			assert(ptr != NULL);
+			extent = item_extent[i];
+			if (extent_arena_ind_get(extent) == locked_arena_ind) {
+				large_dalloc_prep_junked_locked(tsd_tsdn(tsd),
+				    extent);
+			}
+		}
+		if ((config_prof || config_stats) &&
+		    (locked_arena == tcache_arena)) {
+			if (config_prof) {
+				idump = arena_prof_accum(tsd_tsdn(tsd),
+				    tcache_arena, tcache->prof_accumbytes);
+				tcache->prof_accumbytes = 0;
+			}
+			if (config_stats) {
+				merged_stats = true;
+				arena_stats_large_flush_nrequests_add(
+				    tsd_tsdn(tsd), &tcache_arena->stats, binind,
+				    tbin->tstats.nrequests);
+				tbin->tstats.nrequests = 0;
+			}
+		}
+		if (lock_large) {
+			malloc_mutex_unlock(tsd_tsdn(tsd), &locked_arena->large_mtx);
+		}
+
+		unsigned ndeferred = 0;
+		for (unsigned i = 0; i < nflush; i++) {
+			void *ptr = *(tbin->avail - 1 - i);
+			extent = item_extent[i];
+			assert(ptr != NULL && extent != NULL);
+
+			if (extent_arena_ind_get(extent) == locked_arena_ind) {
+				large_dalloc_finish(tsd_tsdn(tsd), extent);
+			} else {
+				/*
+				 * This object was allocated via a different
+				 * arena than the one that is currently locked.
+				 * Stash the object, so that it can be handled
+				 * in a future pass.
+				 */
+				*(tbin->avail - 1 - ndeferred) = ptr;
+				item_extent[ndeferred] = extent;
+				ndeferred++;
+			}
+		}
+		if (config_prof && idump) {
+			prof_idump(tsd_tsdn(tsd));
+		}
+		arena_decay_ticks(tsd_tsdn(tsd), locked_arena, nflush -
+		    ndeferred);
+		nflush = ndeferred;
+	}
+	if (config_stats && !merged_stats) {
+		/*
+		 * The flush loop didn't happen to flush to this thread's
+		 * arena, so the stats didn't get merged.  Manually do so now.
+		 */
+		arena_stats_large_flush_nrequests_add(tsd_tsdn(tsd),
+		    &tcache_arena->stats, binind, tbin->tstats.nrequests);
+		tbin->tstats.nrequests = 0;
+	}
+
+	memmove(tbin->avail - rem, tbin->avail - tbin->ncached, rem *
+	    sizeof(void *));
+	tbin->ncached = rem;
+	if (tbin->ncached < tbin->low_water) {
+		tbin->low_water = tbin->ncached;
+	}
+}
+
+void
+tcache_arena_associate(tsdn_t *tsdn, tcache_t *tcache, arena_t *arena) {
+	assert(tcache->arena == NULL);
+	tcache->arena = arena;
+
+	if (config_stats) {
+		/* Link into list of extant tcaches. */
+		malloc_mutex_lock(tsdn, &arena->tcache_ql_mtx);
+
+		ql_elm_new(tcache, link);
+		ql_tail_insert(&arena->tcache_ql, tcache, link);
+		cache_bin_array_descriptor_init(
+		    &tcache->cache_bin_array_descriptor, tcache->bins_small,
+		    tcache->bins_large);
+		ql_tail_insert(&arena->cache_bin_array_descriptor_ql,
+		    &tcache->cache_bin_array_descriptor, link);
+
+		malloc_mutex_unlock(tsdn, &arena->tcache_ql_mtx);
+	}
+}
+
+static void
+tcache_arena_dissociate(tsdn_t *tsdn, tcache_t *tcache) {
+	arena_t *arena = tcache->arena;
+	assert(arena != NULL);
+	if (config_stats) {
+		/* Unlink from list of extant tcaches. */
+		malloc_mutex_lock(tsdn, &arena->tcache_ql_mtx);
+		if (config_debug) {
+			bool in_ql = false;
+			tcache_t *iter;
+			ql_foreach(iter, &arena->tcache_ql, link) {
+				if (iter == tcache) {
+					in_ql = true;
+					break;
+				}
+			}
+			assert(in_ql);
+		}
+		ql_remove(&arena->tcache_ql, tcache, link);
+		ql_remove(&arena->cache_bin_array_descriptor_ql,
+		    &tcache->cache_bin_array_descriptor, link);
+		tcache_stats_merge(tsdn, tcache, arena);
+		malloc_mutex_unlock(tsdn, &arena->tcache_ql_mtx);
+	}
+	tcache->arena = NULL;
+}
+
+void
+tcache_arena_reassociate(tsdn_t *tsdn, tcache_t *tcache, arena_t *arena) {
+	tcache_arena_dissociate(tsdn, tcache);
+	tcache_arena_associate(tsdn, tcache, arena);
+}
+
+bool
+tsd_tcache_enabled_data_init(tsd_t *tsd) {
+	/* Called upon tsd initialization. */
+	tsd_tcache_enabled_set(tsd, opt_tcache);
+	tsd_slow_update(tsd);
+
+	if (opt_tcache) {
+		/* Trigger tcache init. */
+		tsd_tcache_data_init(tsd);
+	}
+
+	return false;
+}
+
+/* Initialize auto tcache (embedded in TSD). */
+static void
+tcache_init(tsd_t *tsd, tcache_t *tcache, void *avail_stack) {
+	memset(&tcache->link, 0, sizeof(ql_elm(tcache_t)));
+	tcache->prof_accumbytes = 0;
+	tcache->next_gc_bin = 0;
+	tcache->arena = NULL;
+
+	ticker_init(&tcache->gc_ticker, TCACHE_GC_INCR);
+
+	size_t stack_offset = 0;
+	assert((TCACHE_NSLOTS_SMALL_MAX & 1U) == 0);
+	memset(tcache->bins_small, 0, sizeof(cache_bin_t) * SC_NBINS);
+	memset(tcache->bins_large, 0, sizeof(cache_bin_t) * (nhbins - SC_NBINS));
+	unsigned i = 0;
+	for (; i < SC_NBINS; i++) {
+		tcache->lg_fill_div[i] = 1;
+		stack_offset += tcache_bin_info[i].ncached_max * sizeof(void *);
+		/*
+		 * avail points past the available space.  Allocations will
+		 * access the slots toward higher addresses (for the benefit of
+		 * prefetch).
+		 */
+		tcache_small_bin_get(tcache, i)->avail =
+		    (void **)((uintptr_t)avail_stack + (uintptr_t)stack_offset);
+	}
+	for (; i < nhbins; i++) {
+		stack_offset += tcache_bin_info[i].ncached_max * sizeof(void *);
+		tcache_large_bin_get(tcache, i)->avail =
+		    (void **)((uintptr_t)avail_stack + (uintptr_t)stack_offset);
+	}
+	assert(stack_offset == stack_nelms * sizeof(void *));
+}
+
+/* Initialize auto tcache (embedded in TSD). */
+bool
+tsd_tcache_data_init(tsd_t *tsd) {
+	tcache_t *tcache = tsd_tcachep_get_unsafe(tsd);
+	assert(tcache_small_bin_get(tcache, 0)->avail == NULL);
+	size_t size = stack_nelms * sizeof(void *);
+	/* Avoid false cacheline sharing. */
+	size = sz_sa2u(size, CACHELINE);
+
+	void *avail_array = ipallocztm(tsd_tsdn(tsd), size, CACHELINE, true,
+	    NULL, true, arena_get(TSDN_NULL, 0, true));
+	if (avail_array == NULL) {
+		return true;
+	}
+
+	tcache_init(tsd, tcache, avail_array);
+	/*
+	 * Initialization is a bit tricky here.  After malloc init is done, all
+	 * threads can rely on arena_choose and associate tcache accordingly.
+	 * However, the thread that does actual malloc bootstrapping relies on
+	 * functional tsd, and it can only rely on a0.  In that case, we
+	 * associate its tcache to a0 temporarily, and later on
+	 * arena_choose_hard() will re-associate properly.
+	 */
+	tcache->arena = NULL;
+	arena_t *arena;
+	if (!malloc_initialized()) {
+		/* If in initialization, assign to a0. */
+		arena = arena_get(tsd_tsdn(tsd), 0, false);
+		tcache_arena_associate(tsd_tsdn(tsd), tcache, arena);
+	} else {
+		arena = arena_choose(tsd, NULL);
+		/* This may happen if thread.tcache.enabled is used. */
+		if (tcache->arena == NULL) {
+			tcache_arena_associate(tsd_tsdn(tsd), tcache, arena);
+		}
+	}
+	assert(arena == tcache->arena);
+
+	return false;
+}
+
+/* Created manual tcache for tcache.create mallctl. */
+tcache_t *
+tcache_create_explicit(tsd_t *tsd) {
+	tcache_t *tcache;
+	size_t size, stack_offset;
+
+	size = sizeof(tcache_t);
+	/* Naturally align the pointer stacks. */
+	size = PTR_CEILING(size);
+	stack_offset = size;
+	size += stack_nelms * sizeof(void *);
+	/* Avoid false cacheline sharing. */
+	size = sz_sa2u(size, CACHELINE);
+
+	tcache = ipallocztm(tsd_tsdn(tsd), size, CACHELINE, true, NULL, true,
+	    arena_get(TSDN_NULL, 0, true));
+	if (tcache == NULL) {
+		return NULL;
+	}
+
+	tcache_init(tsd, tcache,
+	    (void *)((uintptr_t)tcache + (uintptr_t)stack_offset));
+	tcache_arena_associate(tsd_tsdn(tsd), tcache, arena_ichoose(tsd, NULL));
+
+	return tcache;
+}
+
+static void
+tcache_flush_cache(tsd_t *tsd, tcache_t *tcache) {
+	assert(tcache->arena != NULL);
+
+	for (unsigned i = 0; i < SC_NBINS; i++) {
+		cache_bin_t *tbin = tcache_small_bin_get(tcache, i);
+		tcache_bin_flush_small(tsd, tcache, tbin, i, 0);
+
+		if (config_stats) {
+			assert(tbin->tstats.nrequests == 0);
+		}
+	}
+	for (unsigned i = SC_NBINS; i < nhbins; i++) {
+		cache_bin_t *tbin = tcache_large_bin_get(tcache, i);
+		tcache_bin_flush_large(tsd, tbin, i, 0, tcache);
+
+		if (config_stats) {
+			assert(tbin->tstats.nrequests == 0);
+		}
+	}
+
+	if (config_prof && tcache->prof_accumbytes > 0 &&
+	    arena_prof_accum(tsd_tsdn(tsd), tcache->arena,
+	    tcache->prof_accumbytes)) {
+		prof_idump(tsd_tsdn(tsd));
+	}
+}
+
+void
+tcache_flush(tsd_t *tsd) {
+	assert(tcache_available(tsd));
+	tcache_flush_cache(tsd, tsd_tcachep_get(tsd));
+}
+
+static void
+tcache_destroy(tsd_t *tsd, tcache_t *tcache, bool tsd_tcache) {
+	tcache_flush_cache(tsd, tcache);
+	arena_t *arena = tcache->arena;
+	tcache_arena_dissociate(tsd_tsdn(tsd), tcache);
+
+	if (tsd_tcache) {
+		/* Release the avail array for the TSD embedded auto tcache. */
+		void *avail_array =
+		    (void *)((uintptr_t)tcache_small_bin_get(tcache, 0)->avail -
+		    (uintptr_t)tcache_bin_info[0].ncached_max * sizeof(void *));
+		idalloctm(tsd_tsdn(tsd), avail_array, NULL, NULL, true, true);
+	} else {
+		/* Release both the tcache struct and avail array. */
+		idalloctm(tsd_tsdn(tsd), tcache, NULL, NULL, true, true);
+	}
+
+	/*
+	 * The deallocation and tcache flush above may not trigger decay since
+	 * we are on the tcache shutdown path (potentially with non-nominal
+	 * tsd).  Manually trigger decay to avoid pathological cases.  Also
+	 * include arena 0 because the tcache array is allocated from it.
+	 */
+	arena_decay(tsd_tsdn(tsd), arena_get(tsd_tsdn(tsd), 0, false),
+	    false, false);
+
+	if (arena_nthreads_get(arena, false) == 0 &&
+	    !background_thread_enabled()) {
+		/* Force purging when no threads assigned to the arena anymore. */
+		arena_decay(tsd_tsdn(tsd), arena, false, true);
+	} else {
+		arena_decay(tsd_tsdn(tsd), arena, false, false);
+	}
+}
+
+/* For auto tcache (embedded in TSD) only. */
+void
+tcache_cleanup(tsd_t *tsd) {
+	tcache_t *tcache = tsd_tcachep_get(tsd);
+	if (!tcache_available(tsd)) {
+		assert(tsd_tcache_enabled_get(tsd) == false);
+		if (config_debug) {
+			assert(tcache_small_bin_get(tcache, 0)->avail == NULL);
+		}
+		return;
+	}
+	assert(tsd_tcache_enabled_get(tsd));
+	assert(tcache_small_bin_get(tcache, 0)->avail != NULL);
+
+	tcache_destroy(tsd, tcache, true);
+	if (config_debug) {
+		tcache_small_bin_get(tcache, 0)->avail = NULL;
+	}
+}
+
+void
+tcache_stats_merge(tsdn_t *tsdn, tcache_t *tcache, arena_t *arena) {
+	unsigned i;
+
+	cassert(config_stats);
+
+	/* Merge and reset tcache stats. */
+	for (i = 0; i < SC_NBINS; i++) {
+		cache_bin_t *tbin = tcache_small_bin_get(tcache, i);
+		unsigned binshard;
+		bin_t *bin = arena_bin_choose_lock(tsdn, arena, i, &binshard);
+		bin->stats.nrequests += tbin->tstats.nrequests;
+		malloc_mutex_unlock(tsdn, &bin->lock);
+		tbin->tstats.nrequests = 0;
+	}
+
+	for (; i < nhbins; i++) {
+		cache_bin_t *tbin = tcache_large_bin_get(tcache, i);
+		arena_stats_large_flush_nrequests_add(tsdn, &arena->stats, i,
+		    tbin->tstats.nrequests);
+		tbin->tstats.nrequests = 0;
+	}
+}
+
+static bool
+tcaches_create_prep(tsd_t *tsd) {
+	bool err;
+
+	malloc_mutex_lock(tsd_tsdn(tsd), &tcaches_mtx);
+
+	if (tcaches == NULL) {
+		tcaches = base_alloc(tsd_tsdn(tsd), b0get(), sizeof(tcache_t *)
+		    * (MALLOCX_TCACHE_MAX+1), CACHELINE);
+		if (tcaches == NULL) {
+			err = true;
+			goto label_return;
+		}
+	}
+
+	if (tcaches_avail == NULL && tcaches_past > MALLOCX_TCACHE_MAX) {
+		err = true;
+		goto label_return;
+	}
+
+	err = false;
+label_return:
+	malloc_mutex_unlock(tsd_tsdn(tsd), &tcaches_mtx);
+	return err;
+}
+
+bool
+tcaches_create(tsd_t *tsd, unsigned *r_ind) {
+	witness_assert_depth(tsdn_witness_tsdp_get(tsd_tsdn(tsd)), 0);
+
+	bool err;
+
+	if (tcaches_create_prep(tsd)) {
+		err = true;
+		goto label_return;
+	}
+
+	tcache_t *tcache = tcache_create_explicit(tsd);
+	if (tcache == NULL) {
+		err = true;
+		goto label_return;
+	}
+
+	tcaches_t *elm;
+	malloc_mutex_lock(tsd_tsdn(tsd), &tcaches_mtx);
+	if (tcaches_avail != NULL) {
+		elm = tcaches_avail;
+		tcaches_avail = tcaches_avail->next;
+		elm->tcache = tcache;
+		*r_ind = (unsigned)(elm - tcaches);
+	} else {
+		elm = &tcaches[tcaches_past];
+		elm->tcache = tcache;
+		*r_ind = tcaches_past;
+		tcaches_past++;
+	}
+	malloc_mutex_unlock(tsd_tsdn(tsd), &tcaches_mtx);
+
+	err = false;
+label_return:
+	witness_assert_depth(tsdn_witness_tsdp_get(tsd_tsdn(tsd)), 0);
+	return err;
+}
+
+static tcache_t *
+tcaches_elm_remove(tsd_t *tsd, tcaches_t *elm, bool allow_reinit) {
+	malloc_mutex_assert_owner(tsd_tsdn(tsd), &tcaches_mtx);
+
+	if (elm->tcache == NULL) {
+		return NULL;
+	}
+	tcache_t *tcache = elm->tcache;
+	if (allow_reinit) {
+		elm->tcache = TCACHES_ELM_NEED_REINIT;
+	} else {
+		elm->tcache = NULL;
+	}
+
+	if (tcache == TCACHES_ELM_NEED_REINIT) {
+		return NULL;
+	}
+	return tcache;
+}
+
+void
+tcaches_flush(tsd_t *tsd, unsigned ind) {
+	malloc_mutex_lock(tsd_tsdn(tsd), &tcaches_mtx);
+	tcache_t *tcache = tcaches_elm_remove(tsd, &tcaches[ind], true);
+	malloc_mutex_unlock(tsd_tsdn(tsd), &tcaches_mtx);
+	if (tcache != NULL) {
+		/* Destroy the tcache; recreate in tcaches_get() if needed. */
+		tcache_destroy(tsd, tcache, false);
+	}
+}
+
+void
+tcaches_destroy(tsd_t *tsd, unsigned ind) {
+	malloc_mutex_lock(tsd_tsdn(tsd), &tcaches_mtx);
+	tcaches_t *elm = &tcaches[ind];
+	tcache_t *tcache = tcaches_elm_remove(tsd, elm, false);
+	elm->next = tcaches_avail;
+	tcaches_avail = elm;
+	malloc_mutex_unlock(tsd_tsdn(tsd), &tcaches_mtx);
+	if (tcache != NULL) {
+		tcache_destroy(tsd, tcache, false);
+	}
+}
+
+bool
+tcache_boot(tsdn_t *tsdn) {
+	/* If necessary, clamp opt_lg_tcache_max. */
+	if (opt_lg_tcache_max < 0 || (ZU(1) << opt_lg_tcache_max) <
+	    SC_SMALL_MAXCLASS) {
+		tcache_maxclass = SC_SMALL_MAXCLASS;
+	} else {
+		tcache_maxclass = (ZU(1) << opt_lg_tcache_max);
+	}
+
+	if (malloc_mutex_init(&tcaches_mtx, "tcaches", WITNESS_RANK_TCACHES,
+	    malloc_mutex_rank_exclusive)) {
+		return true;
+	}
+
+	nhbins = sz_size2index(tcache_maxclass) + 1;
+
+	/* Initialize tcache_bin_info. */
+	tcache_bin_info = (cache_bin_info_t *)base_alloc(tsdn, b0get(), nhbins
+	    * sizeof(cache_bin_info_t), CACHELINE);
+	if (tcache_bin_info == NULL) {
+		return true;
+	}
+	stack_nelms = 0;
+	unsigned i;
+	for (i = 0; i < SC_NBINS; i++) {
+		if ((bin_infos[i].nregs << 1) <= TCACHE_NSLOTS_SMALL_MIN) {
+			tcache_bin_info[i].ncached_max =
+			    TCACHE_NSLOTS_SMALL_MIN;
+		} else if ((bin_infos[i].nregs << 1) <=
+		    TCACHE_NSLOTS_SMALL_MAX) {
+			tcache_bin_info[i].ncached_max =
+			    (bin_infos[i].nregs << 1);
+		} else {
+			tcache_bin_info[i].ncached_max =
+			    TCACHE_NSLOTS_SMALL_MAX;
+		}
+		stack_nelms += tcache_bin_info[i].ncached_max;
+	}
+	for (; i < nhbins; i++) {
+		tcache_bin_info[i].ncached_max = TCACHE_NSLOTS_LARGE;
+		stack_nelms += tcache_bin_info[i].ncached_max;
+	}
+
+	return false;
+}
+
+void
+tcache_prefork(tsdn_t *tsdn) {
+	if (!config_prof && opt_tcache) {
+		malloc_mutex_prefork(tsdn, &tcaches_mtx);
+	}
+}
+
+void
+tcache_postfork_parent(tsdn_t *tsdn) {
+	if (!config_prof && opt_tcache) {
+		malloc_mutex_postfork_parent(tsdn, &tcaches_mtx);
+	}
+}
+
+void
+tcache_postfork_child(tsdn_t *tsdn) {
+	if (!config_prof && opt_tcache) {
+		malloc_mutex_postfork_child(tsdn, &tcaches_mtx);
+	}
+}

deps/jemalloc/src/test_hooks.c

@@ -0,0 +1,12 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+
+/*
+ * The hooks are a little bit screwy -- they're not genuinely exported in the
+ * sense that we want them available to end-users, but we do want them visible
+ * from outside the generated library, so that we can use them in test code.
+ */
+JEMALLOC_EXPORT
+void (*test_hooks_arena_new_hook)() = NULL;
+
+JEMALLOC_EXPORT
+void (*test_hooks_libc_hook)() = NULL;

deps/jemalloc/src/ticker.c

@@ -0,0 +1,3 @@
+#define JEMALLOC_TICKER_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"

deps/jemalloc/src/tsd.c

@@ -0,0 +1,534 @@
+#define JEMALLOC_TSD_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/mutex.h"
+#include "jemalloc/internal/rtree.h"
+
+/******************************************************************************/
+/* Data. */
+
+static unsigned ncleanups;
+static malloc_tsd_cleanup_t cleanups[MALLOC_TSD_CLEANUPS_MAX];
+
+/* TSD_INITIALIZER triggers "-Wmissing-field-initializer" */
+JEMALLOC_DIAGNOSTIC_PUSH
+JEMALLOC_DIAGNOSTIC_IGNORE_MISSING_STRUCT_FIELD_INITIALIZERS
+
+#ifdef JEMALLOC_MALLOC_THREAD_CLEANUP
+JEMALLOC_TSD_TYPE_ATTR(tsd_t) tsd_tls = TSD_INITIALIZER;
+JEMALLOC_TSD_TYPE_ATTR(bool) JEMALLOC_TLS_MODEL tsd_initialized = false;
+bool tsd_booted = false;
+#elif (defined(JEMALLOC_TLS))
+JEMALLOC_TSD_TYPE_ATTR(tsd_t) tsd_tls = TSD_INITIALIZER;
+pthread_key_t tsd_tsd;
+bool tsd_booted = false;
+#elif (defined(_WIN32))
+DWORD tsd_tsd;
+tsd_wrapper_t tsd_boot_wrapper = {false, TSD_INITIALIZER};
+bool tsd_booted = false;
+#else
+
+/*
+ * This contains a mutex, but it's pretty convenient to allow the mutex code to
+ * have a dependency on tsd.  So we define the struct here, and only refer to it
+ * by pointer in the header.
+ */
+struct tsd_init_head_s {
+	ql_head(tsd_init_block_t) blocks;
+	malloc_mutex_t lock;
+};
+
+pthread_key_t tsd_tsd;
+tsd_init_head_t	tsd_init_head = {
+	ql_head_initializer(blocks),
+	MALLOC_MUTEX_INITIALIZER
+};
+
+tsd_wrapper_t tsd_boot_wrapper = {
+	false,
+	TSD_INITIALIZER
+};
+bool tsd_booted = false;
+#endif
+
+JEMALLOC_DIAGNOSTIC_POP
+
+/******************************************************************************/
+
+/* A list of all the tsds in the nominal state. */
+typedef ql_head(tsd_t) tsd_list_t;
+static tsd_list_t tsd_nominal_tsds = ql_head_initializer(tsd_nominal_tsds);
+static malloc_mutex_t tsd_nominal_tsds_lock;
+
+/* How many slow-path-enabling features are turned on. */
+static atomic_u32_t tsd_global_slow_count = ATOMIC_INIT(0);
+
+static bool
+tsd_in_nominal_list(tsd_t *tsd) {
+	tsd_t *tsd_list;
+	bool found = false;
+	/*
+	 * We don't know that tsd is nominal; it might not be safe to get data
+	 * out of it here.
+	 */
+	malloc_mutex_lock(TSDN_NULL, &tsd_nominal_tsds_lock);
+	ql_foreach(tsd_list, &tsd_nominal_tsds, TSD_MANGLE(tcache).tsd_link) {
+		if (tsd == tsd_list) {
+			found = true;
+			break;
+		}
+	}
+	malloc_mutex_unlock(TSDN_NULL, &tsd_nominal_tsds_lock);
+	return found;
+}
+
+static void
+tsd_add_nominal(tsd_t *tsd) {
+	assert(!tsd_in_nominal_list(tsd));
+	assert(tsd_state_get(tsd) <= tsd_state_nominal_max);
+	ql_elm_new(tsd, TSD_MANGLE(tcache).tsd_link);
+	malloc_mutex_lock(tsd_tsdn(tsd), &tsd_nominal_tsds_lock);
+	ql_tail_insert(&tsd_nominal_tsds, tsd, TSD_MANGLE(tcache).tsd_link);
+	malloc_mutex_unlock(tsd_tsdn(tsd), &tsd_nominal_tsds_lock);
+}
+
+static void
+tsd_remove_nominal(tsd_t *tsd) {
+	assert(tsd_in_nominal_list(tsd));
+	assert(tsd_state_get(tsd) <= tsd_state_nominal_max);
+	malloc_mutex_lock(tsd_tsdn(tsd), &tsd_nominal_tsds_lock);
+	ql_remove(&tsd_nominal_tsds, tsd, TSD_MANGLE(tcache).tsd_link);
+	malloc_mutex_unlock(tsd_tsdn(tsd), &tsd_nominal_tsds_lock);
+}
+
+static void
+tsd_force_recompute(tsdn_t *tsdn) {
+	/*
+	 * The stores to tsd->state here need to synchronize with the exchange
+	 * in tsd_slow_update.
+	 */
+	atomic_fence(ATOMIC_RELEASE);
+	malloc_mutex_lock(tsdn, &tsd_nominal_tsds_lock);
+	tsd_t *remote_tsd;
+	ql_foreach(remote_tsd, &tsd_nominal_tsds, TSD_MANGLE(tcache).tsd_link) {
+		assert(tsd_atomic_load(&remote_tsd->state, ATOMIC_RELAXED)
+		    <= tsd_state_nominal_max);
+		tsd_atomic_store(&remote_tsd->state, tsd_state_nominal_recompute,
+		    ATOMIC_RELAXED);
+	}
+	malloc_mutex_unlock(tsdn, &tsd_nominal_tsds_lock);
+}
+
+void
+tsd_global_slow_inc(tsdn_t *tsdn) {
+	atomic_fetch_add_u32(&tsd_global_slow_count, 1, ATOMIC_RELAXED);
+	/*
+	 * We unconditionally force a recompute, even if the global slow count
+	 * was already positive.  If we didn't, then it would be possible for us
+	 * to return to the user, have the user synchronize externally with some
+	 * other thread, and then have that other thread not have picked up the
+	 * update yet (since the original incrementing thread might still be
+	 * making its way through the tsd list).
+	 */
+	tsd_force_recompute(tsdn);
+}
+
+void tsd_global_slow_dec(tsdn_t *tsdn) {
+	atomic_fetch_sub_u32(&tsd_global_slow_count, 1, ATOMIC_RELAXED);
+	/* See the note in ..._inc(). */
+	tsd_force_recompute(tsdn);
+}
+
+static bool
+tsd_local_slow(tsd_t *tsd) {
+	return !tsd_tcache_enabled_get(tsd)
+	    || tsd_reentrancy_level_get(tsd) > 0;
+}
+
+bool
+tsd_global_slow() {
+	return atomic_load_u32(&tsd_global_slow_count, ATOMIC_RELAXED) > 0;
+}
+
+/******************************************************************************/
+
+static uint8_t
+tsd_state_compute(tsd_t *tsd) {
+	if (!tsd_nominal(tsd)) {
+		return tsd_state_get(tsd);
+	}
+	/* We're in *a* nominal state; but which one? */
+	if (malloc_slow || tsd_local_slow(tsd) || tsd_global_slow()) {
+		return tsd_state_nominal_slow;
+	} else {
+		return tsd_state_nominal;
+	}
+}
+
+void
+tsd_slow_update(tsd_t *tsd) {
+	uint8_t old_state;
+	do {
+		uint8_t new_state = tsd_state_compute(tsd);
+		old_state = tsd_atomic_exchange(&tsd->state, new_state,
+		    ATOMIC_ACQUIRE);
+	} while (old_state == tsd_state_nominal_recompute);
+}
+
+void
+tsd_state_set(tsd_t *tsd, uint8_t new_state) {
+	/* Only the tsd module can change the state *to* recompute. */
+	assert(new_state != tsd_state_nominal_recompute);
+	uint8_t old_state = tsd_atomic_load(&tsd->state, ATOMIC_RELAXED);
+	if (old_state > tsd_state_nominal_max) {
+		/*
+		 * Not currently in the nominal list, but it might need to be
+		 * inserted there.
+		 */
+		assert(!tsd_in_nominal_list(tsd));
+		tsd_atomic_store(&tsd->state, new_state, ATOMIC_RELAXED);
+		if (new_state <= tsd_state_nominal_max) {
+			tsd_add_nominal(tsd);
+		}
+	} else {
+		/*
+		 * We're currently nominal.  If the new state is non-nominal,
+		 * great; we take ourselves off the list and just enter the new
+		 * state.
+		 */
+		assert(tsd_in_nominal_list(tsd));
+		if (new_state > tsd_state_nominal_max) {
+			tsd_remove_nominal(tsd);
+			tsd_atomic_store(&tsd->state, new_state,
+			    ATOMIC_RELAXED);
+		} else {
+			/*
+			 * This is the tricky case.  We're transitioning from
+			 * one nominal state to another.  The caller can't know
+			 * about any races that are occuring at the same time,
+			 * so we always have to recompute no matter what.
+			 */
+			tsd_slow_update(tsd);
+		}
+	}
+}
+
+static bool
+tsd_data_init(tsd_t *tsd) {
+	/*
+	 * We initialize the rtree context first (before the tcache), since the
+	 * tcache initialization depends on it.
+	 */
+	rtree_ctx_data_init(tsd_rtree_ctxp_get_unsafe(tsd));
+
+	/*
+	 * A nondeterministic seed based on the address of tsd reduces
+	 * the likelihood of lockstep non-uniform cache index
+	 * utilization among identical concurrent processes, but at the
+	 * cost of test repeatability.  For debug builds, instead use a
+	 * deterministic seed.
+	 */
+	*tsd_offset_statep_get(tsd) = config_debug ? 0 :
+	    (uint64_t)(uintptr_t)tsd;
+
+	return tsd_tcache_enabled_data_init(tsd);
+}
+
+static void
+assert_tsd_data_cleanup_done(tsd_t *tsd) {
+	assert(!tsd_nominal(tsd));
+	assert(!tsd_in_nominal_list(tsd));
+	assert(*tsd_arenap_get_unsafe(tsd) == NULL);
+	assert(*tsd_iarenap_get_unsafe(tsd) == NULL);
+	assert(*tsd_arenas_tdata_bypassp_get_unsafe(tsd) == true);
+	assert(*tsd_arenas_tdatap_get_unsafe(tsd) == NULL);
+	assert(*tsd_tcache_enabledp_get_unsafe(tsd) == false);
+	assert(*tsd_prof_tdatap_get_unsafe(tsd) == NULL);
+}
+
+static bool
+tsd_data_init_nocleanup(tsd_t *tsd) {
+	assert(tsd_state_get(tsd) == tsd_state_reincarnated ||
+	    tsd_state_get(tsd) == tsd_state_minimal_initialized);
+	/*
+	 * During reincarnation, there is no guarantee that the cleanup function
+	 * will be called (deallocation may happen after all tsd destructors).
+	 * We set up tsd in a way that no cleanup is needed.
+	 */
+	rtree_ctx_data_init(tsd_rtree_ctxp_get_unsafe(tsd));
+	*tsd_arenas_tdata_bypassp_get(tsd) = true;
+	*tsd_tcache_enabledp_get_unsafe(tsd) = false;
+	*tsd_reentrancy_levelp_get(tsd) = 1;
+	assert_tsd_data_cleanup_done(tsd);
+
+	return false;
+}
+
+tsd_t *
+tsd_fetch_slow(tsd_t *tsd, bool minimal) {
+	assert(!tsd_fast(tsd));
+
+	if (tsd_state_get(tsd) == tsd_state_nominal_slow) {
+		/*
+		 * On slow path but no work needed.  Note that we can't
+		 * necessarily *assert* that we're slow, because we might be
+		 * slow because of an asynchronous modification to global state,
+		 * which might be asynchronously modified *back*.
+		 */
+	} else if (tsd_state_get(tsd) == tsd_state_nominal_recompute) {
+		tsd_slow_update(tsd);
+	} else if (tsd_state_get(tsd) == tsd_state_uninitialized) {
+		if (!minimal) {
+			if (tsd_booted) {
+				tsd_state_set(tsd, tsd_state_nominal);
+				tsd_slow_update(tsd);
+				/* Trigger cleanup handler registration. */
+				tsd_set(tsd);
+				tsd_data_init(tsd);
+			}
+		} else {
+			tsd_state_set(tsd, tsd_state_minimal_initialized);
+			tsd_set(tsd);
+			tsd_data_init_nocleanup(tsd);
+		}
+	} else if (tsd_state_get(tsd) == tsd_state_minimal_initialized) {
+		if (!minimal) {
+			/* Switch to fully initialized. */
+			tsd_state_set(tsd, tsd_state_nominal);
+			assert(*tsd_reentrancy_levelp_get(tsd) >= 1);
+			(*tsd_reentrancy_levelp_get(tsd))--;
+			tsd_slow_update(tsd);
+			tsd_data_init(tsd);
+		} else {
+			assert_tsd_data_cleanup_done(tsd);
+		}
+	} else if (tsd_state_get(tsd) == tsd_state_purgatory) {
+		tsd_state_set(tsd, tsd_state_reincarnated);
+		tsd_set(tsd);
+		tsd_data_init_nocleanup(tsd);
+	} else {
+		assert(tsd_state_get(tsd) == tsd_state_reincarnated);
+	}
+
+	return tsd;
+}
+
+void *
+malloc_tsd_malloc(size_t size) {
+	return a0malloc(CACHELINE_CEILING(size));
+}
+
+void
+malloc_tsd_dalloc(void *wrapper) {
+	a0dalloc(wrapper);
+}
+
+#if defined(JEMALLOC_MALLOC_THREAD_CLEANUP) || defined(_WIN32)
+#ifndef _WIN32
+JEMALLOC_EXPORT
+#endif
+void
+_malloc_thread_cleanup(void) {
+	bool pending[MALLOC_TSD_CLEANUPS_MAX], again;
+	unsigned i;
+
+	for (i = 0; i < ncleanups; i++) {
+		pending[i] = true;
+	}
+
+	do {
+		again = false;
+		for (i = 0; i < ncleanups; i++) {
+			if (pending[i]) {
+				pending[i] = cleanups[i]();
+				if (pending[i]) {
+					again = true;
+				}
+			}
+		}
+	} while (again);
+}
+#endif
+
+void
+malloc_tsd_cleanup_register(bool (*f)(void)) {
+	assert(ncleanups < MALLOC_TSD_CLEANUPS_MAX);
+	cleanups[ncleanups] = f;
+	ncleanups++;
+}
+
+static void
+tsd_do_data_cleanup(tsd_t *tsd) {
+	prof_tdata_cleanup(tsd);
+	iarena_cleanup(tsd);
+	arena_cleanup(tsd);
+	arenas_tdata_cleanup(tsd);
+	tcache_cleanup(tsd);
+	witnesses_cleanup(tsd_witness_tsdp_get_unsafe(tsd));
+}
+
+void
+tsd_cleanup(void *arg) {
+	tsd_t *tsd = (tsd_t *)arg;
+
+	switch (tsd_state_get(tsd)) {
+	case tsd_state_uninitialized:
+		/* Do nothing. */
+		break;
+	case tsd_state_minimal_initialized:
+		/* This implies the thread only did free() in its life time. */
+		/* Fall through. */
+	case tsd_state_reincarnated:
+		/*
+		 * Reincarnated means another destructor deallocated memory
+		 * after the destructor was called.  Cleanup isn't required but
+		 * is still called for testing and completeness.
+		 */
+		assert_tsd_data_cleanup_done(tsd);
+		/* Fall through. */
+	case tsd_state_nominal:
+	case tsd_state_nominal_slow:
+		tsd_do_data_cleanup(tsd);
+		tsd_state_set(tsd, tsd_state_purgatory);
+		tsd_set(tsd);
+		break;
+	case tsd_state_purgatory:
+		/*
+		 * The previous time this destructor was called, we set the
+		 * state to tsd_state_purgatory so that other destructors
+		 * wouldn't cause re-creation of the tsd.  This time, do
+		 * nothing, and do not request another callback.
+		 */
+		break;
+	default:
+		not_reached();
+	}
+#ifdef JEMALLOC_JET
+	test_callback_t test_callback = *tsd_test_callbackp_get_unsafe(tsd);
+	int *data = tsd_test_datap_get_unsafe(tsd);
+	if (test_callback != NULL) {
+		test_callback(data);
+	}
+#endif
+}
+
+tsd_t *
+malloc_tsd_boot0(void) {
+	tsd_t *tsd;
+
+	ncleanups = 0;
+	if (malloc_mutex_init(&tsd_nominal_tsds_lock, "tsd_nominal_tsds_lock",
+	    WITNESS_RANK_OMIT, malloc_mutex_rank_exclusive)) {
+		return NULL;
+	}
+	if (tsd_boot0()) {
+		return NULL;
+	}
+	tsd = tsd_fetch();
+	*tsd_arenas_tdata_bypassp_get(tsd) = true;
+	return tsd;
+}
+
+void
+malloc_tsd_boot1(void) {
+	tsd_boot1();
+	tsd_t *tsd = tsd_fetch();
+	/* malloc_slow has been set properly.  Update tsd_slow. */
+	tsd_slow_update(tsd);
+	*tsd_arenas_tdata_bypassp_get(tsd) = false;
+}
+
+#ifdef _WIN32
+static BOOL WINAPI
+_tls_callback(HINSTANCE hinstDLL, DWORD fdwReason, LPVOID lpvReserved) {
+	switch (fdwReason) {
+#ifdef JEMALLOC_LAZY_LOCK
+	case DLL_THREAD_ATTACH:
+		isthreaded = true;
+		break;
+#endif
+	case DLL_THREAD_DETACH:
+		_malloc_thread_cleanup();
+		break;
+	default:
+		break;
+	}
+	return true;
+}
+
+/*
+ * We need to be able to say "read" here (in the "pragma section"), but have
+ * hooked "read". We won't read for the rest of the file, so we can get away
+ * with unhooking.
+ */
+#ifdef read
+#  undef read
+#endif
+
+#ifdef _MSC_VER
+#  ifdef _M_IX86
+#    pragma comment(linker, "/INCLUDE:__tls_used")
+#    pragma comment(linker, "/INCLUDE:_tls_callback")
+#  else
+#    pragma comment(linker, "/INCLUDE:_tls_used")
+#    pragma comment(linker, "/INCLUDE:" STRINGIFY(tls_callback) )
+#  endif
+#  pragma section(".CRT$XLY",long,read)
+#endif
+JEMALLOC_SECTION(".CRT$XLY") JEMALLOC_ATTR(used)
+BOOL	(WINAPI *const tls_callback)(HINSTANCE hinstDLL,
+    DWORD fdwReason, LPVOID lpvReserved) = _tls_callback;
+#endif
+
+#if (!defined(JEMALLOC_MALLOC_THREAD_CLEANUP) && !defined(JEMALLOC_TLS) && \
+    !defined(_WIN32))
+void *
+tsd_init_check_recursion(tsd_init_head_t *head, tsd_init_block_t *block) {
+	pthread_t self = pthread_self();
+	tsd_init_block_t *iter;
+
+	/* Check whether this thread has already inserted into the list. */
+	malloc_mutex_lock(TSDN_NULL, &head->lock);
+	ql_foreach(iter, &head->blocks, link) {
+		if (iter->thread == self) {
+			malloc_mutex_unlock(TSDN_NULL, &head->lock);
+			return iter->data;
+		}
+	}
+	/* Insert block into list. */
+	ql_elm_new(block, link);
+	block->thread = self;
+	ql_tail_insert(&head->blocks, block, link);
+	malloc_mutex_unlock(TSDN_NULL, &head->lock);
+	return NULL;
+}
+
+void
+tsd_init_finish(tsd_init_head_t *head, tsd_init_block_t *block) {
+	malloc_mutex_lock(TSDN_NULL, &head->lock);
+	ql_remove(&head->blocks, block, link);
+	malloc_mutex_unlock(TSDN_NULL, &head->lock);
+}
+#endif
+
+void
+tsd_prefork(tsd_t *tsd) {
+	malloc_mutex_prefork(tsd_tsdn(tsd), &tsd_nominal_tsds_lock);
+}
+
+void
+tsd_postfork_parent(tsd_t *tsd) {
+	malloc_mutex_postfork_parent(tsd_tsdn(tsd), &tsd_nominal_tsds_lock);
+}
+
+void
+tsd_postfork_child(tsd_t *tsd) {
+	malloc_mutex_postfork_child(tsd_tsdn(tsd), &tsd_nominal_tsds_lock);
+	ql_new(&tsd_nominal_tsds);
+
+	if (tsd_state_get(tsd) <= tsd_state_nominal_max) {
+		tsd_add_nominal(tsd);
+	}
+}

deps/jemalloc/src/util.c

@@ -0,0 +1,648 @@
+#define	assert(e) do {							\
+	if (config_debug && !(e)) {					\
+		malloc_write("<jemalloc>: Failed assertion\n");		\
+		abort();						\
+	}								\
+} while (0)
+
+#define	not_reached() do {						\
+	if (config_debug) {						\
+		malloc_write("<jemalloc>: Unreachable code reached\n");	\
+		abort();						\
+	}								\
+} while (0)
+
+#define	not_implemented() do {						\
+	if (config_debug) {						\
+		malloc_write("<jemalloc>: Not implemented\n");		\
+		abort();						\
+	}								\
+} while (0)
+
+#define	JEMALLOC_UTIL_C_
+#include "jemalloc/internal/jemalloc_internal.h"
+
+/******************************************************************************/
+/* Function prototypes for non-inline static functions. */
+
+static void	wrtmessage(void *cbopaque, const char *s);
+#define	U2S_BUFSIZE	((1U << (LG_SIZEOF_INTMAX_T + 3)) + 1)
+static char	*u2s(uintmax_t x, unsigned base, bool uppercase, char *s,
+    size_t *slen_p);
+#define	D2S_BUFSIZE	(1 + U2S_BUFSIZE)
+static char	*d2s(intmax_t x, char sign, char *s, size_t *slen_p);
+#define	O2S_BUFSIZE	(1 + U2S_BUFSIZE)
+static char	*o2s(uintmax_t x, bool alt_form, char *s, size_t *slen_p);
+#define	X2S_BUFSIZE	(2 + U2S_BUFSIZE)
+static char	*x2s(uintmax_t x, bool alt_form, bool uppercase, char *s,
+    size_t *slen_p);
+
+/******************************************************************************/
+
+/* malloc_message() setup. */
+static void
+wrtmessage(void *cbopaque, const char *s)
+{
+
+#ifdef SYS_write
+	/*
+	 * Use syscall(2) rather than write(2) when possible in order to avoid
+	 * the possibility of memory allocation within libc.  This is necessary
+	 * on FreeBSD; most operating systems do not have this problem though.
+	 */
+	UNUSED int result = syscall(SYS_write, STDERR_FILENO, s, strlen(s));
+#else
+	UNUSED int result = write(STDERR_FILENO, s, strlen(s));
+#endif
+}
+
+JEMALLOC_EXPORT void	(*je_malloc_message)(void *, const char *s);
+
+/*
+ * Wrapper around malloc_message() that avoids the need for
+ * je_malloc_message(...) throughout the code.
+ */
+void
+malloc_write(const char *s)
+{
+
+	if (je_malloc_message != NULL)
+		je_malloc_message(NULL, s);
+	else
+		wrtmessage(NULL, s);
+}
+
+/*
+ * glibc provides a non-standard strerror_r() when _GNU_SOURCE is defined, so
+ * provide a wrapper.
+ */
+int
+buferror(int err, char *buf, size_t buflen)
+{
+
+#ifdef _WIN32
+	FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM, NULL, GetLastError(), 0,
+	    (LPSTR)buf, buflen, NULL);
+	return (0);
+#elif defined(_GNU_SOURCE)
+	char *b = strerror_r(err, buf, buflen);
+	if (b != buf) {
+		strncpy(buf, b, buflen);
+		buf[buflen-1] = '\0';
+	}
+	return (0);
+#else
+	return (strerror_r(err, buf, buflen));
+#endif
+}
+
+uintmax_t
+malloc_strtoumax(const char *restrict nptr, char **restrict endptr, int base)
+{
+	uintmax_t ret, digit;
+	int b;
+	bool neg;
+	const char *p, *ns;
+
+	p = nptr;
+	if (base < 0 || base == 1 || base > 36) {
+		ns = p;
+		set_errno(EINVAL);
+		ret = UINTMAX_MAX;
+		goto label_return;
+	}
+	b = base;
+
+	/* Swallow leading whitespace and get sign, if any. */
+	neg = false;
+	while (true) {
+		switch (*p) {
+		case '\t': case '\n': case '\v': case '\f': case '\r': case ' ':
+			p++;
+			break;
+		case '-':
+			neg = true;
+			/* Fall through. */
+		case '+':
+			p++;
+			/* Fall through. */
+		default:
+			goto label_prefix;
+		}
+	}
+
+	/* Get prefix, if any. */
+	label_prefix:
+	/*
+	 * Note where the first non-whitespace/sign character is so that it is
+	 * possible to tell whether any digits are consumed (e.g., "  0" vs.
+	 * "  -x").
+	 */
+	ns = p;
+	if (*p == '0') {
+		switch (p[1]) {
+		case '0': case '1': case '2': case '3': case '4': case '5':
+		case '6': case '7':
+			if (b == 0)
+				b = 8;
+			if (b == 8)
+				p++;
+			break;
+		case 'X': case 'x':
+			switch (p[2]) {
+			case '0': case '1': case '2': case '3': case '4':
+			case '5': case '6': case '7': case '8': case '9':
+			case 'A': case 'B': case 'C': case 'D': case 'E':
+			case 'F':
+			case 'a': case 'b': case 'c': case 'd': case 'e':
+			case 'f':
+				if (b == 0)
+					b = 16;
+				if (b == 16)
+					p += 2;
+				break;
+			default:
+				break;
+			}
+			break;
+		default:
+			p++;
+			ret = 0;
+			goto label_return;
+		}
+	}
+	if (b == 0)
+		b = 10;
+
+	/* Convert. */
+	ret = 0;
+	while ((*p >= '0' && *p <= '9' && (digit = *p - '0') < b)
+	    || (*p >= 'A' && *p <= 'Z' && (digit = 10 + *p - 'A') < b)
+	    || (*p >= 'a' && *p <= 'z' && (digit = 10 + *p - 'a') < b)) {
+		uintmax_t pret = ret;
+		ret *= b;
+		ret += digit;
+		if (ret < pret) {
+			/* Overflow. */
+			set_errno(ERANGE);
+			ret = UINTMAX_MAX;
+			goto label_return;
+		}
+		p++;
+	}
+	if (neg)
+		ret = -ret;
+
+	if (p == ns) {
+		/* No conversion performed. */
+		set_errno(EINVAL);
+		ret = UINTMAX_MAX;
+		goto label_return;
+	}
+
+label_return:
+	if (endptr != NULL) {
+		if (p == ns) {
+			/* No characters were converted. */
+			*endptr = (char *)nptr;
+		} else
+			*endptr = (char *)p;
+	}
+	return (ret);
+}
+
+static char *
+u2s(uintmax_t x, unsigned base, bool uppercase, char *s, size_t *slen_p)
+{
+	unsigned i;
+
+	i = U2S_BUFSIZE - 1;
+	s[i] = '\0';
+	switch (base) {
+	case 10:
+		do {
+			i--;
+			s[i] = "0123456789"[x % (uint64_t)10];
+			x /= (uint64_t)10;
+		} while (x > 0);
+		break;
+	case 16: {
+		const char *digits = (uppercase)
+		    ? "0123456789ABCDEF"
+		    : "0123456789abcdef";
+
+		do {
+			i--;
+			s[i] = digits[x & 0xf];
+			x >>= 4;
+		} while (x > 0);
+		break;
+	} default: {
+		const char *digits = (uppercase)
+		    ? "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"
+		    : "0123456789abcdefghijklmnopqrstuvwxyz";
+
+		assert(base >= 2 && base <= 36);
+		do {
+			i--;
+			s[i] = digits[x % (uint64_t)base];
+			x /= (uint64_t)base;
+		} while (x > 0);
+	}}
+
+	*slen_p = U2S_BUFSIZE - 1 - i;
+	return (&s[i]);
+}
+
+static char *
+d2s(intmax_t x, char sign, char *s, size_t *slen_p)
+{
+	bool neg;
+
+	if ((neg = (x < 0)))
+		x = -x;
+	s = u2s(x, 10, false, s, slen_p);
+	if (neg)
+		sign = '-';
+	switch (sign) {
+	case '-':
+		if (neg == false)
+			break;
+		/* Fall through. */
+	case ' ':
+	case '+':
+		s--;
+		(*slen_p)++;
+		*s = sign;
+		break;
+	default: not_reached();
+	}
+	return (s);
+}
+
+static char *
+o2s(uintmax_t x, bool alt_form, char *s, size_t *slen_p)
+{
+
+	s = u2s(x, 8, false, s, slen_p);
+	if (alt_form && *s != '0') {
+		s--;
+		(*slen_p)++;
+		*s = '0';
+	}
+	return (s);
+}
+
+static char *
+x2s(uintmax_t x, bool alt_form, bool uppercase, char *s, size_t *slen_p)
+{
+
+	s = u2s(x, 16, uppercase, s, slen_p);
+	if (alt_form) {
+		s -= 2;
+		(*slen_p) += 2;
+		memcpy(s, uppercase ? "0X" : "0x", 2);
+	}
+	return (s);
+}
+
+int
+malloc_vsnprintf(char *str, size_t size, const char *format, va_list ap)
+{
+	int ret;
+	size_t i;
+	const char *f;
+
+#define	APPEND_C(c) do {						\
+	if (i < size)							\
+		str[i] = (c);						\
+	i++;								\
+} while (0)
+#define	APPEND_S(s, slen) do {						\
+	if (i < size) {							\
+		size_t cpylen = (slen <= size - i) ? slen : size - i;	\
+		memcpy(&str[i], s, cpylen);				\
+	}								\
+	i += slen;							\
+} while (0)
+#define	APPEND_PADDED_S(s, slen, width, left_justify) do {		\
+	/* Left padding. */						\
+	size_t pad_len = (width == -1) ? 0 : ((slen < (size_t)width) ?	\
+	    (size_t)width - slen : 0);					\
+	if (left_justify == false && pad_len != 0) {			\
+		size_t j;						\
+		for (j = 0; j < pad_len; j++)				\
+			APPEND_C(' ');					\
+	}								\
+	/* Value. */							\
+	APPEND_S(s, slen);						\
+	/* Right padding. */						\
+	if (left_justify && pad_len != 0) {				\
+		size_t j;						\
+		for (j = 0; j < pad_len; j++)				\
+			APPEND_C(' ');					\
+	}								\
+} while (0)
+#define	GET_ARG_NUMERIC(val, len) do {					\
+	switch (len) {							\
+	case '?':							\
+		val = va_arg(ap, int);					\
+		break;							\
+	case '?' | 0x80:						\
+		val = va_arg(ap, unsigned int);				\
+		break;							\
+	case 'l':							\
+		val = va_arg(ap, long);					\
+		break;							\
+	case 'l' | 0x80:						\
+		val = va_arg(ap, unsigned long);			\
+		break;							\
+	case 'q':							\
+		val = va_arg(ap, long long);				\
+		break;							\
+	case 'q' | 0x80:						\
+		val = va_arg(ap, unsigned long long);			\
+		break;							\
+	case 'j':							\
+		val = va_arg(ap, intmax_t);				\
+		break;							\
+	case 'j' | 0x80:						\
+		val = va_arg(ap, uintmax_t);				\
+		break;							\
+	case 't':							\
+		val = va_arg(ap, ptrdiff_t);				\
+		break;							\
+	case 'z':							\
+		val = va_arg(ap, ssize_t);				\
+		break;							\
+	case 'z' | 0x80:						\
+		val = va_arg(ap, size_t);				\
+		break;							\
+	case 'p': /* Synthetic; used for %p. */				\
+		val = va_arg(ap, uintptr_t);				\
+		break;							\
+	default: not_reached();						\
+	}								\
+} while (0)
+
+	i = 0;
+	f = format;
+	while (true) {
+		switch (*f) {
+		case '\0': goto label_out;
+		case '%': {
+			bool alt_form = false;
+			bool left_justify = false;
+			bool plus_space = false;
+			bool plus_plus = false;
+			int prec = -1;
+			int width = -1;
+			unsigned char len = '?';
+
+			f++;
+			/* Flags. */
+			while (true) {
+				switch (*f) {
+				case '#':
+					assert(alt_form == false);
+					alt_form = true;
+					break;
+				case '-':
+					assert(left_justify == false);
+					left_justify = true;
+					break;
+				case ' ':
+					assert(plus_space == false);
+					plus_space = true;
+					break;
+				case '+':
+					assert(plus_plus == false);
+					plus_plus = true;
+					break;
+				default: goto label_width;
+				}
+				f++;
+			}
+			/* Width. */
+			label_width:
+			switch (*f) {
+			case '*':
+				width = va_arg(ap, int);
+				f++;
+				if (width < 0) {
+					left_justify = true;
+					width = -width;
+				}
+				break;
+			case '0': case '1': case '2': case '3': case '4':
+			case '5': case '6': case '7': case '8': case '9': {
+				uintmax_t uwidth;
+				set_errno(0);
+				uwidth = malloc_strtoumax(f, (char **)&f, 10);
+				assert(uwidth != UINTMAX_MAX || get_errno() !=
+				    ERANGE);
+				width = (int)uwidth;
+				break;
+			} default:
+				break;
+			}
+			/* Width/precision separator. */
+			if (*f == '.')
+				f++;
+			else
+				goto label_length;
+			/* Precision. */
+			switch (*f) {
+			case '*':
+				prec = va_arg(ap, int);
+				f++;
+				break;
+			case '0': case '1': case '2': case '3': case '4':
+			case '5': case '6': case '7': case '8': case '9': {
+				uintmax_t uprec;
+				set_errno(0);
+				uprec = malloc_strtoumax(f, (char **)&f, 10);
+				assert(uprec != UINTMAX_MAX || get_errno() !=
+				    ERANGE);
+				prec = (int)uprec;
+				break;
+			}
+			default: break;
+			}
+			/* Length. */
+			label_length:
+			switch (*f) {
+			case 'l':
+				f++;
+				if (*f == 'l') {
+					len = 'q';
+					f++;
+				} else
+					len = 'l';
+				break;
+			case 'q': case 'j': case 't': case 'z':
+				len = *f;
+				f++;
+				break;
+			default: break;
+			}
+			/* Conversion specifier. */
+			switch (*f) {
+				char *s;
+				size_t slen;
+			case '%':
+				/* %% */
+				APPEND_C(*f);
+				f++;
+				break;
+			case 'd': case 'i': {
+				intmax_t val JEMALLOC_CC_SILENCE_INIT(0);
+				char buf[D2S_BUFSIZE];
+
+				GET_ARG_NUMERIC(val, len);
+				s = d2s(val, (plus_plus ? '+' : (plus_space ?
+				    ' ' : '-')), buf, &slen);
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			} case 'o': {
+				uintmax_t val JEMALLOC_CC_SILENCE_INIT(0);
+				char buf[O2S_BUFSIZE];
+
+				GET_ARG_NUMERIC(val, len | 0x80);
+				s = o2s(val, alt_form, buf, &slen);
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			} case 'u': {
+				uintmax_t val JEMALLOC_CC_SILENCE_INIT(0);
+				char buf[U2S_BUFSIZE];
+
+				GET_ARG_NUMERIC(val, len | 0x80);
+				s = u2s(val, 10, false, buf, &slen);
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			} case 'x': case 'X': {
+				uintmax_t val JEMALLOC_CC_SILENCE_INIT(0);
+				char buf[X2S_BUFSIZE];
+
+				GET_ARG_NUMERIC(val, len | 0x80);
+				s = x2s(val, alt_form, *f == 'X', buf, &slen);
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			} case 'c': {
+				unsigned char val;
+				char buf[2];
+
+				assert(len == '?' || len == 'l');
+				assert_not_implemented(len != 'l');
+				val = va_arg(ap, int);
+				buf[0] = val;
+				buf[1] = '\0';
+				APPEND_PADDED_S(buf, 1, width, left_justify);
+				f++;
+				break;
+			} case 's':
+				assert(len == '?' || len == 'l');
+				assert_not_implemented(len != 'l');
+				s = va_arg(ap, char *);
+				slen = (prec < 0) ? strlen(s) : prec;
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			case 'p': {
+				uintmax_t val;
+				char buf[X2S_BUFSIZE];
+
+				GET_ARG_NUMERIC(val, 'p');
+				s = x2s(val, true, false, buf, &slen);
+				APPEND_PADDED_S(s, slen, width, left_justify);
+				f++;
+				break;
+			} default: not_reached();
+			}
+			break;
+		} default: {
+			APPEND_C(*f);
+			f++;
+			break;
+		}}
+	}
+	label_out:
+	if (i < size)
+		str[i] = '\0';
+	else
+		str[size - 1] = '\0';
+	ret = i;
+
+#undef APPEND_C
+#undef APPEND_S
+#undef APPEND_PADDED_S
+#undef GET_ARG_NUMERIC
+	return (ret);
+}
+
+JEMALLOC_ATTR(format(printf, 3, 4))
+int
+malloc_snprintf(char *str, size_t size, const char *format, ...)
+{
+	int ret;
+	va_list ap;
+
+	va_start(ap, format);
+	ret = malloc_vsnprintf(str, size, format, ap);
+	va_end(ap);
+
+	return (ret);
+}
+
+void
+malloc_vcprintf(void (*write_cb)(void *, const char *), void *cbopaque,
+    const char *format, va_list ap)
+{
+	char buf[MALLOC_PRINTF_BUFSIZE];
+
+	if (write_cb == NULL) {
+		/*
+		 * The caller did not provide an alternate write_cb callback
+		 * function, so use the default one.  malloc_write() is an
+		 * inline function, so use malloc_message() directly here.
+		 */
+		write_cb = (je_malloc_message != NULL) ? je_malloc_message :
+		    wrtmessage;
+		cbopaque = NULL;
+	}
+
+	malloc_vsnprintf(buf, sizeof(buf), format, ap);
+	write_cb(cbopaque, buf);
+}
+
+/*
+ * Print to a callback function in such a way as to (hopefully) avoid memory
+ * allocation.
+ */
+JEMALLOC_ATTR(format(printf, 3, 4))
+void
+malloc_cprintf(void (*write_cb)(void *, const char *), void *cbopaque,
+    const char *format, ...)
+{
+	va_list ap;
+
+	va_start(ap, format);
+	malloc_vcprintf(write_cb, cbopaque, format, ap);
+	va_end(ap);
+}
+
+/* Print to stderr in such a way as to avoid memory allocation. */
+JEMALLOC_ATTR(format(printf, 1, 2))
+void
+malloc_printf(const char *format, ...)
+{
+	va_list ap;
+
+	va_start(ap, format);
+	malloc_vcprintf(NULL, NULL, format, ap);
+	va_end(ap);
+}

deps/jemalloc/src/witness.c

@@ -0,0 +1,100 @@
+#define JEMALLOC_WITNESS_C_
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/malloc_io.h"
+
+void
+witness_init(witness_t *witness, const char *name, witness_rank_t rank,
+    witness_comp_t *comp, void *opaque) {
+	witness->name = name;
+	witness->rank = rank;
+	witness->comp = comp;
+	witness->opaque = opaque;
+}
+
+static void
+witness_lock_error_impl(const witness_list_t *witnesses,
+    const witness_t *witness) {
+	witness_t *w;
+
+	malloc_printf("<jemalloc>: Lock rank order reversal:");
+	ql_foreach(w, witnesses, link) {
+		malloc_printf(" %s(%u)", w->name, w->rank);
+	}
+	malloc_printf(" %s(%u)\n", witness->name, witness->rank);
+	abort();
+}
+witness_lock_error_t *JET_MUTABLE witness_lock_error = witness_lock_error_impl;
+
+static void
+witness_owner_error_impl(const witness_t *witness) {
+	malloc_printf("<jemalloc>: Should own %s(%u)\n", witness->name,
+	    witness->rank);
+	abort();
+}
+witness_owner_error_t *JET_MUTABLE witness_owner_error =
+    witness_owner_error_impl;
+
+static void
+witness_not_owner_error_impl(const witness_t *witness) {
+	malloc_printf("<jemalloc>: Should not own %s(%u)\n", witness->name,
+	    witness->rank);
+	abort();
+}
+witness_not_owner_error_t *JET_MUTABLE witness_not_owner_error =
+    witness_not_owner_error_impl;
+
+static void
+witness_depth_error_impl(const witness_list_t *witnesses,
+    witness_rank_t rank_inclusive, unsigned depth) {
+	witness_t *w;
+
+	malloc_printf("<jemalloc>: Should own %u lock%s of rank >= %u:", depth,
+	    (depth != 1) ?  "s" : "", rank_inclusive);
+	ql_foreach(w, witnesses, link) {
+		malloc_printf(" %s(%u)", w->name, w->rank);
+	}
+	malloc_printf("\n");
+	abort();
+}
+witness_depth_error_t *JET_MUTABLE witness_depth_error =
+    witness_depth_error_impl;
+
+void
+witnesses_cleanup(witness_tsd_t *witness_tsd) {
+	witness_assert_lockless(witness_tsd_tsdn(witness_tsd));
+
+	/* Do nothing. */
+}
+
+void
+witness_prefork(witness_tsd_t *witness_tsd) {
+	if (!config_debug) {
+		return;
+	}
+	witness_tsd->forking = true;
+}
+
+void
+witness_postfork_parent(witness_tsd_t *witness_tsd) {
+	if (!config_debug) {
+		return;
+	}
+	witness_tsd->forking = false;
+}
+
+void
+witness_postfork_child(witness_tsd_t *witness_tsd) {
+	if (!config_debug) {
+		return;
+	}
+#ifndef JEMALLOC_MUTEX_INIT_CB
+	witness_list_t *witnesses;
+
+	witnesses = &witness_tsd->witnesses;
+	ql_new(witnesses);
+#endif
+	witness_tsd->forking = false;
+}

deps/jemalloc/src/zone.c

@@ -0,0 +1,469 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/assert.h"
+
+#ifndef JEMALLOC_ZONE
+#  error "This source file is for zones on Darwin (OS X)."
+#endif
+
+/* Definitions of the following structs in malloc/malloc.h might be too old
+ * for the built binary to run on newer versions of OSX. So use the newest
+ * possible version of those structs.
+ */
+typedef struct _malloc_zone_t {
+	void *reserved1;
+	void *reserved2;
+	size_t (*size)(struct _malloc_zone_t *, const void *);
+	void *(*malloc)(struct _malloc_zone_t *, size_t);
+	void *(*calloc)(struct _malloc_zone_t *, size_t, size_t);
+	void *(*valloc)(struct _malloc_zone_t *, size_t);
+	void (*free)(struct _malloc_zone_t *, void *);
+	void *(*realloc)(struct _malloc_zone_t *, void *, size_t);
+	void (*destroy)(struct _malloc_zone_t *);
+	const char *zone_name;
+	unsigned (*batch_malloc)(struct _malloc_zone_t *, size_t, void **, unsigned);
+	void (*batch_free)(struct _malloc_zone_t *, void **, unsigned);
+	struct malloc_introspection_t *introspect;
+	unsigned version;
+	void *(*memalign)(struct _malloc_zone_t *, size_t, size_t);
+	void (*free_definite_size)(struct _malloc_zone_t *, void *, size_t);
+	size_t (*pressure_relief)(struct _malloc_zone_t *, size_t);
+} malloc_zone_t;
+
+typedef struct {
+	vm_address_t address;
+	vm_size_t size;
+} vm_range_t;
+
+typedef struct malloc_statistics_t {
+	unsigned blocks_in_use;
+	size_t size_in_use;
+	size_t max_size_in_use;
+	size_t size_allocated;
+} malloc_statistics_t;
+
+typedef kern_return_t memory_reader_t(task_t, vm_address_t, vm_size_t, void **);
+
+typedef void vm_range_recorder_t(task_t, void *, unsigned type, vm_range_t *, unsigned);
+
+typedef struct malloc_introspection_t {
+	kern_return_t (*enumerator)(task_t, void *, unsigned, vm_address_t, memory_reader_t, vm_range_recorder_t);
+	size_t (*good_size)(malloc_zone_t *, size_t);
+	boolean_t (*check)(malloc_zone_t *);
+	void (*print)(malloc_zone_t *, boolean_t);
+	void (*log)(malloc_zone_t *, void *);
+	void (*force_lock)(malloc_zone_t *);
+	void (*force_unlock)(malloc_zone_t *);
+	void (*statistics)(malloc_zone_t *, malloc_statistics_t *);
+	boolean_t (*zone_locked)(malloc_zone_t *);
+	boolean_t (*enable_discharge_checking)(malloc_zone_t *);
+	boolean_t (*disable_discharge_checking)(malloc_zone_t *);
+	void (*discharge)(malloc_zone_t *, void *);
+#ifdef __BLOCKS__
+	void (*enumerate_discharged_pointers)(malloc_zone_t *, void (^)(void *, void *));
+#else
+	void *enumerate_unavailable_without_blocks;
+#endif
+	void (*reinit_lock)(malloc_zone_t *);
+} malloc_introspection_t;
+
+extern kern_return_t malloc_get_all_zones(task_t, memory_reader_t, vm_address_t **, unsigned *);
+
+extern malloc_zone_t *malloc_default_zone(void);
+
+extern void malloc_zone_register(malloc_zone_t *zone);
+
+extern void malloc_zone_unregister(malloc_zone_t *zone);
+
+/*
+ * The malloc_default_purgeable_zone() function is only available on >= 10.6.
+ * We need to check whether it is present at runtime, thus the weak_import.
+ */
+extern malloc_zone_t *malloc_default_purgeable_zone(void)
+JEMALLOC_ATTR(weak_import);
+
+/******************************************************************************/
+/* Data. */
+
+static malloc_zone_t *default_zone, *purgeable_zone;
+static malloc_zone_t jemalloc_zone;
+static struct malloc_introspection_t jemalloc_zone_introspect;
+static pid_t zone_force_lock_pid = -1;
+
+/******************************************************************************/
+/* Function prototypes for non-inline static functions. */
+
+static size_t	zone_size(malloc_zone_t *zone, const void *ptr);
+static void	*zone_malloc(malloc_zone_t *zone, size_t size);
+static void	*zone_calloc(malloc_zone_t *zone, size_t num, size_t size);
+static void	*zone_valloc(malloc_zone_t *zone, size_t size);
+static void	zone_free(malloc_zone_t *zone, void *ptr);
+static void	*zone_realloc(malloc_zone_t *zone, void *ptr, size_t size);
+static void	*zone_memalign(malloc_zone_t *zone, size_t alignment,
+    size_t size);
+static void	zone_free_definite_size(malloc_zone_t *zone, void *ptr,
+    size_t size);
+static void	zone_destroy(malloc_zone_t *zone);
+static unsigned	zone_batch_malloc(struct _malloc_zone_t *zone, size_t size,
+    void **results, unsigned num_requested);
+static void	zone_batch_free(struct _malloc_zone_t *zone,
+    void **to_be_freed, unsigned num_to_be_freed);
+static size_t	zone_pressure_relief(struct _malloc_zone_t *zone, size_t goal);
+static size_t	zone_good_size(malloc_zone_t *zone, size_t size);
+static kern_return_t	zone_enumerator(task_t task, void *data, unsigned type_mask,
+    vm_address_t zone_address, memory_reader_t reader,
+    vm_range_recorder_t recorder);
+static boolean_t	zone_check(malloc_zone_t *zone);
+static void	zone_print(malloc_zone_t *zone, boolean_t verbose);
+static void	zone_log(malloc_zone_t *zone, void *address);
+static void	zone_force_lock(malloc_zone_t *zone);
+static void	zone_force_unlock(malloc_zone_t *zone);
+static void	zone_statistics(malloc_zone_t *zone,
+    malloc_statistics_t *stats);
+static boolean_t	zone_locked(malloc_zone_t *zone);
+static void	zone_reinit_lock(malloc_zone_t *zone);
+
+/******************************************************************************/
+/*
+ * Functions.
+ */
+
+static size_t
+zone_size(malloc_zone_t *zone, const void *ptr) {
+	/*
+	 * There appear to be places within Darwin (such as setenv(3)) that
+	 * cause calls to this function with pointers that *no* zone owns.  If
+	 * we knew that all pointers were owned by *some* zone, we could split
+	 * our zone into two parts, and use one as the default allocator and
+	 * the other as the default deallocator/reallocator.  Since that will
+	 * not work in practice, we must check all pointers to assure that they
+	 * reside within a mapped extent before determining size.
+	 */
+	return ivsalloc(tsdn_fetch(), ptr);
+}
+
+static void *
+zone_malloc(malloc_zone_t *zone, size_t size) {
+	return je_malloc(size);
+}
+
+static void *
+zone_calloc(malloc_zone_t *zone, size_t num, size_t size) {
+	return je_calloc(num, size);
+}
+
+static void *
+zone_valloc(malloc_zone_t *zone, size_t size) {
+	void *ret = NULL; /* Assignment avoids useless compiler warning. */
+
+	je_posix_memalign(&ret, PAGE, size);
+
+	return ret;
+}
+
+static void
+zone_free(malloc_zone_t *zone, void *ptr) {
+	if (ivsalloc(tsdn_fetch(), ptr) != 0) {
+		je_free(ptr);
+		return;
+	}
+
+	free(ptr);
+}
+
+static void *
+zone_realloc(malloc_zone_t *zone, void *ptr, size_t size) {
+	if (ivsalloc(tsdn_fetch(), ptr) != 0) {
+		return je_realloc(ptr, size);
+	}
+
+	return realloc(ptr, size);
+}
+
+static void *
+zone_memalign(malloc_zone_t *zone, size_t alignment, size_t size) {
+	void *ret = NULL; /* Assignment avoids useless compiler warning. */
+
+	je_posix_memalign(&ret, alignment, size);
+
+	return ret;
+}
+
+static void
+zone_free_definite_size(malloc_zone_t *zone, void *ptr, size_t size) {
+	size_t alloc_size;
+
+	alloc_size = ivsalloc(tsdn_fetch(), ptr);
+	if (alloc_size != 0) {
+		assert(alloc_size == size);
+		je_free(ptr);
+		return;
+	}
+
+	free(ptr);
+}
+
+static void
+zone_destroy(malloc_zone_t *zone) {
+	/* This function should never be called. */
+	not_reached();
+}
+
+static unsigned
+zone_batch_malloc(struct _malloc_zone_t *zone, size_t size, void **results,
+    unsigned num_requested) {
+	unsigned i;
+
+	for (i = 0; i < num_requested; i++) {
+		results[i] = je_malloc(size);
+		if (!results[i])
+			break;
+	}
+
+	return i;
+}
+
+static void
+zone_batch_free(struct _malloc_zone_t *zone, void **to_be_freed,
+    unsigned num_to_be_freed) {
+	unsigned i;
+
+	for (i = 0; i < num_to_be_freed; i++) {
+		zone_free(zone, to_be_freed[i]);
+		to_be_freed[i] = NULL;
+	}
+}
+
+static size_t
+zone_pressure_relief(struct _malloc_zone_t *zone, size_t goal) {
+	return 0;
+}
+
+static size_t
+zone_good_size(malloc_zone_t *zone, size_t size) {
+	if (size == 0) {
+		size = 1;
+	}
+	return sz_s2u(size);
+}
+
+static kern_return_t
+zone_enumerator(task_t task, void *data, unsigned type_mask,
+    vm_address_t zone_address, memory_reader_t reader,
+    vm_range_recorder_t recorder) {
+	return KERN_SUCCESS;
+}
+
+static boolean_t
+zone_check(malloc_zone_t *zone) {
+	return true;
+}
+
+static void
+zone_print(malloc_zone_t *zone, boolean_t verbose) {
+}
+
+static void
+zone_log(malloc_zone_t *zone, void *address) {
+}
+
+static void
+zone_force_lock(malloc_zone_t *zone) {
+	if (isthreaded) {
+		/*
+		 * See the note in zone_force_unlock, below, to see why we need
+		 * this.
+		 */
+		assert(zone_force_lock_pid == -1);
+		zone_force_lock_pid = getpid();
+		jemalloc_prefork();
+	}
+}
+
+static void
+zone_force_unlock(malloc_zone_t *zone) {
+	/*
+	 * zone_force_lock and zone_force_unlock are the entry points to the
+	 * forking machinery on OS X.  The tricky thing is, the child is not
+	 * allowed to unlock mutexes locked in the parent, even if owned by the
+	 * forking thread (and the mutex type we use in OS X will fail an assert
+	 * if we try).  In the child, we can get away with reinitializing all
+	 * the mutexes, which has the effect of unlocking them.  In the parent,
+	 * doing this would mean we wouldn't wake any waiters blocked on the
+	 * mutexes we unlock.  So, we record the pid of the current thread in
+	 * zone_force_lock, and use that to detect if we're in the parent or
+	 * child here, to decide which unlock logic we need.
+	 */
+	if (isthreaded) {
+		assert(zone_force_lock_pid != -1);
+		if (getpid() == zone_force_lock_pid) {
+			jemalloc_postfork_parent();
+		} else {
+			jemalloc_postfork_child();
+		}
+		zone_force_lock_pid = -1;
+	}
+}
+
+static void
+zone_statistics(malloc_zone_t *zone, malloc_statistics_t *stats) {
+	/* We make no effort to actually fill the values */
+	stats->blocks_in_use = 0;
+	stats->size_in_use = 0;
+	stats->max_size_in_use = 0;
+	stats->size_allocated = 0;
+}
+
+static boolean_t
+zone_locked(malloc_zone_t *zone) {
+	/* Pretend no lock is being held */
+	return false;
+}
+
+static void
+zone_reinit_lock(malloc_zone_t *zone) {
+	/* As of OSX 10.12, this function is only used when force_unlock would
+	 * be used if the zone version were < 9. So just use force_unlock. */
+	zone_force_unlock(zone);
+}
+
+static void
+zone_init(void) {
+	jemalloc_zone.size = zone_size;
+	jemalloc_zone.malloc = zone_malloc;
+	jemalloc_zone.calloc = zone_calloc;
+	jemalloc_zone.valloc = zone_valloc;
+	jemalloc_zone.free = zone_free;
+	jemalloc_zone.realloc = zone_realloc;
+	jemalloc_zone.destroy = zone_destroy;
+	jemalloc_zone.zone_name = "jemalloc_zone";
+	jemalloc_zone.batch_malloc = zone_batch_malloc;
+	jemalloc_zone.batch_free = zone_batch_free;
+	jemalloc_zone.introspect = &jemalloc_zone_introspect;
+	jemalloc_zone.version = 9;
+	jemalloc_zone.memalign = zone_memalign;
+	jemalloc_zone.free_definite_size = zone_free_definite_size;
+	jemalloc_zone.pressure_relief = zone_pressure_relief;
+
+	jemalloc_zone_introspect.enumerator = zone_enumerator;
+	jemalloc_zone_introspect.good_size = zone_good_size;
+	jemalloc_zone_introspect.check = zone_check;
+	jemalloc_zone_introspect.print = zone_print;
+	jemalloc_zone_introspect.log = zone_log;
+	jemalloc_zone_introspect.force_lock = zone_force_lock;
+	jemalloc_zone_introspect.force_unlock = zone_force_unlock;
+	jemalloc_zone_introspect.statistics = zone_statistics;
+	jemalloc_zone_introspect.zone_locked = zone_locked;
+	jemalloc_zone_introspect.enable_discharge_checking = NULL;
+	jemalloc_zone_introspect.disable_discharge_checking = NULL;
+	jemalloc_zone_introspect.discharge = NULL;
+#ifdef __BLOCKS__
+	jemalloc_zone_introspect.enumerate_discharged_pointers = NULL;
+#else
+	jemalloc_zone_introspect.enumerate_unavailable_without_blocks = NULL;
+#endif
+	jemalloc_zone_introspect.reinit_lock = zone_reinit_lock;
+}
+
+static malloc_zone_t *
+zone_default_get(void) {
+	malloc_zone_t **zones = NULL;
+	unsigned int num_zones = 0;
+
+	/*
+	 * On OSX 10.12, malloc_default_zone returns a special zone that is not
+	 * present in the list of registered zones. That zone uses a "lite zone"
+	 * if one is present (apparently enabled when malloc stack logging is
+	 * enabled), or the first registered zone otherwise. In practice this
+	 * means unless malloc stack logging is enabled, the first registered
+	 * zone is the default.  So get the list of zones to get the first one,
+	 * instead of relying on malloc_default_zone.
+	 */
+	if (KERN_SUCCESS != malloc_get_all_zones(0, NULL,
+	    (vm_address_t**)&zones, &num_zones)) {
+		/*
+		 * Reset the value in case the failure happened after it was
+		 * set.
+		 */
+		num_zones = 0;
+	}
+
+	if (num_zones) {
+		return zones[0];
+	}
+
+	return malloc_default_zone();
+}
+
+/* As written, this function can only promote jemalloc_zone. */
+static void
+zone_promote(void) {
+	malloc_zone_t *zone;
+
+	do {
+		/*
+		 * Unregister and reregister the default zone.  On OSX >= 10.6,
+		 * unregistering takes the last registered zone and places it
+		 * at the location of the specified zone.  Unregistering the
+		 * default zone thus makes the last registered one the default.
+		 * On OSX < 10.6, unregistering shifts all registered zones.
+		 * The first registered zone then becomes the default.
+		 */
+		malloc_zone_unregister(default_zone);
+		malloc_zone_register(default_zone);
+
+		/*
+		 * On OSX 10.6, having the default purgeable zone appear before
+		 * the default zone makes some things crash because it thinks it
+		 * owns the default zone allocated pointers.  We thus
+		 * unregister/re-register it in order to ensure it's always
+		 * after the default zone.  On OSX < 10.6, there is no purgeable
+		 * zone, so this does nothing.  On OSX >= 10.6, unregistering
+		 * replaces the purgeable zone with the last registered zone
+		 * above, i.e. the default zone.  Registering it again then puts
+		 * it at the end, obviously after the default zone.
+		 */
+		if (purgeable_zone != NULL) {
+			malloc_zone_unregister(purgeable_zone);
+			malloc_zone_register(purgeable_zone);
+		}
+
+		zone = zone_default_get();
+	} while (zone != &jemalloc_zone);
+}
+
+JEMALLOC_ATTR(constructor)
+void
+zone_register(void) {
+	/*
+	 * If something else replaced the system default zone allocator, don't
+	 * register jemalloc's.
+	 */
+	default_zone = zone_default_get();
+	if (!default_zone->zone_name || strcmp(default_zone->zone_name,
+	    "DefaultMallocZone") != 0) {
+		return;
+	}
+
+	/*
+	 * The default purgeable zone is created lazily by OSX's libc.  It uses
+	 * the default zone when it is created for "small" allocations
+	 * (< 15 KiB), but assumes the default zone is a scalable_zone.  This
+	 * obviously fails when the default zone is the jemalloc zone, so
+	 * malloc_default_purgeable_zone() is called beforehand so that the
+	 * default purgeable zone is created when the default zone is still
+	 * a scalable_zone.  As purgeable zones only exist on >= 10.6, we need
+	 * to check for the existence of malloc_default_purgeable_zone() at
+	 * run time.
+	 */
+	purgeable_zone = (malloc_default_purgeable_zone == NULL) ? NULL :
+	    malloc_default_purgeable_zone();
+
+	/* Register the custom zone.  At this point it won't be the default. */
+	zone_init();
+	malloc_zone_register(&jemalloc_zone);
+
+	/* Promote the custom zone to be default. */
+	zone_promote();
+}