Unlike prof_sample which is supported only with profiling mode active, prof_threshold is intended to be an always-supported allocation callback with much less overhead. The usage of the threshold allows performance critical callers to change program execution based on the callback: e.g. drop caches when memory becomes high or to predict the program is about to OOM ahead of time using peak memory watermarks.

src/ctl.c

@@ -154,6 +154,7 @@ CTL_PROTO(opt_prof_active)
 CTL_PROTO(opt_prof_thread_active_init)
 CTL_PROTO(opt_prof_bt_max)
 CTL_PROTO(opt_lg_prof_sample)
+CTL_PROTO(opt_experimental_lg_prof_threshold)
 CTL_PROTO(opt_lg_prof_interval)
 CTL_PROTO(opt_prof_gdump)
 CTL_PROTO(opt_prof_final)
@@ -357,6 +358,7 @@ CTL_PROTO(experimental_hooks_prof_backtrace)
 CTL_PROTO(experimental_hooks_prof_dump)
 CTL_PROTO(experimental_hooks_prof_sample)
 CTL_PROTO(experimental_hooks_prof_sample_free)
+CTL_PROTO(experimental_hooks_prof_threshold)
 CTL_PROTO(experimental_hooks_safety_check_abort)
 CTL_PROTO(experimental_thread_activity_callback)
 CTL_PROTO(experimental_utilization_query)
@@ -539,6 +541,7 @@ static const ctl_named_node_t opt_node[] = {
 	{NAME("prof_thread_active_init"), CTL(opt_prof_thread_active_init)},
 	{NAME("prof_bt_max"), CTL(opt_prof_bt_max)},
 	{NAME("lg_prof_sample"), CTL(opt_lg_prof_sample)},
+	{NAME("experimental_lg_prof_threshold"), CTL(opt_experimental_lg_prof_threshold)},
 	{NAME("lg_prof_interval"), CTL(opt_lg_prof_interval)},
 	{NAME("prof_gdump"),	CTL(opt_prof_gdump)},
 	{NAME("prof_final"),	CTL(opt_prof_final)},
@@ -965,6 +968,7 @@ static const ctl_named_node_t experimental_hooks_node[] = {
 	{NAME("prof_dump"),	CTL(experimental_hooks_prof_dump)},
 	{NAME("prof_sample"),	CTL(experimental_hooks_prof_sample)},
 	{NAME("prof_sample_free"),	CTL(experimental_hooks_prof_sample_free)},
+	{NAME("prof_threshold"),	CTL(experimental_hooks_prof_threshold)},
 	{NAME("safety_check_abort"),	CTL(experimental_hooks_safety_check_abort)},
 };
 
@@ -2317,6 +2321,7 @@ CTL_RO_NL_CGEN(config_prof, opt_prof_thread_active_init,
     opt_prof_thread_active_init, bool)
 CTL_RO_NL_CGEN(config_prof, opt_prof_bt_max, opt_prof_bt_max, unsigned)
 CTL_RO_NL_CGEN(config_prof, opt_lg_prof_sample, opt_lg_prof_sample, size_t)
+CTL_RO_NL_CGEN(config_prof, opt_experimental_lg_prof_threshold, opt_experimental_lg_prof_threshold, size_t)
 CTL_RO_NL_CGEN(config_prof, opt_prof_accum, opt_prof_accum, bool)
 CTL_RO_NL_CGEN(config_prof, opt_prof_pid_namespace, opt_prof_pid_namespace,
     bool)
@@ -3778,6 +3783,32 @@ label_return:
 	return ret;
 }
 
+
+static int
+experimental_hooks_prof_threshold_ctl(tsd_t *tsd, const size_t *mib,
+    size_t miblen, void *oldp, size_t *oldlenp, void *newp, size_t newlen) {
+	int ret;
+
+	if (oldp == NULL && newp == NULL) {
+		ret = EINVAL;
+		goto label_return;
+	}
+	if (oldp != NULL) {
+		prof_threshold_hook_t old_hook =
+		    prof_threshold_hook_get();
+		READ(old_hook, prof_threshold_hook_t);
+	}
+	if (newp != NULL) {
+		prof_threshold_hook_t new_hook JEMALLOC_CC_SILENCE_INIT(NULL);
+		WRITE(new_hook, prof_threshold_hook_t);
+		prof_threshold_hook_set(new_hook);
+	}
+	ret = 0;
+label_return:
+	return ret;
+}
+
+
 /* For integration test purpose only.  No plan to move out of experimental. */
 static int
 experimental_hooks_safety_check_abort_ctl(tsd_t *tsd, const size_t *mib,

src/jemalloc.c

@@ -1619,6 +1619,10 @@ malloc_conf_init_helper(sc_data_t *sc_data, unsigned bin_shard_sizes[SC_NBINS],
 				    "lg_prof_sample", 0, (sizeof(uint64_t) << 3)
 				    - 1, CONF_DONT_CHECK_MIN, CONF_CHECK_MAX,
 				    true)
+				CONF_HANDLE_SIZE_T(opt_experimental_lg_prof_threshold,
+				    "experimental_lg_prof_threshold", 0, (sizeof(uint64_t) << 3)
+				    - 1, CONF_DONT_CHECK_MIN, CONF_CHECK_MAX,
+				    true)
 				CONF_HANDLE_BOOL(opt_prof_accum, "prof_accum")
 				CONF_HANDLE_UNSIGNED(opt_prof_bt_max, "prof_bt_max",
 				    1, PROF_BT_MAX_LIMIT, CONF_CHECK_MIN, CONF_CHECK_MAX,

src/prof_threshold.c

@@ -0,0 +1,57 @@
+#include "jemalloc/internal/jemalloc_preamble.h"
+#include "jemalloc/internal/jemalloc_internal_includes.h"
+
+#include "jemalloc/internal/activity_callback.h"
+#include "jemalloc/internal/prof_threshold.h"
+
+#include "jemalloc/internal/prof_externs.h"
+
+/*
+ * Update every 128MB by default.
+ */
+#define PROF_THRESHOLD_LG_WAIT_DEFAULT 27
+
+/* Logically a prof_threshold_hook_t. */
+static atomic_p_t prof_threshold_hook;
+size_t opt_experimental_lg_prof_threshold = PROF_THRESHOLD_LG_WAIT_DEFAULT;
+
+void
+prof_threshold_hook_set(prof_threshold_hook_t hook) {
+	atomic_store_p(&prof_threshold_hook, hook, ATOMIC_RELEASE);
+}
+
+prof_threshold_hook_t
+prof_threshold_hook_get(void) {
+	return (prof_threshold_hook_t)atomic_load_p(&prof_threshold_hook,
+	    ATOMIC_ACQUIRE);
+}
+
+/* Invoke callback for threshold reached */
+static void
+prof_threshold_update(tsd_t *tsd) {
+	prof_threshold_hook_t prof_threshold_hook = prof_threshold_hook_get();
+	if (prof_threshold_hook == NULL) {
+		return;
+        }
+	uint64_t alloc = tsd_thread_allocated_get(tsd);
+	uint64_t dalloc = tsd_thread_deallocated_get(tsd);
+	peak_t *peak = tsd_peakp_get(tsd);
+	pre_reentrancy(tsd, NULL);
+	prof_threshold_hook(alloc, dalloc, peak->cur_max);
+	post_reentrancy(tsd);
+}
+
+uint64_t
+prof_threshold_new_event_wait(tsd_t *tsd) {
+	return 1 << opt_experimental_lg_prof_threshold;
+}
+
+uint64_t
+prof_threshold_postponed_event_wait(tsd_t *tsd) {
+	return TE_MIN_START_WAIT;
+}
+
+void
+prof_threshold_event_handler(tsd_t *tsd, uint64_t elapsed) {
+	prof_threshold_update(tsd);
+}

src/thread_event.c

@@ -69,6 +69,11 @@ peak_dalloc_fetch_elapsed(tsd_t *tsd) {
 	return TE_INVALID_ELAPSED;
 }
 
+static uint64_t
+prof_threshold_fetch_elapsed(tsd_t *tsd) {
+	return TE_INVALID_ELAPSED;
+}
+
 /* Per event facilities done. */
 
 static bool