diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.cpp
index a6d2dc47f6b..aba8f693cf8 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.cpp
@@ -30,6 +30,8 @@
#include "gc/shenandoah/shenandoahFreeSet.hpp"
#include "gc/shenandoah/shenandoahHeap.inline.hpp"
#include "gc/shenandoah/shenandoahGeneration.hpp"
+#include "gc/shenandoah/shenandoahHeapRegion.inline.hpp"
+#include "gc/shenandoah/shenandoahOldGeneration.hpp"
#include "gc/shenandoah/shenandoahYoungGeneration.hpp"
#include "logging/log.hpp"
#include "logging/logTag.hpp"
@@ -56,9 +58,6 @@ const double ShenandoahAdaptiveHeuristics::HIGHEST_EXPECTED_AVAILABLE_AT_END = 0
const double ShenandoahAdaptiveHeuristics::MINIMUM_CONFIDENCE = 0.319; // 25%
const double ShenandoahAdaptiveHeuristics::MAXIMUM_CONFIDENCE = 3.291; // 99.9%
-// TODO: Provide comment here or remove if not used
-const uint ShenandoahAdaptiveHeuristics::MINIMUM_RESIZE_INTERVAL = 10;
-
ShenandoahAdaptiveHeuristics::ShenandoahAdaptiveHeuristics(ShenandoahGeneration* generation) :
ShenandoahHeuristics(generation),
_margin_of_error_sd(ShenandoahAdaptiveInitialConfidence),
@@ -110,10 +109,27 @@ void ShenandoahAdaptiveHeuristics::choose_collection_set_from_regiondata(Shenand
QuickSort::sort<RegionData>(data, (int)size, compare_by_garbage, false);
if (is_generational) {
+ for (size_t idx = 0; idx < size; idx++) {
+ ShenandoahHeapRegion* r = data[idx]._region;
+ if (cset->is_preselected(r->index())) {
+ assert(r->age() >= InitialTenuringThreshold, "Preselected regions must have tenure age");
+ // Entire region will be promoted, This region does not impact young-gen or old-gen evacuation reserve.
+ // This region has been pre-selected and its impact on promotion reserve is already accounted for.
+
+ // r->used() is r->garbage() + r->get_live_data_bytes()
+ // Since all live data in this region is being evacuated from young-gen, it is as if this memory
+ // is garbage insofar as young-gen is concerned. Counting this as garbage reduces the need to
+ // reclaim highly utilized young-gen regions just for the sake of finding min_garbage to reclaim
+ // within youn-gen memory.
+
+ cur_young_garbage += r->garbage();
+ cset->add_region(r);
+ }
+ }
if (is_global) {
size_t max_young_cset = (size_t) (heap->get_young_evac_reserve() / ShenandoahEvacWaste);
size_t young_cur_cset = 0;
- size_t max_old_cset = (size_t) (heap->get_old_evac_reserve() / ShenandoahEvacWaste);
+ size_t max_old_cset = (size_t) (heap->get_old_evac_reserve() / ShenandoahOldEvacWaste);
size_t old_cur_cset = 0;
size_t free_target = (capacity * ShenandoahMinFreeThreshold) / 100 + max_young_cset;
size_t min_garbage = (free_target > actual_free) ? (free_target - actual_free) : 0;
@@ -126,6 +142,9 @@ void ShenandoahAdaptiveHeuristics::choose_collection_set_from_regiondata(Shenand
for (size_t idx = 0; idx < size; idx++) {
ShenandoahHeapRegion* r = data[idx]._region;
+ if (cset->is_preselected(r->index())) {
+ continue;
+ }
bool add_region = false;
if (r->is_old()) {
size_t new_cset = old_cur_cset + r->get_live_data_bytes();
@@ -133,17 +152,6 @@ void ShenandoahAdaptiveHeuristics::choose_collection_set_from_regiondata(Shenand
add_region = true;
old_cur_cset = new_cset;
}
- } else if (cset->is_preselected(r->index())) {
- assert(r->age() >= InitialTenuringThreshold, "Preselected regions must have tenure age");
- // Entire region will be promoted, This region does not impact young-gen or old-gen evacuation reserve.
- // This region has been pre-selected and its impact on promotion reserve is already accounted for.
- add_region = true;
- // r->used() is r->garbage() + r->get_live_data_bytes()
- // Since all live data in this region is being evacuated from young-gen, it is as if this memory
- // is garbage insofar as young-gen is concerned. Counting this as garbage reduces the need to
- // reclaim highly utilized young-gen regions just for the sake of finding min_garbage to reclaim
- // within youn-gen memory.
- cur_young_garbage += r->used();
} else if (r->age() < InitialTenuringThreshold) {
size_t new_cset = young_cur_cset + r->get_live_data_bytes();
size_t region_garbage = r->garbage();
@@ -176,42 +184,29 @@ void ShenandoahAdaptiveHeuristics::choose_collection_set_from_regiondata(Shenand
for (size_t idx = 0; idx < size; idx++) {
ShenandoahHeapRegion* r = data[idx]._region;
- bool add_region = false;
-
- if (!r->is_old()) {
- if (cset->is_preselected(r->index())) {
- assert(r->age() >= InitialTenuringThreshold, "Preselected regions must have tenure age");
- // Entire region will be promoted, This region does not impact young-gen evacuation reserve. Memory has already
- // been set aside to hold evacuation results as advance_promotion_reserve.
- add_region = true;
- // Since all live data in this region is being evacuated from young-gen, it is as if this memory
- // is garbage insofar as young-gen is concerned. Counting this as garbage reduces the need to
- // reclaim highly utilized young-gen regions just for the sake of finding min_garbage to reclaim
- // within youn-gen memory
- cur_young_garbage += r->get_live_data_bytes();
- } else if (r->age() < InitialTenuringThreshold) {
- size_t new_cset = cur_cset + r->get_live_data_bytes();
- size_t region_garbage = r->garbage();
- size_t new_garbage = cur_young_garbage + region_garbage;
- bool add_regardless = (region_garbage > ignore_threshold) && (new_garbage < min_garbage);
- if ((new_cset <= max_cset) && (add_regardless || (region_garbage > garbage_threshold))) {
- add_region = true;
- cur_cset = new_cset;
- cur_young_garbage = new_garbage;
- }
- }
- // Note that we do not add aged regions if they were not pre-selected. The reason they were not preselected
- // is because there is not sufficient room in old-gen to hold their to-be-promoted live objects.
-
- if (add_region) {
+ if (cset->is_preselected(r->index())) {
+ continue;
+ }
+ if (r->age() < InitialTenuringThreshold) {
+ size_t new_cset = cur_cset + r->get_live_data_bytes();
+ size_t region_garbage = r->garbage();
+ size_t new_garbage = cur_young_garbage + region_garbage;
+ bool add_regardless = (region_garbage > ignore_threshold) && (new_garbage < min_garbage);
+ assert(r->is_young(), "Only young candidates expected in the data array");
+ if ((new_cset <= max_cset) && (add_regardless || (region_garbage > garbage_threshold))) {
+ cur_cset = new_cset;
+ cur_young_garbage = new_garbage;
cset->add_region(r);
}
}
+ // Note that we do not add aged regions if they were not pre-selected. The reason they were not preselected
+ // is because there is not sufficient room in old-gen to hold their to-be-promoted live objects or because
+ // they are to be promoted in place.
}
}
} else {
// Traditional Shenandoah (non-generational)
- size_t capacity = ShenandoahHeap::heap()->soft_max_capacity();
+ size_t capacity = ShenandoahHeap::heap()->max_capacity();
size_t max_cset = (size_t)((1.0 * capacity / 100 * ShenandoahEvacReserve) / ShenandoahEvacWaste);
size_t free_target = (capacity * ShenandoahMinFreeThreshold) / 100 + max_cset;
size_t min_garbage = (free_target > actual_free) ? (free_target - actual_free) : 0;
@@ -243,12 +238,21 @@ void ShenandoahAdaptiveHeuristics::choose_collection_set_from_regiondata(Shenand
}
}
}
+
+ size_t collected_old = cset->get_old_bytes_reserved_for_evacuation();
+ size_t collected_promoted = cset->get_young_bytes_to_be_promoted();
+ size_t collected_young = cset->get_young_bytes_reserved_for_evacuation();
+
+ log_info(gc, ergo)("Chosen CSet evacuates young: " SIZE_FORMAT "%s (of which at least: " SIZE_FORMAT "%s are to be promoted), "
+ "old: " SIZE_FORMAT "%s",
+ byte_size_in_proper_unit(collected_young), proper_unit_for_byte_size(collected_young),
+ byte_size_in_proper_unit(collected_promoted), proper_unit_for_byte_size(collected_promoted),
+ byte_size_in_proper_unit(collected_old), proper_unit_for_byte_size(collected_old));
}
void ShenandoahAdaptiveHeuristics::record_cycle_start() {
ShenandoahHeuristics::record_cycle_start();
_allocation_rate.allocation_counter_reset();
- ++_cycles_since_last_resize;
}
void ShenandoahAdaptiveHeuristics::record_success_concurrent(bool abbreviated) {
@@ -324,6 +328,84 @@ static double saturate(double value, double min, double max) {
return MAX2(MIN2(value, max), min);
}
+// Return a conservative estimate of how much memory can be allocated before we need to start GC. The estimate is based
+// on memory that is currently available within young generation plus all of the memory that will be added to the young
+// generation at the end of the current cycle (as represented by young_regions_to_be_reclaimed) and on the anticipated
+// amount of time required to perform a GC.
+size_t ShenandoahAdaptiveHeuristics::bytes_of_allocation_runway_before_gc_trigger(size_t young_regions_to_be_reclaimed) {
+ assert(_generation->is_young(), "Only meaningful for young-gen heuristic");
+
+ size_t max_capacity = _generation->max_capacity();
+ size_t capacity = _generation->soft_max_capacity();
+ size_t usage = _generation->used();
+ size_t available = (capacity > usage)? capacity - usage: 0;
+ size_t allocated = _generation->bytes_allocated_since_gc_start();
+
+ size_t available_young_collected = ShenandoahHeap::heap()->collection_set()->get_young_available_bytes_collected();
+ size_t anticipated_available =
+ available + young_regions_to_be_reclaimed * ShenandoahHeapRegion::region_size_bytes() - available_young_collected;
+ size_t allocation_headroom = anticipated_available;
+ size_t spike_headroom = capacity * ShenandoahAllocSpikeFactor / 100;
+ size_t penalties = capacity * _gc_time_penalties / 100;
+
+ double rate = _allocation_rate.sample(allocated);
+
+ // At what value of available, would avg and spike triggers occur?
+ // if allocation_headroom < avg_cycle_time * avg_alloc_rate, then we experience avg trigger
+ // if allocation_headroom < avg_cycle_time * rate, then we experience spike trigger if is_spiking
+ //
+ // allocation_headroom =
+ // 0, if penalties > available or if penalties + spike_headroom > available
+ // available - penalties - spike_headroom, otherwise
+ //
+ // so we trigger if available - penalties - spike_headroom < avg_cycle_time * avg_alloc_rate, which is to say
+ // available < avg_cycle_time * avg_alloc_rate + penalties + spike_headroom
+ // or if available < penalties + spike_headroom
+ //
+ // since avg_cycle_time * avg_alloc_rate > 0, the first test is sufficient to test both conditions
+ //
+ // thus, evac_slack_avg is MIN2(0, available - avg_cycle_time * avg_alloc_rate + penalties + spike_headroom)
+ //
+ // similarly, evac_slack_spiking is MIN2(0, available - avg_cycle_time * rate + penalties + spike_headroom)
+ // but evac_slack_spiking is only relevant if is_spiking, as defined below.
+
+ double avg_cycle_time = _gc_cycle_time_history->davg() + (_margin_of_error_sd * _gc_cycle_time_history->dsd());
+
+ // TODO: Consider making conservative adjustments to avg_cycle_time, such as: (avg_cycle_time *= 2) in cases where
+ // we expect a longer-than-normal GC duration. This includes mixed evacuations, evacuation that perform promotion
+ // including promotion in place, and OLD GC bootstrap cycles. It has been observed that these cycles sometimes
+ // require twice or more the duration of "normal" GC cycles. We have experimented with this approach. While it
+ // does appear to reduce the frequency of degenerated cycles due to late triggers, it also has the effect of reducing
+ // evacuation slack so that there is less memory available to be transferred to OLD. The result is that we
+ // throttle promotion and it takes too long to move old objects out of the young generation.
+
+ double avg_alloc_rate = _allocation_rate.upper_bound(_margin_of_error_sd);
+ size_t evac_slack_avg;
+ if (anticipated_available > avg_cycle_time * avg_alloc_rate + penalties + spike_headroom) {
+ evac_slack_avg = anticipated_available - (avg_cycle_time * avg_alloc_rate + penalties + spike_headroom);
+ } else {
+ // we have no slack because it's already time to trigger
+ evac_slack_avg = 0;
+ }
+
+ bool is_spiking = _allocation_rate.is_spiking(rate, _spike_threshold_sd);
+ size_t evac_slack_spiking;
+ if (is_spiking) {
+ if (anticipated_available > avg_cycle_time * rate + penalties + spike_headroom) {
+ evac_slack_spiking = anticipated_available - (avg_cycle_time * rate + penalties + spike_headroom);
+ } else {
+ // we have no slack because it's already time to trigger
+ evac_slack_spiking = 0;
+ }
+ } else {
+ evac_slack_spiking = evac_slack_avg;
+ }
+
+ size_t threshold = min_free_threshold();
+ size_t evac_min_threshold = (anticipated_available > threshold)? anticipated_available - threshold: 0;
+ return MIN3(evac_slack_spiking, evac_slack_avg, evac_min_threshold);
+}
+
bool ShenandoahAdaptiveHeuristics::should_start_gc() {
size_t capacity = _generation->soft_max_capacity();
size_t available = _generation->soft_available();
@@ -347,137 +429,144 @@ bool ShenandoahAdaptiveHeuristics::should_start_gc() {
double rate = _allocation_rate.sample(allocated);
_last_trigger = OTHER;
- size_t min_threshold = min_free_threshold();
-
- if (available < min_threshold) {
- log_info(gc)("Trigger (%s): Free (" SIZE_FORMAT "%s) is below minimum threshold (" SIZE_FORMAT "%s)",
- _generation->name(),
- byte_size_in_proper_unit(available), proper_unit_for_byte_size(available),
- byte_size_in_proper_unit(min_threshold), proper_unit_for_byte_size(min_threshold));
- return resize_and_evaluate();
- }
-
- // Check if we need to learn a bit about the application
- const size_t max_learn = ShenandoahLearningSteps;
- if (_gc_times_learned < max_learn) {
- size_t init_threshold = capacity / 100 * ShenandoahInitFreeThreshold;
- if (available < init_threshold) {
- log_info(gc)("Trigger (%s): Learning " SIZE_FORMAT " of " SIZE_FORMAT ". Free (" SIZE_FORMAT "%s) is below initial threshold (" SIZE_FORMAT "%s)",
- _generation->name(), _gc_times_learned + 1, max_learn,
- byte_size_in_proper_unit(available), proper_unit_for_byte_size(available),
- byte_size_in_proper_unit(init_threshold), proper_unit_for_byte_size(init_threshold));
+ // OLD generation is maintained to be as small as possible. Depletion-of-free-pool triggers do not apply to old generation.
+ if (!_generation->is_old()) {
+ size_t min_threshold = min_free_threshold();
+ if (available < min_threshold) {
+ log_info(gc)("Trigger (%s): Free (" SIZE_FORMAT "%s) is below minimum threshold (" SIZE_FORMAT "%s)",
+ _generation->name(),
+ byte_size_in_proper_unit(available), proper_unit_for_byte_size(available),
+ byte_size_in_proper_unit(min_threshold), proper_unit_for_byte_size(min_threshold));
return true;
}
- }
- // Rationale:
- // The idea is that there is an average allocation rate and there are occasional abnormal bursts (or spikes) of
- // allocations that exceed the average allocation rate. What do these spikes look like?
- //
- // 1. At certain phase changes, we may discard large amounts of data and replace it with large numbers of newly
- // allocated objects. This "spike" looks more like a phase change. We were in steady state at M bytes/sec
- // allocation rate and now we're in a "reinitialization phase" that looks like N bytes/sec. We need the "spike"
- // accommodation to give us enough runway to recalibrate our "average allocation rate".
- //
- // 2. The typical workload changes. "Suddenly", our typical workload of N TPS increases to N+delta TPS. This means
- // our average allocation rate needs to be adjusted. Once again, we need the "spike" accomodation to give us
- // enough runway to recalibrate our "average allocation rate".
- //
- // 3. Though there is an "average" allocation rate, a given workload's demand for allocation may be very bursty. We
- // allocate a bunch of LABs during the 5 ms that follow completion of a GC, then we perform no more allocations for
- // the next 150 ms. It seems we want the "spike" to represent the maximum divergence from average within the
- // period of time between consecutive evaluation of the should_start_gc() service. Here's the thinking:
- //
- // a) Between now and the next time I ask whether should_start_gc(), we might experience a spike representing
- // the anticipated burst of allocations. If that would put us over budget, then we should start GC immediately.
- // b) Between now and the anticipated depletion of allocation pool, there may be two or more bursts of allocations.
- // If there are more than one of these bursts, we can "approximate" that these will be separated by spans of
- // time with very little or no allocations so the "average" allocation rate should be a suitable approximation
- // of how this will behave.
- //
- // For cases 1 and 2, we need to "quickly" recalibrate the average allocation rate whenever we detect a change
- // in operation mode. We want some way to decide that the average rate has changed. Make average allocation rate
- // computations an independent effort.
-
-
- // TODO: Account for inherent delays in responding to GC triggers
- // 1. It has been observed that delays of 200 ms or greater are common between the moment we return true from should_start_gc()
- // and the moment at which we begin execution of the concurrent reset phase. Add this time into the calculation of
- // avg_cycle_time below. (What is "this time"? Perhaps we should remember recent history of this delay for the
- // running workload and use the maximum delay recently seen for "this time".)
- // 2. The frequency of inquiries to should_start_gc() is adaptive, ranging between ShenandoahControlIntervalMin and
- // ShenandoahControlIntervalMax. The current control interval (or the max control interval) should also be added into
- // the calculation of avg_cycle_time below.
-
- // Check if allocation headroom is still okay. This also factors in:
- // 1. Some space to absorb allocation spikes (ShenandoahAllocSpikeFactor)
- // 2. Accumulated penalties from Degenerated and Full GC
- size_t allocation_headroom = available;
- size_t spike_headroom = capacity / 100 * ShenandoahAllocSpikeFactor;
- size_t penalties = capacity / 100 * _gc_time_penalties;
-
- allocation_headroom -= MIN2(allocation_headroom, penalties);
- allocation_headroom -= MIN2(allocation_headroom, spike_headroom);
+ // Check if we need to learn a bit about the application
+ const size_t max_learn = ShenandoahLearningSteps;
+ if (_gc_times_learned < max_learn) {
+ size_t init_threshold = capacity / 100 * ShenandoahInitFreeThreshold;
+ if (available < init_threshold) {
+ log_info(gc)("Trigger (%s): Learning " SIZE_FORMAT " of " SIZE_FORMAT ". Free ("
+ SIZE_FORMAT "%s) is below initial threshold (" SIZE_FORMAT "%s)",
+ _generation->name(), _gc_times_learned + 1, max_learn,
+ byte_size_in_proper_unit(available), proper_unit_for_byte_size(available),
+ byte_size_in_proper_unit(init_threshold), proper_unit_for_byte_size(init_threshold));
+ return true;
+ }
+ }
- double avg_cycle_time = _gc_cycle_time_history->davg() + (_margin_of_error_sd * _gc_cycle_time_history->dsd());
+ // Rationale:
+ // The idea is that there is an average allocation rate and there are occasional abnormal bursts (or spikes) of
+ // allocations that exceed the average allocation rate. What do these spikes look like?
+ //
+ // 1. At certain phase changes, we may discard large amounts of data and replace it with large numbers of newly
+ // allocated objects. This "spike" looks more like a phase change. We were in steady state at M bytes/sec
+ // allocation rate and now we're in a "reinitialization phase" that looks like N bytes/sec. We need the "spike"
+ // accomodation to give us enough runway to recalibrate our "average allocation rate".
+ //
+ // 2. The typical workload changes. "Suddenly", our typical workload of N TPS increases to N+delta TPS. This means
+ // our average allocation rate needs to be adjusted. Once again, we need the "spike" accomodation to give us
+ // enough runway to recalibrate our "average allocation rate".
+ //
+ // 3. Though there is an "average" allocation rate, a given workload's demand for allocation may be very bursty. We
+ // allocate a bunch of LABs during the 5 ms that follow completion of a GC, then we perform no more allocations for
+ // the next 150 ms. It seems we want the "spike" to represent the maximum divergence from average within the
+ // period of time between consecutive evaluation of the should_start_gc() service. Here's the thinking:
+ //
+ // a) Between now and the next time I ask whether should_start_gc(), we might experience a spike representing
+ // the anticipated burst of allocations. If that would put us over budget, then we should start GC immediately.
+ // b) Between now and the anticipated depletion of allocation pool, there may be two or more bursts of allocations.
+ // If there are more than one of these bursts, we can "approximate" that these will be separated by spans of
+ // time with very little or no allocations so the "average" allocation rate should be a suitable approximation
+ // of how this will behave.
+ //
+ // For cases 1 and 2, we need to "quickly" recalibrate the average allocation rate whenever we detect a change
+ // in operation mode. We want some way to decide that the average rate has changed. Make average allocation rate
+ // computations an independent effort.
- double avg_alloc_rate = _allocation_rate.upper_bound(_margin_of_error_sd);
- log_debug(gc)("%s: average GC time: %.2f ms, allocation rate: %.0f %s/s",
- _generation->name(), avg_cycle_time * 1000,
- byte_size_in_proper_unit(avg_alloc_rate), proper_unit_for_byte_size(avg_alloc_rate));
-
- if (avg_cycle_time > allocation_headroom / avg_alloc_rate) {
- log_info(gc)("Trigger (%s): Average GC time (%.2f ms) is above the time for average allocation rate (%.0f %sB/s) to deplete free headroom (" SIZE_FORMAT "%s) (margin of error = %.2f)",
- _generation->name(), avg_cycle_time * 1000,
- byte_size_in_proper_unit(avg_alloc_rate), proper_unit_for_byte_size(avg_alloc_rate),
- byte_size_in_proper_unit(allocation_headroom), proper_unit_for_byte_size(allocation_headroom),
- _margin_of_error_sd);
-
- log_info(gc, ergo)("Free headroom: " SIZE_FORMAT "%s (free) - " SIZE_FORMAT "%s (spike) - " SIZE_FORMAT "%s (penalties) = " SIZE_FORMAT "%s",
- byte_size_in_proper_unit(available), proper_unit_for_byte_size(available),
- byte_size_in_proper_unit(spike_headroom), proper_unit_for_byte_size(spike_headroom),
- byte_size_in_proper_unit(penalties), proper_unit_for_byte_size(penalties),
- byte_size_in_proper_unit(allocation_headroom), proper_unit_for_byte_size(allocation_headroom));
-
- _last_trigger = RATE;
- return resize_and_evaluate();
- }
- bool is_spiking = _allocation_rate.is_spiking(rate, _spike_threshold_sd);
- if (is_spiking && avg_cycle_time > allocation_headroom / rate) {
- log_info(gc)("Trigger (%s): Average GC time (%.2f ms) is above the time for instantaneous allocation rate (%.0f %sB/s) to deplete free headroom (" SIZE_FORMAT "%s) (spike threshold = %.2f)",
- _generation->name(), avg_cycle_time * 1000,
- byte_size_in_proper_unit(rate), proper_unit_for_byte_size(rate),
- byte_size_in_proper_unit(allocation_headroom), proper_unit_for_byte_size(allocation_headroom),
- _spike_threshold_sd);
- _last_trigger = SPIKE;
- return resize_and_evaluate();
- }
+ // Check if allocation headroom is still okay. This also factors in:
+ // 1. Some space to absorb allocation spikes (ShenandoahAllocSpikeFactor)
+ // 2. Accumulated penalties from Degenerated and Full GC
- return ShenandoahHeuristics::should_start_gc();
-}
+ size_t allocation_headroom = available;
+ size_t spike_headroom = capacity / 100 * ShenandoahAllocSpikeFactor;
+ size_t penalties = capacity / 100 * _gc_time_penalties;
-bool ShenandoahAdaptiveHeuristics::resize_and_evaluate() {
- ShenandoahHeap* heap = ShenandoahHeap::heap();
- if (!heap->mode()->is_generational()) {
- // We only attempt to resize the generations in generational mode.
- return true;
- }
+ allocation_headroom -= MIN2(allocation_headroom, penalties);
+ allocation_headroom -= MIN2(allocation_headroom, spike_headroom);
- if (_cycles_since_last_resize <= MINIMUM_RESIZE_INTERVAL) {
- log_info(gc, ergo)("Not resizing %s for another " UINT32_FORMAT " cycles",
- _generation->name(), _cycles_since_last_resize);
- return true;
- }
+ double avg_cycle_time = _gc_cycle_time_history->davg() + (_margin_of_error_sd * _gc_cycle_time_history->dsd());
+ double avg_alloc_rate = _allocation_rate.upper_bound(_margin_of_error_sd);
+ log_debug(gc)("%s: average GC time: %.2f ms, allocation rate: %.0f %s/s",
+ _generation->name(),
+ avg_cycle_time * 1000, byte_size_in_proper_unit(avg_alloc_rate), proper_unit_for_byte_size(avg_alloc_rate));
- if (!heap->generation_sizer()->transfer_capacity(_generation)) {
- // We could not enlarge our generation, so we must start a gc cycle.
- log_info(gc, ergo)("Could not increase size of %s, begin gc cycle", _generation->name());
- return true;
- }
+ if (avg_cycle_time > allocation_headroom / avg_alloc_rate) {
+
+ log_info(gc)("Trigger (%s): Average GC time (%.2f ms) is above the time for average allocation rate (%.0f %sB/s)"
+ " to deplete free headroom (" SIZE_FORMAT "%s) (margin of error = %.2f)",
+ _generation->name(), avg_cycle_time * 1000,
+ byte_size_in_proper_unit(avg_alloc_rate), proper_unit_for_byte_size(avg_alloc_rate),
+ byte_size_in_proper_unit(allocation_headroom), proper_unit_for_byte_size(allocation_headroom),
+ _margin_of_error_sd);
+
+ log_info(gc, ergo)("Free headroom: " SIZE_FORMAT "%s (free) - " SIZE_FORMAT "%s (spike) - "
+ SIZE_FORMAT "%s (penalties) = " SIZE_FORMAT "%s",
+ byte_size_in_proper_unit(available), proper_unit_for_byte_size(available),
+ byte_size_in_proper_unit(spike_headroom), proper_unit_for_byte_size(spike_headroom),
+ byte_size_in_proper_unit(penalties), proper_unit_for_byte_size(penalties),
+ byte_size_in_proper_unit(allocation_headroom), proper_unit_for_byte_size(allocation_headroom));
- log_info(gc)("Increased size of %s generation, re-evaluate trigger criteria", _generation->name());
- return should_start_gc();
+ _last_trigger = RATE;
+ return true;
+ }
+
+ bool is_spiking = _allocation_rate.is_spiking(rate, _spike_threshold_sd);
+ if (is_spiking && avg_cycle_time > allocation_headroom / rate) {
+ log_info(gc)("Trigger (%s): Average GC time (%.2f ms) is above the time for instantaneous allocation rate (%.0f %sB/s)"
+ " to deplete free headroom (" SIZE_FORMAT "%s) (spike threshold = %.2f)",
+ _generation->name(), avg_cycle_time * 1000,
+ byte_size_in_proper_unit(rate), proper_unit_for_byte_size(rate),
+ byte_size_in_proper_unit(allocation_headroom), proper_unit_for_byte_size(allocation_headroom),
+ _spike_threshold_sd);
+ _last_trigger = SPIKE;
+ return true;
+ }
+
+ ShenandoahHeap* heap = ShenandoahHeap::heap();
+ if (heap->mode()->is_generational()) {
+ // Get through promotions and mixed evacuations as quickly as possible. These cycles sometimes require significantly
+ // more time than traditional young-generation cycles so start them up as soon as possible. This is a "mitigation"
+ // for the reality that old-gen and young-gen activities are not truly "concurrent". If there is old-gen work to
+ // be done, we start up the young-gen GC threads so they can do some of this old-gen work. As implemented, promotion
+ // gets priority over old-gen marking.
+
+ size_t promo_potential = heap->get_promotion_potential();
+ size_t promo_in_place_potential = heap->get_promotion_in_place_potential();
+ ShenandoahOldHeuristics* old_heuristics = (ShenandoahOldHeuristics*) heap->old_generation()->heuristics();
+ size_t mixed_candidates = old_heuristics->unprocessed_old_collection_candidates();
+ if (promo_potential > 0) {
+ // Detect unsigned arithmetic underflow
+ assert(promo_potential < heap->capacity(), "Sanity");
+ log_info(gc)("Trigger (%s): expedite promotion of " SIZE_FORMAT "%s",
+ _generation->name(), byte_size_in_proper_unit(promo_potential), proper_unit_for_byte_size(promo_potential));
+ return true;
+ } else if (promo_in_place_potential > 0) {
+ // Detect unsigned arithmetic underflow
+ assert(promo_in_place_potential < heap->capacity(), "Sanity");
+ log_info(gc)("Trigger (%s): expedite promotion in place of " SIZE_FORMAT "%s", _generation->name(),
+ byte_size_in_proper_unit(promo_in_place_potential),
+ proper_unit_for_byte_size(promo_in_place_potential));
+ return true;
+ } else if (mixed_candidates > 0) {
+ // We need to run young GC in order to open up some free heap regions so we can finish mixed evacuations.
+ log_info(gc)("Trigger (%s): expedite mixed evacuation of " SIZE_FORMAT " regions",
+ _generation->name(), mixed_candidates);
+ return true;
+ }
+ }
+ }
+ return ShenandoahHeuristics::should_start_gc();
}
void ShenandoahAdaptiveHeuristics::adjust_last_trigger_parameters(double amount) {
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.hpp
index 85b563f146d..494173751ea 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.hpp
@@ -39,7 +39,6 @@ class ShenandoahAllocationRate : public CHeapObj<mtGC> {
double upper_bound(double sds) const;
bool is_spiking(double rate, double threshold) const;
-
private:
double instantaneous_rate(double time, size_t allocated) const;
@@ -72,6 +71,8 @@ class ShenandoahAdaptiveHeuristics : public ShenandoahHeuristics {
virtual bool is_diagnostic() { return false; }
virtual bool is_experimental() { return false; }
+ virtual size_t bytes_of_allocation_runway_before_gc_trigger(size_t young_regions_to_be_recycled);
+
private:
// These are used to adjust the margin of error and the spike threshold
// in response to GC cycle outcomes. These values are shared, but the
@@ -85,13 +86,6 @@ class ShenandoahAdaptiveHeuristics : public ShenandoahHeuristics {
const static double LOWEST_EXPECTED_AVAILABLE_AT_END;
const static double HIGHEST_EXPECTED_AVAILABLE_AT_END;
- // At least this many cycles must execute before the heuristic will attempt
- // to resize its generation. This is to prevent the heuristic from rapidly
- // maxing out the generation size (which only forces the collector for the
- // other generation to run more frequently, defeating the purpose of improving
- // MMU).
- const static uint MINIMUM_RESIZE_INTERVAL;
-
friend class ShenandoahAllocationRate;
// Used to record the last trigger that signaled to start a GC.
@@ -106,8 +100,6 @@ class ShenandoahAdaptiveHeuristics : public ShenandoahHeuristics {
void adjust_margin_of_error(double amount);
void adjust_spike_threshold(double amount);
- bool resize_and_evaluate();
-
ShenandoahAllocationRate _allocation_rate;
// The margin of error expressed in standard deviations to add to our
@@ -135,10 +127,6 @@ class ShenandoahAdaptiveHeuristics : public ShenandoahHeuristics {
// establishes what is 'normal' for the application and is used as a
// source of feedback to adjust trigger parameters.
TruncatedSeq _available;
-
- // Do not attempt to resize the generation for this heuristic until this
- // value is greater than MINIMUM_RESIZE_INTERVAL.
- uint _cycles_since_last_resize;
};
#endif // SHARE_GC_SHENANDOAH_HEURISTICS_SHENANDOAHADAPTIVEHEURISTICS_HPP
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.cpp
index a84c23b678c..250d1317306 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.cpp
@@ -49,7 +49,8 @@ ShenandoahCompactHeuristics::ShenandoahCompactHeuristics(ShenandoahGeneration* g
bool ShenandoahCompactHeuristics::should_start_gc() {
size_t max_capacity = _generation->max_capacity();
size_t capacity = _generation->soft_max_capacity();
- size_t available = _generation->available();
+ size_t usage = _generation->used();
+ size_t available = (capacity > usage)? capacity - usage: 0;
// Make sure the code below treats available without the soft tail.
size_t soft_tail = max_capacity - capacity;
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.cpp
index ec0c365b873..93a4f414325 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.cpp
@@ -39,11 +39,22 @@
#include "logging/log.hpp"
#include "logging/logTag.hpp"
#include "runtime/globals_extension.hpp"
+#include "utilities/quickSort.hpp"
+// sort by decreasing garbage (so most garbage comes first)
int ShenandoahHeuristics::compare_by_garbage(RegionData a, RegionData b) {
- if (a._garbage > b._garbage)
+ if (a._u._garbage > b._u._garbage)
return -1;
- else if (a._garbage < b._garbage)
+ else if (a._u._garbage < b._u._garbage)
+ return 1;
+ else return 0;
+}
+
+// sort by increasing live (so least live comes first)
+int ShenandoahHeuristics::compare_by_live(RegionData a, RegionData b) {
+ if (a._u._live_data < b._u._live_data)
+ return -1;
+ else if (a._u._live_data > b._u._live_data)
return 1;
else return 0;
}
@@ -76,29 +87,170 @@ ShenandoahHeuristics::~ShenandoahHeuristics() {
FREE_C_HEAP_ARRAY(RegionGarbage, _region_data);
}
-size_t ShenandoahHeuristics::select_aged_regions(size_t old_available, size_t num_regions, bool* preselected_regions) {
+typedef struct {
+ ShenandoahHeapRegion* _region;
+ size_t _live_data;
+} AgedRegionData;
+
+static int compare_by_aged_live(AgedRegionData a, AgedRegionData b) {
+ if (a._live_data < b._live_data)
+ return -1;
+ else if (a._live_data > b._live_data)
+ return 1;
+ else return 0;
+}
+
+// Preselect for inclusion into the collection set regions whose age is at or above tenure age which contain more than
+// ShenandoahOldGarbageThreshold amounts of garbage. We identify these regions by setting the appropriate entry of
+// candidate_regions_for_promotion_by_copy[] to true. All entries are initialized to false before calling this
+// function.
+//
+// During the subsequent selection of the collection set, we give priority to these promotion set candidates.
+// Without this prioritization, we found that the aged regions tend to be ignored because they typically have
+// much less garbage and much more live data than the recently allocated "eden" regions. When aged regions are
+// repeatedly excluded from the collection set, the amount of live memory within the young generation tends to
+// accumulate and this has the undesirable side effect of causing young-generation collections to require much more
+// CPU and wall-clock time.
+//
+// A second benefit of treating aged regions differently than other regions during collection set selection is
+// that this allows us to more accurately budget memory to hold the results of evacuation. Memory for evacuation
+// of aged regions must be reserved in the old generations. Memory for evacuation of all other regions must be
+// reserved in the young generation.
+//
+// A side effect performed by this function is to tally up the number of regions and the number of live bytes
+// that we plan to promote-in-place during the current GC cycle. This information, which is stored with
+// an invocation of heap->set_promotion_in_place_potential(), feeds into subsequent decisions about when to
+// trigger the next GC and may identify special work to be done during this GC cycle if we choose to abbreviate it.
+//
+// Returns bytes of old-gen memory consumed by selected aged regions
+size_t ShenandoahHeuristics::select_aged_regions(size_t old_available, size_t num_regions,
+ bool candidate_regions_for_promotion_by_copy[]) {
ShenandoahHeap* heap = ShenandoahHeap::heap();
assert(heap->mode()->is_generational(), "Only in generational mode");
-
+ ShenandoahMarkingContext* const ctx = heap->marking_context();
size_t old_consumed = 0;
+ size_t promo_potential = 0;
+ size_t anticipated_promote_in_place_live = 0;
+
+ heap->clear_promotion_in_place_potential();
+ heap->clear_promotion_potential();
+ size_t candidates = 0;
+ size_t candidates_live = 0;
+ size_t old_garbage_threshold = (ShenandoahHeapRegion::region_size_bytes() * ShenandoahOldGarbageThreshold) / 100;
+ size_t promote_in_place_regions = 0;
+ size_t promote_in_place_live = 0;
+ size_t promote_in_place_pad = 0;
+ size_t anticipated_candidates = 0;
+ size_t anticipated_promote_in_place_regions = 0;
+
+ // Sort the promotion-eligible regions according to live-data-bytes so that we can first reclaim regions that require
+ // less evacuation effort. This prioritizes garbage first, expanding the allocation pool before we begin the work of
+ // reclaiming regions that require more effort.
+ AgedRegionData* sorted_regions = (AgedRegionData*) alloca(num_regions * sizeof(AgedRegionData));
for (size_t i = 0; i < num_regions; i++) {
- ShenandoahHeapRegion* region = heap->get_region(i);
- if (in_generation(region) && !region->is_empty() && region->is_regular() && (region->age() >= InitialTenuringThreshold)) {
- size_t promotion_need = (size_t) (region->get_live_data_bytes() * ShenandoahEvacWaste);
- if (old_consumed + promotion_need < old_available) {
+ ShenandoahHeapRegion* r = heap->get_region(i);
+ if (r->is_empty() || !r->has_live() || !r->is_young() || !r->is_regular()) {
+ continue;
+ }
+ if (r->age() >= InitialTenuringThreshold) {
+ r->save_top_before_promote();
+ if ((r->garbage() < old_garbage_threshold)) {
+ HeapWord* tams = ctx->top_at_mark_start(r);
+ HeapWord* original_top = r->top();
+ if (tams == original_top) {
+ // Fill the remnant memory within this region to assure no allocations prior to promote in place. Otherwise,
+ // newly allocated objects will not be parseable when promote in place tries to register them. Furthermore, any
+ // new allocations would not necessarily be eligible for promotion. This addresses both issues.
+ size_t remnant_size = r->free() / HeapWordSize;
+ if (remnant_size > ShenandoahHeap::min_fill_size()) {
+ ShenandoahHeap::fill_with_object(original_top, remnant_size);
+ r->set_top(r->end());
+ promote_in_place_pad += remnant_size * HeapWordSize;
+ } else {
+ // Since the remnant is so small that it cannot be filled, we don't have to worry about any accidental
+ // allocations occuring within this region before the region is promoted in place.
+ }
+ promote_in_place_regions++;
+ promote_in_place_live += r->get_live_data_bytes();
+ }
+ // Else, we do not promote this region (either in place or by copy) because it has received new allocations.
+
+ // During evacuation, we exclude from promotion regions for which age > tenure threshold, garbage < garbage-threshold,
+ // and get_top_before_promote() != tams
+ } else {
+ // After sorting and selecting best candidates below, we may decide to exclude this promotion-eligible region
+ // from the current collection sets. If this happens, we will consider this region as part of the anticipated
+ // promotion potential for the next GC pass.
+ size_t live_data = r->get_live_data_bytes();
+ candidates_live += live_data;
+ sorted_regions[candidates]._region = r;
+ sorted_regions[candidates++]._live_data = live_data;
+ }
+ } else {
+ // We only anticipate to promote regular regions if garbage() is above threshold. Tenure-aged regions with less
+ // garbage are promoted in place. These take a different path to old-gen. Note that certain regions that are
+ // excluded from anticipated promotion because their garbage content is too low (causing us to anticipate that
+ // the region would be promoted in place) may be eligible for evacuation promotion by the time promotion takes
+ // place during a subsequent GC pass because more garbage is found within the region between now and then. This
+ // should not happen if we are properly adapting the tenure age. The theory behind adaptive tenuring threshold
+ // is to choose the youngest age that demonstrates no "significant" futher loss of population since the previous
+ // age. If not this, we expect the tenure age to demonstrate linear population decay for at least two population
+ // samples, whereas we expect to observe exponetial population decay for ages younger than the tenure age.
+ //
+ // In the case that certain regions which were anticipated to be promoted in place need to be promoted by
+ // evacuation, it may be the case that there is not sufficient reserve within old-gen to hold evacuation of
+ // these regions. The likely outcome is that these regions will not be selected for evacuation or promotion
+ // in the current cycle and we will anticipate that they will be promoted in the next cycle. This will cause
+ // us to reserve more old-gen memory so that these objects can be promoted in the subsequent cycle.
+ //
+ // TODO:
+ // If we are auto-tuning the tenure age and regions that were anticipated to be promoted in place end up
+ // being promoted by evacuation, this event should feed into the tenure-age-selection heuristic so that
+ // the tenure age can be increased.
+ if (heap->is_aging_cycle() && (r->age() + 1 == InitialTenuringThreshold)) {
+ if (r->garbage() >= old_garbage_threshold) {
+ anticipated_candidates++;
+ promo_potential += r->get_live_data_bytes();
+ }
+ else {
+ anticipated_promote_in_place_regions++;
+ anticipated_promote_in_place_live += r->get_live_data_bytes();
+ }
+ }
+ }
+ // Note that we keep going even if one region is excluded from selection.
+ // Subsequent regions may be selected if they have smaller live data.
+ }
+ // Sort in increasing order according to live data bytes. Note that candidates represents the number of regions
+ // that qualify to be promoted by evacuation.
+ if (candidates > 0) {
+ QuickSort::sort<AgedRegionData>(sorted_regions, candidates, compare_by_aged_live, false);
+ for (size_t i = 0; i < candidates; i++) {
+ size_t region_live_data = sorted_regions[i]._live_data;
+ size_t promotion_need = (size_t) (region_live_data * ShenandoahPromoEvacWaste);
+ if (old_consumed + promotion_need <= old_available) {
+ ShenandoahHeapRegion* region = sorted_regions[i]._region;
old_consumed += promotion_need;
- preselected_regions[i] = true;
+ candidate_regions_for_promotion_by_copy[region->index()] = true;
+ } else {
+ // We rejected this promotable region from the collection set because we had no room to hold its copy.
+ // Add this region to promo potential for next GC.
+ promo_potential += region_live_data;
}
- // Note that we keep going even if one region is excluded from selection.
- // Subsequent regions may be selected if they have smaller live data.
+ // We keep going even if one region is excluded from selection because we need to accumulate all eligible
+ // regions that are not preselected into promo_potential
}
}
+ heap->set_pad_for_promote_in_place(promote_in_place_pad);
+ heap->set_promotion_potential(promo_potential);
+ heap->set_promotion_in_place_potential(anticipated_promote_in_place_live);
return old_consumed;
}
void ShenandoahHeuristics::choose_collection_set(ShenandoahCollectionSet* collection_set, ShenandoahOldHeuristics* old_heuristics) {
ShenandoahHeap* heap = ShenandoahHeap::heap();
bool is_generational = heap->mode()->is_generational();
+ size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
assert(collection_set->count() == 0, "Must be empty");
assert(!is_generational || !_generation->is_old(), "Old GC invokes ShenandoahOldHeuristics::choose_collection_set()");
@@ -113,6 +265,7 @@ void ShenandoahHeuristics::choose_collection_set(ShenandoahCollectionSet* collec
RegionData* candidates = _region_data;
size_t cand_idx = 0;
+ size_t preselected_candidates = 0;
size_t total_garbage = 0;
@@ -122,12 +275,21 @@ void ShenandoahHeuristics::choose_collection_set(ShenandoahCollectionSet* collec
size_t free = 0;
size_t free_regions = 0;
+ size_t old_garbage_threshold = (region_size_bytes * ShenandoahOldGarbageThreshold) / 100;
+ // This counts number of humongous regions that we intend to promote in this cycle.
+ size_t humongous_regions_promoted = 0;
+ // This counts bytes of memory used by hunongous regions to be promoted in place.
+ size_t humongous_bytes_promoted = 0;
+ // This counts number of regular regions that will be promoted in place.
+ size_t regular_regions_promoted_in_place = 0;
+ // This counts bytes of memory used by regular regions to be promoted in place.
+ size_t regular_regions_promoted_usage = 0;
+
for (size_t i = 0; i < num_regions; i++) {
ShenandoahHeapRegion* region = heap->get_region(i);
if (is_generational && !in_generation(region)) {
continue;
}
-
size_t garbage = region->garbage();
total_garbage += garbage;
if (region->is_empty()) {
@@ -141,17 +303,37 @@ void ShenandoahHeuristics::choose_collection_set(ShenandoahCollectionSet* collec
region->make_trash_immediate();
} else {
assert(!_generation->is_old(), "OLD is handled elsewhere");
+ bool is_candidate;
// This is our candidate for later consideration.
- candidates[cand_idx]._region = region;
if (is_generational && collection_set->is_preselected(i)) {
- // If region is preselected, we know mode()->is_generational() and region->age() >= InitialTenuringThreshold)
- garbage = ShenandoahHeapRegion::region_size_bytes();
+ // If !is_generational, we cannot ask if is_preselected. If is_preselected, we know
+ // region->age() >= InitialTenuringThreshold).
+ is_candidate = true;
+ preselected_candidates++;
+ // Set garbage value to maximum value to force this into the sorted collection set.
+ garbage = region_size_bytes;
+ } else if (is_generational && region->is_young() && (region->age() >= InitialTenuringThreshold)) {
+ // Note that for GLOBAL GC, region may be OLD, and OLD regions do not qualify for pre-selection
+
+ // This region is old enough to be promoted but it was not preselected, either because its garbage is below
+ // ShenandoahOldGarbageThreshold so it will be promoted in place, or because there is not sufficient room
+ // in old gen to hold the evacuated copies of this region's live data. In both cases, we choose not to
+ // place this region into the collection set.
+ if (region->garbage_before_padded_for_promote() < old_garbage_threshold) {
+ regular_regions_promoted_in_place++;
+ regular_regions_promoted_usage += region->used_before_promote();
+ }
+ is_candidate = false;
+ } else {
+ is_candidate = true;
+ }
+ if (is_candidate) {
+ candidates[cand_idx]._region = region;
+ candidates[cand_idx]._u._garbage = garbage;
+ cand_idx++;
}
- candidates[cand_idx]._garbage = garbage;
- cand_idx++;
}
} else if (region->is_humongous_start()) {
-
// Reclaim humongous regions here, and count them as the immediate garbage
#ifdef ASSERT
bool reg_live = region->has_live();
@@ -166,6 +348,13 @@ void ShenandoahHeuristics::choose_collection_set(ShenandoahCollectionSet* collec
// Count only the start. Continuations would be counted on "trash" path
immediate_regions++;
immediate_garbage += garbage;
+ } else {
+ if (region->is_young() && region->age() >= InitialTenuringThreshold) {
+ oop obj = cast_to_oop(region->bottom());
+ size_t humongous_regions = ShenandoahHeapRegion::required_regions(obj->size() * HeapWordSize);
+ humongous_regions_promoted += humongous_regions;
+ humongous_bytes_promoted += obj->size() * HeapWordSize;
+ }
}
} else if (region->is_trash()) {
// Count in just trashed collection set, during coalesced CM-with-UR
@@ -173,6 +362,14 @@ void ShenandoahHeuristics::choose_collection_set(ShenandoahCollectionSet* collec
immediate_garbage += garbage;
}
}
+ heap->reserve_promotable_humongous_regions(humongous_regions_promoted);
+ heap->reserve_promotable_humongous_usage(humongous_bytes_promoted);
+ heap->reserve_promotable_regular_regions(regular_regions_promoted_in_place);
+ heap->reserve_promotable_regular_usage(regular_regions_promoted_usage);
+ log_info(gc, ergo)("Planning to promote in place " SIZE_FORMAT " humongous regions and " SIZE_FORMAT
+ " regular regions, spanning a total of " SIZE_FORMAT " used bytes",
+ humongous_regions_promoted, regular_regions_promoted_in_place,
+ humongous_regions_promoted * ShenandoahHeapRegion::region_size_bytes() + regular_regions_promoted_usage);
// Step 2. Look back at garbage statistics, and decide if we want to collect anything,
// given the amount of immediately reclaimable garbage. If we do, figure out the collection set.
@@ -185,7 +382,9 @@ void ShenandoahHeuristics::choose_collection_set(ShenandoahCollectionSet* collec
size_t immediate_percent = (total_garbage == 0) ? 0 : (immediate_garbage * 100 / total_garbage);
collection_set->set_immediate_trash(immediate_garbage);
- if (immediate_percent <= ShenandoahImmediateThreshold) {
+ ShenandoahGeneration* young_gen = heap->young_generation();
+ bool doing_promote_in_place = (humongous_regions_promoted + regular_regions_promoted_in_place > 0);
+ if (doing_promote_in_place || (preselected_candidates > 0) || (immediate_percent <= ShenandoahImmediateThreshold)) {
if (old_heuristics != nullptr) {
old_heuristics->prime_collection_set(collection_set);
} else {
@@ -359,6 +558,12 @@ void ShenandoahHeuristics::initialize() {
// Nothing to do by default.
}
+size_t ShenandoahHeuristics::bytes_of_allocation_runway_before_gc_trigger(size_t young_regions_to_be_recycled) {
+ assert(false, "Only implemented for young Adaptive Heuristics");
+ return 0;
+}
+
+
double ShenandoahHeuristics::elapsed_cycle_time() const {
return os::elapsedTime() - _cycle_start;
}
@@ -370,6 +575,10 @@ bool ShenandoahHeuristics::in_generation(ShenandoahHeapRegion* region) {
}
size_t ShenandoahHeuristics::min_free_threshold() {
- size_t min_free_threshold = _generation->is_old() ? ShenandoahOldMinFreeThreshold : ShenandoahMinFreeThreshold;
- return _generation->soft_max_capacity() / 100 * min_free_threshold;
+ assert(!_generation->is_old(), "min_free_threshold is only relevant to young GC");
+ size_t min_free_threshold = ShenandoahMinFreeThreshold;
+ // Note that soft_max_capacity() / 100 * min_free_threshold is smaller than max_capacity() / 100 * min_free_threshold.
+ // We want to behave conservatively here, so use max_capacity(). By returning a larger value, we cause the GC to
+ // trigger when the remaining amount of free shrinks below the larger threshold.
+ return _generation->max_capacity() / 100 * min_free_threshold;
}
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.hpp
index aba842c2455..6b33025dfa8 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.hpp
@@ -70,7 +70,10 @@ class ShenandoahHeuristics : public CHeapObj<mtGC> {
typedef struct {
ShenandoahHeapRegion* _region;
- size_t _garbage;
+ union {
+ size_t _garbage; // Not used by old-gen heuristics.
+ size_t _live_data; // Only used for old-gen heuristics, which prioritizes retention of _live_data over garbage reclaim
+ } _u;
} RegionData;
ShenandoahGeneration* _generation;
@@ -106,6 +109,14 @@ class ShenandoahHeuristics : public CHeapObj<mtGC> {
static int compare_by_garbage(RegionData a, RegionData b);
+ // Compare by live is used to prioritize compaction of old-gen regions. With old-gen compaction, the goal is
+ // to tightly pack long-lived objects into available regions. In most cases, there has not been an accumulation
+ // of garbage within old-gen regions. The more likely opportunity will be to combine multiple sparsely populated
+ // old-gen regions which may have been promoted in place into a smaller number of densely packed old-gen regions.
+ // This improves subsequent allocation efficiency and reduces the likelihood of allocation failure (including
+ // humongous allocation failure) due to fragmentation of the available old-gen allocation pool
+ static int compare_by_live(RegionData a, RegionData b);
+
// TODO: We need to enhance this API to give visibility to accompanying old-gen evacuation effort.
// In the case that the old-gen evacuation effort is small or zero, the young-gen heuristics
// should feel free to dedicate increased efforts to young-gen evacuation.
@@ -156,7 +167,7 @@ class ShenandoahHeuristics : public CHeapObj<mtGC> {
virtual void reset_gc_learning();
- virtual size_t select_aged_regions(size_t old_available, size_t num_regions, bool* preselected_regions);
+ virtual size_t select_aged_regions(size_t old_available, size_t num_regions, bool candidate_regions_for_promotion_by_copy[]);
virtual void choose_collection_set(ShenandoahCollectionSet* collection_set, ShenandoahOldHeuristics* old_heuristics);
@@ -169,6 +180,8 @@ class ShenandoahHeuristics : public CHeapObj<mtGC> {
virtual bool is_experimental() = 0;
virtual void initialize();
+ virtual size_t bytes_of_allocation_runway_before_gc_trigger(size_t region_to_be_recycled);
+
double elapsed_cycle_time() const;
};
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.cpp
index ad8535ebe33..5722103d73a 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.cpp
@@ -43,20 +43,23 @@ ShenandoahOldHeuristics::ShenandoahOldHeuristics(ShenandoahOldGeneration* genera
_next_old_collection_candidate(0),
_last_old_region(0),
_trigger_heuristic(trigger_heuristic),
+ _old_generation(generation),
_promotion_failed(false),
- _old_generation(generation)
+ _cannot_expand_trigger(false),
+ _fragmentation_trigger(false),
+ _growth_trigger(false)
{
assert(_generation->is_old(), "This service only available for old-gc heuristics");
}
bool ShenandoahOldHeuristics::prime_collection_set(ShenandoahCollectionSet* collection_set) {
+ ShenandoahHeap* heap = ShenandoahHeap::heap();
if (unprocessed_old_collection_candidates() == 0) {
return false;
}
_first_pinned_candidate = NOT_FOUND;
- ShenandoahHeap* heap = ShenandoahHeap::heap();
uint included_old_regions = 0;
size_t evacuated_old_bytes = 0;
size_t collected_old_bytes = 0;
@@ -66,15 +69,36 @@ bool ShenandoahOldHeuristics::prime_collection_set(ShenandoahCollectionSet* coll
// of memory that can still be evacuated. We address this by reducing the evacuation budget by the amount
// of live memory in that region and by the amount of unallocated memory in that region if the evacuation
// budget is constrained by availability of free memory.
- size_t old_evacuation_budget = (size_t) ((double) heap->get_old_evac_reserve() / ShenandoahEvacWaste);
+ size_t old_evacuation_budget = (size_t) ((double) heap->get_old_evac_reserve() / ShenandoahOldEvacWaste);
+ size_t unfragmented_available = _old_generation->free_unaffiliated_regions() * ShenandoahHeapRegion::region_size_bytes();
+ size_t fragmented_available;
+ size_t excess_fragmented_available;
+
+ if (unfragmented_available > old_evacuation_budget) {
+ unfragmented_available = old_evacuation_budget;
+ fragmented_available = 0;
+ excess_fragmented_available = 0;
+ } else {
+ assert(_old_generation->available() > old_evacuation_budget, "Cannot budget more than is available");
+ fragmented_available = _old_generation->available() - unfragmented_available;
+ assert(fragmented_available + unfragmented_available >= old_evacuation_budget, "Budgets do not add up");
+ if (fragmented_available + unfragmented_available > old_evacuation_budget) {
+ excess_fragmented_available = (fragmented_available + unfragmented_available) - old_evacuation_budget;
+ fragmented_available -= excess_fragmented_available;
+ }
+ }
+
size_t remaining_old_evacuation_budget = old_evacuation_budget;
- size_t lost_evacuation_capacity = 0;
log_info(gc)("Choose old regions for mixed collection: old evacuation budget: " SIZE_FORMAT "%s, candidates: %u",
byte_size_in_proper_unit(old_evacuation_budget), proper_unit_for_byte_size(old_evacuation_budget),
unprocessed_old_collection_candidates());
+ size_t lost_evacuation_capacity = 0;
+
// The number of old-gen regions that were selected as candidates for collection at the end of the most recent old-gen
- // concurrent marking phase and have not yet been collected is represented by unprocessed_old_collection_candidates()
+ // concurrent marking phase and have not yet been collected is represented by unprocessed_old_collection_candidates().
+ // Candidate regions are ordered according to increasing amount of live data. If there is not sufficient room to
+ // evacuate region N, then there is no need to even consider evacuating region N+1.
while (unprocessed_old_collection_candidates() > 0) {
// Old collection candidates are sorted in order of decreasing garbage contained therein.
ShenandoahHeapRegion* r = next_old_collection_candidate();
@@ -82,31 +106,74 @@ bool ShenandoahOldHeuristics::prime_collection_set(ShenandoahCollectionSet* coll
break;
}
- // If we choose region r to be collected, then we need to decrease the capacity to hold other evacuations by
- // the size of r's free memory.
-
- // It's probably overkill to compensate with lost_evacuation_capacity.
- // But it's the safe thing to do and has minimal impact on content of primed collection set.
- size_t live = r->get_live_data_bytes();
- if (live + lost_evacuation_capacity <= remaining_old_evacuation_budget) {
- // Decrement remaining evacuation budget by bytes that will be copied.
- lost_evacuation_capacity += r->free();
- remaining_old_evacuation_budget -= live;
- collection_set->add_region(r);
- included_old_regions++;
- evacuated_old_bytes += live;
- collected_old_bytes += r->garbage();
- consume_old_collection_candidate();
+ // If region r is evacuated to fragmented memory (to free memory within a partially used region), then we need
+ // to decrease the capacity of the fragmented memory by the scaled loss.
+
+ size_t live_data_for_evacuation = r->get_live_data_bytes();
+ size_t lost_available = r->free();
+
+ if ((lost_available > 0) && (excess_fragmented_available > 0)) {
+ if (lost_available < excess_fragmented_available) {
+ excess_fragmented_available -= lost_available;
+ lost_evacuation_capacity -= lost_available;
+ lost_available = 0;
+ } else {
+ lost_available -= excess_fragmented_available;
+ lost_evacuation_capacity -= excess_fragmented_available;
+ excess_fragmented_available = 0;
+ }
+ }
+ size_t scaled_loss = (size_t) ((double) lost_available / ShenandoahOldEvacWaste);
+ if ((lost_available > 0) && (fragmented_available > 0)) {
+ if (scaled_loss + live_data_for_evacuation < fragmented_available) {
+ fragmented_available -= scaled_loss;
+ scaled_loss = 0;
+ } else {
+ // We will have to allocate this region's evacuation memory from unfragmented memory, so don't bother
+ // to decrement scaled_loss
+ }
+ }
+ if (scaled_loss > 0) {
+ // We were not able to account for the lost free memory within fragmented memory, so we need to take this
+ // allocation out of unfragmented memory. Unfragmented memory does not need to account for loss of free.
+ if (live_data_for_evacuation > unfragmented_available) {
+ // There is not room to evacuate this region or any that come after it in within the candidates array.
+ break;
+ } else {
+ unfragmented_available -= live_data_for_evacuation;
+ }
} else {
- break;
+ // Since scaled_loss == 0, we have accounted for the loss of free memory, so we can allocate from either
+ // fragmented or unfragmented available memory. Use up the fragmented memory budget first.
+ size_t evacuation_need = live_data_for_evacuation;
+
+ if (evacuation_need > fragmented_available) {
+ evacuation_need -= fragmented_available;
+ fragmented_available = 0;
+ } else {
+ fragmented_available -= evacuation_need;
+ evacuation_need = 0;
+ }
+ if (evacuation_need > unfragmented_available) {
+ // There is not room to evacuate this region or any that come after it in within the candidates array.
+ break;
+ } else {
+ unfragmented_available -= evacuation_need;
+ // dead code: evacuation_need == 0;
+ }
}
+ collection_set->add_region(r);
+ included_old_regions++;
+ evacuated_old_bytes += live_data_for_evacuation;
+ collected_old_bytes += r->garbage();
+ consume_old_collection_candidate();
}
if (_first_pinned_candidate != NOT_FOUND) {
// Need to deal with pinned regions
slide_pinned_regions_to_front();
}
-
+ decrease_unprocessed_old_collection_candidates_live_memory(evacuated_old_bytes);
if (included_old_regions > 0) {
log_info(gc)("Old-gen piggyback evac (" UINT32_FORMAT " regions, evacuating " SIZE_FORMAT "%s, reclaiming: " SIZE_FORMAT "%s)",
included_old_regions,
@@ -116,6 +183,8 @@ bool ShenandoahOldHeuristics::prime_collection_set(ShenandoahCollectionSet* coll
if (unprocessed_old_collection_candidates() == 0) {
// We have added the last of our collection candidates to a mixed collection.
+ // Any triggers that occurred during mixed evacuations may no longer be valid. They can retrigger if appropriate.
+ clear_triggers();
_old_generation->transition_to(ShenandoahOldGeneration::IDLE);
} else if (included_old_regions == 0) {
// We have candidates, but none were included for evacuation - are they all pinned?
@@ -202,7 +271,7 @@ void ShenandoahOldHeuristics::slide_pinned_regions_to_front() {
if (skipped._region->is_pinned()) {
RegionData& available_slot = _region_data[write_index];
available_slot._region = skipped._region;
- available_slot._garbage = skipped._garbage;
+ available_slot._u._live_data = skipped._u._live_data;
--write_index;
}
}
@@ -232,6 +301,7 @@ void ShenandoahOldHeuristics::prepare_for_old_collections() {
size_t num_regions = heap->num_regions();
size_t immediate_garbage = 0;
size_t immediate_regions = 0;
+ size_t live_data = 0;
RegionData* candidates = _region_data;
for (size_t i = 0; i < num_regions; i++) {
@@ -241,7 +311,9 @@ void ShenandoahOldHeuristics::prepare_for_old_collections() {
}
size_t garbage = region->garbage();
+ size_t live_bytes = region->get_live_data_bytes();
total_garbage += garbage;
+ live_data += live_bytes;
if (region->is_regular() || region->is_pinned()) {
if (!region->has_live()) {
@@ -252,7 +324,7 @@ void ShenandoahOldHeuristics::prepare_for_old_collections() {
} else {
region->begin_preemptible_coalesce_and_fill();
candidates[cand_idx]._region = region;
- candidates[cand_idx]._garbage = garbage;
+ candidates[cand_idx]._u._live_data = live_bytes;
cand_idx++;
}
} else if (region->is_humongous_start()) {
@@ -273,42 +345,63 @@ void ShenandoahOldHeuristics::prepare_for_old_collections() {
}
}
+ _old_generation->set_live_bytes_after_last_mark(live_data);
+
// TODO: Consider not running mixed collects if we recovered some threshold percentage of memory from immediate garbage.
// This would be similar to young and global collections shortcutting evacuation, though we'd probably want a separate
// threshold for the old generation.
- // Prioritize regions to select garbage-first regions
- QuickSort::sort<RegionData>(candidates, cand_idx, compare_by_garbage, false);
+ // Unlike young, we are more interested in efficiently packing OLD-gen than in reclaiming garbage first. We sort by live-data.
+ // Some regular regions may have been promoted in place with no garbage but also with very little live data. When we "compact"
+ // old-gen, we want to pack these underutilized regions together so we can have more unaffiliated (unfragmented) free regions
+ // in old-gen.
+ QuickSort::sort<RegionData>(candidates, cand_idx, compare_by_live, false);
- // Any old-gen region that contains (ShenandoahOldGarbageThreshold (default value 25))% garbage or more is to
- // be evacuated.
+ // Any old-gen region that contains (ShenandoahOldGarbageThreshold (default value 25)% garbage or more is to be
+ // added to the list of candidates for subsequent mixed evacuations.
//
// TODO: allow ShenandoahOldGarbageThreshold to be determined adaptively, by heuristics.
+ const size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+
+ // The convention is to collect regions that have more than this amount of garbage.
+ const size_t garbage_threshold = region_size_bytes * ShenandoahOldGarbageThreshold / 100;
+
+ // Englightened interpretation: collect regions that have less than this amount of live.
+ const size_t live_threshold = region_size_bytes - garbage_threshold;
- const size_t garbage_threshold = ShenandoahHeapRegion::region_size_bytes() * ShenandoahOldGarbageThreshold / 100;
size_t candidates_garbage = 0;
_last_old_region = (uint)cand_idx;
_last_old_collection_candidate = (uint)cand_idx;
_next_old_collection_candidate = 0;
+ size_t unfragmented = 0;
+
for (size_t i = 0; i < cand_idx; i++) {
- if (candidates[i]._garbage < garbage_threshold) {
- // Candidates are sorted in decreasing order of garbage, so no regions after this will be above the threshold
+ size_t live = candidates[i]._u._live_data;
+ if (live > live_threshold) {
+ // Candidates are sorted in increasing order of live data, so no regions after this will be below the threshold.
_last_old_collection_candidate = (uint)i;
break;
}
- candidates_garbage += candidates[i]._garbage;
+ size_t region_garbage = candidates[i]._region->garbage();
+ size_t region_free = candidates[i]._region->free();
+ candidates_garbage += region_garbage;
+ unfragmented += region_free;
}
// Note that we do not coalesce and fill occupied humongous regions
// HR: humongous regions, RR: regular regions, CF: coalesce and fill regions
size_t collectable_garbage = immediate_garbage + candidates_garbage;
- log_info(gc)("Old-Gen Collectable Garbage: " SIZE_FORMAT "%s over " UINT32_FORMAT " regions, "
+ size_t old_candidates = _last_old_collection_candidate;
+ log_info(gc)("Old-Gen Collectable Garbage: " SIZE_FORMAT "%s "
+ "consolidated with free: " SIZE_FORMAT "%s, over " SIZE_FORMAT " regions, "
"Old-Gen Immediate Garbage: " SIZE_FORMAT "%s over " SIZE_FORMAT " regions.",
- byte_size_in_proper_unit(collectable_garbage), proper_unit_for_byte_size(collectable_garbage), _last_old_collection_candidate,
- byte_size_in_proper_unit(immediate_garbage), proper_unit_for_byte_size(immediate_garbage), immediate_regions);
-
+ byte_size_in_proper_unit(collectable_garbage), proper_unit_for_byte_size(collectable_garbage),
+ byte_size_in_proper_unit(unfragmented), proper_unit_for_byte_size(unfragmented), old_candidates,
+ byte_size_in_proper_unit(immediate_garbage), proper_unit_for_byte_size(immediate_garbage), immediate_regions);
+ size_t mixed_evac_live = old_candidates * region_size_bytes - (candidates_garbage + unfragmented);
+ set_unprocessed_old_collection_candidates_live_memory(mixed_evac_live);
if (unprocessed_old_collection_candidates() == 0) {
_old_generation->transition_to(ShenandoahOldGeneration::IDLE);
} else {
@@ -316,6 +409,20 @@ void ShenandoahOldHeuristics::prepare_for_old_collections() {
}
}
+size_t ShenandoahOldHeuristics::unprocessed_old_collection_candidates_live_memory() const {
+ return _live_bytes_in_unprocessed_candidates;
+}
+
+void ShenandoahOldHeuristics::set_unprocessed_old_collection_candidates_live_memory(size_t initial_live) {
+ _live_bytes_in_unprocessed_candidates = initial_live;
+}
+
+void ShenandoahOldHeuristics::decrease_unprocessed_old_collection_candidates_live_memory(size_t evacuated_live) {
+ assert(evacuated_live <= _live_bytes_in_unprocessed_candidates, "Cannot evacuate more than was present");
+ _live_bytes_in_unprocessed_candidates -= evacuated_live;
+}
+
+
// TODO: Unused?
uint ShenandoahOldHeuristics::last_old_collection_candidate_index() {
return _last_old_collection_candidate;
@@ -356,7 +463,7 @@ unsigned int ShenandoahOldHeuristics::get_coalesce_and_fill_candidates(Shenandoa
while (index < end) {
*buffer++ = _region_data[index++]._region;
}
- return _last_old_region - _next_old_collection_candidate;
+ return (_last_old_region - _next_old_collection_candidate);
}
void ShenandoahOldHeuristics::abandon_collection_candidates() {
@@ -366,24 +473,31 @@ void ShenandoahOldHeuristics::abandon_collection_candidates() {
}
void ShenandoahOldHeuristics::handle_promotion_failure() {
- if (!_promotion_failed) {
- if (ShenandoahHeap::heap()->generation_sizer()->transfer_capacity(_old_generation)) {
- log_info(gc)("Increased size of old generation due to promotion failure.");
- }
- // TODO: Increase tenuring threshold to push back on promotions.
- }
_promotion_failed = true;
}
void ShenandoahOldHeuristics::record_cycle_start() {
- _promotion_failed = false;
_trigger_heuristic->record_cycle_start();
}
void ShenandoahOldHeuristics::record_cycle_end() {
_trigger_heuristic->record_cycle_end();
+ clear_triggers();
+}
+
+void ShenandoahOldHeuristics::trigger_old_has_grown() {
+ _growth_trigger = true;
}
+
+void ShenandoahOldHeuristics::clear_triggers() {
+ // Clear any triggers that were set during mixed evacuations. Conditions may be different now that this phase has finished.
+ _promotion_failed = false;
+ _cannot_expand_trigger = false;
+ _fragmentation_trigger = false;
+ _growth_trigger = false;
+ }
+
bool ShenandoahOldHeuristics::should_start_gc() {
// Cannot start a new old-gen GC until previous one has finished.
//
@@ -393,13 +507,49 @@ bool ShenandoahOldHeuristics::should_start_gc() {
return false;
}
- // If there's been a promotion failure (and we don't have regions already scheduled for evacuation),
- // start a new old generation collection.
- if (_promotion_failed) {
- log_info(gc)("Trigger: Promotion Failure");
+ if (_cannot_expand_trigger) {
+ ShenandoahHeap* heap = ShenandoahHeap::heap();
+ size_t old_gen_capacity = _old_generation->max_capacity();
+ size_t heap_capacity = heap->capacity();
+ double percent = 100.0 * ((double) old_gen_capacity) / heap_capacity;
+ log_info(gc)("Trigger (OLD): Expansion failure, current size: " SIZE_FORMAT "%s which is %.1f%% of total heap size",
+ byte_size_in_proper_unit(old_gen_capacity), proper_unit_for_byte_size(old_gen_capacity), percent);
return true;
}
+ if (_fragmentation_trigger) {
+ ShenandoahHeap* heap = ShenandoahHeap::heap();
+ size_t used = _old_generation->used();
+ size_t used_regions_size = _old_generation->used_regions_size();
+ size_t used_regions = _old_generation->used_regions();
+ assert(used_regions_size > used_regions, "Cannot have more used than used regions");
+ size_t fragmented_free = used_regions_size - used;
+ double percent = 100.0 * ((double) fragmented_free) / used_regions_size;
+ log_info(gc)("Trigger (OLD): Old has become fragmented: "
+ SIZE_FORMAT "%s available bytes spread between " SIZE_FORMAT " regions (%.1f%% free)",
+ byte_size_in_proper_unit(fragmented_free), proper_unit_for_byte_size(fragmented_free), used_regions, percent);
+ return true;
+ }
+
+ if (_growth_trigger) {
+ // Growth may be falsely triggered during mixed evacuations, before the mixed-evacuation candidates have been
+ // evacuated. Before acting on a false trigger, we check to confirm the trigger condition is still satisfied.
+ ShenandoahHeap* heap = ShenandoahHeap::heap();
+ size_t current_usage = _old_generation->used();
+ size_t trigger_threshold = _old_generation->usage_trigger_threshold();
+ if (current_usage > trigger_threshold) {
+ size_t live_at_previous_old = _old_generation->get_live_bytes_after_last_mark();
+ double percent_growth = 100.0 * ((double) current_usage - live_at_previous_old) / live_at_previous_old;
+ log_info(gc)("Trigger (OLD): Old has overgrown, live at end of previous OLD marking: "
+ SIZE_FORMAT "%s, current usage: " SIZE_FORMAT "%s, percent growth: %.1f%%",
+ byte_size_in_proper_unit(live_at_previous_old), proper_unit_for_byte_size(live_at_previous_old),
+ byte_size_in_proper_unit(current_usage), proper_unit_for_byte_size(current_usage), percent_growth);
+ return true;
+ } else {
+ _growth_trigger = false;
+ }
+ }
+
// Otherwise, defer to configured heuristic for gc trigger.
return _trigger_heuristic->should_start_gc();
}
@@ -409,14 +559,20 @@ bool ShenandoahOldHeuristics::should_degenerate_cycle() {
}
void ShenandoahOldHeuristics::record_success_concurrent(bool abbreviated) {
+ // Forget any triggers that occured while OLD GC was ongoing. If we really need to start another, it will retrigger.
+ clear_triggers();
_trigger_heuristic->record_success_concurrent(abbreviated);
}
void ShenandoahOldHeuristics::record_success_degenerated() {
+ // Forget any triggers that occured while OLD GC was ongoing. If we really need to start another, it will retrigger.
+ clear_triggers();
_trigger_heuristic->record_success_degenerated();
}
void ShenandoahOldHeuristics::record_success_full() {
+ // Forget any triggers that occured while OLD GC was ongoing. If we really need to start another, it will retrigger.
+ clear_triggers();
_trigger_heuristic->record_success_full();
}
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.hpp
index f8581c2cd01..d6789cb87ec 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.hpp
@@ -69,15 +69,24 @@ class ShenandoahOldHeuristics : public ShenandoahHeuristics {
// the end of old final mark.
uint _last_old_region;
+ // How much live data must be evacuated from within the unprocessed mixed evacuation candidates?
+ size_t _live_bytes_in_unprocessed_candidates;
+
// This can be the 'static' or 'adaptive' heuristic.
ShenandoahHeuristics* _trigger_heuristic;
+ // Keep a pointer to our generation that we can use without down casting a protected member from the base class.
+ ShenandoahOldGeneration* _old_generation;
+
// Flag is set when promotion failure is detected (by gc thread), and cleared when
// old generation collection begins (by control thread).
volatile bool _promotion_failed;
- // Keep a pointer to our generation that we can use without down casting a protected member from the base class.
- ShenandoahOldGeneration* _old_generation;
+ // Flags are set when promotion failure is detected (by gc thread), and cleared when
+ // old generation collection begins (by control thread). Flags are set and cleared at safepoints.
+ bool _cannot_expand_trigger;
+ bool _fragmentation_trigger;
+ bool _growth_trigger;
protected:
virtual void choose_collection_set_from_regiondata(ShenandoahCollectionSet* set, RegionData* data, size_t data_size,
@@ -97,6 +106,13 @@ class ShenandoahOldHeuristics : public ShenandoahHeuristics {
// How many old-collection candidates have not yet been processed?
uint unprocessed_old_collection_candidates();
+ // How much live memory must be evacuated from within old-collection candidates that have not yet been processed?
+ size_t unprocessed_old_collection_candidates_live_memory() const;
+
+ void set_unprocessed_old_collection_candidates_live_memory(size_t initial_live);
+
+ void decrease_unprocessed_old_collection_candidates_live_memory(size_t evacuated_live);
+
// How many old or hidden collection candidates have not yet been processed?
uint last_old_collection_candidate_index();
@@ -122,11 +138,17 @@ class ShenandoahOldHeuristics : public ShenandoahHeuristics {
// held by this heuristic for supplying mixed collections.
void abandon_collection_candidates();
- // Notify the heuristic of promotion failures. The promotion attempt will be skipped and the object will
- // be evacuated into the young generation. The collection should complete normally, but we want to schedule
- // an old collection as soon as possible.
+ // Promotion failure does not currently trigger old-gen collections. Often, promotion failures occur because
+ // old-gen is sized too small rather than because it is necessary to collect old gen. We keep the method
+ // here in case we decide to feed this signal to sizing or triggering heuristics in the future.
void handle_promotion_failure();
+ void trigger_cannot_expand() { _cannot_expand_trigger = true; };
+ void trigger_old_is_fragmented() { _fragmentation_trigger = true; }
+ void trigger_old_has_grown();
+
+ void clear_triggers();
+
virtual void record_cycle_start() override;
virtual void record_cycle_end() override;
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahCollectionSet.cpp b/src/hotspot/share/gc/shenandoah/shenandoahCollectionSet.cpp
index c797ad21a7d..736354e9282 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahCollectionSet.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahCollectionSet.cpp
@@ -90,13 +90,13 @@ void ShenandoahCollectionSet::add_region(ShenandoahHeapRegion* r) {
assert(!r->is_humongous(), "Only add regular regions to the collection set");
_cset_map[r->index()] = 1;
-
- size_t live = r->get_live_data_bytes();
+ size_t live = r->get_live_data_bytes();
size_t garbage = r->garbage();
-
+ size_t free = r->free();
if (r->is_young()) {
_young_region_count++;
_young_bytes_to_evacuate += live;
+ _young_available_bytes_collected += free;
if (r->age() >= InitialTenuringThreshold) {
_young_bytes_to_promote += live;
}
@@ -104,6 +104,7 @@ void ShenandoahCollectionSet::add_region(ShenandoahHeapRegion* r) {
_old_region_count++;
_old_bytes_to_evacuate += live;
_old_garbage += garbage;
+ _old_available_bytes_collected += free;
}
_region_count++;
@@ -117,6 +118,7 @@ void ShenandoahCollectionSet::add_region(ShenandoahHeapRegion* r) {
void ShenandoahCollectionSet::clear() {
assert(ShenandoahSafepoint::is_at_shenandoah_safepoint(), "Must be at a safepoint");
+
Copy::zero_to_bytes(_cset_map, _map_size);
#ifdef ASSERT
@@ -140,6 +142,9 @@ void ShenandoahCollectionSet::clear() {
_old_region_count = 0;
_old_bytes_to_evacuate = 0;
+ _young_available_bytes_collected = 0;
+ _old_available_bytes_collected = 0;
+
_has_old_regions = false;
}
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahCollectionSet.hpp b/src/hotspot/share/gc/shenandoah/shenandoahCollectionSet.hpp
index 112fff383ba..80a5e9dc028 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahCollectionSet.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahCollectionSet.hpp
@@ -66,6 +66,11 @@ class ShenandoahCollectionSet : public CHeapObj<mtGC> {
// spans of time while collection set is being constructed.
bool* _preselected_regions;
+ // When a region having memory available to be allocated is added to the collection set, the region's available memory
+ // should be subtracted from what's available.
+ size_t _young_available_bytes_collected;
+ size_t _old_available_bytes_collected;
+
shenandoah_padding(0);
volatile size_t _current_index;
shenandoah_padding(1);
@@ -111,6 +116,10 @@ class ShenandoahCollectionSet : public CHeapObj<mtGC> {
inline size_t get_young_bytes_to_be_promoted();
+ size_t get_young_available_bytes_collected() { return _young_available_bytes_collected; }
+
+ size_t get_old_available_bytes_collected() { return _old_available_bytes_collected; }
+
inline size_t get_old_region_count();
inline size_t get_young_region_count();
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.cpp b/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.cpp
index 73e1d5dbf2c..918aff26add 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.cpp
@@ -128,6 +128,12 @@ bool ShenandoahCollectorPolicy::is_at_shutdown() {
return _in_shutdown.is_set();
}
+// This may be called by mutator threads. We declare _success_full_gcs volatile to force the value not to be cached
+// in a local register or variable by a mutator thread that is checking this value in a loop.
+size_t ShenandoahCollectorPolicy::get_fullgc_count() {
+ return _success_full_gcs + _alloc_failure_degenerated_upgrade_to_full;
+}
+
void ShenandoahCollectorPolicy::print_gc_stats(outputStream* out) const {
out->print_cr("Under allocation pressure, concurrent cycles may cancel, and either continue cycle");
out->print_cr("under stop-the-world pause or result in stop-the-world Full GC. Increase heap size,");
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.hpp b/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.hpp
index 3ca9965dbda..244663e6ae3 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.hpp
@@ -45,9 +45,9 @@ class ShenandoahCollectorPolicy : public CHeapObj<mtGC> {
size_t _success_old_gcs;
size_t _interrupted_old_gcs;
size_t _success_degenerated_gcs;
- size_t _success_full_gcs;
+ volatile size_t _success_full_gcs;
size_t _alloc_failure_degenerated;
- size_t _alloc_failure_degenerated_upgrade_to_full;
+ volatile size_t _alloc_failure_degenerated_upgrade_to_full;
size_t _alloc_failure_full;
size_t _explicit_concurrent;
size_t _explicit_full;
@@ -88,6 +88,8 @@ class ShenandoahCollectorPolicy : public CHeapObj<mtGC> {
size_t cycle_counter() const;
+ size_t get_fullgc_count();
+
void print_gc_stats(outputStream* out) const;
};
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahConcurrentGC.cpp b/src/hotspot/share/gc/shenandoah/shenandoahConcurrentGC.cpp
index 6c165ade4e3..b09b736ce6e 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahConcurrentGC.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahConcurrentGC.cpp
@@ -217,28 +217,56 @@ bool ShenandoahConcurrentGC::collect(GCCause::Cause cause) {
_abbreviated = true;
}
+ // We defer generation resizing actions until after cset regions have been recycled. We do this even following an
+ // abbreviated cycle.
if (heap->mode()->is_generational()) {
+ bool success;
+ size_t region_xfer;
+ const char* region_destination;
+ ShenandoahYoungGeneration* young_gen = heap->young_generation();
+ ShenandoahGeneration* old_gen = heap->old_generation();
{
- ShenandoahYoungGeneration* young_gen = heap->young_generation();
- ShenandoahGeneration* old_gen = heap->old_generation();
ShenandoahHeapLocker locker(heap->lock());
+ size_t old_region_surplus = heap->get_old_region_surplus();
+ size_t old_region_deficit = heap->get_old_region_deficit();
+ if (old_region_surplus) {
+ success = heap->generation_sizer()->transfer_to_young(old_region_surplus);
+ region_destination = "young";
+ region_xfer = old_region_surplus;
+ } else if (old_region_deficit) {
+ success = heap->generation_sizer()->transfer_to_old(old_region_deficit);
+ region_destination = "old";
+ region_xfer = old_region_deficit;
+ if (!success) {
+ ((ShenandoahOldHeuristics *) old_gen->heuristics())->trigger_cannot_expand();
+ }
+ } else {
+ region_destination = "none";
+ region_xfer = 0;
+ success = true;
+ }
+ heap->set_old_region_surplus(0);
+ heap->set_old_region_deficit(0);
+
size_t old_usage_before_evac = heap->capture_old_usage(0);
size_t old_usage_now = old_gen->used();
size_t promoted_bytes = old_usage_now - old_usage_before_evac;
heap->set_previous_promotion(promoted_bytes);
-
- young_gen->unadjust_available();
- old_gen->unadjust_available();
- // No need to old_gen->increase_used().
- // That was done when plabs were allocated, accounting for both old evacs and promotions.
-
- heap->set_alloc_supplement_reserve(0);
heap->set_young_evac_reserve(0);
heap->set_old_evac_reserve(0);
heap->reset_old_evac_expended();
heap->set_promoted_reserve(0);
}
+
+ // Report outside the heap lock
+ size_t young_available = young_gen->available();
+ size_t old_available = old_gen->available();
+ log_info(gc, ergo)("After cleanup, %s " SIZE_FORMAT " regions to %s to prepare for next gc, old available: "
+ SIZE_FORMAT "%s, young_available: " SIZE_FORMAT "%s",
+ success? "successfully transferred": "failed to transfer", region_xfer, region_destination,
+ byte_size_in_proper_unit(old_available), proper_unit_for_byte_size(old_available),
+ byte_size_in_proper_unit(young_available), proper_unit_for_byte_size(young_available));
}
return true;
}
@@ -733,64 +761,95 @@ void ShenandoahConcurrentGC::op_final_mark() {
// been set aside to hold objects evacuated from the young-gen collection set. Conservatively, this value
// equals the entire amount of live young-gen memory within the collection set, even though some of this memory
// will likely be promoted.
- //
- // heap->get_alloc_supplement_reserve() represents the amount of old-gen memory that can be allocated during evacuation
- // and update-refs phases of gc. The young evacuation reserve has already been removed from this quantity.
// Has to be done after cset selection
heap->prepare_concurrent_roots();
- if (!heap->collection_set()->is_empty()) {
- LogTarget(Debug, gc, cset) lt;
- if (lt.is_enabled()) {
- ResourceMark rm;
- LogStream ls(lt);
- heap->collection_set()->print_on(&ls);
- }
+ if (heap->mode()->is_generational()) {
+ ShenandoahGeneration* young_gen = heap->young_generation();
+ size_t humongous_regions_promoted = heap->get_promotable_humongous_regions();
+ size_t regular_regions_promoted_in_place = heap->get_regular_regions_promoted_in_place();
+ if (!heap->collection_set()->is_empty() || (humongous_regions_promoted + regular_regions_promoted_in_place > 0)) {
+ // Even if the collection set is empty, we need to do evacuation if there are regions to be promoted in place.
+ // Concurrent evacuation takes responsibility for registering objects and setting the remembered set cards to dirty.
+
+ LogTarget(Debug, gc, cset) lt;
+ if (lt.is_enabled()) {
+ ResourceMark rm;
+ LogStream ls(lt);
+ heap->collection_set()->print_on(&ls);
+ }
- if (ShenandoahVerify) {
- heap->verifier()->verify_before_evacuation();
- }
+ if (ShenandoahVerify) {
+ heap->verifier()->verify_before_evacuation();
+ }
+ // TODO: we do not need to run update-references following evacuation if collection_set->is_empty().
- heap->set_evacuation_in_progress(true);
- // From here on, we need to update references.
- heap->set_has_forwarded_objects(true);
+ heap->set_evacuation_in_progress(true);
+ // From here on, we need to update references.
+ heap->set_has_forwarded_objects(true);
- // Verify before arming for concurrent processing.
- // Otherwise, verification can trigger stack processing.
- if (ShenandoahVerify) {
- heap->verifier()->verify_during_evacuation();
- }
+ // Verify before arming for concurrent processing.
+ // Otherwise, verification can trigger stack processing.
+ if (ShenandoahVerify) {
+ heap->verifier()->verify_during_evacuation();
+ }
- // Arm nmethods/stack for concurrent processing
- ShenandoahCodeRoots::arm_nmethods();
- ShenandoahStackWatermark::change_epoch_id();
-
- if (heap->mode()->is_generational()) {
- // Calculate the temporary evacuation allowance supplement to young-gen memory capacity (for allocations
- // and young-gen evacuations).
- intptr_t adjustment = heap->get_alloc_supplement_reserve();
- size_t young_available = heap->young_generation()->adjust_available(adjustment);
- // old_available is memory that can hold promotions and evacuations. Subtract out the memory that is being
- // loaned for young-gen allocations or evacuations.
- size_t old_available = heap->old_generation()->adjust_available(-adjustment);
-
- log_info(gc, ergo)("After generational memory budget adjustments, old available: " SIZE_FORMAT
- "%s, young_available: " SIZE_FORMAT "%s",
- byte_size_in_proper_unit(old_available), proper_unit_for_byte_size(old_available),
- byte_size_in_proper_unit(young_available), proper_unit_for_byte_size(young_available));
- }
+ // Arm nmethods/stack for concurrent processing
+ ShenandoahCodeRoots::arm_nmethods();
+ ShenandoahStackWatermark::change_epoch_id();
- if (ShenandoahPacing) {
- heap->pacer()->setup_for_evac();
+ if (ShenandoahPacing) {
+ heap->pacer()->setup_for_evac();
+ }
+ } else {
+ if (ShenandoahVerify) {
+ heap->verifier()->verify_after_concmark();
+ }
+
+ if (VerifyAfterGC) {
+ Universe::verify();
+ }
}
} else {
- if (ShenandoahVerify) {
- heap->verifier()->verify_after_concmark();
- }
+ // Not is_generational()
+ if (!heap->collection_set()->is_empty()) {
+ LogTarget(Info, gc, ergo) lt;
+ if (lt.is_enabled()) {
+ ResourceMark rm;
+ LogStream ls(lt);
+ heap->collection_set()->print_on(&ls);
+ }
- if (VerifyAfterGC) {
- Universe::verify();
+ if (ShenandoahVerify) {
+ heap->verifier()->verify_before_evacuation();
+ }
+
+ heap->set_evacuation_in_progress(true);
+ // From here on, we need to update references.
+ heap->set_has_forwarded_objects(true);
+
+ // Verify before arming for concurrent processing.
+ // Otherwise, verification can trigger stack processing.
+ if (ShenandoahVerify) {
+ heap->verifier()->verify_during_evacuation();
+ }
+
+ // Arm nmethods/stack for concurrent processing
+ ShenandoahCodeRoots::arm_nmethods();
+ ShenandoahStackWatermark::change_epoch_id();
+
+ if (ShenandoahPacing) {
+ heap->pacer()->setup_for_evac();
+ }
+ } else {
+ if (ShenandoahVerify) {
+ heap->verifier()->verify_after_concmark();
+ }
+
+ if (VerifyAfterGC) {
+ Universe::verify();
+ }
}
}
}
@@ -812,6 +871,7 @@ ShenandoahConcurrentEvacThreadClosure::ShenandoahConcurrentEvacThreadClosure(Oop
void ShenandoahConcurrentEvacThreadClosure::do_thread(Thread* thread) {
JavaThread* const jt = JavaThread::cast(thread);
StackWatermarkSet::finish_processing(jt, _oops, StackWatermarkKind::gc);
+ ShenandoahThreadLocalData::enable_plab_promotions(thread);
}
class ShenandoahConcurrentEvacUpdateThreadTask : public WorkerTask {
@@ -825,6 +885,9 @@ class ShenandoahConcurrentEvacUpdateThreadTask : public WorkerTask {
}
void work(uint worker_id) {
+ Thread* worker_thread = Thread::current();
+ ShenandoahThreadLocalData::enable_plab_promotions(worker_thread);
+
// ShenandoahEvacOOMScope has to be setup by ShenandoahContextEvacuateUpdateRootsClosure.
// Otherwise, may deadlock with watermark lock
ShenandoahContextEvacuateUpdateRootsClosure oops_cl;
@@ -1200,7 +1263,6 @@ void ShenandoahConcurrentGC::op_final_updaterefs() {
Universe::verify();
}
- heap->adjust_generation_sizes();
heap->rebuild_free_set(true /*concurrent*/);
}
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahControlThread.cpp b/src/hotspot/share/gc/shenandoah/shenandoahControlThread.cpp
index 144d9095b0b..09e15a09568 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahControlThread.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahControlThread.cpp
@@ -113,6 +113,7 @@ void ShenandoahControlThread::run_service() {
// degenerated cycle should be 'promoted' to a full cycle. The decision to
// trigger a cycle or not is evaluated on the regulator thread.
ShenandoahHeuristics* global_heuristics = heap->global_generation()->heuristics();
+ bool old_bootstrap_requested = false;
while (!in_graceful_shutdown() && !should_terminate()) {
// Figure out if we have pending requests.
bool alloc_failure_pending = _alloc_failure_gc.is_set();
@@ -207,10 +208,19 @@ void ShenandoahControlThread::run_service() {
// the heuristic to run a young collection so that we can evacuate some old regions.
assert(!heap->is_concurrent_old_mark_in_progress(), "Should not be running mixed collections and concurrent marking");
generation = YOUNG;
+ } else if (_requested_generation == OLD && !old_bootstrap_requested) {
+ // Arrange to perform a young GC immediately followed by a bootstrap OLD GC. OLD GC typically requires more
+ // than twice the time required for YOUNG GC, so we run a YOUNG GC to replenish the YOUNG allocation pool before
+ // we start the longer OLD GC effort.
+ old_bootstrap_requested = true;
+ generation = YOUNG;
} else {
+ // if (old_bootstrap_requested && (_requested_generation == OLD)), this starts the bootstrap GC that
+ // immediately follows the preparatory young GC.
+ // But we will abandon the planned bootstrap GC if a GLOBAL GC has been now been requested.
generation = _requested_generation;
+ old_bootstrap_requested = false;
}
-
// preemption was requested or this is a regular cycle
cause = GCCause::_shenandoah_concurrent_gc;
set_gc_mode(default_mode);
@@ -391,10 +401,18 @@ void ShenandoahControlThread::run_service() {
// Don't wait around if there was an allocation failure - start the next cycle immediately.
if (!is_alloc_failure_gc()) {
- // The timed wait is necessary because this thread has a responsibility to send
- // 'alloc_words' to the pacer when it does not perform a GC.
- MonitorLocker lock(&_control_lock, Mutex::_no_safepoint_check_flag);
- lock.wait(ShenandoahControlIntervalMax);
+ if (old_bootstrap_requested) {
+ _requested_generation = OLD;
+ _requested_gc_cause = GCCause::_shenandoah_concurrent_gc;
+ } else {
+ // The timed wait is necessary because this thread has a responsibility to send
+ // 'alloc_words' to the pacer when it does not perform a GC.
+ MonitorLocker lock(&_control_lock, Mutex::_no_safepoint_check_flag);
+ lock.wait(ShenandoahControlIntervalMax);
+ }
+ } else {
+ // in case of alloc_failure, abandon any plans to do immediate OLD Bootstrap
+ old_bootstrap_requested = false;
}
}
@@ -457,10 +475,11 @@ void ShenandoahControlThread::process_phase_timings(const ShenandoahHeap* heap)
// | v v |
// +---> Global Degen +--------------------> Full <----+
//
-void ShenandoahControlThread::service_concurrent_normal_cycle(const ShenandoahHeap* heap,
+void ShenandoahControlThread::service_concurrent_normal_cycle(ShenandoahHeap* heap,
const ShenandoahGenerationType generation,
GCCause::Cause cause) {
GCIdMark gc_id_mark;
+ ShenandoahGeneration* the_generation = nullptr;
switch (generation) {
case YOUNG: {
// Run a young cycle. This might or might not, have interrupted an ongoing
@@ -469,44 +488,34 @@ void ShenandoahControlThread::service_concurrent_normal_cycle(const ShenandoahHe
// they end up in, but we have to be sure we don't promote into any regions
// that are in the cset.
log_info(gc, ergo)("Start GC cycle (YOUNG)");
- service_concurrent_cycle(heap->young_generation(), cause, false);
+ the_generation = heap->young_generation();
+ service_concurrent_cycle(the_generation, cause, false);
break;
}
case OLD: {
log_info(gc, ergo)("Start GC cycle (OLD)");
+ the_generation = heap->old_generation();
service_concurrent_old_cycle(heap, cause);
break;
}
case GLOBAL_GEN: {
log_info(gc, ergo)("Start GC cycle (GLOBAL)");
- service_concurrent_cycle(heap->global_generation(), cause, false);
+ the_generation = heap->global_generation();
+ service_concurrent_cycle(the_generation, cause, false);
break;
}
case GLOBAL_NON_GEN: {
log_info(gc, ergo)("Start GC cycle");
- service_concurrent_cycle(heap->global_generation(), cause, false);
+ the_generation = heap->global_generation();
+ service_concurrent_cycle(the_generation, cause, false);
break;
}
default:
ShouldNotReachHere();
}
- const char* msg;
- if (heap->mode()->is_generational()) {
- if (heap->cancelled_gc()) {
- msg = (generation == YOUNG) ? "At end of Interrupted Concurrent Young GC" :
- "At end of Interrupted Concurrent Bootstrap GC";
- } else {
- msg = (generation == YOUNG) ? "At end of Concurrent Young GC" :
- "At end of Concurrent Bootstrap GC";
- }
- } else {
- msg = heap->cancelled_gc() ? "At end of cancelled GC" :
- "At end of GC";
- }
- heap->log_heap_status(msg);
}
-void ShenandoahControlThread::service_concurrent_old_cycle(const ShenandoahHeap* heap, GCCause::Cause &cause) {
+void ShenandoahControlThread::service_concurrent_old_cycle(ShenandoahHeap* heap, GCCause::Cause &cause) {
ShenandoahOldGeneration* old_generation = heap->old_generation();
ShenandoahYoungGeneration* young_generation = heap->young_generation();
ShenandoahOldGeneration::State original_state = old_generation->state();
@@ -562,7 +571,7 @@ void ShenandoahControlThread::service_concurrent_old_cycle(const ShenandoahHeap*
set_gc_mode(bootstrapping_old);
young_generation->set_old_gen_task_queues(old_generation->task_queues());
ShenandoahGCSession session(cause, young_generation);
- service_concurrent_cycle(heap,young_generation, cause, true);
+ service_concurrent_cycle(heap, young_generation, cause, true);
process_phase_timings(heap);
if (heap->cancelled_gc()) {
// Young generation bootstrap cycle has failed. Concurrent mark for old generation
@@ -588,9 +597,13 @@ void ShenandoahControlThread::service_concurrent_old_cycle(const ShenandoahHeap*
if (marking_complete) {
assert(old_generation->state() != ShenandoahOldGeneration::MARKING, "Should not still be marking");
if (original_state == ShenandoahOldGeneration::MARKING) {
+ heap->mmu_tracker()->record_old_marking_increment(old_generation, GCId::current(), true,
+ heap->collection_set()->has_old_regions());
heap->log_heap_status("At end of Concurrent Old Marking finishing increment");
}
} else if (original_state == ShenandoahOldGeneration::MARKING) {
+ heap->mmu_tracker()->record_old_marking_increment(old_generation, GCId::current(), false,
+ heap->collection_set()->has_old_regions());
heap->log_heap_status("At end of Concurrent Old Marking increment");
}
break;
@@ -697,7 +710,7 @@ void ShenandoahControlThread::service_concurrent_cycle(ShenandoahGeneration* gen
service_concurrent_cycle(heap, generation, cause, do_old_gc_bootstrap);
}
-void ShenandoahControlThread::service_concurrent_cycle(const ShenandoahHeap* heap,
+void ShenandoahControlThread::service_concurrent_cycle(ShenandoahHeap* heap,
ShenandoahGeneration* generation,
GCCause::Cause& cause,
bool do_old_gc_bootstrap) {
@@ -713,6 +726,32 @@ void ShenandoahControlThread::service_concurrent_cycle(const ShenandoahHeap* hea
// collection. Same for global collections.
_degen_generation = generation;
}
+ const char* msg;
+ if (heap->mode()->is_generational()) {
+ if (heap->cancelled_gc()) {
+ msg = (generation->is_young()) ? "At end of Interrupted Concurrent Young GC" :
+ "At end of Interrupted Concurrent Bootstrap GC";
+ } else {
+ msg = (generation->is_young()) ? "At end of Concurrent Young GC" :
+ "At end of Concurrent Bootstrap GC";
+ // We only record GC results if GC was successful
+ ShenandoahMmuTracker* mmu_tracker = heap->mmu_tracker();
+ if (generation->is_young()) {
+ if (heap->collection_set()->has_old_regions()) {
+ bool mixed_is_done = (heap->old_heuristics()->unprocessed_old_collection_candidates() == 0);
+ mmu_tracker->record_mixed(generation, get_gc_id(), mixed_is_done);
+ } else {
+ mmu_tracker->record_young(generation, get_gc_id());
+ }
+ } else {
+ mmu_tracker->record_bootstrap(generation, get_gc_id(), heap->collection_set()->has_old_regions());
+ }
+ }
+ } else {
+ msg = heap->cancelled_gc() ? "At end of cancelled GC" :
+ "At end of GC";
+ }
+ heap->log_heap_status(msg);
}
bool ShenandoahControlThread::check_cancellation_or_degen(ShenandoahGC::ShenandoahDegenPoint point) {
@@ -928,7 +967,6 @@ void ShenandoahControlThread::handle_alloc_failure(ShenandoahAllocRequest& req)
log_info(gc)("Failed to allocate %s, " SIZE_FORMAT "%s",
req.type_string(),
byte_size_in_proper_unit(req.size() * HeapWordSize), proper_unit_for_byte_size(req.size() * HeapWordSize));
-
// Now that alloc failure GC is scheduled, we can abort everything else
heap->cancel_gc(GCCause::_allocation_failure);
}
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahControlThread.hpp b/src/hotspot/share/gc/shenandoah/shenandoahControlThread.hpp
index f9596feeeb7..339cfa7d5de 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahControlThread.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahControlThread.hpp
@@ -88,6 +88,7 @@ class ShenandoahControlThread: public ConcurrentGCThread {
ShenandoahSharedFlag _preemption_requested;
ShenandoahSharedFlag _gc_requested;
ShenandoahSharedFlag _alloc_failure_gc;
+ ShenandoahSharedFlag _humongous_alloc_failure_gc;
ShenandoahSharedFlag _graceful_shutdown;
ShenandoahSharedFlag _do_counters_update;
ShenandoahSharedFlag _force_counters_update;
@@ -173,11 +174,11 @@ class ShenandoahControlThread: public ConcurrentGCThread {
void prepare_for_graceful_shutdown();
bool in_graceful_shutdown();
- void service_concurrent_normal_cycle(const ShenandoahHeap* heap,
+ void service_concurrent_normal_cycle(ShenandoahHeap* heap,
const ShenandoahGenerationType generation,
GCCause::Cause cause);
- void service_concurrent_old_cycle(const ShenandoahHeap* heap,
+ void service_concurrent_old_cycle(ShenandoahHeap* heap,
GCCause::Cause &cause);
void set_gc_mode(GCMode new_mode);
@@ -191,7 +192,7 @@ class ShenandoahControlThread: public ConcurrentGCThread {
static const char* gc_mode_name(GCMode mode);
void notify_control_thread();
- void service_concurrent_cycle(const ShenandoahHeap* heap,
+ void service_concurrent_cycle(ShenandoahHeap* heap,
ShenandoahGeneration* generation,
GCCause::Cause &cause,
bool do_old_gc_bootstrap);
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahDegeneratedGC.cpp b/src/hotspot/share/gc/shenandoah/shenandoahDegeneratedGC.cpp
index fa62f7a9fe8..0a7354b17be 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahDegeneratedGC.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahDegeneratedGC.cpp
@@ -57,7 +57,11 @@ bool ShenandoahDegenGC::collect(GCCause::Cause cause) {
vmop_degenerated();
ShenandoahHeap* heap = ShenandoahHeap::heap();
if (heap->mode()->is_generational()) {
- heap->log_heap_status("At end of Degenerated GC");
+ bool is_bootstrap_gc = heap->is_concurrent_old_mark_in_progress() && _generation->is_young();
+ heap->mmu_tracker()->record_degenerated(_generation, GCId::current(), is_bootstrap_gc,
+ !heap->collection_set()->has_old_regions());
+ const char* msg = is_bootstrap_gc? "At end of Degenerated Boostrap Old GC": "At end of Degenerated GC";
+ heap->log_heap_status(msg);
}
return true;
}
@@ -272,27 +276,53 @@ void ShenandoahDegenGC::op_degenerated() {
}
op_cleanup_complete();
+ // We defer generation resizing actions until after cset regions have been recycled.
+ if (heap->mode()->is_generational()) {
+ size_t old_region_surplus = heap->get_old_region_surplus();
+ size_t old_region_deficit = heap->get_old_region_deficit();
+ bool success;
+ size_t region_xfer;
+ const char* region_destination;
+ if (old_region_surplus) {
+ region_xfer = old_region_surplus;
+ region_destination = "young";
+ success = heap->generation_sizer()->transfer_to_young(old_region_surplus);
+ } else if (old_region_deficit) {
+ region_xfer = old_region_surplus;
+ region_destination = "old";
+ success = heap->generation_sizer()->transfer_to_old(old_region_deficit);
+ if (!success) {
+ ((ShenandoahOldHeuristics *) heap->old_generation()->heuristics())->trigger_cannot_expand();
+ }
+ } else {
+ region_destination = "none";
+ region_xfer = 0;
+ success = true;
+ }
+
+ size_t young_available = heap->young_generation()->available();
+ size_t old_available = heap->old_generation()->available();
+ log_info(gc, ergo)("After cleanup, %s " SIZE_FORMAT " regions to %s to prepare for next gc, old available: "
+ SIZE_FORMAT "%s, young_available: " SIZE_FORMAT "%s",
+ success? "successfully transferred": "failed to transfer", region_xfer, region_destination,
+ byte_size_in_proper_unit(old_available), proper_unit_for_byte_size(old_available),
+ byte_size_in_proper_unit(young_available), proper_unit_for_byte_size(young_available));
+
+ heap->set_old_region_surplus(0);
+ heap->set_old_region_deficit(0);
+ }
break;
default:
ShouldNotReachHere();
}
if (heap->mode()->is_generational()) {
- // In case degeneration interrupted concurrent evacuation or update references,
- // we need to clean up transient state. Otherwise, these actions have no effect.
-
- heap->young_generation()->unadjust_available();
- heap->old_generation()->unadjust_available();
- // No need to old_gen->increase_used(). That was done when plabs were allocated,
- // accounting for both old evacs and promotions.
-
- heap->set_alloc_supplement_reserve(0);
+ // In case degeneration interrupted concurrent evacuation or update references, we need to clean up transient state.
+ // Otherwise, these actions have no effect.
heap->set_young_evac_reserve(0);
heap->set_old_evac_reserve(0);
heap->reset_old_evac_expended();
heap->set_promoted_reserve(0);
-
- heap->adjust_generation_sizes();
}
if (ShenandoahVerify) {
@@ -354,7 +384,16 @@ void ShenandoahDegenGC::op_prepare_evacuation() {
heap->tlabs_retire(false);
}
- if (!heap->collection_set()->is_empty()) {
+ size_t humongous_regions_promoted = heap->get_promotable_humongous_regions();
+ size_t regular_regions_promoted_in_place = heap->get_regular_regions_promoted_in_place();
+ if (!heap->collection_set()->is_empty() || (humongous_regions_promoted + regular_regions_promoted_in_place > 0)) {
+ // Even if the collection set is empty, we need to do evacuation if there are regions to be promoted in place.
+ // Degenerated evacuation takes responsibility for registering objects and setting the remembered set cards to dirty.
+
+ if (ShenandoahVerify) {
+ heap->verifier()->verify_before_evacuation();
+ }
+
heap->set_evacuation_in_progress(true);
heap->set_has_forwarded_objects(true);
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.cpp b/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.cpp
index 3ac75079411..c8739734597 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.cpp
@@ -38,7 +38,6 @@
#include "memory/resourceArea.hpp"
#include "runtime/orderAccess.hpp"
-
ShenandoahSetsOfFree::ShenandoahSetsOfFree(size_t max_regions, ShenandoahFreeSet* free_set) :
_max(max_regions),
_free_set(free_set),
@@ -162,9 +161,16 @@ void ShenandoahSetsOfFree::move_to_set(size_t idx, ShenandoahFreeMemoryType new_
// During flip_to_gc:
// Mutator empty => Collector
// Mutator empty => Old Collector
- assert (((region_capacity < _region_size_bytes) && (orig_set == Mutator) && (new_set == Collector)) ||
- ((region_capacity == _region_size_bytes) && (orig_set == Mutator) && (new_set == Collector || new_set == OldCollector)),
- "Unexpected movement between sets");
+ // At start of update refs:
+ // Collector => Mutator
+ // OldCollector Empty => Mutator
+ assert (((region_capacity <= _region_size_bytes) &&
+ ((orig_set == Mutator) && (new_set == Collector)) ||
+ ((orig_set == Collector) && (new_set == Mutator))) ||
+ ((region_capacity == _region_size_bytes) &&
+ ((orig_set == Mutator) && (new_set == Collector)) ||
+ ((orig_set == OldCollector) && (new_set == Mutator)) ||
+ (new_set == OldCollector)), "Unexpected movement between sets");
_membership[idx] = new_set;
_capacity_of[orig_set] -= region_capacity;
@@ -398,7 +404,6 @@ void ShenandoahSetsOfFree::assert_bounds() {
}
#endif
-
ShenandoahFreeSet::ShenandoahFreeSet(ShenandoahHeap* heap, size_t max_regions) :
_heap(heap),
_free_sets(max_regions, this)
@@ -412,6 +417,7 @@ ShenandoahFreeSet::ShenandoahFreeSet(ShenandoahHeap* heap, size_t max_regions) :
HeapWord* ShenandoahFreeSet::allocate_old_with_affiliation(ShenandoahAffiliation affiliation,
ShenandoahAllocRequest& req, bool& in_new_region) {
shenandoah_assert_heaplocked();
+
size_t rightmost =
(affiliation == ShenandoahAffiliation::FREE)? _free_sets.rightmost_empty(OldCollector): _free_sets.rightmost(OldCollector);
size_t leftmost =
@@ -450,7 +456,19 @@ HeapWord* ShenandoahFreeSet::allocate_old_with_affiliation(ShenandoahAffiliation
return nullptr;
}
-HeapWord* ShenandoahFreeSet::allocate_with_affiliation(ShenandoahAffiliation affiliation, ShenandoahAllocRequest& req, bool& in_new_region) {
+void ShenandoahFreeSet::add_old_collector_free_region(ShenandoahHeapRegion* region) {
+ shenandoah_assert_heaplocked();
+ size_t idx = region->index();
+ size_t capacity = alloc_capacity(region);
+ assert(_free_sets.membership(idx) == NotFree, "Regions promoted in place should not be in any free set");
+ if (capacity >= PLAB::min_size() * HeapWordSize) {
+ _free_sets.make_free(idx, OldCollector, capacity);
+ _heap->augment_promo_reserve(capacity);
+ }
+}
+
+HeapWord* ShenandoahFreeSet::allocate_with_affiliation(ShenandoahAffiliation affiliation,
+ ShenandoahAllocRequest& req, bool& in_new_region) {
shenandoah_assert_heaplocked();
size_t rightmost =
(affiliation == ShenandoahAffiliation::FREE)? _free_sets.rightmost_empty(Collector): _free_sets.rightmost(Collector);
@@ -469,7 +487,8 @@ HeapWord* ShenandoahFreeSet::allocate_with_affiliation(ShenandoahAffiliation aff
}
}
}
- log_debug(gc, free)("Could not allocate collector region with affiliation: %s for request " PTR_FORMAT, shenandoah_affiliation_name(affiliation), p2i(&req));
+ log_debug(gc, free)("Could not allocate collector region with affiliation: %s for request " PTR_FORMAT,
+ shenandoah_affiliation_name(affiliation), p2i(&req));
return nullptr;
}
@@ -494,15 +513,15 @@ HeapWord* ShenandoahFreeSet::allocate_single(ShenandoahAllocRequest& req, bool&
if (_heap->mode()->is_generational()) {
switch (req.affiliation()) {
case ShenandoahAffiliation::OLD_GENERATION:
- // Note: unsigned result from adjusted_unaffiliated_regions() will never be less than zero, but it may equal zero.
- if (_heap->old_generation()->adjusted_unaffiliated_regions() <= 0) {
+ // Note: unsigned result from free_unaffiliated_regions() will never be less than zero, but it may equal zero.
+ if (_heap->old_generation()->free_unaffiliated_regions() <= 0) {
allow_new_region = false;
}
break;
case ShenandoahAffiliation::YOUNG_GENERATION:
- // Note: unsigned result from adjusted_unaffiliated_regions() will never be less than zero, but it may equal zero.
- if (_heap->young_generation()->adjusted_unaffiliated_regions() <= 0) {
+ // Note: unsigned result from free_unaffiliated_regions() will never be less than zero, but it may equal zero.
+ if (_heap->young_generation()->free_unaffiliated_regions() <= 0) {
allow_new_region = false;
}
break;
@@ -515,7 +534,6 @@ HeapWord* ShenandoahFreeSet::allocate_single(ShenandoahAllocRequest& req, bool&
break;
}
}
-
switch (req.type()) {
case ShenandoahAllocRequest::_alloc_tlab:
case ShenandoahAllocRequest::_alloc_shared: {
@@ -524,8 +542,9 @@ HeapWord* ShenandoahFreeSet::allocate_single(ShenandoahAllocRequest& req, bool&
ShenandoahHeapRegion* r = _heap->get_region(idx);
if (_free_sets.in_free_set(idx, Mutator) && (allow_new_region || r->is_affiliated())) {
// try_allocate_in() increases used if the allocation is successful.
- HeapWord* result = try_allocate_in(r, req, in_new_region);
- if (result != nullptr) {
+ HeapWord* result;
+ size_t min_size = (req.type() == ShenandoahAllocRequest::_alloc_tlab)? req.min_size(): req.size();
+ if ((alloc_capacity(r) >= min_size) && ((result = try_allocate_in(r, req, in_new_region)) != nullptr)) {
return result;
}
}
@@ -576,19 +595,16 @@ HeapWord* ShenandoahFreeSet::allocate_single(ShenandoahAllocRequest& req, bool&
}
}
}
-
// No dice. Can we borrow space from mutator view?
if (!ShenandoahEvacReserveOverflow) {
return nullptr;
}
- // TODO:
- // if (!allow_new_region && req.is_old() && (young_generation->adjusted_unaffiliated_regions() > 0)) {
- // transfer a region from young to old;
- // allow_new_region = true;
- // heap->set_old_evac_reserve(heap->get_old_evac_reserve() + region_size_bytes);
- // }
- //
+ if (!allow_new_region && req.is_old() && (_heap->young_generation()->free_unaffiliated_regions() > 0)) {
+ // This allows us to flip a mutator region to old_collector
+ allow_new_region = true;
+ }
+
// We should expand old-gen if this can prevent an old-gen evacuation failure. We don't care so much about
// promotion failures since they can be mitigated in a subsequent GC pass. Would be nice to know if this
// allocation request is for evacuation or promotion. Individual threads limit their use of PLAB memory for
@@ -706,6 +722,9 @@ HeapWord* ShenandoahFreeSet::try_allocate_in(ShenandoahHeapRegion* r, Shenandoah
// coalesce-and-fill processing.
r->end_preemptible_coalesce_and_fill();
_heap->clear_cards_for(r);
+ _heap->old_generation()->increment_affiliated_region_count();
+ } else {
+ _heap->young_generation()->increment_affiliated_region_count();
}
assert(ctx->top_at_mark_start(r) == r->bottom(), "Newly established allocation region starts with TAMS equal to bottom");
@@ -727,6 +746,8 @@ HeapWord* ShenandoahFreeSet::try_allocate_in(ShenandoahHeapRegion* r, Shenandoah
// req.size() is in words, r->free() is in bytes.
if (ShenandoahElasticTLAB && req.is_lab_alloc()) {
if (req.type() == ShenandoahAllocRequest::_alloc_plab) {
+ assert(_heap->mode()->is_generational(), "PLABs are only for generational mode");
+ assert(_free_sets.in_free_set(r->index(), OldCollector), "PLABS must be allocated in old_collector_free regions");
// Need to assure that plabs are aligned on multiple of card region.
// Since we have Elastic TLABs, align sizes up. They may be decreased to fit in the usable
// memory remaining in the region (which will also be aligned to cards).
@@ -838,7 +859,7 @@ HeapWord* ShenandoahFreeSet::allocate_contiguous(ShenandoahAllocRequest& req) {
// Check if there are enough regions left to satisfy allocation.
if (_heap->mode()->is_generational()) {
- size_t avail_young_regions = generation->adjusted_unaffiliated_regions();
+ size_t avail_young_regions = generation->free_unaffiliated_regions();
if (num > _free_sets.count(Mutator) || (num > avail_young_regions)) {
return nullptr;
}
@@ -908,6 +929,8 @@ HeapWord* ShenandoahFreeSet::allocate_contiguous(ShenandoahAllocRequest& req) {
// While individual regions report their true use, all humongous regions are marked used in the free set.
_free_sets.remove_from_free_sets(r->index());
}
+ _heap->young_generation()->increase_affiliated_region_count(num);
+
size_t total_humongous_size = ShenandoahHeapRegion::region_size_bytes() * num;
_free_sets.increase_used(Mutator, total_humongous_size);
_free_sets.assert_bounds();
@@ -925,6 +948,11 @@ bool ShenandoahFreeSet::can_allocate_from(ShenandoahHeapRegion *r) const {
return r->is_empty() || (r->is_trash() && !_heap->is_concurrent_weak_root_in_progress());
}
+bool ShenandoahFreeSet::can_allocate_from(size_t idx) const {
+ ShenandoahHeapRegion* r = _heap->get_region(idx);
+ return can_allocate_from(r);
+}
+
size_t ShenandoahFreeSet::alloc_capacity(size_t idx) const {
ShenandoahHeapRegion* r = _heap->get_region(idx);
return alloc_capacity(r);
@@ -976,14 +1004,15 @@ void ShenandoahFreeSet::flip_to_old_gc(ShenandoahHeapRegion* r) {
size_t idx = r->index();
assert(_free_sets.in_free_set(idx, Mutator), "Should be in mutator view");
+ // Note: can_allocate_from(r) means r is entirely empty
assert(can_allocate_from(r), "Should not be allocated");
size_t region_capacity = alloc_capacity(r);
_free_sets.move_to_set(idx, OldCollector, region_capacity);
_free_sets.assert_bounds();
-
- // We do not ensure that the region is no longer trash,
- // relying on try_allocate_in(), which always comes next,
+ _heap->generation_sizer()->force_transfer_to_old(1);
+ _heap->augment_old_evac_reserve(region_capacity);
+ // We do not ensure that the region is no longer trash, relying on try_allocate_in(), which always comes next,
// to recycle trash before attempting to allocate anything in the region.
}
@@ -997,8 +1026,7 @@ void ShenandoahFreeSet::flip_to_gc(ShenandoahHeapRegion* r) {
_free_sets.move_to_set(idx, Collector, region_capacity);
_free_sets.assert_bounds();
- // We do not ensure that the region is no longer trash,
- // relying on try_allocate_in(), which always comes next,
+ // We do not ensure that the region is no longer trash, relying on try_allocate_in(), which always comes next,
// to recycle trash before attempting to allocate anything in the region.
}
@@ -1016,10 +1044,21 @@ void ShenandoahFreeSet::clear_internal() {
// move some of the mutator regions into the collector set or old_collector set with the intent of packing
// old_collector memory into the highest (rightmost) addresses of the heap and the collector memory into the
// next highest addresses of the heap, with mutator memory consuming the lowest addresses of the heap.
-void ShenandoahFreeSet::find_regions_with_alloc_capacity() {
+void ShenandoahFreeSet::find_regions_with_alloc_capacity(size_t &young_cset_regions, size_t &old_cset_regions) {
+ old_cset_regions = 0;
+ young_cset_regions = 0;
for (size_t idx = 0; idx < _heap->num_regions(); idx++) {
ShenandoahHeapRegion* region = _heap->get_region(idx);
+ if (region->is_trash()) {
+ // Trashed regions represent regions that had been in the collection set but have not yet been "cleaned up".
+ if (region->is_old()) {
+ old_cset_regions++;
+ } else {
+ assert(region->is_young(), "Trashed region should be old or young");
+ young_cset_regions++;
+ }
+ }
if (region->is_alloc_allowed() || region->is_trash()) {
assert(!region->is_cset(), "Shouldn't be adding cset regions to the free set");
assert(_free_sets.in_free_set(idx, NotFree), "We are about to make region free; it should not be free already");
@@ -1044,20 +1083,124 @@ void ShenandoahFreeSet::find_regions_with_alloc_capacity() {
}
}
-void ShenandoahFreeSet::rebuild() {
+// Move no more than cset_regions from the existing Collector and OldCollector free sets to the Mutator free set.
+// This is called from outside the heap lock.
+void ShenandoahFreeSet::move_collector_sets_to_mutator(size_t max_xfer_regions) {
+ size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+ size_t collector_empty_xfer = 0;
+ size_t collector_not_empty_xfer = 0;
+ size_t old_collector_empty_xfer = 0;
+
+ // Process empty regions within the Collector free set
+ if ((max_xfer_regions > 0) && (_free_sets.leftmost_empty(Collector) <= _free_sets.rightmost_empty(Collector))) {
+ ShenandoahHeapLocker locker(_heap->lock());
+ for (size_t idx = _free_sets.leftmost_empty(Collector);
+ (max_xfer_regions > 0) && (idx <= _free_sets.rightmost_empty(Collector)); idx++) {
+ if (_free_sets.in_free_set(idx, Collector) && can_allocate_from(idx)) {
+ _free_sets.move_to_set(idx, Mutator, region_size_bytes);
+ max_xfer_regions--;
+ collector_empty_xfer += region_size_bytes;
+ }
+ }
+ }
+
+ // Process empty regions within the OldCollector free set
+ size_t old_collector_regions = 0;
+ if ((max_xfer_regions > 0) && (_free_sets.leftmost_empty(OldCollector) <= _free_sets.rightmost_empty(OldCollector))) {
+ ShenandoahHeapLocker locker(_heap->lock());
+ for (size_t idx = _free_sets.leftmost_empty(OldCollector);
+ (max_xfer_regions > 0) && (idx <= _free_sets.rightmost_empty(OldCollector)); idx++) {
+ if (_free_sets.in_free_set(idx, OldCollector) && can_allocate_from(idx)) {
+ _free_sets.move_to_set(idx, Mutator, region_size_bytes);
+ max_xfer_regions--;
+ old_collector_empty_xfer += region_size_bytes;
+ old_collector_regions++;
+ }
+ }
+ if (old_collector_regions > 0) {
+ _heap->generation_sizer()->transfer_to_young(old_collector_regions);
+ }
+ }
+
+ // If there are any non-empty regions within Collector set, we can also move them to the Mutator free set
+ if ((max_xfer_regions > 0) && (_free_sets.leftmost(Collector) <= _free_sets.rightmost(Collector))) {
+ ShenandoahHeapLocker locker(_heap->lock());
+ for (size_t idx = _free_sets.leftmost(Collector); (max_xfer_regions > 0) && (idx <= _free_sets.rightmost(Collector)); idx++) {
+ size_t alloc_capacity = this->alloc_capacity(idx);
+ if (_free_sets.in_free_set(idx, Collector) && (alloc_capacity > 0)) {
+ _free_sets.move_to_set(idx, Mutator, alloc_capacity);
+ max_xfer_regions--;
+ collector_not_empty_xfer += alloc_capacity;
+ }
+ }
+ }
+
+ size_t collector_xfer = collector_empty_xfer + collector_not_empty_xfer;
+ size_t total_xfer = collector_xfer + old_collector_empty_xfer;
+ log_info(gc, free)("At start of update refs, moving " SIZE_FORMAT "%s to Mutator free set from Collector Reserve ("
+ SIZE_FORMAT "%s) and from Old Collector Reserve (" SIZE_FORMAT "%s)",
+ byte_size_in_proper_unit(total_xfer), proper_unit_for_byte_size(total_xfer),
+ byte_size_in_proper_unit(collector_xfer), proper_unit_for_byte_size(collector_xfer),
+ byte_size_in_proper_unit(old_collector_empty_xfer), proper_unit_for_byte_size(old_collector_empty_xfer));
+}
+
+
+// Overwrite arguments to represent the amount of memory in each generation that is about to be recycled
+void ShenandoahFreeSet::prepare_to_rebuild(size_t &young_cset_regions, size_t &old_cset_regions) {
shenandoah_assert_heaplocked();
// This resets all state information, removing all regions from all sets.
clear();
-
log_debug(gc, free)("Rebuilding FreeSet");
// This places regions that have alloc_capacity into the old_collector set if they identify as is_old() or the
// mutator set otherwise.
- find_regions_with_alloc_capacity();
+ find_regions_with_alloc_capacity(young_cset_regions, old_cset_regions);
+}
+
+void ShenandoahFreeSet::rebuild(size_t young_cset_regions, size_t old_cset_regions) {
+ shenandoah_assert_heaplocked();
+ size_t young_reserve, old_reserve;
+ size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+
+ size_t old_capacity = _heap->old_generation()->max_capacity();
+ size_t old_available = _heap->old_generation()->available();
+ size_t old_unaffiliated_regions = _heap->old_generation()->free_unaffiliated_regions();
+ size_t young_capacity = _heap->young_generation()->max_capacity();
+ size_t young_available = _heap->young_generation()->available();
+ size_t young_unaffiliated_regions = _heap->young_generation()->free_unaffiliated_regions();
+
+ old_unaffiliated_regions += old_cset_regions;
+ old_available += old_cset_regions * region_size_bytes;
+ young_unaffiliated_regions += young_cset_regions;
+ young_available += young_cset_regions * region_size_bytes;
+
+ // Consult old-region surplus and deficit to make adjustments to current generation capacities and availability.
+ // The generation region transfers take place after we rebuild.
+ size_t old_region_surplus = _heap->get_old_region_surplus();
+ size_t old_region_deficit = _heap->get_old_region_deficit();
+
+ if (old_region_surplus > 0) {
+ size_t xfer_bytes = old_region_surplus * region_size_bytes;
+ assert(old_region_surplus <= old_unaffiliated_regions, "Cannot transfer regions that are affiliated");
+ old_capacity -= xfer_bytes;
+ old_available -= xfer_bytes;
+ old_unaffiliated_regions -= old_region_surplus;
+ young_capacity += xfer_bytes;
+ young_available += xfer_bytes;
+ young_unaffiliated_regions += old_region_surplus;
+ } else if (old_region_deficit > 0) {
+ size_t xfer_bytes = old_region_deficit * region_size_bytes;
+ assert(old_region_deficit <= young_unaffiliated_regions, "Cannot transfer regions that are affiliated");
+ old_capacity += xfer_bytes;
+ old_available += xfer_bytes;
+ old_unaffiliated_regions += old_region_deficit;
+ young_capacity -= xfer_bytes;;
+ young_available -= xfer_bytes;
+ young_unaffiliated_regions -= old_region_deficit;
+ }
// Evac reserve: reserve trailing space for evacuations, with regions reserved for old evacuations placed to the right
// of regions reserved of young evacuations.
- size_t young_reserve, old_reserve;
if (!_heap->mode()->is_generational()) {
young_reserve = (_heap->max_capacity() / 100) * ShenandoahEvacReserve;
old_reserve = 0;
@@ -1070,15 +1213,32 @@ void ShenandoahFreeSet::rebuild() {
// We are rebuilding at the end of final mark, having already established evacuation budgets for this GC pass.
young_reserve = _heap->get_young_evac_reserve();
old_reserve = _heap->get_promoted_reserve() + _heap->get_old_evac_reserve();
+ assert(old_reserve <= old_available,
+ "Cannot reserve (" SIZE_FORMAT " + " SIZE_FORMAT") more OLD than is available: " SIZE_FORMAT,
+ _heap->get_promoted_reserve(), _heap->get_old_evac_reserve(), old_available);
} else {
// We are rebuilding at end of GC, so we set aside budgets specified on command line (or defaults)
- young_reserve = (_heap->young_generation()->max_capacity() * ShenandoahEvacReserve) / 100;
- old_reserve = MAX2((_heap->old_generation()->max_capacity() * ShenandoahOldEvacReserve) / 100,
- ShenandoahOldCompactionReserve * ShenandoahHeapRegion::region_size_bytes());
+ young_reserve = (young_capacity * ShenandoahEvacReserve) / 100;
+ // The auto-sizer has already made old-gen large enough to hold all anticipated evacuations and promotions.
+ // Affiliated old-gen regions are already in the OldCollector free set. Add in the relevant number of
+ // unaffiliated regions.
+ old_reserve = old_available;
}
}
- reserve_regions(young_reserve, old_reserve);
+ if (old_reserve > _free_sets.capacity_of(OldCollector)) {
+ // Old available regions that have less than PLAB::min_size() of available memory are not placed into the OldCollector
+ // free set. Because of this, old_available may not have enough memory to represent the intended reserve. Adjust
+ // the reserve downward to account for this possibility. This loss is part of the reason why the original budget
+ // was adjusted with ShenandoahOldEvacWaste and ShenandoahOldPromoWaste multipliers.
+ if (old_reserve > _free_sets.capacity_of(OldCollector) + old_unaffiliated_regions * region_size_bytes) {
+ old_reserve = _free_sets.capacity_of(OldCollector) + old_unaffiliated_regions * region_size_bytes;
+ }
+ }
+ if (young_reserve > young_unaffiliated_regions * region_size_bytes) {
+ young_reserve = young_unaffiliated_regions * region_size_bytes;
+ }
+ reserve_regions(young_reserve, old_reserve);
_free_sets.establish_alloc_bias(OldCollector);
_free_sets.assert_bounds();
log_status();
@@ -1090,7 +1250,8 @@ void ShenandoahFreeSet::rebuild() {
// the collector set is at least to_reserve, and the memory available for allocations within the old collector set
// is at least to_reserve_old.
void ShenandoahFreeSet::reserve_regions(size_t to_reserve, size_t to_reserve_old) {
- for (size_t idx = _heap->num_regions() - 1; idx > 0; idx--) {
+ for (size_t i = _heap->num_regions(); i > 0; i--) {
+ size_t idx = i - 1;
ShenandoahHeapRegion* r = _heap->get_region(idx);
if (_free_sets.in_free_set(idx, Mutator)) {
assert (!r->is_old(), "mutator_is_free regions should not be affiliated OLD");
@@ -1128,6 +1289,17 @@ void ShenandoahFreeSet::log_status() {
size_t retired_young = 0;
size_t retired_young_humongous = 0;
size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+ size_t retired_young_waste = 0;
+ size_t retired_old_waste = 0;
+ size_t consumed_collector = 0;
+ size_t consumed_old_collector = 0;
+ size_t consumed_mutator = 0;
+ size_t available_old = 0;
+ size_t available_young = 0;
+ size_t available_mutator = 0;
+ size_t available_collector = 0;
+ size_t available_old_collector = 0;
+
char buffer[BUFFER_SIZE];
for (uint i = 0; i < BUFFER_SIZE; i++) {
buffer[i] = '\0';
@@ -1151,12 +1323,21 @@ void ShenandoahFreeSet::log_status() {
}
if (_free_sets.in_free_set(i, Mutator)) {
assert(!r->is_old(), "Old regions should not be in mutator_free set");
- buffer[idx] = (alloc_capacity(r) == region_size_bytes)? 'M': 'm';
+ size_t capacity = alloc_capacity(r);
+ available_mutator += capacity;
+ consumed_mutator += region_size_bytes - capacity;
+ buffer[idx] = (capacity == region_size_bytes)? 'M': 'm';
} else if (_free_sets.in_free_set(i, Collector)) {
assert(!r->is_old(), "Old regions should not be in collector_free set");
- buffer[idx] = (alloc_capacity(r) == region_size_bytes)? 'C': 'c';
+ size_t capacity = alloc_capacity(r);
+ available_collector += capacity;
+ consumed_collector += region_size_bytes - capacity;
+ buffer[idx] = (capacity == region_size_bytes)? 'C': 'c';
} else if (_free_sets.in_free_set(i, OldCollector)) {
- buffer[idx] = (alloc_capacity(r) == region_size_bytes)? 'O': 'o';
+ size_t capacity = alloc_capacity(r);
+ available_old_collector += capacity;
+ consumed_old_collector += region_size_bytes - capacity;
+ buffer[idx] = (capacity == region_size_bytes)? 'O': 'o';
} else if (r->is_humongous()) {
if (r->is_old()) {
buffer[idx] = 'H';
@@ -1168,9 +1349,11 @@ void ShenandoahFreeSet::log_status() {
} else {
if (r->is_old()) {
buffer[idx] = '~';
+ retired_old_waste += alloc_capacity(r);
retired_old += region_size_bytes;
} else {
buffer[idx] = '_';
+ retired_young_waste += alloc_capacity(r);
retired_young += region_size_bytes;
}
}
@@ -1184,12 +1367,6 @@ void ShenandoahFreeSet::log_status() {
log_info(gc, free)(" %6u: %s", (uint) (_heap->num_regions() - remnant), buffer);
size_t total_young = retired_young + retired_young_humongous;
size_t total_old = retired_old + retired_old_humongous;
- log_info(gc, free)("Retired young: " SIZE_FORMAT "%s (including humongous: " SIZE_FORMAT "%s), old: " SIZE_FORMAT
- "%s (including humongous: " SIZE_FORMAT "%s)",
- byte_size_in_proper_unit(total_young), proper_unit_for_byte_size(total_young),
- byte_size_in_proper_unit(retired_young_humongous), proper_unit_for_byte_size(retired_young_humongous),
- byte_size_in_proper_unit(total_old), proper_unit_for_byte_size(total_old),
- byte_size_in_proper_unit(retired_old_humongous), proper_unit_for_byte_size(retired_old_humongous));
}
#endif
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.hpp b/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.hpp
index ef43c236eab..bbf6f1cafda 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.hpp
@@ -72,7 +72,7 @@ class ShenandoahSetsOfFree {
// Place region idx into free set which_set. Requires that idx is currently NotFree.
void make_free(size_t idx, ShenandoahFreeMemoryType which_set, size_t region_capacity);
- // Place region idx into free set new_set. Requires that idx is currently not NotFRee.
+ // Place region idx into free set new_set. Requires that idx is currently not NotFree.
void move_to_set(size_t idx, ShenandoahFreeMemoryType new_set, size_t region_capacity);
// Returns the ShenandoahFreeMemoryType affiliation of region idx, or NotFree if this region is not currently free. This does
@@ -172,11 +172,20 @@ class ShenandoahFreeSet : public CHeapObj<mtGC> {
void flip_to_gc(ShenandoahHeapRegion* r);
void flip_to_old_gc(ShenandoahHeapRegion* r);
+ void adjust_bounds_for_additional_old_collector_free_region(size_t idx);
+
+ void recompute_bounds();
+ void adjust_bounds();
+ bool touches_bounds(size_t num) const;
+
+ // Used of free set represents the amount of is_mutator_free set that has been consumed since most recent rebuild.
+ void increase_used(size_t amount);
void clear_internal();
void try_recycle_trashed(ShenandoahHeapRegion *r);
bool can_allocate_from(ShenandoahHeapRegion *r) const;
+ bool can_allocate_from(size_t idx) const;
bool has_alloc_capacity(size_t idx) const;
bool has_alloc_capacity(ShenandoahHeapRegion *r) const;
bool has_no_alloc_capacity(ShenandoahHeapRegion *r) const;
@@ -188,7 +197,11 @@ class ShenandoahFreeSet : public CHeapObj<mtGC> {
size_t alloc_capacity(size_t idx) const;
void clear();
- void rebuild();
+ void prepare_to_rebuild(size_t &young_cset_regions, size_t &old_cset_regions);
+ void rebuild(size_t young_cset_regions, size_t old_cset_regions);
+ void move_collector_sets_to_mutator(size_t cset_regions);
+
+ void add_old_collector_free_region(ShenandoahHeapRegion* region);
void recycle_trash();
@@ -209,7 +222,7 @@ class ShenandoahFreeSet : public CHeapObj<mtGC> {
void print_on(outputStream* out) const;
- void find_regions_with_alloc_capacity();
+ void find_regions_with_alloc_capacity(size_t &young_cset_regions, size_t &old_cset_regions);
void reserve_regions(size_t young_reserve, size_t old_reserve);
};
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahFullGC.cpp b/src/hotspot/share/gc/shenandoah/shenandoahFullGC.cpp
index 79c8b1c1345..db886618f38 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahFullGC.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahFullGC.cpp
@@ -175,14 +175,21 @@ void ShenandoahFullGC::op_full(GCCause::Cause cause) {
metrics.snap_after();
if (heap->mode()->is_generational()) {
+ heap->mmu_tracker()->record_full(heap->global_generation(), GCId::current());
heap->log_heap_status("At end of Full GC");
// Since we allow temporary violation of these constraints during Full GC, we want to enforce that the assertions are
// made valid by the time Full GC completes.
- assert(heap->old_generation()->used_regions_size() <= heap->old_generation()->adjusted_capacity(),
+ assert(heap->old_generation()->used_regions_size() <= heap->old_generation()->max_capacity(),
"Old generation affiliated regions must be less than capacity");
- assert(heap->young_generation()->used_regions_size() <= heap->young_generation()->adjusted_capacity(),
+ assert(heap->young_generation()->used_regions_size() <= heap->young_generation()->max_capacity(),
"Young generation affiliated regions must be less than capacity");
+
+ assert((heap->young_generation()->used() + heap->young_generation()->get_humongous_waste())
+ <= heap->young_generation()->used_regions_size(), "Young consumed can be no larger than span of affiliated regions");
+ assert((heap->old_generation()->used() + heap->old_generation()->get_humongous_waste())
+ <= heap->old_generation()->used_regions_size(), "Old consumed can be no larger than span of affiliated regions");
+
}
if (metrics.is_good_progress()) {
ShenandoahHeap::heap()->notify_gc_progress();
@@ -199,11 +206,7 @@ void ShenandoahFullGC::do_it(GCCause::Cause gc_cause) {
heap->set_gc_generation(heap->global_generation());
if (heap->mode()->is_generational()) {
- // Defer unadjust_available() invocations until after Full GC finishes its efforts because Full GC makes use
- // of young-gen memory that may have been loaned from old-gen.
-
// No need for old_gen->increase_used() as this was done when plabs were allocated.
- heap->set_alloc_supplement_reserve(0);
heap->set_young_evac_reserve(0);
heap->set_old_evac_reserve(0);
heap->reset_old_evac_expended();
@@ -342,8 +345,6 @@ void ShenandoahFullGC::do_it(GCCause::Cause gc_cause) {
// Resize metaspace
MetaspaceGC::compute_new_size();
- heap->adjust_generation_sizes();
-
// Free worker slices
for (uint i = 0; i < heap->max_workers(); i++) {
delete worker_slices[i];
@@ -357,10 +358,7 @@ void ShenandoahFullGC::do_it(GCCause::Cause gc_cause) {
heap->verifier()->verify_after_fullgc();
}
- // Having reclaimed all dead memory, it is now safe to restore capacities to original values.
- heap->young_generation()->unadjust_available();
- heap->old_generation()->unadjust_available();
-
+ // Humongous regions are promoted on demand and are accounted for by normal Full GC mechanisms.
if (VerifyAfterGC) {
Universe::verify();
}
@@ -1340,7 +1338,6 @@ class ShenandoahPostCompactClosure : public ShenandoahHeapRegionClosure {
account_for_region(r, _young_regions, _young_usage, _young_humongous_waste);
}
}
-
r->set_live_data(live);
r->reset_alloc_metadata();
}
@@ -1501,12 +1498,84 @@ void ShenandoahFullGC::phase5_epilog() {
ShenandoahPostCompactClosure post_compact;
heap->heap_region_iterate(&post_compact);
post_compact.update_generation_usage();
- log_info(gc)("FullGC done: global usage: " SIZE_FORMAT "%s, young usage: " SIZE_FORMAT "%s, old usage: " SIZE_FORMAT "%s",
- byte_size_in_proper_unit(heap->global_generation()->used()), proper_unit_for_byte_size(heap->global_generation()->used()),
- byte_size_in_proper_unit(heap->young_generation()->used()), proper_unit_for_byte_size(heap->young_generation()->used()),
- byte_size_in_proper_unit(heap->old_generation()->used()), proper_unit_for_byte_size(heap->old_generation()->used()));
+ if (heap->mode()->is_generational()) {
+ size_t old_usage = heap->old_generation()->used_regions_size();
+ size_t old_capacity = heap->old_generation()->max_capacity();
+
+ assert(old_usage % ShenandoahHeapRegion::region_size_bytes() == 0, "Old usage must aligh with region size");
+ assert(old_capacity % ShenandoahHeapRegion::region_size_bytes() == 0, "Old capacity must aligh with region size");
+
+ if (old_capacity > old_usage) {
+ size_t excess_old_regions = (old_capacity - old_usage) / ShenandoahHeapRegion::region_size_bytes();
+ heap->generation_sizer()->transfer_to_young(excess_old_regions);
+ } else if (old_capacity < old_usage) {
+ size_t old_regions_deficit = (old_usage - old_capacity) / ShenandoahHeapRegion::region_size_bytes();
+ heap->generation_sizer()->transfer_to_old(old_regions_deficit);
+ }
+
+ log_info(gc)("FullGC done: young usage: " SIZE_FORMAT "%s, old usage: " SIZE_FORMAT "%s",
+ byte_size_in_proper_unit(heap->young_generation()->used()), proper_unit_for_byte_size(heap->young_generation()->used()),
+ byte_size_in_proper_unit(heap->old_generation()->used()), proper_unit_for_byte_size(heap->old_generation()->used()));
+ }
heap->collection_set()->clear();
- heap->free_set()->rebuild();
+ size_t young_cset_regions, old_cset_regions;
+ heap->free_set()->prepare_to_rebuild(young_cset_regions, old_cset_regions);
+
+ // We also do not expand old generation size following Full GC because we have scrambled age populations and
+ // no longer have objects separated by age into distinct regions.
+
+ // TODO: Do we need to fix FullGC so that it maintains aged segregation of objects into distinct regions?
+ // A partial solution would be to remember how many objects are of tenure age following Full GC, but
+ // this is probably suboptimal, because most of these objects will not reside in a region that will be
+ // selected for the next evacuation phase.
+
+ // In case this Full GC resulted from degeneration, clear the tally on anticipated promotion.
+ heap->clear_promotion_potential();
+ heap->clear_promotion_in_place_potential();
+
+ if (heap->mode()->is_generational()) {
+ // Invoke this in case we are able to transfer memory from OLD to YOUNG.
+ heap->adjust_generation_sizes_for_next_cycle(0, 0, 0);
+ }
+ heap->free_set()->rebuild(young_cset_regions, old_cset_regions);
+
+ // We defer generation resizing actions until after cset regions have been recycled. We do this even following an
+ // abbreviated cycle.
+ if (heap->mode()->is_generational()) {
+ bool success;
+ size_t region_xfer;
+ const char* region_destination;
+ ShenandoahYoungGeneration* young_gen = heap->young_generation();
+ ShenandoahGeneration* old_gen = heap->old_generation();
+
+ size_t old_region_surplus = heap->get_old_region_surplus();
+ size_t old_region_deficit = heap->get_old_region_deficit();
+ if (old_region_surplus) {
+ success = heap->generation_sizer()->transfer_to_young(old_region_surplus);
+ region_destination = "young";
+ region_xfer = old_region_surplus;
+ } else if (old_region_deficit) {
+ success = heap->generation_sizer()->transfer_to_old(old_region_deficit);
+ region_destination = "old";
+ region_xfer = old_region_deficit;
+ if (!success) {
+ ((ShenandoahOldHeuristics *) old_gen->heuristics())->trigger_cannot_expand();
+ }
+ } else {
+ region_destination = "none";
+ region_xfer = 0;
+ success = true;
+ }
+ heap->set_old_region_surplus(0);
+ heap->set_old_region_deficit(0);
+ size_t young_available = young_gen->available();
+ size_t old_available = old_gen->available();
+ log_info(gc, ergo)("After cleanup, %s " SIZE_FORMAT " regions to %s to prepare for next gc, old available: "
+ SIZE_FORMAT "%s, young_available: " SIZE_FORMAT "%s",
+ success? "successfully transferred": "failed to transfer", region_xfer, region_destination,
+ byte_size_in_proper_unit(old_available), proper_unit_for_byte_size(old_available),
+ byte_size_in_proper_unit(young_available), proper_unit_for_byte_size(young_available));
+ }
+ heap->clear_cancelled_gc(true /* clear oom handler */);
}
- heap->clear_cancelled_gc(true /* clear oom handler */);
}
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahGeneration.cpp b/src/hotspot/share/gc/shenandoah/shenandoahGeneration.cpp
index 22281b3250e..3606e2cf632 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahGeneration.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahGeneration.cpp
@@ -225,7 +225,6 @@ void ShenandoahGeneration::prepare_gc() {
void ShenandoahGeneration::compute_evacuation_budgets(ShenandoahHeap* heap, bool* preselected_regions,
ShenandoahCollectionSet* collection_set,
size_t &consumed_by_advance_promotion) {
- assert(heap->mode()->is_generational(), "Only generational mode uses evacuation budgets.");
size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
size_t regions_available_to_loan = 0;
size_t minimum_evacuation_reserve = ShenandoahOldCompactionReserve * region_size_bytes;
@@ -249,160 +248,90 @@ void ShenandoahGeneration::compute_evacuation_budgets(ShenandoahHeap* heap, bool
// Do not fill up old-gen memory with promotions. Reserve some amount of memory for compaction purposes.
size_t young_evac_reserve_max = 0;
- if (heap->doing_mixed_evacuations()) {
- // Compute old_evacuation_reserve: how much memory are we reserving to hold the results of
- // evacuating old-gen heap regions? In order to sustain a consistent pace of young-gen collections,
- // the goal is to maintain a consistent value for this parameter (when the candidate set is not
- // empty). This value is the minimum of:
- // 1. old_gen->available()
- // 2. old-gen->capacity() * ShenandoahOldEvacReserve) / 100
- // (e.g. old evacuation should be no larger than 5% of old_gen capacity)
- // 3. ((young_gen->capacity * ShenandoahEvacReserve / 100) * ShenandoahOldEvacRatioPercent) / 100
- // (e.g. old evacuation should be no larger than 12% of young-gen evacuation)
- old_evacuation_reserve = old_generation->available();
- // This assertion has been disabled because we expect this code to be replaced by 05/2023
- // assert(old_evacuation_reserve > minimum_evacuation_reserve, "Old-gen available has not been preserved!");
- size_t old_evac_reserve_max = old_generation->max_capacity() * ShenandoahOldEvacReserve / 100;
- if (old_evac_reserve_max < old_evacuation_reserve) {
- old_evacuation_reserve = old_evac_reserve_max;
- }
- young_evac_reserve_max =
- (((young_generation->max_capacity() * ShenandoahEvacReserve) / 100) * ShenandoahOldEvacRatioPercent) / 100;
- if (young_evac_reserve_max < old_evacuation_reserve) {
- old_evacuation_reserve = young_evac_reserve_max;
- }
- }
- if (minimum_evacuation_reserve > old_generation->available()) {
- // Due to round-off errors during enforcement of minimum_evacuation_reserve during previous GC passes,
- // there can be slight discrepancies here.
- minimum_evacuation_reserve = old_generation->available();
- }
+ // First priority is to reclaim the easy garbage out of young-gen.
- heap->set_old_evac_reserve(old_evacuation_reserve);
- heap->reset_old_evac_expended();
-
- // Compute the young evacuation reserve: This is how much memory is available for evacuating young-gen objects.
- // We ignore the possible effect of promotions, which reduce demand for young-gen evacuation memory.
- //
- // TODO: We could give special treatment to the regions that have reached promotion age, because we know their
- // live data is entirely eligible for promotion. This knowledge can feed both into calculations of young-gen
- // evacuation reserve and promotion reserve.
- //
- // young_evacuation_reserve for young generation: how much memory are we reserving to hold the results
- // of evacuating young collection set regions? This is typically smaller than the total amount
- // of available memory, and is also smaller than the total amount of marked live memory within
- // young-gen. This value is the smaller of
- //
- // 1. (young_gen->capacity() * ShenandoahEvacReserve) / 100
- // 2. (young_gen->available() + old_gen_memory_available_to_be_loaned
- //
- // ShenandoahEvacReserve represents the configured target size of the evacuation region. We can only honor
- // this target if there is memory available to hold the evacuations. Memory is available if it is already
- // free within young gen, or if it can be borrowed from old gen. Since we have not yet chosen the collection
- // sets, we do not yet know the exact accounting of how many regions will be freed by this collection pass.
- // What we do know is that there will be at least one evacuated young-gen region for each old-gen region that
- // is loaned to the evacuation effort (because regions to be collected consume more memory than the compacted
- // regions that will replace them). In summary, if there are old-gen regions that are available to hold the
- // results of young-gen evacuations, it is safe to loan them for this purpose. At this point, we have not yet
- // established a promoted_reserve. We'll do that after we choose the collection set and analyze its impact
- // on available memory.
- //
- // We do not know the evacuation_supplement until after we have computed the collection set. It is not always
- // the case that young-regions inserted into the collection set will result in net decrease of in-use regions
- // because ShenandoahEvacWaste times multiplied by memory within the region may be larger than the region size.
- // The problem is especially relevant to regions that have been inserted into the collection set because they have
- // reached tenure age. These regions tend to have much higher utilization (e.g. 95%). These regions also offer
- // a unique opportunity because we know that every live object contained within the region is elgible to be
- // promoted. Thus, the following implementation treats these regions specially:
- //
- // 1. Before beginning collection set selection, we tally the total amount of live memory held within regions
- // that are known to have reached tenure age. If this memory times ShenandoahEvacWaste is available within
- // old-gen memory, establish an advance promotion reserve to hold all or some percentage of these objects.
- // This advance promotion reserve is excluded from memory available for holding old-gen evacuations and cannot
- // be "loaned" to young gen.
- //
- // 2. Tenure-aged regions are included in the collection set iff their evacuation size * ShenandoahEvacWaste fits
- // within the advance promotion reserve. It is counter productive to evacuate these regions if they cannot be
- // evacuated directly into old-gen memory. So if there is not sufficient memory to hold copies of their
- // live data right now, we'll just let these regions remain in young for now, to be evacuated by a subsequent
- // evacuation pass.
- //
- // 3. Next, we calculate a young-gen evacuation budget, which is the smaller of the two quantities mentioned
- // above. old_gen_memory_available_to_be_loaned is calculated as:
- // old_gen->available - (advance-promotion-reserve + old-gen_evacuation_reserve)
- //
- // 4. When choosing the collection set, special care is taken to assure that the amount of loaned memory required to
- // hold the results of evacuation is smaller than the total memory occupied by the regions added to the collection
- // set. We need to take these precautions because we do not know how much memory will be reclaimed by evacuation
- // until after the collection set has been constructed. The algorithm is as follows:
- //
- // a. We feed into the algorithm (i) young available at the start of evacuation and (ii) the amount of memory
- // loaned from old-gen that is available to hold the results of evacuation.
- // b. As candidate regions are added into the young-gen collection set, we maintain accumulations of the amount
- // of memory spanned by the collection set regions and the amount of memory that must be reserved to hold
- // evacuation results (by multiplying live-data size by ShenandoahEvacWaste). We process candidate regions
- // in order of decreasing amounts of garbage. We skip over (and do not include into the collection set) any
- // regions that do not satisfy all of the following conditions:
- //
- // i. The amount of live data within the region as scaled by ShenandoahEvacWaste must fit within the
- // relevant evacuation reserve (live data of old-gen regions must fit within the old-evac-reserve, live
- // data of young-gen tenure-aged regions must fit within the advance promotion reserve, live data within
- // other young-gen regions must fit within the youn-gen evacuation reserve).
- // ii. The accumulation of memory consumed by evacuation must not exceed the accumulation of memory reclaimed
- // through evacuation by more than young-gen available.
- // iii. Other conditions may be enforced as appropriate for specific heuristics.
- //
- // Note that regions are considered for inclusion in the selection set in order of decreasing amounts of garbage.
- // It is possible that a region with a larger amount of garbage will be rejected because it also has a larger
- // amount of live data and some region that follows this region in candidate order is included in the collection
- // set (because it has less live data and thus can fit within the evacuation limits even though it has less
- // garbage).
-
- size_t young_evacuation_reserve = (young_generation->max_capacity() * ShenandoahEvacReserve) / 100;
- // old evacuation can pack into existing partially used regions. young evacuation and loans for young allocations
- // need to target regions that do not already hold any old-gen objects. Round down.
- regions_available_to_loan = old_generation->free_unaffiliated_regions();
-
- size_t required_evacuation_reserve;
- // Memory evacuated from old-gen on this pass will be available to hold old-gen evacuations in next pass.
- if (old_evacuation_reserve > minimum_evacuation_reserve) {
- required_evacuation_reserve = 0;
+ // maximum_young_evacuation_reserve is upper bound on memory to be evacuated out of young
+ size_t maximum_young_evacuation_reserve = (young_generation->max_capacity() * ShenandoahEvacReserve) / 100;
+ size_t young_evacuation_reserve = maximum_young_evacuation_reserve;
+ size_t excess_young;
+ if (young_generation->available() > young_evacuation_reserve) {
+ excess_young = young_generation->available() - young_evacuation_reserve;
} else {
- required_evacuation_reserve = minimum_evacuation_reserve - old_evacuation_reserve;
+ young_evacuation_reserve = young_generation->available();
+ excess_young = 0;
}
-
- consumed_by_advance_promotion = _heuristics->select_aged_regions(
- old_generation->available() - old_evacuation_reserve - required_evacuation_reserve, num_regions, preselected_regions);
- size_t net_available_old_regions =
- (old_generation->available() - old_evacuation_reserve - consumed_by_advance_promotion) / region_size_bytes;
-
- if (regions_available_to_loan > net_available_old_regions) {
- regions_available_to_loan = net_available_old_regions;
+ size_t unaffiliated_young = young_generation->free_unaffiliated_regions() * region_size_bytes;
+ if (excess_young > unaffiliated_young) {
+ excess_young = unaffiliated_young;
+ } else {
+ // round down to multiple of region size
+ excess_young /= region_size_bytes;
+ excess_young *= region_size_bytes;
+ }
+ // excess_young is available to be transferred to OLD. Assume that OLD will not request any more than had
+ // already been set aside for its promotion and evacuation needs at the end of previous GC. No need to
+ // hold back memory for allocation runway.
+
+ ShenandoahOldHeuristics* old_heuristics = heap->old_heuristics();
+
+ // maximum_old_evacuation_reserve is an upper bound on memory evacuated from old and evacuated to old (promoted).
+ size_t maximum_old_evacuation_reserve =
+ maximum_young_evacuation_reserve * ShenandoahOldEvacRatioPercent / (100 - ShenandoahOldEvacRatioPercent);
+ // Here's the algebra:
+ // TotalEvacuation = OldEvacuation + YoungEvacuation
+ // OldEvacuation = TotalEvacuation * (ShenandoahOldEvacRatioPercent/100)
+ // OldEvacuation = YoungEvacuation * (ShenandoahOldEvacRatioPercent/100)/(1 - ShenandoahOldEvacRatioPercent/100)
+ // OldEvacuation = YoungEvacuation * ShenandoahOldEvacRatioPercent/(100 - ShenandoahOldEvacRatioPercent)
+
+ if (maximum_old_evacuation_reserve > old_generation->available()) {
+ maximum_old_evacuation_reserve = old_generation->available();
+ }
+
+ // Second priority is to reclaim garbage out of old-gen if there are old-gen collection candidates. Third priority
+ // is to promote as much as we have room to promote. However, if old-gen memory is in short supply, this means young
+ // GC is operating under "duress" and was unable to transfer the memory that we would normally expect. In this case,
+ // old-gen will refrain from compacting itself in order to allow a quicker young-gen cycle (by avoiding the update-refs
+ // through ALL of old-gen). If there is some memory available in old-gen, we will use this for promotions as promotions
+ // do not add to the update-refs burden of GC.
+
+ size_t old_promo_reserve;
+ if (old_heuristics->unprocessed_old_collection_candidates() > 0) {
+ // We reserved all old-gen memory at end of previous GC to hold anticipated evacuations to old-gen. If this is
+ // mixed evacuation, reserve all of this memory for compaction of old-gen and do not promote. Prioritize compaction
+ // over promotion in order to defragment OLD so that it will be better prepared to efficiently receive promoted memory.
+ old_evacuation_reserve = maximum_old_evacuation_reserve;
+ old_promo_reserve = 0;
+ } else {
+ // Make all old-evacuation memory for promotion, but if we can't use it all for promotion, we'll allow some evacuation.
+ old_evacuation_reserve = 0;
+ old_promo_reserve = maximum_old_evacuation_reserve;
}
- // Otherwise, regions_available_to_loan is less than net_available_old_regions because available memory is
- // scattered between multiple partially used regions.
+ // We see too many old-evacuation failures if we force ourselves to evacuate into regions that are not initially empty.
+ // So we limit the old-evacuation reserve to unfragmented memory. Even so, old-evacuation is free to fill in nooks and
+ // crannies within existing partially used regions and it generally tries to do so.
+ size_t old_free_regions = old_generation->free_unaffiliated_regions();
+ size_t old_free_unfragmented = old_free_regions * region_size_bytes;
+ if (old_evacuation_reserve > old_free_unfragmented) {
+ size_t delta = old_evacuation_reserve - old_free_unfragmented;
+ old_evacuation_reserve -= delta;
- if (young_evacuation_reserve > young_generation->available()) {
- size_t short_fall = young_evacuation_reserve - young_generation->available();
- if (regions_available_to_loan * region_size_bytes >= short_fall) {
- old_regions_loaned_for_young_evac = (short_fall + region_size_bytes - 1) / region_size_bytes;
- regions_available_to_loan -= old_regions_loaned_for_young_evac;
- } else {
- old_regions_loaned_for_young_evac = regions_available_to_loan;
- regions_available_to_loan = 0;
- young_evacuation_reserve = young_generation->available() + old_regions_loaned_for_young_evac * region_size_bytes;
- // In this case, there's no memory available for new allocations while evacuating and updating, unless we
- // find more old-gen memory to borrow below.
- }
+ // Let promo consume fragments of old-gen memory.
+ old_promo_reserve += delta;
}
- // In generational mode, we may end up choosing a young collection set that contains so many promotable objects
- // that there is not sufficient space in old generation to hold the promoted objects. That is ok because we have
- // assured there is sufficient space in young generation to hold the rejected promotion candidates. These rejected
- // promotion candidates will presumably be promoted in a future evacuation cycle.
- heap->set_young_evac_reserve(young_evacuation_reserve);
collection_set->establish_preselected(preselected_regions);
+ consumed_by_advance_promotion = _heuristics->select_aged_regions(old_promo_reserve, num_regions, preselected_regions);
+ assert(consumed_by_advance_promotion <= maximum_old_evacuation_reserve, "Cannot promote more than available old-gen memory");
+ if (consumed_by_advance_promotion < old_promo_reserve) {
+ // If we're in a global collection, this memory can be used for old evacuations
+ old_evacuation_reserve += old_promo_reserve - consumed_by_advance_promotion;
+ }
+ heap->set_young_evac_reserve(young_evacuation_reserve);
+ heap->set_old_evac_reserve(old_evacuation_reserve);
+ heap->set_promoted_reserve(consumed_by_advance_promotion);
+
+ // There is no need to expand OLD because all memory used here was set aside at end of previous GC
}
// Having chosen the collection set, adjust the budgets for generational mode based on its composition. Note
@@ -425,356 +354,92 @@ void ShenandoahGeneration::adjust_evacuation_budgets(ShenandoahHeap* heap, Shena
// available that results from a decrease in memory consumed by old evacuation is not necessarily available to be loaned
// to young-gen.
- assert(heap->mode()->is_generational(), "Only generational mode uses evacuation budgets.");
- size_t old_regions_loaned_for_young_evac, regions_available_to_loan;
size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
ShenandoahOldGeneration* old_generation = heap->old_generation();
ShenandoahYoungGeneration* young_generation = heap->young_generation();
- size_t old_evacuated = collection_set->get_old_bytes_reserved_for_evacuation();
- size_t old_evacuated_committed = (size_t) (ShenandoahEvacWaste * old_evacuated);
- size_t old_evacuation_reserve = heap->get_old_evac_reserve();
- // Immediate garbage found during choose_collection_set() is all young
- size_t immediate_garbage = collection_set->get_immediate_trash();
- size_t old_available = old_generation->available();
- size_t young_available = young_generation->available() + immediate_garbage;
- size_t loaned_regions = 0;
- size_t available_loan_remnant = 0; // loaned memory that is not yet dedicated to any particular budget
-
- // We expect this code to be replaced by 05/01/23.
- //
- // assert(((consumed_by_advance_promotion * 33) / 32) >= collection_set->get_young_bytes_to_be_promoted() * ShenandoahEvacWaste,
- // "Advance promotion (" SIZE_FORMAT ") should be at least young_bytes_to_be_promoted (" SIZE_FORMAT
- // ")* ShenandoahEvacWaste, totalling: " SIZE_FORMAT ", within round-off errors of up to 3.125%%",
- // consumed_by_advance_promotion, collection_set->get_young_bytes_to_be_promoted(),
- // (size_t) (collection_set->get_young_bytes_to_be_promoted() * ShenandoahEvacWaste));
-
- // assert(consumed_by_advance_promotion <= (collection_set->get_young_bytes_to_be_promoted() * ShenandoahEvacWaste * 33) / 32,
- // "Round-off errors should be less than 3.125%%, consumed by advance: " SIZE_FORMAT ", promoted: " SIZE_FORMAT,
- // consumed_by_advance_promotion, (size_t) (collection_set->get_young_bytes_to_be_promoted() * ShenandoahEvacWaste));
+ // Preselected regions have been inserted into the collection set, so we no longer need the preselected array.
collection_set->abandon_preselected();
+ size_t old_evacuated = collection_set->get_old_bytes_reserved_for_evacuation();
+ size_t old_evacuated_committed = (size_t) (ShenandoahOldEvacWaste * old_evacuated);
+ size_t old_evacuation_reserve = heap->get_old_evac_reserve();
+
if (old_evacuated_committed > old_evacuation_reserve) {
- // This should only happen due to round-off errors when enforcing ShenandoahEvacWaste
+ // This should only happen due to round-off errors when enforcing ShenandoahOldEvacWaste
assert(old_evacuated_committed <= (33 * old_evacuation_reserve) / 32,
"Round-off errors should be less than 3.125%%, committed: " SIZE_FORMAT ", reserved: " SIZE_FORMAT,
old_evacuated_committed, old_evacuation_reserve);
old_evacuated_committed = old_evacuation_reserve;
+ // Leave old_evac_reserve as previously configured
} else if (old_evacuated_committed < old_evacuation_reserve) {
- // This may happen if the old-gen collection consumes less than full budget.
+ // This happens if the old-gen collection consumes less than full budget.
old_evacuation_reserve = old_evacuated_committed;
heap->set_old_evac_reserve(old_evacuation_reserve);
}
- // Recompute old_regions_loaned_for_young_evac because young-gen collection set may not need all the memory
- // originally reserved.
- size_t young_promoted = collection_set->get_young_bytes_to_be_promoted();
- size_t young_promoted_reserve_used = (size_t) (ShenandoahEvacWaste * young_promoted);
+ size_t young_advance_promoted = collection_set->get_young_bytes_to_be_promoted();
+ size_t young_advance_promoted_reserve_used = (size_t) (ShenandoahPromoEvacWaste * young_advance_promoted);
size_t young_evacuated = collection_set->get_young_bytes_reserved_for_evacuation();
size_t young_evacuated_reserve_used = (size_t) (ShenandoahEvacWaste * young_evacuated);
- // We'll invoke heap->set_young_evac_reserve() further below, after we make additional adjustments to its value
-
- // Adjust old_regions_loaned_for_young_evac to feed into calculations of promoted_reserve
- if (young_evacuated_reserve_used > young_available) {
- size_t short_fall = young_evacuated_reserve_used - young_available;
-
- // region_size_bytes is a power of 2. loan an integral number of regions.
- size_t revised_loan_for_young_evacuation = (short_fall + region_size_bytes - 1) / region_size_bytes;
-
- // available_loan_remnant represents memory loaned from old-gen but not required for young evacuation.
- // This is the excess loaned memory that results from rounding the required loan up to an integral number
- // of heap regions. This will be dedicated to alloc_supplement below.
- available_loan_remnant = (revised_loan_for_young_evacuation * region_size_bytes) - short_fall;
-
- // We previously loaned more than was required by young-gen evacuation. So claw some of this memory back.
- old_regions_loaned_for_young_evac = revised_loan_for_young_evacuation;
- loaned_regions = old_regions_loaned_for_young_evac;
- } else {
- // Undo the previous loan, if any.
- old_regions_loaned_for_young_evac = 0;
- loaned_regions = 0;
- }
-
- size_t old_bytes_loaned_for_young_evac = old_regions_loaned_for_young_evac * region_size_bytes - available_loan_remnant;
-
- // Recompute regions_available_to_loan based on possible changes to old_regions_loaned_for_young_evac and
- // old_evacuation_reserve.
-
- // Any decrease in old_regions_loaned_for_young_evac are immediately available to be loaned
- // However, a change to old_evacuation_reserve() is not necessarily available to loan, because this memory may
- // reside within many fragments scattered throughout old-gen.
-
- regions_available_to_loan = old_generation->free_unaffiliated_regions();
- size_t working_old_available = old_generation->available();
-
- assert(regions_available_to_loan * region_size_bytes <= working_old_available,
- "Regions available to loan must be less than available memory");
-
- // fragmented_old_total is the amount of memory in old-gen beyond regions_available_to_loan that is otherwise not
- // yet dedicated to a particular budget. This memory can be used for promotion_reserve.
- size_t fragmented_old_total = working_old_available - regions_available_to_loan * region_size_bytes;
-
- // fragmented_old_usage is the memory that is dedicated to holding evacuated old-gen objects, which does not need
- // to be an integral number of regions.
- size_t fragmented_old_usage = old_evacuated_committed + consumed_by_advance_promotion;
-
- if (fragmented_old_total >= fragmented_old_usage) {
- // Seems this will be rare. In this case, all of the memory required for old-gen evacuations and promotions can be
- // taken from the existing fragments within old-gen. Reduce this fragmented total by this amount.
- fragmented_old_total -= fragmented_old_usage;
- // And reduce regions_available_to_loan by the regions dedicated to young_evac.
- regions_available_to_loan -= old_regions_loaned_for_young_evac;
- } else {
- // In this case, we need to dedicate some of the regions_available_to_loan to hold the results of old-gen evacuations
- // and promotions.
-
- size_t unaffiliated_memory_required_for_old = fragmented_old_usage - fragmented_old_total;
- size_t unaffiliated_regions_used_by_old = (unaffiliated_memory_required_for_old + region_size_bytes - 1) / region_size_bytes;
- regions_available_to_loan -= (unaffiliated_regions_used_by_old + old_regions_loaned_for_young_evac);
-
- size_t memory_for_promotions_and_old_evac = fragmented_old_total + unaffiliated_regions_used_by_old;
- size_t memory_required_for_promotions_and_old_evac = fragmented_old_usage;
- size_t excess_fragmented = memory_for_promotions_and_old_evac - memory_required_for_promotions_and_old_evac;
- fragmented_old_total = excess_fragmented;
- }
-
- // Subtract from working_old_available old_evacuated_committed and consumed_by_advance_promotion
- working_old_available -= fragmented_old_usage;
- // And also subtract out the regions loaned for young evacuation
- working_old_available -= old_regions_loaned_for_young_evac * region_size_bytes;
-
- // Assure that old_evacuated_committed + old_bytes_loaned_for_young_evac >= the minimum evacuation reserve
- // in order to prevent promotion reserve from violating minimum evacuation reserve.
- size_t old_regions_reserved_for_alloc_supplement = 0;
- size_t old_bytes_reserved_for_alloc_supplement = 0;
- size_t reserved_bytes_for_future_old_evac = 0;
-
- old_bytes_reserved_for_alloc_supplement = available_loan_remnant;
- available_loan_remnant = 0;
-
- // Memory that has been loaned for young evacuations and old-gen regions in the current mixed-evacuation collection
- // set will be available to hold future old-gen evacuations. If this memory is less than the desired amount of memory
- // set aside for old-gen compaction reserve, try to set aside additional memory so that it will be available during
- // the next mixed evacuation cycle. Note that memory loaned to young-gen for allocation supplement is excluded from
- // the old-gen promotion reserve.
- size_t future_evac_reserve_regions = old_regions_loaned_for_young_evac + collection_set->get_old_region_count();
- size_t collected_regions = collection_set->get_young_region_count();
-
- if (future_evac_reserve_regions < ShenandoahOldCompactionReserve) {
- // Require that we loan more memory for holding young evacuations to assure that we have adequate reserves to receive
- // old-gen evacuations during subsequent collections. Loaning this memory for an allocation supplement does not
- // satisfy our needs because newly allocated objects are not necessarily counter-balanced by reclaimed collection
- // set regions.
-
- // Put this memory into reserve by identifying it as old_regions_loaned_for_young_evac
- size_t additional_regions_to_loan = ShenandoahOldCompactionReserve - future_evac_reserve_regions;
-
- // We can loan additional regions to be repaid from the anticipated recycling of young collection set regions
- // provided that these regions are currently available within old-gen memory.
- size_t collected_regions_to_loan;
- if (collected_regions >= additional_regions_to_loan) {
- collected_regions_to_loan = additional_regions_to_loan;
- additional_regions_to_loan = 0;
- } else if (collected_regions > 0) {
- collected_regions_to_loan = collected_regions;
- additional_regions_to_loan -= collected_regions_to_loan;
- } else {
- collected_regions_to_loan = 0;
- }
-
- if (collected_regions_to_loan > 0) {
- // We're evacuating at least this many regions, it's ok to use these regions for allocation supplement since
- // we'll be able to repay the loan at end of this GC pass, assuming the regions are available.
- if (collected_regions_to_loan > regions_available_to_loan) {
- collected_regions_to_loan = regions_available_to_loan;
- }
- old_bytes_reserved_for_alloc_supplement += collected_regions_to_loan * region_size_bytes;
- regions_available_to_loan -= collected_regions_to_loan;
- loaned_regions += collected_regions_to_loan;
- working_old_available -= collected_regions_to_loan * region_size_bytes;
- }
-
- // If there's still memory that we want to exclude from the current promotion reserve, but we are unable to loan
- // this memory because fully empty old-gen regions are not available, decrement the working_old_available to make
- // sure that this memory is not used to hold the results of old-gen evacuation.
- if (additional_regions_to_loan > regions_available_to_loan) {
- size_t unloaned_regions = additional_regions_to_loan - regions_available_to_loan;
- size_t unloaned_bytes = unloaned_regions * region_size_bytes;
-
- if (working_old_available < unloaned_bytes) {
- // We're in dire straits. We won't be able to reserve all the memory that we want to make available for the
- // next old-gen evacuation. We'll reserve as much of it as possible. Setting working_old_available to zero
- // means there will be no promotion except for the advance promotion. Note that if some advance promotion fails,
- // the object will be evacuated to young-gen so we should still end up reclaiming the entire advance promotion
- // collection set.
- reserved_bytes_for_future_old_evac = working_old_available;
- working_old_available = 0;
- } else {
- reserved_bytes_for_future_old_evac = unloaned_bytes;
- working_old_available -= unloaned_bytes;
- }
- size_t regions_reserved_for_future_old_evac =
- (reserved_bytes_for_future_old_evac + region_size_bytes - 1) / region_size_bytes;
-
- if (regions_reserved_for_future_old_evac < regions_available_to_loan) {
- regions_available_to_loan -= regions_reserved_for_future_old_evac;
- } else {
- regions_available_to_loan = 0;
- }
+ assert(young_evacuated_reserve_used <= young_generation->available(), "Cannot evacuate more than is available in young");
+ heap->set_young_evac_reserve(young_evacuated_reserve_used);
- // Since we're in dire straits, zero out fragmented_old_total so this won't be used for promotion;
- if (working_old_available > fragmented_old_total) {
- working_old_available -= fragmented_old_total;
+ size_t old_available = old_generation->available();
+ // Now that we've established the collection set, we know how much memory is really required by old-gen for evacuation
+ // and promotion reserves. Try shrinking OLD now in case that gives us a bit more runway for mutator allocations during
+ // evac and update phases.
+ size_t old_consumed = old_evacuated_committed + young_advance_promoted_reserve_used;
+ assert(old_available >= old_consumed, "Cannot consume more than is available");
+ size_t excess_old = old_available - old_consumed;
+ size_t unaffiliated_old_regions = old_generation->free_unaffiliated_regions();
+ size_t unaffiliated_old = unaffiliated_old_regions * region_size_bytes;
+ assert(old_available >= unaffiliated_old, "Unaffiliated old is a subset of old available");
+
+ // Make sure old_evac_committed is unaffiliated
+ if (old_evacuated_committed > 0) {
+ if (unaffiliated_old > old_evacuated_committed) {
+ size_t giveaway = unaffiliated_old - old_evacuated_committed;
+ size_t giveaway_regions = giveaway / region_size_bytes; // round down
+ if (giveaway_regions > 0) {
+ excess_old = MIN2(excess_old, giveaway_regions * region_size_bytes);
} else {
- working_old_available = 0;
+ excess_old = 0;
}
- fragmented_old_total = 0;
+ } else {
+ excess_old = 0;
}
}
- // Establish young_evac_reserve so that this young-gen memory is not used for new allocations, allowing the memory
- // to be returned to old-gen as soon as the current collection set regions are reclaimed.
- heap->set_young_evac_reserve(young_evacuated_reserve_used);
-
- // Limit promoted_reserve so that we can set aside memory to be loaned from old-gen to young-gen. This
- // value is not "critical". If we underestimate, certain promotions will simply be deferred. If we put
- // "all the rest" of old-gen memory into the promotion reserve, we'll have nothing left to loan to young-gen
- // during the evac and update phases of GC. So we "limit" the sizes of the promotion budget to be the smaller of:
- //
- // 1. old_available
- // (old_available is old_gen->available() -
- // (old_evacuated_committed + consumed_by_advance_promotion + loaned_for_young_evac + reserved_for_alloc_supplement))
- // 2. young bytes reserved for evacuation (we can't promote more than young is evacuating)
- size_t promotion_reserve = working_old_available;
-
- // We experimented with constraining promoted_reserve to be no larger than 4 times the size of previously_promoted,
- // but this constraint was too limiting, resulting in failure of legitimate promotions. This was tried before we
- // had special handling in place for advance promotion. We should retry now that advance promotion is handled
- // specially.
-
- // We had also experimented with constraining promoted_reserve to be no more than young_evacuation_committed
- // divided by promotion_divisor, where:
- // size_t promotion_divisor = (0x02 << InitialTenuringThreshold) - 1;
- // This also was found to be too limiting, resulting in failure of legitimate promotions.
- //
- // Both experiments were conducted in the presence of other bugs which could have been the root cause for
- // the failures identified above as being "too limiting". TODO: conduct new experiments with the more limiting
- // values of young_evacuation_reserved_used.
-
- // young_evacuation_reserve_used already excludes bytes known to be promoted, which equals consumed_by_advance_promotion
- if (young_evacuated_reserve_used < promotion_reserve) {
- // Shrink promotion_reserve if it is larger than the memory to be consumed by evacuating all young objects in
- // collection set, including anticipated waste. There's no benefit in using a larger promotion_reserve.
- // young_evacuation_reserve_used does not include live memory within tenure-aged regions.
- promotion_reserve = young_evacuated_reserve_used;
- }
- assert(working_old_available >= promotion_reserve, "Cannot reserve for promotion more than is available");
- working_old_available -= promotion_reserve;
- // Having reserved this memory for promotion, the regions are no longer available to be loaned.
- size_t regions_consumed_by_promotion_reserve = (promotion_reserve + region_size_bytes - 1) / region_size_bytes;
- if (regions_consumed_by_promotion_reserve > regions_available_to_loan) {
- // This can happen if the promotion reserve makes use of memory that is fragmented between many partially available
- // old-gen regions.
- regions_available_to_loan = 0;
- } else {
- regions_available_to_loan -= regions_consumed_by_promotion_reserve;
+ // If we find that OLD has excess regions, give them back to YOUNG now to reduce likelihood we run out of allocation
+ // runway during evacuation and update-refs.
+ size_t regions_to_xfer = 0;
+ if (excess_old > unaffiliated_old) {
+ // we can give back unaffiliated_old (all of unaffiliated is excess)
+ if (unaffiliated_old_regions > 0) {
+ regions_to_xfer = unaffiliated_old_regions;
+ }
+ } else if (unaffiliated_old_regions > 0) {
+ // excess_old < unaffiliated old: we can give back MIN(excess_old/region_size_bytes, unaffiliated_old_regions)
+ size_t excess_regions = excess_old / region_size_bytes;
+ size_t regions_to_xfer = MIN2(excess_regions, unaffiliated_old_regions);
}
- log_debug(gc)("old_gen->available(): " SIZE_FORMAT " divided between promotion reserve: " SIZE_FORMAT
- ", old evacuation reserve: " SIZE_FORMAT ", advance promotion reserve supplement: " SIZE_FORMAT
- ", old loaned for young evacuation: " SIZE_FORMAT ", old reserved for alloc supplement: " SIZE_FORMAT,
- old_generation->available(), promotion_reserve, old_evacuated_committed, consumed_by_advance_promotion,
- old_regions_loaned_for_young_evac * region_size_bytes, old_bytes_reserved_for_alloc_supplement);
-
- promotion_reserve += consumed_by_advance_promotion;
- heap->set_promoted_reserve(promotion_reserve);
-
- heap->reset_promoted_expended();
- if (collection_set->get_old_bytes_reserved_for_evacuation() == 0) {
- // Setting old evacuation reserve to zero denotes that there is no old-gen evacuation in this pass.
- heap->set_old_evac_reserve(0);
+ if (regions_to_xfer > 0) {
+ bool result = heap->generation_sizer()->transfer_to_young(regions_to_xfer);
+ assert(excess_old > regions_to_xfer * region_size_bytes, "Cannot xfer more than excess old");
+ excess_old -= regions_to_xfer * region_size_bytes;
+ log_info(gc, ergo)("%s transferred " SIZE_FORMAT " excess regions to young before start of evacuation",
+ result? "Successfully": "Unsuccessfully", regions_to_xfer);
}
- size_t old_gen_usage_base = old_generation->used() - collection_set->get_old_garbage();
- heap->capture_old_usage(old_gen_usage_base);
-
- // Compute additional evacuation supplement, which is extra memory borrowed from old-gen that can be allocated
- // by mutators while GC is working on evacuation and update-refs. This memory can be temporarily borrowed
- // from old-gen allotment, then repaid at the end of update-refs from the recycled collection set. After
- // we have computed the collection set based on the parameters established above, we can make additional
- // loans based on our knowledge of the collection set to determine how much allocation we can allow
- // during the evacuation and update-refs phases of execution. The total available supplement is the result
- // of adding old_bytes_reserved_for_alloc_supplement to the smaller of:
- //
- // 1. regions_available_to_loan * region_size_bytes
- // 2. The replenishment budget (number of regions in collection set - the number of regions already
- // under lien for the young_evacuation_reserve)
- //
-
- // Regardless of how many regions may be available to be loaned, we can loan no more regions than
- // the total number of young regions to be evacuated. Call this the regions_for_runway.
-
- if (regions_available_to_loan > 0 && (collected_regions > loaned_regions)) {
- assert(regions_available_to_loan * region_size_bytes <= working_old_available,
- "regions_available_to_loan should not exceed working_old_available");
-
- size_t additional_regions_to_loan = collected_regions - loaned_regions;
- if (additional_regions_to_loan > regions_available_to_loan) {
- additional_regions_to_loan = regions_available_to_loan;
- }
- loaned_regions += additional_regions_to_loan;
- old_bytes_reserved_for_alloc_supplement += additional_regions_to_loan * region_size_bytes;
- working_old_available -= additional_regions_to_loan * region_size_bytes;
- }
- size_t allocation_supplement = old_bytes_reserved_for_alloc_supplement + old_bytes_loaned_for_young_evac;
- assert(allocation_supplement % ShenandoahHeapRegion::region_size_bytes() == 0,
- "allocation_supplement must be multiple of region size");
-
- heap->set_alloc_supplement_reserve(allocation_supplement);
-
- // TODO: young_available, which feeds into alloc_budget_evac_and_update is lacking memory available within
- // existing young-gen regions that were not selected for the collection set. Add this in and adjust the
- // log message (where it says "empty-region allocation budget").
-
-
- log_debug(gc)("Memory reserved for young evacuation: " SIZE_FORMAT "%s for evacuating " SIZE_FORMAT
- "%s out of young available: " SIZE_FORMAT "%s",
- byte_size_in_proper_unit(young_evacuated_reserve_used),
- proper_unit_for_byte_size(young_evacuated_reserve_used),
- byte_size_in_proper_unit(young_evacuated), proper_unit_for_byte_size(young_evacuated),
- byte_size_in_proper_unit(young_available), proper_unit_for_byte_size(young_available));
-
- log_debug(gc)("Memory reserved for old evacuation: " SIZE_FORMAT "%s for evacuating " SIZE_FORMAT
- "%s out of old available: " SIZE_FORMAT "%s",
- byte_size_in_proper_unit(old_evacuated), proper_unit_for_byte_size(old_evacuated),
- byte_size_in_proper_unit(old_evacuated), proper_unit_for_byte_size(old_evacuated),
- byte_size_in_proper_unit(old_available), proper_unit_for_byte_size(old_available));
-
- size_t regular_promotion = promotion_reserve - consumed_by_advance_promotion;
- size_t excess =
- old_available - (old_evacuation_reserve + promotion_reserve + old_bytes_loaned_for_young_evac + allocation_supplement);
- log_info(gc, ergo)("Old available: " SIZE_FORMAT "%s is partitioned into old evacuation budget: " SIZE_FORMAT
- "%s, aged region promotion budget: " SIZE_FORMAT
- "%s, regular region promotion budget: " SIZE_FORMAT
- "%s, loaned for young evacuation: " SIZE_FORMAT
- "%s, loaned for young allocations: " SIZE_FORMAT
- "%s, excess: " SIZE_FORMAT "%s",
- byte_size_in_proper_unit(old_available),
- proper_unit_for_byte_size(old_available),
- byte_size_in_proper_unit(old_evacuation_reserve),
- proper_unit_for_byte_size(old_evacuation_reserve),
- byte_size_in_proper_unit(consumed_by_advance_promotion),
- proper_unit_for_byte_size(consumed_by_advance_promotion),
- byte_size_in_proper_unit(regular_promotion),
- proper_unit_for_byte_size(regular_promotion),
- byte_size_in_proper_unit(old_bytes_loaned_for_young_evac),
- proper_unit_for_byte_size(old_bytes_loaned_for_young_evac),
- byte_size_in_proper_unit(allocation_supplement),
- proper_unit_for_byte_size(allocation_supplement),
- byte_size_in_proper_unit(excess),
- proper_unit_for_byte_size(excess));
+ // Add in the excess_old memory to hold unanticipated promotions, if any. If there are more unanticipated
+ // promotions than fit in reserved memory, they will be deferred until a future GC pass.
+ size_t total_promotion_reserve = young_advance_promoted_reserve_used + excess_old;
+ heap->set_promoted_reserve(total_promotion_reserve);
+ heap->reset_promoted_expended();
}
void ShenandoahGeneration::prepare_regions_and_collection_set(bool concurrent) {
@@ -822,7 +487,6 @@ void ShenandoahGeneration::prepare_regions_and_collection_set(bool concurrent) {
// Budgeting parameters to compute_evacuation_budgets are passed by reference.
compute_evacuation_budgets(heap, preselected_regions, collection_set, consumed_by_advance_promotion);
-
_heuristics->choose_collection_set(collection_set, heap->old_heuristics());
if (!collection_set->is_empty()) {
// only make use of evacuation budgets when we are evacuating
@@ -839,7 +503,11 @@ void ShenandoahGeneration::prepare_regions_and_collection_set(bool concurrent) {
ShenandoahGCPhase phase(concurrent ? ShenandoahPhaseTimings::final_rebuild_freeset :
ShenandoahPhaseTimings::degen_gc_final_rebuild_freeset);
ShenandoahHeapLocker locker(heap->lock());
- heap->free_set()->rebuild();
+ size_t young_cset_regions, old_cset_regions;
+
+ // We are preparing for evacuation. At this time, we ignore cset region tallies.
+ heap->free_set()->prepare_to_rebuild(young_cset_regions, old_cset_regions);
+ heap->free_set()->rebuild(young_cset_regions, old_cset_regions);
}
heap->set_evacuation_reserve_quantities(false);
}
@@ -897,7 +565,7 @@ ShenandoahGeneration::ShenandoahGeneration(ShenandoahGenerationType type,
_collection_thread_time_s(0.0),
_affiliated_region_count(0), _humongous_waste(0), _used(0), _bytes_allocated_since_gc_start(0),
_max_capacity(max_capacity), _soft_max_capacity(soft_max_capacity),
- _adjusted_capacity(max_capacity), _heuristics(nullptr) {
+ _heuristics(nullptr) {
_is_marking_complete.set();
assert(max_workers > 0, "At least one queue");
for (uint i = 0; i < max_workers; ++i) {
@@ -955,6 +623,28 @@ size_t ShenandoahGeneration::decrement_affiliated_region_count() {
// on read and write of _affiliated_region_count. At the end of full gc, a single thread overwrites the count with
// a coherent value.
_affiliated_region_count--;
+ // TODO: REMOVE IS_GLOBAL() QUALIFIER AFTER WE FIX GLOBAL AFFILIATED REGION ACCOUNTING
+ assert(is_global() || ShenandoahHeap::heap()->is_full_gc_in_progress() ||
+ (_used + _humongous_waste <= _affiliated_region_count * ShenandoahHeapRegion::region_size_bytes()),
+ "used + humongous cannot exceed regions");
+ return _affiliated_region_count;
+}
+
+size_t ShenandoahGeneration::increase_affiliated_region_count(size_t delta) {
+ shenandoah_assert_heaplocked_or_fullgc_safepoint();
+ _affiliated_region_count += delta;
+ return _affiliated_region_count;
+}
+
+size_t ShenandoahGeneration::decrease_affiliated_region_count(size_t delta) {
+ shenandoah_assert_heaplocked_or_fullgc_safepoint();
+ assert(_affiliated_region_count > delta, "Affiliated region count cannot be negative");
+
+ _affiliated_region_count -= delta;
+ // TODO: REMOVE IS_GLOBAL() QUALIFIER AFTER WE FIX GLOBAL AFFILIATED REGION ACCOUNTING
+ assert(is_global() || ShenandoahHeap::heap()->is_full_gc_in_progress() ||
+ (_used + _humongous_waste <= _affiliated_region_count * ShenandoahHeapRegion::region_size_bytes()),
+ "used + humongous cannot exceed regions");
return _affiliated_region_count;
}
@@ -967,11 +657,11 @@ void ShenandoahGeneration::establish_usage(size_t num_regions, size_t num_bytes,
void ShenandoahGeneration::increase_used(size_t bytes) {
Atomic::add(&_used, bytes);
-}
-
-void ShenandoahGeneration::decrease_used(size_t bytes) {
- assert(_used >= bytes, "cannot reduce bytes used by generation below zero");
- Atomic::sub(&_used, bytes);
+ // This detects arithmetic wraparound on _used. Non-generational mode does not keep track of _affiliated_region_count
+ // TODO: REMOVE IS_GLOBAL() QUALIFIER AFTER WE FIX GLOBAL AFFILIATED REGION ACCOUNTING
+ assert(is_global() || ShenandoahHeap::heap()->is_full_gc_in_progress() ||
+ (_used + _humongous_waste <= _affiliated_region_count * ShenandoahHeapRegion::region_size_bytes()),
+ "used cannot exceed regions");
}
void ShenandoahGeneration::increase_humongous_waste(size_t bytes) {
@@ -982,13 +672,25 @@ void ShenandoahGeneration::increase_humongous_waste(size_t bytes) {
void ShenandoahGeneration::decrease_humongous_waste(size_t bytes) {
if (bytes > 0) {
- assert(_humongous_waste >= bytes, "Waste cannot be negative");
assert(ShenandoahHeap::heap()->is_full_gc_in_progress() || (_humongous_waste >= bytes),
"Waste (" SIZE_FORMAT ") cannot be negative (after subtracting " SIZE_FORMAT ")", _humongous_waste, bytes);
Atomic::sub(&_humongous_waste, bytes);
}
}
+void ShenandoahGeneration::decrease_used(size_t bytes) {
+ // TODO: REMOVE IS_GLOBAL() QUALIFIER AFTER WE FIX GLOBAL AFFILIATED REGION ACCOUNTING
+ assert(is_global() || ShenandoahHeap::heap()->is_full_gc_in_progress() ||
+ (_used >= bytes), "cannot reduce bytes used by generation below zero");
+ Atomic::sub(&_used, bytes);
+
+ // Non-generational mode does not maintain affiliated region counts
+ // TODO: REMOVE IS_GLOBAL() QUALIFIER AFTER WE FIX GLOBAL AFFILIATED REGION ACCOUNTING
+ assert(is_global() || ShenandoahHeap::heap()->is_full_gc_in_progress() ||
+ (_affiliated_region_count * ShenandoahHeapRegion::region_size_bytes() >= _used),
+ "Affiliated regions must hold more than what is currently used");
+}
+
size_t ShenandoahGeneration::used_regions() const {
return _affiliated_region_count;
}
@@ -996,7 +698,7 @@ size_t ShenandoahGeneration::used_regions() const {
size_t ShenandoahGeneration::free_unaffiliated_regions() const {
size_t result = max_capacity() / ShenandoahHeapRegion::region_size_bytes();
if (_affiliated_region_count > result) {
- result = 0; // If old-gen is loaning regions to young-gen, affiliated regions may exceed capacity temporarily.
+ result = 0;
} else {
result -= _affiliated_region_count;
}
@@ -1019,61 +721,48 @@ size_t ShenandoahGeneration::soft_available() const {
return in_use > soft_capacity ? 0 : soft_capacity - in_use;
}
-size_t ShenandoahGeneration::adjust_available(intptr_t adjustment) {
- assert(adjustment % ShenandoahHeapRegion::region_size_bytes() == 0,
- "Adjustment to generation size must be multiple of region size");
- _adjusted_capacity = max_capacity() + adjustment;
- return _adjusted_capacity;
-}
-
-size_t ShenandoahGeneration::unadjust_available() {
- _adjusted_capacity = max_capacity();
- return _adjusted_capacity;
-}
-
-size_t ShenandoahGeneration::adjusted_available() const {
- size_t in_use = used() + get_humongous_waste();
- size_t capacity = _adjusted_capacity;
- return in_use > capacity ? 0 : capacity - in_use;
-}
-
-size_t ShenandoahGeneration::adjusted_capacity() const {
- return _adjusted_capacity;
-}
-
-size_t ShenandoahGeneration::adjusted_unaffiliated_regions() const {
- // This assertion has been disabled because we expect this code to be replaced by 05/2023
- // assert(adjusted_capacity() >= used_regions_size(), "adjusted_unaffiliated_regions() cannot return negative");
- assert((adjusted_capacity() - used_regions_size()) % ShenandoahHeapRegion::region_size_bytes() == 0,
- "adjusted capacity (" SIZE_FORMAT ") and used regions size (" SIZE_FORMAT ") should be multiples of region_size_bytes",
- adjusted_capacity(), used_regions_size());
- return (adjusted_capacity() - used_regions_size()) / ShenandoahHeapRegion::region_size_bytes();
-}
-
void ShenandoahGeneration::increase_capacity(size_t increment) {
shenandoah_assert_heaplocked_or_safepoint();
- assert(_max_capacity + increment <= ShenandoahHeap::heap()->max_size_for(this), "Cannot increase generation capacity beyond maximum.");
- assert(increment % ShenandoahHeapRegion::region_size_bytes() == 0, "Region-sized changes only");
- // TODO: ysr: remove this check and warning
- if (increment % ShenandoahHeapRegion::region_size_bytes() != 0) {
- log_warning(gc)("Increment (" INTPTR_FORMAT ") should be a multiple of region size (" SIZE_FORMAT ")",
- increment, ShenandoahHeapRegion::region_size_bytes());
- }
+
+ // We do not enforce that new capacity >= heap->max_size_for(this). The maximum generation size is treated as a rule of thumb
+ // which may be violated during certain transitions, such as when we are forcing transfers for the purpose of promoting regions
+ // in place.
+ // TODO: REMOVE IS_GLOBAL() QUALIFIER AFTER WE FIX GLOBAL AFFILIATED REGION ACCOUNTING
+ assert(is_global() || ShenandoahHeap::heap()->is_full_gc_in_progress() ||
+ (_max_capacity + increment <= ShenandoahHeap::heap()->max_capacity()), "Generation cannot be larger than heap size");
+ assert(increment % ShenandoahHeapRegion::region_size_bytes() == 0, "Generation capacity must be multiple of region size");
_max_capacity += increment;
- _adjusted_capacity += increment;
+
+ // This detects arithmetic wraparound on _used
+ // TODO: REMOVE IS_GLOBAL() QUALIFIER AFTER WE FIX GLOBAL AFFILIATED REGION ACCOUNTING
+ assert(is_global() || ShenandoahHeap::heap()->is_full_gc_in_progress() ||
+ (_affiliated_region_count * ShenandoahHeapRegion::region_size_bytes() >= _used),
+ "Affiliated regions must hold more than what is currently used");
}
void ShenandoahGeneration::decrease_capacity(size_t decrement) {
shenandoah_assert_heaplocked_or_safepoint();
- assert(_max_capacity - decrement >= ShenandoahHeap::heap()->min_size_for(this), "Cannot decrease generation capacity beyond minimum.");
- assert(decrement % ShenandoahHeapRegion::region_size_bytes() == 0, "Region-sized changes only");
- // TODO: ysr: remove this check and warning
- if (decrement % ShenandoahHeapRegion::region_size_bytes() != 0) {
- log_warning(gc)("Decrement (" INTPTR_FORMAT ") should be a multiple of region size (" SIZE_FORMAT ")",
- decrement, ShenandoahHeapRegion::region_size_bytes());
- }
+
+ // We do not enforce that new capacity >= heap->min_size_for(this). The minimum generation size is treated as a rule of thumb
+ // which may be violated during certain transitions, such as when we are forcing transfers for the purpose of promoting regions
+ // in place.
+ assert(decrement % ShenandoahHeapRegion::region_size_bytes() == 0, "Generation capacity must be multiple of region size");
+ assert(_max_capacity >= decrement, "Generation capacity cannot be negative");
+
_max_capacity -= decrement;
- _adjusted_capacity -= decrement;
+
+ // This detects arithmetic wraparound on _used
+ // TODO: REMOVE IS_GLOBAL() QUALIFIER AFTER WE FIX GLOBAL AFFILIATED REGION ACCOUNTING
+ assert(is_global() || ShenandoahHeap::heap()->is_full_gc_in_progress() ||
+ (_affiliated_region_count * ShenandoahHeapRegion::region_size_bytes() >= _used),
+ "Affiliated regions must hold more than what is currently used");
+ // TODO: REMOVE IS_GLOBAL() QUALIFIER AFTER WE FIX GLOBAL AFFILIATED REGION ACCOUNTING
+ assert(is_global() || ShenandoahHeap::heap()->is_full_gc_in_progress() ||
+ (_used <= _max_capacity), "Cannot use more than capacity");
+ // TODO: REMOVE IS_GLOBAL() QUALIFIER AFTER WE FIX GLOBAL AFFILIATED REGION ACCOUNTING
+ assert(is_global() || ShenandoahHeap::heap()->is_full_gc_in_progress() ||
+ (_affiliated_region_count * ShenandoahHeapRegion::region_size_bytes() <= _max_capacity),
+ "Cannot use more than capacity");
}
void ShenandoahGeneration::record_success_concurrent(bool abbreviated) {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahGeneration.hpp b/src/hotspot/share/gc/shenandoah/shenandoahGeneration.hpp
index 160a51a443b..2bd85414280 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahGeneration.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahGeneration.hpp
@@ -50,8 +50,6 @@ class ShenandoahGeneration : public CHeapObj<mtGC> {
double _collection_thread_time_s;
-protected:
- // Usage
size_t _affiliated_region_count;
// How much free memory is left in the last region of humongous objects.
@@ -61,13 +59,15 @@ class ShenandoahGeneration : public CHeapObj<mtGC> {
// The units are bytes. The value is only changed on a safepoint or under the
// heap lock.
size_t _humongous_waste;
+
+protected:
+ // Usage
+
volatile size_t _used;
volatile size_t _bytes_allocated_since_gc_start;
size_t _max_capacity;
size_t _soft_max_capacity;
- size_t _adjusted_capacity;
-
ShenandoahHeuristics* _heuristics;
private:
@@ -117,22 +117,6 @@ class ShenandoahGeneration : public CHeapObj<mtGC> {
// max heap size will cause the adaptive heuristic to run more frequent cycles.
size_t soft_available() const;
- // During evacuation and update-refs, some memory may be shifted between generations. In particular, memory
- // may be loaned by old-gen to young-gen based on the promise the loan will be promptly repaid from the memory reclaimed
- // when the current collection set is recycled. The capacity adjustment also takes into consideration memory that is
- // set aside within each generation to hold the results of evacuation, but not promotion, into that region. Promotions
- // into old-gen are bounded by adjusted_available() whereas evacuations into old-gen are pre-committed.
- size_t adjusted_available() const;
- size_t adjusted_capacity() const;
-
- // This is the number of FREE regions that are eligible to be affiliated with this generation according to the current
- // adjusted capacity.
- size_t adjusted_unaffiliated_regions() const;
-
- // Both of following return new value of available
- size_t adjust_available(intptr_t adjustment);
- size_t unadjust_available();
-
size_t bytes_allocated_since_gc_start();
void reset_bytes_allocated_since_gc_start();
void increase_allocated(size_t bytes);
@@ -205,6 +189,12 @@ class ShenandoahGeneration : public CHeapObj<mtGC> {
// Return the updated value of affiliated_region_count
size_t decrement_affiliated_region_count();
+ // Return the updated value of affiliated_region_count
+ size_t increase_affiliated_region_count(size_t delta);
+
+ // Return the updated value of affiliated_region_count
+ size_t decrease_affiliated_region_count(size_t delta);
+
void establish_usage(size_t num_regions, size_t num_bytes, size_t humongous_waste);
void increase_used(size_t bytes);
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp b/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp
index 348e1900e39..eaa51009955 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp
@@ -379,8 +379,11 @@ jint ShenandoahHeap::initialize() {
// Initialize to complete
_marking_context->mark_complete();
+ size_t young_cset_regions, old_cset_regions;
- _free_set->rebuild();
+ // We are initializing free set. We ignore cset region tallies.
+ _free_set->prepare_to_rebuild(young_cset_regions, old_cset_regions);
+ _free_set->rebuild(young_cset_regions, old_cset_regions);
}
if (AlwaysPreTouch) {
@@ -519,15 +522,10 @@ void ShenandoahHeap::initialize_heuristics_generations() {
_young_generation = new ShenandoahYoungGeneration(_max_workers, max_capacity_young, initial_capacity_young);
_old_generation = new ShenandoahOldGeneration(_max_workers, max_capacity_old, initial_capacity_old);
- _global_generation = new ShenandoahGlobalGeneration(_gc_mode->is_generational(), _max_workers, soft_max_capacity(), soft_max_capacity());
-
+ _global_generation = new ShenandoahGlobalGeneration(_gc_mode->is_generational(), _max_workers, max_capacity(), max_capacity());
_global_generation->initialize_heuristics(_gc_mode);
- if (mode()->is_generational()) {
- _young_generation->initialize_heuristics(_gc_mode);
- _old_generation->initialize_heuristics(_gc_mode);
-
- ShenandoahEvacWaste = ShenandoahGenerationalEvacWaste;
- }
+ _young_generation->initialize_heuristics(_gc_mode);
+ _old_generation->initialize_heuristics(_gc_mode);
}
#ifdef _MSC_VER
@@ -540,6 +538,8 @@ ShenandoahHeap::ShenandoahHeap(ShenandoahCollectorPolicy* policy) :
_gc_generation(nullptr),
_prepare_for_old_mark(false),
_initial_size(0),
+ _promotion_potential(0),
+ _promotion_in_place_potential(0),
_committed(0),
_max_workers(MAX3(ConcGCThreads, ParallelGCThreads, 1U)),
_workers(nullptr),
@@ -549,7 +549,6 @@ ShenandoahHeap::ShenandoahHeap(ShenandoahCollectorPolicy* policy) :
_regions(nullptr),
_affiliations(nullptr),
_update_refs_iterator(this),
- _alloc_supplement_reserve(0),
_promoted_reserve(0),
_old_evac_reserve(0),
_old_evac_expended(0),
@@ -581,6 +580,8 @@ ShenandoahHeap::ShenandoahHeap(ShenandoahCollectorPolicy* policy) :
_gc_timer(new ConcurrentGCTimer()),
_soft_ref_policy(),
_log_min_obj_alignment_in_bytes(LogMinObjAlignmentInBytes),
+ _old_regions_surplus(0),
+ _old_regions_deficit(0),
_marking_context(nullptr),
_bitmap_size(0),
_bitmap_regions_per_slice(0),
@@ -818,13 +819,6 @@ void ShenandoahHeap::set_soft_max_capacity(size_t v) {
"Should be in bounds: " SIZE_FORMAT " <= " SIZE_FORMAT " <= " SIZE_FORMAT,
min_capacity(), v, max_capacity());
Atomic::store(&_soft_max_size, v);
-
- if (mode()->is_generational()) {
- size_t max_capacity_young = _generation_sizer.max_young_size();
- size_t min_capacity_young = _generation_sizer.min_young_size();
- size_t new_capacity_young = clamp(v, min_capacity_young, max_capacity_young);
- _young_generation->set_soft_max_capacity(new_capacity_young);
- }
}
size_t ShenandoahHeap::min_capacity() const {
@@ -893,9 +887,7 @@ void ShenandoahHeap::handle_promotion_failure() {
}
void ShenandoahHeap::report_promotion_failure(Thread* thread, size_t size) {
- // We squelch excessive reports to reduce noise in logs. Squelch enforcement is not "perfect" because
- // this same code can be in-lined in multiple contexts, and each context will have its own copy of the static
- // last_report_epoch and this_epoch_report_count variables.
+ // We squelch excessive reports to reduce noise in logs.
const size_t MaxReportsPerEpoch = 4;
static size_t last_report_epoch = 0;
static size_t epoch_report_count = 0;
@@ -915,11 +907,18 @@ void ShenandoahHeap::report_promotion_failure(Thread* thread, size_t size) {
PLAB* plab = ShenandoahThreadLocalData::plab(thread);
size_t words_remaining = (plab == nullptr)? 0: plab->words_remaining();
const char* promote_enabled = ShenandoahThreadLocalData::allow_plab_promotions(thread)? "enabled": "disabled";
+ ShenandoahGeneration* old_gen = old_generation();
+ size_t old_capacity = old_gen->max_capacity();
+ size_t old_usage = old_gen->used();
+ size_t old_free_regions = old_gen->free_unaffiliated_regions();
log_info(gc, ergo)("Promotion failed, size " SIZE_FORMAT ", has plab? %s, PLAB remaining: " SIZE_FORMAT
- ", plab promotions %s, promotion reserve: " SIZE_FORMAT ", promotion expended: " SIZE_FORMAT,
- size, plab == nullptr? "no": "yes",
- words_remaining, promote_enabled, promotion_reserve, promotion_expended);
+ ", plab promotions %s, promotion reserve: " SIZE_FORMAT ", promotion expended: " SIZE_FORMAT
+ ", old capacity: " SIZE_FORMAT ", old_used: " SIZE_FORMAT ", old unaffiliated regions: " SIZE_FORMAT,
+ size * HeapWordSize, plab == nullptr? "no": "yes",
+ words_remaining * HeapWordSize, promote_enabled, promotion_reserve, promotion_expended,
+ old_capacity, old_usage, old_free_regions);
+
if ((gc_id == last_report_epoch) && (epoch_report_count >= MaxReportsPerEpoch)) {
log_info(gc, ergo)("Squelching additional promotion failure reports for current epoch");
} else if (gc_id != last_report_epoch) {
@@ -1029,7 +1028,6 @@ HeapWord* ShenandoahHeap::allocate_from_plab_slow(Thread* thread, size_t size, b
// CAUTION: retire_plab may register the remnant filler object with the remembered set scanner without a lock. This
// is safe iff it is assured that each PLAB is a whole-number multiple of card-mark memory size and each PLAB is
// aligned with the start of a card's memory range.
-
retire_plab(plab, thread);
size_t actual_size = 0;
@@ -1037,7 +1035,12 @@ HeapWord* ShenandoahHeap::allocate_from_plab_slow(Thread* thread, size_t size, b
// less than the remaining evacuation need. It also adjusts plab_preallocated and expend_promoted if appropriate.
HeapWord* plab_buf = allocate_new_plab(min_size, cur_size, &actual_size);
if (plab_buf == nullptr) {
- return nullptr;
+ if (min_size == PLAB::min_size()) {
+ // Disable plab promotions for this thread because we cannot even allocate a plab of minimal size. This allows us
+ // to fail faster on subsequent promotion attempts.
+ ShenandoahThreadLocalData::disable_plab_promotions(thread);
+ }
+ return NULL;
} else {
ShenandoahThreadLocalData::enable_plab_retries(thread);
}
@@ -1056,7 +1059,6 @@ HeapWord* ShenandoahHeap::allocate_from_plab_slow(Thread* thread, size_t size, b
#endif // ASSERT
}
plab->set_buf(plab_buf, actual_size);
-
if (is_promotion && !ShenandoahThreadLocalData::allow_plab_promotions(thread)) {
return nullptr;
}
@@ -1154,11 +1156,93 @@ void ShenandoahHeap::coalesce_and_fill_old_regions() {
parallel_heap_region_iterate(&coalesce);
}
-bool ShenandoahHeap::adjust_generation_sizes() {
- if (mode()->is_generational()) {
- return _generation_sizer.adjust_generation_sizes();
+// xfer_limit is the maximum we're able to transfer from young to old
+void ShenandoahHeap::adjust_generation_sizes_for_next_cycle(
+ size_t xfer_limit, size_t young_cset_regions, size_t old_cset_regions) {
+
+ // Make sure old-generation is large enough, but no larger, than is necessary to hold mixed evacuations
+ // and promotions if we anticipate either.
+ size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+ size_t promo_load = get_promotion_potential();
+ // The free set will reserve this amount of memory to hold young evacuations
+ size_t young_reserve = (young_generation()->max_capacity() * ShenandoahEvacReserve) / 100;
+ size_t old_reserve = 0;
+ size_t mixed_candidates = old_heuristics()->unprocessed_old_collection_candidates();
+ bool doing_mixed = (mixed_candidates > 0);
+ bool doing_promotions = promo_load > 0;
+
+ // round down
+ size_t max_old_region_xfer = xfer_limit / region_size_bytes;
+
+ // We can limit the reserve to the size of anticipated promotions
+ size_t max_old_reserve = young_reserve * ShenandoahOldEvacRatioPercent / (100 - ShenandoahOldEvacRatioPercent);
+ // Here's the algebra:
+ // TotalEvacuation = OldEvacuation + YoungEvacuation
+ // OldEvacuation = TotalEvacuation*(ShenandoahOldEvacRatioPercent/100)
+ // OldEvacuation = YoungEvacuation * (ShenandoahOldEvacRatioPercent/100)/(1 - ShenandoahOldEvacRatioPercent/100)
+ // OldEvacuation = YoungEvacuation * ShenandoahOldEvacRatioPercent/(100 - ShenandoahOldEvacRatioPercent)
+
+ size_t reserve_for_mixed, reserve_for_promo;
+ if (doing_mixed) {
+ assert(old_generation()->available() >= old_generation()->free_unaffiliated_regions() * region_size_bytes,
+ "Unaffiliated available must be less than total available");
+
+ // We want this much memory to be unfragmented in order to reliably evacuate old. This is conservative because we
+ // may not evacuate the entirety of unprocessed candidates in a single mixed evacuation.
+ size_t max_evac_need = (size_t)
+ (old_heuristics()->unprocessed_old_collection_candidates_live_memory() * ShenandoahOldEvacWaste);
+ size_t old_fragmented_available =
+ old_generation()->available() - old_generation()->free_unaffiliated_regions() * region_size_bytes;
+ reserve_for_mixed = max_evac_need + old_fragmented_available;
+ if (reserve_for_mixed > max_old_reserve) {
+ reserve_for_mixed = max_old_reserve;
+ }
+ } else {
+ reserve_for_mixed = 0;
}
- return false;
+
+ size_t available_for_promotions = max_old_reserve - reserve_for_mixed;
+ if (doing_promotions) {
+ // We're only promoting and we have a maximum bound on the amount to be promoted
+ reserve_for_promo = (size_t) (promo_load * ShenandoahPromoEvacWaste);
+ if (reserve_for_promo > available_for_promotions) {
+ reserve_for_promo = available_for_promotions;
+ }
+ } else {
+ reserve_for_promo = 0;
+ }
+ old_reserve = reserve_for_mixed + reserve_for_promo;
+ assert(old_reserve <= max_old_reserve, "cannot reserve more than max for old evacuations");
+ size_t old_available = old_generation()->available() + old_cset_regions * region_size_bytes;
+ size_t young_available = young_generation()->available() + young_cset_regions * region_size_bytes;
+ size_t old_region_deficit = 0;
+ size_t old_region_surplus = 0;
+ if (old_available >= old_reserve) {
+ size_t old_excess = old_available - old_reserve;
+ size_t excess_regions = old_excess / region_size_bytes;
+ size_t unaffiliated_old_regions = old_generation()->free_unaffiliated_regions() + old_cset_regions;
+ size_t unaffiliated_old = unaffiliated_old_regions * region_size_bytes;
+ if (unaffiliated_old_regions < excess_regions) {
+ // We'll give only unaffiliated old to young, which is known to be less than the excess.
+ old_region_surplus = unaffiliated_old_regions;
+ } else {
+ // unaffiliated_old_regions > excess_regions, so we only give away the excess.
+ old_region_surplus = excess_regions;
+ }
+ } else {
+ // We need to request transfer from YOUNG. Ignore that this will directly impact young_generation()->max_capacity(),
+ // indirectly impacting young_reserve and old_reserve. These computations are conservative.
+ size_t old_need = old_reserve - old_available;
+ // Round up the number of regions needed from YOUNG
+ old_region_deficit = (old_need + region_size_bytes - 1) / region_size_bytes;
+ }
+ if (old_region_deficit > max_old_region_xfer) {
+ // If we're running short on young-gen memory, limit the xfer. Old-gen collection activities will be curtailed
+ // if the budget is smaller than desired.
+ old_region_deficit = max_old_region_xfer;
+ }
+ set_old_region_surplus(old_region_surplus);
+ set_old_region_deficit(old_region_deficit);
}
// Called from stubs in JIT code or interpreter
@@ -1193,7 +1277,7 @@ HeapWord* ShenandoahHeap::allocate_new_plab(size_t min_size,
size_t* actual_size) {
ShenandoahAllocRequest req = ShenandoahAllocRequest::for_plab(min_size, word_size);
// Note that allocate_memory() sets a thread-local flag to prohibit further promotions by this thread
- // if we are at risk of exceeding the old-gen evacuation budget.
+ // if we are at risk of infringing on the old-gen evacuation budget.
HeapWord* res = allocate_memory(req, false);
if (res != nullptr) {
*actual_size = req.actual_size();
@@ -1228,14 +1312,16 @@ HeapWord* ShenandoahHeap::allocate_memory(ShenandoahAllocRequest& req, bool is_p
// strategy is to try again, as long as GC makes progress.
//
// Then, we need to make sure the allocation was retried after at least one
- // Full GC, which means we want to try more than ShenandoahFullGCThreshold times.
+ // Full GC.
size_t tries = 0;
+ size_t original_fullgc_count = shenandoah_policy()->get_fullgc_count();
while (result == nullptr && _progress_last_gc.is_set()) {
tries++;
control_thread()->handle_alloc_failure(req);
result = allocate_memory_under_lock(req, in_new_region, is_promotion);
}
- while (result == nullptr && tries <= ShenandoahFullGCThreshold) {
+ while (result == nullptr &&
+ ((shenandoah_policy()->get_fullgc_count() == original_fullgc_count) || (tries <= ShenandoahOOMGCRetries))) {
tries++;
control_thread()->handle_alloc_failure(req);
result = allocate_memory_under_lock(req, in_new_region, is_promotion);
@@ -1297,20 +1383,18 @@ HeapWord* ShenandoahHeap::allocate_memory_under_lock(ShenandoahAllocRequest& req
if (mode()->is_generational()) {
if (req.affiliation() == YOUNG_GENERATION) {
if (req.is_mutator_alloc()) {
- size_t young_available = young_generation()->adjusted_available();
- if (requested_bytes > young_available) {
- // We know this is not a GCLAB. This must be a TLAB or a shared allocation.
- if (req.is_lab_alloc() && (young_available >= req.min_size())) {
- try_smaller_lab_size = true;
- smaller_lab_size = young_available / HeapWordSize;
- } else {
- // Can't allocate because even min_size() is larger than remaining young_available
- log_info(gc, ergo)("Unable to shrink %s alloc request of minimum size: " SIZE_FORMAT
- ", young available: " SIZE_FORMAT,
- req.is_lab_alloc()? "TLAB": "shared",
- HeapWordSize * (req.is_lab_alloc()? req.min_size(): req.size()), young_available);
- return nullptr;
- }
+ size_t young_words_available = young_generation()->available() / HeapWordSize;
+ if (ShenandoahElasticTLAB && req.is_lab_alloc() && (req.min_size() < young_words_available)) {
+ // Allow ourselves to try a smaller lab size even if requested_bytes <= young_available. We may need a smaller
+ // lab size because young memory has become too fragmented.
+ try_smaller_lab_size = true;
+ smaller_lab_size = (young_words_available < req.size())? young_words_available: req.size();
+ } else if (req.size() > young_words_available) {
+ // Can't allocate because even min_size() is larger than remaining young_available
+ log_info(gc, ergo)("Unable to shrink %s alloc request of minimum size: " SIZE_FORMAT
+ ", young words available: " SIZE_FORMAT, req.type_string(),
+ HeapWordSize * (req.is_lab_alloc()? req.min_size(): req.size()), young_words_available);
+ return nullptr;
}
}
} else { // reg.affiliation() == OLD_GENERATION
@@ -1351,66 +1435,83 @@ HeapWord* ShenandoahHeap::allocate_memory_under_lock(ShenandoahAllocRequest& req
}
} // This ends the is_generational() block
- if (!try_smaller_lab_size) {
- result = (allow_allocation)? _free_set->allocate(req, in_new_region): nullptr;
- if (result != nullptr) {
- if (req.is_old()) {
- ShenandoahThreadLocalData::reset_plab_promoted(thread);
- if (req.is_gc_alloc()) {
- if (req.type() == ShenandoahAllocRequest::_alloc_plab) {
- if (promotion_eligible) {
- size_t actual_size = req.actual_size() * HeapWordSize;
+ // First try the original request. If TLAB request size is greater than available, allocate() will attempt to downsize
+ // request to fit within available memory.
+ result = (allow_allocation)? _free_set->allocate(req, in_new_region): nullptr;
+ if (result != nullptr) {
+ if (req.is_old()) {
+ ShenandoahThreadLocalData::reset_plab_promoted(thread);
+ if (req.is_gc_alloc()) {
+ bool disable_plab_promotions = false;
+ if (req.type() == ShenandoahAllocRequest::_alloc_plab) {
+ if (promotion_eligible) {
+ size_t actual_size = req.actual_size() * HeapWordSize;
+ // The actual size of the allocation may be larger than the requested bytes (due to alignment on card boundaries).
+ // If this puts us over our promotion budget, we need to disable future PLAB promotions for this thread.
+ if (get_promoted_expended() + actual_size <= get_promoted_reserve()) {
// Assume the entirety of this PLAB will be used for promotion. This prevents promotion from overreach.
// When we retire this plab, we'll unexpend what we don't really use.
ShenandoahThreadLocalData::enable_plab_promotions(thread);
expend_promoted(actual_size);
- // This assert has been disabled because we expect this code to be replaced by 05/2023.
- // assert(get_promoted_expended() <= get_promoted_reserve(), "Do not expend more promotion than budgeted");
+ assert(get_promoted_expended() <= get_promoted_reserve(), "Do not expend more promotion than budgeted");
ShenandoahThreadLocalData::set_plab_preallocated_promoted(thread, actual_size);
} else {
- // Disable promotions in this thread because entirety of this PLAB must be available to hold old-gen evacuations.
- ShenandoahThreadLocalData::disable_plab_promotions(thread);
- ShenandoahThreadLocalData::set_plab_preallocated_promoted(thread, 0);
+ disable_plab_promotions = true;
}
- } else if (is_promotion) {
- // Shared promotion. Assume size is requested_bytes.
- expend_promoted(requested_bytes);
- assert(get_promoted_expended() <= get_promoted_reserve(), "Do not expend more promotion than budgeted");
+ } else {
+ disable_plab_promotions = true;
}
+ if (disable_plab_promotions) {
+ // Disable promotions in this thread because entirety of this PLAB must be available to hold old-gen evacuations.
+ ShenandoahThreadLocalData::disable_plab_promotions(thread);
+ ShenandoahThreadLocalData::set_plab_preallocated_promoted(thread, 0);
+ }
+ } else if (is_promotion) {
+ // Shared promotion. Assume size is requested_bytes.
+ expend_promoted(requested_bytes);
+ assert(get_promoted_expended() <= get_promoted_reserve(), "Do not expend more promotion than budgeted");
}
-
- // Register the newly allocated object while we're holding the global lock since there's no synchronization
- // built in to the implementation of register_object(). There are potential races when multiple independent
- // threads are allocating objects, some of which might span the same card region. For example, consider
- // a card table's memory region within which three objects are being allocated by three different threads:
- //
- // objects being "concurrently" allocated:
- // [-----a------][-----b-----][--------------c------------------]
- // [---- card table memory range --------------]
- //
- // Before any objects are allocated, this card's memory range holds no objects. Note that allocation of object a
- // wants to set the has-object, first-start, and last-start attributes of the preceding card region.
- // allocation of object b wants to set the has-object, first-start, and last-start attributes of this card region.
- // allocation of object c also wants to set the has-object, first-start, and last-start attributes of this card region.
- //
- // The thread allocating b and the thread allocating c can "race" in various ways, resulting in confusion, such as
- // last-start representing object b while first-start represents object c. This is why we need to require all
- // register_object() invocations to be "mutually exclusive" with respect to each card's memory range.
- ShenandoahHeap::heap()->card_scan()->register_object(result);
- }
- } else {
- // The allocation failed. If this was a plab allocation, We've already retired it and no longer have a plab.
- if (req.is_old() && req.is_gc_alloc() &&
- (req.type() == ShenandoahAllocRequest::_alloc_plab)) {
- // We don't need to disable PLAB promotions because there is no PLAB. We leave promotions enabled because
- // this allows the surrounding infrastructure to retry alloc_plab_slow() with a smaller PLAB size.
- ShenandoahThreadLocalData::set_plab_preallocated_promoted(thread, 0);
}
+
+ // Register the newly allocated object while we're holding the global lock since there's no synchronization
+ // built in to the implementation of register_object(). There are potential races when multiple independent
+ // threads are allocating objects, some of which might span the same card region. For example, consider
+ // a card table's memory region within which three objects are being allocated by three different threads:
+ //
+ // objects being "concurrently" allocated:
+ // [-----a------][-----b-----][--------------c------------------]
+ // [---- card table memory range --------------]
+ //
+ // Before any objects are allocated, this card's memory range holds no objects. Note that allocation of object a
+ // wants to set the starts-object, first-start, and last-start attributes of the preceding card region.
+ // allocation of object b wants to set the starts-object, first-start, and last-start attributes of this card region.
+ // allocation of object c also wants to set the starts-object, first-start, and last-start attributes of this
+ // card region.
+ //
+ // The thread allocating b and the thread allocating c can "race" in various ways, resulting in confusion, such as
+ // last-start representing object b while first-start represents object c. This is why we need to require all
+ // register_object() invocations to be "mutually exclusive" with respect to each card's memory range.
+ ShenandoahHeap::heap()->card_scan()->register_object(result);
+ }
+ } else {
+ // The allocation failed. If this was a plab allocation, We've already retired it and no longer have a plab.
+ if (req.is_old() && req.is_gc_alloc() && (req.type() == ShenandoahAllocRequest::_alloc_plab)) {
+ // We don't need to disable PLAB promotions because there is no PLAB. We leave promotions enabled because
+ // this allows the surrounding infrastructure to retry alloc_plab_slow() with a smaller PLAB size.
+ ShenandoahThreadLocalData::set_plab_preallocated_promoted(thread, 0);
}
+ }
+ if ((result != nullptr) || !try_smaller_lab_size) {
return result;
}
- // else, try_smaller_lab_size is true so we fall through and recurse with a smaller lab size
- } // This closes the block that holds the heap lock. This releases the lock.
+ // else, fall through to try_smaller_lab_size
+ } // This closes the block that holds the heap lock, releasing the lock.
+
+ // We failed to allocate the originally requested lab size. Let's see if we can allocate a smaller lab size.
+ if (req.size() == smaller_lab_size) {
+ // If we were already trying to allocate min size, no value in attempting to repeat the same. End the recursion.
+ return nullptr;
+ }
// We arrive here if the tlab allocation request can be resized to fit within young_available
assert((req.affiliation() == YOUNG_GENERATION) && req.is_lab_alloc() && req.is_mutator_alloc() &&
@@ -1589,31 +1690,41 @@ class ShenandoahGenerationalEvacuationTask : public WorkerTask {
void do_work() {
ShenandoahConcurrentEvacuateRegionObjectClosure cl(_sh);
ShenandoahHeapRegion* r;
+ ShenandoahMarkingContext* const ctx = ShenandoahHeap::heap()->marking_context();
+ size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+ size_t old_garbage_threshold = (region_size_bytes * ShenandoahOldGarbageThreshold) / 100;
while ((r = _regions->next()) != nullptr) {
- log_debug(gc)("GenerationalEvacuationTask do_work(), looking at %s region " SIZE_FORMAT ", (age: %d) [%s, %s]",
+ log_debug(gc)("GenerationalEvacuationTask do_work(), looking at %s region " SIZE_FORMAT ", (age: %d) [%s, %s, %s]",
r->is_old()? "old": r->is_young()? "young": "free", r->index(), r->age(),
r->is_active()? "active": "inactive",
- r->is_humongous()? (r->is_humongous_start()? "humongous_start": "humongous_continuation"): "regular");
+ r->is_humongous()? (r->is_humongous_start()? "humongous_start": "humongous_continuation"): "regular",
+ r->is_cset()? "cset": "not-cset");
+
if (r->is_cset()) {
assert(r->has_live(), "Region " SIZE_FORMAT " should have been reclaimed early", r->index());
_sh->marked_object_iterate(r, &cl);
if (ShenandoahPacing) {
_sh->pacer()->report_evac(r->used() >> LogHeapWordSize);
}
- } else if (r->is_young() && r->is_active() && r->is_humongous_start() && (r->age() > InitialTenuringThreshold)) {
- // We promote humongous_start regions along with their affiliated continuations during evacuation rather than
- // doing this work during a safepoint. We cannot put humongous regions into the collection set because that
- // triggers the load-reference barrier (LRB) to copy on reference fetch.
- if (r->promote_humongous() == 0) {
- // We chose not to promote because old-gen is out of memory. Report and handle the promotion failure because
- // this suggests need for expanding old-gen and/or performing collection of old-gen.
- ShenandoahHeap* heap = ShenandoahHeap::heap();
- oop obj = cast_to_oop(r->bottom());
- size_t size = obj->size();
- Thread* thread = Thread::current();
- heap->report_promotion_failure(thread, size);
- heap->handle_promotion_failure();
+ } else if (r->is_young() && r->is_active() && (r->age() >= InitialTenuringThreshold)) {
+ HeapWord* tams = ctx->top_at_mark_start(r);
+ if (r->is_humongous_start()) {
+ // We promote humongous_start regions along with their affiliated continuations during evacuation rather than
+ // doing this work during a safepoint. We cannot put humongous regions into the collection set because that
+ // triggers the load-reference barrier (LRB) to copy on reference fetch.
+ r->promote_humongous();
+ } else if (r->is_regular() && (r->garbage_before_padded_for_promote() < old_garbage_threshold) && (r->get_top_before_promote() == tams)) {
+ // Likewise, we cannot put promote-in-place regions into the collection set because that would also trigger
+ // the LRB to copy on reference fetch.
+ r->promote_in_place();
}
+ // Aged humongous continuation regions are handled with their start region. If an aged regular region has
+ // more garbage than ShenandoahOldGarbageTrheshold, we'll promote by evacuation. If there is room for evacuation
+ // in this cycle, the region will be in the collection set. If there is not room, the region will be promoted
+ // by evacuation in some future GC cycle.
+
+ // If an aged regular region has received allocations during the current cycle, we do not promote because the
+ // newly allocated objects do not have appropriate age; this region's age will be reset to zero at end of cycle.
}
// else, region is free, or OLD, or not in collection set, or humongous_continuation,
// or is young humongous_start that is too young to be promoted
@@ -1812,7 +1923,7 @@ void ShenandoahHeap::set_young_lab_region_flags() {
size_t ShenandoahHeap::unsafe_max_tlab_alloc(Thread *thread) const {
if (ShenandoahElasticTLAB) {
if (mode()->is_generational()) {
- return MIN2(ShenandoahHeapRegion::max_tlab_size_bytes(), young_generation()->adjusted_available());
+ return MIN2(ShenandoahHeapRegion::max_tlab_size_bytes(), young_generation()->available());
} else {
// With Elastic TLABs, return the max allowed size, and let the allocation path
// figure out the safe size for current allocation.
@@ -1900,24 +2011,16 @@ void ShenandoahHeap::on_cycle_start(GCCause::Cause cause, ShenandoahGeneration*
shenandoah_policy()->record_cycle_start();
generation->heuristics()->record_cycle_start();
-
- // When a cycle starts, attribute any thread activity when the collector
- // is idle to the global generation.
- _mmu_tracker.record(global_generation());
}
void ShenandoahHeap::on_cycle_end(ShenandoahGeneration* generation) {
generation->heuristics()->record_cycle_end();
-
if (mode()->is_generational() && (generation->is_global() || upgraded_to_full())) {
// If we just completed a GLOBAL GC, claim credit for completion of young-gen and old-gen GC as well
young_generation()->heuristics()->record_cycle_end();
old_generation()->heuristics()->record_cycle_end();
}
set_gc_cause(GCCause::_no_gc);
-
- // When a cycle ends, the thread activity is attributed to the respective generation
- _mmu_tracker.record(generation);
}
void ShenandoahHeap::verify(VerifyOption vo) {
@@ -2676,6 +2779,16 @@ class ShenandoahUpdateHeapRefsTask : public WorkerTask {
template<class T>
void do_work(uint worker_id) {
T cl;
+ if (CONCURRENT && (worker_id == 0)) {
+ // We ask the first worker to replenish the Mutator free set by moving regions previously reserved to hold the
+ // results of evacuation. These reserves are no longer necessary because evacuation has completed.
+ size_t cset_regions = _heap->collection_set()->count();
+ // We cannot transfer any more regions than will be reclaimed when the existing collection set is recycled, because
+ // we need the reclaimed collection set regions to replenish the collector reserves
+ _heap->free_set()->move_collector_sets_to_mutator(cset_regions);
+ }
+ // If !CONCURRENT, there's no value in expanding Mutator free set
+
ShenandoahHeapRegion* r = _regions->next();
// We update references for global, old, and young collections.
assert(_heap->active_generation()->is_mark_complete(), "Expected complete marking");
@@ -2938,12 +3051,58 @@ void ShenandoahHeap::update_heap_region_states(bool concurrent) {
}
void ShenandoahHeap::rebuild_free_set(bool concurrent) {
- {
- ShenandoahGCPhase phase(concurrent ?
- ShenandoahPhaseTimings::final_update_refs_rebuild_freeset :
- ShenandoahPhaseTimings::degen_gc_final_update_refs_rebuild_freeset);
- ShenandoahHeapLocker locker(lock());
- _free_set->rebuild();
+ ShenandoahGCPhase phase(concurrent ?
+ ShenandoahPhaseTimings::final_update_refs_rebuild_freeset :
+ ShenandoahPhaseTimings::degen_gc_final_update_refs_rebuild_freeset);
+ size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+ ShenandoahHeapLocker locker(lock());
+ size_t young_cset_regions, old_cset_regions;
+ _free_set->prepare_to_rebuild(young_cset_regions, old_cset_regions);
+
+ if (mode()->is_generational()) {
+ assert(verify_generation_usage(true, old_generation()->used_regions(),
+ old_generation()->used(), old_generation()->get_humongous_waste(),
+ true, young_generation()->used_regions(),
+ young_generation()->used(), young_generation()->get_humongous_waste()),
+ "Generation accounts are inaccurate");
+
+ // The computation of bytes_of_allocation_runway_before_gc_trigger is quite conservative so consider all of this
+ // available for transfer to old. Note that transfer of humongous regions does not impact available.
+ size_t allocation_runway = young_generation()->heuristics()->bytes_of_allocation_runway_before_gc_trigger(young_cset_regions);
+ adjust_generation_sizes_for_next_cycle(allocation_runway, young_cset_regions, old_cset_regions);
+
+ // Total old_available may have been expanded to hold anticipated promotions. We trigger if the fragmented available
+ // memory represents more than 16 regions worth of data. Note that fragmentation may increase when we promote regular
+ // regions in place when many of these regular regions have an abundant amount of available memory within them. Fragmentation
+ // will decrease as promote-by-copy consumes the available memory within these partially consumed regions.
+ //
+ // We consider old-gen to have excessive fragmentation if more than 12.5% of old-gen is free memory that resides
+ // within partially consumed regions of memory.
+ }
+ // Rebuild free set based on adjusted generation sizes.
+ _free_set->rebuild(young_cset_regions, old_cset_regions);
+
+ if (mode()->is_generational()) {
+ size_t old_available = old_generation()->available();
+ size_t old_unaffiliated_available = old_generation()->free_unaffiliated_regions() * region_size_bytes;
+ size_t old_fragmented_available;
+ assert(old_available >= old_unaffiliated_available, "unaffiliated available is a subset of total available");
+ old_fragmented_available = old_available - old_unaffiliated_available;
+
+ size_t old_capacity = old_generation()->max_capacity();
+ size_t heap_capacity = capacity();
+ if ((old_capacity > heap_capacity / 8) && (old_fragmented_available > old_capacity / 8)) {
+ ((ShenandoahOldHeuristics *) old_generation()->heuristics())->trigger_old_is_fragmented();
+ }
+
+ size_t old_used = old_generation()->used() + old_generation()->get_humongous_waste();
+ size_t trigger_threshold = old_generation()->usage_trigger_threshold();
+ // Detects unsigned arithmetic underflow
+ assert(old_used < ShenandoahHeap::heap()->capacity(), "Old used must be less than heap capacity");
+
+ if (old_used > trigger_threshold) {
+ ((ShenandoahOldHeuristics *) old_generation()->heuristics())->trigger_old_has_grown();
+ }
}
}
@@ -3187,6 +3346,57 @@ void ShenandoahGenerationRegionClosure<GLOBAL_NON_GEN>::heap_region_do(Shenandoa
_cl->heap_region_do(region);
}
+bool ShenandoahHeap::verify_generation_usage(bool verify_old, size_t old_regions, size_t old_bytes, size_t old_waste,
+ bool verify_young, size_t young_regions, size_t young_bytes, size_t young_waste) {
+ size_t tally_old_regions = 0;
+ size_t tally_old_bytes = 0;
+ size_t tally_old_waste = 0;
+ size_t tally_young_regions = 0;
+ size_t tally_young_bytes = 0;
+ size_t tally_young_waste = 0;
+
+ shenandoah_assert_heaplocked_or_safepoint();
+ for (size_t i = 0; i < num_regions(); i++) {
+ ShenandoahHeapRegion* r = get_region(i);
+ if (r->is_old()) {
+ tally_old_regions++;
+ tally_old_bytes += r->used();
+ if (r->is_humongous()) {
+ ShenandoahHeapRegion* start = r->humongous_start_region();
+ HeapWord* obj_addr = start->bottom();
+ oop obj = cast_to_oop(obj_addr);
+ size_t word_size = obj->size();
+ HeapWord* end_addr = obj_addr + word_size;
+ if (end_addr <= r->end()) {
+ tally_old_waste += (r->end() - end_addr) * HeapWordSize;
+ }
+ }
+ } else if (r->is_young()) {
+ tally_young_regions++;
+ tally_young_bytes += r->used();
+ if (r->is_humongous()) {
+ ShenandoahHeapRegion* start = r->humongous_start_region();
+ HeapWord* obj_addr = start->bottom();
+ oop obj = cast_to_oop(obj_addr);
+ size_t word_size = obj->size();
+ HeapWord* end_addr = obj_addr + word_size;
+ if (end_addr <= r->end()) {
+ tally_young_waste += (r->end() - end_addr) * HeapWordSize;
+ }
+ }
+ }
+ }
+ if (verify_young &&
+ ((young_regions != tally_young_regions) || (young_bytes != tally_young_bytes) || (young_waste != tally_young_waste))) {
+ return false;
+ } else if (verify_old &&
+ ((old_regions != tally_old_regions) || (old_bytes != tally_old_bytes) || (old_waste != tally_old_waste))) {
+ return false;
+ } else {
+ return true;
+ }
+}
+
ShenandoahGeneration* ShenandoahHeap::generation_for(ShenandoahAffiliation affiliation) const {
if (!mode()->is_generational()) {
return global_generation();
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp b/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp
index 105835e5728..368ae2a1ab3 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp
@@ -205,6 +205,9 @@ class ShenandoahHeap : public CollectedHeap {
void prepare_for_verify() override;
void verify(VerifyOption vo) override;
+ bool verify_generation_usage(bool verify_old, size_t old_regions, size_t old_bytes, size_t old_waste,
+ bool verify_young, size_t young_regions, size_t young_bytes, size_t young_waste);
+
// WhiteBox testing support.
bool supports_concurrent_gc_breakpoints() const override {
return true;
@@ -215,6 +218,14 @@ class ShenandoahHeap : public CollectedHeap {
private:
size_t _initial_size;
size_t _minimum_size;
+ size_t _promotion_potential;
+ size_t _promotion_in_place_potential;
+ size_t _pad_for_promote_in_place; // bytes of filler
+ size_t _promotable_humongous_regions;
+ size_t _promotable_humongous_usage;
+ size_t _regular_regions_promoted_in_place;
+ size_t _regular_usage_promoted_in_place;
+
volatile size_t _soft_max_size;
shenandoah_padding(0);
volatile size_t _committed;
@@ -284,6 +295,8 @@ class ShenandoahHeap : public CollectedHeap {
void heap_region_iterate(ShenandoahHeapRegionClosure* blk) const;
void parallel_heap_region_iterate(ShenandoahHeapRegionClosure* blk) const;
+ inline ShenandoahMmuTracker* const mmu_tracker() { return &_mmu_tracker; };
+
// ---------- GC state machinery
//
// GC state describes the important parts of collector state, that may be
@@ -332,31 +345,8 @@ class ShenandoahHeap : public CollectedHeap {
ShenandoahSharedFlag _progress_last_gc;
ShenandoahSharedFlag _concurrent_strong_root_in_progress;
- // _alloc_supplement_reserve is a supplemental budget for new_memory allocations. During evacuation and update-references,
- // mutator allocation requests are "authorized" iff young_gen->available() plus _alloc_supplement_reserve minus
- // _young_evac_reserve is greater than request size. The values of _alloc_supplement_reserve and _young_evac_reserve
- // are zero except during evacuation and update-reference phases of GC. Both of these values are established at
- // the start of evacuation, and they remain constant throughout the duration of these two phases of GC. Since these
- // two values are constant throughout each GC phases, we introduce a new service into ShenandoahGeneration. This service
- // provides adjusted_available() based on an adjusted capacity. At the start of evacuation, we adjust young capacity by
- // adding the amount to be borrowed from old-gen and subtracting the _young_evac_reserve, we adjust old capacity by
- // subtracting the amount to be loaned to young-gen.
- //
- // We always use adjusted capacities to determine permission to allocate within young and to promote into old. Note
- // that adjusted capacities equal traditional capacities except during evacuation and update refs.
- //
- // During evacuation, we assure that _old_evac_expended does not exceed _old_evac_reserve.
- //
- // At the end of update references, we perform the following bookkeeping activities:
- //
- // 1. Unadjust the capacity within young-gen and old-gen to undo the effects of borrowing memory from old-gen. Note that
- // the entirety of the collection set is now available, so allocation capacity naturally increase at this time.
- // 2. Clear (reset to zero) _alloc_supplement_reserve, _young_evac_reserve, _old_evac_reserve, and _promoted_reserve
- //
- // _young_evac_reserve and _old_evac_reserve are only non-zero during evacuation and update-references.
- //
- // Allocation of old GCLABs assures that _old_evac_expended + request-size < _old_evac_reserved. If the allocation
- // is authorized, increment _old_evac_expended by request size. This allocation ignores old_gen->available().
+ // TODO: Revisit the following comment. It may not accurately represent the true behavior when evacuations fail due to
+ // difficulty finding memory to hold evacuated objects.
//
// Note that the typical total expenditure on evacuation is less than the associated evacuation reserve because we generally
// reserve ShenandoahEvacWaste (> 1.0) times the anticipated evacuation need. In the case that there is an excessive amount
@@ -364,10 +354,15 @@ class ShenandoahHeap : public CollectedHeap {
// effort. If this happens, the requesting thread blocks until some other thread manages to evacuate the offending object.
// Only after "all" threads fail to evacuate an object do we consider the evacuation effort to have failed.
- intptr_t _alloc_supplement_reserve; // Bytes reserved for young allocations during evac and update refs
+ // How many full-gc cycles have been completed?
+ volatile size_t _completed_fullgc_cycles;
+
size_t _promoted_reserve; // Bytes reserved within old-gen to hold the results of promotion
volatile size_t _promoted_expended; // Bytes of old-gen memory expended on promotions
+ // Allocation of old GCLABs (aka PLABs) assures that _old_evac_expended + request-size < _old_evac_reserved. If the allocation
+ // is authorized, increment _old_evac_expended by request size. This allocation ignores old_gen->available().
+
size_t _old_evac_reserve; // Bytes reserved within old-gen to hold evacuated objects from old-gen collection set
volatile size_t _old_evac_expended; // Bytes of old-gen memory expended on old-gen evacuations
@@ -397,7 +392,6 @@ class ShenandoahHeap : public CollectedHeap {
void set_gc_state_mask(uint mask, bool value);
public:
-
char gc_state() const;
static address gc_state_addr();
@@ -415,6 +409,7 @@ class ShenandoahHeap : public CollectedHeap {
void set_prepare_for_old_mark_in_progress(bool cond);
void set_aging_cycle(bool cond);
+
inline bool is_stable() const;
inline bool is_idle() const;
inline bool has_evacuation_reserve_quantities() const;
@@ -441,9 +436,31 @@ class ShenandoahHeap : public CollectedHeap {
inline void set_previous_promotion(size_t promoted_bytes);
inline size_t get_previous_promotion() const;
+ inline void clear_promotion_potential() { _promotion_potential = 0; };
+ inline void set_promotion_potential(size_t val) { _promotion_potential = val; };
+ inline size_t get_promotion_potential() { return _promotion_potential; };
+
+ inline void clear_promotion_in_place_potential() { _promotion_in_place_potential = 0; };
+ inline void set_promotion_in_place_potential(size_t val) { _promotion_in_place_potential = val; };
+ inline size_t get_promotion_in_place_potential() { return _promotion_in_place_potential; };
+
+ inline void set_pad_for_promote_in_place(size_t pad) { _pad_for_promote_in_place = pad; }
+ inline size_t get_pad_for_promote_in_place() { return _pad_for_promote_in_place; }
+
+ inline void reserve_promotable_humongous_regions(size_t region_count) { _promotable_humongous_regions = region_count; }
+ inline void reserve_promotable_humongous_usage(size_t bytes) { _promotable_humongous_usage = bytes; }
+ inline void reserve_promotable_regular_regions(size_t region_count) { _regular_regions_promoted_in_place = region_count; }
+ inline void reserve_promotable_regular_usage(size_t used_bytes) { _regular_usage_promoted_in_place = used_bytes; }
+
+ inline size_t get_promotable_humongous_regions() { return _promotable_humongous_regions; }
+ inline size_t get_promotable_humongous_usage() { return _promotable_humongous_usage; }
+ inline size_t get_regular_regions_promoted_in_place() { return _regular_regions_promoted_in_place; }
+ inline size_t get_regular_usage_promoted_in_place() { return _regular_usage_promoted_in_place; }
+
// Returns previous value
inline size_t set_promoted_reserve(size_t new_val);
inline size_t get_promoted_reserve() const;
+ inline void augment_promo_reserve(size_t increment);
inline void reset_promoted_expended();
inline size_t expend_promoted(size_t increment);
@@ -453,6 +470,7 @@ class ShenandoahHeap : public CollectedHeap {
// Returns previous value
inline size_t set_old_evac_reserve(size_t new_val);
inline size_t get_old_evac_reserve() const;
+ inline void augment_old_evac_reserve(size_t increment);
inline void reset_old_evac_expended();
inline size_t expend_old_evac(size_t increment);
@@ -462,11 +480,6 @@ class ShenandoahHeap : public CollectedHeap {
inline size_t set_young_evac_reserve(size_t new_val);
inline size_t get_young_evac_reserve() const;
- // Returns previous value. This is a signed value because it is the amount borrowed minus the amount reserved for
- // young-gen evacuation. In case we cannot borrow much, this value might be negative.
- inline intptr_t set_alloc_supplement_reserve(intptr_t new_val);
- inline intptr_t get_alloc_supplement_reserve() const;
-
private:
void manage_satb_barrier(bool active);
@@ -517,11 +530,11 @@ class ShenandoahHeap : public CollectedHeap {
void update_heap_references(bool concurrent);
// Final update region states
void update_heap_region_states(bool concurrent);
- void rebuild_free_set(bool concurrent);
void rendezvous_threads();
void recycle_trash();
public:
+ void rebuild_free_set(bool concurrent);
void notify_gc_progress() { _progress_last_gc.set(); }
void notify_gc_no_progress() { _progress_last_gc.unset(); }
@@ -698,6 +711,10 @@ class ShenandoahHeap : public CollectedHeap {
// ---------- Allocation support
//
private:
+ // How many bytes to transfer between old and young after we have finished recycling collection set regions?
+ size_t _old_regions_surplus;
+ size_t _old_regions_deficit;
+
HeapWord* allocate_memory_under_lock(ShenandoahAllocRequest& request, bool& in_new_region, bool is_promotion);
inline HeapWord* allocate_from_gclab(Thread* thread, size_t size);
@@ -731,6 +748,12 @@ class ShenandoahHeap : public CollectedHeap {
void set_young_lab_region_flags();
+ inline void set_old_region_surplus(size_t surplus) { _old_regions_surplus = surplus; };
+ inline void set_old_region_deficit(size_t deficit) { _old_regions_deficit = deficit; };
+
+ inline size_t get_old_region_surplus() { return _old_regions_surplus; };
+ inline size_t get_old_region_deficit() { return _old_regions_deficit; };
+
// ---------- Marking support
//
private:
@@ -830,7 +853,7 @@ class ShenandoahHeap : public CollectedHeap {
void cancel_old_gc();
bool is_old_gc_active();
void coalesce_and_fill_old_regions();
- bool adjust_generation_sizes();
+ void adjust_generation_sizes_for_next_cycle(size_t old_xfer_limit, size_t young_cset_regions, size_t old_cset_regions);
// ---------- Helper functions
//
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp b/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp
index d131d336e78..d6dccf01f9d 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp
@@ -291,11 +291,12 @@ inline HeapWord* ShenandoahHeap::allocate_from_plab(Thread* thread, size_t size,
PLAB* plab = ShenandoahThreadLocalData::plab(thread);
HeapWord* obj;
+
if (plab == nullptr) {
assert(!thread->is_Java_thread() && !thread->is_Worker_thread(), "Performance: thread should have PLAB: %s", thread->name());
// No PLABs in this thread, fallback to shared allocation
return nullptr;
- } else if (is_promotion && (plab->words_remaining() > 0) && !ShenandoahThreadLocalData::allow_plab_promotions(thread)) {
+ } else if (is_promotion && !ShenandoahThreadLocalData::allow_plab_promotions(thread)) {
return nullptr;
}
// if plab->word_size() <= 0, thread's plab not yet initialized for this pass, so allow_plab_promotions() is not trustworthy
@@ -382,7 +383,6 @@ inline oop ShenandoahHeap::try_evacuate_object(oop p, Thread* thread, Shenandoah
break;
}
case OLD_GENERATION: {
-
PLAB* plab = ShenandoahThreadLocalData::plab(thread);
if (plab != nullptr) {
has_plab = true;
@@ -510,6 +510,7 @@ inline oop ShenandoahHeap::try_evacuate_object(oop p, Thread* thread, Shenandoah
// For non-LAB allocations, we have no way to retract the allocation, and
// have to explicitly overwrite the copy with the filler object. With that overwrite,
// we have to keep the fwdptr initialized and pointing to our (stale) copy.
+ assert(size >= ShenandoahHeap::min_fill_size(), "previously allocated object known to be larger than min_size");
fill_with_object(copy, size);
shenandoah_assert_correct(nullptr, copy_val);
// For non-LAB allocations, the object has already been registered
@@ -753,6 +754,14 @@ inline size_t ShenandoahHeap::get_old_evac_reserve() const {
return _old_evac_reserve;
}
+inline void ShenandoahHeap::augment_old_evac_reserve(size_t increment) {
+ _old_evac_reserve += increment;
+}
+
+inline void ShenandoahHeap::augment_promo_reserve(size_t increment) {
+ _promoted_reserve += increment;
+}
+
inline void ShenandoahHeap::reset_old_evac_expended() {
Atomic::store(&_old_evac_expended, (size_t) 0);
}
@@ -791,16 +800,6 @@ inline size_t ShenandoahHeap::get_young_evac_reserve() const {
return _young_evac_reserve;
}
-inline intptr_t ShenandoahHeap::set_alloc_supplement_reserve(intptr_t new_val) {
- intptr_t orig = _alloc_supplement_reserve;
- _alloc_supplement_reserve = new_val;
- return orig;
-}
-
-inline intptr_t ShenandoahHeap::get_alloc_supplement_reserve() const {
- return _alloc_supplement_reserve;
-}
-
template<class T>
inline void ShenandoahHeap::marked_object_iterate(ShenandoahHeapRegion* region, T* cl) {
marked_object_iterate(region, cl, region->top());
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp
index 299c2b1c593..941f4f1ec73 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp
@@ -29,6 +29,7 @@
#include "gc/shared/space.inline.hpp"
#include "gc/shared/tlab_globals.hpp"
#include "gc/shenandoah/shenandoahCardTable.hpp"
+#include "gc/shenandoah/shenandoahFreeSet.hpp"
#include "gc/shenandoah/shenandoahHeapRegionSet.inline.hpp"
#include "gc/shenandoah/shenandoahHeap.inline.hpp"
#include "gc/shenandoah/shenandoahHeapRegion.hpp"
@@ -102,7 +103,7 @@ void ShenandoahHeapRegion::make_regular_allocation(ShenandoahAffiliation affilia
case _empty_uncommitted:
do_commit();
case _empty_committed:
- set_affiliation(affiliation);
+ assert(this->affiliation() == affiliation, "Region affiliation should already be established");
set_state(_regular);
case _regular:
case _pinned:
@@ -122,7 +123,13 @@ void ShenandoahHeapRegion::make_young_maybe() {
case _cset:
case _humongous_start:
case _humongous_cont:
- set_affiliation(YOUNG_GENERATION);
+ if (affiliation() != YOUNG_GENERATION) {
+ if (is_old()) {
+ ShenandoahHeap::heap()->old_generation()->decrement_affiliated_region_count();
+ }
+ set_affiliation(YOUNG_GENERATION);
+ ShenandoahHeap::heap()->young_generation()->increment_affiliated_region_count();
+ }
return;
case _pinned_cset:
case _regular:
@@ -175,6 +182,7 @@ void ShenandoahHeapRegion::make_humongous_start() {
void ShenandoahHeapRegion::make_humongous_start_bypass(ShenandoahAffiliation affiliation) {
shenandoah_assert_heaplocked();
assert (ShenandoahHeap::heap()->is_full_gc_in_progress(), "only for full GC");
+ // Don't bother to account for affiliated regions during Full GC. We recompute totals at end.
set_affiliation(affiliation);
reset_age();
switch (_state) {
@@ -207,6 +215,7 @@ void ShenandoahHeapRegion::make_humongous_cont_bypass(ShenandoahAffiliation affi
shenandoah_assert_heaplocked();
assert (ShenandoahHeap::heap()->is_full_gc_in_progress(), "only for full GC");
set_affiliation(affiliation);
+ // Don't bother to account for affiliated regions during Full GC. We recompute totals at end.
reset_age();
switch (_state) {
case _empty_committed:
@@ -469,6 +478,7 @@ bool ShenandoahHeapRegion::oop_fill_and_coalesce_without_cancel() {
HeapWord* next_marked_obj = marking_context->get_next_marked_addr(obj_addr, t);
assert(next_marked_obj <= t, "next marked object cannot exceed top");
size_t fill_size = next_marked_obj - obj_addr;
+ assert(fill_size >= ShenandoahHeap::min_fill_size(), "previously allocated objects known to be larger than min_size");
ShenandoahHeap::fill_with_object(obj_addr, fill_size);
heap->card_scan()->coalesce_objects(obj_addr, fill_size);
obj_addr = next_marked_obj;
@@ -514,6 +524,7 @@ bool ShenandoahHeapRegion::oop_fill_and_coalesce() {
HeapWord* next_marked_obj = marking_context->get_next_marked_addr(obj_addr, t);
assert(next_marked_obj <= t, "next marked object cannot exceed top");
size_t fill_size = next_marked_obj - obj_addr;
+ assert(fill_size >= ShenandoahHeap::min_fill_size(), "previously allocated object known to be larger than min_size");
ShenandoahHeap::fill_with_object(obj_addr, fill_size);
heap->card_scan()->coalesce_objects(obj_addr, fill_size);
obj_addr = next_marked_obj;
@@ -565,8 +576,8 @@ void ShenandoahHeapRegion::global_oop_iterate_objects_and_fill_dead(OopIterateCl
HeapWord* next_marked_obj = marking_context->get_next_marked_addr(obj_addr, t);
assert(next_marked_obj <= t, "next marked object cannot exceed top");
size_t fill_size = next_marked_obj - obj_addr;
+ assert(fill_size >= ShenandoahHeap::min_fill_size(), "previously allocated objects known to be larger than min_size");
ShenandoahHeap::fill_with_object(obj_addr, fill_size);
-
// coalesce_objects() unregisters all but first object subsumed within coalesced range.
rem_set_scanner->coalesce_objects(obj_addr, fill_size);
obj_addr = next_marked_obj;
@@ -670,8 +681,8 @@ void ShenandoahHeapRegion::recycle() {
set_update_watermark(bottom());
make_empty();
+ ShenandoahHeap::heap()->generation_for(affiliation())->decrement_affiliated_region_count();
set_affiliation(FREE);
-
if (ZapUnusedHeapArea) {
SpaceMangler::mangle_region(MemRegion(bottom(), end()));
}
@@ -957,40 +968,22 @@ void ShenandoahHeapRegion::set_affiliation(ShenandoahAffiliation new_affiliation
}
if (!heap->mode()->is_generational()) {
+ log_trace(gc)("Changing affiliation of region %zu from %s to %s",
+ index(), affiliation_name(), shenandoah_affiliation_name(new_affiliation));
heap->set_affiliation(this, new_affiliation);
return;
}
- log_trace(gc)("Changing affiliation of region %zu from %s to %s",
- index(), shenandoah_affiliation_name(region_affiliation), shenandoah_affiliation_name(new_affiliation));
-
- if (region_affiliation == ShenandoahAffiliation::YOUNG_GENERATION) {
- heap->young_generation()->decrement_affiliated_region_count();
- } else if (region_affiliation == ShenandoahAffiliation::OLD_GENERATION) {
- heap->old_generation()->decrement_affiliated_region_count();
- }
-
- size_t regions;
switch (new_affiliation) {
case FREE:
assert(!has_live(), "Free region should not have live data");
break;
case YOUNG_GENERATION:
reset_age();
- regions = heap->young_generation()->increment_affiliated_region_count();
- // During Full GC, we allow temporary violation of this requirement. We enforce that this condition is
- // restored upon completion of Full GC.
- assert(heap->is_full_gc_in_progress() ||
- (regions * ShenandoahHeapRegion::region_size_bytes() <= heap->young_generation()->adjusted_capacity()),
- "Number of young regions cannot exceed adjusted capacity");
break;
case OLD_GENERATION:
- regions = heap->old_generation()->increment_affiliated_region_count();
- // During Full GC, we allow temporary violation of this requirement. We enforce that this condition is
- // restored upon completion of Full GC.
- assert(heap->is_full_gc_in_progress() ||
- (regions * ShenandoahHeapRegion::region_size_bytes() <= heap->old_generation()->adjusted_capacity()),
- "Number of old regions cannot exceed adjusted capacity");
+ // TODO: should we reset_age() for OLD as well? Examine invocations of set_affiliation(). Some contexts redundantly
+ // invoke reset_age().
break;
default:
ShouldNotReachHere();
@@ -999,8 +992,92 @@ void ShenandoahHeapRegion::set_affiliation(ShenandoahAffiliation new_affiliation
heap->set_affiliation(this, new_affiliation);
}
-// Returns number of regions promoted, or zero if we choose not to promote.
-size_t ShenandoahHeapRegion::promote_humongous() {
+// When we promote a region in place, we can continue to use the established marking context to guide subsequent remembered
+// set scans of this region's content. The region will be coalesced and filled prior to the next old-gen marking effort.
+// We identify the entirety of the region as DIRTY to force the next remembered set scan to identify the "interesting poitners"
+// contained herein.
+void ShenandoahHeapRegion::promote_in_place() {
+ ShenandoahHeap* heap = ShenandoahHeap::heap();
+ ShenandoahMarkingContext* marking_context = heap->marking_context();
+ HeapWord* tams = marking_context->top_at_mark_start(this);
+ assert(heap->active_generation()->is_mark_complete(), "sanity");
+ assert(is_young(), "Only young regions can be promoted");
+ assert(is_regular(), "Use different service to promote humongous regions");
+ assert(age() >= InitialTenuringThreshold, "Only promote regions that are sufficiently aged");
+
+ ShenandoahOldGeneration* old_gen = heap->old_generation();
+ ShenandoahYoungGeneration* young_gen = heap->young_generation();
+ size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+
+ {
+ ShenandoahHeapLocker locker(heap->lock());
+
+ HeapWord* update_watermark = get_update_watermark();
+
+ // Now that this region is affiliated with old, we can allow it to receive allocations, though it may not be in the
+ // is_collector_free range.
+ restore_top_before_promote();
+
+ size_t region_capacity = free();
+ size_t region_used = used();
+
+ // The update_watermark was likely established while we had the artificially high value of top. Make it sane now.
+ assert(update_watermark >= top(), "original top cannot exceed preserved update_watermark");
+ set_update_watermark(top());
+
+ // Unconditionally transfer one region from young to old to represent the newly promoted region.
+ // This expands old and shrinks new by the size of one region. Strictly, we do not "need" to expand old
+ // if there are already enough unaffiliated regions in old to account for this newly promoted region.
+ // However, if we do not transfer the capacities, we end up reducing the amount of memory that would have
+ // otherwise been available to hold old evacuations, because old available is max_capacity - used and now
+ // we would be trading a fully empty region for a partially used region.
+
+ young_gen->decrease_used(region_used);
+ young_gen->decrement_affiliated_region_count();
+
+ // transfer_to_old() increases capacity of old and decreases capacity of young
+ heap->generation_sizer()->force_transfer_to_old(1);
+ set_affiliation(OLD_GENERATION);
+
+ old_gen->increment_affiliated_region_count();
+ old_gen->increase_used(region_used);
+
+ // add_old_collector_free_region() increases promoted_reserve() if available space exceeds PLAB::min_size()
+ heap->free_set()->add_old_collector_free_region(this);
+ }
+
+ assert(top() == tams, "Cannot promote regions in place if top has advanced beyond TAMS");
+
+ // Since this region may have served previously as OLD, it may hold obsolete object range info.
+ heap->card_scan()->reset_object_range(bottom(), end());
+ heap->card_scan()->mark_range_as_dirty(bottom(), top() - bottom());
+
+ // TODO: use an existing coalesce-and-fill function rather than
+ // replicating the code here.
+ HeapWord* obj_addr = bottom();
+ while (obj_addr < tams) {
+ oop obj = cast_to_oop(obj_addr);
+ if (marking_context->is_marked(obj)) {
+ assert(obj->klass() != NULL, "klass should not be NULL");
+ // This thread is responsible for registering all objects in this region. No need for lock.
+ heap->card_scan()->register_object_without_lock(obj_addr);
+ obj_addr += obj->size();
+ } else {
+ HeapWord* next_marked_obj = marking_context->get_next_marked_addr(obj_addr, tams);
+ assert(next_marked_obj <= tams, "next marked object cannot exceed tams");
+ size_t fill_size = next_marked_obj - obj_addr;
+ assert(fill_size >= ShenandoahHeap::min_fill_size(), "previously allocated objects known to be larger than min_size");
+ ShenandoahHeap::fill_with_object(obj_addr, fill_size);
+ heap->card_scan()->register_object_without_lock(obj_addr);
+ obj_addr = next_marked_obj;
+ }
+ }
+
+ // We do not need to scan above TAMS because top equals tams
+ assert(obj_addr == tams, "Expect loop to terminate when obj_addr equals tams");
+}
+
+void ShenandoahHeapRegion::promote_humongous() {
ShenandoahHeap* heap = ShenandoahHeap::heap();
ShenandoahMarkingContext* marking_context = heap->marking_context();
assert(heap->active_generation()->is_mark_complete(), "sanity");
@@ -1020,51 +1097,40 @@ size_t ShenandoahHeapRegion::promote_humongous() {
// it becomes garbage. Better to not make this change until sizes of young-gen and old-gen are completely
// adaptive, as leaving primitive arrays in young-gen might be perceived as an "astonishing result" by someone
// has carefully analyzed the required sizes of an application's young-gen and old-gen.
-
- size_t spanned_regions = ShenandoahHeapRegion::required_regions(obj->size() * HeapWordSize);
+ size_t used_bytes = obj->size() * HeapWordSize;
+ size_t spanned_regions = ShenandoahHeapRegion::required_regions(used_bytes);
+ size_t humongous_waste = spanned_regions * ShenandoahHeapRegion::region_size_bytes() - obj->size() * HeapWordSize;
size_t index_limit = index() + spanned_regions;
-
{
// We need to grab the heap lock in order to avoid a race when changing the affiliations of spanned_regions from
// young to old.
ShenandoahHeapLocker locker(heap->lock());
- size_t available_old_regions = old_generation->adjusted_unaffiliated_regions();
- if (spanned_regions <= available_old_regions) {
- log_debug(gc)("promoting humongous region " SIZE_FORMAT ", spanning " SIZE_FORMAT, index(), spanned_regions);
-
- // For this region and each humongous continuation region spanned by this humongous object, change
- // affiliation to OLD_GENERATION and adjust the generation-use tallies. The remnant of memory
- // in the last humongous region that is not spanned by obj is currently not used.
- for (size_t i = index(); i < index_limit; i++) {
- ShenandoahHeapRegion* r = heap->get_region(i);
- log_debug(gc)("promoting humongous region " SIZE_FORMAT ", from " PTR_FORMAT " to " PTR_FORMAT,
- r->index(), p2i(r->bottom()), p2i(r->top()));
- // We mark the entire humongous object's range as dirty after loop terminates, so no need to dirty the range here
- old_generation->increase_used(r->used());
- young_generation->decrease_used(r->used());
- r->set_affiliation(OLD_GENERATION);
- }
- ShenandoahHeapRegion* tail = heap->get_region(index_limit - 1);
- size_t waste = tail->free();
- if (waste != 0) {
- old_generation->increase_humongous_waste(waste);
- young_generation->decrease_humongous_waste(waste);
- }
- // Then fall through to finish the promotion after releasing the heap lock.
- } else {
- // There are not enough available old regions to promote this humongous region at this time, so defer promotion.
- // TODO: Consider allowing the promotion now, with the expectation that we can resize and/or collect OLD
- // momentarily to address the transient violation of budgets. Some problems that need to be addressed in order
- // to allow transient violation of capacity budgets are:
- // 1. Various size_t subtractions assume usage is less than capacity, and thus assume there will be no
- // arithmetic underflow when we subtract usage from capacity. The results of such size_t subtractions
- // would need to be guarded and special handling provided.
- // 2. ShenandoahVerifier enforces that usage is less than capacity. If we are going to relax this constraint,
- // we need to think about what conditions allow the constraint to be violated and document and implement the
- // changes.
- return 0;
+ // We promote humongous objects unconditionally, without checking for availability. We adjust
+ // usage totals, including humongous waste, after evacuation is done.
+ log_debug(gc)("promoting humongous region " SIZE_FORMAT ", spanning " SIZE_FORMAT, index(), spanned_regions);
+
+ young_generation->decrease_used(used_bytes);
+ young_generation->decrease_humongous_waste(humongous_waste);
+ young_generation->decrease_affiliated_region_count(spanned_regions);
+
+ // transfer_to_old() increases capacity of old and decreases capacity of young
+ heap->generation_sizer()->force_transfer_to_old(spanned_regions);
+
+ // For this region and each humongous continuation region spanned by this humongous object, change
+ // affiliation to OLD_GENERATION and adjust the generation-use tallies. The remnant of memory
+ // in the last humongous region that is not spanned by obj is currently not used.
+ for (size_t i = index(); i < index_limit; i++) {
+ ShenandoahHeapRegion* r = heap->get_region(i);
+ log_debug(gc)("promoting humongous region " SIZE_FORMAT ", from " PTR_FORMAT " to " PTR_FORMAT,
+ r->index(), p2i(r->bottom()), p2i(r->top()));
+ // We mark the entire humongous object's range as dirty after loop terminates, so no need to dirty the range here
+ r->set_affiliation(OLD_GENERATION);
}
+
+ old_generation->increase_affiliated_region_count(spanned_regions);
+ old_generation->increase_used(used_bytes);
+ old_generation->increase_humongous_waste(humongous_waste);
}
// Since this region may have served previously as OLD, it may hold obsolete object range info.
@@ -1082,7 +1148,6 @@ size_t ShenandoahHeapRegion::promote_humongous() {
index(), p2i(bottom()), p2i(bottom() + obj->size()));
heap->card_scan()->mark_range_as_dirty(bottom(), obj->size());
}
- return index_limit - index();
}
void ShenandoahHeapRegion::decrement_humongous_waste() const {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.hpp b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.hpp
index 269f0fd239b..306cff58508 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.hpp
@@ -242,6 +242,8 @@ class ShenandoahHeapRegion {
HeapWord* _new_top;
double _empty_time;
+ HeapWord* _top_before_promoted;
+
// Seldom updated fields
RegionState _state;
HeapWord* _coalesce_and_fill_boundary; // for old regions not selected as collection set candidates.
@@ -350,6 +352,11 @@ class ShenandoahHeapRegion {
return _index;
}
+ inline void save_top_before_promote();
+ inline HeapWord* get_top_before_promote() const { return _top_before_promoted; }
+ inline void restore_top_before_promote();
+ inline size_t garbage_before_padded_for_promote() const;
+
// Allocation (return nullptr if full)
inline HeapWord* allocate_aligned(size_t word_size, ShenandoahAllocRequest &req, size_t alignment_in_words);
@@ -429,6 +436,7 @@ class ShenandoahHeapRegion {
size_t capacity() const { return byte_size(bottom(), end()); }
size_t used() const { return byte_size(bottom(), top()); }
+ size_t used_before_promote() const { return byte_size(bottom(), get_top_before_promote()); }
size_t free() const { return byte_size(top(), end()); }
// Does this region contain this address?
@@ -457,8 +465,9 @@ class ShenandoahHeapRegion {
void decrement_age() { if (_age-- == 0) { _age = 0; } }
void reset_age() { _age = 0; }
- // Sets all remembered set cards to dirty. Returns the number of regions spanned by the associated humongous object.
- size_t promote_humongous();
+ // Register all objects. Set all remembered set cards to dirty.
+ void promote_humongous();
+ void promote_in_place();
private:
void decrement_humongous_waste() const;
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.inline.hpp b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.inline.hpp
index 4ef7f36ab1d..c5823b11ddd 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.inline.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.inline.hpp
@@ -82,6 +82,7 @@ HeapWord* ShenandoahHeapRegion::allocate_aligned(size_t size, ShenandoahAllocReq
if (size >= req.min_size()) {
// Even if req.min_size() is not a multiple of card size, we know that size is.
if (pad_words > 0) {
+ assert(pad_words >= ShenandoahHeap::min_fill_size(), "pad_words expanded above to meet size constraint");
ShenandoahHeap::fill_with_object(orig_top, pad_words);
ShenandoahHeap::heap()->card_scan()->register_object(orig_top);
}
@@ -190,6 +191,17 @@ inline size_t ShenandoahHeapRegion::garbage() const {
return result;
}
+inline size_t ShenandoahHeapRegion::garbage_before_padded_for_promote() const {
+ size_t used_before_promote = byte_size(bottom(), get_top_before_promote());
+ assert(get_top_before_promote() != nullptr, "top before promote should not equal null");
+ assert(used_before_promote >= get_live_data_bytes(),
+ "Live Data must be a subset of used before promotion live: " SIZE_FORMAT " used: " SIZE_FORMAT,
+ get_live_data_bytes(), used_before_promote);
+ size_t result = used_before_promote - get_live_data_bytes();
+ return result;
+
+}
+
inline HeapWord* ShenandoahHeapRegion::get_update_watermark() const {
HeapWord* watermark = Atomic::load_acquire(&_update_watermark);
assert(bottom() <= watermark && watermark <= top(), "within bounds");
@@ -240,4 +252,17 @@ inline bool ShenandoahHeapRegion::is_affiliated() const {
return affiliation() != FREE;
}
+inline void ShenandoahHeapRegion::save_top_before_promote() {
+ _top_before_promoted = _top;
+}
+
+inline void ShenandoahHeapRegion::restore_top_before_promote() {
+ _top = _top_before_promoted;
+#ifdef ASSERT
+ _top_before_promoted = nullptr;
+#endif
+ }
+
+
+
#endif // SHARE_GC_SHENANDOAH_SHENANDOAHHEAPREGION_INLINE_HPP
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahInitLogger.cpp b/src/hotspot/share/gc/shenandoah/shenandoahInitLogger.cpp
index e9268063a56..76cc19fd47d 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahInitLogger.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahInitLogger.cpp
@@ -47,14 +47,14 @@ void ShenandoahInitLogger::print_heap() {
log_info(gc, init)("Heuristics: %s", heap->global_generation()->heuristics()->name());
} else {
log_info(gc, init)("Young Heuristics: %s", heap->young_generation()->heuristics()->name());
- log_info(gc, init)("Young Generation Initial Size: " SIZE_FORMAT "%s",
+ log_info(gc, init)("Young Generation Soft Size: " SIZE_FORMAT "%s",
byte_size_in_proper_unit(heap->young_generation()->soft_max_capacity()),
proper_unit_for_byte_size(heap->young_generation()->soft_max_capacity()));
log_info(gc, init)("Young Generation Max: " SIZE_FORMAT "%s",
byte_size_in_proper_unit(heap->young_generation()->max_capacity()),
proper_unit_for_byte_size(heap->young_generation()->max_capacity()));
log_info(gc, init)("Old Heuristics: %s", heap->old_generation()->heuristics()->name());
- log_info(gc, init)("Old Generation Initial Size: " SIZE_FORMAT "%s",
+ log_info(gc, init)("Old Generation Soft Size: " SIZE_FORMAT "%s",
byte_size_in_proper_unit(heap->old_generation()->soft_max_capacity()),
proper_unit_for_byte_size(heap->old_generation()->soft_max_capacity()));
log_info(gc, init)("Old Generation Max: " SIZE_FORMAT "%s",
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahMmuTracker.cpp b/src/hotspot/share/gc/shenandoah/shenandoahMmuTracker.cpp
index b9443f12923..a7688679f29 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahMmuTracker.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahMmuTracker.cpp
@@ -32,7 +32,6 @@
#include "runtime/os.hpp"
#include "runtime/task.hpp"
-
class ShenandoahMmuTask : public PeriodicTask {
ShenandoahMmuTracker* _mmu_tracker;
public:
@@ -53,66 +52,134 @@ class ThreadTimeAccumulator : public ThreadClosure {
}
};
-double ShenandoahMmuTracker::gc_thread_time_seconds() {
+ShenandoahMmuTracker::ShenandoahMmuTracker() :
+ _most_recent_timestamp(0.0),
+ _most_recent_gc_time(0.0),
+ _most_recent_gcu(0.0),
+ _most_recent_mutator_time(0.0),
+ _most_recent_mu(0.0),
+ _most_recent_periodic_time_stamp(0.0),
+ _most_recent_periodic_gc_time(0.0),
+ _most_recent_periodic_mutator_time(0.0),
+ _mmu_periodic_task(new ShenandoahMmuTask(this)) {
+}
+
+ShenandoahMmuTracker::~ShenandoahMmuTracker() {
+ _mmu_periodic_task->disenroll();
+ delete _mmu_periodic_task;
+}
+
+void ShenandoahMmuTracker::fetch_cpu_times(double &gc_time, double &mutator_time) {
ThreadTimeAccumulator cl;
// We include only the gc threads because those are the only threads
// we are responsible for.
ShenandoahHeap::heap()->gc_threads_do(&cl);
- return double(cl.total_time) / NANOSECS_PER_SEC;
-}
+ double most_recent_gc_thread_time = double(cl.total_time) / NANOSECS_PER_SEC;
+ gc_time = most_recent_gc_thread_time;
-double ShenandoahMmuTracker::process_time_seconds() {
double process_real_time(0.0), process_user_time(0.0), process_system_time(0.0);
bool valid = os::getTimesSecs(&process_real_time, &process_user_time, &process_system_time);
- if (valid) {
- return process_user_time + process_system_time;
+ assert(valid, "don't know why this would not be valid");
+ mutator_time =(process_user_time + process_system_time) - most_recent_gc_thread_time;
+}
+
+void ShenandoahMmuTracker::update_utilization(ShenandoahGeneration* generation, size_t gcid, const char *msg) {
+ double current = os::elapsedTime();
+ _most_recent_gcid = gcid;
+ _most_recent_is_full = false;
+
+ if (gcid == 0) {
+ fetch_cpu_times(_most_recent_gc_time, _most_recent_mutator_time);
+
+ _most_recent_timestamp = current;
+ } else {
+ double gc_cycle_period = current - _most_recent_timestamp;
+ _most_recent_timestamp = current;
+
+ double gc_thread_time, mutator_thread_time;
+ fetch_cpu_times(gc_thread_time, mutator_thread_time);
+ double gc_time = gc_thread_time - _most_recent_gc_time;
+ _most_recent_gc_time = gc_thread_time;
+ _most_recent_gcu = gc_time / (_active_processors * gc_cycle_period);
+ double mutator_time = mutator_thread_time - _most_recent_mutator_time;
+ _most_recent_mutator_time = mutator_thread_time;
+ _most_recent_mu = mutator_time / (_active_processors * gc_cycle_period);
+ log_info(gc, ergo)("At end of %s: GCU: %.1f%%, MU: %.1f%% during period of %.3fs",
+ msg, _most_recent_gcu * 100, _most_recent_mu * 100, gc_cycle_period);
}
- return 0.0;
}
-ShenandoahMmuTracker::ShenandoahMmuTracker() :
- _generational_reference_time_s(0.0),
- _process_reference_time_s(0.0),
- _collector_reference_time_s(0.0),
- _mmu_periodic_task(new ShenandoahMmuTask(this)),
- _mmu_average(10, ShenandoahAdaptiveDecayFactor) {
+void ShenandoahMmuTracker::record_young(ShenandoahGeneration* generation, size_t gcid) {
+ update_utilization(generation, gcid, "Concurrent Young GC");
}
-ShenandoahMmuTracker::~ShenandoahMmuTracker() {
- _mmu_periodic_task->disenroll();
- delete _mmu_periodic_task;
+void ShenandoahMmuTracker::record_bootstrap(ShenandoahGeneration* generation, size_t gcid, bool candidates_for_mixed) {
+ // Not likely that this will represent an "ideal" GCU, but doesn't hurt to try
+ update_utilization(generation, gcid, "Bootstrap Old GC");
+}
+
+void ShenandoahMmuTracker::record_old_marking_increment(ShenandoahGeneration* generation, size_t gcid, bool old_marking_done,
+ bool has_old_candidates) {
+ // No special processing for old marking
+ double now = os::elapsedTime();
+ double duration = now - _most_recent_timestamp;
+
+ double gc_time, mutator_time;
+ fetch_cpu_times(gc_time, mutator_time);
+ double gcu = (gc_time - _most_recent_gc_time) / duration;
+ double mu = (mutator_time - _most_recent_mutator_time) / duration;
+ log_info(gc, ergo)("At end of %s: GCU: %.1f%%, MU: %.1f%% for duration %.3fs (totals to be subsumed in next gc report)",
+ old_marking_done? "last OLD marking increment": "OLD marking increment",
+ gcu * 100, mu * 100, duration);
+}
+
+void ShenandoahMmuTracker::record_mixed(ShenandoahGeneration* generation, size_t gcid, bool is_mixed_done) {
+ update_utilization(generation, gcid, "Mixed Concurrent GC");
+}
+
+void ShenandoahMmuTracker::record_degenerated(ShenandoahGeneration* generation,
+ size_t gcid, bool is_old_bootstrap, bool is_mixed_done) {
+ if ((gcid == _most_recent_gcid) && _most_recent_is_full) {
+ // Do nothing. This is a redundant recording for the full gc that just completed.
+ // TODO: avoid making the call to record_degenerated() in the case that this degenerated upgraded to full gc.
+ } else if (is_old_bootstrap) {
+ update_utilization(generation, gcid, "Degenerated Bootstrap Old GC");
+ } else {
+ update_utilization(generation, gcid, "Degenerated Young GC");
+ }
}
-void ShenandoahMmuTracker::record(ShenandoahGeneration* generation) {
- shenandoah_assert_control_or_vm_thread();
- double collector_time_s = gc_thread_time_seconds();
- double elapsed_gc_time_s = collector_time_s - _generational_reference_time_s;
- generation->add_collection_time(elapsed_gc_time_s);
- _generational_reference_time_s = collector_time_s;
+void ShenandoahMmuTracker::record_full(ShenandoahGeneration* generation, size_t gcid) {
+ update_utilization(generation, gcid, "Full GC");
+ _most_recent_is_full = true;
}
void ShenandoahMmuTracker::report() {
// This is only called by the periodic thread.
- double process_time_s = process_time_seconds();
- double elapsed_process_time_s = process_time_s - _process_reference_time_s;
- if (elapsed_process_time_s <= 0.01) {
- // No cpu time for this interval?
- return;
- }
+ double current = os::elapsedTime();
+ double time_delta = current - _most_recent_periodic_time_stamp;
+ _most_recent_periodic_time_stamp = current;
- _process_reference_time_s = process_time_s;
- double collector_time_s = gc_thread_time_seconds();
- double elapsed_collector_time_s = collector_time_s - _collector_reference_time_s;
- _collector_reference_time_s = collector_time_s;
- double minimum_mutator_utilization = ((elapsed_process_time_s - elapsed_collector_time_s) / elapsed_process_time_s) * 100;
- _mmu_average.add(minimum_mutator_utilization);
- log_info(gc)("Average MMU = %.3f", _mmu_average.davg());
+ double gc_time, mutator_time;
+ fetch_cpu_times(gc_time, mutator_time);
+
+ double gc_delta = gc_time - _most_recent_periodic_gc_time;
+ _most_recent_periodic_gc_time = gc_time;
+
+ double mutator_delta = mutator_time - _most_recent_periodic_mutator_time;
+ _most_recent_periodic_mutator_time = mutator_time;
+
+ double mu = mutator_delta / (_active_processors * time_delta);
+ double gcu = gc_delta / (_active_processors * time_delta);
+ log_info(gc)("Periodic Sample: GCU = %.3f%%, MU = %.3f%% during most recent %.1fs", gcu * 100, mu * 100, time_delta);
}
void ShenandoahMmuTracker::initialize() {
- _process_reference_time_s = process_time_seconds();
- _generational_reference_time_s = gc_thread_time_seconds();
- _collector_reference_time_s = _generational_reference_time_s;
+ // initialize static data
+ _active_processors = os::initial_active_processor_count();
+
+ double _most_recent_periodic_time_stamp = os::elapsedTime();
+ fetch_cpu_times(_most_recent_periodic_gc_time, _most_recent_periodic_mutator_time);
_mmu_periodic_task->enroll();
}
@@ -160,12 +227,12 @@ ShenandoahGenerationSizer::ShenandoahGenerationSizer(ShenandoahMmuTracker* mmu_t
size_t ShenandoahGenerationSizer::calculate_min_young_regions(size_t heap_region_count) {
size_t min_young_regions = (heap_region_count * ShenandoahMinYoungPercentage) / 100;
- return MAX2(uint(min_young_regions), 1U);
+ return MAX2(min_young_regions, (size_t) 1U);
}
size_t ShenandoahGenerationSizer::calculate_max_young_regions(size_t heap_region_count) {
size_t max_young_regions = (heap_region_count * ShenandoahMaxYoungPercentage) / 100;
- return MAX2(uint(max_young_regions), 1U);
+ return MAX2(max_young_regions, (size_t) 1U);
}
void ShenandoahGenerationSizer::recalculate_min_max_young_length(size_t heap_region_count) {
@@ -202,110 +269,71 @@ void ShenandoahGenerationSizer::heap_size_changed(size_t heap_size) {
recalculate_min_max_young_length(heap_size / ShenandoahHeapRegion::region_size_bytes());
}
-bool ShenandoahGenerationSizer::adjust_generation_sizes() const {
- shenandoah_assert_generational();
- if (!use_adaptive_sizing()) {
- return false;
- }
-
- if (_mmu_tracker->average() >= double(GCTimeRatio)) {
- return false;
- }
-
+// Returns true iff transfer is successful
+bool ShenandoahGenerationSizer::transfer_to_old(size_t regions) const {
ShenandoahHeap* heap = ShenandoahHeap::heap();
- ShenandoahOldGeneration *old = heap->old_generation();
- ShenandoahYoungGeneration *young = heap->young_generation();
- ShenandoahGeneration *global = heap->global_generation();
- double old_time_s = old->reset_collection_time();
- double young_time_s = young->reset_collection_time();
- double global_time_s = global->reset_collection_time();
-
- const double transfer_threshold = 3.0;
- double delta = young_time_s - old_time_s;
-
- log_info(gc)("Thread Usr+Sys YOUNG = %.3f, OLD = %.3f, GLOBAL = %.3f", young_time_s, old_time_s, global_time_s);
+ ShenandoahGeneration* old_gen = heap->old_generation();
+ ShenandoahGeneration* young_gen = heap->young_generation();
+ size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+ size_t bytes_to_transfer = regions * region_size_bytes;
- if (abs(delta) <= transfer_threshold) {
- log_info(gc, ergo)("Difference (%.3f) for thread utilization for each generation is under threshold (%.3f)", abs(delta), transfer_threshold);
+ if (young_gen->free_unaffiliated_regions() < regions) {
+ return false;
+ } else if (old_gen->max_capacity() + bytes_to_transfer > heap->max_size_for(old_gen)) {
+ return false;
+ } else if (young_gen->max_capacity() - bytes_to_transfer < heap->min_size_for(young_gen)) {
return false;
- }
-
- if (delta > 0) {
- // young is busier than old, increase size of young to raise MMU
- return transfer_capacity(old, young);
} else {
- // old is busier than young, increase size of old to raise MMU
- return transfer_capacity(young, old);
+ young_gen->decrease_capacity(bytes_to_transfer);
+ old_gen->increase_capacity(bytes_to_transfer);
+ size_t new_size = old_gen->max_capacity();
+ log_info(gc)("Transfer " SIZE_FORMAT " region(s) from %s to %s, yielding increased size: " SIZE_FORMAT "%s",
+ regions, young_gen->name(), old_gen->name(),
+ byte_size_in_proper_unit(new_size), proper_unit_for_byte_size(new_size));
+ return true;
}
}
-bool ShenandoahGenerationSizer::transfer_capacity(ShenandoahGeneration* target) const {
- ShenandoahHeapLocker locker(ShenandoahHeap::heap()->lock());
- if (target->is_young()) {
- return transfer_capacity(ShenandoahHeap::heap()->old_generation(), target);
- } else {
- assert(target->is_old(), "Expected old generation, if not young.");
- return transfer_capacity(ShenandoahHeap::heap()->young_generation(), target);
- }
+// This is used when promoting humongous or highly utilized regular regions in place. It is not required in this situation
+// that the transferred regions be unaffiliated.
+void ShenandoahGenerationSizer::force_transfer_to_old(size_t regions) const {
+ ShenandoahHeap* heap = ShenandoahHeap::heap();
+ ShenandoahGeneration* old_gen = heap->old_generation();
+ ShenandoahGeneration* young_gen = heap->young_generation();
+ size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+ size_t bytes_to_transfer = regions * region_size_bytes;
+
+ young_gen->decrease_capacity(bytes_to_transfer);
+ old_gen->increase_capacity(bytes_to_transfer);
+ size_t new_size = old_gen->max_capacity();
+ log_info(gc)("Forcing transfer of " SIZE_FORMAT " region(s) from %s to %s, yielding increased size: " SIZE_FORMAT "%s",
+ regions, young_gen->name(), old_gen->name(),
+ byte_size_in_proper_unit(new_size), proper_unit_for_byte_size(new_size));
}
-bool ShenandoahGenerationSizer::transfer_capacity(ShenandoahGeneration* from, ShenandoahGeneration* to) const {
- shenandoah_assert_heaplocked_or_safepoint();
- size_t available_regions = from->free_unaffiliated_regions();
- if (available_regions <= 0) {
- log_info(gc)("%s has no regions available for transfer to %s", from->name(), to->name());
- return false;
- }
-
- size_t regions_to_transfer = MAX2(1u, uint(double(available_regions) * _resize_increment));
- if (from->is_young()) {
- regions_to_transfer = adjust_transfer_from_young(from, regions_to_transfer);
- } else {
- regions_to_transfer = adjust_transfer_to_young(to, regions_to_transfer);
- }
+bool ShenandoahGenerationSizer::transfer_to_young(size_t regions) const {
+ ShenandoahHeap* heap = ShenandoahHeap::heap();
+ ShenandoahGeneration* old_gen = heap->old_generation();
+ ShenandoahGeneration* young_gen = heap->young_generation();
+ size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+ size_t bytes_to_transfer = regions * region_size_bytes;
- if (regions_to_transfer == 0) {
- log_info(gc)("No capacity available to transfer from: %s (" SIZE_FORMAT "%s) to: %s (" SIZE_FORMAT "%s)",
- from->name(), byte_size_in_proper_unit(from->max_capacity()), proper_unit_for_byte_size(from->max_capacity()),
- to->name(), byte_size_in_proper_unit(to->max_capacity()), proper_unit_for_byte_size(to->max_capacity()));
+ if (old_gen->free_unaffiliated_regions() < regions) {
return false;
+ } else if (young_gen->max_capacity() + bytes_to_transfer > heap->max_size_for(young_gen)) {
+ return false;
+ } else if (old_gen->max_capacity() - bytes_to_transfer < heap->min_size_for(old_gen)) {
+ return false;
+ } else {
+ old_gen->decrease_capacity(bytes_to_transfer);
+ young_gen->increase_capacity(bytes_to_transfer);
+ size_t new_size = young_gen->max_capacity();
+ log_info(gc)("Transfer " SIZE_FORMAT " region(s) from %s to %s, yielding increased size: " SIZE_FORMAT "%s",
+ regions, old_gen->name(), young_gen->name(),
+ byte_size_in_proper_unit(new_size), proper_unit_for_byte_size(new_size));
+ return true;
}
-
- log_info(gc)("Transfer " SIZE_FORMAT " region(s) from %s to %s", regions_to_transfer, from->name(), to->name());
- from->decrease_capacity(regions_to_transfer * ShenandoahHeapRegion::region_size_bytes());
- to->increase_capacity(regions_to_transfer * ShenandoahHeapRegion::region_size_bytes());
- return true;
-}
-
-size_t ShenandoahGenerationSizer::adjust_transfer_from_young(ShenandoahGeneration* from, size_t regions_to_transfer) const {
- assert(from->is_young(), "Expect to transfer from young");
- size_t young_capacity_regions = from->max_capacity() / ShenandoahHeapRegion::region_size_bytes();
- size_t new_young_regions = young_capacity_regions - regions_to_transfer;
- size_t minimum_young_regions = min_young_regions();
- // Check that we are not going to violate the minimum size constraint.
- if (new_young_regions < minimum_young_regions) {
- assert(minimum_young_regions <= young_capacity_regions, "Young is under minimum capacity.");
- // If the transfer violates the minimum size and there is still some capacity to transfer,
- // adjust the transfer to take the size to the minimum. Note that this may be zero.
- regions_to_transfer = young_capacity_regions - minimum_young_regions;
- }
- return regions_to_transfer;
-}
-
-size_t ShenandoahGenerationSizer::adjust_transfer_to_young(ShenandoahGeneration* to, size_t regions_to_transfer) const {
- assert(to->is_young(), "Can only transfer between young and old.");
- size_t young_capacity_regions = to->max_capacity() / ShenandoahHeapRegion::region_size_bytes();
- size_t new_young_regions = young_capacity_regions + regions_to_transfer;
- size_t maximum_young_regions = max_young_regions();
- // Check that we are not going to violate the maximum size constraint.
- if (new_young_regions > maximum_young_regions) {
- assert(maximum_young_regions >= young_capacity_regions, "Young is over maximum capacity");
- // If the transfer violates the maximum size and there is still some capacity to transfer,
- // adjust the transfer to take the size to the maximum. Note that this may be zero.
- regions_to_transfer = maximum_young_regions - young_capacity_regions;
- }
- return regions_to_transfer;
}
size_t ShenandoahGenerationSizer::min_young_size() const {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahMmuTracker.hpp b/src/hotspot/share/gc/shenandoah/shenandoahMmuTracker.hpp
index 9d3c230a6cc..6ab7d180245 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahMmuTracker.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahMmuTracker.hpp
@@ -50,16 +50,31 @@ class ShenandoahMmuTask;
* MMU.
*/
class ShenandoahMmuTracker {
+private:
+ // These variables hold recent snapshots of cumulative quantities that are used for calculating
+ // CPU time consumed by GC and mutator threads during each GC cycle.
+ double _most_recent_timestamp;
+ double _most_recent_gc_time;
+ double _most_recent_gcu;
+ double _most_recent_mutator_time;
+ double _most_recent_mu;
+
+ // These variables hold recent snapshots of cumulative quantities that are used for reporting
+ // periodic consumption of CPU time by GC and mutator threads.
+ double _most_recent_periodic_time_stamp;
+ double _most_recent_periodic_gc_time;
+ double _most_recent_periodic_mutator_time;
- double _generational_reference_time_s;
- double _process_reference_time_s;
- double _collector_reference_time_s;
+ size_t _most_recent_gcid;
+ uint _active_processors;
+
+ bool _most_recent_is_full;
ShenandoahMmuTask* _mmu_periodic_task;
TruncatedSeq _mmu_average;
- static double gc_thread_time_seconds();
- static double process_time_seconds();
+ void update_utilization(ShenandoahGeneration* generation, size_t gcid, const char* msg);
+ static void fetch_cpu_times(double &gc_time, double &mutator_time);
public:
explicit ShenandoahMmuTracker();
@@ -68,22 +83,24 @@ class ShenandoahMmuTracker {
// This enrolls the periodic task after everything is initialized.
void initialize();
- // This is called at the start and end of a GC cycle. The GC thread times
- // will be accumulated in this generation. Note that the bootstrap cycle
- // for an old collection should be counted against the old generation.
- // When the collector is idle, it still runs a regulator and a control.
- // The times for these threads are attributed to the global generation.
- void record(ShenandoahGeneration* generation);
+ // At completion of each GC cycle (not including interrupted cycles), we invoke one of the following to record the
+ // GC utilization during this cycle. Incremental efforts spent in an interrupted GC cycle will be accumulated into
+ // the CPU time reports for the subsequent completed [degenerated or full] GC cycle.
+ //
+ // We may redundantly record degen and full in the case that a degen upgrades to full. When this happens, we will invoke
+ // both record_full() and record_degenerated() with the same value of gcid. record_full() is called first and the log
+ // reports such a cycle as a FULL cycle.
+ void record_young(ShenandoahGeneration* generation, size_t gcid);
+ void record_bootstrap(ShenandoahGeneration* generation, size_t gcid, bool has_old_candidates);
+ void record_old_marking_increment(ShenandoahGeneration* generation, size_t gcid, bool old_marking_done, bool has_old_candidates);
+ void record_mixed(ShenandoahGeneration* generation, size_t gcid, bool is_mixed_done);
+ void record_full(ShenandoahGeneration* generation, size_t gcid);
+ void record_degenerated(ShenandoahGeneration* generation, size_t gcid, bool is_old_boostrap, bool is_mixed_done);
// This is called by the periodic task timer. The interval is defined by
// GCPauseIntervalMillis and defaults to 5 seconds. This method computes
// the MMU over the elapsed interval and records it in a running average.
- // This method also logs the average MMU.
void report();
-
- double average() {
- return _mmu_average.davg();
- }
};
class ShenandoahGenerationSizer {
@@ -114,14 +131,6 @@ class ShenandoahGenerationSizer {
// given the number of heap regions depending on the kind of sizing algorithm.
void recalculate_min_max_young_length(size_t heap_region_count);
- // These two methods are responsible for enforcing the minimum and maximum
- // constraints for the size of the generations.
- size_t adjust_transfer_from_young(ShenandoahGeneration* from, size_t regions_to_transfer) const;
- size_t adjust_transfer_to_young(ShenandoahGeneration* to, size_t regions_to_transfer) const;
-
- // This will attempt to transfer capacity from one generation to the other. It
- // returns true if a transfer is made, false otherwise.
- bool transfer_capacity(ShenandoahGeneration* from, ShenandoahGeneration* to) const;
public:
explicit ShenandoahGenerationSizer(ShenandoahMmuTracker* mmu_tracker);
@@ -145,19 +154,11 @@ class ShenandoahGenerationSizer {
return _use_adaptive_sizing;
}
- // This is invoked at the end of a collection. This happens on a safepoint
- // to avoid any races with allocators (and to avoid interfering with
- // allocators by taking the heap lock). The amount of capacity to move
- // from one generation to another is controlled by YoungGenerationSizeIncrement
- // and defaults to 20% of the available capacity of the donor generation.
- // The minimum and maximum sizes of the young generation are controlled by
- // ShenandoahMinYoungPercentage and ShenandoahMaxYoungPercentage, respectively.
- // The method returns true when an adjustment is made, false otherwise.
- bool adjust_generation_sizes() const;
-
- // This may be invoked by a heuristic (from regulator thread) before it
- // decides to run a collection.
- bool transfer_capacity(ShenandoahGeneration* target) const;
+ bool transfer_to_young(size_t regions) const;
+ bool transfer_to_old(size_t regions) const;
+
+ // force transfer is used when we promote humongous objects. May violate min/max limits on generation sizes
+ void force_transfer_to_old(size_t regions) const;
};
#endif //SHARE_GC_SHENANDOAH_SHENANDOAHMMUTRACKER_HPP
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahOldGC.cpp b/src/hotspot/share/gc/shenandoah/shenandoahOldGC.cpp
index 94e29209bac..da56f2cb683 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahOldGC.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahOldGC.cpp
@@ -30,6 +30,7 @@
#include "gc/shenandoah/shenandoahOldGC.hpp"
#include "gc/shenandoah/shenandoahOopClosures.inline.hpp"
#include "gc/shenandoah/shenandoahGeneration.hpp"
+#include "gc/shenandoah/shenandoahOldGeneration.hpp"
#include "gc/shenandoah/shenandoahYoungGeneration.hpp"
#include "prims/jvmtiTagMap.hpp"
#include "utilities/events.hpp"
@@ -146,5 +147,48 @@ bool ShenandoahOldGC::collect(GCCause::Cause cause) {
// collection.
vmop_entry_final_roots();
+ // We do not rebuild_free following increments of old marking because memory has not been reclaimed.. However, we may
+ // need to transfer memory to OLD in order to efficiently support the mixed evacuations that might immediately follow.
+ size_t allocation_runway = heap->young_generation()->heuristics()->bytes_of_allocation_runway_before_gc_trigger(0);
+ heap->adjust_generation_sizes_for_next_cycle(allocation_runway, 0, 0);
+
+ bool success;
+ size_t region_xfer;
+ const char* region_destination;
+ ShenandoahYoungGeneration* young_gen = heap->young_generation();
+ ShenandoahGeneration* old_gen = heap->old_generation();
+ {
+ ShenandoahHeapLocker locker(heap->lock());
+
+ size_t old_region_surplus = heap->get_old_region_surplus();
+ size_t old_region_deficit = heap->get_old_region_deficit();
+ if (old_region_surplus) {
+ success = heap->generation_sizer()->transfer_to_young(old_region_surplus);
+ region_destination = "young";
+ region_xfer = old_region_surplus;
+ } else if (old_region_deficit) {
+ success = heap->generation_sizer()->transfer_to_old(old_region_deficit);
+ region_destination = "old";
+ region_xfer = old_region_deficit;
+ if (!success) {
+ ((ShenandoahOldHeuristics *) old_gen->heuristics())->trigger_cannot_expand();
+ }
+ } else {
+ region_destination = "none";
+ region_xfer = 0;
+ success = true;
+ }
+ heap->set_old_region_surplus(0);
+ heap->set_old_region_deficit(0);
+ }
+
+ // Report outside the heap lock
+ size_t young_available = young_gen->available();
+ size_t old_available = old_gen->available();
+ log_info(gc, ergo)("After old marking finished, %s " SIZE_FORMAT " regions to %s to prepare for next gc, old available: "
+ SIZE_FORMAT "%s, young_available: " SIZE_FORMAT "%s",
+ success? "successfully transferred": "failed to transfer", region_xfer, region_destination,
+ byte_size_in_proper_unit(old_available), proper_unit_for_byte_size(old_available),
+ byte_size_in_proper_unit(young_available), proper_unit_for_byte_size(young_available));
return true;
}
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.cpp b/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.cpp
index 331499816f8..f8aaf325cbd 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.cpp
@@ -174,12 +174,30 @@ class ShenandoahConcurrentCoalesceAndFillTask : public WorkerTask {
ShenandoahOldGeneration::ShenandoahOldGeneration(uint max_queues, size_t max_capacity, size_t soft_max_capacity)
: ShenandoahGeneration(OLD, max_queues, max_capacity, soft_max_capacity),
_coalesce_and_fill_region_array(NEW_C_HEAP_ARRAY(ShenandoahHeapRegion*, ShenandoahHeap::heap()->num_regions(), mtGC)),
- _state(IDLE)
+ _state(IDLE),
+ _growth_before_compaction(INITIAL_GROWTH_BEFORE_COMPACTION)
{
+ _live_bytes_after_last_mark = ShenandoahHeap::heap()->capacity() * INITIAL_LIVE_FRACTION / FRACTIONAL_DENOMINATOR;
// Always clear references for old generation
ref_processor()->set_soft_reference_policy(true);
}
+size_t ShenandoahOldGeneration::get_live_bytes_after_last_mark() const {
+ return _live_bytes_after_last_mark;
+}
+
+void ShenandoahOldGeneration::set_live_bytes_after_last_mark(size_t bytes) {
+ _live_bytes_after_last_mark = bytes;
+ if (_growth_before_compaction > MINIMUM_GROWTH_BEFORE_COMPACTION) {
+ _growth_before_compaction /= 2;
+ }
+}
+
+size_t ShenandoahOldGeneration::usage_trigger_threshold() const {
+ size_t result = _live_bytes_after_last_mark + (_live_bytes_after_last_mark * _growth_before_compaction) / FRACTIONAL_DENOMINATOR;
+ return result;
+}
+
bool ShenandoahOldGeneration::contains(ShenandoahHeapRegion* region) const {
// TODO: Should this be region->is_old() instead?
return !region->is_young();
@@ -255,6 +273,7 @@ bool ShenandoahOldGeneration::coalesce_and_fill() {
uint nworkers = workers->active_workers();
log_debug(gc)("Starting (or resuming) coalesce-and-fill of old heap regions");
+
// This code will see the same set of regions to fill on each resumption as it did
// on the initial run. That's okay because each region keeps track of its own coalesce
// and fill state. Regions that were filled on a prior attempt will not try to fill again.
@@ -322,7 +341,11 @@ void ShenandoahOldGeneration::prepare_regions_and_collection_set(bool concurrent
ShenandoahPhaseTimings::final_rebuild_freeset :
ShenandoahPhaseTimings::degen_gc_final_rebuild_freeset);
ShenandoahHeapLocker locker(heap->lock());
- heap->free_set()->rebuild();
+ size_t cset_young_regions, cset_old_regions;
+ heap->free_set()->prepare_to_rebuild(cset_young_regions, cset_old_regions);
+ // This is just old-gen completion. No future budgeting required here. The only reason to rebuild the freeset here
+ // is in case there was any immediate old garbage identified.
+ heap->free_set()->rebuild(cset_young_regions, cset_old_regions);
}
}
@@ -405,7 +428,8 @@ void ShenandoahOldGeneration::validate_transition(State new_state) {
case IDLE:
// GC cancellation can send us back to IDLE from any state.
assert(!heap->is_concurrent_old_mark_in_progress(), "Cannot become idle during old mark.");
- assert(_old_heuristics->unprocessed_old_collection_candidates() == 0, "Cannot become idle with collection candidates");
+ assert(!heap->mode()->is_generational() ||
+ (_old_heuristics->unprocessed_old_collection_candidates() == 0), "Cannot become idle with collection candidates");
assert(!heap->is_prepare_for_old_mark_in_progress(), "Cannot become idle while making old generation parseable.");
assert(heap->young_generation()->old_gen_task_queues() == nullptr, "Cannot become idle when setup for bootstrapping.");
break;
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.hpp b/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.hpp
index 0e3fb429c26..1ac7e22c41a 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.hpp
@@ -90,19 +90,45 @@ class ShenandoahOldGeneration : public ShenandoahGeneration {
private:
State _state;
+ static const size_t FRACTIONAL_DENOMINATOR = 64536;
+
+ // During initialization of the JVM, we search for the correct old-gen size by initally performing old-gen
+ // collection when old-gen usage is 50% more (INITIAL_GROWTH_BEFORE_COMPACTION) than the initial old-gen size
+ // estimate (3.125% of heap). The next old-gen trigger occurs when old-gen grows 25% larger than its live
+ // memory at the end of the first old-gen collection. Then we trigger again when old-gen growns 12.5%
+ // more than its live memory at the end of the previous old-gen collection. Thereafter, we trigger each time
+ // old-gen grows more than 12.5% following the end of its previous old-gen collection.
+ static const size_t INITIAL_GROWTH_BEFORE_COMPACTION = FRACTIONAL_DENOMINATOR / 2; // 50.0%
+ static const size_t MINIMUM_GROWTH_BEFORE_COMPACTION = FRACTIONAL_DENOMINATOR / 8; // 12.5%
+
+ // INITIAL_LIVE_FRACTION represents the initial guess of how large old-gen should be. We estimate that old-gen
+ // needs to consume 3.125% of the total heap size. And we "pretend" that we start out with this amount of live
+ // old-gen memory. The first old-collection trigger will occur when old-gen occupies 50% more than this initial
+ // approximation of the old-gen memory requirement, in other words when old-gen usage is 150% of 3.125%, which
+ // is 4.6875% of the total heap size.
+ static const uint16_t INITIAL_LIVE_FRACTION = FRACTIONAL_DENOMINATOR / 32; // 3.125%
+ size_t _live_bytes_after_last_mark;
+ size_t _growth_before_compaction; // How much growth in usage before we trigger old collection, per 65_536
+
+ void validate_transition(State new_state) NOT_DEBUG_RETURN;
+
public:
State state() const {
return _state;
}
+ void transition_to(State new_state);
+
+ size_t get_live_bytes_after_last_mark() const;
+ void set_live_bytes_after_last_mark(size_t new_live);
+
+ size_t usage_trigger_threshold() const;
+
bool can_start_gc() {
return _state == IDLE || _state == WAITING_FOR_FILL;
}
static const char* state_name(State state);
-
- void transition_to(State new_state);
- void validate_transition(State new_state) NOT_DEBUG_RETURN;
};
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahRegulatorThread.cpp b/src/hotspot/share/gc/shenandoah/shenandoahRegulatorThread.cpp
index 7664b75d1d6..7d21fb27972 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahRegulatorThread.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahRegulatorThread.cpp
@@ -140,7 +140,8 @@ void ShenandoahRegulatorThread::regulator_sleep() {
}
bool ShenandoahRegulatorThread::start_old_cycle() {
- return _old_heuristics->should_start_gc() && _control_thread->request_concurrent_gc(OLD);
+ return !ShenandoahHeap::heap()->doing_mixed_evacuations() && !ShenandoahHeap::heap()->collection_set()->has_old_regions() &&
+ _old_heuristics->should_start_gc() && _control_thread->request_concurrent_gc(OLD);
}
bool ShenandoahRegulatorThread::start_young_cycle() {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp b/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp
index dae7ab21b2d..61ea96999d0 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp
@@ -403,24 +403,32 @@ class ShenandoahGenerationStatsClosure : public ShenandoahHeapRegionClosure {
byte_size_in_proper_unit(stats.used()), proper_unit_for_byte_size(stats.used()));
}
- static void validate_usage(const char* label, ShenandoahGeneration* generation, ShenandoahCalculateRegionStatsClosure& stats) {
+ static void validate_usage(const bool adjust_for_padding,
+ const char* label, ShenandoahGeneration* generation, ShenandoahCalculateRegionStatsClosure& stats) {
+ ShenandoahHeap* heap = ShenandoahHeap::heap();
size_t generation_used = generation->used();
+ size_t generation_used_regions = generation->used_regions();
+ if (adjust_for_padding && (generation->is_young() || generation->is_global())) {
+ size_t pad = ShenandoahHeap::heap()->get_pad_for_promote_in_place();
+ generation_used += pad;
+ }
+
guarantee(stats.used() == generation_used,
"%s: generation (%s) used size must be consistent: generation-used: " SIZE_FORMAT "%s, regions-used: " SIZE_FORMAT "%s",
label, generation->name(),
byte_size_in_proper_unit(generation_used), proper_unit_for_byte_size(generation_used),
byte_size_in_proper_unit(stats.used()), proper_unit_for_byte_size(stats.used()));
- guarantee(stats.regions() == generation->used_regions(),
+ guarantee(stats.regions() == generation_used_regions,
"%s: generation (%s) used regions (" SIZE_FORMAT ") must equal regions that are in use (" SIZE_FORMAT ")",
label, generation->name(), generation->used_regions(), stats.regions());
-// This check is disabled because of known issues with this feature. We expect this code to be updated by 05/2023.
-// size_t capacity = generation->adjusted_capacity();
-// guarantee(stats.span() <= capacity,
-// "%s: generation (%s) size spanned by regions (" SIZE_FORMAT ") must not exceed current capacity (" SIZE_FORMAT "%s)",
-// label, generation->name(), stats.regions(),
-// byte_size_in_proper_unit(capacity), proper_unit_for_byte_size(capacity));
+ size_t generation_capacity = generation->max_capacity();
+ size_t humongous_regions_promoted = 0;
+ guarantee(stats.span() <= generation_capacity,
+ "%s: generation (%s) size spanned by regions (" SIZE_FORMAT ") must not exceed current capacity (" SIZE_FORMAT "%s)",
+ label, generation->name(), stats.regions(),
+ byte_size_in_proper_unit(generation_capacity), proper_unit_for_byte_size(generation_capacity));
size_t humongous_waste = generation->get_humongous_waste();
guarantee(stats.waste() == humongous_waste,
@@ -743,6 +751,7 @@ void ShenandoahVerifier::verify_at_safepoint(const char* label,
VerifyForwarded forwarded, VerifyMarked marked,
VerifyCollectionSet cset,
VerifyLiveness liveness, VerifyRegions regions,
+ VerifySize sizeness,
VerifyGCState gcstate) {
guarantee(ShenandoahSafepoint::is_at_shenandoah_safepoint(), "only when nothing else happens");
guarantee(ShenandoahVerify, "only when enabled, and bitmap is initialized in ShenandoahHeap::initialize");
@@ -795,6 +804,7 @@ void ShenandoahVerifier::verify_at_safepoint(const char* label,
if (enabled) {
char actual = _heap->gc_state();
+ // Old generation marking is allowed in all states.
if (!VerifyThreadGCState::verify_gc_state(actual, expected)) {
fatal("%s: Global gc-state: expected %d, actual %d", label, expected, actual);
}
@@ -813,13 +823,20 @@ void ShenandoahVerifier::verify_at_safepoint(const char* label,
ShenandoahCalculateRegionStatsClosure cl;
_heap->heap_region_iterate(&cl);
- size_t heap_used = _heap->used();
- guarantee(cl.used() == heap_used,
- "%s: heap used size must be consistent: heap-used = " SIZE_FORMAT "%s, regions-used = " SIZE_FORMAT "%s",
- label,
- byte_size_in_proper_unit(heap_used), proper_unit_for_byte_size(heap_used),
- byte_size_in_proper_unit(cl.used()), proper_unit_for_byte_size(cl.used()));
-
+ size_t heap_used;
+ if (_heap->mode()->is_generational() && (sizeness == _verify_size_adjusted_for_padding)) {
+ // Prior to evacuation, regular regions that are to be evacuated in place are padded to prevent further allocations
+ heap_used = _heap->used() + _heap->get_pad_for_promote_in_place();
+ } else if (sizeness != _verify_size_disable) {
+ heap_used = _heap->used();
+ }
+ if (sizeness != _verify_size_disable) {
+ guarantee(cl.used() == heap_used,
+ "%s: heap used size must be consistent: heap-used = " SIZE_FORMAT "%s, regions-used = " SIZE_FORMAT "%s",
+ label,
+ byte_size_in_proper_unit(heap_used), proper_unit_for_byte_size(heap_used),
+ byte_size_in_proper_unit(cl.used()), proper_unit_for_byte_size(cl.used()));
+ }
size_t heap_committed = _heap->committed();
guarantee(cl.committed() == heap_committed,
"%s: heap committed size must be consistent: heap-committed = " SIZE_FORMAT "%s, regions-committed = " SIZE_FORMAT "%s",
@@ -868,10 +885,16 @@ void ShenandoahVerifier::verify_at_safepoint(const char* label,
ShenandoahGenerationStatsClosure::log_usage(_heap->young_generation(), cl.young);
ShenandoahGenerationStatsClosure::log_usage(_heap->global_generation(), cl.global);
}
-
- ShenandoahGenerationStatsClosure::validate_usage(label, _heap->old_generation(), cl.old);
- ShenandoahGenerationStatsClosure::validate_usage(label, _heap->young_generation(), cl.young);
- ShenandoahGenerationStatsClosure::validate_usage(label, _heap->global_generation(), cl.global);
+ if (sizeness == _verify_size_adjusted_for_padding) {
+ ShenandoahGenerationStatsClosure::validate_usage(false, label, _heap->old_generation(), cl.old);
+ ShenandoahGenerationStatsClosure::validate_usage(true, label, _heap->young_generation(), cl.young);
+ ShenandoahGenerationStatsClosure::validate_usage(true, label, _heap->global_generation(), cl.global);
+ } else if (sizeness == _verify_size_exact) {
+ ShenandoahGenerationStatsClosure::validate_usage(false, label, _heap->old_generation(), cl.old);
+ ShenandoahGenerationStatsClosure::validate_usage(false, label, _heap->young_generation(), cl.young);
+ ShenandoahGenerationStatsClosure::validate_usage(false, label, _heap->global_generation(), cl.global);
+ }
+ // else: sizeness must equal _verify_size_disable
}
log_debug(gc)("Safepoint verification finished remembered set verification");
@@ -983,19 +1006,22 @@ void ShenandoahVerifier::verify_generic(VerifyOption vo) {
_verify_cset_disable, // cset may be inconsistent
_verify_liveness_disable, // no reliable liveness data
_verify_regions_disable, // no reliable region data
+ _verify_size_exact, // expect generation and heap sizes to match exactly
_verify_gcstate_disable // no data about gcstate
);
}
void ShenandoahVerifier::verify_before_concmark() {
verify_at_safepoint(
- "Before Mark",
- _verify_remembered_before_marking, // verify read-only remembered set from bottom() to top()
+ "Before Mark",
+ _verify_remembered_before_marking,
+ // verify read-only remembered set from bottom() to top()
_verify_forwarded_none, // UR should have fixed up
_verify_marked_disable, // do not verify marked: lots ot time wasted checking dead allocations
_verify_cset_none, // UR should have fixed this
_verify_liveness_disable, // no reliable liveness data
_verify_regions_notrash, // no trash regions
+ _verify_size_exact, // expect generation and heap sizes to match exactly
_verify_gcstate_stable // there are no forwarded objects
);
}
@@ -1005,10 +1031,12 @@ void ShenandoahVerifier::verify_after_concmark() {
"After Mark",
_verify_remembered_disable, // do not verify remembered set
_verify_forwarded_none, // no forwarded references
- _verify_marked_complete_except_references, // bitmaps as precise as we can get, except dangling j.l.r.Refs
+ _verify_marked_complete_except_references,
+ // bitmaps as precise as we can get, except dangling j.l.r.Refs
_verify_cset_none, // no references to cset anymore
_verify_liveness_complete, // liveness data must be complete here
_verify_regions_disable, // trash regions not yet recycled
+ _verify_size_exact, // expect generation and heap sizes to match exactly
_verify_gcstate_stable_weakroots // heap is still stable, weakroots are in progress
);
}
@@ -1022,6 +1050,8 @@ void ShenandoahVerifier::verify_before_evacuation() {
_verify_cset_disable, // non-forwarded references to cset expected
_verify_liveness_complete, // liveness data must be complete here
_verify_regions_disable, // trash regions not yet recycled
+ _verify_size_adjusted_for_padding, // expect generation and heap sizes to match after adjustments
+ // for promote in place padding
_verify_gcstate_stable_weakroots // heap is still stable, weakroots are in progress
);
}
@@ -1035,6 +1065,7 @@ void ShenandoahVerifier::verify_during_evacuation() {
_verify_cset_disable, // some cset references are not forwarded yet
_verify_liveness_disable, // liveness data might be already stale after pre-evacs
_verify_regions_disable, // trash regions not yet recycled
+ _verify_size_disable, // we don't know how much of promote-in-place work has been completed
_verify_gcstate_evacuation // evacuation is in progress
);
}
@@ -1048,6 +1079,7 @@ void ShenandoahVerifier::verify_after_evacuation() {
_verify_cset_forwarded, // all cset refs are fully forwarded
_verify_liveness_disable, // no reliable liveness data anymore
_verify_regions_notrash, // trash regions have been recycled already
+ _verify_size_exact, // expect generation and heap sizes to match exactly
_verify_gcstate_forwarded // evacuation produced some forwarded objects
);
}
@@ -1056,15 +1088,17 @@ void ShenandoahVerifier::verify_before_updaterefs() {
verify_at_safepoint(
"Before Updating References",
_verify_remembered_before_updating_references, // verify read-write remembered set
- _verify_forwarded_allow, // forwarded references allowed
- _verify_marked_complete, // bitmaps might be stale, but alloc-after-mark should be well
- _verify_cset_forwarded, // all cset refs are fully forwarded
- _verify_liveness_disable, // no reliable liveness data anymore
- _verify_regions_notrash, // trash regions have been recycled already
- _verify_gcstate_updating // evacuation should have produced some forwarded objects
+ _verify_forwarded_allow, // forwarded references allowed
+ _verify_marked_complete, // bitmaps might be stale, but alloc-after-mark should be well
+ _verify_cset_forwarded, // all cset refs are fully forwarded
+ _verify_liveness_disable, // no reliable liveness data anymore
+ _verify_regions_notrash, // trash regions have been recycled already
+ _verify_size_exact, // expect generation and heap sizes to match exactly
+ _verify_gcstate_updating // evacuation should have produced some forwarded objects
);
}
+// We have not yet cleanup (reclaimed) the collection set
void ShenandoahVerifier::verify_after_updaterefs() {
verify_at_safepoint(
"After Updating References",
@@ -1074,6 +1108,7 @@ void ShenandoahVerifier::verify_after_updaterefs() {
_verify_cset_none, // no cset references, all updated
_verify_liveness_disable, // no reliable liveness data anymore
_verify_regions_nocset, // no cset regions, trash regions have appeared
+ _verify_size_exact, // expect generation and heap sizes to match exactly
_verify_gcstate_stable // update refs had cleaned up forwarded objects
);
}
@@ -1087,6 +1122,7 @@ void ShenandoahVerifier::verify_after_degenerated() {
_verify_cset_none, // no cset references
_verify_liveness_disable, // no reliable liveness data anymore
_verify_regions_notrash_nocset, // no trash, no cset
+ _verify_size_exact, // expect generation and heap sizes to match exactly
_verify_gcstate_stable // degenerated refs had cleaned up forwarded objects
);
}
@@ -1100,6 +1136,7 @@ void ShenandoahVerifier::verify_before_fullgc() {
_verify_cset_disable, // cset might be foobared
_verify_liveness_disable, // no reliable liveness data anymore
_verify_regions_disable, // no reliable region data here
+ _verify_size_disable, // if we degenerate during evacuation, usage not valid: padding and deferred accounting
_verify_gcstate_disable // no reliable gcstate data
);
}
@@ -1113,6 +1150,7 @@ void ShenandoahVerifier::verify_after_fullgc() {
_verify_cset_none, // no cset references
_verify_liveness_disable, // no reliable liveness data anymore
_verify_regions_notrash_nocset, // no trash, no cset
+ _verify_size_exact, // expect generation and heap sizes to match exactly
_verify_gcstate_stable // full gc cleaned up everything
);
}
@@ -1314,7 +1352,7 @@ void ShenandoahVerifier::verify_rem_set_before_mark() {
}
// else, this humongous object is not marked so no need to verify its internal pointers
if (!scanner->verify_registration(obj_addr, ctx)) {
- ShenandoahAsserts::print_failure(ShenandoahAsserts::_safe_all, obj, obj_addr, nullptr,
+ ShenandoahAsserts::print_failure(ShenandoahAsserts::_safe_all, obj, nullptr, nullptr,
"Verify init-mark remembered set violation", "object not properly registered", __FILE__, __LINE__);
}
} else if (!r->is_humongous()) {
@@ -1330,7 +1368,7 @@ void ShenandoahVerifier::verify_rem_set_before_mark() {
}
// else, object's start is marked dirty and obj is not an objArray, so any interesting pointers are covered
if (!scanner->verify_registration(obj_addr, ctx)) {
- ShenandoahAsserts::print_failure(ShenandoahAsserts::_safe_all, obj, obj_addr, nullptr,
+ ShenandoahAsserts::print_failure(ShenandoahAsserts::_safe_all, obj, nullptr, nullptr,
"Verify init-mark remembered set violation", "object not properly registered", __FILE__, __LINE__);
}
obj_addr += obj->size();
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahVerifier.hpp b/src/hotspot/share/gc/shenandoah/shenandoahVerifier.hpp
index 284746be541..67b35644bf5 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahVerifier.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahVerifier.hpp
@@ -141,6 +141,17 @@ class ShenandoahVerifier : public CHeapObj<mtGC> {
_verify_regions_notrash_nocset
} VerifyRegions;
+ typedef enum {
+ // Disable size verification
+ _verify_size_disable,
+
+ // Enforce exact consistency
+ _verify_size_exact,
+
+ // Expect promote-in-place adjustments: padding inserted to temporarily prevent further allocation in regular regions
+ _verify_size_adjusted_for_padding
+ } VerifySize;
+
typedef enum {
// Disable gc-state verification
_verify_gcstate_disable,
@@ -189,6 +200,7 @@ class ShenandoahVerifier : public CHeapObj<mtGC> {
VerifyCollectionSet cset,
VerifyLiveness liveness,
VerifyRegions regions,
+ VerifySize sizeness,
VerifyGCState gcstate);
public:
diff --git a/src/hotspot/share/gc/shenandoah/shenandoah_globals.hpp b/src/hotspot/share/gc/shenandoah/shenandoah_globals.hpp
index a7f3311b603..4d5d029f4f7 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoah_globals.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoah_globals.hpp
@@ -97,7 +97,7 @@
"collector accepts. In percents of heap region size.") \
range(0,100) \
\
- product(uintx, ShenandoahOldGarbageThreshold, 10, EXPERIMENTAL, \
+ product(uintx, ShenandoahOldGarbageThreshold, 15, EXPERIMENTAL, \
"How much garbage an old region has to contain before it would " \
"be taken for collection.") \
range(0,100) \
@@ -128,13 +128,6 @@
"size.") \
range(0,100) \
\
- product(uintx, ShenandoahOldMinFreeThreshold, 5, EXPERIMENTAL, \
- "Percentage of free old generation heap memory below which most " \
- "heuristics trigger collection independent of other triggers. " \
- "Provides a safety margin for many heuristics. In percents of " \
- "(soft) max heap size.") \
- range(0,100) \
- \
product(uintx, ShenandoahAllocationThreshold, 0, EXPERIMENTAL, \
"How many new allocations should happen since the last GC cycle " \
"before some heuristics trigger the collection. In percents of " \
@@ -210,7 +203,7 @@
"Heuristics may trigger collections more frequently. Time is in " \
"milliseconds. Setting this to 0 disables the feature.") \
\
- product(uintx, ShenandoahGuaranteedYoungGCInterval, 5*60*1000, EXPERIMENTAL, \
+ product(uintx, ShenandoahGuaranteedYoungGCInterval, 5*60*1000, EXPERIMENTAL, \
"Run a collection of the young generation at least this often. " \
"Heuristics may trigger collections more frequently. Time is in " \
"milliseconds. Setting this to 0 disables the feature.") \
@@ -300,17 +293,20 @@
"failures, which will trigger stop-the-world Full GC passes.") \
range(1.0,100.0) \
\
- product(double, ShenandoahGenerationalEvacWaste, 2.0, EXPERIMENTAL, \
- "For generational mode, how much waste evacuations produce " \
- "within the reserved space. Larger values make evacuations " \
- "more resilient against evacuation conflicts, at expense of " \
- "evacuating less on each GC cycle. Smaller values increase " \
- "the risk of evacuation failures, which will trigger " \
- "stop-the-world Full GC passes. The default value for " \
- "generational mode is 2.0. The reason for the higher default " \
- "value in generational mode is because generational mode " \
- "enforces the evacuation budget, triggering degenerated GC " \
- "which upgrades to full GC whenever the budget is exceeded.") \
+ product(double, ShenandoahOldEvacWaste, 1.4, EXPERIMENTAL, \
+ "How much waste evacuations produce within the reserved space. " \
+ "Larger values make evacuations more resilient against " \
+ "evacuation conflicts, at expense of evacuating less on each " \
+ "GC cycle. Smaller values increase the risk of evacuation " \
+ "failures, which will trigger stop-the-world Full GC passes.") \
+ range(1.0,100.0) \
+ \
+ product(double, ShenandoahPromoEvacWaste, 1.2, EXPERIMENTAL, \
+ "How much waste promotions produce within the reserved space. " \
+ "Larger values make evacuations more resilient against " \
+ "evacuation conflicts, at expense of promoting less on each " \
+ "GC cycle. Smaller values increase the risk of evacuation " \
+ "failures, which will trigger stop-the-world Full GC passes.") \
range(1.0,100.0) \
\
product(uintx, ShenandoahMaxEvacLABRatio, 0, EXPERIMENTAL, \
@@ -339,26 +335,16 @@
"reserve/waste is incorrect, at the risk that application " \
"runs out of memory too early.") \
\
- product(uintx, ShenandoahOldEvacReserve, 2, EXPERIMENTAL, \
- "How much of old-generation heap to reserve for old-generation " \
- "evacuations. Larger values allow GC to evacuate more live " \
- "old-generation objects on every cycle, while potentially " \
- "creating greater impact on the cadence at which the young- " \
- "generation allocation pool is replenished. During mixed " \
- "evacuations, the bound on amount of old-generation heap " \
- "regions included in the collecdtion set is the smaller " \
- "of the quantities specified by this parameter and the " \
- "size of ShenandoahEvacReserve as adjusted by the value of " \
- "ShenandoahOldEvacRatioPercent. In percents of total " \
- "old-generation heap size.") \
- range(1,100) \
- \
- product(uintx, ShenandoahOldEvacRatioPercent, 12, EXPERIMENTAL, \
+ product(uintx, ShenandoahOldEvacRatioPercent, 75, EXPERIMENTAL, \
"The maximum proportion of evacuation from old-gen memory, as " \
- "a percent ratio. The default value 12 denotes that no more " \
- "than one eighth (12%) of the collection set evacuation " \
- "workload may be comprised of old-gen heap regions. A larger " \
- "value allows a smaller number of mixed evacuations to process " \
+ "a percent ratio. The default value 75 denotes that no more " \
+ "than 75% of the collection set evacuation " \
+ "workload may be evacuate to old-gen heap regions. This limits " \
+ "both the promotion of aged regions and the compaction of " \
+ "existing old regions. A value of 75 denotes that the normal " \
+ "young-gen evacuation is increased by up to four fold. " \
+ "A larger value allows quicker promotion and allows" \
+ "a smaller number of mixed evacuations to process " \
"the entire list of old-gen collection candidates at the cost " \
"of an increased disruption of the normal cadence of young-gen " \
"collections. A value of 100 allows a mixed evacuation to " \
@@ -378,7 +364,7 @@
"to be less than this.") \
range(0, 100) \
\
- product(uintx, ShenandoahMaxYoungPercentage, 80, EXPERIMENTAL, \
+ product(uintx, ShenandoahMaxYoungPercentage, 100, EXPERIMENTAL, \
"The maximum percentage of the heap to use for the young " \
"generation. Heuristics will not adjust the young generation " \
"to be more than this.") \
@@ -431,10 +417,14 @@
"When running in passive mode, this can be toggled to measure " \
"either Degenerated GC or Full GC costs.") \
\
- product(uintx, ShenandoahFullGCThreshold, 3, EXPERIMENTAL, \
+ product(uintx, ShenandoahFullGCThreshold, 64, EXPERIMENTAL, \
"How many back-to-back Degenerated GCs should happen before " \
"going to a Full GC.") \
\
+ product(uintx, ShenandoahOOMGCRetries, 3, EXPERIMENTAL, \
+ "How many GCs should happen before we throw OutOfMemoryException "\
+ "for allocation request, including at least one Full GC.") \
+ \
product(bool, ShenandoahImplicitGCInvokesConcurrent, false, EXPERIMENTAL, \
"Should internally-caused GC requests invoke concurrent cycles, " \
"should they do the stop-the-world (Degenerated / Full GC)? " \
@@ -518,20 +508,6 @@
"Fix references with load reference barrier. Disabling this " \
"might degrade performance.") \
\
- product(uintx, ShenandoahBorrowPercent, 30, EXPERIMENTAL, \
- "During evacuation and reference updating in generational " \
- "mode, new allocations are allowed to borrow from old-gen " \
- "memory up to ShenandoahBorrowPercent / 100 amount of the " \
- "young-generation content of the current collection set. " \
- "Any memory borrowed from old-gen during evacuation and " \
- "update-references phases of GC will be repaid from the " \
- "abundance of young-gen memory produced when the collection " \
- "set is recycled at the end of updating references. The " \
- "default value of 30 reserves 70% of the to-be-reclaimed " \
- "young collection set memory to be allocated during the " \
- "subsequent concurrent mark phase of GC.") \
- range(0, 100) \
- \
product(uintx, ShenandoahOldCompactionReserve, 8, EXPERIMENTAL, \
"During generational GC, prevent promotions from filling " \
"this number of heap regions. These regions are reserved " \
diff --git a/test/hotspot/jtreg/ProblemList.txt b/test/hotspot/jtreg/ProblemList.txt
index 48495d2a1ea..14c04bfbd66 100644
--- a/test/hotspot/jtreg/ProblemList.txt
+++ b/test/hotspot/jtreg/ProblemList.txt
@@ -83,11 +83,6 @@ gc/stress/gclocker/TestGCLockerWithG1.java 8180622 generic-all
gc/stress/TestJNIBlockFullGC/TestJNIBlockFullGC.java 8192647 generic-all
gc/stress/TestStressG1Humongous.java 8286554 windows-x64
-gc/shenandoah/oom/TestThreadFailure.java 8306335 generic-all
-gc/shenandoah/oom/TestClassLoaderLeak.java 8306336 generic-all
-gc/stress/gclocker/TestGCLockerWithShenandoah.java#generational 8306341 generic-all
-gc/TestAllocHumongousFragment.java#generational 8306342 generic-all
-
#############################################################################
# :hotspot_runtime