--- a/src/hotspot/share/gc/shared/oopStorage.cpp Thu May 03 14:13:20 2018 -0700
+++ b/src/hotspot/share/gc/shared/oopStorage.cpp Thu May 03 17:36:50 2018 -0400
@@ -28,20 +28,22 @@
#include "logging/log.hpp"
#include "logging/logStream.hpp"
#include "memory/allocation.inline.hpp"
-#include "memory/resourceArea.hpp"
#include "runtime/atomic.hpp"
+#include "runtime/globals.hpp"
#include "runtime/handles.inline.hpp"
#include "runtime/mutex.hpp"
#include "runtime/mutexLocker.hpp"
#include "runtime/orderAccess.inline.hpp"
#include "runtime/safepoint.hpp"
#include "runtime/stubRoutines.hpp"
+#include "runtime/thread.hpp"
#include "utilities/align.hpp"
#include "utilities/count_trailing_zeros.hpp"
#include "utilities/debug.hpp"
#include "utilities/globalDefinitions.hpp"
#include "utilities/macros.hpp"
#include "utilities/ostream.hpp"
+#include "utilities/spinYield.hpp"
OopStorage::BlockEntry::BlockEntry() : _prev(NULL), _next(NULL) {}
@@ -108,6 +110,90 @@
}
}
+OopStorage::BlockArray::BlockArray(size_t size) :
+ _size(size),
+ _block_count(0),
+ _refcount(0)
+{}
+
+OopStorage::BlockArray::~BlockArray() {
+ assert(_refcount == 0, "precondition");
+}
+
+OopStorage::BlockArray* OopStorage::BlockArray::create(size_t size, AllocFailType alloc_fail) {
+ size_t size_in_bytes = blocks_offset() + sizeof(Block*) * size;
+ void* mem = NEW_C_HEAP_ARRAY3(char, size_in_bytes, mtGC, CURRENT_PC, alloc_fail);
+ if (mem == NULL) return NULL;
+ return new (mem) BlockArray(size);
+}
+
+void OopStorage::BlockArray::destroy(BlockArray* ba) {
+ ba->~BlockArray();
+ FREE_C_HEAP_ARRAY(char, ba);
+}
+
+size_t OopStorage::BlockArray::size() const {
+ return _size;
+}
+
+size_t OopStorage::BlockArray::block_count() const {
+ return _block_count;
+}
+
+size_t OopStorage::BlockArray::block_count_acquire() const {
+ return OrderAccess::load_acquire(&_block_count);
+}
+
+void OopStorage::BlockArray::increment_refcount() const {
+ int new_value = Atomic::add(1, &_refcount);
+ assert(new_value >= 1, "negative refcount %d", new_value - 1);
+}
+
+bool OopStorage::BlockArray::decrement_refcount() const {
+ int new_value = Atomic::sub(1, &_refcount);
+ assert(new_value >= 0, "negative refcount %d", new_value);
+ return new_value == 0;
+}
+
+bool OopStorage::BlockArray::push(Block* block) {
+ size_t index = _block_count;
+ if (index < _size) {
+ block->set_active_index(index);
+ *block_ptr(index) = block;
+ // Use a release_store to ensure all the setup is complete before
+ // making the block visible.
+ OrderAccess::release_store(&_block_count, index + 1);
+ return true;
+ } else {
+ return false;
+ }
+}
+
+void OopStorage::BlockArray::remove(Block* block) {
+ assert(_block_count > 0, "array is empty");
+ size_t index = block->active_index();
+ assert(*block_ptr(index) == block, "block not present");
+ size_t last_index = _block_count - 1;
+ Block* last_block = *block_ptr(last_index);
+ last_block->set_active_index(index);
+ *block_ptr(index) = last_block;
+ _block_count = last_index;
+}
+
+void OopStorage::BlockArray::copy_from(const BlockArray* from) {
+ assert(_block_count == 0, "array must be empty");
+ size_t count = from->_block_count;
+ assert(count <= _size, "precondition");
+ Block* const* from_ptr = from->block_ptr(0);
+ Block** to_ptr = block_ptr(0);
+ for (size_t i = 0; i < count; ++i) {
+ Block* block = *from_ptr++;
+ assert(block->active_index() == i, "invariant");
+ *to_ptr++ = block;
+ }
+ _block_count = count;
+}
+
// Blocks start with an array of BitsPerWord oop entries. That array
// is divided into conceptual BytesPerWord sections of BitsPerByte
// entries. Blocks are allocated aligned on section boundaries, for
@@ -125,7 +211,7 @@
_allocated_bitmask(0),
_owner(owner),
_memory(memory),
- _active_entry(),
+ _active_index(0),
_allocate_entry(),
_deferred_updates_next(NULL),
_release_refcount(0)
@@ -146,10 +232,6 @@
const_cast<OopStorage* volatile&>(_owner) = NULL;
}
-const OopStorage::BlockEntry& OopStorage::Block::get_active_entry(const Block& block) {
- return block._active_entry;
-}
-
const OopStorage::BlockEntry& OopStorage::Block::get_allocate_entry(const Block& block) {
return block._allocate_entry;
}
@@ -204,6 +286,20 @@
return (base <= ptr) && (ptr < (base + ARRAY_SIZE(_data)));
}
+size_t OopStorage::Block::active_index() const {
+ return _active_index;
+}
+
+void OopStorage::Block::set_active_index(size_t index) {
+ _active_index = index;
+}
+
+size_t OopStorage::Block::active_index_safe(const Block* block) {
+ STATIC_ASSERT(sizeof(intptr_t) == sizeof(block->_active_index));
+ assert(CanUseSafeFetchN(), "precondition");
+ return SafeFetchN((intptr_t*)&block->_active_index, 0);
+}
+
unsigned OopStorage::Block::get_index(const oop* ptr) const {
assert(contains(ptr), PTR_FORMAT " not in block " PTR_FORMAT, p2i(ptr), p2i(this));
return static_cast<unsigned>(ptr - get_pointer(0));
@@ -246,7 +342,7 @@
// This can return a false positive if ptr is not contained by some
// block. For some uses, it is a precondition that ptr is valid,
-// e.g. contained in some block in owner's _active_list. Other uses
+// e.g. contained in some block in owner's _active_array. Other uses
// require additional validation of the result.
OopStorage::Block*
OopStorage::Block::block_for_ptr(const OopStorage* owner, const oop* ptr) {
@@ -280,12 +376,12 @@
// Allocation involves the _allocate_list, which contains a subset of the
// blocks owned by a storage object. This is a doubly-linked list, linked
// through dedicated fields in the blocks. Full blocks are removed from this
-// list, though they are still present in the _active_list. Empty blocks are
+// list, though they are still present in the _active_array. Empty blocks are
// kept at the end of the _allocate_list, to make it easy for empty block
// deletion to find them.
//
// allocate(), and delete_empty_blocks_concurrent() lock the
-// _allocate_mutex while performing any list modifications.
+// _allocate_mutex while performing any list and array modifications.
//
// allocate() and release() update a block's _allocated_bitmask using CAS
// loops. This prevents loss of updates even though release() performs
@@ -299,7 +395,7 @@
//
// release() is performed lock-free. release() first looks up the block for
// the entry, using address alignment to find the enclosing block (thereby
-// avoiding iteration over the _active_list). Once the block has been
+// avoiding iteration over the _active_array). Once the block has been
// determined, its _allocated_bitmask needs to be updated, and its position in
// the _allocate_list may need to be updated. There are two cases:
//
@@ -340,7 +436,7 @@
// Failed to make new block, no other thread made a block
// available while the mutex was released, and didn't get
// one from a deferred update either, so return failure.
- log_info(oopstorage, ref)("%s: failed allocation", name());
+ log_info(oopstorage, ref)("%s: failed block allocation", name());
return NULL;
}
}
@@ -348,17 +444,21 @@
// Add new block to storage.
log_info(oopstorage, blocks)("%s: new block " PTR_FORMAT, name(), p2i(block));
+ // Add new block to the _active_array, growing if needed.
+ if (!_active_array->push(block)) {
+ if (expand_active_array()) {
+ guarantee(_active_array->push(block), "push failed after expansion");
+ } else {
+ log_info(oopstorage, blocks)("%s: failed active array expand", name());
+ Block::delete_block(*block);
+ return NULL;
+ }
+ }
// Add to end of _allocate_list. The mutex release allowed
// other threads to add blocks to the _allocate_list. We prefer
// to allocate from non-empty blocks, to allow empty blocks to
// be deleted.
_allocate_list.push_back(*block);
- // Add to front of _active_list, and then record as the head
- // block, for concurrent iteration protocol.
- _active_list.push_front(*block);
- ++_block_count;
- // Ensure all setup of block is complete before making it visible.
- OrderAccess::release_store(&_active_head, block);
}
block = _allocate_list.head();
}
@@ -383,6 +483,123 @@
return result;
}
+// Create a new, larger, active array with the same content as the
+// current array, and then replace, relinquishing the old array.
+// Return true if the array was successfully expanded, false to
+// indicate allocation failure.
+bool OopStorage::expand_active_array() {
+ assert_lock_strong(_allocate_mutex);
+ BlockArray* old_array = _active_array;
+ size_t new_size = 2 * old_array->size();
+ log_info(oopstorage, blocks)("%s: expand active array " SIZE_FORMAT,
+ name(), new_size);
+ BlockArray* new_array = BlockArray::create(new_size, AllocFailStrategy::RETURN_NULL);
+ if (new_array == NULL) return false;
+ new_array->copy_from(old_array);
+ replace_active_array(new_array);
+ relinquish_block_array(old_array);
+ return true;
+}
+
+OopStorage::ProtectActive::ProtectActive() : _enter(0), _exit() {}
+
+// Begin read-side critical section.
+uint OopStorage::ProtectActive::read_enter() {
+ return Atomic::add(2u, &_enter);
+}
+
+// End read-side critical section.
+void OopStorage::ProtectActive::read_exit(uint enter_value) {
+ Atomic::add(2u, &_exit[enter_value & 1]);
+}
+
+// Wait until all readers that entered the critical section before
+// synchronization have exited that critical section.
+void OopStorage::ProtectActive::write_synchronize() {
+ SpinYield spinner;
+ // Determine old and new exit counters, based on bit0 of the
+ // on-entry _enter counter.
+ uint value = OrderAccess::load_acquire(&_enter);
+ volatile uint* new_ptr = &_exit[(value + 1) & 1];
+ // Atomically change the in-use exit counter to the new counter, by
+ // adding 1 to the _enter counter (flipping bit0 between 0 and 1)
+ // and initializing the new exit counter to that enter value. Note:
+ // The new exit counter is not being used by read operations until
+ // this change succeeds.
+ uint old;
+ do {
+ old = value;
+ *new_ptr = ++value;
+ value = Atomic::cmpxchg(value, &_enter, old);
+ } while (old != value);
+ // Readers that entered the critical section before we changed the
+ // selected exit counter will use the old exit counter. Readers
+ // entering after the change will use the new exit counter. Wait
+ // for all the critical sections started before the change to
+ // complete, e.g. for the value of old_ptr to catch up with old.
+ volatile uint* old_ptr = &_exit[old & 1];
+ while (old != OrderAccess::load_acquire(old_ptr)) {
+ spinner.wait();
+ }
+}
+
+// Make new_array the _active_array. Increments new_array's refcount
+// to account for the new reference. The assignment is atomic wrto
+// obtain_active_array; once this function returns, it is safe for the
+// caller to relinquish the old array.
+void OopStorage::replace_active_array(BlockArray* new_array) {
+ // Caller has the old array that is the current value of _active_array.
+ // Update new_array refcount to account for the new reference.
+ new_array->increment_refcount();
+ // Install new_array, ensuring its initialization is complete first.
+ OrderAccess::release_store(&_active_array, new_array);
+ // Wait for any readers that could read the old array from _active_array.
+ _protect_active.write_synchronize();
+ // All obtain critical sections that could see the old array have
+ // completed, having incremented the refcount of the old array. The
+ // caller can now safely relinquish the old array.
+}
+
+// Atomically (wrto replace_active_array) get the active array and
+// increment its refcount. This provides safe access to the array,
+// even if an allocate operation expands and replaces the value of
+// _active_array. The caller must relinquish the array when done
+// using it.
+OopStorage::BlockArray* OopStorage::obtain_active_array() const {
+ uint enter_value = _protect_active.read_enter();
+ BlockArray* result = OrderAccess::load_acquire(&_active_array);
+ result->increment_refcount();
+ _protect_active.read_exit(enter_value);
+ return result;
+}
+
+// Decrement refcount of array and destroy if refcount is zero.
+void OopStorage::relinquish_block_array(BlockArray* array) const {
+ if (array->decrement_refcount()) {
+ assert(array != _active_array, "invariant");
+ BlockArray::destroy(array);
+ }
+}
+
+class OopStorage::WithActiveArray : public StackObj {
+ const OopStorage* _storage;
+ BlockArray* _active_array;
+
+public:
+ WithActiveArray(const OopStorage* storage) :
+ _storage(storage),
+ _active_array(storage->obtain_active_array())
+ {}
+
+ ~WithActiveArray() {
+ _storage->relinquish_block_array(_active_array);
+ }
+
+ BlockArray& active_array() const {
+ return *_active_array;
+ }
+};
+
OopStorage::Block* OopStorage::find_block_or_null(const oop* ptr) const {
assert(ptr != NULL, "precondition");
return Block::block_for_ptr(this, ptr);
@@ -392,7 +609,6 @@
uintx old_allocated,
const OopStorage* owner,
const void* block) {
- ResourceMark rm;
Log(oopstorage, blocks) log;
LogStream ls(log.debug());
if (is_full_bitmask(old_allocated)) {
@@ -546,20 +762,21 @@
return dup;
}
+const size_t initial_active_array_size = 8;
+
OopStorage::OopStorage(const char* name,
Mutex* allocate_mutex,
Mutex* active_mutex) :
_name(dup_name(name)),
- _active_list(&Block::get_active_entry),
+ _active_array(BlockArray::create(initial_active_array_size)),
_allocate_list(&Block::get_allocate_entry),
- _active_head(NULL),
_deferred_updates(NULL),
_allocate_mutex(allocate_mutex),
_active_mutex(active_mutex),
_allocation_count(0),
- _block_count(0),
_concurrent_iteration_active(false)
{
+ _active_array->increment_refcount();
assert(_active_mutex->rank() < _allocate_mutex->rank(),
"%s: active_mutex must have lower rank than allocate_mutex", _name);
assert(_active_mutex->_safepoint_check_required != Mutex::_safepoint_check_always,
@@ -583,10 +800,13 @@
while ((block = _allocate_list.head()) != NULL) {
_allocate_list.unlink(*block);
}
- while ((block = _active_list.head()) != NULL) {
- _active_list.unlink(*block);
+ bool unreferenced = _active_array->decrement_refcount();
+ assert(unreferenced, "deleting storage while _active_array is referenced");
+ for (size_t i = _active_array->block_count(); 0 < i; ) {
+ block = _active_array->at(--i);
Block::delete_block(*block);
}
+ BlockArray::destroy(_active_array);
FREE_C_HEAP_ARRAY(char, _name);
}
@@ -598,16 +818,13 @@
// Don't interfere with a concurrent iteration.
if (_concurrent_iteration_active) return;
// Delete empty (and otherwise deletable) blocks from end of _allocate_list.
- for (const Block* block = _allocate_list.ctail();
+ for (Block* block = _allocate_list.tail();
(block != NULL) && block->is_deletable();
- block = _allocate_list.ctail()) {
- _active_list.unlink(*block);
+ block = _allocate_list.tail()) {
+ _active_array->remove(block);
_allocate_list.unlink(*block);
delete_empty_block(*block);
- --_block_count;
}
- // Update _active_head, in case current value was in deleted set.
- _active_head = _active_list.head();
}
void OopStorage::delete_empty_blocks_concurrent() {
@@ -616,14 +833,14 @@
// release the mutex across the block deletions. Set an upper bound
// on how many blocks we'll try to release, so other threads can't
// cause an unbounded stay in this function.
- size_t limit = _block_count;
+ size_t limit = block_count();
for (size_t i = 0; i < limit; ++i) {
// Additional updates might become available while we dropped the
// lock. But limit number processed to limit lock duration.
reduce_deferred_updates();
- const Block* block = _allocate_list.ctail();
+ Block* block = _allocate_list.tail();
if ((block == NULL) || !block->is_deletable()) {
// No block to delete, so done. There could be more pending
// deferred updates that could give us more work to do; deal with
@@ -635,12 +852,7 @@
MutexLockerEx aml(_active_mutex, Mutex::_no_safepoint_check_flag);
// Don't interfere with a concurrent iteration.
if (_concurrent_iteration_active) return;
- // Remove block from _active_list, updating head if needed.
- _active_list.unlink(*block);
- --_block_count;
- if (block == _active_head) {
- _active_head = _active_list.head();
- }
+ _active_array->remove(block);
}
// Remove block from _allocate_list and delete it.
_allocate_list.unlink(*block);
@@ -653,18 +865,17 @@
OopStorage::EntryStatus OopStorage::allocation_status(const oop* ptr) const {
const Block* block = find_block_or_null(ptr);
if (block != NULL) {
- // Verify block is a real block. For now, simple linear search.
- // Do something more clever if this is a performance bottleneck.
+ // Prevent block deletion and _active_array modification.
MutexLockerEx ml(_allocate_mutex, Mutex::_no_safepoint_check_flag);
- for (const Block* check_block = _active_list.chead();
- check_block != NULL;
- check_block = _active_list.next(*check_block)) {
- if (check_block == block) {
- if ((block->allocated_bitmask() & block->bitmask_for_entry(ptr)) != 0) {
- return ALLOCATED_ENTRY;
- } else {
- return UNALLOCATED_ENTRY;
- }
+ // Block could be a false positive, so get index carefully.
+ size_t index = Block::active_index_safe(block);
+ if ((index < _active_array->block_count()) &&
+ (block == _active_array->at(index)) &&
+ block->contains(ptr)) {
+ if ((block->allocated_bitmask() & block->bitmask_for_entry(ptr)) != 0) {
+ return ALLOCATED_ENTRY;
+ } else {
+ return UNALLOCATED_ENTRY;
}
}
}
@@ -676,30 +887,50 @@
}
size_t OopStorage::block_count() const {
- return _block_count;
+ WithActiveArray wab(this);
+ // Count access is racy, but don't care.
+ return wab.active_array().block_count();
}
size_t OopStorage::total_memory_usage() const {
size_t total_size = sizeof(OopStorage);
total_size += strlen(name()) + 1;
- total_size += block_count() * Block::allocation_size();
+ total_size += sizeof(BlockArray);
+ WithActiveArray wab(this);
+ const BlockArray& blocks = wab.active_array();
+ // Count access is racy, but don't care.
+ total_size += blocks.block_count() * Block::allocation_size();
+ total_size += blocks.size() * sizeof(Block*);
return total_size;
}
// Parallel iteration support
-static char* not_started_marker_dummy = NULL;
-static void* const not_started_marker = ¬_started_marker_dummy;
+uint OopStorage::BasicParState::default_estimated_thread_count(bool concurrent) {
+ return concurrent ? ConcGCThreads : ParallelGCThreads;
+}
-OopStorage::BasicParState::BasicParState(OopStorage* storage, bool concurrent) :
+OopStorage::BasicParState::BasicParState(const OopStorage* storage,
+ uint estimated_thread_count,
+ bool concurrent) :
_storage(storage),
- _next_block(not_started_marker),
+ _active_array(_storage->obtain_active_array()),
+ _block_count(0), // initialized properly below
+ _next_block(0),
+ _estimated_thread_count(estimated_thread_count),
_concurrent(concurrent)
{
+ assert(estimated_thread_count > 0, "estimated thread count must be positive");
update_iteration_state(true);
+ // Get the block count *after* iteration state updated, so concurrent
+ // empty block deletion is suppressed and can't reduce the count. But
+ // ensure the count we use was written after the block with that count
+ // was fully initialized; see BlockArray::push.
+ _block_count = _active_array->block_count_acquire();
}
OopStorage::BasicParState::~BasicParState() {
+ _storage->relinquish_block_array(_active_array);
update_iteration_state(false);
}
@@ -711,29 +942,49 @@
}
}
-void OopStorage::BasicParState::ensure_iteration_started() {
- if (!_concurrent) {
- assert_at_safepoint();
+bool OopStorage::BasicParState::claim_next_segment(IterationData* data) {
+ data->_processed += data->_segment_end - data->_segment_start;
+ size_t start = OrderAccess::load_acquire(&_next_block);
+ if (start >= _block_count) {
+ return finish_iteration(data); // No more blocks available.
}
- assert(!_concurrent || _storage->_concurrent_iteration_active, "invariant");
- // Ensure _next_block is not the not_started_marker, setting it to
- // the _active_head to start the iteration if necessary.
- if (OrderAccess::load_acquire(&_next_block) == not_started_marker) {
- Atomic::cmpxchg(_storage->_active_head, &_next_block, not_started_marker);
+ // Try to claim several at a time, but not *too* many. We want to
+ // avoid deciding there are many available and selecting a large
+ // quantity, get delayed, and then end up claiming most or all of
+ // the remaining largish amount of work, leaving nothing for other
+ // threads to do. But too small a step can lead to contention
+ // over _next_block, esp. when the work per block is small.
+ size_t max_step = 10;
+ size_t remaining = _block_count - start;
+ size_t step = MIN2(max_step, 1 + (remaining / _estimated_thread_count));
+ // Atomic::add with possible overshoot. This can perform better
+ // than a CAS loop on some platforms when there is contention.
+ // We can cope with the uncertainty by recomputing start/end from
+ // the result of the add, and dealing with potential overshoot.
+ size_t end = Atomic::add(step, &_next_block);
+ // _next_block may have changed, so recompute start from result of add.
+ start = end - step;
+ // _next_block may have changed so much that end has overshot.
+ end = MIN2(end, _block_count);
+ // _next_block may have changed so much that even start has overshot.
+ if (start < _block_count) {
+ // Record claimed segment for iteration.
+ data->_segment_start = start;
+ data->_segment_end = end;
+ return true; // Success.
+ } else {
+ // No more blocks to claim.
+ return finish_iteration(data);
}
- assert(_next_block != not_started_marker, "postcondition");
}
-OopStorage::Block* OopStorage::BasicParState::claim_next_block() {
- assert(_next_block != not_started_marker, "Iteration not started");
- void* next = _next_block;
- while (next != NULL) {
- void* new_next = _storage->_active_list.next(*static_cast<Block*>(next));
- void* fetched = Atomic::cmpxchg(new_next, &_next_block, next);
- if (fetched == next) break; // Claimed.
- next = fetched;
- }
- return static_cast<Block*>(next);
+bool OopStorage::BasicParState::finish_iteration(const IterationData* data) const {
+ log_debug(oopstorage, blocks, stats)
+ ("Parallel iteration on %s: blocks = " SIZE_FORMAT
+ ", processed = " SIZE_FORMAT " (%2.f%%)",
+ _storage->name(), _block_count, data->_processed,
+ percent_of(data->_processed, _block_count));
+ return false;
}
const char* OopStorage::name() const { return _name; }
@@ -742,7 +993,7 @@
void OopStorage::print_on(outputStream* st) const {
size_t allocations = _allocation_count;
- size_t blocks = _block_count;
+ size_t blocks = _active_array->block_count();
double data_size = section_size * section_count;
double alloc_percentage = percent_of((double)allocations, blocks * data_size);