< prev index next >
src/hotspot/share/runtime/synchronizer.cpp
Print this page
rev 56044 : imported patch 8230184.patch
rev 56046 : v2.00 -> v2.05 (CR5/v2.05/8-for-jdk13) patches combined into one; merge with 8229212.patch; merge with jdk-14+11; merge with 8230184.patch.
rev 56047 : renames, comment cleanups and additions, whitespace and indent fixes; add PaddedObjectMonitor typdef to make 'PaddedEnd<ObjectMonitor' cleanups easier; add a couple of missing 'private' decls; delete unused next() function; merge pieces from dcubed.monitor_deflate_conc.v2.06d in dcubed.monitor_deflate_conc.v2.06[ac]; merge with 8229212.patch; merge with jdk-14+11; merge with 8230184.patch.
rev 56048 : Add OM_CACHE_LINE_SIZE so that ObjectMonitor cache line sizes can be experimented with independently of DEFAULT_CACHE_LINE_SIZE; for SPARC and X64 configs that use 128 for DEFAULT_CACHE_LINE_SIZE, we are experimenting with 64; move _previous_owner_tid and _allocation_state fields to share the cache line with ObjectMonitor::_header; put ObjectMonitor::_ref_count on its own cache line after _owner; add 'int* count_p' parameter to deflate_monitor_list() and deflate_monitor_list_using_JT() and push counter updates down to where the ObjectMonitors are actually removed from the in-use lists; monitors_iterate() async deflation check should use negative ref_count; add 'JavaThread* target' param to deflate_per_thread_idle_monitors_using_JT() add deflate_common_idle_monitors_using_JT() to make it clear which JavaThread* is the target of the work and which is the calling JavaThread* (self); g_free_list, g_om_in_use_list and g_om_in_use_count are now static to synchronizer.cpp (reduce scope); add more diagnostic info to some assert()'s; minor code cleanups and code motion; save_om_ptr() should detect a race with a deflating thread that is bailing out and cause a retry when the ref_count field is not positive; merge with jdk-14+11; add special GC support for TestHumongousClassLoader.java; merge with 8230184.patch.
rev 56049 : Merge the remainder of the lock-free monitor list changes from v2.06 with v2.06a and v2.06b after running the changes through the edit scripts; merge pieces from dcubed.monitor_deflate_conc.v2.06d in dcubed.monitor_deflate_conc.v2.06[ac]; merge pieces from dcubed.monitor_deflate_conc.v2.06e into dcubed.monitor_deflate_conc.v2.06c; merge with jdk-14+11; test work around for test/jdk/tools/jlink/multireleasejar/JLinkMultiReleaseJarTest.java should not been needed anymore.
@@ -116,28 +116,349 @@
#define NINFLATIONLOCKS 256
static volatile intptr_t gInflationLocks[NINFLATIONLOCKS];
// global list of blocks of monitors
PaddedObjectMonitor* volatile ObjectSynchronizer::g_block_list = NULL;
+bool volatile ObjectSynchronizer::_is_async_deflation_requested = false;
+bool volatile ObjectSynchronizer::_is_special_deflation_requested = false;
+jlong ObjectSynchronizer::_last_async_deflation_time_ns = 0;
+
// Global ObjectMonitor free list. Newly allocated and deflated
// ObjectMonitors are prepended here.
-ObjectMonitor* volatile ObjectSynchronizer::g_free_list = NULL;
+static ObjectMonitor* volatile g_free_list = NULL;
// Global ObjectMonitor in-use list. When a JavaThread is exiting,
// ObjectMonitors on its per-thread in-use list are prepended here.
-ObjectMonitor* volatile ObjectSynchronizer::g_om_in_use_list = NULL;
-int ObjectSynchronizer::g_om_in_use_count = 0; // # on g_om_in_use_list
-bool volatile ObjectSynchronizer::_is_async_deflation_requested = false;
-bool volatile ObjectSynchronizer::_is_special_deflation_requested = false;
-jlong ObjectSynchronizer::_last_async_deflation_time_ns = 0;
+static ObjectMonitor* volatile g_om_in_use_list = NULL;
-static volatile intptr_t gListLock = 0; // protects global monitor lists
static volatile int g_om_free_count = 0; // # on g_free_list
+static volatile int g_om_in_use_count = 0; // # on g_om_in_use_list
static volatile int g_om_population = 0; // # Extant -- in circulation
#define CHAINMARKER (cast_to_oop<intptr_t>(-1))
+// =====================> List Management functions
+
+// Return true if the ObjectMonitor's next field is marked.
+// Otherwise returns false.
+static bool is_next_marked(ObjectMonitor* om) {
+ return ((intptr_t)OrderAccess::load_acquire(&om->_next_om) & 0x1) != 0;
+}
+
+// Mark an ObjectMonitor* and return it. Note: the om parameter
+// may or may not have been marked originally.
+static ObjectMonitor* mark_om_ptr(ObjectMonitor* om) {
+ return (ObjectMonitor*)((intptr_t)om | 0x1);
+}
+
+// Mark the next field in an ObjectMonitor. If marking was successful,
+// then the unmarked next field is returned via parameter and true is
+// returned. Otherwise false is returned.
+static bool mark_next(ObjectMonitor* om, ObjectMonitor** next_p) {
+ // Get current next field without any marking value.
+ ObjectMonitor* next = (ObjectMonitor*)
+ ((intptr_t)OrderAccess::load_acquire(&om->_next_om) & ~0x1);
+ if (Atomic::cmpxchg(mark_om_ptr(next), &om->_next_om, next) != next) {
+ return false; // Could not mark the next field or it was already marked.
+ }
+ *next_p = next;
+ return true;
+}
+
+// Loop until we mark the next field in an ObjectMonitor. The unmarked
+// next field is returned.
+static ObjectMonitor* mark_next_loop(ObjectMonitor* om) {
+ ObjectMonitor* next;
+ while (true) {
+ if (mark_next(om, &next)) {
+ // Marked om's next field so return the unmarked value.
+ return next;
+ }
+ }
+}
+
+// Set the next field in an ObjectMonitor to the specified value.
+// The caller of set_next() must be the same thread that marked the
+// ObjectMonitor.
+static void set_next(ObjectMonitor* om, ObjectMonitor* value) {
+ OrderAccess::release_store(&om->_next_om, value);
+}
+
+// Mark the next field in the list head ObjectMonitor. If marking was
+// successful, then the mid and the unmarked next field are returned
+// via parameter and true is returned. Otherwise false is returned.
+static bool mark_list_head(ObjectMonitor* volatile * list_p,
+ ObjectMonitor** mid_p, ObjectMonitor** next_p) {
+ while (true) {
+ ObjectMonitor* mid = OrderAccess::load_acquire(list_p);
+ if (mid == NULL) {
+ return false; // The list is empty so nothing to mark.
+ }
+ if (mark_next(mid, next_p)) {
+ if (OrderAccess::load_acquire(list_p) != mid) {
+ // The list head changed so we have to retry.
+ set_next(mid, *next_p); // unmark mid
+ continue;
+ }
+ // We marked next field to guard against races.
+ *mid_p = mid;
+ return true;
+ }
+ }
+}
+
+// Return the unmarked next field in an ObjectMonitor. Note: the next
+// field may or may not have been marked originally.
+static ObjectMonitor* unmarked_next(ObjectMonitor* om) {
+ return (ObjectMonitor*)((intptr_t)OrderAccess::load_acquire(&om->_next_om) & ~0x1);
+}
+
+#if 0
+// XXX - this is unused
+// Unmark the next field in an ObjectMonitor. Requires that the next
+// field be marked.
+static void unmark_next(ObjectMonitor* om) {
+ ADIM_guarantee(is_next_marked(om), "next field must be marked: next=" INTPTR_FORMAT, p2i(om->_next_om));
+
+ ObjectMonitor* next = unmarked_next(om);
+ set_next(om, next);
+}
+#endif
+
+volatile int visit_counter = 42;
+static void chk_for_list_loop(ObjectMonitor* list, int count) {
+ if (!CheckMonitorLists) {
+ return;
+ }
+ int l_visit_counter = Atomic::add(1, &visit_counter);
+ int l_count = 0;
+ ObjectMonitor* prev = NULL;
+ for (ObjectMonitor* mid = list; mid != NULL; mid = unmarked_next(mid)) {
+ if (mid->visit_marker == l_visit_counter) {
+ log_error(monitorinflation)("ERROR: prev=" INTPTR_FORMAT ", l_count=%d"
+ " refers to an ObjectMonitor that has"
+ " already been visited: mid=" INTPTR_FORMAT,
+ p2i(prev), l_count, p2i(mid));
+ fatal("list=" INTPTR_FORMAT " of %d items has a loop.", p2i(list), count);
+ }
+ mid->visit_marker = l_visit_counter;
+ prev = mid;
+ if (++l_count > count + 1024 * 1024) {
+ fatal("list=" INTPTR_FORMAT " of %d items may have a loop; l_count=%d",
+ p2i(list), count, l_count);
+ }
+ }
+}
+
+static void chk_om_not_on_list(ObjectMonitor* om, ObjectMonitor* list, int count) {
+ if (!CheckMonitorLists) {
+ return;
+ }
+ guarantee(list != om, "ERROR: om=" INTPTR_FORMAT " must not be head of the "
+ "list=" INTPTR_FORMAT ", count=%d", p2i(om), p2i(list), count);
+ int l_count = 0;
+ for (ObjectMonitor* mid = list; mid != NULL; mid = unmarked_next(mid)) {
+ if (unmarked_next(mid) == om) {
+ log_error(monitorinflation)("ERROR: mid=" INTPTR_FORMAT ", l_count=%d"
+ " next_om refers to om=" INTPTR_FORMAT,
+ p2i(mid), l_count, p2i(om));
+ fatal("list=" INTPTR_FORMAT " of %d items has bad next_om value.",
+ p2i(list), count);
+ }
+ if (++l_count > count + 1024 * 1024) {
+ fatal("list=" INTPTR_FORMAT " of %d items may have a loop; l_count=%d",
+ p2i(list), count, l_count);
+ }
+ }
+}
+
+static void chk_om_elems_not_on_list(ObjectMonitor* elems, int elems_count,
+ ObjectMonitor* list, int list_count) {
+ if (!CheckMonitorLists) {
+ return;
+ }
+ chk_for_list_loop(elems, elems_count);
+ for (ObjectMonitor* mid = elems; mid != NULL; mid = unmarked_next(mid)) {
+ chk_om_not_on_list(mid, list, list_count);
+ }
+}
+
+// Prepend a list of ObjectMonitors to the specified *list_p. 'tail' is
+// the last ObjectMonitor in the list and there are 'count' on the list.
+// Also updates the specified *count_p.
+static void prepend_list_to_common(ObjectMonitor* list, ObjectMonitor* tail,
+ int count, ObjectMonitor* volatile* list_p,
+ volatile int* count_p) {
+ chk_for_list_loop(OrderAccess::load_acquire(list_p),
+ OrderAccess::load_acquire(count_p));
+ chk_om_elems_not_on_list(list, count, OrderAccess::load_acquire(list_p),
+ OrderAccess::load_acquire(count_p));
+ while (true) {
+ ObjectMonitor* cur = OrderAccess::load_acquire(list_p);
+ // Prepend list to *list_p.
+ ObjectMonitor* next = NULL;
+ if (!mark_next(tail, &next)) {
+ continue; // failed to mark next field so try it all again
+ }
+ set_next(tail, cur); // tail now points to cur (and unmarks tail)
+ if (cur == NULL) {
+ // No potential race with takers or other prependers since
+ // *list_p is empty.
+ if (Atomic::cmpxchg(list, list_p, cur) == cur) {
+ // Successfully switched *list_p to the list value.
+ Atomic::add(count, count_p);
+ break;
+ }
+ // Implied else: try it all again
+ } else {
+ // Try to mark next field to guard against races:
+ if (!mark_next(cur, &next)) {
+ continue; // failed to mark next field so try it all again
+ }
+ // We marked the next field so try to switch *list_p to the list value.
+ if (Atomic::cmpxchg(list, list_p, cur) != cur) {
+ // The list head has changed so unmark the next field and try again:
+ set_next(cur, next);
+ continue;
+ }
+ Atomic::add(count, count_p);
+ set_next(cur, next); // unmark next field
+ break;
+ }
+ }
+}
+
+// Prepend a newly allocated block of ObjectMonitors to g_block_list and
+// g_free_list. Also updates g_om_population and g_om_free_count.
+void ObjectSynchronizer::prepend_block_to_lists(PaddedObjectMonitor* new_blk) {
+ // First we handle g_block_list:
+ while (true) {
+ PaddedObjectMonitor* cur = OrderAccess::load_acquire(&g_block_list);
+ // Prepend new_blk to g_block_list. The first ObjectMonitor in
+ // a block is reserved for use as linkage to the next block.
+ OrderAccess::release_store(&new_blk[0]._next_om, cur);
+ if (Atomic::cmpxchg(new_blk, &g_block_list, cur) == cur) {
+ // Successfully switched g_block_list to the new_blk value.
+ Atomic::add(_BLOCKSIZE - 1, &g_om_population);
+ break;
+ }
+ // Implied else: try it all again
+ }
+
+ // Second we handle g_free_list:
+ prepend_list_to_common(new_blk + 1, &new_blk[_BLOCKSIZE - 1], _BLOCKSIZE - 1,
+ &g_free_list, &g_om_free_count);
+}
+
+// Prepend a list of ObjectMonitors to g_free_list. 'tail' is the last
+// ObjectMonitor in the list and there are 'count' on the list. Also
+// updates g_om_free_count.
+static void prepend_list_to_g_free_list(ObjectMonitor* list,
+ ObjectMonitor* tail, int count) {
+ prepend_list_to_common(list, tail, count, &g_free_list, &g_om_free_count);
+}
+
+// Prepend a list of ObjectMonitors to g_om_in_use_list. 'tail' is the last
+// ObjectMonitor in the list and there are 'count' on the list. Also
+// updates g_om_in_use_list.
+static void prepend_list_to_g_om_in_use_list(ObjectMonitor* list,
+ ObjectMonitor* tail, int count) {
+ prepend_list_to_common(list, tail, count, &g_om_in_use_list, &g_om_in_use_count);
+}
+
+// Prepend an ObjectMonitor to the specified list. Also updates
+// the specified counter.
+static void prepend_to_common(ObjectMonitor* m, ObjectMonitor* volatile * list_p,
+ int volatile * count_p) {
+ chk_for_list_loop(OrderAccess::load_acquire(list_p),
+ OrderAccess::load_acquire(count_p));
+ chk_om_not_on_list(m, OrderAccess::load_acquire(list_p),
+ OrderAccess::load_acquire(count_p));
+
+ while (true) {
+ ObjectMonitor* cur = OrderAccess::load_acquire(list_p);
+ // Prepend ObjectMonitor to *list_p.
+ ObjectMonitor* next = NULL;
+ if (!mark_next(m, &next)) {
+ continue; // failed to mark next field so try it all again
+ }
+ set_next(m, cur); // m now points to cur (and unmarks m)
+ if (cur == NULL) {
+ // No potential race with other prependers since *list_p is empty.
+ if (Atomic::cmpxchg(m, list_p, cur) == cur) {
+ // Successfully switched *list_p to 'm'.
+ Atomic::inc(count_p);
+ break;
+ }
+ // Implied else: try it all again
+ } else {
+ // Try to mark next field to guard against races:
+ if (!mark_next(cur, &next)) {
+ continue; // failed to mark next field so try it all again
+ }
+ // We marked the next field so try to switch *list_p to 'm'.
+ if (Atomic::cmpxchg(m, list_p, cur) != cur) {
+ // The list head has changed so unmark the next field and try again:
+ set_next(cur, next);
+ continue;
+ }
+ Atomic::inc(count_p);
+ set_next(cur, next); // unmark next field
+ break;
+ }
+ }
+}
+
+// Prepend an ObjectMonitor to a per-thread om_free_list.
+// Also updates the per-thread om_free_count.
+static void prepend_to_om_free_list(Thread* self, ObjectMonitor* m) {
+ prepend_to_common(m, &self->om_free_list, &self->om_free_count);
+}
+
+// Prepend an ObjectMonitor to a per-thread om_in_use_list.
+// Also updates the per-thread om_in_use_count.
+static void prepend_to_om_in_use_list(Thread* self, ObjectMonitor* m) {
+ prepend_to_common(m, &self->om_in_use_list, &self->om_in_use_count);
+}
+
+// Take an ObjectMonitor from the start of the specified list. Also
+// decrements the specified counter. Returns NULL if none are available.
+static ObjectMonitor* take_from_start_of_common(ObjectMonitor* volatile * list_p,
+ int volatile * count_p) {
+ chk_for_list_loop(OrderAccess::load_acquire(list_p),
+ OrderAccess::load_acquire(count_p));
+
+ ObjectMonitor* next = NULL;
+ ObjectMonitor* take = NULL;
+ // Mark the list head to guard against A-B-A race:
+ if (!mark_list_head(list_p, &take, &next)) {
+ return NULL; // None are available.
+ }
+ // Switch marked list head to next (which unmarks the list head, but
+ // leaves take marked):
+ OrderAccess::release_store(list_p, next);
+ Atomic::dec(count_p);
+ // Unmark take, but leave the next value for any lagging list
+ // walkers. It will get cleaned up when take is prepended to
+ // the in-use list:
+ set_next(take, next);
+ return take;
+}
+
+// Take an ObjectMonitor from the start of the global free-list. Also
+// updates g_om_free_count. Returns NULL if none are available.
+static ObjectMonitor* take_from_start_of_g_free_list() {
+ return take_from_start_of_common(&g_free_list, &g_om_free_count);
+}
+
+// Take an ObjectMonitor from the start of a per-thread free-list.
+// Also updates om_free_count. Returns NULL if none are available.
+static ObjectMonitor* take_from_start_of_om_free_list(Thread* self) {
+ return take_from_start_of_common(&self->om_free_list, &self->om_free_count);
+}
+
+
// =====================> Quick functions
// The quick_* forms are special fast-path variants used to improve
// performance. In the simplest case, a "quick_*" implementation could
// simply return false, in which case the caller will perform the necessary
@@ -580,19 +901,19 @@
// As a general policy we use "volatile" to control compiler-based reordering
// and explicit fences (barriers) to control for architectural reordering
// performed by the CPU(s) or platform.
struct SharedGlobals {
- char _pad_prefix[DEFAULT_CACHE_LINE_SIZE];
+ char _pad_prefix[OM_CACHE_LINE_SIZE];
// These are highly shared mostly-read variables.
// To avoid false-sharing they need to be the sole occupants of a cache line.
volatile int stw_random;
volatile int stw_cycle;
- DEFINE_PAD_MINUS_SIZE(1, DEFAULT_CACHE_LINE_SIZE, sizeof(volatile int) * 2);
+ DEFINE_PAD_MINUS_SIZE(1, OM_CACHE_LINE_SIZE, sizeof(volatile int) * 2);
// Hot RW variable -- Sequester to avoid false-sharing
volatile int hc_sequence;
- DEFINE_PAD_MINUS_SIZE(2, DEFAULT_CACHE_LINE_SIZE, sizeof(volatile int));
+ DEFINE_PAD_MINUS_SIZE(2, OM_CACHE_LINE_SIZE, sizeof(volatile int));
};
static SharedGlobals GVars;
static int MonitorScavengeThreshold = 1000000;
static volatile int ForceMonitorScavenge = 0; // Scavenge required and pending
@@ -996,32 +1317,36 @@
ObjectMonitor* mid = (ObjectMonitor *)(block + i);
if (mid->is_active()) {
ObjectMonitorHandle omh(mid);
if (mid->object() == NULL ||
- (AsyncDeflateIdleMonitors && mid->_owner == DEFLATER_MARKER)) {
+ (AsyncDeflateIdleMonitors && mid->ref_count() < 0)) {
// Only process with closure if the object is set.
// For async deflation, race here if monitor is not owned!
// The above ref_count bump (in ObjectMonitorHandle ctr)
// will cause subsequent async deflation to skip it.
- // However, previous or concurrent async deflation is a race.
+ // However, previous or concurrent async deflation is a race
+ // so skip this ObjectMonitor if it is being async deflated.
continue;
}
closure->do_monitor(mid);
}
}
- block = (PaddedObjectMonitor*)block->_next_om;
+ // unmarked_next() is not needed with g_block_list (no next field marking).
+ block = (PaddedObjectMonitor*)OrderAccess::load_acquire(&block->_next_om);
}
}
static bool monitors_used_above_threshold() {
- if (g_om_population == 0) {
+ if (OrderAccess::load_acquire(&g_om_population) == 0) {
return false;
}
if (MonitorUsedDeflationThreshold > 0) {
- int monitors_used = g_om_population - g_om_free_count;
- int monitor_usage = (monitors_used * 100LL) / g_om_population;
+ int monitors_used = OrderAccess::load_acquire(&g_om_population) -
+ OrderAccess::load_acquire(&g_om_free_count);
+ int monitor_usage = (monitors_used * 100LL) /
+ OrderAccess::load_acquire(&g_om_population);
return monitor_usage > MonitorUsedDeflationThreshold;
}
return false;
}
@@ -1048,11 +1373,12 @@
// than AsyncDeflationInterval (unless is_async_deflation_requested)
// in order to not swamp the ServiceThread.
_last_async_deflation_time_ns = os::javaTimeNanos();
return true;
}
- if (is_MonitorBound_exceeded(g_om_population - g_om_free_count)) {
+ if (is_MonitorBound_exceeded(OrderAccess::load_acquire(&g_om_population) -
+ OrderAccess::load_acquire(&g_om_free_count))) {
// Not enough ObjectMonitors on the global free list.
return true;
}
return false;
}
@@ -1083,22 +1409,25 @@
global_used_oops_do(f);
}
void ObjectSynchronizer::global_used_oops_do(OopClosure* f) {
assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
- list_oops_do(g_om_in_use_list, f);
+ list_oops_do(OrderAccess::load_acquire(&g_om_in_use_list), OrderAccess::load_acquire(&g_om_in_use_count), f);
}
void ObjectSynchronizer::thread_local_used_oops_do(Thread* thread, OopClosure* f) {
assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
- list_oops_do(thread->om_in_use_list, f);
+ list_oops_do(OrderAccess::load_acquire(&thread->om_in_use_list), OrderAccess::load_acquire(&thread->om_in_use_count), f);
}
-void ObjectSynchronizer::list_oops_do(ObjectMonitor* list, OopClosure* f) {
+void ObjectSynchronizer::list_oops_do(ObjectMonitor* list, int count, OopClosure* f) {
assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
- ObjectMonitor* mid;
- for (mid = list; mid != NULL; mid = mid->_next_om) {
+ chk_for_list_loop(list, count);
+ // The oops_do() phase does not overlap with monitor deflation
+ // so no need to update the ObjectMonitor's ref_count for this
+ // ObjectMonitor* use.
+ for (ObjectMonitor* mid = list; mid != NULL; mid = unmarked_next(mid)) {
if (mid->object() != NULL) {
f->do_oop((oop*)mid->object_addr());
}
}
}
@@ -1110,13 +1439,10 @@
// Inflation unlinks monitors from the global g_free_list and
// associates them with objects. Deflation -- which occurs at
// STW-time -- disassociates idle monitors from objects. Such
// scavenged monitors are returned to the g_free_list.
//
-// The global list is protected by gListLock. All the critical sections
-// are short and operate in constant-time.
-//
// ObjectMonitors reside in type-stable memory (TSM) and are immortal.
//
// Lifecycle:
// -- unassigned and on the global free list
// -- unassigned and on a thread's private om_free_list
@@ -1155,12 +1481,12 @@
// by a safepoint's cleanup phase or by the ServiceThread at periodic
// intervals when is_async_deflation_needed() returns true. In
// addition to other policies that are checked, if there are not
// enough ObjectMonitors on the global free list, then
// is_async_deflation_needed() will return true. The ServiceThread
-// calls deflate_global_idle_monitors_using_JT() and also sets the
-// per-thread om_request_deflation flag as needed.
+// calls deflate_global_idle_monitors_using_JT() and also calls
+// deflate_per_thread_idle_monitors_using_JT() as needed.
static void InduceScavenge(Thread* self, const char * Whence) {
assert(!AsyncDeflateIdleMonitors, "is not used by async deflation");
// Induce STW safepoint to trim monitors
@@ -1185,62 +1511,42 @@
// number of ObjectMonitors in circulation as well as the STW
// scavenge costs. As usual, we lean toward time in space-time
// tradeoffs.
const int MAXPRIVATE = 1024;
- if (AsyncDeflateIdleMonitors) {
- JavaThread* jt = (JavaThread *)self;
- if (jt->om_request_deflation && jt->om_in_use_count > 0 &&
- cause != inflate_cause_vm_internal) {
- // Deflate any per-thread idle monitors for this JavaThread if
- // this is not an internal inflation; internal inflations can
- // occur in places where it is not safe to pause for a safepoint.
- // Clean up your own mess (Gibbs Rule 45). Otherwise, skip this
- // deflation. deflate_global_idle_monitors_using_JT() is called
- // by the ServiceThread. Per-thread async deflation is triggered
- // by the ServiceThread via om_request_deflation.
- debug_only(jt->check_for_valid_safepoint_state(false);)
- ObjectSynchronizer::deflate_per_thread_idle_monitors_using_JT();
- }
- }
-
stringStream ss;
for (;;) {
ObjectMonitor* m;
// 1: try to allocate from the thread's local om_free_list.
// Threads will attempt to allocate first from their local list, then
- // from the global list, and only after those attempts fail will the thread
- // attempt to instantiate new monitors. Thread-local free lists take
- // heat off the gListLock and improve allocation latency, as well as reducing
- // coherency traffic on the shared global list.
- m = self->om_free_list;
+ // from the global list, and only after those attempts fail will the
+ // thread attempt to instantiate new monitors. Thread-local free lists
+ // improve allocation latency, as well as reducing coherency traffic
+ // on the shared global list.
+ m = take_from_start_of_om_free_list(self);
if (m != NULL) {
- self->om_free_list = m->_next_om;
- self->om_free_count--;
guarantee(m->object() == NULL, "invariant");
m->set_allocation_state(ObjectMonitor::New);
- m->_next_om = self->om_in_use_list;
- self->om_in_use_list = m;
- self->om_in_use_count++;
+ prepend_to_om_in_use_list(self, m);
return m;
}
// 2: try to allocate from the global g_free_list
// CONSIDER: use muxTry() instead of muxAcquire().
// If the muxTry() fails then drop immediately into case 3.
// If we're using thread-local free lists then try
// to reprovision the caller's free list.
- if (g_free_list != NULL) {
+ if (OrderAccess::load_acquire(&g_free_list) != NULL) {
// Reprovision the thread's om_free_list.
// Use bulk transfers to reduce the allocation rate and heat
// on various locks.
- Thread::muxAcquire(&gListLock, "om_alloc(1)");
- for (int i = self->om_free_provision; --i >= 0 && g_free_list != NULL;) {
- g_om_free_count--;
- ObjectMonitor* take = g_free_list;
- g_free_list = take->_next_om;
+ for (int i = self->om_free_provision; --i >= 0;) {
+ ObjectMonitor* take = take_from_start_of_g_free_list();
+ if (take == NULL) {
+ break; // No more are available.
+ }
guarantee(take->object() == NULL, "invariant");
if (AsyncDeflateIdleMonitors) {
// We allowed 3 field values to linger during async deflation.
// We clear header and restore ref_count here, but we leave
// owner == DEFLATER_MARKER so the simple C2 ObjectMonitor
@@ -1258,16 +1564,16 @@
}
take->Recycle();
assert(take->is_free(), "invariant");
om_release(self, take, false);
}
- Thread::muxRelease(&gListLock);
self->om_free_provision += 1 + (self->om_free_provision/2);
if (self->om_free_provision > MAXPRIVATE) self->om_free_provision = MAXPRIVATE;
if (!AsyncDeflateIdleMonitors &&
- is_MonitorBound_exceeded(g_om_population - g_om_free_count)) {
+ is_MonitorBound_exceeded(OrderAccess::load_acquire(&g_om_population) -
+ OrderAccess::load_acquire(&g_om_free_count))) {
// Not enough ObjectMonitors on the global free list.
// We can't safely induce a STW safepoint from om_alloc() as our thread
// state may not be appropriate for such activities and callers may hold
// naked oops, so instead we defer the action.
InduceScavenge(self, "om_alloc");
@@ -1284,14 +1590,14 @@
// A better solution would be to use C++ placement-new.
// BEWARE: As it stands currently, we don't run the ctors!
assert(_BLOCKSIZE > 1, "invariant");
size_t neededsize = sizeof(PaddedObjectMonitor) * _BLOCKSIZE;
PaddedObjectMonitor* temp;
- size_t aligned_size = neededsize + (DEFAULT_CACHE_LINE_SIZE - 1);
+ size_t aligned_size = neededsize + (OM_CACHE_LINE_SIZE - 1);
void* real_malloc_addr = (void*)NEW_C_HEAP_ARRAY(char, aligned_size,
mtInternal);
- temp = (PaddedObjectMonitor*)align_up(real_malloc_addr, DEFAULT_CACHE_LINE_SIZE);
+ temp = (PaddedObjectMonitor*)align_up(real_malloc_addr, OM_CACHE_LINE_SIZE);
// NOTE: (almost) no way to recover if allocation failed.
// We might be able to induce a STW safepoint and scavenge enough
// ObjectMonitors to permit progress.
if (temp == NULL) {
@@ -1307,42 +1613,25 @@
// The trick of using the 1st element in the block as g_block_list
// linkage should be reconsidered. A better implementation would
// look like: class Block { Block * next; int N; ObjectMonitor Body [N] ; }
for (int i = 1; i < _BLOCKSIZE; i++) {
- temp[i]._next_om = (ObjectMonitor *)&temp[i+1];
+ OrderAccess::release_store(&temp[i]._next_om, (ObjectMonitor*)&temp[i+1]);
assert(temp[i].is_free(), "invariant");
}
// terminate the last monitor as the end of list
- temp[_BLOCKSIZE - 1]._next_om = NULL;
+ OrderAccess::release_store(&temp[_BLOCKSIZE - 1]._next_om, (ObjectMonitor*)NULL);
// Element [0] is reserved for global list linkage
temp[0].set_object(CHAINMARKER);
// Consider carving out this thread's current request from the
// block in hand. This avoids some lock traffic and redundant
// list activity.
- // Acquire the gListLock to manipulate g_block_list and g_free_list.
- // An Oyama-Taura-Yonezawa scheme might be more efficient.
- Thread::muxAcquire(&gListLock, "om_alloc(2)");
- g_om_population += _BLOCKSIZE-1;
- g_om_free_count += _BLOCKSIZE-1;
-
- // Add the new block to the list of extant blocks (g_block_list).
- // The very first ObjectMonitor in a block is reserved and dedicated.
- // It serves as blocklist "next" linkage.
- temp[0]._next_om = g_block_list;
- // There are lock-free uses of g_block_list so make sure that
- // the previous stores happen before we update g_block_list.
- OrderAccess::release_store(&g_block_list, temp);
-
- // Add the new string of ObjectMonitors to the global free list
- temp[_BLOCKSIZE - 1]._next_om = g_free_list;
- g_free_list = temp + 1;
- Thread::muxRelease(&gListLock);
+ prepend_block_to_lists(temp);
}
}
// Place "m" on the caller's private per-thread om_free_list.
// In practice there's no need to clamp or limit the number of
@@ -1367,32 +1656,70 @@
m->set_allocation_state(ObjectMonitor::Free);
// _next_om is used for both per-thread in-use and free lists so
// we have to remove 'm' from the in-use list first (as needed).
if (from_per_thread_alloc) {
// Need to remove 'm' from om_in_use_list.
+ // We use the more complicated mark-cur_mid_in_use-and-mid-as-we-go
+ // protocol because async deflation can do list deletions in parallel.
ObjectMonitor* cur_mid_in_use = NULL;
+ ObjectMonitor* mid = NULL;
+ ObjectMonitor* next = NULL;
bool extracted = false;
- for (ObjectMonitor* mid = self->om_in_use_list; mid != NULL; cur_mid_in_use = mid, mid = mid->_next_om) {
+
+ if (!mark_list_head(&self->om_in_use_list, &mid, &next)) {
+ fatal("thread=" INTPTR_FORMAT " in-use list must not be empty.", p2i(self));
+ }
+ while (true) {
if (m == mid) {
- // extract from per-thread in-use list
- if (mid == self->om_in_use_list) {
- self->om_in_use_list = mid->_next_om;
- } else if (cur_mid_in_use != NULL) {
- cur_mid_in_use->_next_om = mid->_next_om; // maintain the current thread in-use list
+ // We found 'm' on the per-thread in-use list so try to extract it.
+ // First try the list head:
+ if (Atomic::cmpxchg(next, &self->om_in_use_list, mid) != mid) {
+ // We could not switch the list head to next.
+ ObjectMonitor* marked_mid = mark_om_ptr(mid);
+ // Switch cur_mid_in_use's next field to next (which also
+ // unmarks cur_mid_in_use):
+ ADIM_guarantee(cur_mid_in_use != NULL, "must not be NULL");
+ if (Atomic::cmpxchg(next, &cur_mid_in_use->_next_om, marked_mid)
+ != marked_mid) {
+ // We could not switch cur_mid_in_use's next field. This
+ // should not be possible since it was marked so we:
+ fatal("mid=" INTPTR_FORMAT " must be referred to by the list "
+ "head: &om_in_use_list=" INTPTR_FORMAT " or by "
+ "cur_mid_in_use's next field: cur_mid_in_use=" INTPTR_FORMAT
+ ", next_om=" INTPTR_FORMAT, p2i(mid),
+ p2i((ObjectMonitor**)&self->om_in_use_list),
+ p2i(cur_mid_in_use), p2i(cur_mid_in_use->_next_om));
+ }
}
extracted = true;
- self->om_in_use_count--;
+ Atomic::dec(&self->om_in_use_count);
+ // Unmark mid, but leave the next value for any lagging list
+ // walkers. It will get cleaned up when mid is prepended to
+ // the thread's free list:
+ set_next(mid, next);
break;
}
+ if (cur_mid_in_use != NULL) {
+ set_next(cur_mid_in_use, mid); // umark cur_mid_in_use
+ }
+ // The next cur_mid_in_use keeps mid's marked next field so
+ // that it is stable for a possible next field change. It
+ // cannot be deflated while it is marked.
+ cur_mid_in_use = mid;
+ mid = next;
+ if (mid == NULL) {
+ // Reached end of the list and didn't find m so:
+ fatal("must find m=" INTPTR_FORMAT "on om_in_use_list=" INTPTR_FORMAT,
+ p2i(m), p2i(self->om_in_use_list));
+ }
+ // Mark mid's next field so we can possibly extract it:
+ next = mark_next_loop(mid);
}
- assert(extracted, "Should have extracted from in-use list");
}
- m->_next_om = self->om_free_list;
+ prepend_to_om_free_list(self, m);
guarantee(m->is_free(), "invariant");
- self->om_free_list = m;
- self->om_free_count++;
}
// Return ObjectMonitors on a moribund thread's free and in-use
// lists to the appropriate global lists. The ObjectMonitors on the
// per-thread in-use list may still be in use by other threads.
@@ -1406,68 +1733,111 @@
// om_flush() is called) or via ObjectSynchronizer::oops_do() (after
// om_flush() is called).
//
// With AsyncDeflateIdleMonitors, deflate_global_idle_monitors_using_JT()
// and deflate_per_thread_idle_monitors_using_JT() (in another thread) can
-// run at the same time as om_flush() so we have to be careful.
+// run at the same time as om_flush() so we have to follow a careful
+// protocol to prevent list corruption.
void ObjectSynchronizer::om_flush(Thread* self) {
- ObjectMonitor* free_list = self->om_free_list;
- ObjectMonitor* free_tail = NULL;
+ // This function can race with an async deflater thread. Since
+ // deflation has to process the per-thread in-use list before
+ // prepending the deflated ObjectMonitors to the global free list,
+ // we process the per-thread lists in the same order to prevent
+ // ordering races.
+ int in_use_count = 0;
+ ObjectMonitor* in_use_list = NULL;
+ ObjectMonitor* in_use_tail = NULL;
+ ObjectMonitor* next = NULL;
+
+ // An async deflation thread checks to see if the target thread
+ // is exiting, but if it has made it past that check before we
+ // started exiting, then it is racing to get to the in-use list.
+ if (mark_list_head(&self->om_in_use_list, &in_use_list, &next)) {
+ chk_for_list_loop(in_use_list, OrderAccess::load_acquire(&self->om_in_use_count));
+ // At this point, we have marked the in-use list head so an
+ // async deflation thread cannot come in after us. If an async
+ // deflation thread is ahead of us, then we'll detect that and
+ // wait for it to finish its work.
+ //
+ // The thread is going away, however the ObjectMonitors on the
+ // om_in_use_list may still be in-use by other threads. Link
+ // them to in_use_tail, which will be linked into the global
+ // in-use list g_om_in_use_list below.
+ //
+ // Account for the in-use list head before the loop since it is
+ // already marked (by this thread):
+ in_use_tail = in_use_list;
+ in_use_count++;
+ for (ObjectMonitor* cur_om = unmarked_next(in_use_list); cur_om != NULL;) {
+ if (is_next_marked(cur_om)) {
+ // This next field is marked so there must be an async deflater
+ // thread ahead of us so we'll give it a chance to finish.
+ while (is_next_marked(cur_om)) {
+ os::naked_short_sleep(1);
+ }
+ // Refetch the possibly changed next field and try again.
+ cur_om = unmarked_next(in_use_tail);
+ continue;
+ }
+ if (!cur_om->is_active()) {
+ // cur_om was deflated and the allocation state was changed
+ // to Free while it was marked. We happened to see it just
+ // after it was unmarked (and added to the free list).
+ // Refetch the possibly changed next field and try again.
+ cur_om = unmarked_next(in_use_tail);
+ continue;
+ }
+ in_use_tail = cur_om;
+ in_use_count++;
+ cur_om = unmarked_next(cur_om);
+ }
+ guarantee(in_use_tail != NULL, "invariant");
+ int l_om_in_use_count = OrderAccess::load_acquire(&self->om_in_use_count);
+ ADIM_guarantee(l_om_in_use_count == in_use_count, "in-use counts don't "
+ "match: l_om_in_use_count=%d, in_use_count=%d",
+ l_om_in_use_count, in_use_count);
+ // Clear the in-use count before unmarking the in-use list head
+ // to avoid races:
+ OrderAccess::release_store(&self->om_in_use_count, 0);
+ // Clear the in-use list head (which also unmarks it):
+ OrderAccess::release_store(&self->om_in_use_list, (ObjectMonitor*)NULL);
+ // Unmark the disconnected list head:
+ set_next(in_use_list, next);
+ }
+
int free_count = 0;
+ ObjectMonitor* free_list = OrderAccess::load_acquire(&self->om_free_list);
+ ObjectMonitor* free_tail = NULL;
if (free_list != NULL) {
- ObjectMonitor* s;
+ chk_for_list_loop(free_list, OrderAccess::load_acquire(&self->om_free_count));
// The thread is going away. Set 'free_tail' to the last per-thread free
- // monitor which will be linked to g_free_list below under the gListLock.
+ // monitor which will be linked to g_free_list below.
stringStream ss;
- for (s = free_list; s != NULL; s = s->_next_om) {
+ for (ObjectMonitor* s = free_list; s != NULL; s = unmarked_next(s)) {
free_count++;
free_tail = s;
guarantee(s->object() == NULL, "invariant");
guarantee(!s->is_busy(), "must be !is_busy: %s", s->is_busy_to_string(&ss));
}
guarantee(free_tail != NULL, "invariant");
- ADIM_guarantee(self->om_free_count == free_count, "free-count off");
- self->om_free_list = NULL;
- self->om_free_count = 0;
+ int l_om_free_count = OrderAccess::load_acquire(&self->om_free_count);
+ ADIM_guarantee(l_om_free_count == free_count, "free counts don't match: "
+ "l_om_free_count=%d, free_count=%d", l_om_free_count,
+ free_count);
+ OrderAccess::release_store(&self->om_free_list, (ObjectMonitor*)NULL);
+ OrderAccess::release_store(&self->om_free_count, 0);
}
- ObjectMonitor* in_use_list = self->om_in_use_list;
- ObjectMonitor* in_use_tail = NULL;
- int in_use_count = 0;
- if (in_use_list != NULL) {
- // The thread is going away, however the ObjectMonitors on the
- // om_in_use_list may still be in-use by other threads. Link
- // them to in_use_tail, which will be linked into the global
- // in-use list g_om_in_use_list below, under the gListLock.
- ObjectMonitor *cur_om;
- for (cur_om = in_use_list; cur_om != NULL; cur_om = cur_om->_next_om) {
- in_use_tail = cur_om;
- in_use_count++;
- ADIM_guarantee(cur_om->is_active(), "invariant");
- }
- guarantee(in_use_tail != NULL, "invariant");
- ADIM_guarantee(self->om_in_use_count == in_use_count, "in-use count off");
- self->om_in_use_list = NULL;
- self->om_in_use_count = 0;
- }
-
- Thread::muxAcquire(&gListLock, "om_flush");
if (free_tail != NULL) {
- free_tail->_next_om = g_free_list;
- g_free_list = free_list;
- g_om_free_count += free_count;
+ prepend_list_to_g_free_list(free_list, free_tail, free_count);
}
if (in_use_tail != NULL) {
- in_use_tail->_next_om = g_om_in_use_list;
- g_om_in_use_list = in_use_list;
- g_om_in_use_count += in_use_count;
+ prepend_list_to_g_om_in_use_list(in_use_list, in_use_tail, in_use_count);
}
- Thread::muxRelease(&gListLock);
-
LogStreamHandle(Debug, monitorinflation) lsh_debug;
LogStreamHandle(Info, monitorinflation) lsh_info;
LogStream* ls = NULL;
if (log_is_enabled(Debug, monitorinflation)) {
ls = &lsh_debug;
@@ -1580,22 +1950,11 @@
// See the comments in om_alloc().
LogStreamHandle(Trace, monitorinflation) lsh;
if (mark.has_locker()) {
- ObjectMonitor* m;
- if (!AsyncDeflateIdleMonitors || cause == inflate_cause_vm_internal) {
- // If !AsyncDeflateIdleMonitors or if an internal inflation, then
- // we won't stop for a potential safepoint in om_alloc.
- m = om_alloc(self, cause);
- } else {
- // If AsyncDeflateIdleMonitors and not an internal inflation, then
- // we may stop for a safepoint in om_alloc() so protect object.
- Handle h_obj(self, object);
- m = om_alloc(self, cause);
- object = h_obj(); // Refresh object.
- }
+ ObjectMonitor* m = om_alloc(self, cause);
// Optimistically prepare the objectmonitor - anticipate successful CAS
// We do this before the CAS in order to minimize the length of time
// in which INFLATING appears in the mark.
m->Recycle();
m->_Responsible = NULL;
@@ -1689,23 +2048,12 @@
// An inflateTry() method that we could call from fast_enter() and slow_enter()
// would be useful.
// Catch if the object's header is not neutral (not locked and
// not marked is what we care about here).
- ADIM_guarantee(mark.is_neutral(), "invariant: header=" INTPTR_FORMAT, mark.value());
- ObjectMonitor* m;
- if (!AsyncDeflateIdleMonitors || cause == inflate_cause_vm_internal) {
- // If !AsyncDeflateIdleMonitors or if an internal inflation, then
- // we won't stop for a potential safepoint in om_alloc.
- m = om_alloc(self, cause);
- } else {
- // If AsyncDeflateIdleMonitors and not an internal inflation, then
- // we may stop for a safepoint in om_alloc() so protect object.
- Handle h_obj(self, object);
- m = om_alloc(self, cause);
- object = h_obj(); // Refresh object.
- }
+ ADIM_guarantee(mark.is_neutral(), "invariant: header=" INTPTR_FORMAT,mark.value());
+ ObjectMonitor* m = om_alloc(self, cause);
// prepare m for installation - set monitor to initial state
m->Recycle();
m->set_header(mark);
// If we leave _owner == DEFLATER_MARKER here, then the simple C2
// ObjectMonitor enter optimization can no longer race with async
@@ -1717,10 +2065,14 @@
omh_p->set_om_ptr(m);
assert(m->is_new(), "freshly allocated monitor must be new");
m->set_allocation_state(ObjectMonitor::Old);
if (object->cas_set_mark(markWord::encode(m), mark) != mark) {
+ guarantee(!m->owner_is_DEFLATER_MARKER() || m->ref_count() >= 0,
+ "race between deflation and om_release() with m=" INTPTR_FORMAT
+ ", _owner=" INTPTR_FORMAT ", ref_count=%d", p2i(m),
+ p2i(m->_owner), m->ref_count());
m->set_header(markWord::zero());
m->set_object(NULL);
m->Recycle();
omh_p->set_om_ptr(NULL);
// om_release() will reset the allocation state
@@ -1775,20 +2127,20 @@
// For async deflation:
// If a special deflation request is made, then the safepoint based
// deflation mechanism is used. Otherwise, an async deflation request
// is registered with the ServiceThread and it is notified.
-void ObjectSynchronizer::do_safepoint_work(DeflateMonitorCounters* _counters) {
+void ObjectSynchronizer::do_safepoint_work(DeflateMonitorCounters* counters) {
assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
// The per-thread in-use lists are handled in
// ParallelSPCleanupThreadClosure::do_thread().
if (!AsyncDeflateIdleMonitors || is_special_deflation_requested()) {
// Use the older mechanism for the global in-use list or if a
// special deflation has been requested before the safepoint.
- ObjectSynchronizer::deflate_idle_monitors(_counters);
+ ObjectSynchronizer::deflate_idle_monitors(counters);
return;
}
log_debug(monitorinflation)("requesting async deflation of idle monitors.");
// Request deflation of idle monitors by the ServiceThread:
@@ -1843,19 +2195,24 @@
assert(mid->object() == NULL, "invariant: object=" INTPTR_FORMAT,
p2i(mid->object()));
assert(mid->is_free(), "invariant");
// Move the deflated ObjectMonitor to the working free list
- // defined by free_head_p and free_tail_p.
+ // defined by free_head_p and free_tail_p. No races on this list
+ // so no need for load_acquire() or store_release().
if (*free_head_p == NULL) *free_head_p = mid;
if (*free_tail_p != NULL) {
// We append to the list so the caller can use mid->_next_om
// to fix the linkages in its context.
ObjectMonitor* prevtail = *free_tail_p;
// Should have been cleaned up by the caller:
- assert(prevtail->_next_om == NULL, "cleaned up deflated?");
- prevtail->_next_om = mid;
+ // Note: Should not have to mark prevtail here since we're at a
+ // safepoint and ObjectMonitors on the local free list should
+ // not be accessed in parallel.
+ assert(prevtail->_next_om == NULL, "must be NULL: _next_om="
+ INTPTR_FORMAT, p2i(prevtail->_next_om));
+ set_next(prevtail, mid);
}
*free_tail_p = mid;
// At this point, mid->_next_om still refers to its current
// value and another ObjectMonitor's _next_om field still
// refers to this ObjectMonitor. Those linkages have to be
@@ -1954,23 +2311,25 @@
p2i(mid->object()));
assert(mid->is_free(), "must be free: allocation_state=%d",
(int) mid->allocation_state());
// Move the deflated ObjectMonitor to the working free list
- // defined by free_head_p and free_tail_p.
+ // defined by free_head_p and free_tail_p. No races on this list
+ // so no need for load_acquire() or store_release().
if (*free_head_p == NULL) {
// First one on the list.
*free_head_p = mid;
}
if (*free_tail_p != NULL) {
// We append to the list so the caller can use mid->_next_om
// to fix the linkages in its context.
ObjectMonitor* prevtail = *free_tail_p;
// Should have been cleaned up by the caller:
- assert(prevtail->_next_om == NULL, "must be NULL: _next_om="
- INTPTR_FORMAT, p2i(prevtail->_next_om));
- prevtail->_next_om = mid;
+ ObjectMonitor* next = mark_next_loop(prevtail);
+ assert(unmarked_next(prevtail) == NULL, "must be NULL: _next_om="
+ INTPTR_FORMAT, p2i(unmarked_next(prevtail)));
+ set_next(prevtail, mid); // prevtail now points to mid (and is unmarked)
}
*free_tail_p = mid;
// At this point, mid->_next_om still refers to its current
// value and another ObjectMonitor's _next_om field still
@@ -2003,135 +2362,239 @@
// The owner field is no longer NULL so we lost the race since the
// ObjectMonitor is now busy.
return false;
}
-// Walk a given monitor list, and deflate idle monitors
-// The given list could be a per-thread list or a global list
-// Caller acquires gListLock as needed.
+// Walk a given monitor list, and deflate idle monitors.
+// The given list could be a per-thread list or a global list.
//
// In the case of parallel processing of thread local monitor lists,
// work is done by Threads::parallel_threads_do() which ensures that
// each Java thread is processed by exactly one worker thread, and
// thus avoid conflicts that would arise when worker threads would
// process the same monitor lists concurrently.
//
// See also ParallelSPCleanupTask and
// SafepointSynchronize::do_cleanup_tasks() in safepoint.cpp and
// Threads::parallel_java_threads_do() in thread.cpp.
-int ObjectSynchronizer::deflate_monitor_list(ObjectMonitor** list_p,
+int ObjectSynchronizer::deflate_monitor_list(ObjectMonitor* volatile * list_p,
+ int volatile * count_p,
ObjectMonitor** free_head_p,
ObjectMonitor** free_tail_p) {
- ObjectMonitor* mid;
- ObjectMonitor* next;
ObjectMonitor* cur_mid_in_use = NULL;
+ ObjectMonitor* mid = NULL;
+ ObjectMonitor* next = NULL;
int deflated_count = 0;
- for (mid = *list_p; mid != NULL;) {
+ // We use the simpler mark-mid-as-we-go protocol since there are no
+ // parallel list deletions since we are at a safepoint.
+ if (!mark_list_head(list_p, &mid, &next)) {
+ return 0; // The list is empty so nothing to deflate.
+ }
+
+ while (true) {
oop obj = (oop) mid->object();
if (obj != NULL && deflate_monitor(mid, obj, free_head_p, free_tail_p)) {
// Deflation succeeded and already updated free_head_p and
// free_tail_p as needed. Finish the move to the local free list
// by unlinking mid from the global or per-thread in-use list.
- if (mid == *list_p) {
- *list_p = mid->_next_om;
- } else if (cur_mid_in_use != NULL) {
- cur_mid_in_use->_next_om = mid->_next_om; // maintain the current thread in-use list
+ if (Atomic::cmpxchg(next, list_p, mid) != mid) {
+ // We could not switch the list head to next.
+ ADIM_guarantee(cur_mid_in_use != NULL, "must not be NULL");
+ if (Atomic::cmpxchg(next, &cur_mid_in_use->_next_om, mid) != mid) {
+ // deflate_monitor_list() is called at a safepoint so the
+ // global or per-thread in-use list should not be modified
+ // in parallel so we:
+ fatal("mid=" INTPTR_FORMAT " must be referred to by the list head: "
+ "list_p=" INTPTR_FORMAT " or by cur_mid_in_use's next field: "
+ "cur_mid_in_use=" INTPTR_FORMAT ", next_om=" INTPTR_FORMAT,
+ p2i(mid), p2i((ObjectMonitor**)list_p), p2i(cur_mid_in_use),
+ p2i(cur_mid_in_use->_next_om));
}
- next = mid->_next_om;
- mid->_next_om = NULL; // This mid is current tail in the free_head_p list
- mid = next;
+ }
+ // At this point mid is disconnected from the in-use list so
+ // its marked next field no longer has any effects.
deflated_count++;
+ Atomic::dec(count_p);
+ chk_for_list_loop(OrderAccess::load_acquire(list_p),
+ OrderAccess::load_acquire(count_p));
+ chk_om_not_on_list(mid, OrderAccess::load_acquire(list_p),
+ OrderAccess::load_acquire(count_p));
+ // mid is current tail in the free_head_p list so NULL terminate it
+ // (which also unmarks it):
+ set_next(mid, NULL);
+
+ // All the list management is done so move on to the next one:
+ mid = next;
} else {
+ set_next(mid, next); // unmark next field
+
+ // All the list management is done so move on to the next one:
cur_mid_in_use = mid;
- mid = mid->_next_om;
+ mid = next;
+ }
+ if (mid == NULL) {
+ break; // Reached end of the list so nothing more to deflate.
}
+ // Mark mid's next field so we can possibly deflate it:
+ next = mark_next_loop(mid);
}
return deflated_count;
}
// Walk a given ObjectMonitor list and deflate idle ObjectMonitors using
// a JavaThread. Returns the number of deflated ObjectMonitors. The given
// list could be a per-thread in-use list or the global in-use list.
-// Caller acquires gListLock as appropriate. If a safepoint has started,
-// then we save state via saved_mid_in_use_p and return to the caller to
-// honor the safepoint.
+// If a safepoint has started, then we save state via saved_mid_in_use_p
+// and return to the caller to honor the safepoint.
//
-int ObjectSynchronizer::deflate_monitor_list_using_JT(ObjectMonitor** list_p,
+int ObjectSynchronizer::deflate_monitor_list_using_JT(ObjectMonitor* volatile * list_p,
+ int volatile * count_p,
ObjectMonitor** free_head_p,
ObjectMonitor** free_tail_p,
ObjectMonitor** saved_mid_in_use_p) {
assert(AsyncDeflateIdleMonitors, "sanity check");
assert(Thread::current()->is_Java_thread(), "precondition");
- ObjectMonitor* mid;
- ObjectMonitor* next;
ObjectMonitor* cur_mid_in_use = NULL;
+ ObjectMonitor* mid = NULL;
+ ObjectMonitor* next = NULL;
+ ObjectMonitor* next_next = NULL;
int deflated_count = 0;
+ // We use the more complicated mark-cur_mid_in_use-and-mid-as-we-go
+ // protocol because om_release() can do list deletions in parallel.
+ // We also mark-next-next-as-we-go to prevent an om_flush() that is
+ // behind this thread from passing us.
if (*saved_mid_in_use_p == NULL) {
// No saved state so start at the beginning.
- mid = *list_p;
+ // Mark the list head's next field so we can possibly deflate it:
+ if (!mark_list_head(list_p, &mid, &next)) {
+ return 0; // The list is empty so nothing to deflate.
+ }
} else {
// We're restarting after a safepoint so restore the necessary state
// before we resume.
cur_mid_in_use = *saved_mid_in_use_p;
- mid = cur_mid_in_use->_next_om;
+ // Mark cur_mid_in_use's next field so we can possibly update its
+ // next field to extract a deflated ObjectMonitor.
+ mid = mark_next_loop(cur_mid_in_use);
+ if (mid == NULL) {
+ set_next(cur_mid_in_use, NULL); // unmark next field
+ *saved_mid_in_use_p = NULL;
+ return 0; // The remainder is empty so nothing more to deflate.
+ }
+ // Mark mid's next field so we can possibly deflate it:
+ next = mark_next_loop(mid);
}
- while (mid != NULL) {
+
+ while (true) {
+ // The current mid's next field is marked at this point. If we have
+ // a cur_mid_in_use, then its next field is also marked at this point.
+
+ if (next != NULL) {
+ // We mark the next -> next field so that an om_flush()
+ // thread that is behind us cannot pass us when we
+ // unmark the current mid's next field.
+ next_next = mark_next_loop(next);
+ }
+
// Only try to deflate if there is an associated Java object and if
// mid is old (is not newly allocated and is not newly freed).
if (mid->object() != NULL && mid->is_old() &&
deflate_monitor_using_JT(mid, free_head_p, free_tail_p)) {
- // Deflation succeeded so update the in-use list.
- if (mid == *list_p) {
- *list_p = mid->_next_om;
- } else if (cur_mid_in_use != NULL) {
- // Maintain the current in-use list.
- cur_mid_in_use->_next_om = mid->_next_om;
- }
- next = mid->_next_om;
- mid->_next_om = NULL;
- // At this point mid is disconnected from the in-use list
- // and is the current tail in the free_head_p list.
- mid = next;
+ // Deflation succeeded and already updated free_head_p and
+ // free_tail_p as needed. Finish the move to the local free list
+ // by unlinking mid from the global or per-thread in-use list.
+ if (Atomic::cmpxchg(next, list_p, mid) != mid) {
+ // We could not switch the list head to next.
+ ObjectMonitor* marked_mid = mark_om_ptr(mid);
+ ObjectMonitor* marked_next = mark_om_ptr(next);
+ // Switch cur_mid_in_use's next field to marked next:
+ ADIM_guarantee(cur_mid_in_use != NULL, "must not be NULL");
+ if (Atomic::cmpxchg(marked_next, &cur_mid_in_use->_next_om,
+ marked_mid) != marked_mid) {
+ // We could not switch cur_mid_in_use's next field. This
+ // should not be possible since it was marked so we:
+ fatal("mid=" INTPTR_FORMAT " must be referred to by the list head: "
+ "&list_p=" INTPTR_FORMAT " or by cur_mid_in_use's next field: "
+ "cur_mid_in_use=" INTPTR_FORMAT ", next_om=" INTPTR_FORMAT,
+ p2i(mid), p2i((ObjectMonitor**)list_p), p2i(cur_mid_in_use),
+ p2i(cur_mid_in_use->_next_om));
+ }
+ }
+ // At this point mid is disconnected from the in-use list so
+ // its marked next field no longer has any effects.
deflated_count++;
+ Atomic::dec(count_p);
+ chk_for_list_loop(OrderAccess::load_acquire(list_p),
+ OrderAccess::load_acquire(count_p));
+ chk_om_not_on_list(mid, OrderAccess::load_acquire(list_p),
+ OrderAccess::load_acquire(count_p));
+ // mid is current tail in the free_head_p list so NULL terminate it
+ // (which also unmarks it):
+ set_next(mid, NULL);
+
+ // All the list management is done so move on to the next one:
+ mid = next; // mid keeps non-NULL next's marked next field
+ next = next_next;
} else {
// mid is considered in-use if it does not have an associated
// Java object or mid is not old or deflation did not succeed.
// A mid->is_new() node can be seen here when it is freshly
// returned by om_alloc() (and skips the deflation code path).
// A mid->is_old() node can be seen here when deflation failed.
// A mid->is_free() node can be seen here when a fresh node from
// om_alloc() is released by om_release() due to losing the race
// in inflate().
+ // All the list management is done so move on to the next one:
+ if (cur_mid_in_use != NULL) {
+ set_next(cur_mid_in_use, mid); // umark cur_mid_in_use
+ }
+ // The next cur_mid_in_use keeps mid's marked next field so
+ // that it is stable for a possible next field change. It
+ // cannot be modified by om_release() while it is marked.
cur_mid_in_use = mid;
- mid = mid->_next_om;
+ mid = next; // mid keeps non-NULL next's marked next field
+ next = next_next;
if (SafepointSynchronize::is_synchronizing() &&
- cur_mid_in_use != *list_p && cur_mid_in_use->is_old()) {
+ cur_mid_in_use != OrderAccess::load_acquire(list_p) &&
+ cur_mid_in_use->is_old()) {
// If a safepoint has started and cur_mid_in_use is not the list
// head and is old, then it is safe to use as saved state. Return
- // to the caller so gListLock can be dropped as appropriate
- // before blocking.
+ // to the caller before blocking.
*saved_mid_in_use_p = cur_mid_in_use;
+ set_next(cur_mid_in_use, mid); // umark cur_mid_in_use
+ if (mid != NULL) {
+ set_next(mid, next); // umark mid
+ }
return deflated_count;
}
}
+ if (mid == NULL) {
+ if (cur_mid_in_use != NULL) {
+ set_next(cur_mid_in_use, mid); // umark cur_mid_in_use
+ }
+ break; // Reached end of the list so nothing more to deflate.
+ }
+
+ // The current mid's next field is marked at this point. If we have
+ // a cur_mid_in_use, then its next field is also marked at this point.
}
// We finished the list without a safepoint starting so there's
// no need to save state.
*saved_mid_in_use_p = NULL;
return deflated_count;
}
void ObjectSynchronizer::prepare_deflate_idle_monitors(DeflateMonitorCounters* counters) {
- counters->n_in_use = 0; // currently associated with objects
- counters->n_in_circulation = 0; // extant
- counters->n_scavenged = 0; // reclaimed (global and per-thread)
- counters->per_thread_scavenged = 0; // per-thread scavenge total
+ OrderAccess::release_store(&counters->n_in_use, 0); // currently associated with objects
+ OrderAccess::release_store(&counters->n_in_circulation, 0); // extant
+ OrderAccess::release_store(&counters->n_scavenged, 0); // reclaimed (global and per-thread)
+ OrderAccess::release_store(&counters->per_thread_scavenged, 0); // per-thread scavenge total
counters->per_thread_times = 0.0; // per-thread scavenge times
}
void ObjectSynchronizer::deflate_idle_monitors(DeflateMonitorCounters* counters) {
assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
@@ -2152,37 +2615,32 @@
if (log_is_enabled(Info, monitorinflation)) {
timer.start();
}
- // Prevent om_flush from changing mids in Thread dtor's during deflation
- // And in case the vm thread is acquiring a lock during a safepoint
- // See e.g. 6320749
- Thread::muxAcquire(&gListLock, "deflate_idle_monitors");
-
// Note: the thread-local monitors lists get deflated in
// a separate pass. See deflate_thread_local_monitors().
// For moribund threads, scan g_om_in_use_list
int deflated_count = 0;
- if (g_om_in_use_list) {
- counters->n_in_circulation += g_om_in_use_count;
- deflated_count = deflate_monitor_list((ObjectMonitor **)&g_om_in_use_list, &free_head_p, &free_tail_p);
- g_om_in_use_count -= deflated_count;
- counters->n_scavenged += deflated_count;
- counters->n_in_use += g_om_in_use_count;
+ if (OrderAccess::load_acquire(&g_om_in_use_list) != NULL) {
+ // Update n_in_circulation before g_om_in_use_count is updated by deflation.
+ Atomic::add(OrderAccess::load_acquire(&g_om_in_use_count), &counters->n_in_circulation);
+
+ deflated_count = deflate_monitor_list(&g_om_in_use_list, &g_om_in_use_count, &free_head_p, &free_tail_p);
+ Atomic::add(OrderAccess::load_acquire(&g_om_in_use_count), &counters->n_in_use);
}
if (free_head_p != NULL) {
// Move the deflated ObjectMonitors back to the global free list.
- guarantee(free_tail_p != NULL && counters->n_scavenged > 0, "invariant");
- assert(free_tail_p->_next_om == NULL, "invariant");
- // constant-time list splice - prepend scavenged segment to g_free_list
- free_tail_p->_next_om = g_free_list;
- g_free_list = free_head_p;
+ // No races on the working free list so no need for load_acquire().
+ guarantee(free_tail_p != NULL && deflated_count > 0, "invariant");
+ assert(free_tail_p->_next_om == NULL, "must be NULL: _next_om="
+ INTPTR_FORMAT, p2i(free_tail_p->_next_om));
+ prepend_list_to_g_free_list(free_head_p, free_tail_p, deflated_count);
+ Atomic::add(deflated_count, &counters->n_scavenged);
}
- Thread::muxRelease(&gListLock);
timer.stop();
LogStreamHandle(Debug, monitorinflation) lsh_debug;
LogStreamHandle(Info, monitorinflation) lsh_info;
LogStream* ls = NULL;
@@ -2204,25 +2662,24 @@
JavaThread* self = JavaThread::current();
deflate_common_idle_monitors_using_JT(true /* is_global */, self);
}
-// Deflate per-thread idle ObjectMonitors using a JavaThread.
+// Deflate the specified JavaThread's idle ObjectMonitors using a JavaThread.
//
-void ObjectSynchronizer::deflate_per_thread_idle_monitors_using_JT() {
+void ObjectSynchronizer::deflate_per_thread_idle_monitors_using_JT(JavaThread* target) {
assert(AsyncDeflateIdleMonitors, "sanity check");
assert(Thread::current()->is_Java_thread(), "precondition");
- JavaThread* self = JavaThread::current();
-
- self->om_request_deflation = false;
- deflate_common_idle_monitors_using_JT(false /* !is_global */, self);
+ deflate_common_idle_monitors_using_JT(false /* !is_global */, target);
}
// Deflate global or per-thread idle ObjectMonitors using a JavaThread.
//
-void ObjectSynchronizer::deflate_common_idle_monitors_using_JT(bool is_global, JavaThread* self) {
+void ObjectSynchronizer::deflate_common_idle_monitors_using_JT(bool is_global, JavaThread* target) {
+ JavaThread* self = JavaThread::current();
+
int deflated_count = 0;
ObjectMonitor* free_head_p = NULL; // Local SLL of scavenged ObjectMonitors
ObjectMonitor* free_tail_p = NULL;
ObjectMonitor* saved_mid_in_use_p = NULL;
elapsedTimer timer;
@@ -2230,75 +2687,64 @@
if (log_is_enabled(Info, monitorinflation)) {
timer.start();
}
if (is_global) {
- Thread::muxAcquire(&gListLock, "deflate_global_idle_monitors_using_JT(1)");
- OM_PERFDATA_OP(MonExtant, set_value(g_om_in_use_count));
+ OM_PERFDATA_OP(MonExtant, set_value(OrderAccess::load_acquire(&g_om_in_use_count)));
} else {
- OM_PERFDATA_OP(MonExtant, inc(self->om_in_use_count));
+ OM_PERFDATA_OP(MonExtant, inc(OrderAccess::load_acquire(&target->om_in_use_count)));
}
do {
int local_deflated_count;
if (is_global) {
- local_deflated_count = deflate_monitor_list_using_JT((ObjectMonitor **)&g_om_in_use_list, &free_head_p, &free_tail_p, &saved_mid_in_use_p);
- g_om_in_use_count -= local_deflated_count;
+ local_deflated_count = deflate_monitor_list_using_JT(&g_om_in_use_list, &g_om_in_use_count, &free_head_p, &free_tail_p, &saved_mid_in_use_p);
} else {
- local_deflated_count = deflate_monitor_list_using_JT(self->om_in_use_list_addr(), &free_head_p, &free_tail_p, &saved_mid_in_use_p);
- self->om_in_use_count -= local_deflated_count;
+ local_deflated_count = deflate_monitor_list_using_JT(&target->om_in_use_list, &target->om_in_use_count, &free_head_p, &free_tail_p, &saved_mid_in_use_p);
}
deflated_count += local_deflated_count;
if (free_head_p != NULL) {
- // Move the scavenged ObjectMonitors to the global free list.
+ // Move the deflated ObjectMonitors to the global free list.
+ // No races on the working list so no need for load_acquire().
guarantee(free_tail_p != NULL && local_deflated_count > 0, "free_tail_p=" INTPTR_FORMAT ", local_deflated_count=%d", p2i(free_tail_p), local_deflated_count);
- assert(free_tail_p->_next_om == NULL, "invariant");
+ // Note: The target thread can be doing an om_alloc() that
+ // is trying to prepend an ObjectMonitor on its in-use list
+ // at the same time that we have deflated the current in-use
+ // list head and put it on the local free list. prepend_to_common()
+ // will detect the race and retry which avoids list corruption,
+ // but the next field in free_tail_p can flicker to marked
+ // and then unmarked while prepend_to_common() is sorting it
+ // all out.
+ assert(unmarked_next(free_tail_p) == NULL, "must be NULL: _next_om="
+ INTPTR_FORMAT, p2i(unmarked_next(free_tail_p)));
- if (!is_global) {
- Thread::muxAcquire(&gListLock, "deflate_per_thread_idle_monitors_using_JT(2)");
- }
- // Constant-time list splice - prepend scavenged segment to g_free_list.
- free_tail_p->_next_om = g_free_list;
- g_free_list = free_head_p;
+ prepend_list_to_g_free_list(free_head_p, free_tail_p, local_deflated_count);
- g_om_free_count += local_deflated_count;
OM_PERFDATA_OP(Deflations, inc(local_deflated_count));
- if (!is_global) {
- Thread::muxRelease(&gListLock);
- }
}
if (saved_mid_in_use_p != NULL) {
// deflate_monitor_list_using_JT() detected a safepoint starting.
- if (is_global) {
- Thread::muxRelease(&gListLock);
- }
timer.stop();
{
if (is_global) {
log_debug(monitorinflation)("pausing deflation of global idle monitors for a safepoint.");
} else {
- log_debug(monitorinflation)("jt=" INTPTR_FORMAT ": pausing deflation of per-thread idle monitors for a safepoint.", p2i(self));
+ log_debug(monitorinflation)("jt=" INTPTR_FORMAT ": pausing deflation of per-thread idle monitors for a safepoint.", p2i(target));
}
assert(SafepointSynchronize::is_synchronizing(), "sanity check");
ThreadBlockInVM blocker(self);
}
// Prepare for another loop after the safepoint.
free_head_p = NULL;
free_tail_p = NULL;
if (log_is_enabled(Info, monitorinflation)) {
timer.start();
}
- if (is_global) {
- Thread::muxAcquire(&gListLock, "deflate_global_idle_monitors_using_JT(3)");
- }
}
} while (saved_mid_in_use_p != NULL);
- if (is_global) {
- Thread::muxRelease(&gListLock);
- }
timer.stop();
LogStreamHandle(Debug, monitorinflation) lsh_debug;
LogStreamHandle(Info, monitorinflation) lsh_info;
LogStream* ls = NULL;
@@ -2309,11 +2755,11 @@
}
if (ls != NULL) {
if (is_global) {
ls->print_cr("async-deflating global idle monitors, %3.7f secs, %d monitors", timer.seconds(), deflated_count);
} else {
- ls->print_cr("jt=" INTPTR_FORMAT ": async-deflating per-thread idle monitors, %3.7f secs, %d monitors", p2i(self), timer.seconds(), deflated_count);
+ ls->print_cr("jt=" INTPTR_FORMAT ": async-deflating per-thread idle monitors, %3.7f secs, %d monitors", p2i(target), timer.seconds(), deflated_count);
}
}
}
void ObjectSynchronizer::finish_deflate_idle_monitors(DeflateMonitorCounters* counters) {
@@ -2321,33 +2767,33 @@
// monitors. Note: if the work is split among more than one
// worker thread, then the reported time will likely be more
// than a beginning to end measurement of the phase.
// Note: AsyncDeflateIdleMonitors only deflates per-thread idle
// monitors at a safepoint when a special deflation has been requested.
- log_info(safepoint, cleanup)("deflating per-thread idle monitors, %3.7f secs, monitors=%d", counters->per_thread_times, counters->per_thread_scavenged);
+ log_info(safepoint, cleanup)("deflating per-thread idle monitors, %3.7f secs, monitors=%d",
+ counters->per_thread_times,
+ OrderAccess::load_acquire(&counters->per_thread_scavenged));
bool needs_special_deflation = is_special_deflation_requested();
if (!AsyncDeflateIdleMonitors || needs_special_deflation) {
// AsyncDeflateIdleMonitors does not use these counters unless
// there is a special deflation request.
- g_om_free_count += counters->n_scavenged;
-
OM_PERFDATA_OP(Deflations, inc(counters->n_scavenged));
OM_PERFDATA_OP(MonExtant, set_value(counters->n_in_circulation));
}
if (log_is_enabled(Debug, monitorinflation)) {
// exit_globals()'s call to audit_and_print_stats() is done
// at the Info level.
ObjectSynchronizer::audit_and_print_stats(false /* on_exit */);
} else if (log_is_enabled(Info, monitorinflation)) {
- Thread::muxAcquire(&gListLock, "finish_deflate_idle_monitors");
log_info(monitorinflation)("g_om_population=%d, g_om_in_use_count=%d, "
- "g_om_free_count=%d", g_om_population,
- g_om_in_use_count, g_om_free_count);
- Thread::muxRelease(&gListLock);
+ "g_om_free_count=%d",
+ OrderAccess::load_acquire(&g_om_population),
+ OrderAccess::load_acquire(&g_om_in_use_count),
+ OrderAccess::load_acquire(&g_om_free_count));
}
ForceMonitorScavenge = 0; // Reset
GVars.stw_random = os::random();
GVars.stw_cycle++;
@@ -2357,60 +2803,47 @@
}
void ObjectSynchronizer::deflate_thread_local_monitors(Thread* thread, DeflateMonitorCounters* counters) {
assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
- if (AsyncDeflateIdleMonitors) {
- if (!is_special_deflation_requested()) {
- // Mark the JavaThread for idle monitor deflation if a special
- // deflation has NOT been requested.
- if (thread->om_in_use_count > 0) {
- // This JavaThread is using monitors so mark it.
- thread->om_request_deflation = true;
- }
+ if (AsyncDeflateIdleMonitors && !is_special_deflation_requested()) {
+ // Nothing to do if a special deflation has NOT been requested.
return;
}
- }
ObjectMonitor* free_head_p = NULL; // Local SLL of scavenged monitors
ObjectMonitor* free_tail_p = NULL;
elapsedTimer timer;
if (log_is_enabled(Info, safepoint, cleanup) ||
log_is_enabled(Info, monitorinflation)) {
timer.start();
}
- int deflated_count = deflate_monitor_list(thread->om_in_use_list_addr(), &free_head_p, &free_tail_p);
-
- Thread::muxAcquire(&gListLock, "deflate_thread_local_monitors");
+ // Update n_in_circulation before om_in_use_count is updated by deflation.
+ Atomic::add(OrderAccess::load_acquire(&thread->om_in_use_count), &counters->n_in_circulation);
- // Adjust counters
- counters->n_in_circulation += thread->om_in_use_count;
- thread->om_in_use_count -= deflated_count;
- counters->n_scavenged += deflated_count;
- counters->n_in_use += thread->om_in_use_count;
- counters->per_thread_scavenged += deflated_count;
+ int deflated_count = deflate_monitor_list(&thread->om_in_use_list, &thread->om_in_use_count, &free_head_p, &free_tail_p);
+ Atomic::add(OrderAccess::load_acquire(&thread->om_in_use_count), &counters->n_in_use);
if (free_head_p != NULL) {
// Move the deflated ObjectMonitors back to the global free list.
+ // No races on the working list so no need for load_acquire().
guarantee(free_tail_p != NULL && deflated_count > 0, "invariant");
- assert(free_tail_p->_next_om == NULL, "invariant");
-
- // constant-time list splice - prepend scavenged segment to g_free_list
- free_tail_p->_next_om = g_free_list;
- g_free_list = free_head_p;
+ assert(free_tail_p->_next_om == NULL, "must be NULL: _next_om="
+ INTPTR_FORMAT, p2i(free_tail_p->_next_om));
+ prepend_list_to_g_free_list(free_head_p, free_tail_p, deflated_count);
+ Atomic::add(deflated_count, &counters->n_scavenged);
+ Atomic::add(deflated_count, &counters->per_thread_scavenged);
}
timer.stop();
// Safepoint logging cares about cumulative per_thread_times and
// we'll capture most of the cost, but not the muxRelease() which
// should be cheap.
counters->per_thread_times += timer.seconds();
- Thread::muxRelease(&gListLock);
-
LogStreamHandle(Debug, monitorinflation) lsh_debug;
LogStreamHandle(Info, monitorinflation) lsh_info;
LogStream* ls = NULL;
if (log_is_enabled(Debug, monitorinflation)) {
ls = &lsh_debug;
@@ -2457,13 +2890,11 @@
void ObjectSynchronizer::release_monitors_owned_by_thread(TRAPS) {
assert(THREAD == JavaThread::current(), "must be current Java thread");
NoSafepointVerifier nsv;
ReleaseJavaMonitorsClosure rjmc(THREAD);
- Thread::muxAcquire(&gListLock, "release_monitors_owned_by_thread");
ObjectSynchronizer::monitors_iterate(&rjmc);
- Thread::muxRelease(&gListLock);
THREAD->clear_pending_exception();
}
const char* ObjectSynchronizer::inflate_cause_name(const InflateCause cause) {
switch (cause) {
@@ -2513,42 +2944,35 @@
} else if (log_is_enabled(Info, monitorinflation)) {
ls = &lsh_info;
}
assert(ls != NULL, "sanity check");
- if (!on_exit) {
- // Not at VM exit so grab the global list lock.
- Thread::muxAcquire(&gListLock, "audit_and_print_stats");
- }
-
// Log counts for the global and per-thread monitor lists:
int chk_om_population = log_monitor_list_counts(ls);
int error_cnt = 0;
ls->print_cr("Checking global lists:");
// Check g_om_population:
- if (g_om_population == chk_om_population) {
+ if (OrderAccess::load_acquire(&g_om_population) == chk_om_population) {
ls->print_cr("g_om_population=%d equals chk_om_population=%d",
- g_om_population, chk_om_population);
+ OrderAccess::load_acquire(&g_om_population),
+ chk_om_population);
} else {
ls->print_cr("ERROR: g_om_population=%d is not equal to "
- "chk_om_population=%d", g_om_population,
+ "chk_om_population=%d",
+ OrderAccess::load_acquire(&g_om_population),
chk_om_population);
error_cnt++;
}
// Check g_om_in_use_list and g_om_in_use_count:
chk_global_in_use_list_and_count(ls, &error_cnt);
// Check g_free_list and g_om_free_count:
chk_global_free_list_and_count(ls, &error_cnt);
- if (!on_exit) {
- Thread::muxRelease(&gListLock);
- }
-
ls->print_cr("Checking per-thread lists:");
for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
// Check om_in_use_list and om_in_use_count:
chk_per_thread_in_use_list_and_count(jt, ls, &error_cnt);
@@ -2566,11 +2990,11 @@
if ((on_exit && log_is_enabled(Info, monitorinflation)) ||
(!on_exit && log_is_enabled(Trace, monitorinflation))) {
// When exiting this log output is at the Info level. When called
// at a safepoint, this log output is at the Trace level since
// there can be a lot of it.
- log_in_use_monitor_details(ls, on_exit);
+ log_in_use_monitor_details(ls);
}
ls->flush();
guarantee(error_cnt == 0, "ERROR: found monitor list errors: error_cnt=%d", error_cnt);
@@ -2622,39 +3046,46 @@
// Check the global free list and count; log the results of the checks.
void ObjectSynchronizer::chk_global_free_list_and_count(outputStream * out,
int *error_cnt_p) {
int chk_om_free_count = 0;
- for (ObjectMonitor* n = g_free_list; n != NULL; n = n->_next_om) {
+ for (ObjectMonitor* n = OrderAccess::load_acquire(&g_free_list); n != NULL; n = unmarked_next(n)) {
chk_free_entry(NULL /* jt */, n, out, error_cnt_p);
chk_om_free_count++;
}
- if (g_om_free_count == chk_om_free_count) {
+ if (OrderAccess::load_acquire(&g_om_free_count) == chk_om_free_count) {
out->print_cr("g_om_free_count=%d equals chk_om_free_count=%d",
- g_om_free_count, chk_om_free_count);
+ OrderAccess::load_acquire(&g_om_free_count),
+ chk_om_free_count);
} else {
- out->print_cr("ERROR: g_om_free_count=%d is not equal to "
- "chk_om_free_count=%d", g_om_free_count,
+ // With lock free access to g_free_list, it is possible for an
+ // ObjectMonitor to be prepended to g_free_list after we started
+ // calculating chk_om_free_count so g_om_free_count may not
+ // match anymore.
+ out->print_cr("WARNING: g_om_free_count=%d is not equal to "
+ "chk_om_free_count=%d",
+ OrderAccess::load_acquire(&g_om_free_count),
chk_om_free_count);
- *error_cnt_p = *error_cnt_p + 1;
}
}
// Check the global in-use list and count; log the results of the checks.
void ObjectSynchronizer::chk_global_in_use_list_and_count(outputStream * out,
int *error_cnt_p) {
int chk_om_in_use_count = 0;
- for (ObjectMonitor* n = g_om_in_use_list; n != NULL; n = n->_next_om) {
+ for (ObjectMonitor* n = OrderAccess::load_acquire(&g_om_in_use_list); n != NULL; n = unmarked_next(n)) {
chk_in_use_entry(NULL /* jt */, n, out, error_cnt_p);
chk_om_in_use_count++;
}
- if (g_om_in_use_count == chk_om_in_use_count) {
- out->print_cr("g_om_in_use_count=%d equals chk_om_in_use_count=%d", g_om_in_use_count,
+ if (OrderAccess::load_acquire(&g_om_in_use_count) == chk_om_in_use_count) {
+ out->print_cr("g_om_in_use_count=%d equals chk_om_in_use_count=%d",
+ OrderAccess::load_acquire(&g_om_in_use_count),
chk_om_in_use_count);
} else {
out->print_cr("ERROR: g_om_in_use_count=%d is not equal to chk_om_in_use_count=%d",
- g_om_in_use_count, chk_om_in_use_count);
+ OrderAccess::load_acquire(&g_om_in_use_count),
+ chk_om_in_use_count);
*error_cnt_p = *error_cnt_p + 1;
}
}
// Check an in-use monitor entry; log any errors.
@@ -2719,64 +3150,63 @@
// Check the thread's free list and count; log the results of the checks.
void ObjectSynchronizer::chk_per_thread_free_list_and_count(JavaThread *jt,
outputStream * out,
int *error_cnt_p) {
int chk_om_free_count = 0;
- for (ObjectMonitor* n = jt->om_free_list; n != NULL; n = n->_next_om) {
+ for (ObjectMonitor* n = OrderAccess::load_acquire(&jt->om_free_list); n != NULL; n = unmarked_next(n)) {
chk_free_entry(jt, n, out, error_cnt_p);
chk_om_free_count++;
}
- if (jt->om_free_count == chk_om_free_count) {
+ if (OrderAccess::load_acquire(&jt->om_free_count) == chk_om_free_count) {
out->print_cr("jt=" INTPTR_FORMAT ": om_free_count=%d equals "
- "chk_om_free_count=%d", p2i(jt), jt->om_free_count, chk_om_free_count);
+ "chk_om_free_count=%d", p2i(jt),
+ OrderAccess::load_acquire(&jt->om_free_count),
+ chk_om_free_count);
} else {
out->print_cr("ERROR: jt=" INTPTR_FORMAT ": om_free_count=%d is not "
- "equal to chk_om_free_count=%d", p2i(jt), jt->om_free_count,
+ "equal to chk_om_free_count=%d", p2i(jt),
+ OrderAccess::load_acquire(&jt->om_free_count),
chk_om_free_count);
*error_cnt_p = *error_cnt_p + 1;
}
}
// Check the thread's in-use list and count; log the results of the checks.
void ObjectSynchronizer::chk_per_thread_in_use_list_and_count(JavaThread *jt,
outputStream * out,
int *error_cnt_p) {
int chk_om_in_use_count = 0;
- for (ObjectMonitor* n = jt->om_in_use_list; n != NULL; n = n->_next_om) {
+ for (ObjectMonitor* n = OrderAccess::load_acquire(&jt->om_in_use_list); n != NULL; n = unmarked_next(n)) {
chk_in_use_entry(jt, n, out, error_cnt_p);
chk_om_in_use_count++;
}
- if (jt->om_in_use_count == chk_om_in_use_count) {
+ if (OrderAccess::load_acquire(&jt->om_in_use_count) == chk_om_in_use_count) {
out->print_cr("jt=" INTPTR_FORMAT ": om_in_use_count=%d equals "
- "chk_om_in_use_count=%d", p2i(jt), jt->om_in_use_count,
+ "chk_om_in_use_count=%d", p2i(jt),
+ OrderAccess::load_acquire(&jt->om_in_use_count),
chk_om_in_use_count);
} else {
out->print_cr("ERROR: jt=" INTPTR_FORMAT ": om_in_use_count=%d is not "
- "equal to chk_om_in_use_count=%d", p2i(jt), jt->om_in_use_count,
+ "equal to chk_om_in_use_count=%d", p2i(jt),
+ OrderAccess::load_acquire(&jt->om_in_use_count),
chk_om_in_use_count);
*error_cnt_p = *error_cnt_p + 1;
}
}
// Log details about ObjectMonitors on the in-use lists. The 'BHL'
// flags indicate why the entry is in-use, 'object' and 'object type'
// indicate the associated object and its type.
-void ObjectSynchronizer::log_in_use_monitor_details(outputStream * out,
- bool on_exit) {
- if (!on_exit) {
- // Not at VM exit so grab the global list lock.
- Thread::muxAcquire(&gListLock, "log_in_use_monitor_details");
- }
-
+void ObjectSynchronizer::log_in_use_monitor_details(outputStream * out) {
stringStream ss;
- if (g_om_in_use_count > 0) {
+ if (OrderAccess::load_acquire(&g_om_in_use_count) > 0) {
out->print_cr("In-use global monitor info:");
out->print_cr("(B -> is_busy, H -> has hash code, L -> lock status)");
out->print_cr("%18s %s %7s %18s %18s",
"monitor", "BHL", "ref_cnt", "object", "object type");
out->print_cr("================== === ======= ================== ==================");
- for (ObjectMonitor* n = g_om_in_use_list; n != NULL; n = n->_next_om) {
+ for (ObjectMonitor* n = OrderAccess::load_acquire(&g_om_in_use_list); n != NULL; n = unmarked_next(n)) {
const oop obj = (oop) n->object();
const markWord mark = n->header();
ResourceMark rm;
out->print(INTPTR_FORMAT " %d%d%d %7d " INTPTR_FORMAT " %s",
p2i(n), n->is_busy() != 0, mark.hash() != 0,
@@ -2788,21 +3218,17 @@
}
out->cr();
}
}
- if (!on_exit) {
- Thread::muxRelease(&gListLock);
- }
-
out->print_cr("In-use per-thread monitor info:");
out->print_cr("(B -> is_busy, H -> has hash code, L -> lock status)");
out->print_cr("%18s %18s %s %7s %18s %18s",
"jt", "monitor", "BHL", "ref_cnt", "object", "object type");
out->print_cr("================== ================== === ======= ================== ==================");
for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
- for (ObjectMonitor* n = jt->om_in_use_list; n != NULL; n = n->_next_om) {
+ for (ObjectMonitor* n = OrderAccess::load_acquire(&jt->om_in_use_list); n != NULL; n = unmarked_next(n)) {
const oop obj = (oop) n->object();
const markWord mark = n->header();
ResourceMark rm;
out->print(INTPTR_FORMAT " " INTPTR_FORMAT " %d%d%d %7d "
INTPTR_FORMAT " %s", p2i(jt), p2i(n), n->is_busy() != 0,
@@ -2825,21 +3251,27 @@
int pop_count = 0;
out->print_cr("%18s %10s %10s %10s",
"Global Lists:", "InUse", "Free", "Total");
out->print_cr("================== ========== ========== ==========");
out->print_cr("%18s %10d %10d %10d", "",
- g_om_in_use_count, g_om_free_count, g_om_population);
- pop_count += g_om_in_use_count + g_om_free_count;
+ OrderAccess::load_acquire(&g_om_in_use_count),
+ OrderAccess::load_acquire(&g_om_free_count),
+ OrderAccess::load_acquire(&g_om_population));
+ pop_count += OrderAccess::load_acquire(&g_om_in_use_count) +
+ OrderAccess::load_acquire(&g_om_free_count);
out->print_cr("%18s %10s %10s %10s",
"Per-Thread Lists:", "InUse", "Free", "Provision");
out->print_cr("================== ========== ========== ==========");
for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
out->print_cr(INTPTR_FORMAT " %10d %10d %10d", p2i(jt),
- jt->om_in_use_count, jt->om_free_count, jt->om_free_provision);
- pop_count += jt->om_in_use_count + jt->om_free_count;
+ OrderAccess::load_acquire(&jt->om_in_use_count),
+ OrderAccess::load_acquire(&jt->om_free_count),
+ jt->om_free_provision);
+ pop_count += OrderAccess::load_acquire(&jt->om_in_use_count) +
+ OrderAccess::load_acquire(&jt->om_free_count);
}
return pop_count;
}
#ifndef PRODUCT
@@ -2857,11 +3289,12 @@
address blk = (address)block;
size_t diff = mon - blk;
assert((diff % sizeof(PaddedObjectMonitor)) == 0, "must be aligned");
return 1;
}
- block = (PaddedObjectMonitor*)block->_next_om;
+ // unmarked_next() is not needed with g_block_list (no next field marking).
+ block = (PaddedObjectMonitor*)OrderAccess::load_acquire(&block->_next_om);
}
return 0;
}
#endif
< prev index next >