提交 2d1fc530 编写于 作者: M mgerdin

8014555: G1: Memory ordering problem with Conc refinement and card marking

Summary: Add a StoreLoad barrier in the G1 post-barrier to fix a race with concurrent refinement. Also-reviewed-by: martin.doerr@sap.com
Reviewed-by: iveresov, tschatzl, brutisso, roland, kvn
上级 4c3396ec
......@@ -37,6 +37,9 @@
#include "runtime/vframeArray.hpp"
#include "utilities/macros.hpp"
#include "vmreg_sparc.inline.hpp"
#if INCLUDE_ALL_GCS
#include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
#endif
// Implementation of StubAssembler
......@@ -912,7 +915,7 @@ OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
Register tmp2 = G3_scratch;
jbyte* byte_map_base = ((CardTableModRefBS*)bs)->byte_map_base;
Label not_already_dirty, restart, refill;
Label not_already_dirty, restart, refill, young_card;
#ifdef _LP64
__ srlx(addr, CardTableModRefBS::card_shift, addr);
......@@ -924,9 +927,15 @@ OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
__ set(rs, cardtable); // cardtable := <card table base>
__ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]
__ cmp_and_br_short(tmp, G1SATBCardTableModRefBS::g1_young_card_val(), Assembler::equal, Assembler::pt, young_card);
__ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
__ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]
assert(CardTableModRefBS::dirty_card_val() == 0, "otherwise check this code");
__ cmp_and_br_short(tmp, G0, Assembler::notEqual, Assembler::pt, not_already_dirty);
__ bind(young_card);
// We didn't take the branch, so we're already dirty: return.
// Use return-from-leaf
__ retl();
......
......@@ -3752,7 +3752,7 @@ static void generate_dirty_card_log_enqueue(jbyte* byte_map_base) {
#define __ masm.
address start = __ pc();
Label not_already_dirty, restart, refill;
Label not_already_dirty, restart, refill, young_card;
#ifdef _LP64
__ srlx(O0, CardTableModRefBS::card_shift, O0);
......@@ -3763,9 +3763,15 @@ static void generate_dirty_card_log_enqueue(jbyte* byte_map_base) {
__ set(addrlit, O1); // O1 := <card table base>
__ ldub(O0, O1, O2); // O2 := [O0 + O1]
__ cmp_and_br_short(O2, G1SATBCardTableModRefBS::g1_young_card_val(), Assembler::equal, Assembler::pt, young_card);
__ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
__ ldub(O0, O1, O2); // O2 := [O0 + O1]
assert(CardTableModRefBS::dirty_card_val() == 0, "otherwise check this code");
__ cmp_and_br_short(O2, G0, Assembler::notEqual, Assembler::pt, not_already_dirty);
__ bind(young_card);
// We didn't take the branch, so we're already dirty: return.
// Use return-from-leaf
__ retl();
......
......@@ -38,6 +38,9 @@
#include "runtime/vframeArray.hpp"
#include "utilities/macros.hpp"
#include "vmreg_x86.inline.hpp"
#if INCLUDE_ALL_GCS
#include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
#endif
// Implementation of StubAssembler
......@@ -1753,13 +1756,17 @@ OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
__ leal(card_addr, __ as_Address(ArrayAddress(cardtable, index)));
#endif
__ cmpb(Address(card_addr, 0), 0);
__ cmpb(Address(card_addr, 0), (int)G1SATBCardTableModRefBS::g1_young_card_val());
__ jcc(Assembler::equal, done);
__ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
__ cmpb(Address(card_addr, 0), (int)CardTableModRefBS::dirty_card_val());
__ jcc(Assembler::equal, done);
// storing region crossing non-NULL, card is clean.
// dirty card and log.
__ movb(Address(card_addr, 0), 0);
__ movb(Address(card_addr, 0), (int)CardTableModRefBS::dirty_card_val());
__ cmpl(queue_index, 0);
__ jcc(Assembler::equal, runtime);
......
......@@ -3389,13 +3389,18 @@ void MacroAssembler::g1_write_barrier_post(Register store_addr,
const Register card_addr = tmp;
lea(card_addr, as_Address(ArrayAddress(cardtable, index)));
#endif
cmpb(Address(card_addr, 0), 0);
cmpb(Address(card_addr, 0), (int)G1SATBCardTableModRefBS::g1_young_card_val());
jcc(Assembler::equal, done);
membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
cmpb(Address(card_addr, 0), (int)CardTableModRefBS::dirty_card_val());
jcc(Assembler::equal, done);
// storing a region crossing, non-NULL oop, card is clean.
// dirty card and log.
movb(Address(card_addr, 0), 0);
movb(Address(card_addr, 0), (int)CardTableModRefBS::dirty_card_val());
cmpl(queue_index, 0);
jcc(Assembler::equal, runtime);
......
......@@ -6035,7 +6035,11 @@ void G1CollectedHeap::verify_dirty_region(HeapRegion* hr) {
// is dirty.
G1SATBCardTableModRefBS* ct_bs = g1_barrier_set();
MemRegion mr(hr->bottom(), hr->pre_dummy_top());
ct_bs->verify_dirty_region(mr);
if (hr->is_young()) {
ct_bs->verify_g1_young_region(mr);
} else {
ct_bs->verify_dirty_region(mr);
}
}
void G1CollectedHeap::verify_dirty_young_list(HeapRegion* head) {
......
......@@ -29,6 +29,7 @@
#include "gc_implementation/g1/g1CollectedHeap.hpp"
#include "gc_implementation/g1/g1AllocRegion.inline.hpp"
#include "gc_implementation/g1/g1CollectorPolicy.hpp"
#include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
#include "gc_implementation/g1/heapRegionSeq.inline.hpp"
#include "utilities/taskqueue.hpp"
......@@ -134,7 +135,7 @@ G1CollectedHeap::dirty_young_block(HeapWord* start, size_t word_size) {
assert(containing_hr->is_in(end - 1), "it should also contain end - 1");
MemRegion mr(start, end);
g1_barrier_set()->dirty(mr);
g1_barrier_set()->g1_mark_as_young(mr);
}
inline RefToScanQueue* G1CollectedHeap::task_queue(int i) const {
......
......@@ -70,6 +70,12 @@ bool G1SATBCardTableModRefBS::mark_card_deferred(size_t card_index) {
if ((val & (clean_card_mask_val() | deferred_card_val())) == deferred_card_val()) {
return false;
}
if (val == g1_young_gen) {
// the card is for a young gen region. We don't need to keep track of all pointers into young
return false;
}
// Cached bit can be installed either on a clean card or on a claimed card.
jbyte new_val = val;
if (val == clean_card_val()) {
......@@ -85,6 +91,19 @@ bool G1SATBCardTableModRefBS::mark_card_deferred(size_t card_index) {
return true;
}
void G1SATBCardTableModRefBS::g1_mark_as_young(const MemRegion& mr) {
jbyte *const first = byte_for(mr.start());
jbyte *const last = byte_after(mr.last());
memset(first, g1_young_gen, last - first);
}
#ifndef PRODUCT
void G1SATBCardTableModRefBS::verify_g1_young_region(MemRegion mr) {
verify_region(mr, g1_young_gen, true);
}
#endif
G1SATBCardTableLoggingModRefBS::
G1SATBCardTableLoggingModRefBS(MemRegion whole_heap,
int max_covered_regions) :
......@@ -97,7 +116,11 @@ G1SATBCardTableLoggingModRefBS(MemRegion whole_heap,
void
G1SATBCardTableLoggingModRefBS::write_ref_field_work(void* field,
oop new_val) {
jbyte* byte = byte_for(field);
volatile jbyte* byte = byte_for(field);
if (*byte == g1_young_gen) {
return;
}
OrderAccess::storeload();
if (*byte != dirty_card) {
*byte = dirty_card;
Thread* thr = Thread::current();
......@@ -129,7 +152,7 @@ G1SATBCardTableLoggingModRefBS::write_ref_field_static(void* field,
void
G1SATBCardTableLoggingModRefBS::invalidate(MemRegion mr, bool whole_heap) {
jbyte* byte = byte_for(mr.start());
volatile jbyte* byte = byte_for(mr.start());
jbyte* last_byte = byte_for(mr.last());
Thread* thr = Thread::current();
if (whole_heap) {
......@@ -138,25 +161,35 @@ G1SATBCardTableLoggingModRefBS::invalidate(MemRegion mr, bool whole_heap) {
byte++;
}
} else {
// Enqueue if necessary.
if (thr->is_Java_thread()) {
JavaThread* jt = (JavaThread*)thr;
while (byte <= last_byte) {
if (*byte != dirty_card) {
*byte = dirty_card;
jt->dirty_card_queue().enqueue(byte);
// skip all consecutive young cards
for (; byte <= last_byte && *byte == g1_young_gen; byte++);
if (byte <= last_byte) {
OrderAccess::storeload();
// Enqueue if necessary.
if (thr->is_Java_thread()) {
JavaThread* jt = (JavaThread*)thr;
for (; byte <= last_byte; byte++) {
if (*byte == g1_young_gen) {
continue;
}
if (*byte != dirty_card) {
*byte = dirty_card;
jt->dirty_card_queue().enqueue(byte);
}
}
byte++;
}
} else {
MutexLockerEx x(Shared_DirtyCardQ_lock,
Mutex::_no_safepoint_check_flag);
while (byte <= last_byte) {
if (*byte != dirty_card) {
*byte = dirty_card;
_dcqs.shared_dirty_card_queue()->enqueue(byte);
} else {
MutexLockerEx x(Shared_DirtyCardQ_lock,
Mutex::_no_safepoint_check_flag);
for (; byte <= last_byte; byte++) {
if (*byte == g1_young_gen) {
continue;
}
if (*byte != dirty_card) {
*byte = dirty_card;
_dcqs.shared_dirty_card_queue()->enqueue(byte);
}
}
byte++;
}
}
}
......
......@@ -38,7 +38,14 @@ class DirtyCardQueueSet;
// snapshot-at-the-beginning marking.
class G1SATBCardTableModRefBS: public CardTableModRefBSForCTRS {
protected:
enum G1CardValues {
g1_young_gen = CT_MR_BS_last_reserved << 1
};
public:
static int g1_young_card_val() { return g1_young_gen; }
// Add "pre_val" to a set of objects that may have been disconnected from the
// pre-marking object graph.
static void enqueue(oop pre_val);
......@@ -118,6 +125,9 @@ public:
_byte_map[card_index] = val;
}
void verify_g1_young_region(MemRegion mr) PRODUCT_RETURN;
void g1_mark_as_young(const MemRegion& mr);
bool mark_card_deferred(size_t card_index);
bool is_card_deferred(size_t card_index) {
......
......@@ -80,6 +80,10 @@ public:
void reset() { if (_buf != NULL) _index = _sz; }
void enqueue(volatile void* ptr) {
enqueue((void*)(ptr));
}
// Enqueues the given "obj".
void enqueue(void* ptr) {
if (!_active) return;
......
......@@ -3713,7 +3713,8 @@ void GraphKit::g1_write_barrier_post(Node* oop_store,
Node* no_base = __ top();
float likely = PROB_LIKELY(0.999);
float unlikely = PROB_UNLIKELY(0.999);
Node* zero = __ ConI(0);
Node* young_card = __ ConI((jint)G1SATBCardTableModRefBS::g1_young_card_val());
Node* dirty_card = __ ConI((jint)CardTableModRefBS::dirty_card_val());
Node* zeroX = __ ConX(0);
// Get the alias_index for raw card-mark memory
......@@ -3769,8 +3770,16 @@ void GraphKit::g1_write_barrier_post(Node* oop_store,
// load the original value of the card
Node* card_val = __ load(__ ctrl(), card_adr, TypeInt::INT, T_BYTE, Compile::AliasIdxRaw);
__ if_then(card_val, BoolTest::ne, zero); {
g1_mark_card(ideal, card_adr, oop_store, alias_idx, index, index_adr, buffer, tf);
__ if_then(card_val, BoolTest::ne, young_card); {
sync_kit(ideal);
// Use Op_MemBarVolatile to achieve the effect of a StoreLoad barrier.
insert_mem_bar(Op_MemBarVolatile, oop_store);
__ sync_kit(this);
Node* card_val_reload = __ load(__ ctrl(), card_adr, TypeInt::INT, T_BYTE, Compile::AliasIdxRaw);
__ if_then(card_val_reload, BoolTest::ne, dirty_card); {
g1_mark_card(ideal, card_adr, oop_store, alias_idx, index, index_adr, buffer, tf);
} __ end_if();
} __ end_if();
} __ end_if();
} __ end_if();
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册