13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "gc/g1/g1FreeIdSet.hpp"
27 #include "memory/allocation.inline.hpp"
28 #include "runtime/atomic.hpp"
29 #include "utilities/debug.hpp"
30 #include "utilities/globalDefinitions.hpp"
31 #include "utilities/macros.hpp"
32
33 const uint claimed = UINT_MAX;
34
35 G1FreeIdSet::G1FreeIdSet(uint start, uint size) :
36 _sem(size), // counting semaphore for available ids
37 _next(NULL), // array of "next" indices
38 _start(start), // first id value
39 _size(size), // number of available ids
40 _head_index_mask(0), // mask for extracting index from a _head value.
41 _head(0) // low part: index; high part: update counter
42 {
43 assert(size != 0, "precondition");
44 assert(start <= (UINT_MAX - size),
45 "start (%u) + size (%u) overflow: ", start, size);
46 // 2^shift must be greater than size. Equal is not permitted, because
47 // size is the "end of list" value, and can be the index part of _head.
48 uint shift = log2_intptr((uintptr_t)size) + 1;
49 assert(shift <= (BitsPerWord / 2), "excessive size %u", size);
50 _head_index_mask = (uintx(1) << shift) - 1;
51 assert(size <= _head_index_mask, "invariant");
52 _next = NEW_C_HEAP_ARRAY(uint, size, mtGC);
53 for (uint i = 0; i < size; ++i) {
54 _next[i] = i + 1;
55 }
56 }
57
58 G1FreeIdSet::~G1FreeIdSet() {
59 FREE_C_HEAP_ARRAY(uint, _next);
60 }
61
62 uint G1FreeIdSet::head_index(uintx head) const {
63 return head & _head_index_mask;
64 }
65
66 uintx G1FreeIdSet::make_head(uint index, uintx old_head) const {
67 // Include incremented old update counter to avoid ABA problem.
68 return index | ((old_head & ~_head_index_mask) + 1 + _head_index_mask);
69 }
70
71 uint G1FreeIdSet::claim_par_id() {
72 _sem.wait();
73 // Semaphore gate permits passage by no more than the number of
74 // available ids, so there must be one that we can claim. But there
75 // may be multiple threads trying to claim ids at the same time.
76 uintx old_head = Atomic::load(&_head);
77 uint index;
78 while (true) {
79 index = head_index(old_head);
80 assert(index < _size, "invariant");
81 uintx new_head = make_head(_next[index], old_head);
82 new_head = Atomic::cmpxchg(new_head, &_head, old_head);
83 if (new_head == old_head) break;
84 old_head = new_head;
85 }
86 DEBUG_ONLY(_next[index] = claimed;)
87 return _start + index;
88 }
89
90 void G1FreeIdSet::release_par_id(uint id) {
91 uint index = id - _start;
92 assert(index < _size, "invalid id %u", id);
93 assert(_next[index] == claimed, "precondition");
94 uintx old_head = Atomic::load(&_head);
95 while (true) {
96 _next[index] = head_index(old_head);
97 uintx new_head = make_head(index, old_head);
98 new_head = Atomic::cmpxchg(new_head, &_head, old_head);
99 if (new_head == old_head) break;
100 old_head = new_head;
101 }
102 // Now that id has been released, permit another thread through the gate.
103 _sem.signal();
104 }
|
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "gc/g1/g1FreeIdSet.hpp"
27 #include "memory/allocation.inline.hpp"
28 #include "runtime/atomic.hpp"
29 #include "utilities/debug.hpp"
30 #include "utilities/globalDefinitions.hpp"
31 #include "utilities/macros.hpp"
32
33 G1FreeIdSet::G1FreeIdSet(uint start, uint size) :
34 _sem(size), // counting semaphore for available ids
35 _next(NULL), // array of "next" indices
36 _start(start), // first id value
37 _size(size), // number of available ids
38 _head_index_mask(0), // mask for extracting index from a _head value.
39 _head(0) // low part: index; high part: update counter
40 {
41 assert(size != 0, "precondition");
42 assert(start <= (UINT_MAX - size),
43 "start (%u) + size (%u) overflow: ", start, size);
44 // 2^shift must be greater than size. Equal is not permitted, because
45 // size is the "end of list" value, and can be the index part of _head.
46 uint shift = log2_intptr((uintptr_t)size) + 1;
47 assert(shift <= (BitsPerWord / 2), "excessive size %u", size);
48 _head_index_mask = (uintx(1) << shift) - 1;
49 assert(size <= _head_index_mask, "invariant");
50 _next = NEW_C_HEAP_ARRAY(uint, size, mtGC);
51 for (uint i = 0; i < size; ++i) {
52 _next[i] = i + 1;
53 }
54 }
55
56 G1FreeIdSet::~G1FreeIdSet() {
57 FREE_C_HEAP_ARRAY(uint, _next);
58 }
59
60 uint G1FreeIdSet::head_index(uintx head) const {
61 return head & _head_index_mask;
62 }
63
64 uintx G1FreeIdSet::make_head(uint index, uintx old_head) const {
65 // Include incremented old update counter to avoid ABA problem.
66 return index | ((old_head & ~_head_index_mask) + 1 + _head_index_mask);
67 }
68
69 const uint Claimed = UINT_MAX;
70
71 uint G1FreeIdSet::claim_par_id() {
72 _sem.wait();
73 // Semaphore gate permits passage by no more than the number of
74 // available ids, so there must be one that we can claim. But there
75 // may be multiple threads trying to claim ids at the same time.
76 uintx old_head = Atomic::load(&_head);
77 uint index;
78 while (true) {
79 index = head_index(old_head);
80 assert(index < _size, "invariant");
81 uintx new_head = make_head(_next[index], old_head);
82 new_head = Atomic::cmpxchg(new_head, &_head, old_head);
83 if (new_head == old_head) break;
84 old_head = new_head;
85 }
86 DEBUG_ONLY(_next[index] = Claimed;)
87 return _start + index;
88 }
89
90 void G1FreeIdSet::release_par_id(uint id) {
91 uint index = id - _start;
92 assert(index < _size, "invalid id %u", id);
93 assert(_next[index] == Claimed, "precondition");
94 uintx old_head = Atomic::load(&_head);
95 while (true) {
96 _next[index] = head_index(old_head);
97 uintx new_head = make_head(index, old_head);
98 new_head = Atomic::cmpxchg(new_head, &_head, old_head);
99 if (new_head == old_head) break;
100 old_head = new_head;
101 }
102 // Now that id has been released, permit another thread through the gate.
103 _sem.signal();
104 }
|