1 /*
   2  * Copyright (c) 2015, 2019, Red Hat, Inc. All rights reserved.
   3  *
   4  * This code is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License version 2 only, as
   6  * published by the Free Software Foundation.
   7  *
   8  * This code is distributed in the hope that it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  11  * version 2 for more details (a copy is included in the LICENSE file that
  12  * accompanied this code).
  13  *
  14  * You should have received a copy of the GNU General Public License version
  15  * 2 along with this work; if not, write to the Free Software Foundation,
  16  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  17  *
  18  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  19  * or visit www.oracle.com if you need additional information or have any
  20  * questions.
  21  *
  22  */
  23 
  24 #ifndef SHARE_GC_SHENANDOAH_SHENANDOAHCONCURRENTMARK_INLINE_HPP
  25 #define SHARE_GC_SHENANDOAH_SHENANDOAHCONCURRENTMARK_INLINE_HPP
  26 
  27 #include "gc/shenandoah/shenandoahAsserts.hpp"
  28 #include "gc/shenandoah/shenandoahBarrierSet.inline.hpp"
  29 #include "gc/shenandoah/shenandoahConcurrentMark.hpp"
  30 #include "gc/shenandoah/shenandoahMarkingContext.inline.hpp"
  31 #include "gc/shenandoah/shenandoahStringDedup.inline.hpp"
  32 #include "gc/shenandoah/shenandoahTaskqueue.inline.hpp"
  33 #include "memory/iterator.inline.hpp"
  34 #include "oops/compressedOops.inline.hpp"
  35 #include "oops/oop.inline.hpp"
  36 #include "runtime/prefetch.inline.hpp"
  37 
  38 template <class T>
  39 void ShenandoahConcurrentMark::do_task(ShenandoahObjToScanQueue* q, T* cl, jushort* live_data, ShenandoahMarkTask* task) {
  40   oop obj = task->obj();
  41 
  42   shenandoah_assert_not_forwarded_except(NULL, obj, _heap->is_concurrent_traversal_in_progress() && _heap->cancelled_gc());
  43   shenandoah_assert_marked(NULL, obj);
  44   shenandoah_assert_not_in_cset_except(NULL, obj, _heap->cancelled_gc());
  45 
  46   if (task->is_not_chunked()) {
  47     if (obj->is_instance()) {
  48       // Case 1: Normal oop, process as usual.
  49       obj->oop_iterate(cl);
  50     } else if (obj->is_objArray()) {
  51       // Case 2: Object array instance and no chunk is set. Must be the first
  52       // time we visit it, start the chunked processing.
  53       do_chunked_array_start<T>(q, cl, obj);
  54     } else {
  55       // Case 3: Primitive array. Do nothing, no oops there. We use the same
  56       // performance tweak TypeArrayKlass::oop_oop_iterate_impl is using:
  57       // We skip iterating over the klass pointer since we know that
  58       // Universe::TypeArrayKlass never moves.
  59       assert (obj->is_typeArray(), "should be type array");
  60     }
  61     // Count liveness the last: push the outstanding work to the queues first
  62     count_liveness(live_data, obj);
  63   } else {
  64     // Case 4: Array chunk, has sensible chunk id. Process it.
  65     do_chunked_array<T>(q, cl, obj, task->chunk(), task->pow());
  66   }
  67 }
  68 
  69 inline void ShenandoahConcurrentMark::count_liveness(jushort* live_data, oop obj) {
  70   size_t region_idx = _heap->heap_region_index_containing(obj);
  71   ShenandoahHeapRegion* region = _heap->get_region(region_idx);
  72   size_t size = obj->size();
  73 
  74   if (!region->is_humongous_start()) {
  75     assert(!region->is_humongous(), "Cannot have continuations here");
  76     size_t max = (1 << (sizeof(jushort) * 8)) - 1;
  77     if (size >= max) {
  78       // too big, add to region data directly
  79       region->increase_live_data_gc_words(size);
  80     } else {
  81       jushort cur = live_data[region_idx];
  82       size_t new_val = cur + size;
  83       if (new_val >= max) {
  84         // overflow, flush to region data
  85         region->increase_live_data_gc_words(new_val);
  86         live_data[region_idx] = 0;
  87       } else {
  88         // still good, remember in locals
  89         live_data[region_idx] = (jushort) new_val;
  90       }
  91     }
  92   } else {
  93     shenandoah_assert_in_correct_region(NULL, obj);
  94     size_t num_regions = ShenandoahHeapRegion::required_regions(size * HeapWordSize);
  95 
  96     for (size_t i = region_idx; i < region_idx + num_regions; i++) {
  97       ShenandoahHeapRegion* chain_reg = _heap->get_region(i);
  98       assert(chain_reg->is_humongous(), "Expecting a humongous region");
  99       chain_reg->increase_live_data_gc_words(chain_reg->used() >> LogHeapWordSize);
 100     }
 101   }
 102 }
 103 
 104 template <class T>
 105 inline void ShenandoahConcurrentMark::do_chunked_array_start(ShenandoahObjToScanQueue* q, T* cl, oop obj) {
 106   assert(obj->is_objArray(), "expect object array");
 107   objArrayOop array = objArrayOop(obj);
 108   int len = array->length();
 109 
 110   if (len <= (int) ObjArrayMarkingStride*2) {
 111     // A few slices only, process directly
 112     array->oop_iterate_range(cl, 0, len);
 113   } else {
 114     int bits = log2_long((size_t) len);
 115     // Compensate for non-power-of-two arrays, cover the array in excess:
 116     if (len != (1 << bits)) bits++;
 117 
 118     // Only allow full chunks on the queue. This frees do_chunked_array() from checking from/to
 119     // boundaries against array->length(), touching the array header on every chunk.
 120     //
 121     // To do this, we cut the prefix in full-sized chunks, and submit them on the queue.
 122     // If the array is not divided in chunk sizes, then there would be an irregular tail,
 123     // which we will process separately.
 124 
 125     int last_idx = 0;
 126 
 127     int chunk = 1;
 128     int pow = bits;
 129 
 130     // Handle overflow
 131     if (pow >= 31) {
 132       assert (pow == 31, "sanity");
 133       pow--;
 134       chunk = 2;
 135       last_idx = (1 << pow);
 136       bool pushed = q->push(ShenandoahMarkTask(array, 1, pow));
 137       assert(pushed, "overflow queue should always succeed pushing");
 138     }
 139 
 140     // Split out tasks, as suggested in ObjArrayChunkedTask docs. Record the last
 141     // successful right boundary to figure out the irregular tail.
 142     while ((1 << pow) > (int)ObjArrayMarkingStride &&
 143            (chunk*2 < ShenandoahMarkTask::chunk_size())) {
 144       pow--;
 145       int left_chunk = chunk*2 - 1;
 146       int right_chunk = chunk*2;
 147       int left_chunk_end = left_chunk * (1 << pow);
 148       if (left_chunk_end < len) {
 149         bool pushed = q->push(ShenandoahMarkTask(array, left_chunk, pow));
 150         assert(pushed, "overflow queue should always succeed pushing");
 151         chunk = right_chunk;
 152         last_idx = left_chunk_end;
 153       } else {
 154         chunk = left_chunk;
 155       }
 156     }
 157 
 158     // Process the irregular tail, if present
 159     int from = last_idx;
 160     if (from < len) {
 161       array->oop_iterate_range(cl, from, len);
 162     }
 163   }
 164 }
 165 
 166 template <class T>
 167 inline void ShenandoahConcurrentMark::do_chunked_array(ShenandoahObjToScanQueue* q, T* cl, oop obj, int chunk, int pow) {
 168   assert(obj->is_objArray(), "expect object array");
 169   objArrayOop array = objArrayOop(obj);
 170 
 171   assert (ObjArrayMarkingStride > 0, "sanity");
 172 
 173   // Split out tasks, as suggested in ObjArrayChunkedTask docs. Avoid pushing tasks that
 174   // are known to start beyond the array.
 175   while ((1 << pow) > (int)ObjArrayMarkingStride && (chunk*2 < ShenandoahMarkTask::chunk_size())) {
 176     pow--;
 177     chunk *= 2;
 178     bool pushed = q->push(ShenandoahMarkTask(array, chunk - 1, pow));
 179     assert(pushed, "overflow queue should always succeed pushing");
 180   }
 181 
 182   int chunk_size = 1 << pow;
 183 
 184   int from = (chunk - 1) * chunk_size;
 185   int to = chunk * chunk_size;
 186 
 187 #ifdef ASSERT
 188   int len = array->length();
 189   assert (0 <= from && from < len, "from is sane: %d/%d", from, len);
 190   assert (0 < to && to <= len, "to is sane: %d/%d", to, len);
 191 #endif
 192 
 193   array->oop_iterate_range(cl, from, to);
 194 }
 195 
 196 class ShenandoahSATBBufferClosure : public SATBBufferClosure {
 197 private:
 198   ShenandoahObjToScanQueue* _queue;
 199   ShenandoahHeap* _heap;
 200   ShenandoahMarkingContext* const _mark_context;
 201 public:
 202   ShenandoahSATBBufferClosure(ShenandoahObjToScanQueue* q) :
 203     _queue(q),
 204     _heap(ShenandoahHeap::heap()),
 205     _mark_context(_heap->marking_context())
 206   {
 207   }
 208 
 209   void do_buffer(void **buffer, size_t size) {
 210     if (_heap->has_forwarded_objects()) {
 211       if (ShenandoahStringDedup::is_enabled()) {
 212         do_buffer_impl<RESOLVE, ENQUEUE_DEDUP>(buffer, size);
 213       } else {
 214         do_buffer_impl<RESOLVE, NO_DEDUP>(buffer, size);
 215       }
 216     } else {
 217       if (ShenandoahStringDedup::is_enabled()) {
 218         do_buffer_impl<NONE, ENQUEUE_DEDUP>(buffer, size);
 219       } else {
 220         do_buffer_impl<NONE, NO_DEDUP>(buffer, size);
 221       }
 222     }
 223   }
 224 
 225   template<UpdateRefsMode UPDATE_REFS, StringDedupMode STRING_DEDUP>
 226   void do_buffer_impl(void **buffer, size_t size) {
 227     for (size_t i = 0; i < size; ++i) {
 228       oop *p = (oop *) &buffer[i];
 229       ShenandoahConcurrentMark::mark_through_ref<oop, UPDATE_REFS, STRING_DEDUP>(p, _heap, _queue, _mark_context);
 230     }
 231   }
 232 };
 233 
 234 template<class T, UpdateRefsMode UPDATE_REFS, StringDedupMode STRING_DEDUP>
 235 inline void ShenandoahConcurrentMark::mark_through_ref(T *p, ShenandoahHeap* heap, ShenandoahObjToScanQueue* q, ShenandoahMarkingContext* const mark_context) {
 236   T o = RawAccess<>::oop_load(p);
 237   if (!CompressedOops::is_null(o)) {
 238     oop obj = CompressedOops::decode_not_null(o);
 239     switch (UPDATE_REFS) {
 240     case NONE:
 241       break;
 242     case RESOLVE:
 243       obj = ShenandoahBarrierSet::resolve_forwarded_not_null(obj);
 244       break;
 245     case SIMPLE:
 246       // We piggy-back reference updating to the marking tasks.
 247       obj = heap->update_with_forwarded_not_null(p, obj);
 248       break;
 249     case CONCURRENT:
 250       obj = heap->maybe_update_with_forwarded_not_null(p, obj);
 251       break;
 252     default:
 253       ShouldNotReachHere();
 254     }
 255 
 256     // Note: Only when concurrently updating references can obj be different
 257     // (that is, really different, not just different from-/to-space copies of the same)
 258     // from the one we originally loaded. Mutator thread can beat us by writing something
 259     // else into the location. In that case, we would mark through that updated value,
 260     // on the off-chance it is not handled by other means (e.g. via SATB). However,
 261     // if that write was NULL, we don't need to do anything else.
 262     if (UPDATE_REFS != CONCURRENT || !CompressedOops::is_null(obj)) {
 263       shenandoah_assert_not_forwarded(p, obj);
 264       shenandoah_assert_not_in_cset_except(p, obj, heap->cancelled_gc());
 265 
 266       if (mark_context->mark(obj)) {
 267         bool pushed = q->push(ShenandoahMarkTask(obj));
 268         assert(pushed, "overflow queue should always succeed pushing");
 269 
 270         if ((STRING_DEDUP == ENQUEUE_DEDUP) && ShenandoahStringDedup::is_candidate(obj)) {
 271           assert(ShenandoahStringDedup::is_enabled(), "Must be enabled");
 272           ShenandoahStringDedup::enqueue_candidate(obj);
 273         }
 274       }
 275 
 276       shenandoah_assert_marked(p, obj);
 277     }
 278   }
 279 }
 280 
 281 #endif // SHARE_GC_SHENANDOAH_SHENANDOAHCONCURRENTMARK_INLINE_HPP
--- EOF ---