1 /*
2 * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
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 #ifndef SHARE_VM_GC_IMPLEMENTATION_SHARED_MUTABLENUMASPACE_HPP
26 #define SHARE_VM_GC_IMPLEMENTATION_SHARED_MUTABLENUMASPACE_HPP
27
28 #ifndef SERIALGC
29 #include "gc_implementation/shared/gcUtil.hpp"
30 #include "gc_implementation/shared/mutableSpace.hpp"
31 #endif
32
33 /*
34 * The NUMA-aware allocator (MutableNUMASpace) is basically a modification
35 * of MutableSpace which preserves interfaces but implements different
36 * functionality. The space is split into chunks for each locality group
37 * (resizing for adaptive size policy is also supported). For each thread
38 * allocations are performed in the chunk corresponding to the home locality
39 * group of the thread. Whenever any chunk fills-in the young generation
40 * collection occurs.
41 * The chunks can be also be adaptively resized. The idea behind the adaptive
42 * sizing is to reduce the loss of the space in the eden due to fragmentation.
43 * The main cause of fragmentation is uneven allocation rates of threads.
44 * The allocation rate difference between locality groups may be caused either by
45 * application specifics or by uneven LWP distribution by the OS. Besides,
46 * application can have less threads then the number of locality groups.
47 * In order to resize the chunk we measure the allocation rate of the
48 * application between collections. After that we reshape the chunks to reflect
49 * the allocation rate pattern. The AdaptiveWeightedAverage exponentially
50 * decaying average is used to smooth the measurements. The NUMASpaceResizeRate
51 * parameter is used to control the adaptation speed by restricting the number of
52 * bytes that can be moved during the adaptation phase.
53 * Chunks may contain pages from a wrong locality group. The page-scanner has
54 * been introduced to address the problem. Remote pages typically appear due to
55 * the memory shortage in the target locality group. Besides Solaris would
56 * allocate a large page from the remote locality group even if there are small
57 * local pages available. The page-scanner scans the pages right after the
58 * collection and frees remote pages in hope that subsequent reallocation would
59 * be more successful. This approach proved to be useful on systems with high
60 * load where multiple processes are competing for the memory.
61 */
62
63 class MutableNUMASpace : public MutableSpace {
64 friend class VMStructs;
65
66 class LGRPSpace : public CHeapObj {
67 int _lgrp_id;
68 MutableSpace* _space;
69 MemRegion _invalid_region;
70 AdaptiveWeightedAverage *_alloc_rate;
71 bool _allocation_failed;
72
73 struct SpaceStats {
74 size_t _local_space, _remote_space, _unbiased_space, _uncommited_space;
75 size_t _large_pages, _small_pages;
76
77 SpaceStats() {
78 _local_space = 0;
79 _remote_space = 0;
80 _unbiased_space = 0;
81 _uncommited_space = 0;
82 _large_pages = 0;
83 _small_pages = 0;
84 }
85 };
86
87 SpaceStats _space_stats;
88
89 char* _last_page_scanned;
90 char* last_page_scanned() { return _last_page_scanned; }
91 void set_last_page_scanned(char* p) { _last_page_scanned = p; }
92 public:
93 LGRPSpace(int l, size_t alignment) : _lgrp_id(l), _last_page_scanned(NULL), _allocation_failed(false) {
94 _space = new MutableSpace(alignment);
95 _alloc_rate = new AdaptiveWeightedAverage(NUMAChunkResizeWeight);
96 }
97 ~LGRPSpace() {
98 delete _space;
99 delete _alloc_rate;
100 }
101
102 void add_invalid_region(MemRegion r) {
103 if (!_invalid_region.is_empty()) {
104 _invalid_region.set_start(MIN2(_invalid_region.start(), r.start()));
105 _invalid_region.set_end(MAX2(_invalid_region.end(), r.end()));
106 } else {
107 _invalid_region = r;
108 }
109 }
110
111 static bool equals(void* lgrp_id_value, LGRPSpace* p) {
112 return *(int*)lgrp_id_value == p->lgrp_id();
113 }
114
115 // Report a failed allocation.
116 void set_allocation_failed() { _allocation_failed = true; }
117
118 void sample() {
119 // If there was a failed allocation make allocation rate equal
120 // to the size of the whole chunk. This ensures the progress of
121 // the adaptation process.
122 size_t alloc_rate_sample;
123 if (_allocation_failed) {
124 alloc_rate_sample = space()->capacity_in_bytes();
125 _allocation_failed = false;
126 } else {
127 alloc_rate_sample = space()->used_in_bytes();
128 }
129 alloc_rate()->sample(alloc_rate_sample);
130 }
131
132 MemRegion invalid_region() const { return _invalid_region; }
133 void set_invalid_region(MemRegion r) { _invalid_region = r; }
134 int lgrp_id() const { return _lgrp_id; }
135 MutableSpace* space() const { return _space; }
136 AdaptiveWeightedAverage* alloc_rate() const { return _alloc_rate; }
137 void clear_alloc_rate() { _alloc_rate->clear(); }
138 SpaceStats* space_stats() { return &_space_stats; }
139 void clear_space_stats() { _space_stats = SpaceStats(); }
140
141 void accumulate_statistics(size_t page_size);
142 void scan_pages(size_t page_size, size_t page_count);
143 };
144
145 GrowableArray<LGRPSpace*>* _lgrp_spaces;
146 size_t _page_size;
147 unsigned _adaptation_cycles, _samples_count;
148
149 void set_page_size(size_t psz) { _page_size = psz; }
150 size_t page_size() const { return _page_size; }
151
152 unsigned adaptation_cycles() { return _adaptation_cycles; }
153 void set_adaptation_cycles(int v) { _adaptation_cycles = v; }
154
155 unsigned samples_count() { return _samples_count; }
156 void increment_samples_count() { ++_samples_count; }
157
158 size_t _base_space_size;
159 void set_base_space_size(size_t v) { _base_space_size = v; }
160 size_t base_space_size() const { return _base_space_size; }
161
162 // Check if the NUMA topology has changed. Add and remove spaces if needed.
163 // The update can be forced by setting the force parameter equal to true.
164 bool update_layout(bool force);
165 // Bias region towards the lgrp.
166 void bias_region(MemRegion mr, int lgrp_id);
167 // Free pages in a given region.
168 void free_region(MemRegion mr);
169 // Get current chunk size.
170 size_t current_chunk_size(int i);
171 // Get default chunk size (equally divide the space).
172 size_t default_chunk_size();
173 // Adapt the chunk size to follow the allocation rate.
174 size_t adaptive_chunk_size(int i, size_t limit);
175 // Scan and free invalid pages.
176 void scan_pages(size_t page_count);
177 // Return the bottom_region and the top_region. Align them to page_size() boundary.
178 // |------------------new_region---------------------------------|
179 // |----bottom_region--|---intersection---|------top_region------|
180 void select_tails(MemRegion new_region, MemRegion intersection,
181 MemRegion* bottom_region, MemRegion *top_region);
182 // Try to merge the invalid region with the bottom or top region by decreasing
183 // the intersection area. Return the invalid_region aligned to the page_size()
184 // boundary if it's inside the intersection. Return non-empty invalid_region
185 // if it lies inside the intersection (also page-aligned).
186 // |------------------new_region---------------------------------|
187 // |----------------|-------invalid---|--------------------------|
188 // |----bottom_region--|---intersection---|------top_region------|
189 void merge_regions(MemRegion new_region, MemRegion* intersection,
190 MemRegion *invalid_region);
191
192 public:
193 GrowableArray<LGRPSpace*>* lgrp_spaces() const { return _lgrp_spaces; }
194 MutableNUMASpace(size_t alignment);
195 virtual ~MutableNUMASpace();
196 // Space initialization.
197 virtual void initialize(MemRegion mr, bool clear_space, bool mangle_space, bool setup_pages = SetupPages);
198 // Update space layout if necessary. Do all adaptive resizing job.
199 virtual void update();
200 // Update allocation rate averages.
201 virtual void accumulate_statistics();
202
203 virtual void clear(bool mangle_space);
204 virtual void mangle_unused_area() PRODUCT_RETURN;
205 virtual void mangle_unused_area_complete() PRODUCT_RETURN;
206 virtual void mangle_region(MemRegion mr) PRODUCT_RETURN;
207 virtual void check_mangled_unused_area(HeapWord* limit) PRODUCT_RETURN;
208 virtual void check_mangled_unused_area_complete() PRODUCT_RETURN;
209 virtual void set_top_for_allocations(HeapWord* v) PRODUCT_RETURN;
210 virtual void set_top_for_allocations() PRODUCT_RETURN;
211
212 virtual void ensure_parsability();
213 virtual size_t used_in_words() const;
214 virtual size_t free_in_words() const;
215
216 using MutableSpace::capacity_in_words;
217 virtual size_t capacity_in_words(Thread* thr) const;
218 virtual size_t tlab_capacity(Thread* thr) const;
219 virtual size_t unsafe_max_tlab_alloc(Thread* thr) const;
220
221 // Allocation (return NULL if full)
222 virtual HeapWord* allocate(size_t word_size);
223 virtual HeapWord* cas_allocate(size_t word_size);
224
225 // Debugging
226 virtual void print_on(outputStream* st) const;
227 virtual void print_short_on(outputStream* st) const;
228 virtual void verify(bool allow_dirty);
229
230 virtual void set_top(HeapWord* value);
231 };
232
233 #endif // SHARE_VM_GC_IMPLEMENTATION_SHARED_MUTABLENUMASPACE_HPP