Thu, 05 Jun 2008 15:57:56 -0700
6711316: Open source the Garbage-First garbage collector
Summary: First mercurial integration of the code for the Garbage-First garbage collector.
Reviewed-by: apetrusenko, iveresov, jmasa, sgoldman, tonyp, ysr
duke@435 | 1 | /* |
duke@435 | 2 | * Copyright 2006-2007 Sun Microsystems, Inc. All Rights Reserved. |
duke@435 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
duke@435 | 4 | * |
duke@435 | 5 | * This code is free software; you can redistribute it and/or modify it |
duke@435 | 6 | * under the terms of the GNU General Public License version 2 only, as |
duke@435 | 7 | * published by the Free Software Foundation. |
duke@435 | 8 | * |
duke@435 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
duke@435 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
duke@435 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
duke@435 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
duke@435 | 13 | * accompanied this code). |
duke@435 | 14 | * |
duke@435 | 15 | * You should have received a copy of the GNU General Public License version |
duke@435 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
duke@435 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
duke@435 | 18 | * |
duke@435 | 19 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
duke@435 | 20 | * CA 95054 USA or visit www.sun.com if you need additional information or |
duke@435 | 21 | * have any questions. |
duke@435 | 22 | * |
duke@435 | 23 | */ |
duke@435 | 24 | |
duke@435 | 25 | /* |
duke@435 | 26 | * The NUMA-aware allocator (MutableNUMASpace) is basically a modification |
duke@435 | 27 | * of MutableSpace which preserves interfaces but implements different |
duke@435 | 28 | * functionality. The space is split into chunks for each locality group |
duke@435 | 29 | * (resizing for adaptive size policy is also supported). For each thread |
duke@435 | 30 | * allocations are performed in the chunk corresponding to the home locality |
duke@435 | 31 | * group of the thread. Whenever any chunk fills-in the young generation |
duke@435 | 32 | * collection occurs. |
duke@435 | 33 | * The chunks can be also be adaptively resized. The idea behind the adaptive |
duke@435 | 34 | * sizing is to reduce the loss of the space in the eden due to fragmentation. |
duke@435 | 35 | * The main cause of fragmentation is uneven allocation rates of threads. |
duke@435 | 36 | * The allocation rate difference between locality groups may be caused either by |
duke@435 | 37 | * application specifics or by uneven LWP distribution by the OS. Besides, |
duke@435 | 38 | * application can have less threads then the number of locality groups. |
duke@435 | 39 | * In order to resize the chunk we measure the allocation rate of the |
duke@435 | 40 | * application between collections. After that we reshape the chunks to reflect |
duke@435 | 41 | * the allocation rate pattern. The AdaptiveWeightedAverage exponentially |
duke@435 | 42 | * decaying average is used to smooth the measurements. The NUMASpaceResizeRate |
duke@435 | 43 | * parameter is used to control the adaptation speed by restricting the number of |
duke@435 | 44 | * bytes that can be moved during the adaptation phase. |
duke@435 | 45 | * Chunks may contain pages from a wrong locality group. The page-scanner has |
duke@435 | 46 | * been introduced to address the problem. Remote pages typically appear due to |
duke@435 | 47 | * the memory shortage in the target locality group. Besides Solaris would |
duke@435 | 48 | * allocate a large page from the remote locality group even if there are small |
duke@435 | 49 | * local pages available. The page-scanner scans the pages right after the |
duke@435 | 50 | * collection and frees remote pages in hope that subsequent reallocation would |
duke@435 | 51 | * be more successful. This approach proved to be useful on systems with high |
duke@435 | 52 | * load where multiple processes are competing for the memory. |
duke@435 | 53 | */ |
duke@435 | 54 | |
duke@435 | 55 | class MutableNUMASpace : public MutableSpace { |
duke@435 | 56 | friend class VMStructs; |
duke@435 | 57 | |
duke@435 | 58 | class LGRPSpace : public CHeapObj { |
duke@435 | 59 | int _lgrp_id; |
duke@435 | 60 | MutableSpace* _space; |
duke@435 | 61 | MemRegion _invalid_region; |
duke@435 | 62 | AdaptiveWeightedAverage *_alloc_rate; |
duke@435 | 63 | |
duke@435 | 64 | struct SpaceStats { |
duke@435 | 65 | size_t _local_space, _remote_space, _unbiased_space, _uncommited_space; |
duke@435 | 66 | size_t _large_pages, _small_pages; |
duke@435 | 67 | |
duke@435 | 68 | SpaceStats() { |
duke@435 | 69 | _local_space = 0; |
duke@435 | 70 | _remote_space = 0; |
duke@435 | 71 | _unbiased_space = 0; |
duke@435 | 72 | _uncommited_space = 0; |
duke@435 | 73 | _large_pages = 0; |
duke@435 | 74 | _small_pages = 0; |
duke@435 | 75 | } |
duke@435 | 76 | }; |
duke@435 | 77 | |
duke@435 | 78 | SpaceStats _space_stats; |
duke@435 | 79 | |
duke@435 | 80 | char* _last_page_scanned; |
duke@435 | 81 | char* last_page_scanned() { return _last_page_scanned; } |
duke@435 | 82 | void set_last_page_scanned(char* p) { _last_page_scanned = p; } |
duke@435 | 83 | public: |
duke@435 | 84 | LGRPSpace(int l) : _lgrp_id(l), _last_page_scanned(NULL) { |
duke@435 | 85 | _space = new MutableSpace(); |
duke@435 | 86 | _alloc_rate = new AdaptiveWeightedAverage(NUMAChunkResizeWeight); |
duke@435 | 87 | } |
duke@435 | 88 | ~LGRPSpace() { |
duke@435 | 89 | delete _space; |
duke@435 | 90 | delete _alloc_rate; |
duke@435 | 91 | } |
duke@435 | 92 | |
duke@435 | 93 | void add_invalid_region(MemRegion r) { |
duke@435 | 94 | if (!_invalid_region.is_empty()) { |
duke@435 | 95 | _invalid_region.set_start(MIN2(_invalid_region.start(), r.start())); |
duke@435 | 96 | _invalid_region.set_end(MAX2(_invalid_region.end(), r.end())); |
duke@435 | 97 | } else { |
duke@435 | 98 | _invalid_region = r; |
duke@435 | 99 | } |
duke@435 | 100 | } |
duke@435 | 101 | |
duke@435 | 102 | static bool equals(void* lgrp_id_value, LGRPSpace* p) { |
duke@435 | 103 | return *(int*)lgrp_id_value == p->lgrp_id(); |
duke@435 | 104 | } |
duke@435 | 105 | |
duke@435 | 106 | void sample() { |
duke@435 | 107 | alloc_rate()->sample(space()->used_in_bytes()); |
duke@435 | 108 | } |
duke@435 | 109 | |
duke@435 | 110 | MemRegion invalid_region() const { return _invalid_region; } |
duke@435 | 111 | void set_invalid_region(MemRegion r) { _invalid_region = r; } |
duke@435 | 112 | int lgrp_id() const { return _lgrp_id; } |
duke@435 | 113 | MutableSpace* space() const { return _space; } |
duke@435 | 114 | AdaptiveWeightedAverage* alloc_rate() const { return _alloc_rate; } |
duke@435 | 115 | SpaceStats* space_stats() { return &_space_stats; } |
duke@435 | 116 | void clear_space_stats() { _space_stats = SpaceStats(); } |
duke@435 | 117 | |
duke@435 | 118 | void accumulate_statistics(size_t page_size); |
duke@435 | 119 | void scan_pages(size_t page_size, size_t page_count); |
duke@435 | 120 | }; |
duke@435 | 121 | |
duke@435 | 122 | GrowableArray<LGRPSpace*>* _lgrp_spaces; |
duke@435 | 123 | size_t _page_size; |
duke@435 | 124 | unsigned _adaptation_cycles, _samples_count; |
duke@435 | 125 | |
duke@435 | 126 | void set_page_size(size_t psz) { _page_size = psz; } |
duke@435 | 127 | size_t page_size() const { return _page_size; } |
duke@435 | 128 | |
duke@435 | 129 | unsigned adaptation_cycles() { return _adaptation_cycles; } |
duke@435 | 130 | void set_adaptation_cycles(int v) { _adaptation_cycles = v; } |
duke@435 | 131 | |
duke@435 | 132 | unsigned samples_count() { return _samples_count; } |
duke@435 | 133 | void increment_samples_count() { ++_samples_count; } |
duke@435 | 134 | |
duke@435 | 135 | size_t _base_space_size; |
duke@435 | 136 | void set_base_space_size(size_t v) { _base_space_size = v; } |
duke@435 | 137 | size_t base_space_size() const { return _base_space_size; } |
duke@435 | 138 | |
duke@435 | 139 | // Check if the NUMA topology has changed. Add and remove spaces if needed. |
duke@435 | 140 | // The update can be forced by setting the force parameter equal to true. |
duke@435 | 141 | bool update_layout(bool force); |
iveresov@576 | 142 | // Bias region towards the lgrp. |
iveresov@576 | 143 | void bias_region(MemRegion mr, int lgrp_id); |
duke@435 | 144 | // Free pages in a given region. |
duke@435 | 145 | void free_region(MemRegion mr); |
duke@435 | 146 | // Get current chunk size. |
duke@435 | 147 | size_t current_chunk_size(int i); |
duke@435 | 148 | // Get default chunk size (equally divide the space). |
duke@435 | 149 | size_t default_chunk_size(); |
duke@435 | 150 | // Adapt the chunk size to follow the allocation rate. |
duke@435 | 151 | size_t adaptive_chunk_size(int i, size_t limit); |
duke@435 | 152 | // Scan and free invalid pages. |
duke@435 | 153 | void scan_pages(size_t page_count); |
duke@435 | 154 | // Return the bottom_region and the top_region. Align them to page_size() boundary. |
duke@435 | 155 | // |------------------new_region---------------------------------| |
duke@435 | 156 | // |----bottom_region--|---intersection---|------top_region------| |
duke@435 | 157 | void select_tails(MemRegion new_region, MemRegion intersection, |
duke@435 | 158 | MemRegion* bottom_region, MemRegion *top_region); |
duke@435 | 159 | // Try to merge the invalid region with the bottom or top region by decreasing |
duke@435 | 160 | // the intersection area. Return the invalid_region aligned to the page_size() |
duke@435 | 161 | // boundary if it's inside the intersection. Return non-empty invalid_region |
duke@435 | 162 | // if it lies inside the intersection (also page-aligned). |
duke@435 | 163 | // |------------------new_region---------------------------------| |
duke@435 | 164 | // |----------------|-------invalid---|--------------------------| |
duke@435 | 165 | // |----bottom_region--|---intersection---|------top_region------| |
duke@435 | 166 | void merge_regions(MemRegion new_region, MemRegion* intersection, |
duke@435 | 167 | MemRegion *invalid_region); |
duke@435 | 168 | |
duke@435 | 169 | public: |
duke@435 | 170 | GrowableArray<LGRPSpace*>* lgrp_spaces() const { return _lgrp_spaces; } |
duke@435 | 171 | MutableNUMASpace(); |
duke@435 | 172 | virtual ~MutableNUMASpace(); |
duke@435 | 173 | // Space initialization. |
duke@435 | 174 | virtual void initialize(MemRegion mr, bool clear_space); |
duke@435 | 175 | // Update space layout if necessary. Do all adaptive resizing job. |
duke@435 | 176 | virtual void update(); |
duke@435 | 177 | // Update allocation rate averages. |
duke@435 | 178 | virtual void accumulate_statistics(); |
duke@435 | 179 | |
duke@435 | 180 | virtual void clear(); |
duke@435 | 181 | virtual void mangle_unused_area(); |
duke@435 | 182 | virtual void ensure_parsability(); |
duke@435 | 183 | virtual size_t used_in_words() const; |
duke@435 | 184 | virtual size_t free_in_words() const; |
duke@435 | 185 | virtual size_t tlab_capacity(Thread* thr) const; |
duke@435 | 186 | virtual size_t unsafe_max_tlab_alloc(Thread* thr) const; |
duke@435 | 187 | |
duke@435 | 188 | // Allocation (return NULL if full) |
duke@435 | 189 | virtual HeapWord* allocate(size_t word_size); |
duke@435 | 190 | virtual HeapWord* cas_allocate(size_t word_size); |
duke@435 | 191 | |
duke@435 | 192 | // Debugging |
duke@435 | 193 | virtual void print_on(outputStream* st) const; |
duke@435 | 194 | virtual void print_short_on(outputStream* st) const; |
duke@435 | 195 | virtual void verify(bool allow_dirty) const; |
duke@435 | 196 | |
duke@435 | 197 | virtual void set_top(HeapWord* value); |
duke@435 | 198 | }; |