• file
• revisions
• annotate
• diff
• comparison
• raw

src/share/vm/gc_implementation/shared/mutableNUMASpace.hpp@0fbdb4381b99 (annotated)

src/share/vm/gc_implementation/shared/mutableNUMASpace.hpp

Mon, 09 Mar 2009 13:28:46 -0700

author

xdono

date

Mon, 09 Mar 2009 13:28:46 -0700

changeset 1014

0fbdb4381b99

parent 970

4e400c36026f

child 1907

c18cbe5936b8

permissions

-rw-r--r--

6814575: Update copyright year
Summary: Update copyright for files that have been modified in 2009, up to 03/09
Reviewed-by: katleman, tbell, ohair

duke@435	1	/*
xdono@1014	2	* Copyright 2006-2009 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;
iveresov@808	63	bool _allocation_failed;
duke@435	64
duke@435	65	struct SpaceStats {
duke@435	66	size_t _local_space, _remote_space, _unbiased_space, _uncommited_space;
duke@435	67	size_t _large_pages, _small_pages;
duke@435	68
duke@435	69	SpaceStats() {
duke@435	70	_local_space = 0;
duke@435	71	_remote_space = 0;
duke@435	72	_unbiased_space = 0;
duke@435	73	_uncommited_space = 0;
duke@435	74	_large_pages = 0;
duke@435	75	_small_pages = 0;
duke@435	76	}
duke@435	77	};
duke@435	78
duke@435	79	SpaceStats _space_stats;
duke@435	80
duke@435	81	char* _last_page_scanned;
duke@435	82	char* last_page_scanned() { return _last_page_scanned; }
duke@435	83	void set_last_page_scanned(char* p) { _last_page_scanned = p; }
duke@435	84	public:
iveresov@970	85	LGRPSpace(int l, size_t alignment) : _lgrp_id(l), _last_page_scanned(NULL), _allocation_failed(false) {
iveresov@970	86	_space = new MutableSpace(alignment);
duke@435	87	_alloc_rate = new AdaptiveWeightedAverage(NUMAChunkResizeWeight);
duke@435	88	}
duke@435	89	~LGRPSpace() {
duke@435	90	delete _space;
duke@435	91	delete _alloc_rate;
duke@435	92	}
duke@435	93
duke@435	94	void add_invalid_region(MemRegion r) {
duke@435	95	if (!_invalid_region.is_empty()) {
duke@435	96	_invalid_region.set_start(MIN2(_invalid_region.start(), r.start()));
duke@435	97	_invalid_region.set_end(MAX2(_invalid_region.end(), r.end()));
duke@435	98	} else {
duke@435	99	_invalid_region = r;
duke@435	100	}
duke@435	101	}
duke@435	102
duke@435	103	static bool equals(void* lgrp_id_value, LGRPSpace* p) {
duke@435	104	return *(int*)lgrp_id_value == p->lgrp_id();
duke@435	105	}
duke@435	106
iveresov@808	107	// Report a failed allocation.
iveresov@808	108	void set_allocation_failed() { _allocation_failed = true; }
iveresov@808	109
duke@435	110	void sample() {
iveresov@808	111	// If there was a failed allocation make allocation rate equal
iveresov@808	112	// to the size of the whole chunk. This ensures the progress of
iveresov@808	113	// the adaptation process.
iveresov@808	114	size_t alloc_rate_sample;
iveresov@808	115	if (_allocation_failed) {
iveresov@808	116	alloc_rate_sample = space()->capacity_in_bytes();
iveresov@808	117	_allocation_failed = false;
iveresov@808	118	} else {
iveresov@808	119	alloc_rate_sample = space()->used_in_bytes();
iveresov@808	120	}
iveresov@808	121	alloc_rate()->sample(alloc_rate_sample);
duke@435	122	}
duke@435	123
duke@435	124	MemRegion invalid_region() const { return _invalid_region; }
duke@435	125	void set_invalid_region(MemRegion r) { _invalid_region = r; }
duke@435	126	int lgrp_id() const { return _lgrp_id; }
duke@435	127	MutableSpace* space() const { return _space; }
duke@435	128	AdaptiveWeightedAverage* alloc_rate() const { return _alloc_rate; }
iveresov@703	129	void clear_alloc_rate() { _alloc_rate->clear(); }
duke@435	130	SpaceStats* space_stats() { return &_space_stats; }
duke@435	131	void clear_space_stats() { _space_stats = SpaceStats(); }
duke@435	132
duke@435	133	void accumulate_statistics(size_t page_size);
duke@435	134	void scan_pages(size_t page_size, size_t page_count);
duke@435	135	};
duke@435	136
duke@435	137	GrowableArray<LGRPSpace*>* _lgrp_spaces;
duke@435	138	size_t _page_size;
duke@435	139	unsigned _adaptation_cycles, _samples_count;
duke@435	140
duke@435	141	void set_page_size(size_t psz) { _page_size = psz; }
duke@435	142	size_t page_size() const { return _page_size; }
duke@435	143
duke@435	144	unsigned adaptation_cycles() { return _adaptation_cycles; }
duke@435	145	void set_adaptation_cycles(int v) { _adaptation_cycles = v; }
duke@435	146
duke@435	147	unsigned samples_count() { return _samples_count; }
duke@435	148	void increment_samples_count() { ++_samples_count; }
duke@435	149
duke@435	150	size_t _base_space_size;
duke@435	151	void set_base_space_size(size_t v) { _base_space_size = v; }
duke@435	152	size_t base_space_size() const { return _base_space_size; }
duke@435	153
duke@435	154	// Check if the NUMA topology has changed. Add and remove spaces if needed.
duke@435	155	// The update can be forced by setting the force parameter equal to true.
duke@435	156	bool update_layout(bool force);
iveresov@576	157	// Bias region towards the lgrp.
iveresov@576	158	void bias_region(MemRegion mr, int lgrp_id);
duke@435	159	// Free pages in a given region.
duke@435	160	void free_region(MemRegion mr);
duke@435	161	// Get current chunk size.
duke@435	162	size_t current_chunk_size(int i);
duke@435	163	// Get default chunk size (equally divide the space).
duke@435	164	size_t default_chunk_size();
duke@435	165	// Adapt the chunk size to follow the allocation rate.
duke@435	166	size_t adaptive_chunk_size(int i, size_t limit);
duke@435	167	// Scan and free invalid pages.
duke@435	168	void scan_pages(size_t page_count);
duke@435	169	// Return the bottom_region and the top_region. Align them to page_size() boundary.
duke@435	170	// |------------------new_region---------------------------------|
duke@435	171	// |----bottom_region--|---intersection---|------top_region------|
duke@435	172	void select_tails(MemRegion new_region, MemRegion intersection,
duke@435	173	MemRegion* bottom_region, MemRegion *top_region);
duke@435	174	// Try to merge the invalid region with the bottom or top region by decreasing
duke@435	175	// the intersection area. Return the invalid_region aligned to the page_size()
duke@435	176	// boundary if it's inside the intersection. Return non-empty invalid_region
duke@435	177	// if it lies inside the intersection (also page-aligned).
duke@435	178	// |------------------new_region---------------------------------|
duke@435	179	// |----------------|-------invalid---|--------------------------|
duke@435	180	// |----bottom_region--|---intersection---|------top_region------|
duke@435	181	void merge_regions(MemRegion new_region, MemRegion* intersection,
duke@435	182	MemRegion *invalid_region);
duke@435	183
duke@435	184	public:
duke@435	185	GrowableArray<LGRPSpace*>* lgrp_spaces() const { return _lgrp_spaces; }
iveresov@970	186	MutableNUMASpace(size_t alignment);
duke@435	187	virtual ~MutableNUMASpace();
duke@435	188	// Space initialization.
iveresov@970	189	virtual void initialize(MemRegion mr, bool clear_space, bool mangle_space, bool setup_pages = SetupPages);
duke@435	190	// Update space layout if necessary. Do all adaptive resizing job.
duke@435	191	virtual void update();
duke@435	192	// Update allocation rate averages.
duke@435	193	virtual void accumulate_statistics();
duke@435	194
jmasa@698	195	virtual void clear(bool mangle_space);
jmasa@698	196	virtual void mangle_unused_area() PRODUCT_RETURN;
jmasa@698	197	virtual void mangle_unused_area_complete() PRODUCT_RETURN;
jmasa@698	198	virtual void mangle_region(MemRegion mr) PRODUCT_RETURN;
jmasa@698	199	virtual void check_mangled_unused_area(HeapWord* limit) PRODUCT_RETURN;
jmasa@698	200	virtual void check_mangled_unused_area_complete() PRODUCT_RETURN;
jmasa@698	201	virtual void set_top_for_allocations(HeapWord* v) PRODUCT_RETURN;
jmasa@698	202	virtual void set_top_for_allocations() PRODUCT_RETURN;
jmasa@698	203
duke@435	204	virtual void ensure_parsability();
duke@435	205	virtual size_t used_in_words() const;
duke@435	206	virtual size_t free_in_words() const;
iveresov@808	207
iveresov@808	208	using MutableSpace::capacity_in_words;
iveresov@808	209	virtual size_t capacity_in_words(Thread* thr) const;
duke@435	210	virtual size_t tlab_capacity(Thread* thr) const;
duke@435	211	virtual size_t unsafe_max_tlab_alloc(Thread* thr) const;
duke@435	212
duke@435	213	// Allocation (return NULL if full)
duke@435	214	virtual HeapWord* allocate(size_t word_size);
duke@435	215	virtual HeapWord* cas_allocate(size_t word_size);
duke@435	216
duke@435	217	// Debugging
duke@435	218	virtual void print_on(outputStream* st) const;
duke@435	219	virtual void print_short_on(outputStream* st) const;
iveresov@625	220	virtual void verify(bool allow_dirty);
duke@435	221
duke@435	222	virtual void set_top(HeapWord* value);
duke@435	223	};