duke@435: /*
xdono@631:  * Copyright 2000-2008 Sun Microsystems, Inc.  All Rights Reserved.
duke@435:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435:  *
duke@435:  * This code is free software; you can redistribute it and/or modify it
duke@435:  * under the terms of the GNU General Public License version 2 only, as
duke@435:  * published by the Free Software Foundation.
duke@435:  *
duke@435:  * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
duke@435:  * version 2 for more details (a copy is included in the LICENSE file that
duke@435:  * accompanied this code).
duke@435:  *
duke@435:  * You should have received a copy of the GNU General Public License version
duke@435:  * 2 along with this work; if not, write to the Free Software Foundation,
duke@435:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435:  *
duke@435:  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
duke@435:  * CA 95054 USA or visit www.sun.com if you need additional information or
duke@435:  * have any questions.
duke@435:  *
duke@435:  */
duke@435: 
duke@435: // This kind of "BarrierSet" allows a "CollectedHeap" to detect and
duke@435: // enumerate ref fields that have been modified (since the last
duke@435: // enumeration.)
duke@435: 
duke@435: // As it currently stands, this barrier is *imprecise*: when a ref field in
duke@435: // an object "o" is modified, the card table entry for the card containing
duke@435: // the head of "o" is dirtied, not necessarily the card containing the
duke@435: // modified field itself.  For object arrays, however, the barrier *is*
duke@435: // precise; only the card containing the modified element is dirtied.
duke@435: // Any MemRegionClosures used to scan dirty cards should take these
duke@435: // considerations into account.
duke@435: 
duke@435: class Generation;
duke@435: class OopsInGenClosure;
duke@435: class DirtyCardToOopClosure;
duke@435: 
duke@435: class CardTableModRefBS: public ModRefBarrierSet {
duke@435:   // Some classes get to look at some private stuff.
duke@435:   friend class BytecodeInterpreter;
duke@435:   friend class VMStructs;
duke@435:   friend class CardTableRS;
duke@435:   friend class CheckForUnmarkedOops; // Needs access to raw card bytes.
duke@435: #ifndef PRODUCT
duke@435:   // For debugging.
duke@435:   friend class GuaranteeNotModClosure;
duke@435: #endif
duke@435:  protected:
duke@435: 
duke@435:   enum CardValues {
duke@435:     clean_card                  = -1,
duke@435:     dirty_card                  =  0,
duke@435:     precleaned_card             =  1,
duke@435:     last_card                   =  4,
duke@435:     CT_MR_BS_last_reserved      = 10
duke@435:   };
duke@435: 
duke@435:   // dirty and precleaned are equivalent wrt younger_refs_iter.
duke@435:   static bool card_is_dirty_wrt_gen_iter(jbyte cv) {
duke@435:     return cv == dirty_card || cv == precleaned_card;
duke@435:   }
duke@435: 
duke@435:   // Returns "true" iff the value "cv" will cause the card containing it
duke@435:   // to be scanned in the current traversal.  May be overridden by
duke@435:   // subtypes.
duke@435:   virtual bool card_will_be_scanned(jbyte cv) {
duke@435:     return CardTableModRefBS::card_is_dirty_wrt_gen_iter(cv);
duke@435:   }
duke@435: 
duke@435:   // Returns "true" iff the value "cv" may have represented a dirty card at
duke@435:   // some point.
duke@435:   virtual bool card_may_have_been_dirty(jbyte cv) {
duke@435:     return card_is_dirty_wrt_gen_iter(cv);
duke@435:   }
duke@435: 
duke@435:   // The declaration order of these const fields is important; see the
duke@435:   // constructor before changing.
duke@435:   const MemRegion _whole_heap;       // the region covered by the card table
duke@435:   const size_t    _guard_index;      // index of very last element in the card
duke@435:                                      // table; it is set to a guard value
duke@435:                                      // (last_card) and should never be modified
duke@435:   const size_t    _last_valid_index; // index of the last valid element
duke@435:   const size_t    _page_size;        // page size used when mapping _byte_map
duke@435:   const size_t    _byte_map_size;    // in bytes
duke@435:   jbyte*          _byte_map;         // the card marking array
duke@435: 
duke@435:   int _cur_covered_regions;
duke@435:   // The covered regions should be in address order.
duke@435:   MemRegion* _covered;
duke@435:   // The committed regions correspond one-to-one to the covered regions.
duke@435:   // They represent the card-table memory that has been committed to service
duke@435:   // the corresponding covered region.  It may be that committed region for
duke@435:   // one covered region corresponds to a larger region because of page-size
duke@435:   // roundings.  Thus, a committed region for one covered region may
duke@435:   // actually extend onto the card-table space for the next covered region.
duke@435:   MemRegion* _committed;
duke@435: 
duke@435:   // The last card is a guard card, and we commit the page for it so
duke@435:   // we can use the card for verification purposes. We make sure we never
duke@435:   // uncommit the MemRegion for that page.
duke@435:   MemRegion _guard_region;
duke@435: 
duke@435:  protected:
duke@435:   // Initialization utilities; covered_words is the size of the covered region
duke@435:   // in, um, words.
duke@435:   inline size_t cards_required(size_t covered_words);
duke@435:   inline size_t compute_byte_map_size();
duke@435: 
duke@435:   // Finds and return the index of the region, if any, to which the given
duke@435:   // region would be contiguous.  If none exists, assign a new region and
duke@435:   // returns its index.  Requires that no more than the maximum number of
duke@435:   // covered regions defined in the constructor are ever in use.
duke@435:   int find_covering_region_by_base(HeapWord* base);
duke@435: 
duke@435:   // Same as above, but finds the region containing the given address
duke@435:   // instead of starting at a given base address.
duke@435:   int find_covering_region_containing(HeapWord* addr);
duke@435: 
duke@435:   // Resize one of the regions covered by the remembered set.
duke@435:   void resize_covered_region(MemRegion new_region);
duke@435: 
duke@435:   // Returns the leftmost end of a committed region corresponding to a
duke@435:   // covered region before covered region "ind", or else "NULL" if "ind" is
duke@435:   // the first covered region.
duke@435:   HeapWord* largest_prev_committed_end(int ind) const;
duke@435: 
duke@435:   // Returns the part of the region mr that doesn't intersect with
duke@435:   // any committed region other than self.  Used to prevent uncommitting
duke@435:   // regions that are also committed by other regions.  Also protects
duke@435:   // against uncommitting the guard region.
duke@435:   MemRegion committed_unique_to_self(int self, MemRegion mr) const;
duke@435: 
duke@435:   // Mapping from address to card marking array entry
duke@435:   jbyte* byte_for(const void* p) const {
duke@435:     assert(_whole_heap.contains(p),
duke@435:            "out of bounds access to card marking array");
duke@435:     jbyte* result = &byte_map_base[uintptr_t(p) >> card_shift];
duke@435:     assert(result >= _byte_map && result < _byte_map + _byte_map_size,
duke@435:            "out of bounds accessor for card marking array");
duke@435:     return result;
duke@435:   }
duke@435: 
duke@435:   // The card table byte one after the card marking array
duke@435:   // entry for argument address. Typically used for higher bounds
duke@435:   // for loops iterating through the card table.
duke@435:   jbyte* byte_after(const void* p) const {
duke@435:     return byte_for(p) + 1;
duke@435:   }
duke@435: 
duke@435:   // Mapping from card marking array entry to address of first word
duke@435:   HeapWord* addr_for(const jbyte* p) const {
duke@435:     assert(p >= _byte_map && p < _byte_map + _byte_map_size,
duke@435:            "out of bounds access to card marking array");
duke@435:     size_t delta = pointer_delta(p, byte_map_base, sizeof(jbyte));
duke@435:     HeapWord* result = (HeapWord*) (delta << card_shift);
duke@435:     assert(_whole_heap.contains(result),
duke@435:            "out of bounds accessor from card marking array");
duke@435:     return result;
duke@435:   }
duke@435: 
duke@435:   // Iterate over the portion of the card-table which covers the given
duke@435:   // region mr in the given space and apply cl to any dirty sub-regions
duke@435:   // of mr. cl and dcto_cl must either be the same closure or cl must
duke@435:   // wrap dcto_cl. Both are required - neither may be NULL. Also, dcto_cl
duke@435:   // may be modified. Note that this function will operate in a parallel
duke@435:   // mode if worker threads are available.
duke@435:   void non_clean_card_iterate(Space* sp, MemRegion mr,
duke@435:                               DirtyCardToOopClosure* dcto_cl,
duke@435:                               MemRegionClosure* cl,
duke@435:                               bool clear);
duke@435: 
duke@435:   // Utility function used to implement the other versions below.
duke@435:   void non_clean_card_iterate_work(MemRegion mr, MemRegionClosure* cl,
duke@435:                                    bool clear);
duke@435: 
duke@435:   void par_non_clean_card_iterate_work(Space* sp, MemRegion mr,
duke@435:                                        DirtyCardToOopClosure* dcto_cl,
duke@435:                                        MemRegionClosure* cl,
duke@435:                                        bool clear,
duke@435:                                        int n_threads);
duke@435: 
duke@435:   // Dirty the bytes corresponding to "mr" (not all of which must be
duke@435:   // covered.)
duke@435:   void dirty_MemRegion(MemRegion mr);
duke@435: 
duke@435:   // Clear (to clean_card) the bytes entirely contained within "mr" (not
duke@435:   // all of which must be covered.)
duke@435:   void clear_MemRegion(MemRegion mr);
duke@435: 
duke@435:   // *** Support for parallel card scanning.
duke@435: 
duke@435:   enum SomeConstantsForParallelism {
duke@435:     StridesPerThread    = 2,
duke@435:     CardsPerStrideChunk = 256
duke@435:   };
duke@435: 
duke@435:   // This is an array, one element per covered region of the card table.
duke@435:   // Each entry is itself an array, with one element per chunk in the
duke@435:   // covered region.  Each entry of these arrays is the lowest non-clean
duke@435:   // card of the corresponding chunk containing part of an object from the
duke@435:   // previous chunk, or else NULL.
duke@435:   typedef jbyte*  CardPtr;
duke@435:   typedef CardPtr* CardArr;
duke@435:   CardArr* _lowest_non_clean;
duke@435:   size_t*  _lowest_non_clean_chunk_size;
duke@435:   uintptr_t* _lowest_non_clean_base_chunk_index;
duke@435:   int* _last_LNC_resizing_collection;
duke@435: 
duke@435:   // Initializes "lowest_non_clean" to point to the array for the region
duke@435:   // covering "sp", and "lowest_non_clean_base_chunk_index" to the chunk
duke@435:   // index of the corresponding to the first element of that array.
duke@435:   // Ensures that these arrays are of sufficient size, allocating if necessary.
duke@435:   // May be called by several threads concurrently.
duke@435:   void get_LNC_array_for_space(Space* sp,
duke@435:                                jbyte**& lowest_non_clean,
duke@435:                                uintptr_t& lowest_non_clean_base_chunk_index,
duke@435:                                size_t& lowest_non_clean_chunk_size);
duke@435: 
duke@435:   // Returns the number of chunks necessary to cover "mr".
duke@435:   size_t chunks_to_cover(MemRegion mr) {
duke@435:     return (size_t)(addr_to_chunk_index(mr.last()) -
duke@435:                     addr_to_chunk_index(mr.start()) + 1);
duke@435:   }
duke@435: 
duke@435:   // Returns the index of the chunk in a stride which
duke@435:   // covers the given address.
duke@435:   uintptr_t addr_to_chunk_index(const void* addr) {
duke@435:     uintptr_t card = (uintptr_t) byte_for(addr);
duke@435:     return card / CardsPerStrideChunk;
duke@435:   }
duke@435: 
duke@435:   // Apply cl, which must either itself apply dcto_cl or be dcto_cl,
duke@435:   // to the cards in the stride (of n_strides) within the given space.
duke@435:   void process_stride(Space* sp,
duke@435:                       MemRegion used,
duke@435:                       jint stride, int n_strides,
duke@435:                       DirtyCardToOopClosure* dcto_cl,
duke@435:                       MemRegionClosure* cl,
duke@435:                       bool clear,
duke@435:                       jbyte** lowest_non_clean,
duke@435:                       uintptr_t lowest_non_clean_base_chunk_index,
duke@435:                       size_t lowest_non_clean_chunk_size);
duke@435: 
duke@435:   // Makes sure that chunk boundaries are handled appropriately, by
duke@435:   // adjusting the min_done of dcto_cl, and by using a special card-table
duke@435:   // value to indicate how min_done should be set.
duke@435:   void process_chunk_boundaries(Space* sp,
duke@435:                                 DirtyCardToOopClosure* dcto_cl,
duke@435:                                 MemRegion chunk_mr,
duke@435:                                 MemRegion used,
duke@435:                                 jbyte** lowest_non_clean,
duke@435:                                 uintptr_t lowest_non_clean_base_chunk_index,
duke@435:                                 size_t    lowest_non_clean_chunk_size);
duke@435: 
duke@435: public:
duke@435:   // Constants
duke@435:   enum SomePublicConstants {
duke@435:     card_shift                  = 9,
duke@435:     card_size                   = 1 << card_shift,
duke@435:     card_size_in_words          = card_size / sizeof(HeapWord)
duke@435:   };
duke@435: 
duke@435:   // For RTTI simulation.
duke@435:   BarrierSet::Name kind() { return BarrierSet::CardTableModRef; }
duke@435:   bool is_a(BarrierSet::Name bsn) {
duke@435:     return bsn == BarrierSet::CardTableModRef || bsn == BarrierSet::ModRef;
duke@435:   }
duke@435: 
duke@435:   CardTableModRefBS(MemRegion whole_heap, int max_covered_regions);
duke@435: 
duke@435:   // *** Barrier set functions.
duke@435: 
coleenp@548:   inline bool write_ref_needs_barrier(void* field, oop new_val) {
duke@435:     // Note that this assumes the perm gen is the highest generation
duke@435:     // in the address space
duke@435:     return new_val != NULL && !new_val->is_perm();
duke@435:   }
duke@435: 
duke@435:   // Record a reference update. Note that these versions are precise!
duke@435:   // The scanning code has to handle the fact that the write barrier may be
duke@435:   // either precise or imprecise. We make non-virtual inline variants of
duke@435:   // these functions here for performance.
duke@435: protected:
duke@435:   void write_ref_field_work(oop obj, size_t offset, oop newVal);
coleenp@548:   void write_ref_field_work(void* field, oop newVal);
duke@435: public:
duke@435: 
duke@435:   bool has_write_ref_array_opt() { return true; }
duke@435:   bool has_write_region_opt() { return true; }
duke@435: 
duke@435:   inline void inline_write_region(MemRegion mr) {
duke@435:     dirty_MemRegion(mr);
duke@435:   }
duke@435: protected:
duke@435:   void write_region_work(MemRegion mr) {
duke@435:     inline_write_region(mr);
duke@435:   }
duke@435: public:
duke@435: 
duke@435:   inline void inline_write_ref_array(MemRegion mr) {
duke@435:     dirty_MemRegion(mr);
duke@435:   }
duke@435: protected:
duke@435:   void write_ref_array_work(MemRegion mr) {
duke@435:     inline_write_ref_array(mr);
duke@435:   }
duke@435: public:
duke@435: 
duke@435:   bool is_aligned(HeapWord* addr) {
duke@435:     return is_card_aligned(addr);
duke@435:   }
duke@435: 
duke@435:   // *** Card-table-barrier-specific things.
duke@435: 
coleenp@548:   inline void inline_write_ref_field(void* field, oop newVal) {
duke@435:     jbyte* byte = byte_for(field);
duke@435:     *byte = dirty_card;
duke@435:   }
duke@435: 
duke@435:   // Card marking array base (adjusted for heap low boundary)
duke@435:   // This would be the 0th element of _byte_map, if the heap started at 0x0.
duke@435:   // But since the heap starts at some higher address, this points to somewhere
duke@435:   // before the beginning of the actual _byte_map.
duke@435:   jbyte* byte_map_base;
duke@435: 
duke@435:   // Return true if "p" is at the start of a card.
duke@435:   bool is_card_aligned(HeapWord* p) {
duke@435:     jbyte* pcard = byte_for(p);
duke@435:     return (addr_for(pcard) == p);
duke@435:   }
duke@435: 
duke@435:   // The kinds of precision a CardTableModRefBS may offer.
duke@435:   enum PrecisionStyle {
duke@435:     Precise,
duke@435:     ObjHeadPreciseArray
duke@435:   };
duke@435: 
duke@435:   // Tells what style of precision this card table offers.
duke@435:   PrecisionStyle precision() {
duke@435:     return ObjHeadPreciseArray; // Only one supported for now.
duke@435:   }
duke@435: 
duke@435:   // ModRefBS functions.
duke@435:   void invalidate(MemRegion mr);
duke@435:   void clear(MemRegion mr);
duke@435:   void mod_oop_in_space_iterate(Space* sp, OopClosure* cl,
duke@435:                                 bool clear = false,
duke@435:                                 bool before_save_marks = false);
duke@435: 
duke@435:   // *** Card-table-RemSet-specific things.
duke@435: 
duke@435:   // Invoke "cl.do_MemRegion" on a set of MemRegions that collectively
duke@435:   // includes all the modified cards (expressing each card as a
duke@435:   // MemRegion).  Thus, several modified cards may be lumped into one
duke@435:   // region.  The regions are non-overlapping, and are visited in
duke@435:   // *decreasing* address order.  (This order aids with imprecise card
duke@435:   // marking, where a dirty card may cause scanning, and summarization
duke@435:   // marking, of objects that extend onto subsequent cards.)
duke@435:   // If "clear" is true, the card is (conceptually) marked unmodified before
duke@435:   // applying the closure.
duke@435:   void mod_card_iterate(MemRegionClosure* cl, bool clear = false) {
duke@435:     non_clean_card_iterate_work(_whole_heap, cl, clear);
duke@435:   }
duke@435: 
duke@435:   // Like the "mod_cards_iterate" above, except only invokes the closure
duke@435:   // for cards within the MemRegion "mr" (which is required to be
duke@435:   // card-aligned and sized.)
duke@435:   void mod_card_iterate(MemRegion mr, MemRegionClosure* cl,
duke@435:                         bool clear = false) {
duke@435:     non_clean_card_iterate_work(mr, cl, clear);
duke@435:   }
duke@435: 
duke@435:   static uintx ct_max_alignment_constraint();
duke@435: 
duke@435:   // Apply closure cl to the dirty cards lying completely
duke@435:   // within MemRegion mr, setting the cards to precleaned.
duke@435:   void      dirty_card_iterate(MemRegion mr, MemRegionClosure* cl);
duke@435: 
duke@435:   // Return the MemRegion corresponding to the first maximal run
duke@435:   // of dirty cards lying completely within MemRegion mr, after
duke@435:   // marking those cards precleaned.
duke@435:   MemRegion dirty_card_range_after_preclean(MemRegion mr);
duke@435: 
duke@435:   // Set all the dirty cards in the given region to precleaned state.
duke@435:   void preclean_dirty_cards(MemRegion mr);
duke@435: 
duke@435:   // Mapping from address to card marking array index.
duke@435:   int index_for(void* p) {
duke@435:     assert(_whole_heap.contains(p),
duke@435:            "out of bounds access to card marking array");
duke@435:     return byte_for(p) - _byte_map;
duke@435:   }
duke@435: 
duke@435:   void verify();
duke@435:   void verify_guard();
duke@435: 
duke@435:   void verify_clean_region(MemRegion mr) PRODUCT_RETURN;
duke@435: 
duke@435:   static size_t par_chunk_heapword_alignment() {
duke@435:     return CardsPerStrideChunk * card_size_in_words;
duke@435:   }
duke@435: };
duke@435: 
duke@435: class CardTableRS;
duke@435: 
duke@435: // A specialization for the CardTableRS gen rem set.
duke@435: class CardTableModRefBSForCTRS: public CardTableModRefBS {
duke@435:   CardTableRS* _rs;
duke@435: protected:
duke@435:   bool card_will_be_scanned(jbyte cv);
duke@435:   bool card_may_have_been_dirty(jbyte cv);
duke@435: public:
duke@435:   CardTableModRefBSForCTRS(MemRegion whole_heap,
duke@435:                            int max_covered_regions) :
duke@435:     CardTableModRefBS(whole_heap, max_covered_regions) {}
duke@435: 
duke@435:   void set_CTRS(CardTableRS* rs) { _rs = rs; }
duke@435: };