1e7c033c3SPaolo Bonzini /* 2e7c033c3SPaolo Bonzini * Hierarchical Bitmap Data Type 3e7c033c3SPaolo Bonzini * 4e7c033c3SPaolo Bonzini * Copyright Red Hat, Inc., 2012 5e7c033c3SPaolo Bonzini * 6e7c033c3SPaolo Bonzini * Author: Paolo Bonzini <pbonzini@redhat.com> 7e7c033c3SPaolo Bonzini * 8e7c033c3SPaolo Bonzini * This work is licensed under the terms of the GNU GPL, version 2 or 9e7c033c3SPaolo Bonzini * later. See the COPYING file in the top-level directory. 10e7c033c3SPaolo Bonzini */ 11e7c033c3SPaolo Bonzini 12e7c033c3SPaolo Bonzini #include "qemu/osdep.h" 13e7c033c3SPaolo Bonzini #include "qemu/hbitmap.h" 14e7c033c3SPaolo Bonzini #include "qemu/host-utils.h" 15e7c033c3SPaolo Bonzini #include "trace.h" 16a3b52535SVladimir Sementsov-Ogievskiy #include "crypto/hash.h" 17e7c033c3SPaolo Bonzini 18e7c033c3SPaolo Bonzini /* HBitmaps provides an array of bits. The bits are stored as usual in an 19e7c033c3SPaolo Bonzini * array of unsigned longs, but HBitmap is also optimized to provide fast 20e7c033c3SPaolo Bonzini * iteration over set bits; going from one bit to the next is O(logB n) 21e7c033c3SPaolo Bonzini * worst case, with B = sizeof(long) * CHAR_BIT: the result is low enough 22e7c033c3SPaolo Bonzini * that the number of levels is in fact fixed. 23e7c033c3SPaolo Bonzini * 24e7c033c3SPaolo Bonzini * In order to do this, it stacks multiple bitmaps with progressively coarser 25e7c033c3SPaolo Bonzini * granularity; in all levels except the last, bit N is set iff the N-th 26e7c033c3SPaolo Bonzini * unsigned long is nonzero in the immediately next level. When iteration 27e7c033c3SPaolo Bonzini * completes on the last level it can examine the 2nd-last level to quickly 28e7c033c3SPaolo Bonzini * skip entire words, and even do so recursively to skip blocks of 64 words or 29e7c033c3SPaolo Bonzini * powers thereof (32 on 32-bit machines). 30e7c033c3SPaolo Bonzini * 31e7c033c3SPaolo Bonzini * Given an index in the bitmap, it can be split in group of bits like 32e7c033c3SPaolo Bonzini * this (for the 64-bit case): 33e7c033c3SPaolo Bonzini * 34e7c033c3SPaolo Bonzini * bits 0-57 => word in the last bitmap | bits 58-63 => bit in the word 35e7c033c3SPaolo Bonzini * bits 0-51 => word in the 2nd-last bitmap | bits 52-57 => bit in the word 36e7c033c3SPaolo Bonzini * bits 0-45 => word in the 3rd-last bitmap | bits 46-51 => bit in the word 37e7c033c3SPaolo Bonzini * 38e7c033c3SPaolo Bonzini * So it is easy to move up simply by shifting the index right by 39e7c033c3SPaolo Bonzini * log2(BITS_PER_LONG) bits. To move down, you shift the index left 40e7c033c3SPaolo Bonzini * similarly, and add the word index within the group. Iteration uses 41e7c033c3SPaolo Bonzini * ffs (find first set bit) to find the next word to examine; this 42e7c033c3SPaolo Bonzini * operation can be done in constant time in most current architectures. 43e7c033c3SPaolo Bonzini * 44e7c033c3SPaolo Bonzini * Setting or clearing a range of m bits on all levels, the work to perform 45e7c033c3SPaolo Bonzini * is O(m + m/W + m/W^2 + ...), which is O(m) like on a regular bitmap. 46e7c033c3SPaolo Bonzini * 47e7c033c3SPaolo Bonzini * When iterating on a bitmap, each bit (on any level) is only visited 48e7c033c3SPaolo Bonzini * once. Hence, The total cost of visiting a bitmap with m bits in it is 49e7c033c3SPaolo Bonzini * the number of bits that are set in all bitmaps. Unless the bitmap is 50e7c033c3SPaolo Bonzini * extremely sparse, this is also O(m + m/W + m/W^2 + ...), so the amortized 51e7c033c3SPaolo Bonzini * cost of advancing from one bit to the next is usually constant (worst case 52e7c033c3SPaolo Bonzini * O(logB n) as in the non-amortized complexity). 53e7c033c3SPaolo Bonzini */ 54e7c033c3SPaolo Bonzini 55e7c033c3SPaolo Bonzini struct HBitmap { 5676d570dcSVladimir Sementsov-Ogievskiy /* Size of the bitmap, as requested in hbitmap_alloc. */ 5776d570dcSVladimir Sementsov-Ogievskiy uint64_t orig_size; 5876d570dcSVladimir Sementsov-Ogievskiy 59e7c033c3SPaolo Bonzini /* Number of total bits in the bottom level. */ 60e7c033c3SPaolo Bonzini uint64_t size; 61e7c033c3SPaolo Bonzini 62e7c033c3SPaolo Bonzini /* Number of set bits in the bottom level. */ 63e7c033c3SPaolo Bonzini uint64_t count; 64e7c033c3SPaolo Bonzini 65e7c033c3SPaolo Bonzini /* A scaling factor. Given a granularity of G, each bit in the bitmap will 66e7c033c3SPaolo Bonzini * will actually represent a group of 2^G elements. Each operation on a 67e7c033c3SPaolo Bonzini * range of bits first rounds the bits to determine which group they land 68e7c033c3SPaolo Bonzini * in, and then affect the entire page; iteration will only visit the first 69e7c033c3SPaolo Bonzini * bit of each group. Here is an example of operations in a size-16, 70e7c033c3SPaolo Bonzini * granularity-1 HBitmap: 71e7c033c3SPaolo Bonzini * 72e7c033c3SPaolo Bonzini * initial state 00000000 73e7c033c3SPaolo Bonzini * set(start=0, count=9) 11111000 (iter: 0, 2, 4, 6, 8) 74e7c033c3SPaolo Bonzini * reset(start=1, count=3) 00111000 (iter: 4, 6, 8) 75e7c033c3SPaolo Bonzini * set(start=9, count=2) 00111100 (iter: 4, 6, 8, 10) 76e7c033c3SPaolo Bonzini * reset(start=5, count=5) 00000000 77e7c033c3SPaolo Bonzini * 78e7c033c3SPaolo Bonzini * From an implementation point of view, when setting or resetting bits, 79e7c033c3SPaolo Bonzini * the bitmap will scale bit numbers right by this amount of bits. When 80e7c033c3SPaolo Bonzini * iterating, the bitmap will scale bit numbers left by this amount of 81e7c033c3SPaolo Bonzini * bits. 82e7c033c3SPaolo Bonzini */ 83e7c033c3SPaolo Bonzini int granularity; 84e7c033c3SPaolo Bonzini 8507ac4cdbSFam Zheng /* A meta dirty bitmap to track the dirtiness of bits in this HBitmap. */ 8607ac4cdbSFam Zheng HBitmap *meta; 8707ac4cdbSFam Zheng 88e7c033c3SPaolo Bonzini /* A number of progressively less coarse bitmaps (i.e. level 0 is the 89e7c033c3SPaolo Bonzini * coarsest). Each bit in level N represents a word in level N+1 that 90e7c033c3SPaolo Bonzini * has a set bit, except the last level where each bit represents the 91e7c033c3SPaolo Bonzini * actual bitmap. 92e7c033c3SPaolo Bonzini * 93e7c033c3SPaolo Bonzini * Note that all bitmaps have the same number of levels. Even a 1-bit 94e7c033c3SPaolo Bonzini * bitmap will still allocate HBITMAP_LEVELS arrays. 95e7c033c3SPaolo Bonzini */ 96e7c033c3SPaolo Bonzini unsigned long *levels[HBITMAP_LEVELS]; 978515efbeSJohn Snow 988515efbeSJohn Snow /* The length of each levels[] array. */ 998515efbeSJohn Snow uint64_t sizes[HBITMAP_LEVELS]; 100e7c033c3SPaolo Bonzini }; 101e7c033c3SPaolo Bonzini 102e7c033c3SPaolo Bonzini /* Advance hbi to the next nonzero word and return it. hbi->pos 103e7c033c3SPaolo Bonzini * is updated. Returns zero if we reach the end of the bitmap. 104e7c033c3SPaolo Bonzini */ 105e7c033c3SPaolo Bonzini unsigned long hbitmap_iter_skip_words(HBitmapIter *hbi) 106e7c033c3SPaolo Bonzini { 107e7c033c3SPaolo Bonzini size_t pos = hbi->pos; 108e7c033c3SPaolo Bonzini const HBitmap *hb = hbi->hb; 109e7c033c3SPaolo Bonzini unsigned i = HBITMAP_LEVELS - 1; 110e7c033c3SPaolo Bonzini 111e7c033c3SPaolo Bonzini unsigned long cur; 112e7c033c3SPaolo Bonzini do { 113f63ea4e9SVladimir Sementsov-Ogievskiy i--; 114e7c033c3SPaolo Bonzini pos >>= BITS_PER_LEVEL; 115f63ea4e9SVladimir Sementsov-Ogievskiy cur = hbi->cur[i] & hb->levels[i][pos]; 116e7c033c3SPaolo Bonzini } while (cur == 0); 117e7c033c3SPaolo Bonzini 118e7c033c3SPaolo Bonzini /* Check for end of iteration. We always use fewer than BITS_PER_LONG 119e7c033c3SPaolo Bonzini * bits in the level 0 bitmap; thus we can repurpose the most significant 120e7c033c3SPaolo Bonzini * bit as a sentinel. The sentinel is set in hbitmap_alloc and ensures 121e7c033c3SPaolo Bonzini * that the above loop ends even without an explicit check on i. 122e7c033c3SPaolo Bonzini */ 123e7c033c3SPaolo Bonzini 124e7c033c3SPaolo Bonzini if (i == 0 && cur == (1UL << (BITS_PER_LONG - 1))) { 125e7c033c3SPaolo Bonzini return 0; 126e7c033c3SPaolo Bonzini } 127e7c033c3SPaolo Bonzini for (; i < HBITMAP_LEVELS - 1; i++) { 128e7c033c3SPaolo Bonzini /* Shift back pos to the left, matching the right shifts above. 129e7c033c3SPaolo Bonzini * The index of this word's least significant set bit provides 130e7c033c3SPaolo Bonzini * the low-order bits. 131e7c033c3SPaolo Bonzini */ 13218331e7cSRichard Henderson assert(cur); 13318331e7cSRichard Henderson pos = (pos << BITS_PER_LEVEL) + ctzl(cur); 134e7c033c3SPaolo Bonzini hbi->cur[i] = cur & (cur - 1); 135e7c033c3SPaolo Bonzini 136e7c033c3SPaolo Bonzini /* Set up next level for iteration. */ 137e7c033c3SPaolo Bonzini cur = hb->levels[i + 1][pos]; 138e7c033c3SPaolo Bonzini } 139e7c033c3SPaolo Bonzini 140e7c033c3SPaolo Bonzini hbi->pos = pos; 141e7c033c3SPaolo Bonzini trace_hbitmap_iter_skip_words(hbi->hb, hbi, pos, cur); 142e7c033c3SPaolo Bonzini 143e7c033c3SPaolo Bonzini assert(cur); 144e7c033c3SPaolo Bonzini return cur; 145e7c033c3SPaolo Bonzini } 146e7c033c3SPaolo Bonzini 147a33fbb4fSMax Reitz int64_t hbitmap_iter_next(HBitmapIter *hbi, bool advance) 148f63ea4e9SVladimir Sementsov-Ogievskiy { 149f63ea4e9SVladimir Sementsov-Ogievskiy unsigned long cur = hbi->cur[HBITMAP_LEVELS - 1] & 150f63ea4e9SVladimir Sementsov-Ogievskiy hbi->hb->levels[HBITMAP_LEVELS - 1][hbi->pos]; 151f63ea4e9SVladimir Sementsov-Ogievskiy int64_t item; 152f63ea4e9SVladimir Sementsov-Ogievskiy 153f63ea4e9SVladimir Sementsov-Ogievskiy if (cur == 0) { 154f63ea4e9SVladimir Sementsov-Ogievskiy cur = hbitmap_iter_skip_words(hbi); 155f63ea4e9SVladimir Sementsov-Ogievskiy if (cur == 0) { 156f63ea4e9SVladimir Sementsov-Ogievskiy return -1; 157f63ea4e9SVladimir Sementsov-Ogievskiy } 158f63ea4e9SVladimir Sementsov-Ogievskiy } 159f63ea4e9SVladimir Sementsov-Ogievskiy 160a33fbb4fSMax Reitz if (advance) { 161f63ea4e9SVladimir Sementsov-Ogievskiy /* The next call will resume work from the next bit. */ 162f63ea4e9SVladimir Sementsov-Ogievskiy hbi->cur[HBITMAP_LEVELS - 1] = cur & (cur - 1); 163a33fbb4fSMax Reitz } else { 164a33fbb4fSMax Reitz hbi->cur[HBITMAP_LEVELS - 1] = cur; 165a33fbb4fSMax Reitz } 166f63ea4e9SVladimir Sementsov-Ogievskiy item = ((uint64_t)hbi->pos << BITS_PER_LEVEL) + ctzl(cur); 167f63ea4e9SVladimir Sementsov-Ogievskiy 168f63ea4e9SVladimir Sementsov-Ogievskiy return item << hbi->granularity; 169f63ea4e9SVladimir Sementsov-Ogievskiy } 170f63ea4e9SVladimir Sementsov-Ogievskiy 171e7c033c3SPaolo Bonzini void hbitmap_iter_init(HBitmapIter *hbi, const HBitmap *hb, uint64_t first) 172e7c033c3SPaolo Bonzini { 173e7c033c3SPaolo Bonzini unsigned i, bit; 174e7c033c3SPaolo Bonzini uint64_t pos; 175e7c033c3SPaolo Bonzini 176e7c033c3SPaolo Bonzini hbi->hb = hb; 177e7c033c3SPaolo Bonzini pos = first >> hb->granularity; 1781b095244SPaolo Bonzini assert(pos < hb->size); 179e7c033c3SPaolo Bonzini hbi->pos = pos >> BITS_PER_LEVEL; 180e7c033c3SPaolo Bonzini hbi->granularity = hb->granularity; 181e7c033c3SPaolo Bonzini 182e7c033c3SPaolo Bonzini for (i = HBITMAP_LEVELS; i-- > 0; ) { 183e7c033c3SPaolo Bonzini bit = pos & (BITS_PER_LONG - 1); 184e7c033c3SPaolo Bonzini pos >>= BITS_PER_LEVEL; 185e7c033c3SPaolo Bonzini 186e7c033c3SPaolo Bonzini /* Drop bits representing items before first. */ 187e7c033c3SPaolo Bonzini hbi->cur[i] = hb->levels[i][pos] & ~((1UL << bit) - 1); 188e7c033c3SPaolo Bonzini 189e7c033c3SPaolo Bonzini /* We have already added level i+1, so the lowest set bit has 190e7c033c3SPaolo Bonzini * been processed. Clear it. 191e7c033c3SPaolo Bonzini */ 192e7c033c3SPaolo Bonzini if (i != HBITMAP_LEVELS - 1) { 193e7c033c3SPaolo Bonzini hbi->cur[i] &= ~(1UL << bit); 194e7c033c3SPaolo Bonzini } 195e7c033c3SPaolo Bonzini } 196e7c033c3SPaolo Bonzini } 197e7c033c3SPaolo Bonzini 19876d570dcSVladimir Sementsov-Ogievskiy int64_t hbitmap_next_zero(const HBitmap *hb, uint64_t start, uint64_t count) 19956207df5SVladimir Sementsov-Ogievskiy { 20056207df5SVladimir Sementsov-Ogievskiy size_t pos = (start >> hb->granularity) >> BITS_PER_LEVEL; 20156207df5SVladimir Sementsov-Ogievskiy unsigned long *last_lev = hb->levels[HBITMAP_LEVELS - 1]; 20256207df5SVladimir Sementsov-Ogievskiy unsigned long cur = last_lev[pos]; 20376d570dcSVladimir Sementsov-Ogievskiy unsigned start_bit_offset; 20476d570dcSVladimir Sementsov-Ogievskiy uint64_t end_bit, sz; 20556207df5SVladimir Sementsov-Ogievskiy int64_t res; 20656207df5SVladimir Sementsov-Ogievskiy 20776d570dcSVladimir Sementsov-Ogievskiy if (start >= hb->orig_size || count == 0) { 20876d570dcSVladimir Sementsov-Ogievskiy return -1; 20976d570dcSVladimir Sementsov-Ogievskiy } 21076d570dcSVladimir Sementsov-Ogievskiy 21176d570dcSVladimir Sementsov-Ogievskiy end_bit = count > hb->orig_size - start ? 21276d570dcSVladimir Sementsov-Ogievskiy hb->size : 21376d570dcSVladimir Sementsov-Ogievskiy ((start + count - 1) >> hb->granularity) + 1; 21476d570dcSVladimir Sementsov-Ogievskiy sz = (end_bit + BITS_PER_LONG - 1) >> BITS_PER_LEVEL; 21576d570dcSVladimir Sementsov-Ogievskiy 21676d570dcSVladimir Sementsov-Ogievskiy /* There may be some zero bits in @cur before @start. We are not interested 21776d570dcSVladimir Sementsov-Ogievskiy * in them, let's set them. 21876d570dcSVladimir Sementsov-Ogievskiy */ 21976d570dcSVladimir Sementsov-Ogievskiy start_bit_offset = (start >> hb->granularity) & (BITS_PER_LONG - 1); 22056207df5SVladimir Sementsov-Ogievskiy cur |= (1UL << start_bit_offset) - 1; 22156207df5SVladimir Sementsov-Ogievskiy assert((start >> hb->granularity) < hb->size); 22256207df5SVladimir Sementsov-Ogievskiy 22356207df5SVladimir Sementsov-Ogievskiy if (cur == (unsigned long)-1) { 22456207df5SVladimir Sementsov-Ogievskiy do { 22556207df5SVladimir Sementsov-Ogievskiy pos++; 22656207df5SVladimir Sementsov-Ogievskiy } while (pos < sz && last_lev[pos] == (unsigned long)-1); 22756207df5SVladimir Sementsov-Ogievskiy 22856207df5SVladimir Sementsov-Ogievskiy if (pos >= sz) { 22956207df5SVladimir Sementsov-Ogievskiy return -1; 23056207df5SVladimir Sementsov-Ogievskiy } 23156207df5SVladimir Sementsov-Ogievskiy 23256207df5SVladimir Sementsov-Ogievskiy cur = last_lev[pos]; 23356207df5SVladimir Sementsov-Ogievskiy } 23456207df5SVladimir Sementsov-Ogievskiy 23556207df5SVladimir Sementsov-Ogievskiy res = (pos << BITS_PER_LEVEL) + ctol(cur); 23676d570dcSVladimir Sementsov-Ogievskiy if (res >= end_bit) { 23756207df5SVladimir Sementsov-Ogievskiy return -1; 23856207df5SVladimir Sementsov-Ogievskiy } 23956207df5SVladimir Sementsov-Ogievskiy 24056207df5SVladimir Sementsov-Ogievskiy res = res << hb->granularity; 24156207df5SVladimir Sementsov-Ogievskiy if (res < start) { 24256207df5SVladimir Sementsov-Ogievskiy assert(((start - res) >> hb->granularity) == 0); 24356207df5SVladimir Sementsov-Ogievskiy return start; 24456207df5SVladimir Sementsov-Ogievskiy } 24556207df5SVladimir Sementsov-Ogievskiy 24656207df5SVladimir Sementsov-Ogievskiy return res; 24756207df5SVladimir Sementsov-Ogievskiy } 24856207df5SVladimir Sementsov-Ogievskiy 249*a78a1a48SVladimir Sementsov-Ogievskiy bool hbitmap_next_dirty_area(const HBitmap *hb, uint64_t *start, 250*a78a1a48SVladimir Sementsov-Ogievskiy uint64_t *count) 251*a78a1a48SVladimir Sementsov-Ogievskiy { 252*a78a1a48SVladimir Sementsov-Ogievskiy HBitmapIter hbi; 253*a78a1a48SVladimir Sementsov-Ogievskiy int64_t firt_dirty_off, area_end; 254*a78a1a48SVladimir Sementsov-Ogievskiy uint32_t granularity = 1UL << hb->granularity; 255*a78a1a48SVladimir Sementsov-Ogievskiy uint64_t end; 256*a78a1a48SVladimir Sementsov-Ogievskiy 257*a78a1a48SVladimir Sementsov-Ogievskiy if (*start >= hb->orig_size || *count == 0) { 258*a78a1a48SVladimir Sementsov-Ogievskiy return false; 259*a78a1a48SVladimir Sementsov-Ogievskiy } 260*a78a1a48SVladimir Sementsov-Ogievskiy 261*a78a1a48SVladimir Sementsov-Ogievskiy end = *count > hb->orig_size - *start ? hb->orig_size : *start + *count; 262*a78a1a48SVladimir Sementsov-Ogievskiy 263*a78a1a48SVladimir Sementsov-Ogievskiy hbitmap_iter_init(&hbi, hb, *start); 264*a78a1a48SVladimir Sementsov-Ogievskiy firt_dirty_off = hbitmap_iter_next(&hbi, false); 265*a78a1a48SVladimir Sementsov-Ogievskiy 266*a78a1a48SVladimir Sementsov-Ogievskiy if (firt_dirty_off < 0 || firt_dirty_off >= end) { 267*a78a1a48SVladimir Sementsov-Ogievskiy return false; 268*a78a1a48SVladimir Sementsov-Ogievskiy } 269*a78a1a48SVladimir Sementsov-Ogievskiy 270*a78a1a48SVladimir Sementsov-Ogievskiy if (firt_dirty_off + granularity >= end) { 271*a78a1a48SVladimir Sementsov-Ogievskiy area_end = end; 272*a78a1a48SVladimir Sementsov-Ogievskiy } else { 273*a78a1a48SVladimir Sementsov-Ogievskiy area_end = hbitmap_next_zero(hb, firt_dirty_off + granularity, 274*a78a1a48SVladimir Sementsov-Ogievskiy end - firt_dirty_off - granularity); 275*a78a1a48SVladimir Sementsov-Ogievskiy if (area_end < 0) { 276*a78a1a48SVladimir Sementsov-Ogievskiy area_end = end; 277*a78a1a48SVladimir Sementsov-Ogievskiy } 278*a78a1a48SVladimir Sementsov-Ogievskiy } 279*a78a1a48SVladimir Sementsov-Ogievskiy 280*a78a1a48SVladimir Sementsov-Ogievskiy if (firt_dirty_off > *start) { 281*a78a1a48SVladimir Sementsov-Ogievskiy *start = firt_dirty_off; 282*a78a1a48SVladimir Sementsov-Ogievskiy } 283*a78a1a48SVladimir Sementsov-Ogievskiy *count = area_end - *start; 284*a78a1a48SVladimir Sementsov-Ogievskiy 285*a78a1a48SVladimir Sementsov-Ogievskiy return true; 286*a78a1a48SVladimir Sementsov-Ogievskiy } 287*a78a1a48SVladimir Sementsov-Ogievskiy 288e7c033c3SPaolo Bonzini bool hbitmap_empty(const HBitmap *hb) 289e7c033c3SPaolo Bonzini { 290e7c033c3SPaolo Bonzini return hb->count == 0; 291e7c033c3SPaolo Bonzini } 292e7c033c3SPaolo Bonzini 293e7c033c3SPaolo Bonzini int hbitmap_granularity(const HBitmap *hb) 294e7c033c3SPaolo Bonzini { 295e7c033c3SPaolo Bonzini return hb->granularity; 296e7c033c3SPaolo Bonzini } 297e7c033c3SPaolo Bonzini 298e7c033c3SPaolo Bonzini uint64_t hbitmap_count(const HBitmap *hb) 299e7c033c3SPaolo Bonzini { 300e7c033c3SPaolo Bonzini return hb->count << hb->granularity; 301e7c033c3SPaolo Bonzini } 302e7c033c3SPaolo Bonzini 303e7c033c3SPaolo Bonzini /* Count the number of set bits between start and end, not accounting for 304e7c033c3SPaolo Bonzini * the granularity. Also an example of how to use hbitmap_iter_next_word. 305e7c033c3SPaolo Bonzini */ 306e7c033c3SPaolo Bonzini static uint64_t hb_count_between(HBitmap *hb, uint64_t start, uint64_t last) 307e7c033c3SPaolo Bonzini { 308e7c033c3SPaolo Bonzini HBitmapIter hbi; 309e7c033c3SPaolo Bonzini uint64_t count = 0; 310e7c033c3SPaolo Bonzini uint64_t end = last + 1; 311e7c033c3SPaolo Bonzini unsigned long cur; 312e7c033c3SPaolo Bonzini size_t pos; 313e7c033c3SPaolo Bonzini 314e7c033c3SPaolo Bonzini hbitmap_iter_init(&hbi, hb, start << hb->granularity); 315e7c033c3SPaolo Bonzini for (;;) { 316e7c033c3SPaolo Bonzini pos = hbitmap_iter_next_word(&hbi, &cur); 317e7c033c3SPaolo Bonzini if (pos >= (end >> BITS_PER_LEVEL)) { 318e7c033c3SPaolo Bonzini break; 319e7c033c3SPaolo Bonzini } 320591b320aSPeter Maydell count += ctpopl(cur); 321e7c033c3SPaolo Bonzini } 322e7c033c3SPaolo Bonzini 323e7c033c3SPaolo Bonzini if (pos == (end >> BITS_PER_LEVEL)) { 324e7c033c3SPaolo Bonzini /* Drop bits representing the END-th and subsequent items. */ 325e7c033c3SPaolo Bonzini int bit = end & (BITS_PER_LONG - 1); 326e7c033c3SPaolo Bonzini cur &= (1UL << bit) - 1; 327591b320aSPeter Maydell count += ctpopl(cur); 328e7c033c3SPaolo Bonzini } 329e7c033c3SPaolo Bonzini 330e7c033c3SPaolo Bonzini return count; 331e7c033c3SPaolo Bonzini } 332e7c033c3SPaolo Bonzini 333e7c033c3SPaolo Bonzini /* Setting starts at the last layer and propagates up if an element 33407ac4cdbSFam Zheng * changes. 335e7c033c3SPaolo Bonzini */ 336e7c033c3SPaolo Bonzini static inline bool hb_set_elem(unsigned long *elem, uint64_t start, uint64_t last) 337e7c033c3SPaolo Bonzini { 338e7c033c3SPaolo Bonzini unsigned long mask; 33907ac4cdbSFam Zheng unsigned long old; 340e7c033c3SPaolo Bonzini 341e7c033c3SPaolo Bonzini assert((last >> BITS_PER_LEVEL) == (start >> BITS_PER_LEVEL)); 342e7c033c3SPaolo Bonzini assert(start <= last); 343e7c033c3SPaolo Bonzini 344e7c033c3SPaolo Bonzini mask = 2UL << (last & (BITS_PER_LONG - 1)); 345e7c033c3SPaolo Bonzini mask -= 1UL << (start & (BITS_PER_LONG - 1)); 34607ac4cdbSFam Zheng old = *elem; 347e7c033c3SPaolo Bonzini *elem |= mask; 34807ac4cdbSFam Zheng return old != *elem; 349e7c033c3SPaolo Bonzini } 350e7c033c3SPaolo Bonzini 35107ac4cdbSFam Zheng /* The recursive workhorse (the depth is limited to HBITMAP_LEVELS)... 35207ac4cdbSFam Zheng * Returns true if at least one bit is changed. */ 35307ac4cdbSFam Zheng static bool hb_set_between(HBitmap *hb, int level, uint64_t start, 35407ac4cdbSFam Zheng uint64_t last) 355e7c033c3SPaolo Bonzini { 356e7c033c3SPaolo Bonzini size_t pos = start >> BITS_PER_LEVEL; 357e7c033c3SPaolo Bonzini size_t lastpos = last >> BITS_PER_LEVEL; 358e7c033c3SPaolo Bonzini bool changed = false; 359e7c033c3SPaolo Bonzini size_t i; 360e7c033c3SPaolo Bonzini 361e7c033c3SPaolo Bonzini i = pos; 362e7c033c3SPaolo Bonzini if (i < lastpos) { 363e7c033c3SPaolo Bonzini uint64_t next = (start | (BITS_PER_LONG - 1)) + 1; 364e7c033c3SPaolo Bonzini changed |= hb_set_elem(&hb->levels[level][i], start, next - 1); 365e7c033c3SPaolo Bonzini for (;;) { 366e7c033c3SPaolo Bonzini start = next; 367e7c033c3SPaolo Bonzini next += BITS_PER_LONG; 368e7c033c3SPaolo Bonzini if (++i == lastpos) { 369e7c033c3SPaolo Bonzini break; 370e7c033c3SPaolo Bonzini } 371e7c033c3SPaolo Bonzini changed |= (hb->levels[level][i] == 0); 372e7c033c3SPaolo Bonzini hb->levels[level][i] = ~0UL; 373e7c033c3SPaolo Bonzini } 374e7c033c3SPaolo Bonzini } 375e7c033c3SPaolo Bonzini changed |= hb_set_elem(&hb->levels[level][i], start, last); 376e7c033c3SPaolo Bonzini 377e7c033c3SPaolo Bonzini /* If there was any change in this layer, we may have to update 378e7c033c3SPaolo Bonzini * the one above. 379e7c033c3SPaolo Bonzini */ 380e7c033c3SPaolo Bonzini if (level > 0 && changed) { 381e7c033c3SPaolo Bonzini hb_set_between(hb, level - 1, pos, lastpos); 382e7c033c3SPaolo Bonzini } 38307ac4cdbSFam Zheng return changed; 384e7c033c3SPaolo Bonzini } 385e7c033c3SPaolo Bonzini 386e7c033c3SPaolo Bonzini void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count) 387e7c033c3SPaolo Bonzini { 388e7c033c3SPaolo Bonzini /* Compute range in the last layer. */ 38907ac4cdbSFam Zheng uint64_t first, n; 390e7c033c3SPaolo Bonzini uint64_t last = start + count - 1; 391e7c033c3SPaolo Bonzini 392e7c033c3SPaolo Bonzini trace_hbitmap_set(hb, start, count, 393e7c033c3SPaolo Bonzini start >> hb->granularity, last >> hb->granularity); 394e7c033c3SPaolo Bonzini 39507ac4cdbSFam Zheng first = start >> hb->granularity; 396e7c033c3SPaolo Bonzini last >>= hb->granularity; 3970e321191SVladimir Sementsov-Ogievskiy assert(last < hb->size); 39807ac4cdbSFam Zheng n = last - first + 1; 399e7c033c3SPaolo Bonzini 40007ac4cdbSFam Zheng hb->count += n - hb_count_between(hb, first, last); 40107ac4cdbSFam Zheng if (hb_set_between(hb, HBITMAP_LEVELS - 1, first, last) && 40207ac4cdbSFam Zheng hb->meta) { 40307ac4cdbSFam Zheng hbitmap_set(hb->meta, start, count); 40407ac4cdbSFam Zheng } 405e7c033c3SPaolo Bonzini } 406e7c033c3SPaolo Bonzini 407e7c033c3SPaolo Bonzini /* Resetting works the other way round: propagate up if the new 408e7c033c3SPaolo Bonzini * value is zero. 409e7c033c3SPaolo Bonzini */ 410e7c033c3SPaolo Bonzini static inline bool hb_reset_elem(unsigned long *elem, uint64_t start, uint64_t last) 411e7c033c3SPaolo Bonzini { 412e7c033c3SPaolo Bonzini unsigned long mask; 413e7c033c3SPaolo Bonzini bool blanked; 414e7c033c3SPaolo Bonzini 415e7c033c3SPaolo Bonzini assert((last >> BITS_PER_LEVEL) == (start >> BITS_PER_LEVEL)); 416e7c033c3SPaolo Bonzini assert(start <= last); 417e7c033c3SPaolo Bonzini 418e7c033c3SPaolo Bonzini mask = 2UL << (last & (BITS_PER_LONG - 1)); 419e7c033c3SPaolo Bonzini mask -= 1UL << (start & (BITS_PER_LONG - 1)); 420e7c033c3SPaolo Bonzini blanked = *elem != 0 && ((*elem & ~mask) == 0); 421e7c033c3SPaolo Bonzini *elem &= ~mask; 422e7c033c3SPaolo Bonzini return blanked; 423e7c033c3SPaolo Bonzini } 424e7c033c3SPaolo Bonzini 42507ac4cdbSFam Zheng /* The recursive workhorse (the depth is limited to HBITMAP_LEVELS)... 42607ac4cdbSFam Zheng * Returns true if at least one bit is changed. */ 42707ac4cdbSFam Zheng static bool hb_reset_between(HBitmap *hb, int level, uint64_t start, 42807ac4cdbSFam Zheng uint64_t last) 429e7c033c3SPaolo Bonzini { 430e7c033c3SPaolo Bonzini size_t pos = start >> BITS_PER_LEVEL; 431e7c033c3SPaolo Bonzini size_t lastpos = last >> BITS_PER_LEVEL; 432e7c033c3SPaolo Bonzini bool changed = false; 433e7c033c3SPaolo Bonzini size_t i; 434e7c033c3SPaolo Bonzini 435e7c033c3SPaolo Bonzini i = pos; 436e7c033c3SPaolo Bonzini if (i < lastpos) { 437e7c033c3SPaolo Bonzini uint64_t next = (start | (BITS_PER_LONG - 1)) + 1; 438e7c033c3SPaolo Bonzini 439e7c033c3SPaolo Bonzini /* Here we need a more complex test than when setting bits. Even if 440e7c033c3SPaolo Bonzini * something was changed, we must not blank bits in the upper level 441e7c033c3SPaolo Bonzini * unless the lower-level word became entirely zero. So, remove pos 442e7c033c3SPaolo Bonzini * from the upper-level range if bits remain set. 443e7c033c3SPaolo Bonzini */ 444e7c033c3SPaolo Bonzini if (hb_reset_elem(&hb->levels[level][i], start, next - 1)) { 445e7c033c3SPaolo Bonzini changed = true; 446e7c033c3SPaolo Bonzini } else { 447e7c033c3SPaolo Bonzini pos++; 448e7c033c3SPaolo Bonzini } 449e7c033c3SPaolo Bonzini 450e7c033c3SPaolo Bonzini for (;;) { 451e7c033c3SPaolo Bonzini start = next; 452e7c033c3SPaolo Bonzini next += BITS_PER_LONG; 453e7c033c3SPaolo Bonzini if (++i == lastpos) { 454e7c033c3SPaolo Bonzini break; 455e7c033c3SPaolo Bonzini } 456e7c033c3SPaolo Bonzini changed |= (hb->levels[level][i] != 0); 457e7c033c3SPaolo Bonzini hb->levels[level][i] = 0UL; 458e7c033c3SPaolo Bonzini } 459e7c033c3SPaolo Bonzini } 460e7c033c3SPaolo Bonzini 461e7c033c3SPaolo Bonzini /* Same as above, this time for lastpos. */ 462e7c033c3SPaolo Bonzini if (hb_reset_elem(&hb->levels[level][i], start, last)) { 463e7c033c3SPaolo Bonzini changed = true; 464e7c033c3SPaolo Bonzini } else { 465e7c033c3SPaolo Bonzini lastpos--; 466e7c033c3SPaolo Bonzini } 467e7c033c3SPaolo Bonzini 468e7c033c3SPaolo Bonzini if (level > 0 && changed) { 469e7c033c3SPaolo Bonzini hb_reset_between(hb, level - 1, pos, lastpos); 470e7c033c3SPaolo Bonzini } 47107ac4cdbSFam Zheng 47207ac4cdbSFam Zheng return changed; 47307ac4cdbSFam Zheng 474e7c033c3SPaolo Bonzini } 475e7c033c3SPaolo Bonzini 476e7c033c3SPaolo Bonzini void hbitmap_reset(HBitmap *hb, uint64_t start, uint64_t count) 477e7c033c3SPaolo Bonzini { 478e7c033c3SPaolo Bonzini /* Compute range in the last layer. */ 47907ac4cdbSFam Zheng uint64_t first; 480e7c033c3SPaolo Bonzini uint64_t last = start + count - 1; 481e7c033c3SPaolo Bonzini 482e7c033c3SPaolo Bonzini trace_hbitmap_reset(hb, start, count, 483e7c033c3SPaolo Bonzini start >> hb->granularity, last >> hb->granularity); 484e7c033c3SPaolo Bonzini 48507ac4cdbSFam Zheng first = start >> hb->granularity; 486e7c033c3SPaolo Bonzini last >>= hb->granularity; 4870e321191SVladimir Sementsov-Ogievskiy assert(last < hb->size); 488e7c033c3SPaolo Bonzini 48907ac4cdbSFam Zheng hb->count -= hb_count_between(hb, first, last); 49007ac4cdbSFam Zheng if (hb_reset_between(hb, HBITMAP_LEVELS - 1, first, last) && 49107ac4cdbSFam Zheng hb->meta) { 49207ac4cdbSFam Zheng hbitmap_set(hb->meta, start, count); 49307ac4cdbSFam Zheng } 494e7c033c3SPaolo Bonzini } 495e7c033c3SPaolo Bonzini 496c6a8c328SWen Congyang void hbitmap_reset_all(HBitmap *hb) 497c6a8c328SWen Congyang { 498c6a8c328SWen Congyang unsigned int i; 499c6a8c328SWen Congyang 500c6a8c328SWen Congyang /* Same as hbitmap_alloc() except for memset() instead of malloc() */ 501c6a8c328SWen Congyang for (i = HBITMAP_LEVELS; --i >= 1; ) { 502c6a8c328SWen Congyang memset(hb->levels[i], 0, hb->sizes[i] * sizeof(unsigned long)); 503c6a8c328SWen Congyang } 504c6a8c328SWen Congyang 505c6a8c328SWen Congyang hb->levels[0][0] = 1UL << (BITS_PER_LONG - 1); 506c6a8c328SWen Congyang hb->count = 0; 507c6a8c328SWen Congyang } 508c6a8c328SWen Congyang 50920a579deSMax Reitz bool hbitmap_is_serializable(const HBitmap *hb) 51020a579deSMax Reitz { 51120a579deSMax Reitz /* Every serialized chunk must be aligned to 64 bits so that endianness 51220a579deSMax Reitz * requirements can be fulfilled on both 64 bit and 32 bit hosts. 513ecbfa281SEric Blake * We have hbitmap_serialization_align() which converts this 51420a579deSMax Reitz * alignment requirement from bitmap bits to items covered (e.g. sectors). 51520a579deSMax Reitz * That value is: 51620a579deSMax Reitz * 64 << hb->granularity 51720a579deSMax Reitz * Since this value must not exceed UINT64_MAX, hb->granularity must be 51820a579deSMax Reitz * less than 58 (== 64 - 6, where 6 is ld(64), i.e. 1 << 6 == 64). 51920a579deSMax Reitz * 520ecbfa281SEric Blake * In order for hbitmap_serialization_align() to always return a 52120a579deSMax Reitz * meaningful value, bitmaps that are to be serialized must have a 52220a579deSMax Reitz * granularity of less than 58. */ 52320a579deSMax Reitz 52420a579deSMax Reitz return hb->granularity < 58; 52520a579deSMax Reitz } 52620a579deSMax Reitz 527e7c033c3SPaolo Bonzini bool hbitmap_get(const HBitmap *hb, uint64_t item) 528e7c033c3SPaolo Bonzini { 529e7c033c3SPaolo Bonzini /* Compute position and bit in the last layer. */ 530e7c033c3SPaolo Bonzini uint64_t pos = item >> hb->granularity; 531e7c033c3SPaolo Bonzini unsigned long bit = 1UL << (pos & (BITS_PER_LONG - 1)); 5320e321191SVladimir Sementsov-Ogievskiy assert(pos < hb->size); 533e7c033c3SPaolo Bonzini 534e7c033c3SPaolo Bonzini return (hb->levels[HBITMAP_LEVELS - 1][pos >> BITS_PER_LEVEL] & bit) != 0; 535e7c033c3SPaolo Bonzini } 536e7c033c3SPaolo Bonzini 537ecbfa281SEric Blake uint64_t hbitmap_serialization_align(const HBitmap *hb) 5388258888eSVladimir Sementsov-Ogievskiy { 53920a579deSMax Reitz assert(hbitmap_is_serializable(hb)); 5406725f887SMax Reitz 5418258888eSVladimir Sementsov-Ogievskiy /* Require at least 64 bit granularity to be safe on both 64 bit and 32 bit 5428258888eSVladimir Sementsov-Ogievskiy * hosts. */ 5436725f887SMax Reitz return UINT64_C(64) << hb->granularity; 5448258888eSVladimir Sementsov-Ogievskiy } 5458258888eSVladimir Sementsov-Ogievskiy 5468258888eSVladimir Sementsov-Ogievskiy /* Start should be aligned to serialization granularity, chunk size should be 5478258888eSVladimir Sementsov-Ogievskiy * aligned to serialization granularity too, except for last chunk. 5488258888eSVladimir Sementsov-Ogievskiy */ 5498258888eSVladimir Sementsov-Ogievskiy static void serialization_chunk(const HBitmap *hb, 5508258888eSVladimir Sementsov-Ogievskiy uint64_t start, uint64_t count, 5518258888eSVladimir Sementsov-Ogievskiy unsigned long **first_el, uint64_t *el_count) 5528258888eSVladimir Sementsov-Ogievskiy { 5538258888eSVladimir Sementsov-Ogievskiy uint64_t last = start + count - 1; 554ecbfa281SEric Blake uint64_t gran = hbitmap_serialization_align(hb); 5558258888eSVladimir Sementsov-Ogievskiy 5568258888eSVladimir Sementsov-Ogievskiy assert((start & (gran - 1)) == 0); 5578258888eSVladimir Sementsov-Ogievskiy assert((last >> hb->granularity) < hb->size); 5588258888eSVladimir Sementsov-Ogievskiy if ((last >> hb->granularity) != hb->size - 1) { 5598258888eSVladimir Sementsov-Ogievskiy assert((count & (gran - 1)) == 0); 5608258888eSVladimir Sementsov-Ogievskiy } 5618258888eSVladimir Sementsov-Ogievskiy 5628258888eSVladimir Sementsov-Ogievskiy start = (start >> hb->granularity) >> BITS_PER_LEVEL; 5638258888eSVladimir Sementsov-Ogievskiy last = (last >> hb->granularity) >> BITS_PER_LEVEL; 5648258888eSVladimir Sementsov-Ogievskiy 5658258888eSVladimir Sementsov-Ogievskiy *first_el = &hb->levels[HBITMAP_LEVELS - 1][start]; 5668258888eSVladimir Sementsov-Ogievskiy *el_count = last - start + 1; 5678258888eSVladimir Sementsov-Ogievskiy } 5688258888eSVladimir Sementsov-Ogievskiy 5698258888eSVladimir Sementsov-Ogievskiy uint64_t hbitmap_serialization_size(const HBitmap *hb, 5708258888eSVladimir Sementsov-Ogievskiy uint64_t start, uint64_t count) 5718258888eSVladimir Sementsov-Ogievskiy { 5728258888eSVladimir Sementsov-Ogievskiy uint64_t el_count; 5738258888eSVladimir Sementsov-Ogievskiy unsigned long *cur; 5748258888eSVladimir Sementsov-Ogievskiy 5758258888eSVladimir Sementsov-Ogievskiy if (!count) { 5768258888eSVladimir Sementsov-Ogievskiy return 0; 5778258888eSVladimir Sementsov-Ogievskiy } 5788258888eSVladimir Sementsov-Ogievskiy serialization_chunk(hb, start, count, &cur, &el_count); 5798258888eSVladimir Sementsov-Ogievskiy 5808258888eSVladimir Sementsov-Ogievskiy return el_count * sizeof(unsigned long); 5818258888eSVladimir Sementsov-Ogievskiy } 5828258888eSVladimir Sementsov-Ogievskiy 5838258888eSVladimir Sementsov-Ogievskiy void hbitmap_serialize_part(const HBitmap *hb, uint8_t *buf, 5848258888eSVladimir Sementsov-Ogievskiy uint64_t start, uint64_t count) 5858258888eSVladimir Sementsov-Ogievskiy { 5868258888eSVladimir Sementsov-Ogievskiy uint64_t el_count; 5878258888eSVladimir Sementsov-Ogievskiy unsigned long *cur, *end; 5888258888eSVladimir Sementsov-Ogievskiy 5898258888eSVladimir Sementsov-Ogievskiy if (!count) { 5908258888eSVladimir Sementsov-Ogievskiy return; 5918258888eSVladimir Sementsov-Ogievskiy } 5928258888eSVladimir Sementsov-Ogievskiy serialization_chunk(hb, start, count, &cur, &el_count); 5938258888eSVladimir Sementsov-Ogievskiy end = cur + el_count; 5948258888eSVladimir Sementsov-Ogievskiy 5958258888eSVladimir Sementsov-Ogievskiy while (cur != end) { 5968258888eSVladimir Sementsov-Ogievskiy unsigned long el = 5978258888eSVladimir Sementsov-Ogievskiy (BITS_PER_LONG == 32 ? cpu_to_le32(*cur) : cpu_to_le64(*cur)); 5988258888eSVladimir Sementsov-Ogievskiy 5998258888eSVladimir Sementsov-Ogievskiy memcpy(buf, &el, sizeof(el)); 6008258888eSVladimir Sementsov-Ogievskiy buf += sizeof(el); 6018258888eSVladimir Sementsov-Ogievskiy cur++; 6028258888eSVladimir Sementsov-Ogievskiy } 6038258888eSVladimir Sementsov-Ogievskiy } 6048258888eSVladimir Sementsov-Ogievskiy 6058258888eSVladimir Sementsov-Ogievskiy void hbitmap_deserialize_part(HBitmap *hb, uint8_t *buf, 6068258888eSVladimir Sementsov-Ogievskiy uint64_t start, uint64_t count, 6078258888eSVladimir Sementsov-Ogievskiy bool finish) 6088258888eSVladimir Sementsov-Ogievskiy { 6098258888eSVladimir Sementsov-Ogievskiy uint64_t el_count; 6108258888eSVladimir Sementsov-Ogievskiy unsigned long *cur, *end; 6118258888eSVladimir Sementsov-Ogievskiy 6128258888eSVladimir Sementsov-Ogievskiy if (!count) { 6138258888eSVladimir Sementsov-Ogievskiy return; 6148258888eSVladimir Sementsov-Ogievskiy } 6158258888eSVladimir Sementsov-Ogievskiy serialization_chunk(hb, start, count, &cur, &el_count); 6168258888eSVladimir Sementsov-Ogievskiy end = cur + el_count; 6178258888eSVladimir Sementsov-Ogievskiy 6188258888eSVladimir Sementsov-Ogievskiy while (cur != end) { 6198258888eSVladimir Sementsov-Ogievskiy memcpy(cur, buf, sizeof(*cur)); 6208258888eSVladimir Sementsov-Ogievskiy 6218258888eSVladimir Sementsov-Ogievskiy if (BITS_PER_LONG == 32) { 6228258888eSVladimir Sementsov-Ogievskiy le32_to_cpus((uint32_t *)cur); 6238258888eSVladimir Sementsov-Ogievskiy } else { 6248258888eSVladimir Sementsov-Ogievskiy le64_to_cpus((uint64_t *)cur); 6258258888eSVladimir Sementsov-Ogievskiy } 6268258888eSVladimir Sementsov-Ogievskiy 6278258888eSVladimir Sementsov-Ogievskiy buf += sizeof(unsigned long); 6288258888eSVladimir Sementsov-Ogievskiy cur++; 6298258888eSVladimir Sementsov-Ogievskiy } 6308258888eSVladimir Sementsov-Ogievskiy if (finish) { 6318258888eSVladimir Sementsov-Ogievskiy hbitmap_deserialize_finish(hb); 6328258888eSVladimir Sementsov-Ogievskiy } 6338258888eSVladimir Sementsov-Ogievskiy } 6348258888eSVladimir Sementsov-Ogievskiy 6358258888eSVladimir Sementsov-Ogievskiy void hbitmap_deserialize_zeroes(HBitmap *hb, uint64_t start, uint64_t count, 6368258888eSVladimir Sementsov-Ogievskiy bool finish) 6378258888eSVladimir Sementsov-Ogievskiy { 6388258888eSVladimir Sementsov-Ogievskiy uint64_t el_count; 6398258888eSVladimir Sementsov-Ogievskiy unsigned long *first; 6408258888eSVladimir Sementsov-Ogievskiy 6418258888eSVladimir Sementsov-Ogievskiy if (!count) { 6428258888eSVladimir Sementsov-Ogievskiy return; 6438258888eSVladimir Sementsov-Ogievskiy } 6448258888eSVladimir Sementsov-Ogievskiy serialization_chunk(hb, start, count, &first, &el_count); 6458258888eSVladimir Sementsov-Ogievskiy 6468258888eSVladimir Sementsov-Ogievskiy memset(first, 0, el_count * sizeof(unsigned long)); 6478258888eSVladimir Sementsov-Ogievskiy if (finish) { 6488258888eSVladimir Sementsov-Ogievskiy hbitmap_deserialize_finish(hb); 6498258888eSVladimir Sementsov-Ogievskiy } 6508258888eSVladimir Sementsov-Ogievskiy } 6518258888eSVladimir Sementsov-Ogievskiy 6526bdc8b71SVladimir Sementsov-Ogievskiy void hbitmap_deserialize_ones(HBitmap *hb, uint64_t start, uint64_t count, 6536bdc8b71SVladimir Sementsov-Ogievskiy bool finish) 6546bdc8b71SVladimir Sementsov-Ogievskiy { 6556bdc8b71SVladimir Sementsov-Ogievskiy uint64_t el_count; 6566bdc8b71SVladimir Sementsov-Ogievskiy unsigned long *first; 6576bdc8b71SVladimir Sementsov-Ogievskiy 6586bdc8b71SVladimir Sementsov-Ogievskiy if (!count) { 6596bdc8b71SVladimir Sementsov-Ogievskiy return; 6606bdc8b71SVladimir Sementsov-Ogievskiy } 6616bdc8b71SVladimir Sementsov-Ogievskiy serialization_chunk(hb, start, count, &first, &el_count); 6626bdc8b71SVladimir Sementsov-Ogievskiy 6636bdc8b71SVladimir Sementsov-Ogievskiy memset(first, 0xff, el_count * sizeof(unsigned long)); 6646bdc8b71SVladimir Sementsov-Ogievskiy if (finish) { 6656bdc8b71SVladimir Sementsov-Ogievskiy hbitmap_deserialize_finish(hb); 6666bdc8b71SVladimir Sementsov-Ogievskiy } 6676bdc8b71SVladimir Sementsov-Ogievskiy } 6686bdc8b71SVladimir Sementsov-Ogievskiy 6698258888eSVladimir Sementsov-Ogievskiy void hbitmap_deserialize_finish(HBitmap *bitmap) 6708258888eSVladimir Sementsov-Ogievskiy { 6718258888eSVladimir Sementsov-Ogievskiy int64_t i, size, prev_size; 6728258888eSVladimir Sementsov-Ogievskiy int lev; 6738258888eSVladimir Sementsov-Ogievskiy 6748258888eSVladimir Sementsov-Ogievskiy /* restore levels starting from penultimate to zero level, assuming 6758258888eSVladimir Sementsov-Ogievskiy * that the last level is ok */ 6768258888eSVladimir Sementsov-Ogievskiy size = MAX((bitmap->size + BITS_PER_LONG - 1) >> BITS_PER_LEVEL, 1); 6778258888eSVladimir Sementsov-Ogievskiy for (lev = HBITMAP_LEVELS - 1; lev-- > 0; ) { 6788258888eSVladimir Sementsov-Ogievskiy prev_size = size; 6798258888eSVladimir Sementsov-Ogievskiy size = MAX((size + BITS_PER_LONG - 1) >> BITS_PER_LEVEL, 1); 6808258888eSVladimir Sementsov-Ogievskiy memset(bitmap->levels[lev], 0, size * sizeof(unsigned long)); 6818258888eSVladimir Sementsov-Ogievskiy 6828258888eSVladimir Sementsov-Ogievskiy for (i = 0; i < prev_size; ++i) { 6838258888eSVladimir Sementsov-Ogievskiy if (bitmap->levels[lev + 1][i]) { 6848258888eSVladimir Sementsov-Ogievskiy bitmap->levels[lev][i >> BITS_PER_LEVEL] |= 6858258888eSVladimir Sementsov-Ogievskiy 1UL << (i & (BITS_PER_LONG - 1)); 6868258888eSVladimir Sementsov-Ogievskiy } 6878258888eSVladimir Sementsov-Ogievskiy } 6888258888eSVladimir Sementsov-Ogievskiy } 6898258888eSVladimir Sementsov-Ogievskiy 6908258888eSVladimir Sementsov-Ogievskiy bitmap->levels[0][0] |= 1UL << (BITS_PER_LONG - 1); 6913260cdffSLiang Li bitmap->count = hb_count_between(bitmap, 0, bitmap->size - 1); 6928258888eSVladimir Sementsov-Ogievskiy } 6938258888eSVladimir Sementsov-Ogievskiy 694e7c033c3SPaolo Bonzini void hbitmap_free(HBitmap *hb) 695e7c033c3SPaolo Bonzini { 696e7c033c3SPaolo Bonzini unsigned i; 69707ac4cdbSFam Zheng assert(!hb->meta); 698e7c033c3SPaolo Bonzini for (i = HBITMAP_LEVELS; i-- > 0; ) { 699e7c033c3SPaolo Bonzini g_free(hb->levels[i]); 700e7c033c3SPaolo Bonzini } 701e7c033c3SPaolo Bonzini g_free(hb); 702e7c033c3SPaolo Bonzini } 703e7c033c3SPaolo Bonzini 704e7c033c3SPaolo Bonzini HBitmap *hbitmap_alloc(uint64_t size, int granularity) 705e7c033c3SPaolo Bonzini { 706e1cf5582SMarkus Armbruster HBitmap *hb = g_new0(struct HBitmap, 1); 707e7c033c3SPaolo Bonzini unsigned i; 708e7c033c3SPaolo Bonzini 70976d570dcSVladimir Sementsov-Ogievskiy hb->orig_size = size; 71076d570dcSVladimir Sementsov-Ogievskiy 711e7c033c3SPaolo Bonzini assert(granularity >= 0 && granularity < 64); 712e7c033c3SPaolo Bonzini size = (size + (1ULL << granularity) - 1) >> granularity; 713e7c033c3SPaolo Bonzini assert(size <= ((uint64_t)1 << HBITMAP_LOG_MAX_SIZE)); 714e7c033c3SPaolo Bonzini 715e7c033c3SPaolo Bonzini hb->size = size; 716e7c033c3SPaolo Bonzini hb->granularity = granularity; 717e7c033c3SPaolo Bonzini for (i = HBITMAP_LEVELS; i-- > 0; ) { 718e7c033c3SPaolo Bonzini size = MAX((size + BITS_PER_LONG - 1) >> BITS_PER_LEVEL, 1); 7198515efbeSJohn Snow hb->sizes[i] = size; 720e1cf5582SMarkus Armbruster hb->levels[i] = g_new0(unsigned long, size); 721e7c033c3SPaolo Bonzini } 722e7c033c3SPaolo Bonzini 723e7c033c3SPaolo Bonzini /* We necessarily have free bits in level 0 due to the definition 724e7c033c3SPaolo Bonzini * of HBITMAP_LEVELS, so use one for a sentinel. This speeds up 725e7c033c3SPaolo Bonzini * hbitmap_iter_skip_words. 726e7c033c3SPaolo Bonzini */ 727e7c033c3SPaolo Bonzini assert(size == 1); 728e7c033c3SPaolo Bonzini hb->levels[0][0] |= 1UL << (BITS_PER_LONG - 1); 729e7c033c3SPaolo Bonzini return hb; 730e7c033c3SPaolo Bonzini } 731be58721dSJohn Snow 732ce1ffea8SJohn Snow void hbitmap_truncate(HBitmap *hb, uint64_t size) 733ce1ffea8SJohn Snow { 734ce1ffea8SJohn Snow bool shrink; 735ce1ffea8SJohn Snow unsigned i; 736ce1ffea8SJohn Snow uint64_t num_elements = size; 737ce1ffea8SJohn Snow uint64_t old; 738ce1ffea8SJohn Snow 739ce1ffea8SJohn Snow /* Size comes in as logical elements, adjust for granularity. */ 740ce1ffea8SJohn Snow size = (size + (1ULL << hb->granularity) - 1) >> hb->granularity; 741ce1ffea8SJohn Snow assert(size <= ((uint64_t)1 << HBITMAP_LOG_MAX_SIZE)); 742ce1ffea8SJohn Snow shrink = size < hb->size; 743ce1ffea8SJohn Snow 744ce1ffea8SJohn Snow /* bit sizes are identical; nothing to do. */ 745ce1ffea8SJohn Snow if (size == hb->size) { 746ce1ffea8SJohn Snow return; 747ce1ffea8SJohn Snow } 748ce1ffea8SJohn Snow 749ce1ffea8SJohn Snow /* If we're losing bits, let's clear those bits before we invalidate all of 750ce1ffea8SJohn Snow * our invariants. This helps keep the bitcount consistent, and will prevent 751ce1ffea8SJohn Snow * us from carrying around garbage bits beyond the end of the map. 752ce1ffea8SJohn Snow */ 753ce1ffea8SJohn Snow if (shrink) { 754ce1ffea8SJohn Snow /* Don't clear partial granularity groups; 755ce1ffea8SJohn Snow * start at the first full one. */ 7566725f887SMax Reitz uint64_t start = ROUND_UP(num_elements, UINT64_C(1) << hb->granularity); 757ce1ffea8SJohn Snow uint64_t fix_count = (hb->size << hb->granularity) - start; 758ce1ffea8SJohn Snow 759ce1ffea8SJohn Snow assert(fix_count); 760ce1ffea8SJohn Snow hbitmap_reset(hb, start, fix_count); 761ce1ffea8SJohn Snow } 762ce1ffea8SJohn Snow 763ce1ffea8SJohn Snow hb->size = size; 764ce1ffea8SJohn Snow for (i = HBITMAP_LEVELS; i-- > 0; ) { 765ce1ffea8SJohn Snow size = MAX(BITS_TO_LONGS(size), 1); 766ce1ffea8SJohn Snow if (hb->sizes[i] == size) { 767ce1ffea8SJohn Snow break; 768ce1ffea8SJohn Snow } 769ce1ffea8SJohn Snow old = hb->sizes[i]; 770ce1ffea8SJohn Snow hb->sizes[i] = size; 771ce1ffea8SJohn Snow hb->levels[i] = g_realloc(hb->levels[i], size * sizeof(unsigned long)); 772ce1ffea8SJohn Snow if (!shrink) { 773ce1ffea8SJohn Snow memset(&hb->levels[i][old], 0x00, 774ce1ffea8SJohn Snow (size - old) * sizeof(*hb->levels[i])); 775ce1ffea8SJohn Snow } 776ce1ffea8SJohn Snow } 77707ac4cdbSFam Zheng if (hb->meta) { 77807ac4cdbSFam Zheng hbitmap_truncate(hb->meta, hb->size << hb->granularity); 77907ac4cdbSFam Zheng } 780ce1ffea8SJohn Snow } 781ce1ffea8SJohn Snow 782fa000f2fSVladimir Sementsov-Ogievskiy bool hbitmap_can_merge(const HBitmap *a, const HBitmap *b) 783fa000f2fSVladimir Sementsov-Ogievskiy { 784fa000f2fSVladimir Sementsov-Ogievskiy return (a->size == b->size) && (a->granularity == b->granularity); 785fa000f2fSVladimir Sementsov-Ogievskiy } 786ce1ffea8SJohn Snow 787be58721dSJohn Snow /** 788be58721dSJohn Snow * Given HBitmaps A and B, let A := A (BITOR) B. 789be58721dSJohn Snow * Bitmap B will not be modified. 790be58721dSJohn Snow * 791be58721dSJohn Snow * @return true if the merge was successful, 792be58721dSJohn Snow * false if it was not attempted. 793be58721dSJohn Snow */ 794fa000f2fSVladimir Sementsov-Ogievskiy bool hbitmap_merge(const HBitmap *a, const HBitmap *b, HBitmap *result) 795be58721dSJohn Snow { 796be58721dSJohn Snow int i; 797be58721dSJohn Snow uint64_t j; 798be58721dSJohn Snow 799fa000f2fSVladimir Sementsov-Ogievskiy if (!hbitmap_can_merge(a, b) || !hbitmap_can_merge(a, result)) { 800be58721dSJohn Snow return false; 801be58721dSJohn Snow } 802fa000f2fSVladimir Sementsov-Ogievskiy assert(hbitmap_can_merge(b, result)); 803be58721dSJohn Snow 804be58721dSJohn Snow if (hbitmap_count(b) == 0) { 805be58721dSJohn Snow return true; 806be58721dSJohn Snow } 807be58721dSJohn Snow 808be58721dSJohn Snow /* This merge is O(size), as BITS_PER_LONG and HBITMAP_LEVELS are constant. 809be58721dSJohn Snow * It may be possible to improve running times for sparsely populated maps 810be58721dSJohn Snow * by using hbitmap_iter_next, but this is suboptimal for dense maps. 811be58721dSJohn Snow */ 812be58721dSJohn Snow for (i = HBITMAP_LEVELS - 1; i >= 0; i--) { 813be58721dSJohn Snow for (j = 0; j < a->sizes[i]; j++) { 814fa000f2fSVladimir Sementsov-Ogievskiy result->levels[i][j] = a->levels[i][j] | b->levels[i][j]; 815be58721dSJohn Snow } 816be58721dSJohn Snow } 817be58721dSJohn Snow 818d1dde714SEric Blake /* Recompute the dirty count */ 819d1dde714SEric Blake result->count = hb_count_between(result, 0, result->size - 1); 820d1dde714SEric Blake 821be58721dSJohn Snow return true; 822be58721dSJohn Snow } 82307ac4cdbSFam Zheng 82407ac4cdbSFam Zheng HBitmap *hbitmap_create_meta(HBitmap *hb, int chunk_size) 82507ac4cdbSFam Zheng { 82607ac4cdbSFam Zheng assert(!(chunk_size & (chunk_size - 1))); 82707ac4cdbSFam Zheng assert(!hb->meta); 82807ac4cdbSFam Zheng hb->meta = hbitmap_alloc(hb->size << hb->granularity, 82907ac4cdbSFam Zheng hb->granularity + ctz32(chunk_size)); 83007ac4cdbSFam Zheng return hb->meta; 83107ac4cdbSFam Zheng } 83207ac4cdbSFam Zheng 83307ac4cdbSFam Zheng void hbitmap_free_meta(HBitmap *hb) 83407ac4cdbSFam Zheng { 83507ac4cdbSFam Zheng assert(hb->meta); 83607ac4cdbSFam Zheng hbitmap_free(hb->meta); 83707ac4cdbSFam Zheng hb->meta = NULL; 83807ac4cdbSFam Zheng } 839a3b52535SVladimir Sementsov-Ogievskiy 840a3b52535SVladimir Sementsov-Ogievskiy char *hbitmap_sha256(const HBitmap *bitmap, Error **errp) 841a3b52535SVladimir Sementsov-Ogievskiy { 842a3b52535SVladimir Sementsov-Ogievskiy size_t size = bitmap->sizes[HBITMAP_LEVELS - 1] * sizeof(unsigned long); 843a3b52535SVladimir Sementsov-Ogievskiy char *data = (char *)bitmap->levels[HBITMAP_LEVELS - 1]; 844a3b52535SVladimir Sementsov-Ogievskiy char *hash = NULL; 845a3b52535SVladimir Sementsov-Ogievskiy qcrypto_hash_digest(QCRYPTO_HASH_ALG_SHA256, data, size, &hash, errp); 846a3b52535SVladimir Sementsov-Ogievskiy 847a3b52535SVladimir Sementsov-Ogievskiy return hash; 848a3b52535SVladimir Sementsov-Ogievskiy } 849