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 { 564e4de222SVladimir Sementsov-Ogievskiy /* 574e4de222SVladimir Sementsov-Ogievskiy * Size of the bitmap, as requested in hbitmap_alloc or in hbitmap_truncate. 584e4de222SVladimir Sementsov-Ogievskiy */ 5976d570dcSVladimir Sementsov-Ogievskiy uint64_t orig_size; 6076d570dcSVladimir Sementsov-Ogievskiy 61e7c033c3SPaolo Bonzini /* Number of total bits in the bottom level. */ 62e7c033c3SPaolo Bonzini uint64_t size; 63e7c033c3SPaolo Bonzini 64e7c033c3SPaolo Bonzini /* Number of set bits in the bottom level. */ 65e7c033c3SPaolo Bonzini uint64_t count; 66e7c033c3SPaolo Bonzini 67e7c033c3SPaolo Bonzini /* A scaling factor. Given a granularity of G, each bit in the bitmap will 68e7c033c3SPaolo Bonzini * will actually represent a group of 2^G elements. Each operation on a 69e7c033c3SPaolo Bonzini * range of bits first rounds the bits to determine which group they land 70e7c033c3SPaolo Bonzini * in, and then affect the entire page; iteration will only visit the first 71e7c033c3SPaolo Bonzini * bit of each group. Here is an example of operations in a size-16, 72e7c033c3SPaolo Bonzini * granularity-1 HBitmap: 73e7c033c3SPaolo Bonzini * 74e7c033c3SPaolo Bonzini * initial state 00000000 75e7c033c3SPaolo Bonzini * set(start=0, count=9) 11111000 (iter: 0, 2, 4, 6, 8) 76e7c033c3SPaolo Bonzini * reset(start=1, count=3) 00111000 (iter: 4, 6, 8) 77e7c033c3SPaolo Bonzini * set(start=9, count=2) 00111100 (iter: 4, 6, 8, 10) 78e7c033c3SPaolo Bonzini * reset(start=5, count=5) 00000000 79e7c033c3SPaolo Bonzini * 80e7c033c3SPaolo Bonzini * From an implementation point of view, when setting or resetting bits, 81e7c033c3SPaolo Bonzini * the bitmap will scale bit numbers right by this amount of bits. When 82e7c033c3SPaolo Bonzini * iterating, the bitmap will scale bit numbers left by this amount of 83e7c033c3SPaolo Bonzini * bits. 84e7c033c3SPaolo Bonzini */ 85e7c033c3SPaolo Bonzini int granularity; 86e7c033c3SPaolo Bonzini 8707ac4cdbSFam Zheng /* A meta dirty bitmap to track the dirtiness of bits in this HBitmap. */ 8807ac4cdbSFam Zheng HBitmap *meta; 8907ac4cdbSFam Zheng 90e7c033c3SPaolo Bonzini /* A number of progressively less coarse bitmaps (i.e. level 0 is the 91e7c033c3SPaolo Bonzini * coarsest). Each bit in level N represents a word in level N+1 that 92e7c033c3SPaolo Bonzini * has a set bit, except the last level where each bit represents the 93e7c033c3SPaolo Bonzini * actual bitmap. 94e7c033c3SPaolo Bonzini * 95e7c033c3SPaolo Bonzini * Note that all bitmaps have the same number of levels. Even a 1-bit 96e7c033c3SPaolo Bonzini * bitmap will still allocate HBITMAP_LEVELS arrays. 97e7c033c3SPaolo Bonzini */ 98e7c033c3SPaolo Bonzini unsigned long *levels[HBITMAP_LEVELS]; 998515efbeSJohn Snow 1008515efbeSJohn Snow /* The length of each levels[] array. */ 1018515efbeSJohn Snow uint64_t sizes[HBITMAP_LEVELS]; 102e7c033c3SPaolo Bonzini }; 103e7c033c3SPaolo Bonzini 104e7c033c3SPaolo Bonzini /* Advance hbi to the next nonzero word and return it. hbi->pos 105e7c033c3SPaolo Bonzini * is updated. Returns zero if we reach the end of the bitmap. 106e7c033c3SPaolo Bonzini */ 10730b8346cSVladimir Sementsov-Ogievskiy static unsigned long hbitmap_iter_skip_words(HBitmapIter *hbi) 108e7c033c3SPaolo Bonzini { 109e7c033c3SPaolo Bonzini size_t pos = hbi->pos; 110e7c033c3SPaolo Bonzini const HBitmap *hb = hbi->hb; 111e7c033c3SPaolo Bonzini unsigned i = HBITMAP_LEVELS - 1; 112e7c033c3SPaolo Bonzini 113e7c033c3SPaolo Bonzini unsigned long cur; 114e7c033c3SPaolo Bonzini do { 115f63ea4e9SVladimir Sementsov-Ogievskiy i--; 116e7c033c3SPaolo Bonzini pos >>= BITS_PER_LEVEL; 117f63ea4e9SVladimir Sementsov-Ogievskiy cur = hbi->cur[i] & hb->levels[i][pos]; 118e7c033c3SPaolo Bonzini } while (cur == 0); 119e7c033c3SPaolo Bonzini 120e7c033c3SPaolo Bonzini /* Check for end of iteration. We always use fewer than BITS_PER_LONG 121e7c033c3SPaolo Bonzini * bits in the level 0 bitmap; thus we can repurpose the most significant 122e7c033c3SPaolo Bonzini * bit as a sentinel. The sentinel is set in hbitmap_alloc and ensures 123e7c033c3SPaolo Bonzini * that the above loop ends even without an explicit check on i. 124e7c033c3SPaolo Bonzini */ 125e7c033c3SPaolo Bonzini 126e7c033c3SPaolo Bonzini if (i == 0 && cur == (1UL << (BITS_PER_LONG - 1))) { 127e7c033c3SPaolo Bonzini return 0; 128e7c033c3SPaolo Bonzini } 129e7c033c3SPaolo Bonzini for (; i < HBITMAP_LEVELS - 1; i++) { 130e7c033c3SPaolo Bonzini /* Shift back pos to the left, matching the right shifts above. 131e7c033c3SPaolo Bonzini * The index of this word's least significant set bit provides 132e7c033c3SPaolo Bonzini * the low-order bits. 133e7c033c3SPaolo Bonzini */ 13418331e7cSRichard Henderson assert(cur); 13518331e7cSRichard Henderson pos = (pos << BITS_PER_LEVEL) + ctzl(cur); 136e7c033c3SPaolo Bonzini hbi->cur[i] = cur & (cur - 1); 137e7c033c3SPaolo Bonzini 138e7c033c3SPaolo Bonzini /* Set up next level for iteration. */ 139e7c033c3SPaolo Bonzini cur = hb->levels[i + 1][pos]; 140e7c033c3SPaolo Bonzini } 141e7c033c3SPaolo Bonzini 142e7c033c3SPaolo Bonzini hbi->pos = pos; 143e7c033c3SPaolo Bonzini trace_hbitmap_iter_skip_words(hbi->hb, hbi, pos, cur); 144e7c033c3SPaolo Bonzini 145e7c033c3SPaolo Bonzini assert(cur); 146e7c033c3SPaolo Bonzini return cur; 147e7c033c3SPaolo Bonzini } 148e7c033c3SPaolo Bonzini 14919c021e1SVladimir Sementsov-Ogievskiy int64_t hbitmap_iter_next(HBitmapIter *hbi) 150f63ea4e9SVladimir Sementsov-Ogievskiy { 151f63ea4e9SVladimir Sementsov-Ogievskiy unsigned long cur = hbi->cur[HBITMAP_LEVELS - 1] & 152f63ea4e9SVladimir Sementsov-Ogievskiy hbi->hb->levels[HBITMAP_LEVELS - 1][hbi->pos]; 153f63ea4e9SVladimir Sementsov-Ogievskiy int64_t item; 154f63ea4e9SVladimir Sementsov-Ogievskiy 155f63ea4e9SVladimir Sementsov-Ogievskiy if (cur == 0) { 156f63ea4e9SVladimir Sementsov-Ogievskiy cur = hbitmap_iter_skip_words(hbi); 157f63ea4e9SVladimir Sementsov-Ogievskiy if (cur == 0) { 158f63ea4e9SVladimir Sementsov-Ogievskiy return -1; 159f63ea4e9SVladimir Sementsov-Ogievskiy } 160f63ea4e9SVladimir Sementsov-Ogievskiy } 161f63ea4e9SVladimir Sementsov-Ogievskiy 162f63ea4e9SVladimir Sementsov-Ogievskiy /* The next call will resume work from the next bit. */ 163f63ea4e9SVladimir Sementsov-Ogievskiy hbi->cur[HBITMAP_LEVELS - 1] = cur & (cur - 1); 164f63ea4e9SVladimir Sementsov-Ogievskiy item = ((uint64_t)hbi->pos << BITS_PER_LEVEL) + ctzl(cur); 165f63ea4e9SVladimir Sementsov-Ogievskiy 166f63ea4e9SVladimir Sementsov-Ogievskiy return item << hbi->granularity; 167f63ea4e9SVladimir Sementsov-Ogievskiy } 168f63ea4e9SVladimir Sementsov-Ogievskiy 169e7c033c3SPaolo Bonzini void hbitmap_iter_init(HBitmapIter *hbi, const HBitmap *hb, uint64_t first) 170e7c033c3SPaolo Bonzini { 171e7c033c3SPaolo Bonzini unsigned i, bit; 172e7c033c3SPaolo Bonzini uint64_t pos; 173e7c033c3SPaolo Bonzini 174e7c033c3SPaolo Bonzini hbi->hb = hb; 175e7c033c3SPaolo Bonzini pos = first >> hb->granularity; 1761b095244SPaolo Bonzini assert(pos < hb->size); 177e7c033c3SPaolo Bonzini hbi->pos = pos >> BITS_PER_LEVEL; 178e7c033c3SPaolo Bonzini hbi->granularity = hb->granularity; 179e7c033c3SPaolo Bonzini 180e7c033c3SPaolo Bonzini for (i = HBITMAP_LEVELS; i-- > 0; ) { 181e7c033c3SPaolo Bonzini bit = pos & (BITS_PER_LONG - 1); 182e7c033c3SPaolo Bonzini pos >>= BITS_PER_LEVEL; 183e7c033c3SPaolo Bonzini 184e7c033c3SPaolo Bonzini /* Drop bits representing items before first. */ 185e7c033c3SPaolo Bonzini hbi->cur[i] = hb->levels[i][pos] & ~((1UL << bit) - 1); 186e7c033c3SPaolo Bonzini 187e7c033c3SPaolo Bonzini /* We have already added level i+1, so the lowest set bit has 188e7c033c3SPaolo Bonzini * been processed. Clear it. 189e7c033c3SPaolo Bonzini */ 190e7c033c3SPaolo Bonzini if (i != HBITMAP_LEVELS - 1) { 191e7c033c3SPaolo Bonzini hbi->cur[i] &= ~(1UL << bit); 192e7c033c3SPaolo Bonzini } 193e7c033c3SPaolo Bonzini } 194e7c033c3SPaolo Bonzini } 195e7c033c3SPaolo Bonzini 196*9399c54bSVladimir Sementsov-Ogievskiy int64_t hbitmap_next_dirty(const HBitmap *hb, int64_t start, int64_t count) 197*9399c54bSVladimir Sementsov-Ogievskiy { 198*9399c54bSVladimir Sementsov-Ogievskiy HBitmapIter hbi; 199*9399c54bSVladimir Sementsov-Ogievskiy int64_t first_dirty_off; 200*9399c54bSVladimir Sementsov-Ogievskiy uint64_t end; 201*9399c54bSVladimir Sementsov-Ogievskiy 202*9399c54bSVladimir Sementsov-Ogievskiy assert(start >= 0 && count >= 0); 203*9399c54bSVladimir Sementsov-Ogievskiy 204*9399c54bSVladimir Sementsov-Ogievskiy if (start >= hb->orig_size || count == 0) { 205*9399c54bSVladimir Sementsov-Ogievskiy return -1; 206*9399c54bSVladimir Sementsov-Ogievskiy } 207*9399c54bSVladimir Sementsov-Ogievskiy 208*9399c54bSVladimir Sementsov-Ogievskiy end = count > hb->orig_size - start ? hb->orig_size : start + count; 209*9399c54bSVladimir Sementsov-Ogievskiy 210*9399c54bSVladimir Sementsov-Ogievskiy hbitmap_iter_init(&hbi, hb, start); 211*9399c54bSVladimir Sementsov-Ogievskiy first_dirty_off = hbitmap_iter_next(&hbi); 212*9399c54bSVladimir Sementsov-Ogievskiy 213*9399c54bSVladimir Sementsov-Ogievskiy if (first_dirty_off < 0 || first_dirty_off >= end) { 214*9399c54bSVladimir Sementsov-Ogievskiy return -1; 215*9399c54bSVladimir Sementsov-Ogievskiy } 216*9399c54bSVladimir Sementsov-Ogievskiy 217*9399c54bSVladimir Sementsov-Ogievskiy return MAX(start, first_dirty_off); 218*9399c54bSVladimir Sementsov-Ogievskiy } 219*9399c54bSVladimir Sementsov-Ogievskiy 220642700fdSVladimir Sementsov-Ogievskiy int64_t hbitmap_next_zero(const HBitmap *hb, int64_t start, int64_t count) 22156207df5SVladimir Sementsov-Ogievskiy { 22256207df5SVladimir Sementsov-Ogievskiy size_t pos = (start >> hb->granularity) >> BITS_PER_LEVEL; 22356207df5SVladimir Sementsov-Ogievskiy unsigned long *last_lev = hb->levels[HBITMAP_LEVELS - 1]; 22456207df5SVladimir Sementsov-Ogievskiy unsigned long cur = last_lev[pos]; 22576d570dcSVladimir Sementsov-Ogievskiy unsigned start_bit_offset; 22676d570dcSVladimir Sementsov-Ogievskiy uint64_t end_bit, sz; 22756207df5SVladimir Sementsov-Ogievskiy int64_t res; 22856207df5SVladimir Sementsov-Ogievskiy 229642700fdSVladimir Sementsov-Ogievskiy assert(start >= 0 && count >= 0); 230642700fdSVladimir Sementsov-Ogievskiy 23176d570dcSVladimir Sementsov-Ogievskiy if (start >= hb->orig_size || count == 0) { 23276d570dcSVladimir Sementsov-Ogievskiy return -1; 23376d570dcSVladimir Sementsov-Ogievskiy } 23476d570dcSVladimir Sementsov-Ogievskiy 23576d570dcSVladimir Sementsov-Ogievskiy end_bit = count > hb->orig_size - start ? 23676d570dcSVladimir Sementsov-Ogievskiy hb->size : 23776d570dcSVladimir Sementsov-Ogievskiy ((start + count - 1) >> hb->granularity) + 1; 23876d570dcSVladimir Sementsov-Ogievskiy sz = (end_bit + BITS_PER_LONG - 1) >> BITS_PER_LEVEL; 23976d570dcSVladimir Sementsov-Ogievskiy 24076d570dcSVladimir Sementsov-Ogievskiy /* There may be some zero bits in @cur before @start. We are not interested 24176d570dcSVladimir Sementsov-Ogievskiy * in them, let's set them. 24276d570dcSVladimir Sementsov-Ogievskiy */ 24376d570dcSVladimir Sementsov-Ogievskiy start_bit_offset = (start >> hb->granularity) & (BITS_PER_LONG - 1); 24456207df5SVladimir Sementsov-Ogievskiy cur |= (1UL << start_bit_offset) - 1; 24556207df5SVladimir Sementsov-Ogievskiy assert((start >> hb->granularity) < hb->size); 24656207df5SVladimir Sementsov-Ogievskiy 24756207df5SVladimir Sementsov-Ogievskiy if (cur == (unsigned long)-1) { 24856207df5SVladimir Sementsov-Ogievskiy do { 24956207df5SVladimir Sementsov-Ogievskiy pos++; 25056207df5SVladimir Sementsov-Ogievskiy } while (pos < sz && last_lev[pos] == (unsigned long)-1); 25156207df5SVladimir Sementsov-Ogievskiy 25256207df5SVladimir Sementsov-Ogievskiy if (pos >= sz) { 25356207df5SVladimir Sementsov-Ogievskiy return -1; 25456207df5SVladimir Sementsov-Ogievskiy } 25556207df5SVladimir Sementsov-Ogievskiy 25656207df5SVladimir Sementsov-Ogievskiy cur = last_lev[pos]; 25756207df5SVladimir Sementsov-Ogievskiy } 25856207df5SVladimir Sementsov-Ogievskiy 25956207df5SVladimir Sementsov-Ogievskiy res = (pos << BITS_PER_LEVEL) + ctol(cur); 26076d570dcSVladimir Sementsov-Ogievskiy if (res >= end_bit) { 26156207df5SVladimir Sementsov-Ogievskiy return -1; 26256207df5SVladimir Sementsov-Ogievskiy } 26356207df5SVladimir Sementsov-Ogievskiy 26456207df5SVladimir Sementsov-Ogievskiy res = res << hb->granularity; 26556207df5SVladimir Sementsov-Ogievskiy if (res < start) { 26656207df5SVladimir Sementsov-Ogievskiy assert(((start - res) >> hb->granularity) == 0); 26756207df5SVladimir Sementsov-Ogievskiy return start; 26856207df5SVladimir Sementsov-Ogievskiy } 26956207df5SVladimir Sementsov-Ogievskiy 27056207df5SVladimir Sementsov-Ogievskiy return res; 27156207df5SVladimir Sementsov-Ogievskiy } 27256207df5SVladimir Sementsov-Ogievskiy 273642700fdSVladimir Sementsov-Ogievskiy bool hbitmap_next_dirty_area(const HBitmap *hb, int64_t *start, int64_t *count) 274a78a1a48SVladimir Sementsov-Ogievskiy { 275*9399c54bSVladimir Sementsov-Ogievskiy int64_t area_start, area_end; 276a78a1a48SVladimir Sementsov-Ogievskiy 277*9399c54bSVladimir Sementsov-Ogievskiy area_start = hbitmap_next_dirty(hb, *start, *count); 278*9399c54bSVladimir Sementsov-Ogievskiy if (area_start < 0) { 279a78a1a48SVladimir Sementsov-Ogievskiy return false; 280a78a1a48SVladimir Sementsov-Ogievskiy } 281a78a1a48SVladimir Sementsov-Ogievskiy 282*9399c54bSVladimir Sementsov-Ogievskiy area_end = hbitmap_next_zero(hb, area_start, *start + *count - area_start); 283a78a1a48SVladimir Sementsov-Ogievskiy if (area_end < 0) { 284*9399c54bSVladimir Sementsov-Ogievskiy area_end = MIN(hb->orig_size, *start + *count); 285a78a1a48SVladimir Sementsov-Ogievskiy } 286a78a1a48SVladimir Sementsov-Ogievskiy 287*9399c54bSVladimir Sementsov-Ogievskiy *start = area_start; 288*9399c54bSVladimir Sementsov-Ogievskiy *count = area_end - area_start; 289a78a1a48SVladimir Sementsov-Ogievskiy 290a78a1a48SVladimir Sementsov-Ogievskiy return true; 291a78a1a48SVladimir Sementsov-Ogievskiy } 292a78a1a48SVladimir Sementsov-Ogievskiy 293e7c033c3SPaolo Bonzini bool hbitmap_empty(const HBitmap *hb) 294e7c033c3SPaolo Bonzini { 295e7c033c3SPaolo Bonzini return hb->count == 0; 296e7c033c3SPaolo Bonzini } 297e7c033c3SPaolo Bonzini 298e7c033c3SPaolo Bonzini int hbitmap_granularity(const HBitmap *hb) 299e7c033c3SPaolo Bonzini { 300e7c033c3SPaolo Bonzini return hb->granularity; 301e7c033c3SPaolo Bonzini } 302e7c033c3SPaolo Bonzini 303e7c033c3SPaolo Bonzini uint64_t hbitmap_count(const HBitmap *hb) 304e7c033c3SPaolo Bonzini { 305e7c033c3SPaolo Bonzini return hb->count << hb->granularity; 306e7c033c3SPaolo Bonzini } 307e7c033c3SPaolo Bonzini 308be24c714SVladimir Sementsov-Ogievskiy /** 309be24c714SVladimir Sementsov-Ogievskiy * hbitmap_iter_next_word: 310be24c714SVladimir Sementsov-Ogievskiy * @hbi: HBitmapIter to operate on. 311be24c714SVladimir Sementsov-Ogievskiy * @p_cur: Location where to store the next non-zero word. 312be24c714SVladimir Sementsov-Ogievskiy * 313be24c714SVladimir Sementsov-Ogievskiy * Return the index of the next nonzero word that is set in @hbi's 314be24c714SVladimir Sementsov-Ogievskiy * associated HBitmap, and set *p_cur to the content of that word 315be24c714SVladimir Sementsov-Ogievskiy * (bits before the index that was passed to hbitmap_iter_init are 316be24c714SVladimir Sementsov-Ogievskiy * trimmed on the first call). Return -1, and set *p_cur to zero, 317be24c714SVladimir Sementsov-Ogievskiy * if all remaining words are zero. 318be24c714SVladimir Sementsov-Ogievskiy */ 319be24c714SVladimir Sementsov-Ogievskiy static size_t hbitmap_iter_next_word(HBitmapIter *hbi, unsigned long *p_cur) 320be24c714SVladimir Sementsov-Ogievskiy { 321be24c714SVladimir Sementsov-Ogievskiy unsigned long cur = hbi->cur[HBITMAP_LEVELS - 1]; 322be24c714SVladimir Sementsov-Ogievskiy 323be24c714SVladimir Sementsov-Ogievskiy if (cur == 0) { 324be24c714SVladimir Sementsov-Ogievskiy cur = hbitmap_iter_skip_words(hbi); 325be24c714SVladimir Sementsov-Ogievskiy if (cur == 0) { 326be24c714SVladimir Sementsov-Ogievskiy *p_cur = 0; 327be24c714SVladimir Sementsov-Ogievskiy return -1; 328be24c714SVladimir Sementsov-Ogievskiy } 329be24c714SVladimir Sementsov-Ogievskiy } 330be24c714SVladimir Sementsov-Ogievskiy 331be24c714SVladimir Sementsov-Ogievskiy /* The next call will resume work from the next word. */ 332be24c714SVladimir Sementsov-Ogievskiy hbi->cur[HBITMAP_LEVELS - 1] = 0; 333be24c714SVladimir Sementsov-Ogievskiy *p_cur = cur; 334be24c714SVladimir Sementsov-Ogievskiy return hbi->pos; 335be24c714SVladimir Sementsov-Ogievskiy } 336be24c714SVladimir Sementsov-Ogievskiy 337e7c033c3SPaolo Bonzini /* Count the number of set bits between start and end, not accounting for 338e7c033c3SPaolo Bonzini * the granularity. Also an example of how to use hbitmap_iter_next_word. 339e7c033c3SPaolo Bonzini */ 340e7c033c3SPaolo Bonzini static uint64_t hb_count_between(HBitmap *hb, uint64_t start, uint64_t last) 341e7c033c3SPaolo Bonzini { 342e7c033c3SPaolo Bonzini HBitmapIter hbi; 343e7c033c3SPaolo Bonzini uint64_t count = 0; 344e7c033c3SPaolo Bonzini uint64_t end = last + 1; 345e7c033c3SPaolo Bonzini unsigned long cur; 346e7c033c3SPaolo Bonzini size_t pos; 347e7c033c3SPaolo Bonzini 348e7c033c3SPaolo Bonzini hbitmap_iter_init(&hbi, hb, start << hb->granularity); 349e7c033c3SPaolo Bonzini for (;;) { 350e7c033c3SPaolo Bonzini pos = hbitmap_iter_next_word(&hbi, &cur); 351e7c033c3SPaolo Bonzini if (pos >= (end >> BITS_PER_LEVEL)) { 352e7c033c3SPaolo Bonzini break; 353e7c033c3SPaolo Bonzini } 354591b320aSPeter Maydell count += ctpopl(cur); 355e7c033c3SPaolo Bonzini } 356e7c033c3SPaolo Bonzini 357e7c033c3SPaolo Bonzini if (pos == (end >> BITS_PER_LEVEL)) { 358e7c033c3SPaolo Bonzini /* Drop bits representing the END-th and subsequent items. */ 359e7c033c3SPaolo Bonzini int bit = end & (BITS_PER_LONG - 1); 360e7c033c3SPaolo Bonzini cur &= (1UL << bit) - 1; 361591b320aSPeter Maydell count += ctpopl(cur); 362e7c033c3SPaolo Bonzini } 363e7c033c3SPaolo Bonzini 364e7c033c3SPaolo Bonzini return count; 365e7c033c3SPaolo Bonzini } 366e7c033c3SPaolo Bonzini 367e7c033c3SPaolo Bonzini /* Setting starts at the last layer and propagates up if an element 36807ac4cdbSFam Zheng * changes. 369e7c033c3SPaolo Bonzini */ 370e7c033c3SPaolo Bonzini static inline bool hb_set_elem(unsigned long *elem, uint64_t start, uint64_t last) 371e7c033c3SPaolo Bonzini { 372e7c033c3SPaolo Bonzini unsigned long mask; 37307ac4cdbSFam Zheng unsigned long old; 374e7c033c3SPaolo Bonzini 375e7c033c3SPaolo Bonzini assert((last >> BITS_PER_LEVEL) == (start >> BITS_PER_LEVEL)); 376e7c033c3SPaolo Bonzini assert(start <= last); 377e7c033c3SPaolo Bonzini 378e7c033c3SPaolo Bonzini mask = 2UL << (last & (BITS_PER_LONG - 1)); 379e7c033c3SPaolo Bonzini mask -= 1UL << (start & (BITS_PER_LONG - 1)); 38007ac4cdbSFam Zheng old = *elem; 381e7c033c3SPaolo Bonzini *elem |= mask; 38207ac4cdbSFam Zheng return old != *elem; 383e7c033c3SPaolo Bonzini } 384e7c033c3SPaolo Bonzini 38507ac4cdbSFam Zheng /* The recursive workhorse (the depth is limited to HBITMAP_LEVELS)... 38607ac4cdbSFam Zheng * Returns true if at least one bit is changed. */ 38707ac4cdbSFam Zheng static bool hb_set_between(HBitmap *hb, int level, uint64_t start, 38807ac4cdbSFam Zheng uint64_t last) 389e7c033c3SPaolo Bonzini { 390e7c033c3SPaolo Bonzini size_t pos = start >> BITS_PER_LEVEL; 391e7c033c3SPaolo Bonzini size_t lastpos = last >> BITS_PER_LEVEL; 392e7c033c3SPaolo Bonzini bool changed = false; 393e7c033c3SPaolo Bonzini size_t i; 394e7c033c3SPaolo Bonzini 395e7c033c3SPaolo Bonzini i = pos; 396e7c033c3SPaolo Bonzini if (i < lastpos) { 397e7c033c3SPaolo Bonzini uint64_t next = (start | (BITS_PER_LONG - 1)) + 1; 398e7c033c3SPaolo Bonzini changed |= hb_set_elem(&hb->levels[level][i], start, next - 1); 399e7c033c3SPaolo Bonzini for (;;) { 400e7c033c3SPaolo Bonzini start = next; 401e7c033c3SPaolo Bonzini next += BITS_PER_LONG; 402e7c033c3SPaolo Bonzini if (++i == lastpos) { 403e7c033c3SPaolo Bonzini break; 404e7c033c3SPaolo Bonzini } 405e7c033c3SPaolo Bonzini changed |= (hb->levels[level][i] == 0); 406e7c033c3SPaolo Bonzini hb->levels[level][i] = ~0UL; 407e7c033c3SPaolo Bonzini } 408e7c033c3SPaolo Bonzini } 409e7c033c3SPaolo Bonzini changed |= hb_set_elem(&hb->levels[level][i], start, last); 410e7c033c3SPaolo Bonzini 411e7c033c3SPaolo Bonzini /* If there was any change in this layer, we may have to update 412e7c033c3SPaolo Bonzini * the one above. 413e7c033c3SPaolo Bonzini */ 414e7c033c3SPaolo Bonzini if (level > 0 && changed) { 415e7c033c3SPaolo Bonzini hb_set_between(hb, level - 1, pos, lastpos); 416e7c033c3SPaolo Bonzini } 41707ac4cdbSFam Zheng return changed; 418e7c033c3SPaolo Bonzini } 419e7c033c3SPaolo Bonzini 420e7c033c3SPaolo Bonzini void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count) 421e7c033c3SPaolo Bonzini { 422e7c033c3SPaolo Bonzini /* Compute range in the last layer. */ 42307ac4cdbSFam Zheng uint64_t first, n; 424e7c033c3SPaolo Bonzini uint64_t last = start + count - 1; 425e7c033c3SPaolo Bonzini 426fed33bd1SVladimir Sementsov-Ogievskiy if (count == 0) { 427fed33bd1SVladimir Sementsov-Ogievskiy return; 428fed33bd1SVladimir Sementsov-Ogievskiy } 429fed33bd1SVladimir Sementsov-Ogievskiy 430e7c033c3SPaolo Bonzini trace_hbitmap_set(hb, start, count, 431e7c033c3SPaolo Bonzini start >> hb->granularity, last >> hb->granularity); 432e7c033c3SPaolo Bonzini 43307ac4cdbSFam Zheng first = start >> hb->granularity; 434e7c033c3SPaolo Bonzini last >>= hb->granularity; 4350e321191SVladimir Sementsov-Ogievskiy assert(last < hb->size); 43607ac4cdbSFam Zheng n = last - first + 1; 437e7c033c3SPaolo Bonzini 43807ac4cdbSFam Zheng hb->count += n - hb_count_between(hb, first, last); 43907ac4cdbSFam Zheng if (hb_set_between(hb, HBITMAP_LEVELS - 1, first, last) && 44007ac4cdbSFam Zheng hb->meta) { 44107ac4cdbSFam Zheng hbitmap_set(hb->meta, start, count); 44207ac4cdbSFam Zheng } 443e7c033c3SPaolo Bonzini } 444e7c033c3SPaolo Bonzini 445e7c033c3SPaolo Bonzini /* Resetting works the other way round: propagate up if the new 446e7c033c3SPaolo Bonzini * value is zero. 447e7c033c3SPaolo Bonzini */ 448e7c033c3SPaolo Bonzini static inline bool hb_reset_elem(unsigned long *elem, uint64_t start, uint64_t last) 449e7c033c3SPaolo Bonzini { 450e7c033c3SPaolo Bonzini unsigned long mask; 451e7c033c3SPaolo Bonzini bool blanked; 452e7c033c3SPaolo Bonzini 453e7c033c3SPaolo Bonzini assert((last >> BITS_PER_LEVEL) == (start >> BITS_PER_LEVEL)); 454e7c033c3SPaolo Bonzini assert(start <= last); 455e7c033c3SPaolo Bonzini 456e7c033c3SPaolo Bonzini mask = 2UL << (last & (BITS_PER_LONG - 1)); 457e7c033c3SPaolo Bonzini mask -= 1UL << (start & (BITS_PER_LONG - 1)); 458e7c033c3SPaolo Bonzini blanked = *elem != 0 && ((*elem & ~mask) == 0); 459e7c033c3SPaolo Bonzini *elem &= ~mask; 460e7c033c3SPaolo Bonzini return blanked; 461e7c033c3SPaolo Bonzini } 462e7c033c3SPaolo Bonzini 46307ac4cdbSFam Zheng /* The recursive workhorse (the depth is limited to HBITMAP_LEVELS)... 46407ac4cdbSFam Zheng * Returns true if at least one bit is changed. */ 46507ac4cdbSFam Zheng static bool hb_reset_between(HBitmap *hb, int level, uint64_t start, 46607ac4cdbSFam Zheng uint64_t last) 467e7c033c3SPaolo Bonzini { 468e7c033c3SPaolo Bonzini size_t pos = start >> BITS_PER_LEVEL; 469e7c033c3SPaolo Bonzini size_t lastpos = last >> BITS_PER_LEVEL; 470e7c033c3SPaolo Bonzini bool changed = false; 471e7c033c3SPaolo Bonzini size_t i; 472e7c033c3SPaolo Bonzini 473e7c033c3SPaolo Bonzini i = pos; 474e7c033c3SPaolo Bonzini if (i < lastpos) { 475e7c033c3SPaolo Bonzini uint64_t next = (start | (BITS_PER_LONG - 1)) + 1; 476e7c033c3SPaolo Bonzini 477e7c033c3SPaolo Bonzini /* Here we need a more complex test than when setting bits. Even if 478e7c033c3SPaolo Bonzini * something was changed, we must not blank bits in the upper level 479e7c033c3SPaolo Bonzini * unless the lower-level word became entirely zero. So, remove pos 480e7c033c3SPaolo Bonzini * from the upper-level range if bits remain set. 481e7c033c3SPaolo Bonzini */ 482e7c033c3SPaolo Bonzini if (hb_reset_elem(&hb->levels[level][i], start, next - 1)) { 483e7c033c3SPaolo Bonzini changed = true; 484e7c033c3SPaolo Bonzini } else { 485e7c033c3SPaolo Bonzini pos++; 486e7c033c3SPaolo Bonzini } 487e7c033c3SPaolo Bonzini 488e7c033c3SPaolo Bonzini for (;;) { 489e7c033c3SPaolo Bonzini start = next; 490e7c033c3SPaolo Bonzini next += BITS_PER_LONG; 491e7c033c3SPaolo Bonzini if (++i == lastpos) { 492e7c033c3SPaolo Bonzini break; 493e7c033c3SPaolo Bonzini } 494e7c033c3SPaolo Bonzini changed |= (hb->levels[level][i] != 0); 495e7c033c3SPaolo Bonzini hb->levels[level][i] = 0UL; 496e7c033c3SPaolo Bonzini } 497e7c033c3SPaolo Bonzini } 498e7c033c3SPaolo Bonzini 499e7c033c3SPaolo Bonzini /* Same as above, this time for lastpos. */ 500e7c033c3SPaolo Bonzini if (hb_reset_elem(&hb->levels[level][i], start, last)) { 501e7c033c3SPaolo Bonzini changed = true; 502e7c033c3SPaolo Bonzini } else { 503e7c033c3SPaolo Bonzini lastpos--; 504e7c033c3SPaolo Bonzini } 505e7c033c3SPaolo Bonzini 506e7c033c3SPaolo Bonzini if (level > 0 && changed) { 507e7c033c3SPaolo Bonzini hb_reset_between(hb, level - 1, pos, lastpos); 508e7c033c3SPaolo Bonzini } 50907ac4cdbSFam Zheng 51007ac4cdbSFam Zheng return changed; 51107ac4cdbSFam Zheng 512e7c033c3SPaolo Bonzini } 513e7c033c3SPaolo Bonzini 514e7c033c3SPaolo Bonzini void hbitmap_reset(HBitmap *hb, uint64_t start, uint64_t count) 515e7c033c3SPaolo Bonzini { 516e7c033c3SPaolo Bonzini /* Compute range in the last layer. */ 51707ac4cdbSFam Zheng uint64_t first; 518e7c033c3SPaolo Bonzini uint64_t last = start + count - 1; 51948557b13SVladimir Sementsov-Ogievskiy uint64_t gran = 1ULL << hb->granularity; 52048557b13SVladimir Sementsov-Ogievskiy 521fed33bd1SVladimir Sementsov-Ogievskiy if (count == 0) { 522fed33bd1SVladimir Sementsov-Ogievskiy return; 523fed33bd1SVladimir Sementsov-Ogievskiy } 524fed33bd1SVladimir Sementsov-Ogievskiy 52548557b13SVladimir Sementsov-Ogievskiy assert(QEMU_IS_ALIGNED(start, gran)); 52648557b13SVladimir Sementsov-Ogievskiy assert(QEMU_IS_ALIGNED(count, gran) || (start + count == hb->orig_size)); 527e7c033c3SPaolo Bonzini 528e7c033c3SPaolo Bonzini trace_hbitmap_reset(hb, start, count, 529e7c033c3SPaolo Bonzini start >> hb->granularity, last >> hb->granularity); 530e7c033c3SPaolo Bonzini 53107ac4cdbSFam Zheng first = start >> hb->granularity; 532e7c033c3SPaolo Bonzini last >>= hb->granularity; 5330e321191SVladimir Sementsov-Ogievskiy assert(last < hb->size); 534e7c033c3SPaolo Bonzini 53507ac4cdbSFam Zheng hb->count -= hb_count_between(hb, first, last); 53607ac4cdbSFam Zheng if (hb_reset_between(hb, HBITMAP_LEVELS - 1, first, last) && 53707ac4cdbSFam Zheng hb->meta) { 53807ac4cdbSFam Zheng hbitmap_set(hb->meta, start, count); 53907ac4cdbSFam Zheng } 540e7c033c3SPaolo Bonzini } 541e7c033c3SPaolo Bonzini 542c6a8c328SWen Congyang void hbitmap_reset_all(HBitmap *hb) 543c6a8c328SWen Congyang { 544c6a8c328SWen Congyang unsigned int i; 545c6a8c328SWen Congyang 546c6a8c328SWen Congyang /* Same as hbitmap_alloc() except for memset() instead of malloc() */ 547c6a8c328SWen Congyang for (i = HBITMAP_LEVELS; --i >= 1; ) { 548c6a8c328SWen Congyang memset(hb->levels[i], 0, hb->sizes[i] * sizeof(unsigned long)); 549c6a8c328SWen Congyang } 550c6a8c328SWen Congyang 551c6a8c328SWen Congyang hb->levels[0][0] = 1UL << (BITS_PER_LONG - 1); 552c6a8c328SWen Congyang hb->count = 0; 553c6a8c328SWen Congyang } 554c6a8c328SWen Congyang 55520a579deSMax Reitz bool hbitmap_is_serializable(const HBitmap *hb) 55620a579deSMax Reitz { 55720a579deSMax Reitz /* Every serialized chunk must be aligned to 64 bits so that endianness 55820a579deSMax Reitz * requirements can be fulfilled on both 64 bit and 32 bit hosts. 559ecbfa281SEric Blake * We have hbitmap_serialization_align() which converts this 56020a579deSMax Reitz * alignment requirement from bitmap bits to items covered (e.g. sectors). 56120a579deSMax Reitz * That value is: 56220a579deSMax Reitz * 64 << hb->granularity 56320a579deSMax Reitz * Since this value must not exceed UINT64_MAX, hb->granularity must be 56420a579deSMax Reitz * less than 58 (== 64 - 6, where 6 is ld(64), i.e. 1 << 6 == 64). 56520a579deSMax Reitz * 566ecbfa281SEric Blake * In order for hbitmap_serialization_align() to always return a 56720a579deSMax Reitz * meaningful value, bitmaps that are to be serialized must have a 56820a579deSMax Reitz * granularity of less than 58. */ 56920a579deSMax Reitz 57020a579deSMax Reitz return hb->granularity < 58; 57120a579deSMax Reitz } 57220a579deSMax Reitz 573e7c033c3SPaolo Bonzini bool hbitmap_get(const HBitmap *hb, uint64_t item) 574e7c033c3SPaolo Bonzini { 575e7c033c3SPaolo Bonzini /* Compute position and bit in the last layer. */ 576e7c033c3SPaolo Bonzini uint64_t pos = item >> hb->granularity; 577e7c033c3SPaolo Bonzini unsigned long bit = 1UL << (pos & (BITS_PER_LONG - 1)); 5780e321191SVladimir Sementsov-Ogievskiy assert(pos < hb->size); 579e7c033c3SPaolo Bonzini 580e7c033c3SPaolo Bonzini return (hb->levels[HBITMAP_LEVELS - 1][pos >> BITS_PER_LEVEL] & bit) != 0; 581e7c033c3SPaolo Bonzini } 582e7c033c3SPaolo Bonzini 583ecbfa281SEric Blake uint64_t hbitmap_serialization_align(const HBitmap *hb) 5848258888eSVladimir Sementsov-Ogievskiy { 58520a579deSMax Reitz assert(hbitmap_is_serializable(hb)); 5866725f887SMax Reitz 5878258888eSVladimir Sementsov-Ogievskiy /* Require at least 64 bit granularity to be safe on both 64 bit and 32 bit 5888258888eSVladimir Sementsov-Ogievskiy * hosts. */ 5896725f887SMax Reitz return UINT64_C(64) << hb->granularity; 5908258888eSVladimir Sementsov-Ogievskiy } 5918258888eSVladimir Sementsov-Ogievskiy 5928258888eSVladimir Sementsov-Ogievskiy /* Start should be aligned to serialization granularity, chunk size should be 5938258888eSVladimir Sementsov-Ogievskiy * aligned to serialization granularity too, except for last chunk. 5948258888eSVladimir Sementsov-Ogievskiy */ 5958258888eSVladimir Sementsov-Ogievskiy static void serialization_chunk(const HBitmap *hb, 5968258888eSVladimir Sementsov-Ogievskiy uint64_t start, uint64_t count, 5978258888eSVladimir Sementsov-Ogievskiy unsigned long **first_el, uint64_t *el_count) 5988258888eSVladimir Sementsov-Ogievskiy { 5998258888eSVladimir Sementsov-Ogievskiy uint64_t last = start + count - 1; 600ecbfa281SEric Blake uint64_t gran = hbitmap_serialization_align(hb); 6018258888eSVladimir Sementsov-Ogievskiy 6028258888eSVladimir Sementsov-Ogievskiy assert((start & (gran - 1)) == 0); 6038258888eSVladimir Sementsov-Ogievskiy assert((last >> hb->granularity) < hb->size); 6048258888eSVladimir Sementsov-Ogievskiy if ((last >> hb->granularity) != hb->size - 1) { 6058258888eSVladimir Sementsov-Ogievskiy assert((count & (gran - 1)) == 0); 6068258888eSVladimir Sementsov-Ogievskiy } 6078258888eSVladimir Sementsov-Ogievskiy 6088258888eSVladimir Sementsov-Ogievskiy start = (start >> hb->granularity) >> BITS_PER_LEVEL; 6098258888eSVladimir Sementsov-Ogievskiy last = (last >> hb->granularity) >> BITS_PER_LEVEL; 6108258888eSVladimir Sementsov-Ogievskiy 6118258888eSVladimir Sementsov-Ogievskiy *first_el = &hb->levels[HBITMAP_LEVELS - 1][start]; 6128258888eSVladimir Sementsov-Ogievskiy *el_count = last - start + 1; 6138258888eSVladimir Sementsov-Ogievskiy } 6148258888eSVladimir Sementsov-Ogievskiy 6158258888eSVladimir Sementsov-Ogievskiy uint64_t hbitmap_serialization_size(const HBitmap *hb, 6168258888eSVladimir Sementsov-Ogievskiy uint64_t start, uint64_t count) 6178258888eSVladimir Sementsov-Ogievskiy { 6188258888eSVladimir Sementsov-Ogievskiy uint64_t el_count; 6198258888eSVladimir Sementsov-Ogievskiy unsigned long *cur; 6208258888eSVladimir Sementsov-Ogievskiy 6218258888eSVladimir Sementsov-Ogievskiy if (!count) { 6228258888eSVladimir Sementsov-Ogievskiy return 0; 6238258888eSVladimir Sementsov-Ogievskiy } 6248258888eSVladimir Sementsov-Ogievskiy serialization_chunk(hb, start, count, &cur, &el_count); 6258258888eSVladimir Sementsov-Ogievskiy 6268258888eSVladimir Sementsov-Ogievskiy return el_count * sizeof(unsigned long); 6278258888eSVladimir Sementsov-Ogievskiy } 6288258888eSVladimir Sementsov-Ogievskiy 6298258888eSVladimir Sementsov-Ogievskiy void hbitmap_serialize_part(const HBitmap *hb, uint8_t *buf, 6308258888eSVladimir Sementsov-Ogievskiy uint64_t start, uint64_t count) 6318258888eSVladimir Sementsov-Ogievskiy { 6328258888eSVladimir Sementsov-Ogievskiy uint64_t el_count; 6338258888eSVladimir Sementsov-Ogievskiy unsigned long *cur, *end; 6348258888eSVladimir Sementsov-Ogievskiy 6358258888eSVladimir Sementsov-Ogievskiy if (!count) { 6368258888eSVladimir Sementsov-Ogievskiy return; 6378258888eSVladimir Sementsov-Ogievskiy } 6388258888eSVladimir Sementsov-Ogievskiy serialization_chunk(hb, start, count, &cur, &el_count); 6398258888eSVladimir Sementsov-Ogievskiy end = cur + el_count; 6408258888eSVladimir Sementsov-Ogievskiy 6418258888eSVladimir Sementsov-Ogievskiy while (cur != end) { 6428258888eSVladimir Sementsov-Ogievskiy unsigned long el = 6438258888eSVladimir Sementsov-Ogievskiy (BITS_PER_LONG == 32 ? cpu_to_le32(*cur) : cpu_to_le64(*cur)); 6448258888eSVladimir Sementsov-Ogievskiy 6458258888eSVladimir Sementsov-Ogievskiy memcpy(buf, &el, sizeof(el)); 6468258888eSVladimir Sementsov-Ogievskiy buf += sizeof(el); 6478258888eSVladimir Sementsov-Ogievskiy cur++; 6488258888eSVladimir Sementsov-Ogievskiy } 6498258888eSVladimir Sementsov-Ogievskiy } 6508258888eSVladimir Sementsov-Ogievskiy 6518258888eSVladimir Sementsov-Ogievskiy void hbitmap_deserialize_part(HBitmap *hb, uint8_t *buf, 6528258888eSVladimir Sementsov-Ogievskiy uint64_t start, uint64_t count, 6538258888eSVladimir Sementsov-Ogievskiy bool finish) 6548258888eSVladimir Sementsov-Ogievskiy { 6558258888eSVladimir Sementsov-Ogievskiy uint64_t el_count; 6568258888eSVladimir Sementsov-Ogievskiy unsigned long *cur, *end; 6578258888eSVladimir Sementsov-Ogievskiy 6588258888eSVladimir Sementsov-Ogievskiy if (!count) { 6598258888eSVladimir Sementsov-Ogievskiy return; 6608258888eSVladimir Sementsov-Ogievskiy } 6618258888eSVladimir Sementsov-Ogievskiy serialization_chunk(hb, start, count, &cur, &el_count); 6628258888eSVladimir Sementsov-Ogievskiy end = cur + el_count; 6638258888eSVladimir Sementsov-Ogievskiy 6648258888eSVladimir Sementsov-Ogievskiy while (cur != end) { 6658258888eSVladimir Sementsov-Ogievskiy memcpy(cur, buf, sizeof(*cur)); 6668258888eSVladimir Sementsov-Ogievskiy 6678258888eSVladimir Sementsov-Ogievskiy if (BITS_PER_LONG == 32) { 6688258888eSVladimir Sementsov-Ogievskiy le32_to_cpus((uint32_t *)cur); 6698258888eSVladimir Sementsov-Ogievskiy } else { 6708258888eSVladimir Sementsov-Ogievskiy le64_to_cpus((uint64_t *)cur); 6718258888eSVladimir Sementsov-Ogievskiy } 6728258888eSVladimir Sementsov-Ogievskiy 6738258888eSVladimir Sementsov-Ogievskiy buf += sizeof(unsigned long); 6748258888eSVladimir Sementsov-Ogievskiy cur++; 6758258888eSVladimir Sementsov-Ogievskiy } 6768258888eSVladimir Sementsov-Ogievskiy if (finish) { 6778258888eSVladimir Sementsov-Ogievskiy hbitmap_deserialize_finish(hb); 6788258888eSVladimir Sementsov-Ogievskiy } 6798258888eSVladimir Sementsov-Ogievskiy } 6808258888eSVladimir Sementsov-Ogievskiy 6818258888eSVladimir Sementsov-Ogievskiy void hbitmap_deserialize_zeroes(HBitmap *hb, uint64_t start, uint64_t count, 6828258888eSVladimir Sementsov-Ogievskiy bool finish) 6838258888eSVladimir Sementsov-Ogievskiy { 6848258888eSVladimir Sementsov-Ogievskiy uint64_t el_count; 6858258888eSVladimir Sementsov-Ogievskiy unsigned long *first; 6868258888eSVladimir Sementsov-Ogievskiy 6878258888eSVladimir Sementsov-Ogievskiy if (!count) { 6888258888eSVladimir Sementsov-Ogievskiy return; 6898258888eSVladimir Sementsov-Ogievskiy } 6908258888eSVladimir Sementsov-Ogievskiy serialization_chunk(hb, start, count, &first, &el_count); 6918258888eSVladimir Sementsov-Ogievskiy 6928258888eSVladimir Sementsov-Ogievskiy memset(first, 0, el_count * sizeof(unsigned long)); 6938258888eSVladimir Sementsov-Ogievskiy if (finish) { 6948258888eSVladimir Sementsov-Ogievskiy hbitmap_deserialize_finish(hb); 6958258888eSVladimir Sementsov-Ogievskiy } 6968258888eSVladimir Sementsov-Ogievskiy } 6978258888eSVladimir Sementsov-Ogievskiy 6986bdc8b71SVladimir Sementsov-Ogievskiy void hbitmap_deserialize_ones(HBitmap *hb, uint64_t start, uint64_t count, 6996bdc8b71SVladimir Sementsov-Ogievskiy bool finish) 7006bdc8b71SVladimir Sementsov-Ogievskiy { 7016bdc8b71SVladimir Sementsov-Ogievskiy uint64_t el_count; 7026bdc8b71SVladimir Sementsov-Ogievskiy unsigned long *first; 7036bdc8b71SVladimir Sementsov-Ogievskiy 7046bdc8b71SVladimir Sementsov-Ogievskiy if (!count) { 7056bdc8b71SVladimir Sementsov-Ogievskiy return; 7066bdc8b71SVladimir Sementsov-Ogievskiy } 7076bdc8b71SVladimir Sementsov-Ogievskiy serialization_chunk(hb, start, count, &first, &el_count); 7086bdc8b71SVladimir Sementsov-Ogievskiy 7096bdc8b71SVladimir Sementsov-Ogievskiy memset(first, 0xff, el_count * sizeof(unsigned long)); 7106bdc8b71SVladimir Sementsov-Ogievskiy if (finish) { 7116bdc8b71SVladimir Sementsov-Ogievskiy hbitmap_deserialize_finish(hb); 7126bdc8b71SVladimir Sementsov-Ogievskiy } 7136bdc8b71SVladimir Sementsov-Ogievskiy } 7146bdc8b71SVladimir Sementsov-Ogievskiy 7158258888eSVladimir Sementsov-Ogievskiy void hbitmap_deserialize_finish(HBitmap *bitmap) 7168258888eSVladimir Sementsov-Ogievskiy { 7178258888eSVladimir Sementsov-Ogievskiy int64_t i, size, prev_size; 7188258888eSVladimir Sementsov-Ogievskiy int lev; 7198258888eSVladimir Sementsov-Ogievskiy 7208258888eSVladimir Sementsov-Ogievskiy /* restore levels starting from penultimate to zero level, assuming 7218258888eSVladimir Sementsov-Ogievskiy * that the last level is ok */ 7228258888eSVladimir Sementsov-Ogievskiy size = MAX((bitmap->size + BITS_PER_LONG - 1) >> BITS_PER_LEVEL, 1); 7238258888eSVladimir Sementsov-Ogievskiy for (lev = HBITMAP_LEVELS - 1; lev-- > 0; ) { 7248258888eSVladimir Sementsov-Ogievskiy prev_size = size; 7258258888eSVladimir Sementsov-Ogievskiy size = MAX((size + BITS_PER_LONG - 1) >> BITS_PER_LEVEL, 1); 7268258888eSVladimir Sementsov-Ogievskiy memset(bitmap->levels[lev], 0, size * sizeof(unsigned long)); 7278258888eSVladimir Sementsov-Ogievskiy 7288258888eSVladimir Sementsov-Ogievskiy for (i = 0; i < prev_size; ++i) { 7298258888eSVladimir Sementsov-Ogievskiy if (bitmap->levels[lev + 1][i]) { 7308258888eSVladimir Sementsov-Ogievskiy bitmap->levels[lev][i >> BITS_PER_LEVEL] |= 7318258888eSVladimir Sementsov-Ogievskiy 1UL << (i & (BITS_PER_LONG - 1)); 7328258888eSVladimir Sementsov-Ogievskiy } 7338258888eSVladimir Sementsov-Ogievskiy } 7348258888eSVladimir Sementsov-Ogievskiy } 7358258888eSVladimir Sementsov-Ogievskiy 7368258888eSVladimir Sementsov-Ogievskiy bitmap->levels[0][0] |= 1UL << (BITS_PER_LONG - 1); 7373260cdffSLiang Li bitmap->count = hb_count_between(bitmap, 0, bitmap->size - 1); 7388258888eSVladimir Sementsov-Ogievskiy } 7398258888eSVladimir Sementsov-Ogievskiy 740e7c033c3SPaolo Bonzini void hbitmap_free(HBitmap *hb) 741e7c033c3SPaolo Bonzini { 742e7c033c3SPaolo Bonzini unsigned i; 74307ac4cdbSFam Zheng assert(!hb->meta); 744e7c033c3SPaolo Bonzini for (i = HBITMAP_LEVELS; i-- > 0; ) { 745e7c033c3SPaolo Bonzini g_free(hb->levels[i]); 746e7c033c3SPaolo Bonzini } 747e7c033c3SPaolo Bonzini g_free(hb); 748e7c033c3SPaolo Bonzini } 749e7c033c3SPaolo Bonzini 750e7c033c3SPaolo Bonzini HBitmap *hbitmap_alloc(uint64_t size, int granularity) 751e7c033c3SPaolo Bonzini { 752e1cf5582SMarkus Armbruster HBitmap *hb = g_new0(struct HBitmap, 1); 753e7c033c3SPaolo Bonzini unsigned i; 754e7c033c3SPaolo Bonzini 7556a150995SVladimir Sementsov-Ogievskiy assert(size <= INT64_MAX); 75676d570dcSVladimir Sementsov-Ogievskiy hb->orig_size = size; 75776d570dcSVladimir Sementsov-Ogievskiy 758e7c033c3SPaolo Bonzini assert(granularity >= 0 && granularity < 64); 759e7c033c3SPaolo Bonzini size = (size + (1ULL << granularity) - 1) >> granularity; 760e7c033c3SPaolo Bonzini assert(size <= ((uint64_t)1 << HBITMAP_LOG_MAX_SIZE)); 761e7c033c3SPaolo Bonzini 762e7c033c3SPaolo Bonzini hb->size = size; 763e7c033c3SPaolo Bonzini hb->granularity = granularity; 764e7c033c3SPaolo Bonzini for (i = HBITMAP_LEVELS; i-- > 0; ) { 765e7c033c3SPaolo Bonzini size = MAX((size + BITS_PER_LONG - 1) >> BITS_PER_LEVEL, 1); 7668515efbeSJohn Snow hb->sizes[i] = size; 767e1cf5582SMarkus Armbruster hb->levels[i] = g_new0(unsigned long, size); 768e7c033c3SPaolo Bonzini } 769e7c033c3SPaolo Bonzini 770e7c033c3SPaolo Bonzini /* We necessarily have free bits in level 0 due to the definition 771e7c033c3SPaolo Bonzini * of HBITMAP_LEVELS, so use one for a sentinel. This speeds up 772e7c033c3SPaolo Bonzini * hbitmap_iter_skip_words. 773e7c033c3SPaolo Bonzini */ 774e7c033c3SPaolo Bonzini assert(size == 1); 775e7c033c3SPaolo Bonzini hb->levels[0][0] |= 1UL << (BITS_PER_LONG - 1); 776e7c033c3SPaolo Bonzini return hb; 777e7c033c3SPaolo Bonzini } 778be58721dSJohn Snow 779ce1ffea8SJohn Snow void hbitmap_truncate(HBitmap *hb, uint64_t size) 780ce1ffea8SJohn Snow { 781ce1ffea8SJohn Snow bool shrink; 782ce1ffea8SJohn Snow unsigned i; 783ce1ffea8SJohn Snow uint64_t num_elements = size; 784ce1ffea8SJohn Snow uint64_t old; 785ce1ffea8SJohn Snow 7866a150995SVladimir Sementsov-Ogievskiy assert(size <= INT64_MAX); 7874e4de222SVladimir Sementsov-Ogievskiy hb->orig_size = size; 7884e4de222SVladimir Sementsov-Ogievskiy 789ce1ffea8SJohn Snow /* Size comes in as logical elements, adjust for granularity. */ 790ce1ffea8SJohn Snow size = (size + (1ULL << hb->granularity) - 1) >> hb->granularity; 791ce1ffea8SJohn Snow assert(size <= ((uint64_t)1 << HBITMAP_LOG_MAX_SIZE)); 792ce1ffea8SJohn Snow shrink = size < hb->size; 793ce1ffea8SJohn Snow 794ce1ffea8SJohn Snow /* bit sizes are identical; nothing to do. */ 795ce1ffea8SJohn Snow if (size == hb->size) { 796ce1ffea8SJohn Snow return; 797ce1ffea8SJohn Snow } 798ce1ffea8SJohn Snow 799ce1ffea8SJohn Snow /* If we're losing bits, let's clear those bits before we invalidate all of 800ce1ffea8SJohn Snow * our invariants. This helps keep the bitcount consistent, and will prevent 801ce1ffea8SJohn Snow * us from carrying around garbage bits beyond the end of the map. 802ce1ffea8SJohn Snow */ 803ce1ffea8SJohn Snow if (shrink) { 804ce1ffea8SJohn Snow /* Don't clear partial granularity groups; 805ce1ffea8SJohn Snow * start at the first full one. */ 8066725f887SMax Reitz uint64_t start = ROUND_UP(num_elements, UINT64_C(1) << hb->granularity); 807ce1ffea8SJohn Snow uint64_t fix_count = (hb->size << hb->granularity) - start; 808ce1ffea8SJohn Snow 809ce1ffea8SJohn Snow assert(fix_count); 810ce1ffea8SJohn Snow hbitmap_reset(hb, start, fix_count); 811ce1ffea8SJohn Snow } 812ce1ffea8SJohn Snow 813ce1ffea8SJohn Snow hb->size = size; 814ce1ffea8SJohn Snow for (i = HBITMAP_LEVELS; i-- > 0; ) { 815ce1ffea8SJohn Snow size = MAX(BITS_TO_LONGS(size), 1); 816ce1ffea8SJohn Snow if (hb->sizes[i] == size) { 817ce1ffea8SJohn Snow break; 818ce1ffea8SJohn Snow } 819ce1ffea8SJohn Snow old = hb->sizes[i]; 820ce1ffea8SJohn Snow hb->sizes[i] = size; 821ce1ffea8SJohn Snow hb->levels[i] = g_realloc(hb->levels[i], size * sizeof(unsigned long)); 822ce1ffea8SJohn Snow if (!shrink) { 823ce1ffea8SJohn Snow memset(&hb->levels[i][old], 0x00, 824ce1ffea8SJohn Snow (size - old) * sizeof(*hb->levels[i])); 825ce1ffea8SJohn Snow } 826ce1ffea8SJohn Snow } 82707ac4cdbSFam Zheng if (hb->meta) { 82807ac4cdbSFam Zheng hbitmap_truncate(hb->meta, hb->size << hb->granularity); 82907ac4cdbSFam Zheng } 830ce1ffea8SJohn Snow } 831ce1ffea8SJohn Snow 832fa000f2fSVladimir Sementsov-Ogievskiy bool hbitmap_can_merge(const HBitmap *a, const HBitmap *b) 833fa000f2fSVladimir Sementsov-Ogievskiy { 834c5b40c1fSJohn Snow return (a->orig_size == b->orig_size); 835c5b40c1fSJohn Snow } 836c5b40c1fSJohn Snow 837c5b40c1fSJohn Snow /** 838c5b40c1fSJohn Snow * hbitmap_sparse_merge: performs dst = dst | src 839c5b40c1fSJohn Snow * works with differing granularities. 840c5b40c1fSJohn Snow * best used when src is sparsely populated. 841c5b40c1fSJohn Snow */ 842c5b40c1fSJohn Snow static void hbitmap_sparse_merge(HBitmap *dst, const HBitmap *src) 843c5b40c1fSJohn Snow { 844642700fdSVladimir Sementsov-Ogievskiy int64_t offset = 0; 845642700fdSVladimir Sementsov-Ogievskiy int64_t count = src->orig_size; 846c5b40c1fSJohn Snow 847c5b40c1fSJohn Snow while (hbitmap_next_dirty_area(src, &offset, &count)) { 848c5b40c1fSJohn Snow hbitmap_set(dst, offset, count); 849c5b40c1fSJohn Snow offset += count; 850c5b40c1fSJohn Snow if (offset >= src->orig_size) { 851c5b40c1fSJohn Snow break; 852c5b40c1fSJohn Snow } 853c5b40c1fSJohn Snow count = src->orig_size - offset; 854c5b40c1fSJohn Snow } 855fa000f2fSVladimir Sementsov-Ogievskiy } 856ce1ffea8SJohn Snow 857be58721dSJohn Snow /** 8583bde4b01SJohn Snow * Given HBitmaps A and B, let R := A (BITOR) B. 8593bde4b01SJohn Snow * Bitmaps A and B will not be modified, 8603bde4b01SJohn Snow * except when bitmap R is an alias of A or B. 861be58721dSJohn Snow * 862be58721dSJohn Snow * @return true if the merge was successful, 863be58721dSJohn Snow * false if it was not attempted. 864be58721dSJohn Snow */ 865fa000f2fSVladimir Sementsov-Ogievskiy bool hbitmap_merge(const HBitmap *a, const HBitmap *b, HBitmap *result) 866be58721dSJohn Snow { 867be58721dSJohn Snow int i; 868be58721dSJohn Snow uint64_t j; 869be58721dSJohn Snow 870fa000f2fSVladimir Sementsov-Ogievskiy if (!hbitmap_can_merge(a, b) || !hbitmap_can_merge(a, result)) { 871be58721dSJohn Snow return false; 872be58721dSJohn Snow } 873fa000f2fSVladimir Sementsov-Ogievskiy assert(hbitmap_can_merge(b, result)); 874be58721dSJohn Snow 8753bde4b01SJohn Snow if ((!hbitmap_count(a) && result == b) || 8763bde4b01SJohn Snow (!hbitmap_count(b) && result == a)) { 8773bde4b01SJohn Snow return true; 8783bde4b01SJohn Snow } 8793bde4b01SJohn Snow 8803bde4b01SJohn Snow if (!hbitmap_count(a) && !hbitmap_count(b)) { 8813bde4b01SJohn Snow hbitmap_reset_all(result); 882be58721dSJohn Snow return true; 883be58721dSJohn Snow } 884be58721dSJohn Snow 885c5b40c1fSJohn Snow if (a->granularity != b->granularity) { 886c5b40c1fSJohn Snow if ((a != result) && (b != result)) { 887c5b40c1fSJohn Snow hbitmap_reset_all(result); 888c5b40c1fSJohn Snow } 889c5b40c1fSJohn Snow if (a != result) { 890c5b40c1fSJohn Snow hbitmap_sparse_merge(result, a); 891c5b40c1fSJohn Snow } 892c5b40c1fSJohn Snow if (b != result) { 893c5b40c1fSJohn Snow hbitmap_sparse_merge(result, b); 894c5b40c1fSJohn Snow } 895c5b40c1fSJohn Snow return true; 896c5b40c1fSJohn Snow } 897c5b40c1fSJohn Snow 898be58721dSJohn Snow /* This merge is O(size), as BITS_PER_LONG and HBITMAP_LEVELS are constant. 899be58721dSJohn Snow * It may be possible to improve running times for sparsely populated maps 900be58721dSJohn Snow * by using hbitmap_iter_next, but this is suboptimal for dense maps. 901be58721dSJohn Snow */ 902c5b40c1fSJohn Snow assert(a->size == b->size); 903be58721dSJohn Snow for (i = HBITMAP_LEVELS - 1; i >= 0; i--) { 904be58721dSJohn Snow for (j = 0; j < a->sizes[i]; j++) { 905fa000f2fSVladimir Sementsov-Ogievskiy result->levels[i][j] = a->levels[i][j] | b->levels[i][j]; 906be58721dSJohn Snow } 907be58721dSJohn Snow } 908be58721dSJohn Snow 909d1dde714SEric Blake /* Recompute the dirty count */ 910d1dde714SEric Blake result->count = hb_count_between(result, 0, result->size - 1); 911d1dde714SEric Blake 912be58721dSJohn Snow return true; 913be58721dSJohn Snow } 91407ac4cdbSFam Zheng 915a3b52535SVladimir Sementsov-Ogievskiy char *hbitmap_sha256(const HBitmap *bitmap, Error **errp) 916a3b52535SVladimir Sementsov-Ogievskiy { 917a3b52535SVladimir Sementsov-Ogievskiy size_t size = bitmap->sizes[HBITMAP_LEVELS - 1] * sizeof(unsigned long); 918a3b52535SVladimir Sementsov-Ogievskiy char *data = (char *)bitmap->levels[HBITMAP_LEVELS - 1]; 919a3b52535SVladimir Sementsov-Ogievskiy char *hash = NULL; 920a3b52535SVladimir Sementsov-Ogievskiy qcrypto_hash_digest(QCRYPTO_HASH_ALG_SHA256, data, size, &hash, errp); 921a3b52535SVladimir Sementsov-Ogievskiy 922a3b52535SVladimir Sementsov-Ogievskiy return hash; 923a3b52535SVladimir Sementsov-Ogievskiy } 924