xref: /openbmc/qemu/util/hbitmap.c (revision 48557b138383aaf69c2617ca9a88bfb394fc50ec)
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  */
107e7c033c3SPaolo Bonzini 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 
19676d570dcSVladimir Sementsov-Ogievskiy int64_t hbitmap_next_zero(const HBitmap *hb, uint64_t start, uint64_t count)
19756207df5SVladimir Sementsov-Ogievskiy {
19856207df5SVladimir Sementsov-Ogievskiy     size_t pos = (start >> hb->granularity) >> BITS_PER_LEVEL;
19956207df5SVladimir Sementsov-Ogievskiy     unsigned long *last_lev = hb->levels[HBITMAP_LEVELS - 1];
20056207df5SVladimir Sementsov-Ogievskiy     unsigned long cur = last_lev[pos];
20176d570dcSVladimir Sementsov-Ogievskiy     unsigned start_bit_offset;
20276d570dcSVladimir Sementsov-Ogievskiy     uint64_t end_bit, sz;
20356207df5SVladimir Sementsov-Ogievskiy     int64_t res;
20456207df5SVladimir Sementsov-Ogievskiy 
20576d570dcSVladimir Sementsov-Ogievskiy     if (start >= hb->orig_size || count == 0) {
20676d570dcSVladimir Sementsov-Ogievskiy         return -1;
20776d570dcSVladimir Sementsov-Ogievskiy     }
20876d570dcSVladimir Sementsov-Ogievskiy 
20976d570dcSVladimir Sementsov-Ogievskiy     end_bit = count > hb->orig_size - start ?
21076d570dcSVladimir Sementsov-Ogievskiy                 hb->size :
21176d570dcSVladimir Sementsov-Ogievskiy                 ((start + count - 1) >> hb->granularity) + 1;
21276d570dcSVladimir Sementsov-Ogievskiy     sz = (end_bit + BITS_PER_LONG - 1) >> BITS_PER_LEVEL;
21376d570dcSVladimir Sementsov-Ogievskiy 
21476d570dcSVladimir Sementsov-Ogievskiy     /* There may be some zero bits in @cur before @start. We are not interested
21576d570dcSVladimir Sementsov-Ogievskiy      * in them, let's set them.
21676d570dcSVladimir Sementsov-Ogievskiy      */
21776d570dcSVladimir Sementsov-Ogievskiy     start_bit_offset = (start >> hb->granularity) & (BITS_PER_LONG - 1);
21856207df5SVladimir Sementsov-Ogievskiy     cur |= (1UL << start_bit_offset) - 1;
21956207df5SVladimir Sementsov-Ogievskiy     assert((start >> hb->granularity) < hb->size);
22056207df5SVladimir Sementsov-Ogievskiy 
22156207df5SVladimir Sementsov-Ogievskiy     if (cur == (unsigned long)-1) {
22256207df5SVladimir Sementsov-Ogievskiy         do {
22356207df5SVladimir Sementsov-Ogievskiy             pos++;
22456207df5SVladimir Sementsov-Ogievskiy         } while (pos < sz && last_lev[pos] == (unsigned long)-1);
22556207df5SVladimir Sementsov-Ogievskiy 
22656207df5SVladimir Sementsov-Ogievskiy         if (pos >= sz) {
22756207df5SVladimir Sementsov-Ogievskiy             return -1;
22856207df5SVladimir Sementsov-Ogievskiy         }
22956207df5SVladimir Sementsov-Ogievskiy 
23056207df5SVladimir Sementsov-Ogievskiy         cur = last_lev[pos];
23156207df5SVladimir Sementsov-Ogievskiy     }
23256207df5SVladimir Sementsov-Ogievskiy 
23356207df5SVladimir Sementsov-Ogievskiy     res = (pos << BITS_PER_LEVEL) + ctol(cur);
23476d570dcSVladimir Sementsov-Ogievskiy     if (res >= end_bit) {
23556207df5SVladimir Sementsov-Ogievskiy         return -1;
23656207df5SVladimir Sementsov-Ogievskiy     }
23756207df5SVladimir Sementsov-Ogievskiy 
23856207df5SVladimir Sementsov-Ogievskiy     res = res << hb->granularity;
23956207df5SVladimir Sementsov-Ogievskiy     if (res < start) {
24056207df5SVladimir Sementsov-Ogievskiy         assert(((start - res) >> hb->granularity) == 0);
24156207df5SVladimir Sementsov-Ogievskiy         return start;
24256207df5SVladimir Sementsov-Ogievskiy     }
24356207df5SVladimir Sementsov-Ogievskiy 
24456207df5SVladimir Sementsov-Ogievskiy     return res;
24556207df5SVladimir Sementsov-Ogievskiy }
24656207df5SVladimir Sementsov-Ogievskiy 
247a78a1a48SVladimir Sementsov-Ogievskiy bool hbitmap_next_dirty_area(const HBitmap *hb, uint64_t *start,
248a78a1a48SVladimir Sementsov-Ogievskiy                              uint64_t *count)
249a78a1a48SVladimir Sementsov-Ogievskiy {
250a78a1a48SVladimir Sementsov-Ogievskiy     HBitmapIter hbi;
251a78a1a48SVladimir Sementsov-Ogievskiy     int64_t firt_dirty_off, area_end;
252a78a1a48SVladimir Sementsov-Ogievskiy     uint32_t granularity = 1UL << hb->granularity;
253a78a1a48SVladimir Sementsov-Ogievskiy     uint64_t end;
254a78a1a48SVladimir Sementsov-Ogievskiy 
255a78a1a48SVladimir Sementsov-Ogievskiy     if (*start >= hb->orig_size || *count == 0) {
256a78a1a48SVladimir Sementsov-Ogievskiy         return false;
257a78a1a48SVladimir Sementsov-Ogievskiy     }
258a78a1a48SVladimir Sementsov-Ogievskiy 
259a78a1a48SVladimir Sementsov-Ogievskiy     end = *count > hb->orig_size - *start ? hb->orig_size : *start + *count;
260a78a1a48SVladimir Sementsov-Ogievskiy 
261a78a1a48SVladimir Sementsov-Ogievskiy     hbitmap_iter_init(&hbi, hb, *start);
26219c021e1SVladimir Sementsov-Ogievskiy     firt_dirty_off = hbitmap_iter_next(&hbi);
263a78a1a48SVladimir Sementsov-Ogievskiy 
264a78a1a48SVladimir Sementsov-Ogievskiy     if (firt_dirty_off < 0 || firt_dirty_off >= end) {
265a78a1a48SVladimir Sementsov-Ogievskiy         return false;
266a78a1a48SVladimir Sementsov-Ogievskiy     }
267a78a1a48SVladimir Sementsov-Ogievskiy 
268a78a1a48SVladimir Sementsov-Ogievskiy     if (firt_dirty_off + granularity >= end) {
269a78a1a48SVladimir Sementsov-Ogievskiy         area_end = end;
270a78a1a48SVladimir Sementsov-Ogievskiy     } else {
271a78a1a48SVladimir Sementsov-Ogievskiy         area_end = hbitmap_next_zero(hb, firt_dirty_off + granularity,
272a78a1a48SVladimir Sementsov-Ogievskiy                                      end - firt_dirty_off - granularity);
273a78a1a48SVladimir Sementsov-Ogievskiy         if (area_end < 0) {
274a78a1a48SVladimir Sementsov-Ogievskiy             area_end = end;
275a78a1a48SVladimir Sementsov-Ogievskiy         }
276a78a1a48SVladimir Sementsov-Ogievskiy     }
277a78a1a48SVladimir Sementsov-Ogievskiy 
278a78a1a48SVladimir Sementsov-Ogievskiy     if (firt_dirty_off > *start) {
279a78a1a48SVladimir Sementsov-Ogievskiy         *start = firt_dirty_off;
280a78a1a48SVladimir Sementsov-Ogievskiy     }
281a78a1a48SVladimir Sementsov-Ogievskiy     *count = area_end - *start;
282a78a1a48SVladimir Sementsov-Ogievskiy 
283a78a1a48SVladimir Sementsov-Ogievskiy     return true;
284a78a1a48SVladimir Sementsov-Ogievskiy }
285a78a1a48SVladimir Sementsov-Ogievskiy 
286e7c033c3SPaolo Bonzini bool hbitmap_empty(const HBitmap *hb)
287e7c033c3SPaolo Bonzini {
288e7c033c3SPaolo Bonzini     return hb->count == 0;
289e7c033c3SPaolo Bonzini }
290e7c033c3SPaolo Bonzini 
291e7c033c3SPaolo Bonzini int hbitmap_granularity(const HBitmap *hb)
292e7c033c3SPaolo Bonzini {
293e7c033c3SPaolo Bonzini     return hb->granularity;
294e7c033c3SPaolo Bonzini }
295e7c033c3SPaolo Bonzini 
296e7c033c3SPaolo Bonzini uint64_t hbitmap_count(const HBitmap *hb)
297e7c033c3SPaolo Bonzini {
298e7c033c3SPaolo Bonzini     return hb->count << hb->granularity;
299e7c033c3SPaolo Bonzini }
300e7c033c3SPaolo Bonzini 
301e7c033c3SPaolo Bonzini /* Count the number of set bits between start and end, not accounting for
302e7c033c3SPaolo Bonzini  * the granularity.  Also an example of how to use hbitmap_iter_next_word.
303e7c033c3SPaolo Bonzini  */
304e7c033c3SPaolo Bonzini static uint64_t hb_count_between(HBitmap *hb, uint64_t start, uint64_t last)
305e7c033c3SPaolo Bonzini {
306e7c033c3SPaolo Bonzini     HBitmapIter hbi;
307e7c033c3SPaolo Bonzini     uint64_t count = 0;
308e7c033c3SPaolo Bonzini     uint64_t end = last + 1;
309e7c033c3SPaolo Bonzini     unsigned long cur;
310e7c033c3SPaolo Bonzini     size_t pos;
311e7c033c3SPaolo Bonzini 
312e7c033c3SPaolo Bonzini     hbitmap_iter_init(&hbi, hb, start << hb->granularity);
313e7c033c3SPaolo Bonzini     for (;;) {
314e7c033c3SPaolo Bonzini         pos = hbitmap_iter_next_word(&hbi, &cur);
315e7c033c3SPaolo Bonzini         if (pos >= (end >> BITS_PER_LEVEL)) {
316e7c033c3SPaolo Bonzini             break;
317e7c033c3SPaolo Bonzini         }
318591b320aSPeter Maydell         count += ctpopl(cur);
319e7c033c3SPaolo Bonzini     }
320e7c033c3SPaolo Bonzini 
321e7c033c3SPaolo Bonzini     if (pos == (end >> BITS_PER_LEVEL)) {
322e7c033c3SPaolo Bonzini         /* Drop bits representing the END-th and subsequent items.  */
323e7c033c3SPaolo Bonzini         int bit = end & (BITS_PER_LONG - 1);
324e7c033c3SPaolo Bonzini         cur &= (1UL << bit) - 1;
325591b320aSPeter Maydell         count += ctpopl(cur);
326e7c033c3SPaolo Bonzini     }
327e7c033c3SPaolo Bonzini 
328e7c033c3SPaolo Bonzini     return count;
329e7c033c3SPaolo Bonzini }
330e7c033c3SPaolo Bonzini 
331e7c033c3SPaolo Bonzini /* Setting starts at the last layer and propagates up if an element
33207ac4cdbSFam Zheng  * changes.
333e7c033c3SPaolo Bonzini  */
334e7c033c3SPaolo Bonzini static inline bool hb_set_elem(unsigned long *elem, uint64_t start, uint64_t last)
335e7c033c3SPaolo Bonzini {
336e7c033c3SPaolo Bonzini     unsigned long mask;
33707ac4cdbSFam Zheng     unsigned long old;
338e7c033c3SPaolo Bonzini 
339e7c033c3SPaolo Bonzini     assert((last >> BITS_PER_LEVEL) == (start >> BITS_PER_LEVEL));
340e7c033c3SPaolo Bonzini     assert(start <= last);
341e7c033c3SPaolo Bonzini 
342e7c033c3SPaolo Bonzini     mask = 2UL << (last & (BITS_PER_LONG - 1));
343e7c033c3SPaolo Bonzini     mask -= 1UL << (start & (BITS_PER_LONG - 1));
34407ac4cdbSFam Zheng     old = *elem;
345e7c033c3SPaolo Bonzini     *elem |= mask;
34607ac4cdbSFam Zheng     return old != *elem;
347e7c033c3SPaolo Bonzini }
348e7c033c3SPaolo Bonzini 
34907ac4cdbSFam Zheng /* The recursive workhorse (the depth is limited to HBITMAP_LEVELS)...
35007ac4cdbSFam Zheng  * Returns true if at least one bit is changed. */
35107ac4cdbSFam Zheng static bool hb_set_between(HBitmap *hb, int level, uint64_t start,
35207ac4cdbSFam Zheng                            uint64_t last)
353e7c033c3SPaolo Bonzini {
354e7c033c3SPaolo Bonzini     size_t pos = start >> BITS_PER_LEVEL;
355e7c033c3SPaolo Bonzini     size_t lastpos = last >> BITS_PER_LEVEL;
356e7c033c3SPaolo Bonzini     bool changed = false;
357e7c033c3SPaolo Bonzini     size_t i;
358e7c033c3SPaolo Bonzini 
359e7c033c3SPaolo Bonzini     i = pos;
360e7c033c3SPaolo Bonzini     if (i < lastpos) {
361e7c033c3SPaolo Bonzini         uint64_t next = (start | (BITS_PER_LONG - 1)) + 1;
362e7c033c3SPaolo Bonzini         changed |= hb_set_elem(&hb->levels[level][i], start, next - 1);
363e7c033c3SPaolo Bonzini         for (;;) {
364e7c033c3SPaolo Bonzini             start = next;
365e7c033c3SPaolo Bonzini             next += BITS_PER_LONG;
366e7c033c3SPaolo Bonzini             if (++i == lastpos) {
367e7c033c3SPaolo Bonzini                 break;
368e7c033c3SPaolo Bonzini             }
369e7c033c3SPaolo Bonzini             changed |= (hb->levels[level][i] == 0);
370e7c033c3SPaolo Bonzini             hb->levels[level][i] = ~0UL;
371e7c033c3SPaolo Bonzini         }
372e7c033c3SPaolo Bonzini     }
373e7c033c3SPaolo Bonzini     changed |= hb_set_elem(&hb->levels[level][i], start, last);
374e7c033c3SPaolo Bonzini 
375e7c033c3SPaolo Bonzini     /* If there was any change in this layer, we may have to update
376e7c033c3SPaolo Bonzini      * the one above.
377e7c033c3SPaolo Bonzini      */
378e7c033c3SPaolo Bonzini     if (level > 0 && changed) {
379e7c033c3SPaolo Bonzini         hb_set_between(hb, level - 1, pos, lastpos);
380e7c033c3SPaolo Bonzini     }
38107ac4cdbSFam Zheng     return changed;
382e7c033c3SPaolo Bonzini }
383e7c033c3SPaolo Bonzini 
384e7c033c3SPaolo Bonzini void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count)
385e7c033c3SPaolo Bonzini {
386e7c033c3SPaolo Bonzini     /* Compute range in the last layer.  */
38707ac4cdbSFam Zheng     uint64_t first, n;
388e7c033c3SPaolo Bonzini     uint64_t last = start + count - 1;
389e7c033c3SPaolo Bonzini 
390e7c033c3SPaolo Bonzini     trace_hbitmap_set(hb, start, count,
391e7c033c3SPaolo Bonzini                       start >> hb->granularity, last >> hb->granularity);
392e7c033c3SPaolo Bonzini 
39307ac4cdbSFam Zheng     first = start >> hb->granularity;
394e7c033c3SPaolo Bonzini     last >>= hb->granularity;
3950e321191SVladimir Sementsov-Ogievskiy     assert(last < hb->size);
39607ac4cdbSFam Zheng     n = last - first + 1;
397e7c033c3SPaolo Bonzini 
39807ac4cdbSFam Zheng     hb->count += n - hb_count_between(hb, first, last);
39907ac4cdbSFam Zheng     if (hb_set_between(hb, HBITMAP_LEVELS - 1, first, last) &&
40007ac4cdbSFam Zheng         hb->meta) {
40107ac4cdbSFam Zheng         hbitmap_set(hb->meta, start, count);
40207ac4cdbSFam Zheng     }
403e7c033c3SPaolo Bonzini }
404e7c033c3SPaolo Bonzini 
405e7c033c3SPaolo Bonzini /* Resetting works the other way round: propagate up if the new
406e7c033c3SPaolo Bonzini  * value is zero.
407e7c033c3SPaolo Bonzini  */
408e7c033c3SPaolo Bonzini static inline bool hb_reset_elem(unsigned long *elem, uint64_t start, uint64_t last)
409e7c033c3SPaolo Bonzini {
410e7c033c3SPaolo Bonzini     unsigned long mask;
411e7c033c3SPaolo Bonzini     bool blanked;
412e7c033c3SPaolo Bonzini 
413e7c033c3SPaolo Bonzini     assert((last >> BITS_PER_LEVEL) == (start >> BITS_PER_LEVEL));
414e7c033c3SPaolo Bonzini     assert(start <= last);
415e7c033c3SPaolo Bonzini 
416e7c033c3SPaolo Bonzini     mask = 2UL << (last & (BITS_PER_LONG - 1));
417e7c033c3SPaolo Bonzini     mask -= 1UL << (start & (BITS_PER_LONG - 1));
418e7c033c3SPaolo Bonzini     blanked = *elem != 0 && ((*elem & ~mask) == 0);
419e7c033c3SPaolo Bonzini     *elem &= ~mask;
420e7c033c3SPaolo Bonzini     return blanked;
421e7c033c3SPaolo Bonzini }
422e7c033c3SPaolo Bonzini 
42307ac4cdbSFam Zheng /* The recursive workhorse (the depth is limited to HBITMAP_LEVELS)...
42407ac4cdbSFam Zheng  * Returns true if at least one bit is changed. */
42507ac4cdbSFam Zheng static bool hb_reset_between(HBitmap *hb, int level, uint64_t start,
42607ac4cdbSFam Zheng                              uint64_t last)
427e7c033c3SPaolo Bonzini {
428e7c033c3SPaolo Bonzini     size_t pos = start >> BITS_PER_LEVEL;
429e7c033c3SPaolo Bonzini     size_t lastpos = last >> BITS_PER_LEVEL;
430e7c033c3SPaolo Bonzini     bool changed = false;
431e7c033c3SPaolo Bonzini     size_t i;
432e7c033c3SPaolo Bonzini 
433e7c033c3SPaolo Bonzini     i = pos;
434e7c033c3SPaolo Bonzini     if (i < lastpos) {
435e7c033c3SPaolo Bonzini         uint64_t next = (start | (BITS_PER_LONG - 1)) + 1;
436e7c033c3SPaolo Bonzini 
437e7c033c3SPaolo Bonzini         /* Here we need a more complex test than when setting bits.  Even if
438e7c033c3SPaolo Bonzini          * something was changed, we must not blank bits in the upper level
439e7c033c3SPaolo Bonzini          * unless the lower-level word became entirely zero.  So, remove pos
440e7c033c3SPaolo Bonzini          * from the upper-level range if bits remain set.
441e7c033c3SPaolo Bonzini          */
442e7c033c3SPaolo Bonzini         if (hb_reset_elem(&hb->levels[level][i], start, next - 1)) {
443e7c033c3SPaolo Bonzini             changed = true;
444e7c033c3SPaolo Bonzini         } else {
445e7c033c3SPaolo Bonzini             pos++;
446e7c033c3SPaolo Bonzini         }
447e7c033c3SPaolo Bonzini 
448e7c033c3SPaolo Bonzini         for (;;) {
449e7c033c3SPaolo Bonzini             start = next;
450e7c033c3SPaolo Bonzini             next += BITS_PER_LONG;
451e7c033c3SPaolo Bonzini             if (++i == lastpos) {
452e7c033c3SPaolo Bonzini                 break;
453e7c033c3SPaolo Bonzini             }
454e7c033c3SPaolo Bonzini             changed |= (hb->levels[level][i] != 0);
455e7c033c3SPaolo Bonzini             hb->levels[level][i] = 0UL;
456e7c033c3SPaolo Bonzini         }
457e7c033c3SPaolo Bonzini     }
458e7c033c3SPaolo Bonzini 
459e7c033c3SPaolo Bonzini     /* Same as above, this time for lastpos.  */
460e7c033c3SPaolo Bonzini     if (hb_reset_elem(&hb->levels[level][i], start, last)) {
461e7c033c3SPaolo Bonzini         changed = true;
462e7c033c3SPaolo Bonzini     } else {
463e7c033c3SPaolo Bonzini         lastpos--;
464e7c033c3SPaolo Bonzini     }
465e7c033c3SPaolo Bonzini 
466e7c033c3SPaolo Bonzini     if (level > 0 && changed) {
467e7c033c3SPaolo Bonzini         hb_reset_between(hb, level - 1, pos, lastpos);
468e7c033c3SPaolo Bonzini     }
46907ac4cdbSFam Zheng 
47007ac4cdbSFam Zheng     return changed;
47107ac4cdbSFam Zheng 
472e7c033c3SPaolo Bonzini }
473e7c033c3SPaolo Bonzini 
474e7c033c3SPaolo Bonzini void hbitmap_reset(HBitmap *hb, uint64_t start, uint64_t count)
475e7c033c3SPaolo Bonzini {
476e7c033c3SPaolo Bonzini     /* Compute range in the last layer.  */
47707ac4cdbSFam Zheng     uint64_t first;
478e7c033c3SPaolo Bonzini     uint64_t last = start + count - 1;
479*48557b13SVladimir Sementsov-Ogievskiy     uint64_t gran = 1ULL << hb->granularity;
480*48557b13SVladimir Sementsov-Ogievskiy 
481*48557b13SVladimir Sementsov-Ogievskiy     assert(QEMU_IS_ALIGNED(start, gran));
482*48557b13SVladimir Sementsov-Ogievskiy     assert(QEMU_IS_ALIGNED(count, gran) || (start + count == hb->orig_size));
483e7c033c3SPaolo Bonzini 
484e7c033c3SPaolo Bonzini     trace_hbitmap_reset(hb, start, count,
485e7c033c3SPaolo Bonzini                         start >> hb->granularity, last >> hb->granularity);
486e7c033c3SPaolo Bonzini 
48707ac4cdbSFam Zheng     first = start >> hb->granularity;
488e7c033c3SPaolo Bonzini     last >>= hb->granularity;
4890e321191SVladimir Sementsov-Ogievskiy     assert(last < hb->size);
490e7c033c3SPaolo Bonzini 
49107ac4cdbSFam Zheng     hb->count -= hb_count_between(hb, first, last);
49207ac4cdbSFam Zheng     if (hb_reset_between(hb, HBITMAP_LEVELS - 1, first, last) &&
49307ac4cdbSFam Zheng         hb->meta) {
49407ac4cdbSFam Zheng         hbitmap_set(hb->meta, start, count);
49507ac4cdbSFam Zheng     }
496e7c033c3SPaolo Bonzini }
497e7c033c3SPaolo Bonzini 
498c6a8c328SWen Congyang void hbitmap_reset_all(HBitmap *hb)
499c6a8c328SWen Congyang {
500c6a8c328SWen Congyang     unsigned int i;
501c6a8c328SWen Congyang 
502c6a8c328SWen Congyang     /* Same as hbitmap_alloc() except for memset() instead of malloc() */
503c6a8c328SWen Congyang     for (i = HBITMAP_LEVELS; --i >= 1; ) {
504c6a8c328SWen Congyang         memset(hb->levels[i], 0, hb->sizes[i] * sizeof(unsigned long));
505c6a8c328SWen Congyang     }
506c6a8c328SWen Congyang 
507c6a8c328SWen Congyang     hb->levels[0][0] = 1UL << (BITS_PER_LONG - 1);
508c6a8c328SWen Congyang     hb->count = 0;
509c6a8c328SWen Congyang }
510c6a8c328SWen Congyang 
51120a579deSMax Reitz bool hbitmap_is_serializable(const HBitmap *hb)
51220a579deSMax Reitz {
51320a579deSMax Reitz     /* Every serialized chunk must be aligned to 64 bits so that endianness
51420a579deSMax Reitz      * requirements can be fulfilled on both 64 bit and 32 bit hosts.
515ecbfa281SEric Blake      * We have hbitmap_serialization_align() which converts this
51620a579deSMax Reitz      * alignment requirement from bitmap bits to items covered (e.g. sectors).
51720a579deSMax Reitz      * That value is:
51820a579deSMax Reitz      *    64 << hb->granularity
51920a579deSMax Reitz      * Since this value must not exceed UINT64_MAX, hb->granularity must be
52020a579deSMax Reitz      * less than 58 (== 64 - 6, where 6 is ld(64), i.e. 1 << 6 == 64).
52120a579deSMax Reitz      *
522ecbfa281SEric Blake      * In order for hbitmap_serialization_align() to always return a
52320a579deSMax Reitz      * meaningful value, bitmaps that are to be serialized must have a
52420a579deSMax Reitz      * granularity of less than 58. */
52520a579deSMax Reitz 
52620a579deSMax Reitz     return hb->granularity < 58;
52720a579deSMax Reitz }
52820a579deSMax Reitz 
529e7c033c3SPaolo Bonzini bool hbitmap_get(const HBitmap *hb, uint64_t item)
530e7c033c3SPaolo Bonzini {
531e7c033c3SPaolo Bonzini     /* Compute position and bit in the last layer.  */
532e7c033c3SPaolo Bonzini     uint64_t pos = item >> hb->granularity;
533e7c033c3SPaolo Bonzini     unsigned long bit = 1UL << (pos & (BITS_PER_LONG - 1));
5340e321191SVladimir Sementsov-Ogievskiy     assert(pos < hb->size);
535e7c033c3SPaolo Bonzini 
536e7c033c3SPaolo Bonzini     return (hb->levels[HBITMAP_LEVELS - 1][pos >> BITS_PER_LEVEL] & bit) != 0;
537e7c033c3SPaolo Bonzini }
538e7c033c3SPaolo Bonzini 
539ecbfa281SEric Blake uint64_t hbitmap_serialization_align(const HBitmap *hb)
5408258888eSVladimir Sementsov-Ogievskiy {
54120a579deSMax Reitz     assert(hbitmap_is_serializable(hb));
5426725f887SMax Reitz 
5438258888eSVladimir Sementsov-Ogievskiy     /* Require at least 64 bit granularity to be safe on both 64 bit and 32 bit
5448258888eSVladimir Sementsov-Ogievskiy      * hosts. */
5456725f887SMax Reitz     return UINT64_C(64) << hb->granularity;
5468258888eSVladimir Sementsov-Ogievskiy }
5478258888eSVladimir Sementsov-Ogievskiy 
5488258888eSVladimir Sementsov-Ogievskiy /* Start should be aligned to serialization granularity, chunk size should be
5498258888eSVladimir Sementsov-Ogievskiy  * aligned to serialization granularity too, except for last chunk.
5508258888eSVladimir Sementsov-Ogievskiy  */
5518258888eSVladimir Sementsov-Ogievskiy static void serialization_chunk(const HBitmap *hb,
5528258888eSVladimir Sementsov-Ogievskiy                                 uint64_t start, uint64_t count,
5538258888eSVladimir Sementsov-Ogievskiy                                 unsigned long **first_el, uint64_t *el_count)
5548258888eSVladimir Sementsov-Ogievskiy {
5558258888eSVladimir Sementsov-Ogievskiy     uint64_t last = start + count - 1;
556ecbfa281SEric Blake     uint64_t gran = hbitmap_serialization_align(hb);
5578258888eSVladimir Sementsov-Ogievskiy 
5588258888eSVladimir Sementsov-Ogievskiy     assert((start & (gran - 1)) == 0);
5598258888eSVladimir Sementsov-Ogievskiy     assert((last >> hb->granularity) < hb->size);
5608258888eSVladimir Sementsov-Ogievskiy     if ((last >> hb->granularity) != hb->size - 1) {
5618258888eSVladimir Sementsov-Ogievskiy         assert((count & (gran - 1)) == 0);
5628258888eSVladimir Sementsov-Ogievskiy     }
5638258888eSVladimir Sementsov-Ogievskiy 
5648258888eSVladimir Sementsov-Ogievskiy     start = (start >> hb->granularity) >> BITS_PER_LEVEL;
5658258888eSVladimir Sementsov-Ogievskiy     last = (last >> hb->granularity) >> BITS_PER_LEVEL;
5668258888eSVladimir Sementsov-Ogievskiy 
5678258888eSVladimir Sementsov-Ogievskiy     *first_el = &hb->levels[HBITMAP_LEVELS - 1][start];
5688258888eSVladimir Sementsov-Ogievskiy     *el_count = last - start + 1;
5698258888eSVladimir Sementsov-Ogievskiy }
5708258888eSVladimir Sementsov-Ogievskiy 
5718258888eSVladimir Sementsov-Ogievskiy uint64_t hbitmap_serialization_size(const HBitmap *hb,
5728258888eSVladimir Sementsov-Ogievskiy                                     uint64_t start, uint64_t count)
5738258888eSVladimir Sementsov-Ogievskiy {
5748258888eSVladimir Sementsov-Ogievskiy     uint64_t el_count;
5758258888eSVladimir Sementsov-Ogievskiy     unsigned long *cur;
5768258888eSVladimir Sementsov-Ogievskiy 
5778258888eSVladimir Sementsov-Ogievskiy     if (!count) {
5788258888eSVladimir Sementsov-Ogievskiy         return 0;
5798258888eSVladimir Sementsov-Ogievskiy     }
5808258888eSVladimir Sementsov-Ogievskiy     serialization_chunk(hb, start, count, &cur, &el_count);
5818258888eSVladimir Sementsov-Ogievskiy 
5828258888eSVladimir Sementsov-Ogievskiy     return el_count * sizeof(unsigned long);
5838258888eSVladimir Sementsov-Ogievskiy }
5848258888eSVladimir Sementsov-Ogievskiy 
5858258888eSVladimir Sementsov-Ogievskiy void hbitmap_serialize_part(const HBitmap *hb, uint8_t *buf,
5868258888eSVladimir Sementsov-Ogievskiy                             uint64_t start, uint64_t count)
5878258888eSVladimir Sementsov-Ogievskiy {
5888258888eSVladimir Sementsov-Ogievskiy     uint64_t el_count;
5898258888eSVladimir Sementsov-Ogievskiy     unsigned long *cur, *end;
5908258888eSVladimir Sementsov-Ogievskiy 
5918258888eSVladimir Sementsov-Ogievskiy     if (!count) {
5928258888eSVladimir Sementsov-Ogievskiy         return;
5938258888eSVladimir Sementsov-Ogievskiy     }
5948258888eSVladimir Sementsov-Ogievskiy     serialization_chunk(hb, start, count, &cur, &el_count);
5958258888eSVladimir Sementsov-Ogievskiy     end = cur + el_count;
5968258888eSVladimir Sementsov-Ogievskiy 
5978258888eSVladimir Sementsov-Ogievskiy     while (cur != end) {
5988258888eSVladimir Sementsov-Ogievskiy         unsigned long el =
5998258888eSVladimir Sementsov-Ogievskiy             (BITS_PER_LONG == 32 ? cpu_to_le32(*cur) : cpu_to_le64(*cur));
6008258888eSVladimir Sementsov-Ogievskiy 
6018258888eSVladimir Sementsov-Ogievskiy         memcpy(buf, &el, sizeof(el));
6028258888eSVladimir Sementsov-Ogievskiy         buf += sizeof(el);
6038258888eSVladimir Sementsov-Ogievskiy         cur++;
6048258888eSVladimir Sementsov-Ogievskiy     }
6058258888eSVladimir Sementsov-Ogievskiy }
6068258888eSVladimir Sementsov-Ogievskiy 
6078258888eSVladimir Sementsov-Ogievskiy void hbitmap_deserialize_part(HBitmap *hb, uint8_t *buf,
6088258888eSVladimir Sementsov-Ogievskiy                               uint64_t start, uint64_t count,
6098258888eSVladimir Sementsov-Ogievskiy                               bool finish)
6108258888eSVladimir Sementsov-Ogievskiy {
6118258888eSVladimir Sementsov-Ogievskiy     uint64_t el_count;
6128258888eSVladimir Sementsov-Ogievskiy     unsigned long *cur, *end;
6138258888eSVladimir Sementsov-Ogievskiy 
6148258888eSVladimir Sementsov-Ogievskiy     if (!count) {
6158258888eSVladimir Sementsov-Ogievskiy         return;
6168258888eSVladimir Sementsov-Ogievskiy     }
6178258888eSVladimir Sementsov-Ogievskiy     serialization_chunk(hb, start, count, &cur, &el_count);
6188258888eSVladimir Sementsov-Ogievskiy     end = cur + el_count;
6198258888eSVladimir Sementsov-Ogievskiy 
6208258888eSVladimir Sementsov-Ogievskiy     while (cur != end) {
6218258888eSVladimir Sementsov-Ogievskiy         memcpy(cur, buf, sizeof(*cur));
6228258888eSVladimir Sementsov-Ogievskiy 
6238258888eSVladimir Sementsov-Ogievskiy         if (BITS_PER_LONG == 32) {
6248258888eSVladimir Sementsov-Ogievskiy             le32_to_cpus((uint32_t *)cur);
6258258888eSVladimir Sementsov-Ogievskiy         } else {
6268258888eSVladimir Sementsov-Ogievskiy             le64_to_cpus((uint64_t *)cur);
6278258888eSVladimir Sementsov-Ogievskiy         }
6288258888eSVladimir Sementsov-Ogievskiy 
6298258888eSVladimir Sementsov-Ogievskiy         buf += sizeof(unsigned long);
6308258888eSVladimir Sementsov-Ogievskiy         cur++;
6318258888eSVladimir Sementsov-Ogievskiy     }
6328258888eSVladimir Sementsov-Ogievskiy     if (finish) {
6338258888eSVladimir Sementsov-Ogievskiy         hbitmap_deserialize_finish(hb);
6348258888eSVladimir Sementsov-Ogievskiy     }
6358258888eSVladimir Sementsov-Ogievskiy }
6368258888eSVladimir Sementsov-Ogievskiy 
6378258888eSVladimir Sementsov-Ogievskiy void hbitmap_deserialize_zeroes(HBitmap *hb, uint64_t start, uint64_t count,
6388258888eSVladimir Sementsov-Ogievskiy                                 bool finish)
6398258888eSVladimir Sementsov-Ogievskiy {
6408258888eSVladimir Sementsov-Ogievskiy     uint64_t el_count;
6418258888eSVladimir Sementsov-Ogievskiy     unsigned long *first;
6428258888eSVladimir Sementsov-Ogievskiy 
6438258888eSVladimir Sementsov-Ogievskiy     if (!count) {
6448258888eSVladimir Sementsov-Ogievskiy         return;
6458258888eSVladimir Sementsov-Ogievskiy     }
6468258888eSVladimir Sementsov-Ogievskiy     serialization_chunk(hb, start, count, &first, &el_count);
6478258888eSVladimir Sementsov-Ogievskiy 
6488258888eSVladimir Sementsov-Ogievskiy     memset(first, 0, el_count * sizeof(unsigned long));
6498258888eSVladimir Sementsov-Ogievskiy     if (finish) {
6508258888eSVladimir Sementsov-Ogievskiy         hbitmap_deserialize_finish(hb);
6518258888eSVladimir Sementsov-Ogievskiy     }
6528258888eSVladimir Sementsov-Ogievskiy }
6538258888eSVladimir Sementsov-Ogievskiy 
6546bdc8b71SVladimir Sementsov-Ogievskiy void hbitmap_deserialize_ones(HBitmap *hb, uint64_t start, uint64_t count,
6556bdc8b71SVladimir Sementsov-Ogievskiy                               bool finish)
6566bdc8b71SVladimir Sementsov-Ogievskiy {
6576bdc8b71SVladimir Sementsov-Ogievskiy     uint64_t el_count;
6586bdc8b71SVladimir Sementsov-Ogievskiy     unsigned long *first;
6596bdc8b71SVladimir Sementsov-Ogievskiy 
6606bdc8b71SVladimir Sementsov-Ogievskiy     if (!count) {
6616bdc8b71SVladimir Sementsov-Ogievskiy         return;
6626bdc8b71SVladimir Sementsov-Ogievskiy     }
6636bdc8b71SVladimir Sementsov-Ogievskiy     serialization_chunk(hb, start, count, &first, &el_count);
6646bdc8b71SVladimir Sementsov-Ogievskiy 
6656bdc8b71SVladimir Sementsov-Ogievskiy     memset(first, 0xff, el_count * sizeof(unsigned long));
6666bdc8b71SVladimir Sementsov-Ogievskiy     if (finish) {
6676bdc8b71SVladimir Sementsov-Ogievskiy         hbitmap_deserialize_finish(hb);
6686bdc8b71SVladimir Sementsov-Ogievskiy     }
6696bdc8b71SVladimir Sementsov-Ogievskiy }
6706bdc8b71SVladimir Sementsov-Ogievskiy 
6718258888eSVladimir Sementsov-Ogievskiy void hbitmap_deserialize_finish(HBitmap *bitmap)
6728258888eSVladimir Sementsov-Ogievskiy {
6738258888eSVladimir Sementsov-Ogievskiy     int64_t i, size, prev_size;
6748258888eSVladimir Sementsov-Ogievskiy     int lev;
6758258888eSVladimir Sementsov-Ogievskiy 
6768258888eSVladimir Sementsov-Ogievskiy     /* restore levels starting from penultimate to zero level, assuming
6778258888eSVladimir Sementsov-Ogievskiy      * that the last level is ok */
6788258888eSVladimir Sementsov-Ogievskiy     size = MAX((bitmap->size + BITS_PER_LONG - 1) >> BITS_PER_LEVEL, 1);
6798258888eSVladimir Sementsov-Ogievskiy     for (lev = HBITMAP_LEVELS - 1; lev-- > 0; ) {
6808258888eSVladimir Sementsov-Ogievskiy         prev_size = size;
6818258888eSVladimir Sementsov-Ogievskiy         size = MAX((size + BITS_PER_LONG - 1) >> BITS_PER_LEVEL, 1);
6828258888eSVladimir Sementsov-Ogievskiy         memset(bitmap->levels[lev], 0, size * sizeof(unsigned long));
6838258888eSVladimir Sementsov-Ogievskiy 
6848258888eSVladimir Sementsov-Ogievskiy         for (i = 0; i < prev_size; ++i) {
6858258888eSVladimir Sementsov-Ogievskiy             if (bitmap->levels[lev + 1][i]) {
6868258888eSVladimir Sementsov-Ogievskiy                 bitmap->levels[lev][i >> BITS_PER_LEVEL] |=
6878258888eSVladimir Sementsov-Ogievskiy                     1UL << (i & (BITS_PER_LONG - 1));
6888258888eSVladimir Sementsov-Ogievskiy             }
6898258888eSVladimir Sementsov-Ogievskiy         }
6908258888eSVladimir Sementsov-Ogievskiy     }
6918258888eSVladimir Sementsov-Ogievskiy 
6928258888eSVladimir Sementsov-Ogievskiy     bitmap->levels[0][0] |= 1UL << (BITS_PER_LONG - 1);
6933260cdffSLiang Li     bitmap->count = hb_count_between(bitmap, 0, bitmap->size - 1);
6948258888eSVladimir Sementsov-Ogievskiy }
6958258888eSVladimir Sementsov-Ogievskiy 
696e7c033c3SPaolo Bonzini void hbitmap_free(HBitmap *hb)
697e7c033c3SPaolo Bonzini {
698e7c033c3SPaolo Bonzini     unsigned i;
69907ac4cdbSFam Zheng     assert(!hb->meta);
700e7c033c3SPaolo Bonzini     for (i = HBITMAP_LEVELS; i-- > 0; ) {
701e7c033c3SPaolo Bonzini         g_free(hb->levels[i]);
702e7c033c3SPaolo Bonzini     }
703e7c033c3SPaolo Bonzini     g_free(hb);
704e7c033c3SPaolo Bonzini }
705e7c033c3SPaolo Bonzini 
706e7c033c3SPaolo Bonzini HBitmap *hbitmap_alloc(uint64_t size, int granularity)
707e7c033c3SPaolo Bonzini {
708e1cf5582SMarkus Armbruster     HBitmap *hb = g_new0(struct HBitmap, 1);
709e7c033c3SPaolo Bonzini     unsigned i;
710e7c033c3SPaolo Bonzini 
71176d570dcSVladimir Sementsov-Ogievskiy     hb->orig_size = size;
71276d570dcSVladimir Sementsov-Ogievskiy 
713e7c033c3SPaolo Bonzini     assert(granularity >= 0 && granularity < 64);
714e7c033c3SPaolo Bonzini     size = (size + (1ULL << granularity) - 1) >> granularity;
715e7c033c3SPaolo Bonzini     assert(size <= ((uint64_t)1 << HBITMAP_LOG_MAX_SIZE));
716e7c033c3SPaolo Bonzini 
717e7c033c3SPaolo Bonzini     hb->size = size;
718e7c033c3SPaolo Bonzini     hb->granularity = granularity;
719e7c033c3SPaolo Bonzini     for (i = HBITMAP_LEVELS; i-- > 0; ) {
720e7c033c3SPaolo Bonzini         size = MAX((size + BITS_PER_LONG - 1) >> BITS_PER_LEVEL, 1);
7218515efbeSJohn Snow         hb->sizes[i] = size;
722e1cf5582SMarkus Armbruster         hb->levels[i] = g_new0(unsigned long, size);
723e7c033c3SPaolo Bonzini     }
724e7c033c3SPaolo Bonzini 
725e7c033c3SPaolo Bonzini     /* We necessarily have free bits in level 0 due to the definition
726e7c033c3SPaolo Bonzini      * of HBITMAP_LEVELS, so use one for a sentinel.  This speeds up
727e7c033c3SPaolo Bonzini      * hbitmap_iter_skip_words.
728e7c033c3SPaolo Bonzini      */
729e7c033c3SPaolo Bonzini     assert(size == 1);
730e7c033c3SPaolo Bonzini     hb->levels[0][0] |= 1UL << (BITS_PER_LONG - 1);
731e7c033c3SPaolo Bonzini     return hb;
732e7c033c3SPaolo Bonzini }
733be58721dSJohn Snow 
734ce1ffea8SJohn Snow void hbitmap_truncate(HBitmap *hb, uint64_t size)
735ce1ffea8SJohn Snow {
736ce1ffea8SJohn Snow     bool shrink;
737ce1ffea8SJohn Snow     unsigned i;
738ce1ffea8SJohn Snow     uint64_t num_elements = size;
739ce1ffea8SJohn Snow     uint64_t old;
740ce1ffea8SJohn Snow 
7414e4de222SVladimir Sementsov-Ogievskiy     hb->orig_size = size;
7424e4de222SVladimir Sementsov-Ogievskiy 
743ce1ffea8SJohn Snow     /* Size comes in as logical elements, adjust for granularity. */
744ce1ffea8SJohn Snow     size = (size + (1ULL << hb->granularity) - 1) >> hb->granularity;
745ce1ffea8SJohn Snow     assert(size <= ((uint64_t)1 << HBITMAP_LOG_MAX_SIZE));
746ce1ffea8SJohn Snow     shrink = size < hb->size;
747ce1ffea8SJohn Snow 
748ce1ffea8SJohn Snow     /* bit sizes are identical; nothing to do. */
749ce1ffea8SJohn Snow     if (size == hb->size) {
750ce1ffea8SJohn Snow         return;
751ce1ffea8SJohn Snow     }
752ce1ffea8SJohn Snow 
753ce1ffea8SJohn Snow     /* If we're losing bits, let's clear those bits before we invalidate all of
754ce1ffea8SJohn Snow      * our invariants. This helps keep the bitcount consistent, and will prevent
755ce1ffea8SJohn Snow      * us from carrying around garbage bits beyond the end of the map.
756ce1ffea8SJohn Snow      */
757ce1ffea8SJohn Snow     if (shrink) {
758ce1ffea8SJohn Snow         /* Don't clear partial granularity groups;
759ce1ffea8SJohn Snow          * start at the first full one. */
7606725f887SMax Reitz         uint64_t start = ROUND_UP(num_elements, UINT64_C(1) << hb->granularity);
761ce1ffea8SJohn Snow         uint64_t fix_count = (hb->size << hb->granularity) - start;
762ce1ffea8SJohn Snow 
763ce1ffea8SJohn Snow         assert(fix_count);
764ce1ffea8SJohn Snow         hbitmap_reset(hb, start, fix_count);
765ce1ffea8SJohn Snow     }
766ce1ffea8SJohn Snow 
767ce1ffea8SJohn Snow     hb->size = size;
768ce1ffea8SJohn Snow     for (i = HBITMAP_LEVELS; i-- > 0; ) {
769ce1ffea8SJohn Snow         size = MAX(BITS_TO_LONGS(size), 1);
770ce1ffea8SJohn Snow         if (hb->sizes[i] == size) {
771ce1ffea8SJohn Snow             break;
772ce1ffea8SJohn Snow         }
773ce1ffea8SJohn Snow         old = hb->sizes[i];
774ce1ffea8SJohn Snow         hb->sizes[i] = size;
775ce1ffea8SJohn Snow         hb->levels[i] = g_realloc(hb->levels[i], size * sizeof(unsigned long));
776ce1ffea8SJohn Snow         if (!shrink) {
777ce1ffea8SJohn Snow             memset(&hb->levels[i][old], 0x00,
778ce1ffea8SJohn Snow                    (size - old) * sizeof(*hb->levels[i]));
779ce1ffea8SJohn Snow         }
780ce1ffea8SJohn Snow     }
78107ac4cdbSFam Zheng     if (hb->meta) {
78207ac4cdbSFam Zheng         hbitmap_truncate(hb->meta, hb->size << hb->granularity);
78307ac4cdbSFam Zheng     }
784ce1ffea8SJohn Snow }
785ce1ffea8SJohn Snow 
786fa000f2fSVladimir Sementsov-Ogievskiy bool hbitmap_can_merge(const HBitmap *a, const HBitmap *b)
787fa000f2fSVladimir Sementsov-Ogievskiy {
788c5b40c1fSJohn Snow     return (a->orig_size == b->orig_size);
789c5b40c1fSJohn Snow }
790c5b40c1fSJohn Snow 
791c5b40c1fSJohn Snow /**
792c5b40c1fSJohn Snow  * hbitmap_sparse_merge: performs dst = dst | src
793c5b40c1fSJohn Snow  * works with differing granularities.
794c5b40c1fSJohn Snow  * best used when src is sparsely populated.
795c5b40c1fSJohn Snow  */
796c5b40c1fSJohn Snow static void hbitmap_sparse_merge(HBitmap *dst, const HBitmap *src)
797c5b40c1fSJohn Snow {
798c5b40c1fSJohn Snow     uint64_t offset = 0;
799c5b40c1fSJohn Snow     uint64_t count = src->orig_size;
800c5b40c1fSJohn Snow 
801c5b40c1fSJohn Snow     while (hbitmap_next_dirty_area(src, &offset, &count)) {
802c5b40c1fSJohn Snow         hbitmap_set(dst, offset, count);
803c5b40c1fSJohn Snow         offset += count;
804c5b40c1fSJohn Snow         if (offset >= src->orig_size) {
805c5b40c1fSJohn Snow             break;
806c5b40c1fSJohn Snow         }
807c5b40c1fSJohn Snow         count = src->orig_size - offset;
808c5b40c1fSJohn Snow     }
809fa000f2fSVladimir Sementsov-Ogievskiy }
810ce1ffea8SJohn Snow 
811be58721dSJohn Snow /**
8123bde4b01SJohn Snow  * Given HBitmaps A and B, let R := A (BITOR) B.
8133bde4b01SJohn Snow  * Bitmaps A and B will not be modified,
8143bde4b01SJohn Snow  *     except when bitmap R is an alias of A or B.
815be58721dSJohn Snow  *
816be58721dSJohn Snow  * @return true if the merge was successful,
817be58721dSJohn Snow  *         false if it was not attempted.
818be58721dSJohn Snow  */
819fa000f2fSVladimir Sementsov-Ogievskiy bool hbitmap_merge(const HBitmap *a, const HBitmap *b, HBitmap *result)
820be58721dSJohn Snow {
821be58721dSJohn Snow     int i;
822be58721dSJohn Snow     uint64_t j;
823be58721dSJohn Snow 
824fa000f2fSVladimir Sementsov-Ogievskiy     if (!hbitmap_can_merge(a, b) || !hbitmap_can_merge(a, result)) {
825be58721dSJohn Snow         return false;
826be58721dSJohn Snow     }
827fa000f2fSVladimir Sementsov-Ogievskiy     assert(hbitmap_can_merge(b, result));
828be58721dSJohn Snow 
8293bde4b01SJohn Snow     if ((!hbitmap_count(a) && result == b) ||
8303bde4b01SJohn Snow         (!hbitmap_count(b) && result == a)) {
8313bde4b01SJohn Snow         return true;
8323bde4b01SJohn Snow     }
8333bde4b01SJohn Snow 
8343bde4b01SJohn Snow     if (!hbitmap_count(a) && !hbitmap_count(b)) {
8353bde4b01SJohn Snow         hbitmap_reset_all(result);
836be58721dSJohn Snow         return true;
837be58721dSJohn Snow     }
838be58721dSJohn Snow 
839c5b40c1fSJohn Snow     if (a->granularity != b->granularity) {
840c5b40c1fSJohn Snow         if ((a != result) && (b != result)) {
841c5b40c1fSJohn Snow             hbitmap_reset_all(result);
842c5b40c1fSJohn Snow         }
843c5b40c1fSJohn Snow         if (a != result) {
844c5b40c1fSJohn Snow             hbitmap_sparse_merge(result, a);
845c5b40c1fSJohn Snow         }
846c5b40c1fSJohn Snow         if (b != result) {
847c5b40c1fSJohn Snow             hbitmap_sparse_merge(result, b);
848c5b40c1fSJohn Snow         }
849c5b40c1fSJohn Snow         return true;
850c5b40c1fSJohn Snow     }
851c5b40c1fSJohn Snow 
852be58721dSJohn Snow     /* This merge is O(size), as BITS_PER_LONG and HBITMAP_LEVELS are constant.
853be58721dSJohn Snow      * It may be possible to improve running times for sparsely populated maps
854be58721dSJohn Snow      * by using hbitmap_iter_next, but this is suboptimal for dense maps.
855be58721dSJohn Snow      */
856c5b40c1fSJohn Snow     assert(a->size == b->size);
857be58721dSJohn Snow     for (i = HBITMAP_LEVELS - 1; i >= 0; i--) {
858be58721dSJohn Snow         for (j = 0; j < a->sizes[i]; j++) {
859fa000f2fSVladimir Sementsov-Ogievskiy             result->levels[i][j] = a->levels[i][j] | b->levels[i][j];
860be58721dSJohn Snow         }
861be58721dSJohn Snow     }
862be58721dSJohn Snow 
863d1dde714SEric Blake     /* Recompute the dirty count */
864d1dde714SEric Blake     result->count = hb_count_between(result, 0, result->size - 1);
865d1dde714SEric Blake 
866be58721dSJohn Snow     return true;
867be58721dSJohn Snow }
86807ac4cdbSFam Zheng 
86907ac4cdbSFam Zheng HBitmap *hbitmap_create_meta(HBitmap *hb, int chunk_size)
87007ac4cdbSFam Zheng {
87107ac4cdbSFam Zheng     assert(!(chunk_size & (chunk_size - 1)));
87207ac4cdbSFam Zheng     assert(!hb->meta);
87307ac4cdbSFam Zheng     hb->meta = hbitmap_alloc(hb->size << hb->granularity,
87407ac4cdbSFam Zheng                              hb->granularity + ctz32(chunk_size));
87507ac4cdbSFam Zheng     return hb->meta;
87607ac4cdbSFam Zheng }
87707ac4cdbSFam Zheng 
87807ac4cdbSFam Zheng void hbitmap_free_meta(HBitmap *hb)
87907ac4cdbSFam Zheng {
88007ac4cdbSFam Zheng     assert(hb->meta);
88107ac4cdbSFam Zheng     hbitmap_free(hb->meta);
88207ac4cdbSFam Zheng     hb->meta = NULL;
88307ac4cdbSFam Zheng }
884a3b52535SVladimir Sementsov-Ogievskiy 
885a3b52535SVladimir Sementsov-Ogievskiy char *hbitmap_sha256(const HBitmap *bitmap, Error **errp)
886a3b52535SVladimir Sementsov-Ogievskiy {
887a3b52535SVladimir Sementsov-Ogievskiy     size_t size = bitmap->sizes[HBITMAP_LEVELS - 1] * sizeof(unsigned long);
888a3b52535SVladimir Sementsov-Ogievskiy     char *data = (char *)bitmap->levels[HBITMAP_LEVELS - 1];
889a3b52535SVladimir Sementsov-Ogievskiy     char *hash = NULL;
890a3b52535SVladimir Sementsov-Ogievskiy     qcrypto_hash_digest(QCRYPTO_HASH_ALG_SHA256, data, size, &hash, errp);
891a3b52535SVladimir Sementsov-Ogievskiy 
892a3b52535SVladimir Sementsov-Ogievskiy     return hash;
893a3b52535SVladimir Sementsov-Ogievskiy }
894