xref: /openbmc/qemu/util/bitmap.c (revision 59a3a1c0)
1 /*
2  * Bitmap Module
3  *
4  * Stolen from linux/src/lib/bitmap.c
5  *
6  * Copyright (C) 2010 Corentin Chary
7  *
8  * This source code is licensed under the GNU General Public License,
9  * Version 2.
10  */
11 
12 #include "qemu/osdep.h"
13 #include "qemu/bitops.h"
14 #include "qemu/bitmap.h"
15 #include "qemu/atomic.h"
16 
17 /*
18  * bitmaps provide an array of bits, implemented using an
19  * array of unsigned longs.  The number of valid bits in a
20  * given bitmap does _not_ need to be an exact multiple of
21  * BITS_PER_LONG.
22  *
23  * The possible unused bits in the last, partially used word
24  * of a bitmap are 'don't care'.  The implementation makes
25  * no particular effort to keep them zero.  It ensures that
26  * their value will not affect the results of any operation.
27  * The bitmap operations that return Boolean (bitmap_empty,
28  * for example) or scalar (bitmap_weight, for example) results
29  * carefully filter out these unused bits from impacting their
30  * results.
31  *
32  * These operations actually hold to a slightly stronger rule:
33  * if you don't input any bitmaps to these ops that have some
34  * unused bits set, then they won't output any set unused bits
35  * in output bitmaps.
36  *
37  * The byte ordering of bitmaps is more natural on little
38  * endian architectures.
39  */
40 
41 int slow_bitmap_empty(const unsigned long *bitmap, long bits)
42 {
43     long k, lim = bits/BITS_PER_LONG;
44 
45     for (k = 0; k < lim; ++k) {
46         if (bitmap[k]) {
47             return 0;
48         }
49     }
50     if (bits % BITS_PER_LONG) {
51         if (bitmap[k] & BITMAP_LAST_WORD_MASK(bits)) {
52             return 0;
53         }
54     }
55 
56     return 1;
57 }
58 
59 int slow_bitmap_full(const unsigned long *bitmap, long bits)
60 {
61     long k, lim = bits/BITS_PER_LONG;
62 
63     for (k = 0; k < lim; ++k) {
64         if (~bitmap[k]) {
65             return 0;
66         }
67     }
68 
69     if (bits % BITS_PER_LONG) {
70         if (~bitmap[k] & BITMAP_LAST_WORD_MASK(bits)) {
71             return 0;
72         }
73     }
74 
75     return 1;
76 }
77 
78 int slow_bitmap_equal(const unsigned long *bitmap1,
79                       const unsigned long *bitmap2, long bits)
80 {
81     long k, lim = bits/BITS_PER_LONG;
82 
83     for (k = 0; k < lim; ++k) {
84         if (bitmap1[k] != bitmap2[k]) {
85             return 0;
86         }
87     }
88 
89     if (bits % BITS_PER_LONG) {
90         if ((bitmap1[k] ^ bitmap2[k]) & BITMAP_LAST_WORD_MASK(bits)) {
91             return 0;
92         }
93     }
94 
95     return 1;
96 }
97 
98 void slow_bitmap_complement(unsigned long *dst, const unsigned long *src,
99                             long bits)
100 {
101     long k, lim = bits/BITS_PER_LONG;
102 
103     for (k = 0; k < lim; ++k) {
104         dst[k] = ~src[k];
105     }
106 
107     if (bits % BITS_PER_LONG) {
108         dst[k] = ~src[k] & BITMAP_LAST_WORD_MASK(bits);
109     }
110 }
111 
112 int slow_bitmap_and(unsigned long *dst, const unsigned long *bitmap1,
113                     const unsigned long *bitmap2, long bits)
114 {
115     long k;
116     long nr = BITS_TO_LONGS(bits);
117     unsigned long result = 0;
118 
119     for (k = 0; k < nr; k++) {
120         result |= (dst[k] = bitmap1[k] & bitmap2[k]);
121     }
122     return result != 0;
123 }
124 
125 void slow_bitmap_or(unsigned long *dst, const unsigned long *bitmap1,
126                     const unsigned long *bitmap2, long bits)
127 {
128     long k;
129     long nr = BITS_TO_LONGS(bits);
130 
131     for (k = 0; k < nr; k++) {
132         dst[k] = bitmap1[k] | bitmap2[k];
133     }
134 }
135 
136 void slow_bitmap_xor(unsigned long *dst, const unsigned long *bitmap1,
137                      const unsigned long *bitmap2, long bits)
138 {
139     long k;
140     long nr = BITS_TO_LONGS(bits);
141 
142     for (k = 0; k < nr; k++) {
143         dst[k] = bitmap1[k] ^ bitmap2[k];
144     }
145 }
146 
147 int slow_bitmap_andnot(unsigned long *dst, const unsigned long *bitmap1,
148                        const unsigned long *bitmap2, long bits)
149 {
150     long k;
151     long nr = BITS_TO_LONGS(bits);
152     unsigned long result = 0;
153 
154     for (k = 0; k < nr; k++) {
155         result |= (dst[k] = bitmap1[k] & ~bitmap2[k]);
156     }
157     return result != 0;
158 }
159 
160 void bitmap_set(unsigned long *map, long start, long nr)
161 {
162     unsigned long *p = map + BIT_WORD(start);
163     const long size = start + nr;
164     int bits_to_set = BITS_PER_LONG - (start % BITS_PER_LONG);
165     unsigned long mask_to_set = BITMAP_FIRST_WORD_MASK(start);
166 
167     assert(start >= 0 && nr >= 0);
168 
169     while (nr - bits_to_set >= 0) {
170         *p |= mask_to_set;
171         nr -= bits_to_set;
172         bits_to_set = BITS_PER_LONG;
173         mask_to_set = ~0UL;
174         p++;
175     }
176     if (nr) {
177         mask_to_set &= BITMAP_LAST_WORD_MASK(size);
178         *p |= mask_to_set;
179     }
180 }
181 
182 void bitmap_set_atomic(unsigned long *map, long start, long nr)
183 {
184     unsigned long *p = map + BIT_WORD(start);
185     const long size = start + nr;
186     int bits_to_set = BITS_PER_LONG - (start % BITS_PER_LONG);
187     unsigned long mask_to_set = BITMAP_FIRST_WORD_MASK(start);
188 
189     assert(start >= 0 && nr >= 0);
190 
191     /* First word */
192     if (nr - bits_to_set > 0) {
193         atomic_or(p, mask_to_set);
194         nr -= bits_to_set;
195         bits_to_set = BITS_PER_LONG;
196         mask_to_set = ~0UL;
197         p++;
198     }
199 
200     /* Full words */
201     if (bits_to_set == BITS_PER_LONG) {
202         while (nr >= BITS_PER_LONG) {
203             *p = ~0UL;
204             nr -= BITS_PER_LONG;
205             p++;
206         }
207     }
208 
209     /* Last word */
210     if (nr) {
211         mask_to_set &= BITMAP_LAST_WORD_MASK(size);
212         atomic_or(p, mask_to_set);
213     } else {
214         /* If we avoided the full barrier in atomic_or(), issue a
215          * barrier to account for the assignments in the while loop.
216          */
217         smp_mb();
218     }
219 }
220 
221 void bitmap_clear(unsigned long *map, long start, long nr)
222 {
223     unsigned long *p = map + BIT_WORD(start);
224     const long size = start + nr;
225     int bits_to_clear = BITS_PER_LONG - (start % BITS_PER_LONG);
226     unsigned long mask_to_clear = BITMAP_FIRST_WORD_MASK(start);
227 
228     assert(start >= 0 && nr >= 0);
229 
230     while (nr - bits_to_clear >= 0) {
231         *p &= ~mask_to_clear;
232         nr -= bits_to_clear;
233         bits_to_clear = BITS_PER_LONG;
234         mask_to_clear = ~0UL;
235         p++;
236     }
237     if (nr) {
238         mask_to_clear &= BITMAP_LAST_WORD_MASK(size);
239         *p &= ~mask_to_clear;
240     }
241 }
242 
243 bool bitmap_test_and_clear_atomic(unsigned long *map, long start, long nr)
244 {
245     unsigned long *p = map + BIT_WORD(start);
246     const long size = start + nr;
247     int bits_to_clear = BITS_PER_LONG - (start % BITS_PER_LONG);
248     unsigned long mask_to_clear = BITMAP_FIRST_WORD_MASK(start);
249     unsigned long dirty = 0;
250     unsigned long old_bits;
251 
252     assert(start >= 0 && nr >= 0);
253 
254     /* First word */
255     if (nr - bits_to_clear > 0) {
256         old_bits = atomic_fetch_and(p, ~mask_to_clear);
257         dirty |= old_bits & mask_to_clear;
258         nr -= bits_to_clear;
259         bits_to_clear = BITS_PER_LONG;
260         mask_to_clear = ~0UL;
261         p++;
262     }
263 
264     /* Full words */
265     if (bits_to_clear == BITS_PER_LONG) {
266         while (nr >= BITS_PER_LONG) {
267             if (*p) {
268                 old_bits = atomic_xchg(p, 0);
269                 dirty |= old_bits;
270             }
271             nr -= BITS_PER_LONG;
272             p++;
273         }
274     }
275 
276     /* Last word */
277     if (nr) {
278         mask_to_clear &= BITMAP_LAST_WORD_MASK(size);
279         old_bits = atomic_fetch_and(p, ~mask_to_clear);
280         dirty |= old_bits & mask_to_clear;
281     } else {
282         if (!dirty) {
283             smp_mb();
284         }
285     }
286 
287     return dirty != 0;
288 }
289 
290 void bitmap_copy_and_clear_atomic(unsigned long *dst, unsigned long *src,
291                                   long nr)
292 {
293     while (nr > 0) {
294         *dst = atomic_xchg(src, 0);
295         dst++;
296         src++;
297         nr -= BITS_PER_LONG;
298     }
299 }
300 
301 #define ALIGN_MASK(x,mask)      (((x)+(mask))&~(mask))
302 
303 /**
304  * bitmap_find_next_zero_area - find a contiguous aligned zero area
305  * @map: The address to base the search on
306  * @size: The bitmap size in bits
307  * @start: The bitnumber to start searching at
308  * @nr: The number of zeroed bits we're looking for
309  * @align_mask: Alignment mask for zero area
310  *
311  * The @align_mask should be one less than a power of 2; the effect is that
312  * the bit offset of all zero areas this function finds is multiples of that
313  * power of 2. A @align_mask of 0 means no alignment is required.
314  */
315 unsigned long bitmap_find_next_zero_area(unsigned long *map,
316                                          unsigned long size,
317                                          unsigned long start,
318                                          unsigned long nr,
319                                          unsigned long align_mask)
320 {
321     unsigned long index, end, i;
322 again:
323     index = find_next_zero_bit(map, size, start);
324 
325     /* Align allocation */
326     index = ALIGN_MASK(index, align_mask);
327 
328     end = index + nr;
329     if (end > size) {
330         return end;
331     }
332     i = find_next_bit(map, end, index);
333     if (i < end) {
334         start = i + 1;
335         goto again;
336     }
337     return index;
338 }
339 
340 int slow_bitmap_intersects(const unsigned long *bitmap1,
341                            const unsigned long *bitmap2, long bits)
342 {
343     long k, lim = bits/BITS_PER_LONG;
344 
345     for (k = 0; k < lim; ++k) {
346         if (bitmap1[k] & bitmap2[k]) {
347             return 1;
348         }
349     }
350 
351     if (bits % BITS_PER_LONG) {
352         if ((bitmap1[k] & bitmap2[k]) & BITMAP_LAST_WORD_MASK(bits)) {
353             return 1;
354         }
355     }
356     return 0;
357 }
358 
359 long slow_bitmap_count_one(const unsigned long *bitmap, long nbits)
360 {
361     long k, lim = nbits / BITS_PER_LONG, result = 0;
362 
363     for (k = 0; k < lim; k++) {
364         result += ctpopl(bitmap[k]);
365     }
366 
367     if (nbits % BITS_PER_LONG) {
368         result += ctpopl(bitmap[k] & BITMAP_LAST_WORD_MASK(nbits));
369     }
370 
371     return result;
372 }
373 
374 static void bitmap_to_from_le(unsigned long *dst,
375                               const unsigned long *src, long nbits)
376 {
377     long len = BITS_TO_LONGS(nbits);
378 
379 #ifdef HOST_WORDS_BIGENDIAN
380     long index;
381 
382     for (index = 0; index < len; index++) {
383 # if HOST_LONG_BITS == 64
384         dst[index] = bswap64(src[index]);
385 # else
386         dst[index] = bswap32(src[index]);
387 # endif
388     }
389 #else
390     memcpy(dst, src, len * sizeof(unsigned long));
391 #endif
392 }
393 
394 void bitmap_from_le(unsigned long *dst, const unsigned long *src,
395                     long nbits)
396 {
397     bitmap_to_from_le(dst, src, nbits);
398 }
399 
400 void bitmap_to_le(unsigned long *dst, const unsigned long *src,
401                   long nbits)
402 {
403     bitmap_to_from_le(dst, src, nbits);
404 }
405 
406 /*
407  * Copy "src" bitmap with a positive offset and put it into the "dst"
408  * bitmap.  The caller needs to make sure the bitmap size of "src"
409  * is bigger than (shift + nbits).
410  */
411 void bitmap_copy_with_src_offset(unsigned long *dst, const unsigned long *src,
412                                  unsigned long shift, unsigned long nbits)
413 {
414     unsigned long left_mask, right_mask, last_mask;
415 
416     /* Proper shift src pointer to the first word to copy from */
417     src += BIT_WORD(shift);
418     shift %= BITS_PER_LONG;
419 
420     if (!shift) {
421         /* Fast path */
422         bitmap_copy(dst, src, nbits);
423         return;
424     }
425 
426     right_mask = (1ul << shift) - 1;
427     left_mask = ~right_mask;
428 
429     while (nbits >= BITS_PER_LONG) {
430         *dst = (*src & left_mask) >> shift;
431         *dst |= (src[1] & right_mask) << (BITS_PER_LONG - shift);
432         dst++;
433         src++;
434         nbits -= BITS_PER_LONG;
435     }
436 
437     if (nbits > BITS_PER_LONG - shift) {
438         *dst = (*src & left_mask) >> shift;
439         nbits -= BITS_PER_LONG - shift;
440         last_mask = (1ul << nbits) - 1;
441         *dst |= (src[1] & last_mask) << (BITS_PER_LONG - shift);
442     } else if (nbits) {
443         last_mask = (1ul << nbits) - 1;
444         *dst = (*src >> shift) & last_mask;
445     }
446 }
447 
448 /*
449  * Copy "src" bitmap into the "dst" bitmap with an offset in the
450  * "dst".  The caller needs to make sure the bitmap size of "dst" is
451  * bigger than (shift + nbits).
452  */
453 void bitmap_copy_with_dst_offset(unsigned long *dst, const unsigned long *src,
454                                  unsigned long shift, unsigned long nbits)
455 {
456     unsigned long left_mask, right_mask, last_mask;
457 
458     /* Proper shift dst pointer to the first word to copy from */
459     dst += BIT_WORD(shift);
460     shift %= BITS_PER_LONG;
461 
462     if (!shift) {
463         /* Fast path */
464         bitmap_copy(dst, src, nbits);
465         return;
466     }
467 
468     right_mask = (1ul << (BITS_PER_LONG - shift)) - 1;
469     left_mask = ~right_mask;
470 
471     *dst &= (1ul << shift) - 1;
472     while (nbits >= BITS_PER_LONG) {
473         *dst |= (*src & right_mask) << shift;
474         dst[1] = (*src & left_mask) >> (BITS_PER_LONG - shift);
475         dst++;
476         src++;
477         nbits -= BITS_PER_LONG;
478     }
479 
480     if (nbits > BITS_PER_LONG - shift) {
481         *dst |= (*src & right_mask) << shift;
482         nbits -= BITS_PER_LONG - shift;
483         last_mask = ((1ul << nbits) - 1) << (BITS_PER_LONG - shift);
484         dst[1] = (*src & last_mask) >> (BITS_PER_LONG - shift);
485     } else if (nbits) {
486         last_mask = (1ul << nbits) - 1;
487         *dst |= (*src & last_mask) << shift;
488     }
489 }
490