s390/bitops: implement cache friendly test_and_set_bit_lock
[deliverable/linux.git] / arch / s390 / include / asm / bitops.h
1 /*
2 * Copyright IBM Corp. 1999,2013
3 *
4 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>,
5 *
6 * The description below was taken in large parts from the powerpc
7 * bitops header file:
8 * Within a word, bits are numbered LSB first. Lot's of places make
9 * this assumption by directly testing bits with (val & (1<<nr)).
10 * This can cause confusion for large (> 1 word) bitmaps on a
11 * big-endian system because, unlike little endian, the number of each
12 * bit depends on the word size.
13 *
14 * The bitop functions are defined to work on unsigned longs, so for an
15 * s390x system the bits end up numbered:
16 * |63..............0|127............64|191...........128|255...........192|
17 * and on s390:
18 * |31.....0|63....32|95....64|127...96|159..128|191..160|223..192|255..224|
19 *
20 * There are a few little-endian macros used mostly for filesystem
21 * bitmaps, these work on similar bit arrays layouts, but
22 * byte-oriented:
23 * |7...0|15...8|23...16|31...24|39...32|47...40|55...48|63...56|
24 *
25 * The main difference is that bit 3-5 (64b) or 3-4 (32b) in the bit
26 * number field needs to be reversed compared to the big-endian bit
27 * fields. This can be achieved by XOR with 0x38 (64b) or 0x18 (32b).
28 *
29 * We also have special functions which work with an MSB0 encoding:
30 * on an s390x system the bits are numbered:
31 * |0..............63|64............127|128...........191|192...........255|
32 * and on s390:
33 * |0.....31|32....63|64....95|96...127|128..159|160..191|192..223|224..255|
34 *
35 * The main difference is that bit 0-63 (64b) or 0-31 (32b) in the bit
36 * number field needs to be reversed compared to the LSB0 encoded bit
37 * fields. This can be achieved by XOR with 0x3f (64b) or 0x1f (32b).
38 *
39 */
40
41 #ifndef _S390_BITOPS_H
42 #define _S390_BITOPS_H
43
44 #ifndef _LINUX_BITOPS_H
45 #error only <linux/bitops.h> can be included directly
46 #endif
47
48 #include <linux/typecheck.h>
49 #include <linux/compiler.h>
50 #include <asm/barrier.h>
51
52 #define __BITOPS_NO_BARRIER "\n"
53
54 #ifdef CONFIG_HAVE_MARCH_Z196_FEATURES
55
56 #define __BITOPS_OR "laog"
57 #define __BITOPS_AND "lang"
58 #define __BITOPS_XOR "laxg"
59 #define __BITOPS_BARRIER "bcr 14,0\n"
60
61 #define __BITOPS_LOOP(__addr, __val, __op_string, __barrier) \
62 ({ \
63 unsigned long __old; \
64 \
65 typecheck(unsigned long *, (__addr)); \
66 asm volatile( \
67 __barrier \
68 __op_string " %0,%2,%1\n" \
69 __barrier \
70 : "=d" (__old), "+Q" (*(__addr)) \
71 : "d" (__val) \
72 : "cc", "memory"); \
73 __old; \
74 })
75
76 #else /* CONFIG_HAVE_MARCH_Z196_FEATURES */
77
78 #define __BITOPS_OR "ogr"
79 #define __BITOPS_AND "ngr"
80 #define __BITOPS_XOR "xgr"
81 #define __BITOPS_BARRIER "\n"
82
83 #define __BITOPS_LOOP(__addr, __val, __op_string, __barrier) \
84 ({ \
85 unsigned long __old, __new; \
86 \
87 typecheck(unsigned long *, (__addr)); \
88 asm volatile( \
89 " lg %0,%2\n" \
90 "0: lgr %1,%0\n" \
91 __op_string " %1,%3\n" \
92 " csg %0,%1,%2\n" \
93 " jl 0b" \
94 : "=&d" (__old), "=&d" (__new), "+Q" (*(__addr))\
95 : "d" (__val) \
96 : "cc", "memory"); \
97 __old; \
98 })
99
100 #endif /* CONFIG_HAVE_MARCH_Z196_FEATURES */
101
102 #define __BITOPS_WORDS(bits) (((bits) + BITS_PER_LONG - 1) / BITS_PER_LONG)
103
104 static inline unsigned long *
105 __bitops_word(unsigned long nr, volatile unsigned long *ptr)
106 {
107 unsigned long addr;
108
109 addr = (unsigned long)ptr + ((nr ^ (nr & (BITS_PER_LONG - 1))) >> 3);
110 return (unsigned long *)addr;
111 }
112
113 static inline unsigned char *
114 __bitops_byte(unsigned long nr, volatile unsigned long *ptr)
115 {
116 return ((unsigned char *)ptr) + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
117 }
118
119 static inline void set_bit(unsigned long nr, volatile unsigned long *ptr)
120 {
121 unsigned long *addr = __bitops_word(nr, ptr);
122 unsigned long mask;
123
124 #ifdef CONFIG_HAVE_MARCH_ZEC12_FEATURES
125 if (__builtin_constant_p(nr)) {
126 unsigned char *caddr = __bitops_byte(nr, ptr);
127
128 asm volatile(
129 "oi %0,%b1\n"
130 : "+Q" (*caddr)
131 : "i" (1 << (nr & 7))
132 : "cc", "memory");
133 return;
134 }
135 #endif
136 mask = 1UL << (nr & (BITS_PER_LONG - 1));
137 __BITOPS_LOOP(addr, mask, __BITOPS_OR, __BITOPS_NO_BARRIER);
138 }
139
140 static inline void clear_bit(unsigned long nr, volatile unsigned long *ptr)
141 {
142 unsigned long *addr = __bitops_word(nr, ptr);
143 unsigned long mask;
144
145 #ifdef CONFIG_HAVE_MARCH_ZEC12_FEATURES
146 if (__builtin_constant_p(nr)) {
147 unsigned char *caddr = __bitops_byte(nr, ptr);
148
149 asm volatile(
150 "ni %0,%b1\n"
151 : "+Q" (*caddr)
152 : "i" (~(1 << (nr & 7)))
153 : "cc", "memory");
154 return;
155 }
156 #endif
157 mask = ~(1UL << (nr & (BITS_PER_LONG - 1)));
158 __BITOPS_LOOP(addr, mask, __BITOPS_AND, __BITOPS_NO_BARRIER);
159 }
160
161 static inline void change_bit(unsigned long nr, volatile unsigned long *ptr)
162 {
163 unsigned long *addr = __bitops_word(nr, ptr);
164 unsigned long mask;
165
166 #ifdef CONFIG_HAVE_MARCH_ZEC12_FEATURES
167 if (__builtin_constant_p(nr)) {
168 unsigned char *caddr = __bitops_byte(nr, ptr);
169
170 asm volatile(
171 "xi %0,%b1\n"
172 : "+Q" (*caddr)
173 : "i" (1 << (nr & 7))
174 : "cc", "memory");
175 return;
176 }
177 #endif
178 mask = 1UL << (nr & (BITS_PER_LONG - 1));
179 __BITOPS_LOOP(addr, mask, __BITOPS_XOR, __BITOPS_NO_BARRIER);
180 }
181
182 static inline int
183 test_and_set_bit(unsigned long nr, volatile unsigned long *ptr)
184 {
185 unsigned long *addr = __bitops_word(nr, ptr);
186 unsigned long old, mask;
187
188 mask = 1UL << (nr & (BITS_PER_LONG - 1));
189 old = __BITOPS_LOOP(addr, mask, __BITOPS_OR, __BITOPS_BARRIER);
190 return (old & mask) != 0;
191 }
192
193 static inline int
194 test_and_clear_bit(unsigned long nr, volatile unsigned long *ptr)
195 {
196 unsigned long *addr = __bitops_word(nr, ptr);
197 unsigned long old, mask;
198
199 mask = ~(1UL << (nr & (BITS_PER_LONG - 1)));
200 old = __BITOPS_LOOP(addr, mask, __BITOPS_AND, __BITOPS_BARRIER);
201 return (old & ~mask) != 0;
202 }
203
204 static inline int
205 test_and_change_bit(unsigned long nr, volatile unsigned long *ptr)
206 {
207 unsigned long *addr = __bitops_word(nr, ptr);
208 unsigned long old, mask;
209
210 mask = 1UL << (nr & (BITS_PER_LONG - 1));
211 old = __BITOPS_LOOP(addr, mask, __BITOPS_XOR, __BITOPS_BARRIER);
212 return (old & mask) != 0;
213 }
214
215 static inline void __set_bit(unsigned long nr, volatile unsigned long *ptr)
216 {
217 unsigned char *addr = __bitops_byte(nr, ptr);
218
219 *addr |= 1 << (nr & 7);
220 }
221
222 static inline void
223 __clear_bit(unsigned long nr, volatile unsigned long *ptr)
224 {
225 unsigned char *addr = __bitops_byte(nr, ptr);
226
227 *addr &= ~(1 << (nr & 7));
228 }
229
230 static inline void __change_bit(unsigned long nr, volatile unsigned long *ptr)
231 {
232 unsigned char *addr = __bitops_byte(nr, ptr);
233
234 *addr ^= 1 << (nr & 7);
235 }
236
237 static inline int
238 __test_and_set_bit(unsigned long nr, volatile unsigned long *ptr)
239 {
240 unsigned char *addr = __bitops_byte(nr, ptr);
241 unsigned char ch;
242
243 ch = *addr;
244 *addr |= 1 << (nr & 7);
245 return (ch >> (nr & 7)) & 1;
246 }
247
248 static inline int
249 __test_and_clear_bit(unsigned long nr, volatile unsigned long *ptr)
250 {
251 unsigned char *addr = __bitops_byte(nr, ptr);
252 unsigned char ch;
253
254 ch = *addr;
255 *addr &= ~(1 << (nr & 7));
256 return (ch >> (nr & 7)) & 1;
257 }
258
259 static inline int
260 __test_and_change_bit(unsigned long nr, volatile unsigned long *ptr)
261 {
262 unsigned char *addr = __bitops_byte(nr, ptr);
263 unsigned char ch;
264
265 ch = *addr;
266 *addr ^= 1 << (nr & 7);
267 return (ch >> (nr & 7)) & 1;
268 }
269
270 static inline int test_bit(unsigned long nr, const volatile unsigned long *ptr)
271 {
272 const volatile unsigned char *addr;
273
274 addr = ((const volatile unsigned char *)ptr);
275 addr += (nr ^ (BITS_PER_LONG - 8)) >> 3;
276 return (*addr >> (nr & 7)) & 1;
277 }
278
279 static inline int test_and_set_bit_lock(unsigned long nr,
280 volatile unsigned long *ptr)
281 {
282 if (test_bit(nr, ptr))
283 return 1;
284 return test_and_set_bit(nr, ptr);
285 }
286
287 static inline void clear_bit_unlock(unsigned long nr,
288 volatile unsigned long *ptr)
289 {
290 smp_mb__before_atomic();
291 clear_bit(nr, ptr);
292 }
293
294 static inline void __clear_bit_unlock(unsigned long nr,
295 volatile unsigned long *ptr)
296 {
297 smp_mb();
298 __clear_bit(nr, ptr);
299 }
300
301 /*
302 * Functions which use MSB0 bit numbering.
303 * On an s390x system the bits are numbered:
304 * |0..............63|64............127|128...........191|192...........255|
305 * and on s390:
306 * |0.....31|32....63|64....95|96...127|128..159|160..191|192..223|224..255|
307 */
308 unsigned long find_first_bit_inv(const unsigned long *addr, unsigned long size);
309 unsigned long find_next_bit_inv(const unsigned long *addr, unsigned long size,
310 unsigned long offset);
311
312 static inline void set_bit_inv(unsigned long nr, volatile unsigned long *ptr)
313 {
314 return set_bit(nr ^ (BITS_PER_LONG - 1), ptr);
315 }
316
317 static inline void clear_bit_inv(unsigned long nr, volatile unsigned long *ptr)
318 {
319 return clear_bit(nr ^ (BITS_PER_LONG - 1), ptr);
320 }
321
322 static inline void __set_bit_inv(unsigned long nr, volatile unsigned long *ptr)
323 {
324 return __set_bit(nr ^ (BITS_PER_LONG - 1), ptr);
325 }
326
327 static inline void __clear_bit_inv(unsigned long nr, volatile unsigned long *ptr)
328 {
329 return __clear_bit(nr ^ (BITS_PER_LONG - 1), ptr);
330 }
331
332 static inline int test_bit_inv(unsigned long nr,
333 const volatile unsigned long *ptr)
334 {
335 return test_bit(nr ^ (BITS_PER_LONG - 1), ptr);
336 }
337
338 #ifdef CONFIG_HAVE_MARCH_Z9_109_FEATURES
339
340 /**
341 * __flogr - find leftmost one
342 * @word - The word to search
343 *
344 * Returns the bit number of the most significant bit set,
345 * where the most significant bit has bit number 0.
346 * If no bit is set this function returns 64.
347 */
348 static inline unsigned char __flogr(unsigned long word)
349 {
350 if (__builtin_constant_p(word)) {
351 unsigned long bit = 0;
352
353 if (!word)
354 return 64;
355 if (!(word & 0xffffffff00000000UL)) {
356 word <<= 32;
357 bit += 32;
358 }
359 if (!(word & 0xffff000000000000UL)) {
360 word <<= 16;
361 bit += 16;
362 }
363 if (!(word & 0xff00000000000000UL)) {
364 word <<= 8;
365 bit += 8;
366 }
367 if (!(word & 0xf000000000000000UL)) {
368 word <<= 4;
369 bit += 4;
370 }
371 if (!(word & 0xc000000000000000UL)) {
372 word <<= 2;
373 bit += 2;
374 }
375 if (!(word & 0x8000000000000000UL)) {
376 word <<= 1;
377 bit += 1;
378 }
379 return bit;
380 } else {
381 register unsigned long bit asm("4") = word;
382 register unsigned long out asm("5");
383
384 asm volatile(
385 " flogr %[bit],%[bit]\n"
386 : [bit] "+d" (bit), [out] "=d" (out) : : "cc");
387 return bit;
388 }
389 }
390
391 /**
392 * __ffs - find first bit in word.
393 * @word: The word to search
394 *
395 * Undefined if no bit exists, so code should check against 0 first.
396 */
397 static inline unsigned long __ffs(unsigned long word)
398 {
399 return __flogr(-word & word) ^ (BITS_PER_LONG - 1);
400 }
401
402 /**
403 * ffs - find first bit set
404 * @word: the word to search
405 *
406 * This is defined the same way as the libc and
407 * compiler builtin ffs routines (man ffs).
408 */
409 static inline int ffs(int word)
410 {
411 unsigned long mask = 2 * BITS_PER_LONG - 1;
412 unsigned int val = (unsigned int)word;
413
414 return (1 + (__flogr(-val & val) ^ (BITS_PER_LONG - 1))) & mask;
415 }
416
417 /**
418 * __fls - find last (most-significant) set bit in a long word
419 * @word: the word to search
420 *
421 * Undefined if no set bit exists, so code should check against 0 first.
422 */
423 static inline unsigned long __fls(unsigned long word)
424 {
425 return __flogr(word) ^ (BITS_PER_LONG - 1);
426 }
427
428 /**
429 * fls64 - find last set bit in a 64-bit word
430 * @word: the word to search
431 *
432 * This is defined in a similar way as the libc and compiler builtin
433 * ffsll, but returns the position of the most significant set bit.
434 *
435 * fls64(value) returns 0 if value is 0 or the position of the last
436 * set bit if value is nonzero. The last (most significant) bit is
437 * at position 64.
438 */
439 static inline int fls64(unsigned long word)
440 {
441 unsigned long mask = 2 * BITS_PER_LONG - 1;
442
443 return (1 + (__flogr(word) ^ (BITS_PER_LONG - 1))) & mask;
444 }
445
446 /**
447 * fls - find last (most-significant) bit set
448 * @word: the word to search
449 *
450 * This is defined the same way as ffs.
451 * Note fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32.
452 */
453 static inline int fls(int word)
454 {
455 return fls64((unsigned int)word);
456 }
457
458 #else /* CONFIG_HAVE_MARCH_Z9_109_FEATURES */
459
460 #include <asm-generic/bitops/__ffs.h>
461 #include <asm-generic/bitops/ffs.h>
462 #include <asm-generic/bitops/__fls.h>
463 #include <asm-generic/bitops/fls.h>
464 #include <asm-generic/bitops/fls64.h>
465
466 #endif /* CONFIG_HAVE_MARCH_Z9_109_FEATURES */
467
468 #include <asm-generic/bitops/ffz.h>
469 #include <asm-generic/bitops/find.h>
470 #include <asm-generic/bitops/hweight.h>
471 #include <asm-generic/bitops/sched.h>
472 #include <asm-generic/bitops/le.h>
473 #include <asm-generic/bitops/ext2-atomic-setbit.h>
474
475 #endif /* _S390_BITOPS_H */
This page took 0.054411 seconds and 5 git commands to generate.