xref: /openbmc/linux/arch/s390/include/asm/bitops.h (revision b6bec26c)
1 /*
2  *  S390 version
3  *    Copyright IBM Corp. 1999
4  *    Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
5  *
6  *  Derived from "include/asm-i386/bitops.h"
7  *    Copyright (C) 1992, Linus Torvalds
8  *
9  */
10 
11 #ifndef _S390_BITOPS_H
12 #define _S390_BITOPS_H
13 
14 #ifndef _LINUX_BITOPS_H
15 #error only <linux/bitops.h> can be included directly
16 #endif
17 
18 #include <linux/compiler.h>
19 
20 /*
21  * 32 bit bitops format:
22  * bit 0 is the LSB of *addr; bit 31 is the MSB of *addr;
23  * bit 32 is the LSB of *(addr+4). That combined with the
24  * big endian byte order on S390 give the following bit
25  * order in memory:
26  *    1f 1e 1d 1c 1b 1a 19 18 17 16 15 14 13 12 11 10 \
27  *    0f 0e 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00
28  * after that follows the next long with bit numbers
29  *    3f 3e 3d 3c 3b 3a 39 38 37 36 35 34 33 32 31 30
30  *    2f 2e 2d 2c 2b 2a 29 28 27 26 25 24 23 22 21 20
31  * The reason for this bit ordering is the fact that
32  * in the architecture independent code bits operations
33  * of the form "flags |= (1 << bitnr)" are used INTERMIXED
34  * with operation of the form "set_bit(bitnr, flags)".
35  *
36  * 64 bit bitops format:
37  * bit 0 is the LSB of *addr; bit 63 is the MSB of *addr;
38  * bit 64 is the LSB of *(addr+8). That combined with the
39  * big endian byte order on S390 give the following bit
40  * order in memory:
41  *    3f 3e 3d 3c 3b 3a 39 38 37 36 35 34 33 32 31 30
42  *    2f 2e 2d 2c 2b 2a 29 28 27 26 25 24 23 22 21 20
43  *    1f 1e 1d 1c 1b 1a 19 18 17 16 15 14 13 12 11 10
44  *    0f 0e 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00
45  * after that follows the next long with bit numbers
46  *    7f 7e 7d 7c 7b 7a 79 78 77 76 75 74 73 72 71 70
47  *    6f 6e 6d 6c 6b 6a 69 68 67 66 65 64 63 62 61 60
48  *    5f 5e 5d 5c 5b 5a 59 58 57 56 55 54 53 52 51 50
49  *    4f 4e 4d 4c 4b 4a 49 48 47 46 45 44 43 42 41 40
50  * The reason for this bit ordering is the fact that
51  * in the architecture independent code bits operations
52  * of the form "flags |= (1 << bitnr)" are used INTERMIXED
53  * with operation of the form "set_bit(bitnr, flags)".
54  */
55 
56 /* bitmap tables from arch/s390/kernel/bitmap.c */
57 extern const char _oi_bitmap[];
58 extern const char _ni_bitmap[];
59 extern const char _zb_findmap[];
60 extern const char _sb_findmap[];
61 
62 #ifndef CONFIG_64BIT
63 
64 #define __BITOPS_ALIGN		3
65 #define __BITOPS_WORDSIZE	32
66 #define __BITOPS_OR		"or"
67 #define __BITOPS_AND		"nr"
68 #define __BITOPS_XOR		"xr"
69 
70 #define __BITOPS_LOOP(__old, __new, __addr, __val, __op_string)	\
71 	asm volatile(						\
72 		"	l	%0,%2\n"			\
73 		"0:	lr	%1,%0\n"			\
74 		__op_string "	%1,%3\n"			\
75 		"	cs	%0,%1,%2\n"			\
76 		"	jl	0b"				\
77 		: "=&d" (__old), "=&d" (__new),			\
78 		  "=Q" (*(unsigned long *) __addr)		\
79 		: "d" (__val), "Q" (*(unsigned long *) __addr)	\
80 		: "cc");
81 
82 #else /* CONFIG_64BIT */
83 
84 #define __BITOPS_ALIGN		7
85 #define __BITOPS_WORDSIZE	64
86 #define __BITOPS_OR		"ogr"
87 #define __BITOPS_AND		"ngr"
88 #define __BITOPS_XOR		"xgr"
89 
90 #define __BITOPS_LOOP(__old, __new, __addr, __val, __op_string)	\
91 	asm volatile(						\
92 		"	lg	%0,%2\n"			\
93 		"0:	lgr	%1,%0\n"			\
94 		__op_string "	%1,%3\n"			\
95 		"	csg	%0,%1,%2\n"			\
96 		"	jl	0b"				\
97 		: "=&d" (__old), "=&d" (__new),			\
98 		  "=Q" (*(unsigned long *) __addr)		\
99 		: "d" (__val), "Q" (*(unsigned long *) __addr)	\
100 		: "cc");
101 
102 #endif /* CONFIG_64BIT */
103 
104 #define __BITOPS_WORDS(bits) (((bits)+__BITOPS_WORDSIZE-1)/__BITOPS_WORDSIZE)
105 #define __BITOPS_BARRIER() asm volatile("" : : : "memory")
106 
107 #ifdef CONFIG_SMP
108 /*
109  * SMP safe set_bit routine based on compare and swap (CS)
110  */
111 static inline void set_bit_cs(unsigned long nr, volatile unsigned long *ptr)
112 {
113         unsigned long addr, old, new, mask;
114 
115 	addr = (unsigned long) ptr;
116 	/* calculate address for CS */
117 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
118 	/* make OR mask */
119 	mask = 1UL << (nr & (__BITOPS_WORDSIZE - 1));
120 	/* Do the atomic update. */
121 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_OR);
122 }
123 
124 /*
125  * SMP safe clear_bit routine based on compare and swap (CS)
126  */
127 static inline void clear_bit_cs(unsigned long nr, volatile unsigned long *ptr)
128 {
129         unsigned long addr, old, new, mask;
130 
131 	addr = (unsigned long) ptr;
132 	/* calculate address for CS */
133 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
134 	/* make AND mask */
135 	mask = ~(1UL << (nr & (__BITOPS_WORDSIZE - 1)));
136 	/* Do the atomic update. */
137 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_AND);
138 }
139 
140 /*
141  * SMP safe change_bit routine based on compare and swap (CS)
142  */
143 static inline void change_bit_cs(unsigned long nr, volatile unsigned long *ptr)
144 {
145         unsigned long addr, old, new, mask;
146 
147 	addr = (unsigned long) ptr;
148 	/* calculate address for CS */
149 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
150 	/* make XOR mask */
151 	mask = 1UL << (nr & (__BITOPS_WORDSIZE - 1));
152 	/* Do the atomic update. */
153 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_XOR);
154 }
155 
156 /*
157  * SMP safe test_and_set_bit routine based on compare and swap (CS)
158  */
159 static inline int
160 test_and_set_bit_cs(unsigned long nr, volatile unsigned long *ptr)
161 {
162         unsigned long addr, old, new, mask;
163 
164 	addr = (unsigned long) ptr;
165 	/* calculate address for CS */
166 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
167 	/* make OR/test mask */
168 	mask = 1UL << (nr & (__BITOPS_WORDSIZE - 1));
169 	/* Do the atomic update. */
170 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_OR);
171 	__BITOPS_BARRIER();
172 	return (old & mask) != 0;
173 }
174 
175 /*
176  * SMP safe test_and_clear_bit routine based on compare and swap (CS)
177  */
178 static inline int
179 test_and_clear_bit_cs(unsigned long nr, volatile unsigned long *ptr)
180 {
181         unsigned long addr, old, new, mask;
182 
183 	addr = (unsigned long) ptr;
184 	/* calculate address for CS */
185 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
186 	/* make AND/test mask */
187 	mask = ~(1UL << (nr & (__BITOPS_WORDSIZE - 1)));
188 	/* Do the atomic update. */
189 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_AND);
190 	__BITOPS_BARRIER();
191 	return (old ^ new) != 0;
192 }
193 
194 /*
195  * SMP safe test_and_change_bit routine based on compare and swap (CS)
196  */
197 static inline int
198 test_and_change_bit_cs(unsigned long nr, volatile unsigned long *ptr)
199 {
200         unsigned long addr, old, new, mask;
201 
202 	addr = (unsigned long) ptr;
203 	/* calculate address for CS */
204 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
205 	/* make XOR/test mask */
206 	mask = 1UL << (nr & (__BITOPS_WORDSIZE - 1));
207 	/* Do the atomic update. */
208 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_XOR);
209 	__BITOPS_BARRIER();
210 	return (old & mask) != 0;
211 }
212 #endif /* CONFIG_SMP */
213 
214 /*
215  * fast, non-SMP set_bit routine
216  */
217 static inline void __set_bit(unsigned long nr, volatile unsigned long *ptr)
218 {
219 	unsigned long addr;
220 
221 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
222 	asm volatile(
223 		"	oc	%O0(1,%R0),%1"
224 		: "=Q" (*(char *) addr) : "Q" (_oi_bitmap[nr & 7]) : "cc" );
225 }
226 
227 static inline void
228 __constant_set_bit(const unsigned long nr, volatile unsigned long *ptr)
229 {
230 	unsigned long addr;
231 
232 	addr = ((unsigned long) ptr) + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
233 	*(unsigned char *) addr |= 1 << (nr & 7);
234 }
235 
236 #define set_bit_simple(nr,addr) \
237 (__builtin_constant_p((nr)) ? \
238  __constant_set_bit((nr),(addr)) : \
239  __set_bit((nr),(addr)) )
240 
241 /*
242  * fast, non-SMP clear_bit routine
243  */
244 static inline void
245 __clear_bit(unsigned long nr, volatile unsigned long *ptr)
246 {
247 	unsigned long addr;
248 
249 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
250 	asm volatile(
251 		"	nc	%O0(1,%R0),%1"
252 		: "=Q" (*(char *) addr) : "Q" (_ni_bitmap[nr & 7]) : "cc" );
253 }
254 
255 static inline void
256 __constant_clear_bit(const unsigned long nr, volatile unsigned long *ptr)
257 {
258 	unsigned long addr;
259 
260 	addr = ((unsigned long) ptr) + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
261 	*(unsigned char *) addr &= ~(1 << (nr & 7));
262 }
263 
264 #define clear_bit_simple(nr,addr) \
265 (__builtin_constant_p((nr)) ? \
266  __constant_clear_bit((nr),(addr)) : \
267  __clear_bit((nr),(addr)) )
268 
269 /*
270  * fast, non-SMP change_bit routine
271  */
272 static inline void __change_bit(unsigned long nr, volatile unsigned long *ptr)
273 {
274 	unsigned long addr;
275 
276 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
277 	asm volatile(
278 		"	xc	%O0(1,%R0),%1"
279 		: "=Q" (*(char *) addr) : "Q" (_oi_bitmap[nr & 7]) : "cc" );
280 }
281 
282 static inline void
283 __constant_change_bit(const unsigned long nr, volatile unsigned long *ptr)
284 {
285 	unsigned long addr;
286 
287 	addr = ((unsigned long) ptr) + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
288 	*(unsigned char *) addr ^= 1 << (nr & 7);
289 }
290 
291 #define change_bit_simple(nr,addr) \
292 (__builtin_constant_p((nr)) ? \
293  __constant_change_bit((nr),(addr)) : \
294  __change_bit((nr),(addr)) )
295 
296 /*
297  * fast, non-SMP test_and_set_bit routine
298  */
299 static inline int
300 test_and_set_bit_simple(unsigned long nr, volatile unsigned long *ptr)
301 {
302 	unsigned long addr;
303 	unsigned char ch;
304 
305 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
306 	ch = *(unsigned char *) addr;
307 	asm volatile(
308 		"	oc	%O0(1,%R0),%1"
309 		: "=Q" (*(char *) addr)	: "Q" (_oi_bitmap[nr & 7])
310 		: "cc", "memory");
311 	return (ch >> (nr & 7)) & 1;
312 }
313 #define __test_and_set_bit(X,Y)		test_and_set_bit_simple(X,Y)
314 
315 /*
316  * fast, non-SMP test_and_clear_bit routine
317  */
318 static inline int
319 test_and_clear_bit_simple(unsigned long nr, volatile unsigned long *ptr)
320 {
321 	unsigned long addr;
322 	unsigned char ch;
323 
324 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
325 	ch = *(unsigned char *) addr;
326 	asm volatile(
327 		"	nc	%O0(1,%R0),%1"
328 		: "=Q" (*(char *) addr)	: "Q" (_ni_bitmap[nr & 7])
329 		: "cc", "memory");
330 	return (ch >> (nr & 7)) & 1;
331 }
332 #define __test_and_clear_bit(X,Y)	test_and_clear_bit_simple(X,Y)
333 
334 /*
335  * fast, non-SMP test_and_change_bit routine
336  */
337 static inline int
338 test_and_change_bit_simple(unsigned long nr, volatile unsigned long *ptr)
339 {
340 	unsigned long addr;
341 	unsigned char ch;
342 
343 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
344 	ch = *(unsigned char *) addr;
345 	asm volatile(
346 		"	xc	%O0(1,%R0),%1"
347 		: "=Q" (*(char *) addr)	: "Q" (_oi_bitmap[nr & 7])
348 		: "cc", "memory");
349 	return (ch >> (nr & 7)) & 1;
350 }
351 #define __test_and_change_bit(X,Y)	test_and_change_bit_simple(X,Y)
352 
353 #ifdef CONFIG_SMP
354 #define set_bit             set_bit_cs
355 #define clear_bit           clear_bit_cs
356 #define change_bit          change_bit_cs
357 #define test_and_set_bit    test_and_set_bit_cs
358 #define test_and_clear_bit  test_and_clear_bit_cs
359 #define test_and_change_bit test_and_change_bit_cs
360 #else
361 #define set_bit             set_bit_simple
362 #define clear_bit           clear_bit_simple
363 #define change_bit          change_bit_simple
364 #define test_and_set_bit    test_and_set_bit_simple
365 #define test_and_clear_bit  test_and_clear_bit_simple
366 #define test_and_change_bit test_and_change_bit_simple
367 #endif
368 
369 
370 /*
371  * This routine doesn't need to be atomic.
372  */
373 
374 static inline int __test_bit(unsigned long nr, const volatile unsigned long *ptr)
375 {
376 	unsigned long addr;
377 	unsigned char ch;
378 
379 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
380 	ch = *(volatile unsigned char *) addr;
381 	return (ch >> (nr & 7)) & 1;
382 }
383 
384 static inline int
385 __constant_test_bit(unsigned long nr, const volatile unsigned long *addr) {
386     return (((volatile char *) addr)
387 	    [(nr^(__BITOPS_WORDSIZE-8))>>3] & (1<<(nr&7))) != 0;
388 }
389 
390 #define test_bit(nr,addr) \
391 (__builtin_constant_p((nr)) ? \
392  __constant_test_bit((nr),(addr)) : \
393  __test_bit((nr),(addr)) )
394 
395 /*
396  * Optimized find bit helper functions.
397  */
398 
399 /**
400  * __ffz_word_loop - find byte offset of first long != -1UL
401  * @addr: pointer to array of unsigned long
402  * @size: size of the array in bits
403  */
404 static inline unsigned long __ffz_word_loop(const unsigned long *addr,
405 					    unsigned long size)
406 {
407 	typedef struct { long _[__BITOPS_WORDS(size)]; } addrtype;
408 	unsigned long bytes = 0;
409 
410 	asm volatile(
411 #ifndef CONFIG_64BIT
412 		"	ahi	%1,-1\n"
413 		"	sra	%1,5\n"
414 		"	jz	1f\n"
415 		"0:	c	%2,0(%0,%3)\n"
416 		"	jne	1f\n"
417 		"	la	%0,4(%0)\n"
418 		"	brct	%1,0b\n"
419 		"1:\n"
420 #else
421 		"	aghi	%1,-1\n"
422 		"	srag	%1,%1,6\n"
423 		"	jz	1f\n"
424 		"0:	cg	%2,0(%0,%3)\n"
425 		"	jne	1f\n"
426 		"	la	%0,8(%0)\n"
427 		"	brct	%1,0b\n"
428 		"1:\n"
429 #endif
430 		: "+&a" (bytes), "+&d" (size)
431 		: "d" (-1UL), "a" (addr), "m" (*(addrtype *) addr)
432 		: "cc" );
433 	return bytes;
434 }
435 
436 /**
437  * __ffs_word_loop - find byte offset of first long != 0UL
438  * @addr: pointer to array of unsigned long
439  * @size: size of the array in bits
440  */
441 static inline unsigned long __ffs_word_loop(const unsigned long *addr,
442 					    unsigned long size)
443 {
444 	typedef struct { long _[__BITOPS_WORDS(size)]; } addrtype;
445 	unsigned long bytes = 0;
446 
447 	asm volatile(
448 #ifndef CONFIG_64BIT
449 		"	ahi	%1,-1\n"
450 		"	sra	%1,5\n"
451 		"	jz	1f\n"
452 		"0:	c	%2,0(%0,%3)\n"
453 		"	jne	1f\n"
454 		"	la	%0,4(%0)\n"
455 		"	brct	%1,0b\n"
456 		"1:\n"
457 #else
458 		"	aghi	%1,-1\n"
459 		"	srag	%1,%1,6\n"
460 		"	jz	1f\n"
461 		"0:	cg	%2,0(%0,%3)\n"
462 		"	jne	1f\n"
463 		"	la	%0,8(%0)\n"
464 		"	brct	%1,0b\n"
465 		"1:\n"
466 #endif
467 		: "+&a" (bytes), "+&a" (size)
468 		: "d" (0UL), "a" (addr), "m" (*(addrtype *) addr)
469 		: "cc" );
470 	return bytes;
471 }
472 
473 /**
474  * __ffz_word - add number of the first unset bit
475  * @nr: base value the bit number is added to
476  * @word: the word that is searched for unset bits
477  */
478 static inline unsigned long __ffz_word(unsigned long nr, unsigned long word)
479 {
480 #ifdef CONFIG_64BIT
481 	if ((word & 0xffffffff) == 0xffffffff) {
482 		word >>= 32;
483 		nr += 32;
484 	}
485 #endif
486 	if ((word & 0xffff) == 0xffff) {
487 		word >>= 16;
488 		nr += 16;
489 	}
490 	if ((word & 0xff) == 0xff) {
491 		word >>= 8;
492 		nr += 8;
493 	}
494 	return nr + _zb_findmap[(unsigned char) word];
495 }
496 
497 /**
498  * __ffs_word - add number of the first set bit
499  * @nr: base value the bit number is added to
500  * @word: the word that is searched for set bits
501  */
502 static inline unsigned long __ffs_word(unsigned long nr, unsigned long word)
503 {
504 #ifdef CONFIG_64BIT
505 	if ((word & 0xffffffff) == 0) {
506 		word >>= 32;
507 		nr += 32;
508 	}
509 #endif
510 	if ((word & 0xffff) == 0) {
511 		word >>= 16;
512 		nr += 16;
513 	}
514 	if ((word & 0xff) == 0) {
515 		word >>= 8;
516 		nr += 8;
517 	}
518 	return nr + _sb_findmap[(unsigned char) word];
519 }
520 
521 
522 /**
523  * __load_ulong_be - load big endian unsigned long
524  * @p: pointer to array of unsigned long
525  * @offset: byte offset of source value in the array
526  */
527 static inline unsigned long __load_ulong_be(const unsigned long *p,
528 					    unsigned long offset)
529 {
530 	p = (unsigned long *)((unsigned long) p + offset);
531 	return *p;
532 }
533 
534 /**
535  * __load_ulong_le - load little endian unsigned long
536  * @p: pointer to array of unsigned long
537  * @offset: byte offset of source value in the array
538  */
539 static inline unsigned long __load_ulong_le(const unsigned long *p,
540 					    unsigned long offset)
541 {
542 	unsigned long word;
543 
544 	p = (unsigned long *)((unsigned long) p + offset);
545 #ifndef CONFIG_64BIT
546 	asm volatile(
547 		"	ic	%0,%O1(%R1)\n"
548 		"	icm	%0,2,%O1+1(%R1)\n"
549 		"	icm	%0,4,%O1+2(%R1)\n"
550 		"	icm	%0,8,%O1+3(%R1)"
551 		: "=&d" (word) : "Q" (*p) : "cc");
552 #else
553 	asm volatile(
554 		"	lrvg	%0,%1"
555 		: "=d" (word) : "m" (*p) );
556 #endif
557 	return word;
558 }
559 
560 /*
561  * The various find bit functions.
562  */
563 
564 /*
565  * ffz - find first zero in word.
566  * @word: The word to search
567  *
568  * Undefined if no zero exists, so code should check against ~0UL first.
569  */
570 static inline unsigned long ffz(unsigned long word)
571 {
572 	return __ffz_word(0, word);
573 }
574 
575 /**
576  * __ffs - find first bit in word.
577  * @word: The word to search
578  *
579  * Undefined if no bit exists, so code should check against 0 first.
580  */
581 static inline unsigned long __ffs (unsigned long word)
582 {
583 	return __ffs_word(0, word);
584 }
585 
586 /**
587  * ffs - find first bit set
588  * @x: the word to search
589  *
590  * This is defined the same way as
591  * the libc and compiler builtin ffs routines, therefore
592  * differs in spirit from the above ffz (man ffs).
593  */
594 static inline int ffs(int x)
595 {
596 	if (!x)
597 		return 0;
598 	return __ffs_word(1, x);
599 }
600 
601 /**
602  * find_first_zero_bit - find the first zero bit in a memory region
603  * @addr: The address to start the search at
604  * @size: The maximum size to search
605  *
606  * Returns the bit-number of the first zero bit, not the number of the byte
607  * containing a bit.
608  */
609 static inline unsigned long find_first_zero_bit(const unsigned long *addr,
610 						unsigned long size)
611 {
612 	unsigned long bytes, bits;
613 
614         if (!size)
615                 return 0;
616 	bytes = __ffz_word_loop(addr, size);
617 	bits = __ffz_word(bytes*8, __load_ulong_be(addr, bytes));
618 	return (bits < size) ? bits : size;
619 }
620 #define find_first_zero_bit find_first_zero_bit
621 
622 /**
623  * find_first_bit - find the first set bit in a memory region
624  * @addr: The address to start the search at
625  * @size: The maximum size to search
626  *
627  * Returns the bit-number of the first set bit, not the number of the byte
628  * containing a bit.
629  */
630 static inline unsigned long find_first_bit(const unsigned long * addr,
631 					   unsigned long size)
632 {
633 	unsigned long bytes, bits;
634 
635         if (!size)
636                 return 0;
637 	bytes = __ffs_word_loop(addr, size);
638 	bits = __ffs_word(bytes*8, __load_ulong_be(addr, bytes));
639 	return (bits < size) ? bits : size;
640 }
641 #define find_first_bit find_first_bit
642 
643 /*
644  * Big endian variant whichs starts bit counting from left using
645  * the flogr (find leftmost one) instruction.
646  */
647 static inline unsigned long __flo_word(unsigned long nr, unsigned long val)
648 {
649 	register unsigned long bit asm("2") = val;
650 	register unsigned long out asm("3");
651 
652 	asm volatile (
653 		"	.insn	rre,0xb9830000,%[bit],%[bit]\n"
654 		: [bit] "+d" (bit), [out] "=d" (out) : : "cc");
655 	return nr + bit;
656 }
657 
658 /*
659  * 64 bit special left bitops format:
660  * order in memory:
661  *    00 01 02 03 04 05 06 07 08 09 0a 0b 0c 0d 0e 0f
662  *    10 11 12 13 14 15 16 17 18 19 1a 1b 1c 1d 1e 1f
663  *    20 21 22 23 24 25 26 27 28 29 2a 2b 2c 2d 2e 2f
664  *    30 31 32 33 34 35 36 37 38 39 3a 3b 3c 3d 3e 3f
665  * after that follows the next long with bit numbers
666  *    40 41 42 43 44 45 46 47 48 49 4a 4b 4c 4d 4e 4f
667  *    50 51 52 53 54 55 56 57 58 59 5a 5b 5c 5d 5e 5f
668  *    60 61 62 63 64 65 66 67 68 69 6a 6b 6c 6d 6e 6f
669  *    70 71 72 73 74 75 76 77 78 79 7a 7b 7c 7d 7e 7f
670  * The reason for this bit ordering is the fact that
671  * the hardware sets bits in a bitmap starting at bit 0
672  * and we don't want to scan the bitmap from the 'wrong
673  * end'.
674  */
675 static inline unsigned long find_first_bit_left(const unsigned long *addr,
676 						unsigned long size)
677 {
678 	unsigned long bytes, bits;
679 
680 	if (!size)
681 		return 0;
682 	bytes = __ffs_word_loop(addr, size);
683 	bits = __flo_word(bytes * 8, __load_ulong_be(addr, bytes));
684 	return (bits < size) ? bits : size;
685 }
686 
687 static inline int find_next_bit_left(const unsigned long *addr,
688 				     unsigned long size,
689 				     unsigned long offset)
690 {
691 	const unsigned long *p;
692 	unsigned long bit, set;
693 
694 	if (offset >= size)
695 		return size;
696 	bit = offset & (__BITOPS_WORDSIZE - 1);
697 	offset -= bit;
698 	size -= offset;
699 	p = addr + offset / __BITOPS_WORDSIZE;
700 	if (bit) {
701 		set = __flo_word(0, *p & (~0UL << bit));
702 		if (set >= size)
703 			return size + offset;
704 		if (set < __BITOPS_WORDSIZE)
705 			return set + offset;
706 		offset += __BITOPS_WORDSIZE;
707 		size -= __BITOPS_WORDSIZE;
708 		p++;
709 	}
710 	return offset + find_first_bit_left(p, size);
711 }
712 
713 #define for_each_set_bit_left(bit, addr, size)				\
714 	for ((bit) = find_first_bit_left((addr), (size));		\
715 	     (bit) < (size);						\
716 	     (bit) = find_next_bit_left((addr), (size), (bit) + 1))
717 
718 /* same as for_each_set_bit() but use bit as value to start with */
719 #define for_each_set_bit_left_cont(bit, addr, size)			\
720 	for ((bit) = find_next_bit_left((addr), (size), (bit));		\
721 	     (bit) < (size);						\
722 	     (bit) = find_next_bit_left((addr), (size), (bit) + 1))
723 
724 /**
725  * find_next_zero_bit - find the first zero bit in a memory region
726  * @addr: The address to base the search on
727  * @offset: The bitnumber to start searching at
728  * @size: The maximum size to search
729  */
730 static inline int find_next_zero_bit (const unsigned long * addr,
731 				      unsigned long size,
732 				      unsigned long offset)
733 {
734         const unsigned long *p;
735 	unsigned long bit, set;
736 
737 	if (offset >= size)
738 		return size;
739 	bit = offset & (__BITOPS_WORDSIZE - 1);
740 	offset -= bit;
741 	size -= offset;
742 	p = addr + offset / __BITOPS_WORDSIZE;
743 	if (bit) {
744 		/*
745 		 * __ffz_word returns __BITOPS_WORDSIZE
746 		 * if no zero bit is present in the word.
747 		 */
748 		set = __ffz_word(bit, *p >> bit);
749 		if (set >= size)
750 			return size + offset;
751 		if (set < __BITOPS_WORDSIZE)
752 			return set + offset;
753 		offset += __BITOPS_WORDSIZE;
754 		size -= __BITOPS_WORDSIZE;
755 		p++;
756 	}
757 	return offset + find_first_zero_bit(p, size);
758 }
759 #define find_next_zero_bit find_next_zero_bit
760 
761 /**
762  * find_next_bit - find the first set bit in a memory region
763  * @addr: The address to base the search on
764  * @offset: The bitnumber to start searching at
765  * @size: The maximum size to search
766  */
767 static inline int find_next_bit (const unsigned long * addr,
768 				 unsigned long size,
769 				 unsigned long offset)
770 {
771         const unsigned long *p;
772 	unsigned long bit, set;
773 
774 	if (offset >= size)
775 		return size;
776 	bit = offset & (__BITOPS_WORDSIZE - 1);
777 	offset -= bit;
778 	size -= offset;
779 	p = addr + offset / __BITOPS_WORDSIZE;
780 	if (bit) {
781 		/*
782 		 * __ffs_word returns __BITOPS_WORDSIZE
783 		 * if no one bit is present in the word.
784 		 */
785 		set = __ffs_word(0, *p & (~0UL << bit));
786 		if (set >= size)
787 			return size + offset;
788 		if (set < __BITOPS_WORDSIZE)
789 			return set + offset;
790 		offset += __BITOPS_WORDSIZE;
791 		size -= __BITOPS_WORDSIZE;
792 		p++;
793 	}
794 	return offset + find_first_bit(p, size);
795 }
796 #define find_next_bit find_next_bit
797 
798 /*
799  * Every architecture must define this function. It's the fastest
800  * way of searching a 140-bit bitmap where the first 100 bits are
801  * unlikely to be set. It's guaranteed that at least one of the 140
802  * bits is cleared.
803  */
804 static inline int sched_find_first_bit(unsigned long *b)
805 {
806 	return find_first_bit(b, 140);
807 }
808 
809 #include <asm-generic/bitops/fls.h>
810 #include <asm-generic/bitops/__fls.h>
811 #include <asm-generic/bitops/fls64.h>
812 
813 #include <asm-generic/bitops/hweight.h>
814 #include <asm-generic/bitops/lock.h>
815 
816 /*
817  * ATTENTION: intel byte ordering convention for ext2 and minix !!
818  * bit 0 is the LSB of addr; bit 31 is the MSB of addr;
819  * bit 32 is the LSB of (addr+4).
820  * That combined with the little endian byte order of Intel gives the
821  * following bit order in memory:
822  *    07 06 05 04 03 02 01 00 15 14 13 12 11 10 09 08 \
823  *    23 22 21 20 19 18 17 16 31 30 29 28 27 26 25 24
824  */
825 
826 static inline int find_first_zero_bit_le(void *vaddr, unsigned int size)
827 {
828 	unsigned long bytes, bits;
829 
830         if (!size)
831                 return 0;
832 	bytes = __ffz_word_loop(vaddr, size);
833 	bits = __ffz_word(bytes*8, __load_ulong_le(vaddr, bytes));
834 	return (bits < size) ? bits : size;
835 }
836 #define find_first_zero_bit_le find_first_zero_bit_le
837 
838 static inline int find_next_zero_bit_le(void *vaddr, unsigned long size,
839 					  unsigned long offset)
840 {
841         unsigned long *addr = vaddr, *p;
842 	unsigned long bit, set;
843 
844         if (offset >= size)
845                 return size;
846 	bit = offset & (__BITOPS_WORDSIZE - 1);
847 	offset -= bit;
848 	size -= offset;
849 	p = addr + offset / __BITOPS_WORDSIZE;
850         if (bit) {
851 		/*
852 		 * s390 version of ffz returns __BITOPS_WORDSIZE
853 		 * if no zero bit is present in the word.
854 		 */
855 		set = __ffz_word(bit, __load_ulong_le(p, 0) >> bit);
856 		if (set >= size)
857 			return size + offset;
858 		if (set < __BITOPS_WORDSIZE)
859 			return set + offset;
860 		offset += __BITOPS_WORDSIZE;
861 		size -= __BITOPS_WORDSIZE;
862 		p++;
863         }
864 	return offset + find_first_zero_bit_le(p, size);
865 }
866 #define find_next_zero_bit_le find_next_zero_bit_le
867 
868 static inline unsigned long find_first_bit_le(void *vaddr, unsigned long size)
869 {
870 	unsigned long bytes, bits;
871 
872 	if (!size)
873 		return 0;
874 	bytes = __ffs_word_loop(vaddr, size);
875 	bits = __ffs_word(bytes*8, __load_ulong_le(vaddr, bytes));
876 	return (bits < size) ? bits : size;
877 }
878 #define find_first_bit_le find_first_bit_le
879 
880 static inline int find_next_bit_le(void *vaddr, unsigned long size,
881 				     unsigned long offset)
882 {
883 	unsigned long *addr = vaddr, *p;
884 	unsigned long bit, set;
885 
886 	if (offset >= size)
887 		return size;
888 	bit = offset & (__BITOPS_WORDSIZE - 1);
889 	offset -= bit;
890 	size -= offset;
891 	p = addr + offset / __BITOPS_WORDSIZE;
892 	if (bit) {
893 		/*
894 		 * s390 version of ffz returns __BITOPS_WORDSIZE
895 		 * if no zero bit is present in the word.
896 		 */
897 		set = __ffs_word(0, __load_ulong_le(p, 0) & (~0UL << bit));
898 		if (set >= size)
899 			return size + offset;
900 		if (set < __BITOPS_WORDSIZE)
901 			return set + offset;
902 		offset += __BITOPS_WORDSIZE;
903 		size -= __BITOPS_WORDSIZE;
904 		p++;
905 	}
906 	return offset + find_first_bit_le(p, size);
907 }
908 #define find_next_bit_le find_next_bit_le
909 
910 #include <asm-generic/bitops/le.h>
911 
912 #include <asm-generic/bitops/ext2-atomic-setbit.h>
913 
914 #endif /* _S390_BITOPS_H */
915