xref: /openbmc/linux/arch/s390/include/asm/bitops.h (revision 261a9af6)
1 #ifndef _S390_BITOPS_H
2 #define _S390_BITOPS_H
3 
4 /*
5  *  include/asm-s390/bitops.h
6  *
7  *  S390 version
8  *    Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
9  *    Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
10  *
11  *  Derived from "include/asm-i386/bitops.h"
12  *    Copyright (C) 1992, Linus Torvalds
13  *
14  */
15 
16 #ifdef __KERNEL__
17 
18 #ifndef _LINUX_BITOPS_H
19 #error only <linux/bitops.h> can be included directly
20 #endif
21 
22 #include <linux/compiler.h>
23 
24 /*
25  * 32 bit bitops format:
26  * bit 0 is the LSB of *addr; bit 31 is the MSB of *addr;
27  * bit 32 is the LSB of *(addr+4). That combined with the
28  * big endian byte order on S390 give the following bit
29  * order in memory:
30  *    1f 1e 1d 1c 1b 1a 19 18 17 16 15 14 13 12 11 10 \
31  *    0f 0e 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00
32  * after that follows the next long with bit numbers
33  *    3f 3e 3d 3c 3b 3a 39 38 37 36 35 34 33 32 31 30
34  *    2f 2e 2d 2c 2b 2a 29 28 27 26 25 24 23 22 21 20
35  * The reason for this bit ordering is the fact that
36  * in the architecture independent code bits operations
37  * of the form "flags |= (1 << bitnr)" are used INTERMIXED
38  * with operation of the form "set_bit(bitnr, flags)".
39  *
40  * 64 bit bitops format:
41  * bit 0 is the LSB of *addr; bit 63 is the MSB of *addr;
42  * bit 64 is the LSB of *(addr+8). That combined with the
43  * big endian byte order on S390 give the following bit
44  * order in memory:
45  *    3f 3e 3d 3c 3b 3a 39 38 37 36 35 34 33 32 31 30
46  *    2f 2e 2d 2c 2b 2a 29 28 27 26 25 24 23 22 21 20
47  *    1f 1e 1d 1c 1b 1a 19 18 17 16 15 14 13 12 11 10
48  *    0f 0e 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00
49  * after that follows the next long with bit numbers
50  *    7f 7e 7d 7c 7b 7a 79 78 77 76 75 74 73 72 71 70
51  *    6f 6e 6d 6c 6b 6a 69 68 67 66 65 64 63 62 61 60
52  *    5f 5e 5d 5c 5b 5a 59 58 57 56 55 54 53 52 51 50
53  *    4f 4e 4d 4c 4b 4a 49 48 47 46 45 44 43 42 41 40
54  * The reason for this bit ordering is the fact that
55  * in the architecture independent code bits operations
56  * of the form "flags |= (1 << bitnr)" are used INTERMIXED
57  * with operation of the form "set_bit(bitnr, flags)".
58  */
59 
60 /* bitmap tables from arch/s390/kernel/bitmap.c */
61 extern const char _oi_bitmap[];
62 extern const char _ni_bitmap[];
63 extern const char _zb_findmap[];
64 extern const char _sb_findmap[];
65 
66 #ifndef __s390x__
67 
68 #define __BITOPS_ALIGN		3
69 #define __BITOPS_WORDSIZE	32
70 #define __BITOPS_OR		"or"
71 #define __BITOPS_AND		"nr"
72 #define __BITOPS_XOR		"xr"
73 
74 #define __BITOPS_LOOP(__old, __new, __addr, __val, __op_string)	\
75 	asm volatile(						\
76 		"	l	%0,%2\n"			\
77 		"0:	lr	%1,%0\n"			\
78 		__op_string "	%1,%3\n"			\
79 		"	cs	%0,%1,%2\n"			\
80 		"	jl	0b"				\
81 		: "=&d" (__old), "=&d" (__new),			\
82 		  "=Q" (*(unsigned long *) __addr)		\
83 		: "d" (__val), "Q" (*(unsigned long *) __addr)	\
84 		: "cc");
85 
86 #else /* __s390x__ */
87 
88 #define __BITOPS_ALIGN		7
89 #define __BITOPS_WORDSIZE	64
90 #define __BITOPS_OR		"ogr"
91 #define __BITOPS_AND		"ngr"
92 #define __BITOPS_XOR		"xgr"
93 
94 #define __BITOPS_LOOP(__old, __new, __addr, __val, __op_string)	\
95 	asm volatile(						\
96 		"	lg	%0,%2\n"			\
97 		"0:	lgr	%1,%0\n"			\
98 		__op_string "	%1,%3\n"			\
99 		"	csg	%0,%1,%2\n"			\
100 		"	jl	0b"				\
101 		: "=&d" (__old), "=&d" (__new),			\
102 		  "=Q" (*(unsigned long *) __addr)		\
103 		: "d" (__val), "Q" (*(unsigned long *) __addr)	\
104 		: "cc");
105 
106 #endif /* __s390x__ */
107 
108 #define __BITOPS_WORDS(bits) (((bits)+__BITOPS_WORDSIZE-1)/__BITOPS_WORDSIZE)
109 #define __BITOPS_BARRIER() asm volatile("" : : : "memory")
110 
111 #ifdef CONFIG_SMP
112 /*
113  * SMP safe set_bit routine based on compare and swap (CS)
114  */
115 static inline void set_bit_cs(unsigned long nr, volatile unsigned long *ptr)
116 {
117         unsigned long addr, old, new, mask;
118 
119 	addr = (unsigned long) ptr;
120 	/* calculate address for CS */
121 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
122 	/* make OR mask */
123 	mask = 1UL << (nr & (__BITOPS_WORDSIZE - 1));
124 	/* Do the atomic update. */
125 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_OR);
126 }
127 
128 /*
129  * SMP safe clear_bit routine based on compare and swap (CS)
130  */
131 static inline void clear_bit_cs(unsigned long nr, volatile unsigned long *ptr)
132 {
133         unsigned long addr, old, new, mask;
134 
135 	addr = (unsigned long) ptr;
136 	/* calculate address for CS */
137 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
138 	/* make AND mask */
139 	mask = ~(1UL << (nr & (__BITOPS_WORDSIZE - 1)));
140 	/* Do the atomic update. */
141 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_AND);
142 }
143 
144 /*
145  * SMP safe change_bit routine based on compare and swap (CS)
146  */
147 static inline void change_bit_cs(unsigned long nr, volatile unsigned long *ptr)
148 {
149         unsigned long addr, old, new, mask;
150 
151 	addr = (unsigned long) ptr;
152 	/* calculate address for CS */
153 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
154 	/* make XOR mask */
155 	mask = 1UL << (nr & (__BITOPS_WORDSIZE - 1));
156 	/* Do the atomic update. */
157 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_XOR);
158 }
159 
160 /*
161  * SMP safe test_and_set_bit routine based on compare and swap (CS)
162  */
163 static inline int
164 test_and_set_bit_cs(unsigned long nr, volatile unsigned long *ptr)
165 {
166         unsigned long addr, old, new, mask;
167 
168 	addr = (unsigned long) ptr;
169 	/* calculate address for CS */
170 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
171 	/* make OR/test mask */
172 	mask = 1UL << (nr & (__BITOPS_WORDSIZE - 1));
173 	/* Do the atomic update. */
174 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_OR);
175 	__BITOPS_BARRIER();
176 	return (old & mask) != 0;
177 }
178 
179 /*
180  * SMP safe test_and_clear_bit routine based on compare and swap (CS)
181  */
182 static inline int
183 test_and_clear_bit_cs(unsigned long nr, volatile unsigned long *ptr)
184 {
185         unsigned long addr, old, new, mask;
186 
187 	addr = (unsigned long) ptr;
188 	/* calculate address for CS */
189 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
190 	/* make AND/test mask */
191 	mask = ~(1UL << (nr & (__BITOPS_WORDSIZE - 1)));
192 	/* Do the atomic update. */
193 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_AND);
194 	__BITOPS_BARRIER();
195 	return (old ^ new) != 0;
196 }
197 
198 /*
199  * SMP safe test_and_change_bit routine based on compare and swap (CS)
200  */
201 static inline int
202 test_and_change_bit_cs(unsigned long nr, volatile unsigned long *ptr)
203 {
204         unsigned long addr, old, new, mask;
205 
206 	addr = (unsigned long) ptr;
207 	/* calculate address for CS */
208 	addr += (nr ^ (nr & (__BITOPS_WORDSIZE - 1))) >> 3;
209 	/* make XOR/test mask */
210 	mask = 1UL << (nr & (__BITOPS_WORDSIZE - 1));
211 	/* Do the atomic update. */
212 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_XOR);
213 	__BITOPS_BARRIER();
214 	return (old & mask) != 0;
215 }
216 #endif /* CONFIG_SMP */
217 
218 /*
219  * fast, non-SMP set_bit routine
220  */
221 static inline void __set_bit(unsigned long nr, volatile unsigned long *ptr)
222 {
223 	unsigned long addr;
224 
225 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
226 	asm volatile(
227 		"	oc	%O0(1,%R0),%1"
228 		: "=Q" (*(char *) addr) : "Q" (_oi_bitmap[nr & 7]) : "cc" );
229 }
230 
231 static inline void
232 __constant_set_bit(const unsigned long nr, volatile unsigned long *ptr)
233 {
234 	unsigned long addr;
235 
236 	addr = ((unsigned long) ptr) + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
237 	*(unsigned char *) addr |= 1 << (nr & 7);
238 }
239 
240 #define set_bit_simple(nr,addr) \
241 (__builtin_constant_p((nr)) ? \
242  __constant_set_bit((nr),(addr)) : \
243  __set_bit((nr),(addr)) )
244 
245 /*
246  * fast, non-SMP clear_bit routine
247  */
248 static inline void
249 __clear_bit(unsigned long nr, volatile unsigned long *ptr)
250 {
251 	unsigned long addr;
252 
253 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
254 	asm volatile(
255 		"	nc	%O0(1,%R0),%1"
256 		: "=Q" (*(char *) addr) : "Q" (_ni_bitmap[nr & 7]) : "cc" );
257 }
258 
259 static inline void
260 __constant_clear_bit(const unsigned long nr, volatile unsigned long *ptr)
261 {
262 	unsigned long addr;
263 
264 	addr = ((unsigned long) ptr) + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
265 	*(unsigned char *) addr &= ~(1 << (nr & 7));
266 }
267 
268 #define clear_bit_simple(nr,addr) \
269 (__builtin_constant_p((nr)) ? \
270  __constant_clear_bit((nr),(addr)) : \
271  __clear_bit((nr),(addr)) )
272 
273 /*
274  * fast, non-SMP change_bit routine
275  */
276 static inline void __change_bit(unsigned long nr, volatile unsigned long *ptr)
277 {
278 	unsigned long addr;
279 
280 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
281 	asm volatile(
282 		"	xc	%O0(1,%R0),%1"
283 		: "=Q" (*(char *) addr) : "Q" (_oi_bitmap[nr & 7]) : "cc" );
284 }
285 
286 static inline void
287 __constant_change_bit(const unsigned long nr, volatile unsigned long *ptr)
288 {
289 	unsigned long addr;
290 
291 	addr = ((unsigned long) ptr) + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
292 	*(unsigned char *) addr ^= 1 << (nr & 7);
293 }
294 
295 #define change_bit_simple(nr,addr) \
296 (__builtin_constant_p((nr)) ? \
297  __constant_change_bit((nr),(addr)) : \
298  __change_bit((nr),(addr)) )
299 
300 /*
301  * fast, non-SMP test_and_set_bit routine
302  */
303 static inline int
304 test_and_set_bit_simple(unsigned long nr, volatile unsigned long *ptr)
305 {
306 	unsigned long addr;
307 	unsigned char ch;
308 
309 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
310 	ch = *(unsigned char *) addr;
311 	asm volatile(
312 		"	oc	%O0(1,%R0),%1"
313 		: "=Q" (*(char *) addr)	: "Q" (_oi_bitmap[nr & 7])
314 		: "cc", "memory");
315 	return (ch >> (nr & 7)) & 1;
316 }
317 #define __test_and_set_bit(X,Y)		test_and_set_bit_simple(X,Y)
318 
319 /*
320  * fast, non-SMP test_and_clear_bit routine
321  */
322 static inline int
323 test_and_clear_bit_simple(unsigned long nr, volatile unsigned long *ptr)
324 {
325 	unsigned long addr;
326 	unsigned char ch;
327 
328 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
329 	ch = *(unsigned char *) addr;
330 	asm volatile(
331 		"	nc	%O0(1,%R0),%1"
332 		: "=Q" (*(char *) addr)	: "Q" (_ni_bitmap[nr & 7])
333 		: "cc", "memory");
334 	return (ch >> (nr & 7)) & 1;
335 }
336 #define __test_and_clear_bit(X,Y)	test_and_clear_bit_simple(X,Y)
337 
338 /*
339  * fast, non-SMP test_and_change_bit routine
340  */
341 static inline int
342 test_and_change_bit_simple(unsigned long nr, volatile unsigned long *ptr)
343 {
344 	unsigned long addr;
345 	unsigned char ch;
346 
347 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
348 	ch = *(unsigned char *) addr;
349 	asm volatile(
350 		"	xc	%O0(1,%R0),%1"
351 		: "=Q" (*(char *) addr)	: "Q" (_oi_bitmap[nr & 7])
352 		: "cc", "memory");
353 	return (ch >> (nr & 7)) & 1;
354 }
355 #define __test_and_change_bit(X,Y)	test_and_change_bit_simple(X,Y)
356 
357 #ifdef CONFIG_SMP
358 #define set_bit             set_bit_cs
359 #define clear_bit           clear_bit_cs
360 #define change_bit          change_bit_cs
361 #define test_and_set_bit    test_and_set_bit_cs
362 #define test_and_clear_bit  test_and_clear_bit_cs
363 #define test_and_change_bit test_and_change_bit_cs
364 #else
365 #define set_bit             set_bit_simple
366 #define clear_bit           clear_bit_simple
367 #define change_bit          change_bit_simple
368 #define test_and_set_bit    test_and_set_bit_simple
369 #define test_and_clear_bit  test_and_clear_bit_simple
370 #define test_and_change_bit test_and_change_bit_simple
371 #endif
372 
373 
374 /*
375  * This routine doesn't need to be atomic.
376  */
377 
378 static inline int __test_bit(unsigned long nr, const volatile unsigned long *ptr)
379 {
380 	unsigned long addr;
381 	unsigned char ch;
382 
383 	addr = (unsigned long) ptr + ((nr ^ (__BITOPS_WORDSIZE - 8)) >> 3);
384 	ch = *(volatile unsigned char *) addr;
385 	return (ch >> (nr & 7)) & 1;
386 }
387 
388 static inline int
389 __constant_test_bit(unsigned long nr, const volatile unsigned long *addr) {
390     return (((volatile char *) addr)
391 	    [(nr^(__BITOPS_WORDSIZE-8))>>3] & (1<<(nr&7))) != 0;
392 }
393 
394 #define test_bit(nr,addr) \
395 (__builtin_constant_p((nr)) ? \
396  __constant_test_bit((nr),(addr)) : \
397  __test_bit((nr),(addr)) )
398 
399 /*
400  * Optimized find bit helper functions.
401  */
402 
403 /**
404  * __ffz_word_loop - find byte offset of first long != -1UL
405  * @addr: pointer to array of unsigned long
406  * @size: size of the array in bits
407  */
408 static inline unsigned long __ffz_word_loop(const unsigned long *addr,
409 					    unsigned long size)
410 {
411 	typedef struct { long _[__BITOPS_WORDS(size)]; } addrtype;
412 	unsigned long bytes = 0;
413 
414 	asm volatile(
415 #ifndef __s390x__
416 		"	ahi	%1,-1\n"
417 		"	sra	%1,5\n"
418 		"	jz	1f\n"
419 		"0:	c	%2,0(%0,%3)\n"
420 		"	jne	1f\n"
421 		"	la	%0,4(%0)\n"
422 		"	brct	%1,0b\n"
423 		"1:\n"
424 #else
425 		"	aghi	%1,-1\n"
426 		"	srag	%1,%1,6\n"
427 		"	jz	1f\n"
428 		"0:	cg	%2,0(%0,%3)\n"
429 		"	jne	1f\n"
430 		"	la	%0,8(%0)\n"
431 		"	brct	%1,0b\n"
432 		"1:\n"
433 #endif
434 		: "+&a" (bytes), "+&d" (size)
435 		: "d" (-1UL), "a" (addr), "m" (*(addrtype *) addr)
436 		: "cc" );
437 	return bytes;
438 }
439 
440 /**
441  * __ffs_word_loop - find byte offset of first long != 0UL
442  * @addr: pointer to array of unsigned long
443  * @size: size of the array in bits
444  */
445 static inline unsigned long __ffs_word_loop(const unsigned long *addr,
446 					    unsigned long size)
447 {
448 	typedef struct { long _[__BITOPS_WORDS(size)]; } addrtype;
449 	unsigned long bytes = 0;
450 
451 	asm volatile(
452 #ifndef __s390x__
453 		"	ahi	%1,-1\n"
454 		"	sra	%1,5\n"
455 		"	jz	1f\n"
456 		"0:	c	%2,0(%0,%3)\n"
457 		"	jne	1f\n"
458 		"	la	%0,4(%0)\n"
459 		"	brct	%1,0b\n"
460 		"1:\n"
461 #else
462 		"	aghi	%1,-1\n"
463 		"	srag	%1,%1,6\n"
464 		"	jz	1f\n"
465 		"0:	cg	%2,0(%0,%3)\n"
466 		"	jne	1f\n"
467 		"	la	%0,8(%0)\n"
468 		"	brct	%1,0b\n"
469 		"1:\n"
470 #endif
471 		: "+&a" (bytes), "+&a" (size)
472 		: "d" (0UL), "a" (addr), "m" (*(addrtype *) addr)
473 		: "cc" );
474 	return bytes;
475 }
476 
477 /**
478  * __ffz_word - add number of the first unset bit
479  * @nr: base value the bit number is added to
480  * @word: the word that is searched for unset bits
481  */
482 static inline unsigned long __ffz_word(unsigned long nr, unsigned long word)
483 {
484 #ifdef __s390x__
485 	if ((word & 0xffffffff) == 0xffffffff) {
486 		word >>= 32;
487 		nr += 32;
488 	}
489 #endif
490 	if ((word & 0xffff) == 0xffff) {
491 		word >>= 16;
492 		nr += 16;
493 	}
494 	if ((word & 0xff) == 0xff) {
495 		word >>= 8;
496 		nr += 8;
497 	}
498 	return nr + _zb_findmap[(unsigned char) word];
499 }
500 
501 /**
502  * __ffs_word - add number of the first set bit
503  * @nr: base value the bit number is added to
504  * @word: the word that is searched for set bits
505  */
506 static inline unsigned long __ffs_word(unsigned long nr, unsigned long word)
507 {
508 #ifdef __s390x__
509 	if ((word & 0xffffffff) == 0) {
510 		word >>= 32;
511 		nr += 32;
512 	}
513 #endif
514 	if ((word & 0xffff) == 0) {
515 		word >>= 16;
516 		nr += 16;
517 	}
518 	if ((word & 0xff) == 0) {
519 		word >>= 8;
520 		nr += 8;
521 	}
522 	return nr + _sb_findmap[(unsigned char) word];
523 }
524 
525 
526 /**
527  * __load_ulong_be - load big endian unsigned long
528  * @p: pointer to array of unsigned long
529  * @offset: byte offset of source value in the array
530  */
531 static inline unsigned long __load_ulong_be(const unsigned long *p,
532 					    unsigned long offset)
533 {
534 	p = (unsigned long *)((unsigned long) p + offset);
535 	return *p;
536 }
537 
538 /**
539  * __load_ulong_le - load little endian unsigned long
540  * @p: pointer to array of unsigned long
541  * @offset: byte offset of source value in the array
542  */
543 static inline unsigned long __load_ulong_le(const unsigned long *p,
544 					    unsigned long offset)
545 {
546 	unsigned long word;
547 
548 	p = (unsigned long *)((unsigned long) p + offset);
549 #ifndef __s390x__
550 	asm volatile(
551 		"	ic	%0,%O1(%R1)\n"
552 		"	icm	%0,2,%O1+1(%R1)\n"
553 		"	icm	%0,4,%O1+2(%R1)\n"
554 		"	icm	%0,8,%O1+3(%R1)"
555 		: "=&d" (word) : "Q" (*p) : "cc");
556 #else
557 	asm volatile(
558 		"	lrvg	%0,%1"
559 		: "=d" (word) : "m" (*p) );
560 #endif
561 	return word;
562 }
563 
564 /*
565  * The various find bit functions.
566  */
567 
568 /*
569  * ffz - find first zero in word.
570  * @word: The word to search
571  *
572  * Undefined if no zero exists, so code should check against ~0UL first.
573  */
574 static inline unsigned long ffz(unsigned long word)
575 {
576 	return __ffz_word(0, word);
577 }
578 
579 /**
580  * __ffs - find first bit in word.
581  * @word: The word to search
582  *
583  * Undefined if no bit exists, so code should check against 0 first.
584  */
585 static inline unsigned long __ffs (unsigned long word)
586 {
587 	return __ffs_word(0, word);
588 }
589 
590 /**
591  * ffs - find first bit set
592  * @x: the word to search
593  *
594  * This is defined the same way as
595  * the libc and compiler builtin ffs routines, therefore
596  * differs in spirit from the above ffz (man ffs).
597  */
598 static inline int ffs(int x)
599 {
600 	if (!x)
601 		return 0;
602 	return __ffs_word(1, x);
603 }
604 
605 /**
606  * find_first_zero_bit - find the first zero bit in a memory region
607  * @addr: The address to start the search at
608  * @size: The maximum size to search
609  *
610  * Returns the bit-number of the first zero bit, not the number of the byte
611  * containing a bit.
612  */
613 static inline unsigned long find_first_zero_bit(const unsigned long *addr,
614 						unsigned long size)
615 {
616 	unsigned long bytes, bits;
617 
618         if (!size)
619                 return 0;
620 	bytes = __ffz_word_loop(addr, size);
621 	bits = __ffz_word(bytes*8, __load_ulong_be(addr, bytes));
622 	return (bits < size) ? bits : size;
623 }
624 #define find_first_zero_bit find_first_zero_bit
625 
626 /**
627  * find_first_bit - find the first set bit in a memory region
628  * @addr: The address to start the search at
629  * @size: The maximum size to search
630  *
631  * Returns the bit-number of the first set bit, not the number of the byte
632  * containing a bit.
633  */
634 static inline unsigned long find_first_bit(const unsigned long * addr,
635 					   unsigned long size)
636 {
637 	unsigned long bytes, bits;
638 
639         if (!size)
640                 return 0;
641 	bytes = __ffs_word_loop(addr, size);
642 	bits = __ffs_word(bytes*8, __load_ulong_be(addr, bytes));
643 	return (bits < size) ? bits : size;
644 }
645 #define find_first_bit find_first_bit
646 
647 /**
648  * find_next_zero_bit - find the first zero bit in a memory region
649  * @addr: The address to base the search on
650  * @offset: The bitnumber to start searching at
651  * @size: The maximum size to search
652  */
653 static inline int find_next_zero_bit (const unsigned long * addr,
654 				      unsigned long size,
655 				      unsigned long offset)
656 {
657         const unsigned long *p;
658 	unsigned long bit, set;
659 
660 	if (offset >= size)
661 		return size;
662 	bit = offset & (__BITOPS_WORDSIZE - 1);
663 	offset -= bit;
664 	size -= offset;
665 	p = addr + offset / __BITOPS_WORDSIZE;
666 	if (bit) {
667 		/*
668 		 * __ffz_word returns __BITOPS_WORDSIZE
669 		 * if no zero bit is present in the word.
670 		 */
671 		set = __ffz_word(bit, *p >> bit);
672 		if (set >= size)
673 			return size + offset;
674 		if (set < __BITOPS_WORDSIZE)
675 			return set + offset;
676 		offset += __BITOPS_WORDSIZE;
677 		size -= __BITOPS_WORDSIZE;
678 		p++;
679 	}
680 	return offset + find_first_zero_bit(p, size);
681 }
682 #define find_next_zero_bit find_next_zero_bit
683 
684 /**
685  * find_next_bit - find the first set bit in a memory region
686  * @addr: The address to base the search on
687  * @offset: The bitnumber to start searching at
688  * @size: The maximum size to search
689  */
690 static inline int find_next_bit (const unsigned long * addr,
691 				 unsigned long size,
692 				 unsigned long offset)
693 {
694         const unsigned long *p;
695 	unsigned long bit, set;
696 
697 	if (offset >= size)
698 		return size;
699 	bit = offset & (__BITOPS_WORDSIZE - 1);
700 	offset -= bit;
701 	size -= offset;
702 	p = addr + offset / __BITOPS_WORDSIZE;
703 	if (bit) {
704 		/*
705 		 * __ffs_word returns __BITOPS_WORDSIZE
706 		 * if no one bit is present in the word.
707 		 */
708 		set = __ffs_word(0, *p & (~0UL << bit));
709 		if (set >= size)
710 			return size + offset;
711 		if (set < __BITOPS_WORDSIZE)
712 			return set + offset;
713 		offset += __BITOPS_WORDSIZE;
714 		size -= __BITOPS_WORDSIZE;
715 		p++;
716 	}
717 	return offset + find_first_bit(p, size);
718 }
719 #define find_next_bit find_next_bit
720 
721 /*
722  * Every architecture must define this function. It's the fastest
723  * way of searching a 140-bit bitmap where the first 100 bits are
724  * unlikely to be set. It's guaranteed that at least one of the 140
725  * bits is cleared.
726  */
727 static inline int sched_find_first_bit(unsigned long *b)
728 {
729 	return find_first_bit(b, 140);
730 }
731 
732 #include <asm-generic/bitops/fls.h>
733 #include <asm-generic/bitops/__fls.h>
734 #include <asm-generic/bitops/fls64.h>
735 
736 #include <asm-generic/bitops/hweight.h>
737 #include <asm-generic/bitops/lock.h>
738 
739 /*
740  * ATTENTION: intel byte ordering convention for ext2 and minix !!
741  * bit 0 is the LSB of addr; bit 31 is the MSB of addr;
742  * bit 32 is the LSB of (addr+4).
743  * That combined with the little endian byte order of Intel gives the
744  * following bit order in memory:
745  *    07 06 05 04 03 02 01 00 15 14 13 12 11 10 09 08 \
746  *    23 22 21 20 19 18 17 16 31 30 29 28 27 26 25 24
747  */
748 
749 static inline int find_first_zero_bit_le(void *vaddr, unsigned int size)
750 {
751 	unsigned long bytes, bits;
752 
753         if (!size)
754                 return 0;
755 	bytes = __ffz_word_loop(vaddr, size);
756 	bits = __ffz_word(bytes*8, __load_ulong_le(vaddr, bytes));
757 	return (bits < size) ? bits : size;
758 }
759 #define find_first_zero_bit_le find_first_zero_bit_le
760 
761 static inline int find_next_zero_bit_le(void *vaddr, unsigned long size,
762 					  unsigned long offset)
763 {
764         unsigned long *addr = vaddr, *p;
765 	unsigned long bit, set;
766 
767         if (offset >= size)
768                 return size;
769 	bit = offset & (__BITOPS_WORDSIZE - 1);
770 	offset -= bit;
771 	size -= offset;
772 	p = addr + offset / __BITOPS_WORDSIZE;
773         if (bit) {
774 		/*
775 		 * s390 version of ffz returns __BITOPS_WORDSIZE
776 		 * if no zero bit is present in the word.
777 		 */
778 		set = __ffz_word(bit, __load_ulong_le(p, 0) >> bit);
779 		if (set >= size)
780 			return size + offset;
781 		if (set < __BITOPS_WORDSIZE)
782 			return set + offset;
783 		offset += __BITOPS_WORDSIZE;
784 		size -= __BITOPS_WORDSIZE;
785 		p++;
786         }
787 	return offset + find_first_zero_bit_le(p, size);
788 }
789 #define find_next_zero_bit_le find_next_zero_bit_le
790 
791 static inline unsigned long find_first_bit_le(void *vaddr, unsigned long size)
792 {
793 	unsigned long bytes, bits;
794 
795 	if (!size)
796 		return 0;
797 	bytes = __ffs_word_loop(vaddr, size);
798 	bits = __ffs_word(bytes*8, __load_ulong_le(vaddr, bytes));
799 	return (bits < size) ? bits : size;
800 }
801 #define find_first_bit_le find_first_bit_le
802 
803 static inline int find_next_bit_le(void *vaddr, unsigned long size,
804 				     unsigned long offset)
805 {
806 	unsigned long *addr = vaddr, *p;
807 	unsigned long bit, set;
808 
809 	if (offset >= size)
810 		return size;
811 	bit = offset & (__BITOPS_WORDSIZE - 1);
812 	offset -= bit;
813 	size -= offset;
814 	p = addr + offset / __BITOPS_WORDSIZE;
815 	if (bit) {
816 		/*
817 		 * s390 version of ffz returns __BITOPS_WORDSIZE
818 		 * if no zero bit is present in the word.
819 		 */
820 		set = __ffs_word(0, __load_ulong_le(p, 0) & (~0UL << bit));
821 		if (set >= size)
822 			return size + offset;
823 		if (set < __BITOPS_WORDSIZE)
824 			return set + offset;
825 		offset += __BITOPS_WORDSIZE;
826 		size -= __BITOPS_WORDSIZE;
827 		p++;
828 	}
829 	return offset + find_first_bit_le(p, size);
830 }
831 #define find_next_bit_le find_next_bit_le
832 
833 #include <asm-generic/bitops/le.h>
834 
835 #define ext2_set_bit_atomic(lock, nr, addr)	\
836 	test_and_set_bit_le(nr, addr)
837 #define ext2_clear_bit_atomic(lock, nr, addr)	\
838 	test_and_clear_bit_le(nr, addr)
839 
840 
841 #endif /* __KERNEL__ */
842 
843 #endif /* _S390_BITOPS_H */
844