xref: /openbmc/linux/arch/s390/include/asm/bitops.h (revision cd5d5810)
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_OR		"or"
65 #define __BITOPS_AND		"nr"
66 #define __BITOPS_XOR		"xr"
67 
68 #define __BITOPS_LOOP(__old, __new, __addr, __val, __op_string)	\
69 	asm volatile(						\
70 		"	l	%0,%2\n"			\
71 		"0:	lr	%1,%0\n"			\
72 		__op_string "	%1,%3\n"			\
73 		"	cs	%0,%1,%2\n"			\
74 		"	jl	0b"				\
75 		: "=&d" (__old), "=&d" (__new),			\
76 		  "=Q" (*(unsigned long *) __addr)		\
77 		: "d" (__val), "Q" (*(unsigned long *) __addr)	\
78 		: "cc");
79 
80 #else /* CONFIG_64BIT */
81 
82 #define __BITOPS_OR		"ogr"
83 #define __BITOPS_AND		"ngr"
84 #define __BITOPS_XOR		"xgr"
85 
86 #define __BITOPS_LOOP(__old, __new, __addr, __val, __op_string)	\
87 	asm volatile(						\
88 		"	lg	%0,%2\n"			\
89 		"0:	lgr	%1,%0\n"			\
90 		__op_string "	%1,%3\n"			\
91 		"	csg	%0,%1,%2\n"			\
92 		"	jl	0b"				\
93 		: "=&d" (__old), "=&d" (__new),			\
94 		  "=Q" (*(unsigned long *) __addr)		\
95 		: "d" (__val), "Q" (*(unsigned long *) __addr)	\
96 		: "cc");
97 
98 #endif /* CONFIG_64BIT */
99 
100 #define __BITOPS_WORDS(bits) (((bits) + BITS_PER_LONG - 1) / BITS_PER_LONG)
101 
102 #ifdef CONFIG_SMP
103 /*
104  * SMP safe set_bit routine based on compare and swap (CS)
105  */
106 static inline void set_bit_cs(unsigned long nr, volatile unsigned long *ptr)
107 {
108         unsigned long addr, old, new, mask;
109 
110 	addr = (unsigned long) ptr;
111 	/* calculate address for CS */
112 	addr += (nr ^ (nr & (BITS_PER_LONG - 1))) >> 3;
113 	/* make OR mask */
114 	mask = 1UL << (nr & (BITS_PER_LONG - 1));
115 	/* Do the atomic update. */
116 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_OR);
117 }
118 
119 /*
120  * SMP safe clear_bit routine based on compare and swap (CS)
121  */
122 static inline void clear_bit_cs(unsigned long nr, volatile unsigned long *ptr)
123 {
124         unsigned long addr, old, new, mask;
125 
126 	addr = (unsigned long) ptr;
127 	/* calculate address for CS */
128 	addr += (nr ^ (nr & (BITS_PER_LONG - 1))) >> 3;
129 	/* make AND mask */
130 	mask = ~(1UL << (nr & (BITS_PER_LONG - 1)));
131 	/* Do the atomic update. */
132 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_AND);
133 }
134 
135 /*
136  * SMP safe change_bit routine based on compare and swap (CS)
137  */
138 static inline void change_bit_cs(unsigned long nr, volatile unsigned long *ptr)
139 {
140         unsigned long addr, old, new, mask;
141 
142 	addr = (unsigned long) ptr;
143 	/* calculate address for CS */
144 	addr += (nr ^ (nr & (BITS_PER_LONG - 1))) >> 3;
145 	/* make XOR mask */
146 	mask = 1UL << (nr & (BITS_PER_LONG - 1));
147 	/* Do the atomic update. */
148 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_XOR);
149 }
150 
151 /*
152  * SMP safe test_and_set_bit routine based on compare and swap (CS)
153  */
154 static inline int
155 test_and_set_bit_cs(unsigned long nr, volatile unsigned long *ptr)
156 {
157         unsigned long addr, old, new, mask;
158 
159 	addr = (unsigned long) ptr;
160 	/* calculate address for CS */
161 	addr += (nr ^ (nr & (BITS_PER_LONG - 1))) >> 3;
162 	/* make OR/test mask */
163 	mask = 1UL << (nr & (BITS_PER_LONG - 1));
164 	/* Do the atomic update. */
165 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_OR);
166 	barrier();
167 	return (old & mask) != 0;
168 }
169 
170 /*
171  * SMP safe test_and_clear_bit routine based on compare and swap (CS)
172  */
173 static inline int
174 test_and_clear_bit_cs(unsigned long nr, volatile unsigned long *ptr)
175 {
176         unsigned long addr, old, new, mask;
177 
178 	addr = (unsigned long) ptr;
179 	/* calculate address for CS */
180 	addr += (nr ^ (nr & (BITS_PER_LONG - 1))) >> 3;
181 	/* make AND/test mask */
182 	mask = ~(1UL << (nr & (BITS_PER_LONG - 1)));
183 	/* Do the atomic update. */
184 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_AND);
185 	barrier();
186 	return (old ^ new) != 0;
187 }
188 
189 /*
190  * SMP safe test_and_change_bit routine based on compare and swap (CS)
191  */
192 static inline int
193 test_and_change_bit_cs(unsigned long nr, volatile unsigned long *ptr)
194 {
195         unsigned long addr, old, new, mask;
196 
197 	addr = (unsigned long) ptr;
198 	/* calculate address for CS */
199 	addr += (nr ^ (nr & (BITS_PER_LONG - 1))) >> 3;
200 	/* make XOR/test mask */
201 	mask = 1UL << (nr & (BITS_PER_LONG - 1));
202 	/* Do the atomic update. */
203 	__BITOPS_LOOP(old, new, addr, mask, __BITOPS_XOR);
204 	barrier();
205 	return (old & mask) != 0;
206 }
207 #endif /* CONFIG_SMP */
208 
209 /*
210  * fast, non-SMP set_bit routine
211  */
212 static inline void __set_bit(unsigned long nr, volatile unsigned long *ptr)
213 {
214 	unsigned long addr;
215 
216 	addr = (unsigned long) ptr + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
217 	asm volatile(
218 		"	oc	%O0(1,%R0),%1"
219 		: "+Q" (*(char *) addr) : "Q" (_oi_bitmap[nr & 7]) : "cc");
220 }
221 
222 static inline void
223 __constant_set_bit(const unsigned long nr, volatile unsigned long *ptr)
224 {
225 	unsigned long addr;
226 
227 	addr = ((unsigned long) ptr) + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
228 	*(unsigned char *) addr |= 1 << (nr & 7);
229 }
230 
231 #define set_bit_simple(nr,addr) \
232 (__builtin_constant_p((nr)) ? \
233  __constant_set_bit((nr),(addr)) : \
234  __set_bit((nr),(addr)) )
235 
236 /*
237  * fast, non-SMP clear_bit routine
238  */
239 static inline void
240 __clear_bit(unsigned long nr, volatile unsigned long *ptr)
241 {
242 	unsigned long addr;
243 
244 	addr = (unsigned long) ptr + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
245 	asm volatile(
246 		"	nc	%O0(1,%R0),%1"
247 		: "+Q" (*(char *) addr) : "Q" (_ni_bitmap[nr & 7]) : "cc");
248 }
249 
250 static inline void
251 __constant_clear_bit(const unsigned long nr, volatile unsigned long *ptr)
252 {
253 	unsigned long addr;
254 
255 	addr = ((unsigned long) ptr) + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
256 	*(unsigned char *) addr &= ~(1 << (nr & 7));
257 }
258 
259 #define clear_bit_simple(nr,addr) \
260 (__builtin_constant_p((nr)) ? \
261  __constant_clear_bit((nr),(addr)) : \
262  __clear_bit((nr),(addr)) )
263 
264 /*
265  * fast, non-SMP change_bit routine
266  */
267 static inline void __change_bit(unsigned long nr, volatile unsigned long *ptr)
268 {
269 	unsigned long addr;
270 
271 	addr = (unsigned long) ptr + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
272 	asm volatile(
273 		"	xc	%O0(1,%R0),%1"
274 		: "+Q" (*(char *) addr) : "Q" (_oi_bitmap[nr & 7]) : "cc");
275 }
276 
277 static inline void
278 __constant_change_bit(const unsigned long nr, volatile unsigned long *ptr)
279 {
280 	unsigned long addr;
281 
282 	addr = ((unsigned long) ptr) + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
283 	*(unsigned char *) addr ^= 1 << (nr & 7);
284 }
285 
286 #define change_bit_simple(nr,addr) \
287 (__builtin_constant_p((nr)) ? \
288  __constant_change_bit((nr),(addr)) : \
289  __change_bit((nr),(addr)) )
290 
291 /*
292  * fast, non-SMP test_and_set_bit routine
293  */
294 static inline int
295 test_and_set_bit_simple(unsigned long nr, volatile unsigned long *ptr)
296 {
297 	unsigned long addr;
298 	unsigned char ch;
299 
300 	addr = (unsigned long) ptr + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
301 	ch = *(unsigned char *) addr;
302 	asm volatile(
303 		"	oc	%O0(1,%R0),%1"
304 		: "+Q" (*(char *) addr)	: "Q" (_oi_bitmap[nr & 7])
305 		: "cc", "memory");
306 	return (ch >> (nr & 7)) & 1;
307 }
308 #define __test_and_set_bit(X,Y)		test_and_set_bit_simple(X,Y)
309 
310 /*
311  * fast, non-SMP test_and_clear_bit routine
312  */
313 static inline int
314 test_and_clear_bit_simple(unsigned long nr, volatile unsigned long *ptr)
315 {
316 	unsigned long addr;
317 	unsigned char ch;
318 
319 	addr = (unsigned long) ptr + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
320 	ch = *(unsigned char *) addr;
321 	asm volatile(
322 		"	nc	%O0(1,%R0),%1"
323 		: "+Q" (*(char *) addr)	: "Q" (_ni_bitmap[nr & 7])
324 		: "cc", "memory");
325 	return (ch >> (nr & 7)) & 1;
326 }
327 #define __test_and_clear_bit(X,Y)	test_and_clear_bit_simple(X,Y)
328 
329 /*
330  * fast, non-SMP test_and_change_bit routine
331  */
332 static inline int
333 test_and_change_bit_simple(unsigned long nr, volatile unsigned long *ptr)
334 {
335 	unsigned long addr;
336 	unsigned char ch;
337 
338 	addr = (unsigned long) ptr + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
339 	ch = *(unsigned char *) addr;
340 	asm volatile(
341 		"	xc	%O0(1,%R0),%1"
342 		: "+Q" (*(char *) addr)	: "Q" (_oi_bitmap[nr & 7])
343 		: "cc", "memory");
344 	return (ch >> (nr & 7)) & 1;
345 }
346 #define __test_and_change_bit(X,Y)	test_and_change_bit_simple(X,Y)
347 
348 #ifdef CONFIG_SMP
349 #define set_bit             set_bit_cs
350 #define clear_bit           clear_bit_cs
351 #define change_bit          change_bit_cs
352 #define test_and_set_bit    test_and_set_bit_cs
353 #define test_and_clear_bit  test_and_clear_bit_cs
354 #define test_and_change_bit test_and_change_bit_cs
355 #else
356 #define set_bit             set_bit_simple
357 #define clear_bit           clear_bit_simple
358 #define change_bit          change_bit_simple
359 #define test_and_set_bit    test_and_set_bit_simple
360 #define test_and_clear_bit  test_and_clear_bit_simple
361 #define test_and_change_bit test_and_change_bit_simple
362 #endif
363 
364 
365 /*
366  * This routine doesn't need to be atomic.
367  */
368 
369 static inline int __test_bit(unsigned long nr, const volatile unsigned long *ptr)
370 {
371 	unsigned long addr;
372 	unsigned char ch;
373 
374 	addr = (unsigned long) ptr + ((nr ^ (BITS_PER_LONG - 8)) >> 3);
375 	ch = *(volatile unsigned char *) addr;
376 	return (ch >> (nr & 7)) & 1;
377 }
378 
379 static inline int
380 __constant_test_bit(unsigned long nr, const volatile unsigned long *addr) {
381     return (((volatile char *) addr)
382 	    [(nr^(BITS_PER_LONG-8))>>3] & (1<<(nr&7))) != 0;
383 }
384 
385 #define test_bit(nr,addr) \
386 (__builtin_constant_p((nr)) ? \
387  __constant_test_bit((nr),(addr)) : \
388  __test_bit((nr),(addr)) )
389 
390 /*
391  * Optimized find bit helper functions.
392  */
393 
394 /**
395  * __ffz_word_loop - find byte offset of first long != -1UL
396  * @addr: pointer to array of unsigned long
397  * @size: size of the array in bits
398  */
399 static inline unsigned long __ffz_word_loop(const unsigned long *addr,
400 					    unsigned long size)
401 {
402 	typedef struct { long _[__BITOPS_WORDS(size)]; } addrtype;
403 	unsigned long bytes = 0;
404 
405 	asm volatile(
406 #ifndef CONFIG_64BIT
407 		"	ahi	%1,-1\n"
408 		"	sra	%1,5\n"
409 		"	jz	1f\n"
410 		"0:	c	%2,0(%0,%3)\n"
411 		"	jne	1f\n"
412 		"	la	%0,4(%0)\n"
413 		"	brct	%1,0b\n"
414 		"1:\n"
415 #else
416 		"	aghi	%1,-1\n"
417 		"	srag	%1,%1,6\n"
418 		"	jz	1f\n"
419 		"0:	cg	%2,0(%0,%3)\n"
420 		"	jne	1f\n"
421 		"	la	%0,8(%0)\n"
422 		"	brct	%1,0b\n"
423 		"1:\n"
424 #endif
425 		: "+&a" (bytes), "+&d" (size)
426 		: "d" (-1UL), "a" (addr), "m" (*(addrtype *) addr)
427 		: "cc" );
428 	return bytes;
429 }
430 
431 /**
432  * __ffs_word_loop - find byte offset of first long != 0UL
433  * @addr: pointer to array of unsigned long
434  * @size: size of the array in bits
435  */
436 static inline unsigned long __ffs_word_loop(const unsigned long *addr,
437 					    unsigned long size)
438 {
439 	typedef struct { long _[__BITOPS_WORDS(size)]; } addrtype;
440 	unsigned long bytes = 0;
441 
442 	asm volatile(
443 #ifndef CONFIG_64BIT
444 		"	ahi	%1,-1\n"
445 		"	sra	%1,5\n"
446 		"	jz	1f\n"
447 		"0:	c	%2,0(%0,%3)\n"
448 		"	jne	1f\n"
449 		"	la	%0,4(%0)\n"
450 		"	brct	%1,0b\n"
451 		"1:\n"
452 #else
453 		"	aghi	%1,-1\n"
454 		"	srag	%1,%1,6\n"
455 		"	jz	1f\n"
456 		"0:	cg	%2,0(%0,%3)\n"
457 		"	jne	1f\n"
458 		"	la	%0,8(%0)\n"
459 		"	brct	%1,0b\n"
460 		"1:\n"
461 #endif
462 		: "+&a" (bytes), "+&a" (size)
463 		: "d" (0UL), "a" (addr), "m" (*(addrtype *) addr)
464 		: "cc" );
465 	return bytes;
466 }
467 
468 /**
469  * __ffz_word - add number of the first unset bit
470  * @nr: base value the bit number is added to
471  * @word: the word that is searched for unset bits
472  */
473 static inline unsigned long __ffz_word(unsigned long nr, unsigned long word)
474 {
475 #ifdef CONFIG_64BIT
476 	if ((word & 0xffffffff) == 0xffffffff) {
477 		word >>= 32;
478 		nr += 32;
479 	}
480 #endif
481 	if ((word & 0xffff) == 0xffff) {
482 		word >>= 16;
483 		nr += 16;
484 	}
485 	if ((word & 0xff) == 0xff) {
486 		word >>= 8;
487 		nr += 8;
488 	}
489 	return nr + _zb_findmap[(unsigned char) word];
490 }
491 
492 /**
493  * __ffs_word - add number of the first set bit
494  * @nr: base value the bit number is added to
495  * @word: the word that is searched for set bits
496  */
497 static inline unsigned long __ffs_word(unsigned long nr, unsigned long word)
498 {
499 #ifdef CONFIG_64BIT
500 	if ((word & 0xffffffff) == 0) {
501 		word >>= 32;
502 		nr += 32;
503 	}
504 #endif
505 	if ((word & 0xffff) == 0) {
506 		word >>= 16;
507 		nr += 16;
508 	}
509 	if ((word & 0xff) == 0) {
510 		word >>= 8;
511 		nr += 8;
512 	}
513 	return nr + _sb_findmap[(unsigned char) word];
514 }
515 
516 
517 /**
518  * __load_ulong_be - load big endian unsigned long
519  * @p: pointer to array of unsigned long
520  * @offset: byte offset of source value in the array
521  */
522 static inline unsigned long __load_ulong_be(const unsigned long *p,
523 					    unsigned long offset)
524 {
525 	p = (unsigned long *)((unsigned long) p + offset);
526 	return *p;
527 }
528 
529 /**
530  * __load_ulong_le - load little endian unsigned long
531  * @p: pointer to array of unsigned long
532  * @offset: byte offset of source value in the array
533  */
534 static inline unsigned long __load_ulong_le(const unsigned long *p,
535 					    unsigned long offset)
536 {
537 	unsigned long word;
538 
539 	p = (unsigned long *)((unsigned long) p + offset);
540 #ifndef CONFIG_64BIT
541 	asm volatile(
542 		"	ic	%0,%O1(%R1)\n"
543 		"	icm	%0,2,%O1+1(%R1)\n"
544 		"	icm	%0,4,%O1+2(%R1)\n"
545 		"	icm	%0,8,%O1+3(%R1)"
546 		: "=&d" (word) : "Q" (*p) : "cc");
547 #else
548 	asm volatile(
549 		"	lrvg	%0,%1"
550 		: "=d" (word) : "m" (*p) );
551 #endif
552 	return word;
553 }
554 
555 /*
556  * The various find bit functions.
557  */
558 
559 /*
560  * ffz - find first zero in word.
561  * @word: The word to search
562  *
563  * Undefined if no zero exists, so code should check against ~0UL first.
564  */
565 static inline unsigned long ffz(unsigned long word)
566 {
567 	return __ffz_word(0, word);
568 }
569 
570 /**
571  * __ffs - find first bit in word.
572  * @word: The word to search
573  *
574  * Undefined if no bit exists, so code should check against 0 first.
575  */
576 static inline unsigned long __ffs (unsigned long word)
577 {
578 	return __ffs_word(0, word);
579 }
580 
581 /**
582  * ffs - find first bit set
583  * @x: the word to search
584  *
585  * This is defined the same way as
586  * the libc and compiler builtin ffs routines, therefore
587  * differs in spirit from the above ffz (man ffs).
588  */
589 static inline int ffs(int x)
590 {
591 	if (!x)
592 		return 0;
593 	return __ffs_word(1, x);
594 }
595 
596 /**
597  * find_first_zero_bit - find the first zero bit in a memory region
598  * @addr: The address to start the search at
599  * @size: The maximum size to search
600  *
601  * Returns the bit-number of the first zero bit, not the number of the byte
602  * containing a bit.
603  */
604 static inline unsigned long find_first_zero_bit(const unsigned long *addr,
605 						unsigned long size)
606 {
607 	unsigned long bytes, bits;
608 
609         if (!size)
610                 return 0;
611 	bytes = __ffz_word_loop(addr, size);
612 	bits = __ffz_word(bytes*8, __load_ulong_be(addr, bytes));
613 	return (bits < size) ? bits : size;
614 }
615 #define find_first_zero_bit find_first_zero_bit
616 
617 /**
618  * find_first_bit - find the first set bit in a memory region
619  * @addr: The address to start the search at
620  * @size: The maximum size to search
621  *
622  * Returns the bit-number of the first set bit, not the number of the byte
623  * containing a bit.
624  */
625 static inline unsigned long find_first_bit(const unsigned long * addr,
626 					   unsigned long size)
627 {
628 	unsigned long bytes, bits;
629 
630         if (!size)
631                 return 0;
632 	bytes = __ffs_word_loop(addr, size);
633 	bits = __ffs_word(bytes*8, __load_ulong_be(addr, bytes));
634 	return (bits < size) ? bits : size;
635 }
636 #define find_first_bit find_first_bit
637 
638 /*
639  * Big endian variant whichs starts bit counting from left using
640  * the flogr (find leftmost one) instruction.
641  */
642 static inline unsigned long __flo_word(unsigned long nr, unsigned long val)
643 {
644 	register unsigned long bit asm("2") = val;
645 	register unsigned long out asm("3");
646 
647 	asm volatile (
648 		"	.insn	rre,0xb9830000,%[bit],%[bit]\n"
649 		: [bit] "+d" (bit), [out] "=d" (out) : : "cc");
650 	return nr + bit;
651 }
652 
653 /*
654  * 64 bit special left bitops format:
655  * order in memory:
656  *    00 01 02 03 04 05 06 07 08 09 0a 0b 0c 0d 0e 0f
657  *    10 11 12 13 14 15 16 17 18 19 1a 1b 1c 1d 1e 1f
658  *    20 21 22 23 24 25 26 27 28 29 2a 2b 2c 2d 2e 2f
659  *    30 31 32 33 34 35 36 37 38 39 3a 3b 3c 3d 3e 3f
660  * after that follows the next long with bit numbers
661  *    40 41 42 43 44 45 46 47 48 49 4a 4b 4c 4d 4e 4f
662  *    50 51 52 53 54 55 56 57 58 59 5a 5b 5c 5d 5e 5f
663  *    60 61 62 63 64 65 66 67 68 69 6a 6b 6c 6d 6e 6f
664  *    70 71 72 73 74 75 76 77 78 79 7a 7b 7c 7d 7e 7f
665  * The reason for this bit ordering is the fact that
666  * the hardware sets bits in a bitmap starting at bit 0
667  * and we don't want to scan the bitmap from the 'wrong
668  * end'.
669  */
670 static inline unsigned long find_first_bit_left(const unsigned long *addr,
671 						unsigned long size)
672 {
673 	unsigned long bytes, bits;
674 
675 	if (!size)
676 		return 0;
677 	bytes = __ffs_word_loop(addr, size);
678 	bits = __flo_word(bytes * 8, __load_ulong_be(addr, bytes));
679 	return (bits < size) ? bits : size;
680 }
681 
682 static inline int find_next_bit_left(const unsigned long *addr,
683 				     unsigned long size,
684 				     unsigned long offset)
685 {
686 	const unsigned long *p;
687 	unsigned long bit, set;
688 
689 	if (offset >= size)
690 		return size;
691 	bit = offset & (BITS_PER_LONG - 1);
692 	offset -= bit;
693 	size -= offset;
694 	p = addr + offset / BITS_PER_LONG;
695 	if (bit) {
696 		set = __flo_word(0, *p & (~0UL >> bit));
697 		if (set >= size)
698 			return size + offset;
699 		if (set < BITS_PER_LONG)
700 			return set + offset;
701 		offset += BITS_PER_LONG;
702 		size -= BITS_PER_LONG;
703 		p++;
704 	}
705 	return offset + find_first_bit_left(p, size);
706 }
707 
708 #define for_each_set_bit_left(bit, addr, size)				\
709 	for ((bit) = find_first_bit_left((addr), (size));		\
710 	     (bit) < (size);						\
711 	     (bit) = find_next_bit_left((addr), (size), (bit) + 1))
712 
713 /* same as for_each_set_bit() but use bit as value to start with */
714 #define for_each_set_bit_left_cont(bit, addr, size)			\
715 	for ((bit) = find_next_bit_left((addr), (size), (bit));		\
716 	     (bit) < (size);						\
717 	     (bit) = find_next_bit_left((addr), (size), (bit) + 1))
718 
719 /**
720  * find_next_zero_bit - find the first zero bit in a memory region
721  * @addr: The address to base the search on
722  * @offset: The bitnumber to start searching at
723  * @size: The maximum size to search
724  */
725 static inline int find_next_zero_bit (const unsigned long * addr,
726 				      unsigned long size,
727 				      unsigned long offset)
728 {
729         const unsigned long *p;
730 	unsigned long bit, set;
731 
732 	if (offset >= size)
733 		return size;
734 	bit = offset & (BITS_PER_LONG - 1);
735 	offset -= bit;
736 	size -= offset;
737 	p = addr + offset / BITS_PER_LONG;
738 	if (bit) {
739 		/*
740 		 * __ffz_word returns BITS_PER_LONG
741 		 * if no zero bit is present in the word.
742 		 */
743 		set = __ffz_word(bit, *p >> bit);
744 		if (set >= size)
745 			return size + offset;
746 		if (set < BITS_PER_LONG)
747 			return set + offset;
748 		offset += BITS_PER_LONG;
749 		size -= BITS_PER_LONG;
750 		p++;
751 	}
752 	return offset + find_first_zero_bit(p, size);
753 }
754 #define find_next_zero_bit find_next_zero_bit
755 
756 /**
757  * find_next_bit - find the first set bit in a memory region
758  * @addr: The address to base the search on
759  * @offset: The bitnumber to start searching at
760  * @size: The maximum size to search
761  */
762 static inline int find_next_bit (const unsigned long * addr,
763 				 unsigned long size,
764 				 unsigned long offset)
765 {
766         const unsigned long *p;
767 	unsigned long bit, set;
768 
769 	if (offset >= size)
770 		return size;
771 	bit = offset & (BITS_PER_LONG - 1);
772 	offset -= bit;
773 	size -= offset;
774 	p = addr + offset / BITS_PER_LONG;
775 	if (bit) {
776 		/*
777 		 * __ffs_word returns BITS_PER_LONG
778 		 * if no one bit is present in the word.
779 		 */
780 		set = __ffs_word(0, *p & (~0UL << bit));
781 		if (set >= size)
782 			return size + offset;
783 		if (set < BITS_PER_LONG)
784 			return set + offset;
785 		offset += BITS_PER_LONG;
786 		size -= BITS_PER_LONG;
787 		p++;
788 	}
789 	return offset + find_first_bit(p, size);
790 }
791 #define find_next_bit find_next_bit
792 
793 /*
794  * Every architecture must define this function. It's the fastest
795  * way of searching a 140-bit bitmap where the first 100 bits are
796  * unlikely to be set. It's guaranteed that at least one of the 140
797  * bits is cleared.
798  */
799 static inline int sched_find_first_bit(unsigned long *b)
800 {
801 	return find_first_bit(b, 140);
802 }
803 
804 #include <asm-generic/bitops/fls.h>
805 #include <asm-generic/bitops/__fls.h>
806 #include <asm-generic/bitops/fls64.h>
807 
808 #include <asm-generic/bitops/hweight.h>
809 #include <asm-generic/bitops/lock.h>
810 
811 /*
812  * ATTENTION: intel byte ordering convention for ext2 and minix !!
813  * bit 0 is the LSB of addr; bit 31 is the MSB of addr;
814  * bit 32 is the LSB of (addr+4).
815  * That combined with the little endian byte order of Intel gives the
816  * following bit order in memory:
817  *    07 06 05 04 03 02 01 00 15 14 13 12 11 10 09 08 \
818  *    23 22 21 20 19 18 17 16 31 30 29 28 27 26 25 24
819  */
820 
821 static inline int find_first_zero_bit_le(void *vaddr, unsigned int size)
822 {
823 	unsigned long bytes, bits;
824 
825         if (!size)
826                 return 0;
827 	bytes = __ffz_word_loop(vaddr, size);
828 	bits = __ffz_word(bytes*8, __load_ulong_le(vaddr, bytes));
829 	return (bits < size) ? bits : size;
830 }
831 #define find_first_zero_bit_le find_first_zero_bit_le
832 
833 static inline int find_next_zero_bit_le(void *vaddr, unsigned long size,
834 					  unsigned long offset)
835 {
836         unsigned long *addr = vaddr, *p;
837 	unsigned long bit, set;
838 
839         if (offset >= size)
840                 return size;
841 	bit = offset & (BITS_PER_LONG - 1);
842 	offset -= bit;
843 	size -= offset;
844 	p = addr + offset / BITS_PER_LONG;
845         if (bit) {
846 		/*
847 		 * s390 version of ffz returns BITS_PER_LONG
848 		 * if no zero bit is present in the word.
849 		 */
850 		set = __ffz_word(bit, __load_ulong_le(p, 0) >> bit);
851 		if (set >= size)
852 			return size + offset;
853 		if (set < BITS_PER_LONG)
854 			return set + offset;
855 		offset += BITS_PER_LONG;
856 		size -= BITS_PER_LONG;
857 		p++;
858         }
859 	return offset + find_first_zero_bit_le(p, size);
860 }
861 #define find_next_zero_bit_le find_next_zero_bit_le
862 
863 static inline unsigned long find_first_bit_le(void *vaddr, unsigned long size)
864 {
865 	unsigned long bytes, bits;
866 
867 	if (!size)
868 		return 0;
869 	bytes = __ffs_word_loop(vaddr, size);
870 	bits = __ffs_word(bytes*8, __load_ulong_le(vaddr, bytes));
871 	return (bits < size) ? bits : size;
872 }
873 #define find_first_bit_le find_first_bit_le
874 
875 static inline int find_next_bit_le(void *vaddr, unsigned long size,
876 				     unsigned long offset)
877 {
878 	unsigned long *addr = vaddr, *p;
879 	unsigned long bit, set;
880 
881 	if (offset >= size)
882 		return size;
883 	bit = offset & (BITS_PER_LONG - 1);
884 	offset -= bit;
885 	size -= offset;
886 	p = addr + offset / BITS_PER_LONG;
887 	if (bit) {
888 		/*
889 		 * s390 version of ffz returns BITS_PER_LONG
890 		 * if no zero bit is present in the word.
891 		 */
892 		set = __ffs_word(0, __load_ulong_le(p, 0) & (~0UL << bit));
893 		if (set >= size)
894 			return size + offset;
895 		if (set < BITS_PER_LONG)
896 			return set + offset;
897 		offset += BITS_PER_LONG;
898 		size -= BITS_PER_LONG;
899 		p++;
900 	}
901 	return offset + find_first_bit_le(p, size);
902 }
903 #define find_next_bit_le find_next_bit_le
904 
905 #include <asm-generic/bitops/le.h>
906 
907 #include <asm-generic/bitops/ext2-atomic-setbit.h>
908 
909 #endif /* _S390_BITOPS_H */
910