xref: /linux/arch/powerpc/include/asm/bitops.h (revision 4949009eb8d40a441dcddcd96e101e77d31cf1b2)
1 /*
2  * PowerPC atomic bit operations.
3  *
4  * Merged version by David Gibson <david@gibson.dropbear.id.au>.
5  * Based on ppc64 versions by: Dave Engebretsen, Todd Inglett, Don
6  * Reed, Pat McCarthy, Peter Bergner, Anton Blanchard.  They
7  * originally took it from the ppc32 code.
8  *
9  * Within a word, bits are numbered LSB first.  Lot's of places make
10  * this assumption by directly testing bits with (val & (1<<nr)).
11  * This can cause confusion for large (> 1 word) bitmaps on a
12  * big-endian system because, unlike little endian, the number of each
13  * bit depends on the word size.
14  *
15  * The bitop functions are defined to work on unsigned longs, so for a
16  * ppc64 system the bits end up numbered:
17  *   |63..............0|127............64|191...........128|255...........192|
18  * and on ppc32:
19  *   |31.....0|63....32|95....64|127...96|159..128|191..160|223..192|255..224|
20  *
21  * There are a few little-endian macros used mostly for filesystem
22  * bitmaps, these work on similar bit arrays layouts, but
23  * byte-oriented:
24  *   |7...0|15...8|23...16|31...24|39...32|47...40|55...48|63...56|
25  *
26  * The main difference is that bit 3-5 (64b) or 3-4 (32b) in the bit
27  * number field needs to be reversed compared to the big-endian bit
28  * fields. This can be achieved by XOR with 0x38 (64b) or 0x18 (32b).
29  *
30  * This program is free software; you can redistribute it and/or
31  * modify it under the terms of the GNU General Public License
32  * as published by the Free Software Foundation; either version
33  * 2 of the License, or (at your option) any later version.
34  */
35 
36 #ifndef _ASM_POWERPC_BITOPS_H
37 #define _ASM_POWERPC_BITOPS_H
38 
39 #ifdef __KERNEL__
40 
41 #ifndef _LINUX_BITOPS_H
42 #error only <linux/bitops.h> can be included directly
43 #endif
44 
45 #include <linux/compiler.h>
46 #include <asm/asm-compat.h>
47 #include <asm/synch.h>
48 
49 /* PPC bit number conversion */
50 #define PPC_BITLSHIFT(be)	(BITS_PER_LONG - 1 - (be))
51 #define PPC_BIT(bit)		(1UL << PPC_BITLSHIFT(bit))
52 #define PPC_BITMASK(bs, be)	((PPC_BIT(bs) - PPC_BIT(be)) | PPC_BIT(bs))
53 
54 #include <asm/barrier.h>
55 
56 /* Macro for generating the ***_bits() functions */
57 #define DEFINE_BITOP(fn, op, prefix)		\
58 static __inline__ void fn(unsigned long mask,	\
59 		volatile unsigned long *_p)	\
60 {						\
61 	unsigned long old;			\
62 	unsigned long *p = (unsigned long *)_p;	\
63 	__asm__ __volatile__ (			\
64 	prefix					\
65 "1:"	PPC_LLARX(%0,0,%3,0) "\n"		\
66 	stringify_in_c(op) "%0,%0,%2\n"		\
67 	PPC405_ERR77(0,%3)			\
68 	PPC_STLCX "%0,0,%3\n"			\
69 	"bne- 1b\n"				\
70 	: "=&r" (old), "+m" (*p)		\
71 	: "r" (mask), "r" (p)			\
72 	: "cc", "memory");			\
73 }
74 
75 DEFINE_BITOP(set_bits, or, "")
76 DEFINE_BITOP(clear_bits, andc, "")
77 DEFINE_BITOP(clear_bits_unlock, andc, PPC_RELEASE_BARRIER)
78 DEFINE_BITOP(change_bits, xor, "")
79 
80 static __inline__ void set_bit(int nr, volatile unsigned long *addr)
81 {
82 	set_bits(BIT_MASK(nr), addr + BIT_WORD(nr));
83 }
84 
85 static __inline__ void clear_bit(int nr, volatile unsigned long *addr)
86 {
87 	clear_bits(BIT_MASK(nr), addr + BIT_WORD(nr));
88 }
89 
90 static __inline__ void clear_bit_unlock(int nr, volatile unsigned long *addr)
91 {
92 	clear_bits_unlock(BIT_MASK(nr), addr + BIT_WORD(nr));
93 }
94 
95 static __inline__ void change_bit(int nr, volatile unsigned long *addr)
96 {
97 	change_bits(BIT_MASK(nr), addr + BIT_WORD(nr));
98 }
99 
100 /* Like DEFINE_BITOP(), with changes to the arguments to 'op' and the output
101  * operands. */
102 #define DEFINE_TESTOP(fn, op, prefix, postfix, eh)	\
103 static __inline__ unsigned long fn(			\
104 		unsigned long mask,			\
105 		volatile unsigned long *_p)		\
106 {							\
107 	unsigned long old, t;				\
108 	unsigned long *p = (unsigned long *)_p;		\
109 	__asm__ __volatile__ (				\
110 	prefix						\
111 "1:"	PPC_LLARX(%0,0,%3,eh) "\n"			\
112 	stringify_in_c(op) "%1,%0,%2\n"			\
113 	PPC405_ERR77(0,%3)				\
114 	PPC_STLCX "%1,0,%3\n"				\
115 	"bne- 1b\n"					\
116 	postfix						\
117 	: "=&r" (old), "=&r" (t)			\
118 	: "r" (mask), "r" (p)				\
119 	: "cc", "memory");				\
120 	return (old & mask);				\
121 }
122 
123 DEFINE_TESTOP(test_and_set_bits, or, PPC_ATOMIC_ENTRY_BARRIER,
124 	      PPC_ATOMIC_EXIT_BARRIER, 0)
125 DEFINE_TESTOP(test_and_set_bits_lock, or, "",
126 	      PPC_ACQUIRE_BARRIER, 1)
127 DEFINE_TESTOP(test_and_clear_bits, andc, PPC_ATOMIC_ENTRY_BARRIER,
128 	      PPC_ATOMIC_EXIT_BARRIER, 0)
129 DEFINE_TESTOP(test_and_change_bits, xor, PPC_ATOMIC_ENTRY_BARRIER,
130 	      PPC_ATOMIC_EXIT_BARRIER, 0)
131 
132 static __inline__ int test_and_set_bit(unsigned long nr,
133 				       volatile unsigned long *addr)
134 {
135 	return test_and_set_bits(BIT_MASK(nr), addr + BIT_WORD(nr)) != 0;
136 }
137 
138 static __inline__ int test_and_set_bit_lock(unsigned long nr,
139 				       volatile unsigned long *addr)
140 {
141 	return test_and_set_bits_lock(BIT_MASK(nr),
142 				addr + BIT_WORD(nr)) != 0;
143 }
144 
145 static __inline__ int test_and_clear_bit(unsigned long nr,
146 					 volatile unsigned long *addr)
147 {
148 	return test_and_clear_bits(BIT_MASK(nr), addr + BIT_WORD(nr)) != 0;
149 }
150 
151 static __inline__ int test_and_change_bit(unsigned long nr,
152 					  volatile unsigned long *addr)
153 {
154 	return test_and_change_bits(BIT_MASK(nr), addr + BIT_WORD(nr)) != 0;
155 }
156 
157 #include <asm-generic/bitops/non-atomic.h>
158 
159 static __inline__ void __clear_bit_unlock(int nr, volatile unsigned long *addr)
160 {
161 	__asm__ __volatile__(PPC_RELEASE_BARRIER "" ::: "memory");
162 	__clear_bit(nr, addr);
163 }
164 
165 /*
166  * Return the zero-based bit position (LE, not IBM bit numbering) of
167  * the most significant 1-bit in a double word.
168  */
169 static __inline__ __attribute__((const))
170 int __ilog2(unsigned long x)
171 {
172 	int lz;
173 
174 	asm (PPC_CNTLZL "%0,%1" : "=r" (lz) : "r" (x));
175 	return BITS_PER_LONG - 1 - lz;
176 }
177 
178 static inline __attribute__((const))
179 int __ilog2_u32(u32 n)
180 {
181 	int bit;
182 	asm ("cntlzw %0,%1" : "=r" (bit) : "r" (n));
183 	return 31 - bit;
184 }
185 
186 #ifdef __powerpc64__
187 static inline __attribute__((const))
188 int __ilog2_u64(u64 n)
189 {
190 	int bit;
191 	asm ("cntlzd %0,%1" : "=r" (bit) : "r" (n));
192 	return 63 - bit;
193 }
194 #endif
195 
196 /*
197  * Determines the bit position of the least significant 0 bit in the
198  * specified double word. The returned bit position will be
199  * zero-based, starting from the right side (63/31 - 0).
200  */
201 static __inline__ unsigned long ffz(unsigned long x)
202 {
203 	/* no zero exists anywhere in the 8 byte area. */
204 	if ((x = ~x) == 0)
205 		return BITS_PER_LONG;
206 
207 	/*
208 	 * Calculate the bit position of the least significant '1' bit in x
209 	 * (since x has been changed this will actually be the least significant
210 	 * '0' bit in * the original x).  Note: (x & -x) gives us a mask that
211 	 * is the least significant * (RIGHT-most) 1-bit of the value in x.
212 	 */
213 	return __ilog2(x & -x);
214 }
215 
216 static __inline__ unsigned long __ffs(unsigned long x)
217 {
218 	return __ilog2(x & -x);
219 }
220 
221 /*
222  * ffs: find first bit set. This is defined the same way as
223  * the libc and compiler builtin ffs routines, therefore
224  * differs in spirit from the above ffz (man ffs).
225  */
226 static __inline__ int ffs(int x)
227 {
228 	unsigned long i = (unsigned long)x;
229 	return __ilog2(i & -i) + 1;
230 }
231 
232 /*
233  * fls: find last (most-significant) bit set.
234  * Note fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32.
235  */
236 static __inline__ int fls(unsigned int x)
237 {
238 	int lz;
239 
240 	asm ("cntlzw %0,%1" : "=r" (lz) : "r" (x));
241 	return 32 - lz;
242 }
243 
244 static __inline__ unsigned long __fls(unsigned long x)
245 {
246 	return __ilog2(x);
247 }
248 
249 /*
250  * 64-bit can do this using one cntlzd (count leading zeroes doubleword)
251  * instruction; for 32-bit we use the generic version, which does two
252  * 32-bit fls calls.
253  */
254 #ifdef __powerpc64__
255 static __inline__ int fls64(__u64 x)
256 {
257 	int lz;
258 
259 	asm ("cntlzd %0,%1" : "=r" (lz) : "r" (x));
260 	return 64 - lz;
261 }
262 #else
263 #include <asm-generic/bitops/fls64.h>
264 #endif /* __powerpc64__ */
265 
266 #ifdef CONFIG_PPC64
267 unsigned int __arch_hweight8(unsigned int w);
268 unsigned int __arch_hweight16(unsigned int w);
269 unsigned int __arch_hweight32(unsigned int w);
270 unsigned long __arch_hweight64(__u64 w);
271 #include <asm-generic/bitops/const_hweight.h>
272 #else
273 #include <asm-generic/bitops/hweight.h>
274 #endif
275 
276 #include <asm-generic/bitops/find.h>
277 
278 /* Little-endian versions */
279 #include <asm-generic/bitops/le.h>
280 
281 /* Bitmap functions for the ext2 filesystem */
282 
283 #include <asm-generic/bitops/ext2-atomic-setbit.h>
284 
285 #include <asm-generic/bitops/sched.h>
286 
287 #endif /* __KERNEL__ */
288 
289 #endif /* _ASM_POWERPC_BITOPS_H */
290