1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3 * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
4 */
5
6 #ifndef _ASM_BITOPS_H
7 #define _ASM_BITOPS_H
8
9 #ifndef _LINUX_BITOPS_H
10 #error only <linux/bitops.h> can be included directly
11 #endif
12
13 #ifndef __ASSEMBLY__
14
15 #include <linux/types.h>
16 #include <linux/compiler.h>
17 #include <asm/barrier.h>
18 #ifndef CONFIG_ARC_HAS_LLSC
19 #include <asm/smp.h>
20 #endif
21
22 #ifdef CONFIG_ARC_HAS_LLSC
23
24 /*
25 * Hardware assisted Atomic-R-M-W
26 */
27
28 #define BIT_OP(op, c_op, asm_op) \
29 static inline void op##_bit(unsigned long nr, volatile unsigned long *m)\
30 { \
31 unsigned int temp; \
32 \
33 m += nr >> 5; \
34 \
35 nr &= 0x1f; \
36 \
37 __asm__ __volatile__( \
38 "1: llock %0, [%1] \n" \
39 " " #asm_op " %0, %0, %2 \n" \
40 " scond %0, [%1] \n" \
41 " bnz 1b \n" \
42 : "=&r"(temp) /* Early clobber, to prevent reg reuse */ \
43 : "r"(m), /* Not "m": llock only supports reg direct addr mode */ \
44 "ir"(nr) \
45 : "cc"); \
46 }
47
48 /*
49 * Semantically:
50 * Test the bit
51 * if clear
52 * set it and return 0 (old value)
53 * else
54 * return 1 (old value).
55 *
56 * Since ARC lacks a equivalent h/w primitive, the bit is set unconditionally
57 * and the old value of bit is returned
58 */
59 #define TEST_N_BIT_OP(op, c_op, asm_op) \
60 static inline int test_and_##op##_bit(unsigned long nr, volatile unsigned long *m)\
61 { \
62 unsigned long old, temp; \
63 \
64 m += nr >> 5; \
65 \
66 nr &= 0x1f; \
67 \
68 /* \
69 * Explicit full memory barrier needed before/after as \
70 * LLOCK/SCOND themselves don't provide any such smenatic \
71 */ \
72 smp_mb(); \
73 \
74 __asm__ __volatile__( \
75 "1: llock %0, [%2] \n" \
76 " " #asm_op " %1, %0, %3 \n" \
77 " scond %1, [%2] \n" \
78 " bnz 1b \n" \
79 : "=&r"(old), "=&r"(temp) \
80 : "r"(m), "ir"(nr) \
81 : "cc"); \
82 \
83 smp_mb(); \
84 \
85 return (old & (1 << nr)) != 0; \
86 }
87
88 #else /* !CONFIG_ARC_HAS_LLSC */
89
90 /*
91 * Non hardware assisted Atomic-R-M-W
92 * Locking would change to irq-disabling only (UP) and spinlocks (SMP)
93 *
94 * There's "significant" micro-optimization in writing our own variants of
95 * bitops (over generic variants)
96 *
97 * (1) The generic APIs have "signed" @nr while we have it "unsigned"
98 * This avoids extra code to be generated for pointer arithmatic, since
99 * is "not sure" that index is NOT -ve
100 * (2) Utilize the fact that ARCompact bit fidding insn (BSET/BCLR/ASL) etc
101 * only consider bottom 5 bits of @nr, so NO need to mask them off.
102 * (GCC Quirk: however for constant @nr we still need to do the masking
103 * at compile time)
104 */
105
106 #define BIT_OP(op, c_op, asm_op) \
107 static inline void op##_bit(unsigned long nr, volatile unsigned long *m)\
108 { \
109 unsigned long temp, flags; \
110 m += nr >> 5; \
111 \
112 /* \
113 * spin lock/unlock provide the needed smp_mb() before/after \
114 */ \
115 bitops_lock(flags); \
116 \
117 temp = *m; \
118 *m = temp c_op (1UL << (nr & 0x1f)); \
119 \
120 bitops_unlock(flags); \
121 }
122
123 #define TEST_N_BIT_OP(op, c_op, asm_op) \
124 static inline int test_and_##op##_bit(unsigned long nr, volatile unsigned long *m)\
125 { \
126 unsigned long old, flags; \
127 m += nr >> 5; \
128 \
129 bitops_lock(flags); \
130 \
131 old = *m; \
132 *m = old c_op (1UL << (nr & 0x1f)); \
133 \
134 bitops_unlock(flags); \
135 \
136 return (old & (1UL << (nr & 0x1f))) != 0; \
137 }
138
139 #endif
140
141 /***************************************
142 * Non atomic variants
143 **************************************/
144
145 #define __BIT_OP(op, c_op, asm_op) \
146 static inline void __##op##_bit(unsigned long nr, volatile unsigned long *m) \
147 { \
148 unsigned long temp; \
149 m += nr >> 5; \
150 \
151 temp = *m; \
152 *m = temp c_op (1UL << (nr & 0x1f)); \
153 }
154
155 #define __TEST_N_BIT_OP(op, c_op, asm_op) \
156 static inline int __test_and_##op##_bit(unsigned long nr, volatile unsigned long *m)\
157 { \
158 unsigned long old; \
159 m += nr >> 5; \
160 \
161 old = *m; \
162 *m = old c_op (1UL << (nr & 0x1f)); \
163 \
164 return (old & (1UL << (nr & 0x1f))) != 0; \
165 }
166
167 #define BIT_OPS(op, c_op, asm_op) \
168 \
169 /* set_bit(), clear_bit(), change_bit() */ \
170 BIT_OP(op, c_op, asm_op) \
171 \
172 /* test_and_set_bit(), test_and_clear_bit(), test_and_change_bit() */\
173 TEST_N_BIT_OP(op, c_op, asm_op) \
174 \
175 /* __set_bit(), __clear_bit(), __change_bit() */ \
176 __BIT_OP(op, c_op, asm_op) \
177 \
178 /* __test_and_set_bit(), __test_and_clear_bit(), __test_and_change_bit() */\
179 __TEST_N_BIT_OP(op, c_op, asm_op)
180
181 BIT_OPS(set, |, bset)
182 BIT_OPS(clear, & ~, bclr)
183 BIT_OPS(change, ^, bxor)
184
185 /*
186 * This routine doesn't need to be atomic.
187 */
188 static inline int
test_bit(unsigned int nr,const volatile unsigned long * addr)189 test_bit(unsigned int nr, const volatile unsigned long *addr)
190 {
191 unsigned long mask;
192
193 addr += nr >> 5;
194
195 mask = 1UL << (nr & 0x1f);
196
197 return ((mask & *addr) != 0);
198 }
199
200 #ifdef CONFIG_ISA_ARCOMPACT
201
202 /*
203 * Count the number of zeros, starting from MSB
204 * Helper for fls( ) friends
205 * This is a pure count, so (1-32) or (0-31) doesn't apply
206 * It could be 0 to 32, based on num of 0's in there
207 * clz(0x8000_0000) = 0, clz(0xFFFF_FFFF)=0, clz(0) = 32, clz(1) = 31
208 */
clz(unsigned int x)209 static inline __attribute__ ((const)) int clz(unsigned int x)
210 {
211 unsigned int res;
212
213 __asm__ __volatile__(
214 " norm.f %0, %1 \n"
215 " mov.n %0, 0 \n"
216 " add.p %0, %0, 1 \n"
217 : "=r"(res)
218 : "r"(x)
219 : "cc");
220
221 return res;
222 }
223
constant_fls(unsigned int x)224 static inline int constant_fls(unsigned int x)
225 {
226 int r = 32;
227
228 if (!x)
229 return 0;
230 if (!(x & 0xffff0000u)) {
231 x <<= 16;
232 r -= 16;
233 }
234 if (!(x & 0xff000000u)) {
235 x <<= 8;
236 r -= 8;
237 }
238 if (!(x & 0xf0000000u)) {
239 x <<= 4;
240 r -= 4;
241 }
242 if (!(x & 0xc0000000u)) {
243 x <<= 2;
244 r -= 2;
245 }
246 if (!(x & 0x80000000u))
247 r -= 1;
248 return r;
249 }
250
251 /*
252 * fls = Find Last Set in word
253 * @result: [1-32]
254 * fls(1) = 1, fls(0x80000000) = 32, fls(0) = 0
255 */
fls(unsigned int x)256 static inline __attribute__ ((const)) int fls(unsigned int x)
257 {
258 if (__builtin_constant_p(x))
259 return constant_fls(x);
260
261 return 32 - clz(x);
262 }
263
264 /*
265 * __fls: Similar to fls, but zero based (0-31)
266 */
__fls(unsigned long x)267 static inline __attribute__ ((const)) int __fls(unsigned long x)
268 {
269 if (!x)
270 return 0;
271 else
272 return fls(x) - 1;
273 }
274
275 /*
276 * ffs = Find First Set in word (LSB to MSB)
277 * @result: [1-32], 0 if all 0's
278 */
279 #define ffs(x) ({ unsigned long __t = (x); fls(__t & -__t); })
280
281 /*
282 * __ffs: Similar to ffs, but zero based (0-31)
283 */
__ffs(unsigned long word)284 static inline __attribute__ ((const)) unsigned long __ffs(unsigned long word)
285 {
286 if (!word)
287 return word;
288
289 return ffs(word) - 1;
290 }
291
292 #else /* CONFIG_ISA_ARCV2 */
293
294 /*
295 * fls = Find Last Set in word
296 * @result: [1-32]
297 * fls(1) = 1, fls(0x80000000) = 32, fls(0) = 0
298 */
fls(unsigned long x)299 static inline __attribute__ ((const)) int fls(unsigned long x)
300 {
301 int n;
302
303 asm volatile(
304 " fls.f %0, %1 \n" /* 0:31; 0(Z) if src 0 */
305 " add.nz %0, %0, 1 \n" /* 0:31 -> 1:32 */
306 : "=r"(n) /* Early clobber not needed */
307 : "r"(x)
308 : "cc");
309
310 return n;
311 }
312
313 /*
314 * __fls: Similar to fls, but zero based (0-31). Also 0 if no bit set
315 */
__fls(unsigned long x)316 static inline __attribute__ ((const)) int __fls(unsigned long x)
317 {
318 /* FLS insn has exactly same semantics as the API */
319 return __builtin_arc_fls(x);
320 }
321
322 /*
323 * ffs = Find First Set in word (LSB to MSB)
324 * @result: [1-32], 0 if all 0's
325 */
ffs(unsigned long x)326 static inline __attribute__ ((const)) int ffs(unsigned long x)
327 {
328 int n;
329
330 asm volatile(
331 " ffs.f %0, %1 \n" /* 0:31; 31(Z) if src 0 */
332 " add.nz %0, %0, 1 \n" /* 0:31 -> 1:32 */
333 " mov.z %0, 0 \n" /* 31(Z)-> 0 */
334 : "=r"(n) /* Early clobber not needed */
335 : "r"(x)
336 : "cc");
337
338 return n;
339 }
340
341 /*
342 * __ffs: Similar to ffs, but zero based (0-31)
343 */
__ffs(unsigned long x)344 static inline __attribute__ ((const)) unsigned long __ffs(unsigned long x)
345 {
346 unsigned long n;
347
348 asm volatile(
349 " ffs.f %0, %1 \n" /* 0:31; 31(Z) if src 0 */
350 " mov.z %0, 0 \n" /* 31(Z)-> 0 */
351 : "=r"(n)
352 : "r"(x)
353 : "cc");
354
355 return n;
356
357 }
358
359 #endif /* CONFIG_ISA_ARCOMPACT */
360
361 /*
362 * ffz = Find First Zero in word.
363 * @return:[0-31], 32 if all 1's
364 */
365 #define ffz(x) __ffs(~(x))
366
367 #include <asm-generic/bitops/hweight.h>
368 #include <asm-generic/bitops/fls64.h>
369 #include <asm-generic/bitops/sched.h>
370 #include <asm-generic/bitops/lock.h>
371
372 #include <asm-generic/bitops/find.h>
373 #include <asm-generic/bitops/le.h>
374 #include <asm-generic/bitops/ext2-atomic-setbit.h>
375
376 #endif /* !__ASSEMBLY__ */
377
378 #endif
379
