xref: /openbmc/linux/arch/arc/include/asm/bitops.h (revision 22d55f02)
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 #elif !defined(CONFIG_ARC_PLAT_EZNPS)
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 #else /* CONFIG_ARC_PLAT_EZNPS */
140 
141 #define BIT_OP(op, c_op, asm_op)					\
142 static inline void op##_bit(unsigned long nr, volatile unsigned long *m)\
143 {									\
144 	m += nr >> 5;							\
145 									\
146 	nr = (1UL << (nr & 0x1f));					\
147 	if (asm_op == CTOP_INST_AAND_DI_R2_R2_R3)			\
148 		nr = ~nr;						\
149 									\
150 	__asm__ __volatile__(						\
151 	"	mov r2, %0\n"						\
152 	"	mov r3, %1\n"						\
153 	"	.word %2\n"						\
154 	:								\
155 	: "r"(nr), "r"(m), "i"(asm_op)					\
156 	: "r2", "r3", "memory");					\
157 }
158 
159 #define TEST_N_BIT_OP(op, c_op, asm_op)					\
160 static inline int test_and_##op##_bit(unsigned long nr, volatile unsigned long *m)\
161 {									\
162 	unsigned long old;						\
163 									\
164 	m += nr >> 5;							\
165 									\
166 	nr = old = (1UL << (nr & 0x1f));				\
167 	if (asm_op == CTOP_INST_AAND_DI_R2_R2_R3)			\
168 		old = ~old;						\
169 									\
170 	/* Explicit full memory barrier needed before/after */		\
171 	smp_mb();							\
172 									\
173 	__asm__ __volatile__(						\
174 	"	mov r2, %0\n"						\
175 	"	mov r3, %1\n"						\
176 	"       .word %2\n"						\
177 	"	mov %0, r2"						\
178 	: "+r"(old)							\
179 	: "r"(m), "i"(asm_op)						\
180 	: "r2", "r3", "memory");					\
181 									\
182 	smp_mb();							\
183 									\
184 	return (old & nr) != 0;					\
185 }
186 
187 #endif /* CONFIG_ARC_PLAT_EZNPS */
188 
189 /***************************************
190  * Non atomic variants
191  **************************************/
192 
193 #define __BIT_OP(op, c_op, asm_op)					\
194 static inline void __##op##_bit(unsigned long nr, volatile unsigned long *m)	\
195 {									\
196 	unsigned long temp;						\
197 	m += nr >> 5;							\
198 									\
199 	temp = *m;							\
200 	*m = temp c_op (1UL << (nr & 0x1f));				\
201 }
202 
203 #define __TEST_N_BIT_OP(op, c_op, asm_op)				\
204 static inline int __test_and_##op##_bit(unsigned long nr, volatile unsigned long *m)\
205 {									\
206 	unsigned long old;						\
207 	m += nr >> 5;							\
208 									\
209 	old = *m;							\
210 	*m = old c_op (1UL << (nr & 0x1f));				\
211 									\
212 	return (old & (1UL << (nr & 0x1f))) != 0;			\
213 }
214 
215 #define BIT_OPS(op, c_op, asm_op)					\
216 									\
217 	/* set_bit(), clear_bit(), change_bit() */			\
218 	BIT_OP(op, c_op, asm_op)					\
219 									\
220 	/* test_and_set_bit(), test_and_clear_bit(), test_and_change_bit() */\
221 	TEST_N_BIT_OP(op, c_op, asm_op)					\
222 									\
223 	/* __set_bit(), __clear_bit(), __change_bit() */		\
224 	__BIT_OP(op, c_op, asm_op)					\
225 									\
226 	/* __test_and_set_bit(), __test_and_clear_bit(), __test_and_change_bit() */\
227 	__TEST_N_BIT_OP(op, c_op, asm_op)
228 
229 #ifndef CONFIG_ARC_PLAT_EZNPS
230 BIT_OPS(set, |, bset)
231 BIT_OPS(clear, & ~, bclr)
232 BIT_OPS(change, ^, bxor)
233 #else
234 BIT_OPS(set, |, CTOP_INST_AOR_DI_R2_R2_R3)
235 BIT_OPS(clear, & ~, CTOP_INST_AAND_DI_R2_R2_R3)
236 BIT_OPS(change, ^, CTOP_INST_AXOR_DI_R2_R2_R3)
237 #endif
238 
239 /*
240  * This routine doesn't need to be atomic.
241  */
242 static inline int
243 test_bit(unsigned int nr, const volatile unsigned long *addr)
244 {
245 	unsigned long mask;
246 
247 	addr += nr >> 5;
248 
249 	mask = 1UL << (nr & 0x1f);
250 
251 	return ((mask & *addr) != 0);
252 }
253 
254 #ifdef CONFIG_ISA_ARCOMPACT
255 
256 /*
257  * Count the number of zeros, starting from MSB
258  * Helper for fls( ) friends
259  * This is a pure count, so (1-32) or (0-31) doesn't apply
260  * It could be 0 to 32, based on num of 0's in there
261  * clz(0x8000_0000) = 0, clz(0xFFFF_FFFF)=0, clz(0) = 32, clz(1) = 31
262  */
263 static inline __attribute__ ((const)) int clz(unsigned int x)
264 {
265 	unsigned int res;
266 
267 	__asm__ __volatile__(
268 	"	norm.f  %0, %1		\n"
269 	"	mov.n   %0, 0		\n"
270 	"	add.p   %0, %0, 1	\n"
271 	: "=r"(res)
272 	: "r"(x)
273 	: "cc");
274 
275 	return res;
276 }
277 
278 static inline int constant_fls(unsigned int x)
279 {
280 	int r = 32;
281 
282 	if (!x)
283 		return 0;
284 	if (!(x & 0xffff0000u)) {
285 		x <<= 16;
286 		r -= 16;
287 	}
288 	if (!(x & 0xff000000u)) {
289 		x <<= 8;
290 		r -= 8;
291 	}
292 	if (!(x & 0xf0000000u)) {
293 		x <<= 4;
294 		r -= 4;
295 	}
296 	if (!(x & 0xc0000000u)) {
297 		x <<= 2;
298 		r -= 2;
299 	}
300 	if (!(x & 0x80000000u)) {
301 		x <<= 1;
302 		r -= 1;
303 	}
304 	return r;
305 }
306 
307 /*
308  * fls = Find Last Set in word
309  * @result: [1-32]
310  * fls(1) = 1, fls(0x80000000) = 32, fls(0) = 0
311  */
312 static inline __attribute__ ((const)) int fls(unsigned int x)
313 {
314 	if (__builtin_constant_p(x))
315 	       return constant_fls(x);
316 
317 	return 32 - clz(x);
318 }
319 
320 /*
321  * __fls: Similar to fls, but zero based (0-31)
322  */
323 static inline __attribute__ ((const)) int __fls(unsigned long x)
324 {
325 	if (!x)
326 		return 0;
327 	else
328 		return fls(x) - 1;
329 }
330 
331 /*
332  * ffs = Find First Set in word (LSB to MSB)
333  * @result: [1-32], 0 if all 0's
334  */
335 #define ffs(x)	({ unsigned long __t = (x); fls(__t & -__t); })
336 
337 /*
338  * __ffs: Similar to ffs, but zero based (0-31)
339  */
340 static inline __attribute__ ((const)) unsigned long __ffs(unsigned long word)
341 {
342 	if (!word)
343 		return word;
344 
345 	return ffs(word) - 1;
346 }
347 
348 #else	/* CONFIG_ISA_ARCV2 */
349 
350 /*
351  * fls = Find Last Set in word
352  * @result: [1-32]
353  * fls(1) = 1, fls(0x80000000) = 32, fls(0) = 0
354  */
355 static inline __attribute__ ((const)) int fls(unsigned long x)
356 {
357 	int n;
358 
359 	asm volatile(
360 	"	fls.f	%0, %1		\n"  /* 0:31; 0(Z) if src 0 */
361 	"	add.nz	%0, %0, 1	\n"  /* 0:31 -> 1:32 */
362 	: "=r"(n)	/* Early clobber not needed */
363 	: "r"(x)
364 	: "cc");
365 
366 	return n;
367 }
368 
369 /*
370  * __fls: Similar to fls, but zero based (0-31). Also 0 if no bit set
371  */
372 static inline __attribute__ ((const)) int __fls(unsigned long x)
373 {
374 	/* FLS insn has exactly same semantics as the API */
375 	return	__builtin_arc_fls(x);
376 }
377 
378 /*
379  * ffs = Find First Set in word (LSB to MSB)
380  * @result: [1-32], 0 if all 0's
381  */
382 static inline __attribute__ ((const)) int ffs(unsigned long x)
383 {
384 	int n;
385 
386 	asm volatile(
387 	"	ffs.f	%0, %1		\n"  /* 0:31; 31(Z) if src 0 */
388 	"	add.nz	%0, %0, 1	\n"  /* 0:31 -> 1:32 */
389 	"	mov.z	%0, 0		\n"  /* 31(Z)-> 0 */
390 	: "=r"(n)	/* Early clobber not needed */
391 	: "r"(x)
392 	: "cc");
393 
394 	return n;
395 }
396 
397 /*
398  * __ffs: Similar to ffs, but zero based (0-31)
399  */
400 static inline __attribute__ ((const)) unsigned long __ffs(unsigned long x)
401 {
402 	unsigned long n;
403 
404 	asm volatile(
405 	"	ffs.f	%0, %1		\n"  /* 0:31; 31(Z) if src 0 */
406 	"	mov.z	%0, 0		\n"  /* 31(Z)-> 0 */
407 	: "=r"(n)
408 	: "r"(x)
409 	: "cc");
410 
411 	return n;
412 
413 }
414 
415 #endif	/* CONFIG_ISA_ARCOMPACT */
416 
417 /*
418  * ffz = Find First Zero in word.
419  * @return:[0-31], 32 if all 1's
420  */
421 #define ffz(x)	__ffs(~(x))
422 
423 #include <asm-generic/bitops/hweight.h>
424 #include <asm-generic/bitops/fls64.h>
425 #include <asm-generic/bitops/sched.h>
426 #include <asm-generic/bitops/lock.h>
427 
428 #include <asm-generic/bitops/find.h>
429 #include <asm-generic/bitops/le.h>
430 #include <asm-generic/bitops/ext2-atomic-setbit.h>
431 
432 #endif /* !__ASSEMBLY__ */
433 
434 #endif
435