xref: /linux/arch/hexagon/include/asm/bitops.h (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * Bit operations for the Hexagon architecture
3  *
4  * Copyright (c) 2010-2011, The Linux Foundation. All rights reserved.
5  *
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 and
9  * only version 2 as published by the Free Software Foundation.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19  * 02110-1301, USA.
20  */
21 
22 #ifndef _ASM_BITOPS_H
23 #define _ASM_BITOPS_H
24 
25 #include <linux/compiler.h>
26 #include <asm/byteorder.h>
27 #include <asm/atomic.h>
28 #include <asm/barrier.h>
29 
30 #ifdef __KERNEL__
31 
32 /*
33  * The offset calculations for these are based on BITS_PER_LONG == 32
34  * (i.e. I get to shift by #5-2 (32 bits per long, 4 bytes per access),
35  * mask by 0x0000001F)
36  *
37  * Typically, R10 is clobbered for address, R11 bit nr, and R12 is temp
38  */
39 
40 /**
41  * test_and_clear_bit - clear a bit and return its old value
42  * @nr:  bit number to clear
43  * @addr:  pointer to memory
44  */
45 static inline int test_and_clear_bit(int nr, volatile void *addr)
46 {
47 	int oldval;
48 
49 	__asm__ __volatile__ (
50 	"	{R10 = %1; R11 = asr(%2,#5); }\n"
51 	"	{R10 += asl(R11,#2); R11 = and(%2,#0x1f)}\n"
52 	"1:	R12 = memw_locked(R10);\n"
53 	"	{ P0 = tstbit(R12,R11); R12 = clrbit(R12,R11); }\n"
54 	"	memw_locked(R10,P1) = R12;\n"
55 	"	{if !P1 jump 1b; %0 = mux(P0,#1,#0);}\n"
56 	: "=&r" (oldval)
57 	: "r" (addr), "r" (nr)
58 	: "r10", "r11", "r12", "p0", "p1", "memory"
59 	);
60 
61 	return oldval;
62 }
63 
64 /**
65  * test_and_set_bit - set a bit and return its old value
66  * @nr:  bit number to set
67  * @addr:  pointer to memory
68  */
69 static inline int test_and_set_bit(int nr, volatile void *addr)
70 {
71 	int oldval;
72 
73 	__asm__ __volatile__ (
74 	"	{R10 = %1; R11 = asr(%2,#5); }\n"
75 	"	{R10 += asl(R11,#2); R11 = and(%2,#0x1f)}\n"
76 	"1:	R12 = memw_locked(R10);\n"
77 	"	{ P0 = tstbit(R12,R11); R12 = setbit(R12,R11); }\n"
78 	"	memw_locked(R10,P1) = R12;\n"
79 	"	{if !P1 jump 1b; %0 = mux(P0,#1,#0);}\n"
80 	: "=&r" (oldval)
81 	: "r" (addr), "r" (nr)
82 	: "r10", "r11", "r12", "p0", "p1", "memory"
83 	);
84 
85 
86 	return oldval;
87 
88 }
89 
90 /**
91  * test_and_change_bit - toggle a bit and return its old value
92  * @nr:  bit number to set
93  * @addr:  pointer to memory
94  */
95 static inline int test_and_change_bit(int nr, volatile void *addr)
96 {
97 	int oldval;
98 
99 	__asm__ __volatile__ (
100 	"	{R10 = %1; R11 = asr(%2,#5); }\n"
101 	"	{R10 += asl(R11,#2); R11 = and(%2,#0x1f)}\n"
102 	"1:	R12 = memw_locked(R10);\n"
103 	"	{ P0 = tstbit(R12,R11); R12 = togglebit(R12,R11); }\n"
104 	"	memw_locked(R10,P1) = R12;\n"
105 	"	{if !P1 jump 1b; %0 = mux(P0,#1,#0);}\n"
106 	: "=&r" (oldval)
107 	: "r" (addr), "r" (nr)
108 	: "r10", "r11", "r12", "p0", "p1", "memory"
109 	);
110 
111 	return oldval;
112 
113 }
114 
115 /*
116  * Atomic, but doesn't care about the return value.
117  * Rewrite later to save a cycle or two.
118  */
119 
120 static inline void clear_bit(int nr, volatile void *addr)
121 {
122 	test_and_clear_bit(nr, addr);
123 }
124 
125 static inline void set_bit(int nr, volatile void *addr)
126 {
127 	test_and_set_bit(nr, addr);
128 }
129 
130 static inline void change_bit(int nr, volatile void *addr)
131 {
132 	test_and_change_bit(nr, addr);
133 }
134 
135 
136 /*
137  * These are allowed to be non-atomic.  In fact the generic flavors are
138  * in non-atomic.h.  Would it be better to use intrinsics for this?
139  *
140  * OK, writes in our architecture do not invalidate LL/SC, so this has to
141  * be atomic, particularly for things like slab_lock and slab_unlock.
142  *
143  */
144 static inline void __clear_bit(int nr, volatile unsigned long *addr)
145 {
146 	test_and_clear_bit(nr, addr);
147 }
148 
149 static inline void __set_bit(int nr, volatile unsigned long *addr)
150 {
151 	test_and_set_bit(nr, addr);
152 }
153 
154 static inline void __change_bit(int nr, volatile unsigned long *addr)
155 {
156 	test_and_change_bit(nr, addr);
157 }
158 
159 /*  Apparently, at least some of these are allowed to be non-atomic  */
160 static inline int __test_and_clear_bit(int nr, volatile unsigned long *addr)
161 {
162 	return test_and_clear_bit(nr, addr);
163 }
164 
165 static inline int __test_and_set_bit(int nr, volatile unsigned long *addr)
166 {
167 	return test_and_set_bit(nr, addr);
168 }
169 
170 static inline int __test_and_change_bit(int nr, volatile unsigned long *addr)
171 {
172 	return test_and_change_bit(nr, addr);
173 }
174 
175 static inline int __test_bit(int nr, const volatile unsigned long *addr)
176 {
177 	int retval;
178 
179 	asm volatile(
180 	"{P0 = tstbit(%1,%2); if (P0.new) %0 = #1; if (!P0.new) %0 = #0;}\n"
181 	: "=&r" (retval)
182 	: "r" (addr[BIT_WORD(nr)]), "r" (nr % BITS_PER_LONG)
183 	: "p0"
184 	);
185 
186 	return retval;
187 }
188 
189 #define test_bit(nr, addr) __test_bit(nr, addr)
190 
191 /*
192  * ffz - find first zero in word.
193  * @word: The word to search
194  *
195  * Undefined if no zero exists, so code should check against ~0UL first.
196  */
197 static inline long ffz(int x)
198 {
199 	int r;
200 
201 	asm("%0 = ct1(%1);\n"
202 		: "=&r" (r)
203 		: "r" (x));
204 	return r;
205 }
206 
207 /*
208  * fls - find last (most-significant) bit set
209  * @x: the word to search
210  *
211  * This is defined the same way as ffs.
212  * Note fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32.
213  */
214 static inline long fls(int x)
215 {
216 	int r;
217 
218 	asm("{ %0 = cl0(%1);}\n"
219 		"%0 = sub(#32,%0);\n"
220 		: "=&r" (r)
221 		: "r" (x)
222 		: "p0");
223 
224 	return r;
225 }
226 
227 /*
228  * ffs - find first bit set
229  * @x: the word to search
230  *
231  * This is defined the same way as
232  * the libc and compiler builtin ffs routines, therefore
233  * differs in spirit from the above ffz (man ffs).
234  */
235 static inline long ffs(int x)
236 {
237 	int r;
238 
239 	asm("{ P0 = cmp.eq(%1,#0); %0 = ct0(%1);}\n"
240 		"{ if P0 %0 = #0; if !P0 %0 = add(%0,#1);}\n"
241 		: "=&r" (r)
242 		: "r" (x)
243 		: "p0");
244 
245 	return r;
246 }
247 
248 /*
249  * __ffs - find first bit in word.
250  * @word: The word to search
251  *
252  * Undefined if no bit exists, so code should check against 0 first.
253  *
254  * bits_per_long assumed to be 32
255  * numbering starts at 0 I think (instead of 1 like ffs)
256  */
257 static inline unsigned long __ffs(unsigned long word)
258 {
259 	int num;
260 
261 	asm("%0 = ct0(%1);\n"
262 		: "=&r" (num)
263 		: "r" (word));
264 
265 	return num;
266 }
267 
268 /*
269  * __fls - find last (most-significant) set bit in a long word
270  * @word: the word to search
271  *
272  * Undefined if no set bit exists, so code should check against 0 first.
273  * bits_per_long assumed to be 32
274  */
275 static inline unsigned long __fls(unsigned long word)
276 {
277 	int num;
278 
279 	asm("%0 = cl0(%1);\n"
280 		"%0 = sub(#31,%0);\n"
281 		: "=&r" (num)
282 		: "r" (word));
283 
284 	return num;
285 }
286 
287 #include <asm-generic/bitops/lock.h>
288 #include <asm-generic/bitops/find.h>
289 
290 #include <asm-generic/bitops/fls64.h>
291 #include <asm-generic/bitops/sched.h>
292 #include <asm-generic/bitops/hweight.h>
293 
294 #include <asm-generic/bitops/le.h>
295 #include <asm-generic/bitops/ext2-atomic.h>
296 
297 #endif /* __KERNEL__ */
298 #endif
299