1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _PARISC_BITOPS_H
3#define _PARISC_BITOPS_H
4
5#ifndef _LINUX_BITOPS_H
6#error only <linux/bitops.h> can be included directly
7#endif
8
9#include <linux/compiler.h>
10#include <asm/types.h>
11#include <asm/byteorder.h>
12#include <asm/barrier.h>
13#include <linux/atomic.h>
14
15/* See http://marc.theaimsgroup.com/?t=108826637900003 for discussion
16 * on use of volatile and __*_bit() (set/clear/change):
17 *	*_bit() want use of volatile.
18 *	__*_bit() are "relaxed" and don't use spinlock or volatile.
19 */
20
21static __inline__ void set_bit(int nr, volatile unsigned long * addr)
22{
23	unsigned long mask = BIT_MASK(nr);
24	unsigned long flags;
25
26	addr += BIT_WORD(nr);
27	_atomic_spin_lock_irqsave(addr, flags);
28	*addr |= mask;
29	_atomic_spin_unlock_irqrestore(addr, flags);
30}
31
32static __inline__ void clear_bit(int nr, volatile unsigned long * addr)
33{
34	unsigned long mask = BIT_MASK(nr);
35	unsigned long flags;
36
37	addr += BIT_WORD(nr);
38	_atomic_spin_lock_irqsave(addr, flags);
39	*addr &= ~mask;
40	_atomic_spin_unlock_irqrestore(addr, flags);
41}
42
43static __inline__ void change_bit(int nr, volatile unsigned long * addr)
44{
45	unsigned long mask = BIT_MASK(nr);
46	unsigned long flags;
47
48	addr += BIT_WORD(nr);
49	_atomic_spin_lock_irqsave(addr, flags);
50	*addr ^= mask;
51	_atomic_spin_unlock_irqrestore(addr, flags);
52}
53
54static __inline__ int test_and_set_bit(int nr, volatile unsigned long * addr)
55{
56	unsigned long mask = BIT_MASK(nr);
57	unsigned long old;
58	unsigned long flags;
59	int set;
60
61	addr += BIT_WORD(nr);
62	_atomic_spin_lock_irqsave(addr, flags);
63	old = *addr;
64	set = (old & mask) ? 1 : 0;
65	if (!set)
66		*addr = old | mask;
67	_atomic_spin_unlock_irqrestore(addr, flags);
68
69	return set;
70}
71
72static __inline__ int test_and_clear_bit(int nr, volatile unsigned long * addr)
73{
74	unsigned long mask = BIT_MASK(nr);
75	unsigned long old;
76	unsigned long flags;
77	int set;
78
79	addr += BIT_WORD(nr);
80	_atomic_spin_lock_irqsave(addr, flags);
81	old = *addr;
82	set = (old & mask) ? 1 : 0;
83	if (set)
84		*addr = old & ~mask;
85	_atomic_spin_unlock_irqrestore(addr, flags);
86
87	return set;
88}
89
90static __inline__ int test_and_change_bit(int nr, volatile unsigned long * addr)
91{
92	unsigned long mask = BIT_MASK(nr);
93	unsigned long oldbit;
94	unsigned long flags;
95
96	addr += BIT_WORD(nr);
97	_atomic_spin_lock_irqsave(addr, flags);
98	oldbit = *addr;
99	*addr = oldbit ^ mask;
100	_atomic_spin_unlock_irqrestore(addr, flags);
101
102	return (oldbit & mask) ? 1 : 0;
103}
104
105#include <asm-generic/bitops/non-atomic.h>
106
107/**
108 * __ffs - find first bit in word. returns 0 to "BITS_PER_LONG-1".
109 * @word: The word to search
110 *
111 * __ffs() return is undefined if no bit is set.
112 *
113 * 32-bit fast __ffs by LaMont Jones "lamont At hp com".
114 * 64-bit enhancement by Grant Grundler "grundler At parisc-linux org".
115 * (with help from willy/jejb to get the semantics right)
116 *
117 * This algorithm avoids branches by making use of nullification.
118 * One side effect of "extr" instructions is it sets PSW[N] bit.
119 * How PSW[N] (nullify next insn) gets set is determined by the
120 * "condition" field (eg "<>" or "TR" below) in the extr* insn.
121 * Only the 1st and one of either the 2cd or 3rd insn will get executed.
122 * Each set of 3 insn will get executed in 2 cycles on PA8x00 vs 16 or so
123 * cycles for each mispredicted branch.
124 */
125
126static __inline__ unsigned long __ffs(unsigned long x)
127{
128	unsigned long ret;
129
130	__asm__(
131#ifdef CONFIG_64BIT
132		" ldi       63,%1\n"
133		" extrd,u,*<>  %0,63,32,%%r0\n"
134		" extrd,u,*TR  %0,31,32,%0\n"	/* move top 32-bits down */
135		" addi    -32,%1,%1\n"
136#else
137		" ldi       31,%1\n"
138#endif
139		" extru,<>  %0,31,16,%%r0\n"
140		" extru,TR  %0,15,16,%0\n"	/* xxxx0000 -> 0000xxxx */
141		" addi    -16,%1,%1\n"
142		" extru,<>  %0,31,8,%%r0\n"
143		" extru,TR  %0,23,8,%0\n"	/* 0000xx00 -> 000000xx */
144		" addi    -8,%1,%1\n"
145		" extru,<>  %0,31,4,%%r0\n"
146		" extru,TR  %0,27,4,%0\n"	/* 000000x0 -> 0000000x */
147		" addi    -4,%1,%1\n"
148		" extru,<>  %0,31,2,%%r0\n"
149		" extru,TR  %0,29,2,%0\n"	/* 0000000y, 1100b -> 0011b */
150		" addi    -2,%1,%1\n"
151		" extru,=  %0,31,1,%%r0\n"	/* check last bit */
152		" addi    -1,%1,%1\n"
153			: "+r" (x), "=r" (ret) );
154	return ret;
155}
156
157#include <asm-generic/bitops/ffz.h>
158
159/*
160 * ffs: find first bit set. returns 1 to BITS_PER_LONG or 0 (if none set)
161 * This is defined the same way as the libc and compiler builtin
162 * ffs routines, therefore differs in spirit from the above ffz (man ffs).
163 */
164static __inline__ int ffs(int x)
165{
166	return x ? (__ffs((unsigned long)x) + 1) : 0;
167}
168
169/*
170 * fls: find last (most significant) bit set.
171 * fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32.
172 */
173
174static __inline__ int fls(unsigned int x)
175{
176	int ret;
177	if (!x)
178		return 0;
179
180	__asm__(
181	"	ldi		1,%1\n"
182	"	extru,<>	%0,15,16,%%r0\n"
183	"	zdep,TR		%0,15,16,%0\n"		/* xxxx0000 */
184	"	addi		16,%1,%1\n"
185	"	extru,<>	%0,7,8,%%r0\n"
186	"	zdep,TR		%0,23,24,%0\n"		/* xx000000 */
187	"	addi		8,%1,%1\n"
188	"	extru,<>	%0,3,4,%%r0\n"
189	"	zdep,TR		%0,27,28,%0\n"		/* x0000000 */
190	"	addi		4,%1,%1\n"
191	"	extru,<>	%0,1,2,%%r0\n"
192	"	zdep,TR		%0,29,30,%0\n"		/* y0000000 (y&3 = 0) */
193	"	addi		2,%1,%1\n"
194	"	extru,=		%0,0,1,%%r0\n"
195	"	addi		1,%1,%1\n"		/* if y & 8, add 1 */
196		: "+r" (x), "=r" (ret) );
197
198	return ret;
199}
200
201#include <asm-generic/bitops/__fls.h>
202#include <asm-generic/bitops/fls64.h>
203#include <asm-generic/bitops/hweight.h>
204#include <asm-generic/bitops/lock.h>
205#include <asm-generic/bitops/sched.h>
206#include <asm-generic/bitops/le.h>
207#include <asm-generic/bitops/ext2-atomic-setbit.h>
208
209#endif /* _PARISC_BITOPS_H */
210