1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _ASM_X86_BITOPS_H
3#define _ASM_X86_BITOPS_H
4
5/*
6 * Copyright 1992, Linus Torvalds.
7 *
8 * Note: inlines with more than a single statement should be marked
9 * __always_inline to avoid problems with older gcc's inlining heuristics.
10 */
11
12#ifndef _LINUX_BITOPS_H
13#error only <linux/bitops.h> can be included directly
14#endif
15
16#include <linux/compiler.h>
17#include <asm/alternative.h>
18#include <asm/rmwcc.h>
19#include <asm/barrier.h>
20
21#if BITS_PER_LONG == 32
22# define _BITOPS_LONG_SHIFT 5
23#elif BITS_PER_LONG == 64
24# define _BITOPS_LONG_SHIFT 6
25#else
26# error "Unexpected BITS_PER_LONG"
27#endif
28
29#define BIT_64(n) (U64_C(1) << (n))
30
31/*
32 * These have to be done with inline assembly: that way the bit-setting
33 * is guaranteed to be atomic. All bit operations return 0 if the bit
34 * was cleared before the operation and != 0 if it was not.
35 *
36 * bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1).
37 */
38
39#define RLONG_ADDR(x) "m" (*(volatile long *) (x))
40#define WBYTE_ADDR(x) "+m" (*(volatile char *) (x))
41
42#define ADDR RLONG_ADDR(addr)
43
44/*
45 * We do the locked ops that don't return the old value as
46 * a mask operation on a byte.
47 */
48#define CONST_MASK_ADDR(nr, addr) WBYTE_ADDR((void *)(addr) + ((nr)>>3))
49#define CONST_MASK(nr) (1 << ((nr) & 7))
50
51static __always_inline void
52arch_set_bit(long nr, volatile unsigned long *addr)
53{
54 if (__builtin_constant_p(nr)) {
55 asm volatile(LOCK_PREFIX "orb %b1,%0"
56 : CONST_MASK_ADDR(nr, addr)
57 : "iq" (CONST_MASK(nr))
58 : "memory");
59 } else {
60 asm volatile(LOCK_PREFIX __ASM_SIZE(bts) " %1,%0"
61 : : RLONG_ADDR(addr), "Ir" (nr) : "memory");
62 }
63}
64
65static __always_inline void
66arch___set_bit(unsigned long nr, volatile unsigned long *addr)
67{
68 asm volatile(__ASM_SIZE(bts) " %1,%0" : : ADDR, "Ir" (nr) : "memory");
69}
70
71static __always_inline void
72arch_clear_bit(long nr, volatile unsigned long *addr)
73{
74 if (__builtin_constant_p(nr)) {
75 asm volatile(LOCK_PREFIX "andb %b1,%0"
76 : CONST_MASK_ADDR(nr, addr)
77 : "iq" (~CONST_MASK(nr)));
78 } else {
79 asm volatile(LOCK_PREFIX __ASM_SIZE(btr) " %1,%0"
80 : : RLONG_ADDR(addr), "Ir" (nr) : "memory");
81 }
82}
83
84static __always_inline void
85arch_clear_bit_unlock(long nr, volatile unsigned long *addr)
86{
87 barrier();
88 arch_clear_bit(nr, addr);
89}
90
91static __always_inline void
92arch___clear_bit(unsigned long nr, volatile unsigned long *addr)
93{
94 asm volatile(__ASM_SIZE(btr) " %1,%0" : : ADDR, "Ir" (nr) : "memory");
95}
96
97static __always_inline bool arch_xor_unlock_is_negative_byte(unsigned long mask,
98 volatile unsigned long *addr)
99{
100 bool negative;
101 asm volatile(LOCK_PREFIX "xorb %2,%1"
102 CC_SET(s)
103 : CC_OUT(s) (negative), WBYTE_ADDR(addr)
104 : "iq" ((char)mask) : "memory");
105 return negative;
106}
107#define arch_xor_unlock_is_negative_byte arch_xor_unlock_is_negative_byte
108
109static __always_inline void
110arch___clear_bit_unlock(long nr, volatile unsigned long *addr)
111{
112 arch___clear_bit(nr, addr);
113}
114
115static __always_inline void
116arch___change_bit(unsigned long nr, volatile unsigned long *addr)
117{
118 asm volatile(__ASM_SIZE(btc) " %1,%0" : : ADDR, "Ir" (nr) : "memory");
119}
120
121static __always_inline void
122arch_change_bit(long nr, volatile unsigned long *addr)
123{
124 if (__builtin_constant_p(nr)) {
125 asm volatile(LOCK_PREFIX "xorb %b1,%0"
126 : CONST_MASK_ADDR(nr, addr)
127 : "iq" (CONST_MASK(nr)));
128 } else {
129 asm volatile(LOCK_PREFIX __ASM_SIZE(btc) " %1,%0"
130 : : RLONG_ADDR(addr), "Ir" (nr) : "memory");
131 }
132}
133
134static __always_inline bool
135arch_test_and_set_bit(long nr, volatile unsigned long *addr)
136{
137 return GEN_BINARY_RMWcc(LOCK_PREFIX __ASM_SIZE(bts), *addr, c, "Ir", nr);
138}
139
140static __always_inline bool
141arch_test_and_set_bit_lock(long nr, volatile unsigned long *addr)
142{
143 return arch_test_and_set_bit(nr, addr);
144}
145
146static __always_inline bool
147arch___test_and_set_bit(unsigned long nr, volatile unsigned long *addr)
148{
149 bool oldbit;
150
151 asm(__ASM_SIZE(bts) " %2,%1"
152 CC_SET(c)
153 : CC_OUT(c) (oldbit)
154 : ADDR, "Ir" (nr) : "memory");
155 return oldbit;
156}
157
158static __always_inline bool
159arch_test_and_clear_bit(long nr, volatile unsigned long *addr)
160{
161 return GEN_BINARY_RMWcc(LOCK_PREFIX __ASM_SIZE(btr), *addr, c, "Ir", nr);
162}
163
164/*
165 * Note: the operation is performed atomically with respect to
166 * the local CPU, but not other CPUs. Portable code should not
167 * rely on this behaviour.
168 * KVM relies on this behaviour on x86 for modifying memory that is also
169 * accessed from a hypervisor on the same CPU if running in a VM: don't change
170 * this without also updating arch/x86/kernel/kvm.c
171 */
172static __always_inline bool
173arch___test_and_clear_bit(unsigned long nr, volatile unsigned long *addr)
174{
175 bool oldbit;
176
177 asm volatile(__ASM_SIZE(btr) " %2,%1"
178 CC_SET(c)
179 : CC_OUT(c) (oldbit)
180 : ADDR, "Ir" (nr) : "memory");
181 return oldbit;
182}
183
184static __always_inline bool
185arch___test_and_change_bit(unsigned long nr, volatile unsigned long *addr)
186{
187 bool oldbit;
188
189 asm volatile(__ASM_SIZE(btc) " %2,%1"
190 CC_SET(c)
191 : CC_OUT(c) (oldbit)
192 : ADDR, "Ir" (nr) : "memory");
193
194 return oldbit;
195}
196
197static __always_inline bool
198arch_test_and_change_bit(long nr, volatile unsigned long *addr)
199{
200 return GEN_BINARY_RMWcc(LOCK_PREFIX __ASM_SIZE(btc), *addr, c, "Ir", nr);
201}
202
203static __always_inline bool constant_test_bit(long nr, const volatile unsigned long *addr)
204{
205 return ((1UL << (nr & (BITS_PER_LONG-1))) &
206 (addr[nr >> _BITOPS_LONG_SHIFT])) != 0;
207}
208
209static __always_inline bool constant_test_bit_acquire(long nr, const volatile unsigned long *addr)
210{
211 bool oldbit;
212
213 asm volatile("testb %2,%1"
214 CC_SET(nz)
215 : CC_OUT(nz) (oldbit)
216 : "m" (((unsigned char *)addr)[nr >> 3]),
217 "i" (1 << (nr & 7))
218 :"memory");
219
220 return oldbit;
221}
222
223static __always_inline bool variable_test_bit(long nr, volatile const unsigned long *addr)
224{
225 bool oldbit;
226
227 asm volatile(__ASM_SIZE(bt) " %2,%1"
228 CC_SET(c)
229 : CC_OUT(c) (oldbit)
230 : "m" (*(unsigned long *)addr), "Ir" (nr) : "memory");
231
232 return oldbit;
233}
234
235static __always_inline bool
236arch_test_bit(unsigned long nr, const volatile unsigned long *addr)
237{
238 return __builtin_constant_p(nr) ? constant_test_bit(nr, addr) :
239 variable_test_bit(nr, addr);
240}
241
242static __always_inline bool
243arch_test_bit_acquire(unsigned long nr, const volatile unsigned long *addr)
244{
245 return __builtin_constant_p(nr) ? constant_test_bit_acquire(nr, addr) :
246 variable_test_bit(nr, addr);
247}
248
249static __always_inline unsigned long variable__ffs(unsigned long word)
250{
251 asm("rep; bsf %1,%0"
252 : "=r" (word)
253 : "rm" (word));
254 return word;
255}
256
257/**
258 * __ffs - find first set bit in word
259 * @word: The word to search
260 *
261 * Undefined if no bit exists, so code should check against 0 first.
262 */
263#define __ffs(word) \
264 (__builtin_constant_p(word) ? \
265 (unsigned long)__builtin_ctzl(word) : \
266 variable__ffs(word))
267
268static __always_inline unsigned long variable_ffz(unsigned long word)
269{
270 asm("rep; bsf %1,%0"
271 : "=r" (word)
272 : "r" (~word));
273 return word;
274}
275
276/**
277 * ffz - find first zero bit in word
278 * @word: The word to search
279 *
280 * Undefined if no zero exists, so code should check against ~0UL first.
281 */
282#define ffz(word) \
283 (__builtin_constant_p(word) ? \
284 (unsigned long)__builtin_ctzl(~word) : \
285 variable_ffz(word))
286
287/*
288 * __fls: find last set bit in word
289 * @word: The word to search
290 *
291 * Undefined if no set bit exists, so code should check against 0 first.
292 */
293static __always_inline unsigned long __fls(unsigned long word)
294{
295 if (__builtin_constant_p(word))
296 return BITS_PER_LONG - 1 - __builtin_clzl(word);
297
298 asm("bsr %1,%0"
299 : "=r" (word)
300 : "rm" (word));
301 return word;
302}
303
304#undef ADDR
305
306#ifdef __KERNEL__
307static __always_inline int variable_ffs(int x)
308{
309 int r;
310
311#ifdef CONFIG_X86_64
312 /*
313 * AMD64 says BSFL won't clobber the dest reg if x==0; Intel64 says the
314 * dest reg is undefined if x==0, but their CPU architect says its
315 * value is written to set it to the same as before, except that the
316 * top 32 bits will be cleared.
317 *
318 * We cannot do this on 32 bits because at the very least some
319 * 486 CPUs did not behave this way.
320 */
321 asm("bsfl %1,%0"
322 : "=r" (r)
323 : "rm" (x), "0" (-1));
324#elif defined(CONFIG_X86_CMOV)
325 asm("bsfl %1,%0\n\t"
326 "cmovzl %2,%0"
327 : "=&r" (r) : "rm" (x), "r" (-1));
328#else
329 asm("bsfl %1,%0\n\t"
330 "jnz 1f\n\t"
331 "movl $-1,%0\n"
332 "1:" : "=r" (r) : "rm" (x));
333#endif
334 return r + 1;
335}
336
337/**
338 * ffs - find first set bit in word
339 * @x: the word to search
340 *
341 * This is defined the same way as the libc and compiler builtin ffs
342 * routines, therefore differs in spirit from the other bitops.
343 *
344 * ffs(value) returns 0 if value is 0 or the position of the first
345 * set bit if value is nonzero. The first (least significant) bit
346 * is at position 1.
347 */
348#define ffs(x) (__builtin_constant_p(x) ? __builtin_ffs(x) : variable_ffs(x))
349
350/**
351 * fls - find last set bit in word
352 * @x: the word to search
353 *
354 * This is defined in a similar way as the libc and compiler builtin
355 * ffs, but returns the position of the most significant set bit.
356 *
357 * fls(value) returns 0 if value is 0 or the position of the last
358 * set bit if value is nonzero. The last (most significant) bit is
359 * at position 32.
360 */
361static __always_inline int fls(unsigned int x)
362{
363 int r;
364
365 if (__builtin_constant_p(x))
366 return x ? 32 - __builtin_clz(x) : 0;
367
368#ifdef CONFIG_X86_64
369 /*
370 * AMD64 says BSRL won't clobber the dest reg if x==0; Intel64 says the
371 * dest reg is undefined if x==0, but their CPU architect says its
372 * value is written to set it to the same as before, except that the
373 * top 32 bits will be cleared.
374 *
375 * We cannot do this on 32 bits because at the very least some
376 * 486 CPUs did not behave this way.
377 */
378 asm("bsrl %1,%0"
379 : "=r" (r)
380 : "rm" (x), "0" (-1));
381#elif defined(CONFIG_X86_CMOV)
382 asm("bsrl %1,%0\n\t"
383 "cmovzl %2,%0"
384 : "=&r" (r) : "rm" (x), "rm" (-1));
385#else
386 asm("bsrl %1,%0\n\t"
387 "jnz 1f\n\t"
388 "movl $-1,%0\n"
389 "1:" : "=r" (r) : "rm" (x));
390#endif
391 return r + 1;
392}
393
394/**
395 * fls64 - find last set bit in a 64-bit word
396 * @x: the word to search
397 *
398 * This is defined in a similar way as the libc and compiler builtin
399 * ffsll, but returns the position of the most significant set bit.
400 *
401 * fls64(value) returns 0 if value is 0 or the position of the last
402 * set bit if value is nonzero. The last (most significant) bit is
403 * at position 64.
404 */
405#ifdef CONFIG_X86_64
406static __always_inline int fls64(__u64 x)
407{
408 int bitpos = -1;
409
410 if (__builtin_constant_p(x))
411 return x ? 64 - __builtin_clzll(x) : 0;
412 /*
413 * AMD64 says BSRQ won't clobber the dest reg if x==0; Intel64 says the
414 * dest reg is undefined if x==0, but their CPU architect says its
415 * value is written to set it to the same as before.
416 */
417 asm("bsrq %1,%q0"
418 : "+r" (bitpos)
419 : "rm" (x));
420 return bitpos + 1;
421}
422#else
423#include <asm-generic/bitops/fls64.h>
424#endif
425
426#include <asm-generic/bitops/sched.h>
427
428#include <asm/arch_hweight.h>
429
430#include <asm-generic/bitops/const_hweight.h>
431
432#include <asm-generic/bitops/instrumented-atomic.h>
433#include <asm-generic/bitops/instrumented-non-atomic.h>
434#include <asm-generic/bitops/instrumented-lock.h>
435
436#include <asm-generic/bitops/le.h>
437
438#include <asm-generic/bitops/ext2-atomic-setbit.h>
439
440#endif /* __KERNEL__ */
441#endif /* _ASM_X86_BITOPS_H */
442

source code of linux/arch/x86/include/asm/bitops.h