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_inline volatile(LOCK_PREFIX "orb %b1,%0"
56 : CONST_MASK_ADDR(nr, addr)
57 : "iq" (CONST_MASK(nr))
58 : "memory");
59 } else {
60 asm_inline 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_inline volatile(LOCK_PREFIX "andb %b1,%0"
76 : CONST_MASK_ADDR(nr, addr)
77 : "iq" (~CONST_MASK(nr)));
78 } else {
79 asm_inline 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_inline volatile(LOCK_PREFIX "xorb %2,%1"
102 : "=@ccs" (negative), WBYTE_ADDR(addr)
103 : "iq" ((char)mask) : "memory");
104 return negative;
105}
106#define arch_xor_unlock_is_negative_byte arch_xor_unlock_is_negative_byte
107
108static __always_inline void
109arch___clear_bit_unlock(long nr, volatile unsigned long *addr)
110{
111 arch___clear_bit(nr, addr);
112}
113
114static __always_inline void
115arch___change_bit(unsigned long nr, volatile unsigned long *addr)
116{
117 asm volatile(__ASM_SIZE(btc) " %1,%0" : : ADDR, "Ir" (nr) : "memory");
118}
119
120static __always_inline void
121arch_change_bit(long nr, volatile unsigned long *addr)
122{
123 if (__builtin_constant_p(nr)) {
124 asm_inline volatile(LOCK_PREFIX "xorb %b1,%0"
125 : CONST_MASK_ADDR(nr, addr)
126 : "iq" (CONST_MASK(nr)));
127 } else {
128 asm_inline volatile(LOCK_PREFIX __ASM_SIZE(btc) " %1,%0"
129 : : RLONG_ADDR(addr), "Ir" (nr) : "memory");
130 }
131}
132
133static __always_inline bool
134arch_test_and_set_bit(long nr, volatile unsigned long *addr)
135{
136 return GEN_BINARY_RMWcc(LOCK_PREFIX __ASM_SIZE(bts), *addr, c, "Ir", nr);
137}
138
139static __always_inline bool
140arch_test_and_set_bit_lock(long nr, volatile unsigned long *addr)
141{
142 return arch_test_and_set_bit(nr, addr);
143}
144
145static __always_inline bool
146arch___test_and_set_bit(unsigned long nr, volatile unsigned long *addr)
147{
148 bool oldbit;
149
150 asm(__ASM_SIZE(bts) " %2,%1"
151 : "=@ccc" (oldbit)
152 : ADDR, "Ir" (nr) : "memory");
153 return oldbit;
154}
155
156static __always_inline bool
157arch_test_and_clear_bit(long nr, volatile unsigned long *addr)
158{
159 return GEN_BINARY_RMWcc(LOCK_PREFIX __ASM_SIZE(btr), *addr, c, "Ir", nr);
160}
161
162/*
163 * Note: the operation is performed atomically with respect to
164 * the local CPU, but not other CPUs. Portable code should not
165 * rely on this behaviour.
166 * KVM relies on this behaviour on x86 for modifying memory that is also
167 * accessed from a hypervisor on the same CPU if running in a VM: don't change
168 * this without also updating arch/x86/kernel/kvm.c
169 */
170static __always_inline bool
171arch___test_and_clear_bit(unsigned long nr, volatile unsigned long *addr)
172{
173 bool oldbit;
174
175 asm volatile(__ASM_SIZE(btr) " %2,%1"
176 : "=@ccc" (oldbit)
177 : ADDR, "Ir" (nr) : "memory");
178 return oldbit;
179}
180
181static __always_inline bool
182arch___test_and_change_bit(unsigned long nr, volatile unsigned long *addr)
183{
184 bool oldbit;
185
186 asm volatile(__ASM_SIZE(btc) " %2,%1"
187 : "=@ccc" (oldbit)
188 : ADDR, "Ir" (nr) : "memory");
189
190 return oldbit;
191}
192
193static __always_inline bool
194arch_test_and_change_bit(long nr, volatile unsigned long *addr)
195{
196 return GEN_BINARY_RMWcc(LOCK_PREFIX __ASM_SIZE(btc), *addr, c, "Ir", nr);
197}
198
199static __always_inline bool constant_test_bit(long nr, const volatile unsigned long *addr)
200{
201 return ((1UL << (nr & (BITS_PER_LONG-1))) &
202 (addr[nr >> _BITOPS_LONG_SHIFT])) != 0;
203}
204
205static __always_inline bool constant_test_bit_acquire(long nr, const volatile unsigned long *addr)
206{
207 bool oldbit;
208
209 asm volatile("testb %2,%1"
210 : "=@ccnz" (oldbit)
211 : "m" (((unsigned char *)addr)[nr >> 3]),
212 "i" (1 << (nr & 7))
213 :"memory");
214
215 return oldbit;
216}
217
218static __always_inline bool variable_test_bit(long nr, volatile const unsigned long *addr)
219{
220 bool oldbit;
221
222 asm volatile(__ASM_SIZE(bt) " %2,%1"
223 : "=@ccc" (oldbit)
224 : "m" (*(unsigned long *)addr), "Ir" (nr) : "memory");
225
226 return oldbit;
227}
228
229static __always_inline bool
230arch_test_bit(unsigned long nr, const volatile unsigned long *addr)
231{
232 return __builtin_constant_p(nr) ? constant_test_bit(nr, addr) :
233 variable_test_bit(nr, addr);
234}
235
236static __always_inline bool
237arch_test_bit_acquire(unsigned long nr, const volatile unsigned long *addr)
238{
239 return __builtin_constant_p(nr) ? constant_test_bit_acquire(nr, addr) :
240 variable_test_bit(nr, addr);
241}
242
243static __always_inline __attribute_const__ unsigned long variable__ffs(unsigned long word)
244{
245 asm("tzcnt %1,%0"
246 : "=r" (word)
247 : ASM_INPUT_RM (word));
248 return word;
249}
250
251/**
252 * __ffs - find first set bit in word
253 * @word: The word to search
254 *
255 * Undefined if no bit exists, so code should check against 0 first.
256 */
257#define __ffs(word) \
258 (__builtin_constant_p(word) ? \
259 (unsigned long)__builtin_ctzl(word) : \
260 variable__ffs(word))
261
262static __always_inline __attribute_const__ unsigned long variable_ffz(unsigned long word)
263{
264 return variable__ffs(word: ~word);
265}
266
267/**
268 * ffz - find first zero bit in word
269 * @word: The word to search
270 *
271 * Undefined if no zero exists, so code should check against ~0UL first.
272 */
273#define ffz(word) \
274 (__builtin_constant_p(word) ? \
275 (unsigned long)__builtin_ctzl(~word) : \
276 variable_ffz(word))
277
278/*
279 * __fls: find last set bit in word
280 * @word: The word to search
281 *
282 * Undefined if no set bit exists, so code should check against 0 first.
283 */
284static __always_inline __attribute_const__ unsigned long __fls(unsigned long word)
285{
286 if (__builtin_constant_p(word))
287 return BITS_PER_LONG - 1 - __builtin_clzl(word);
288
289 asm("bsr %1,%0"
290 : "=r" (word)
291 : ASM_INPUT_RM (word));
292 return word;
293}
294
295#undef ADDR
296
297#ifdef __KERNEL__
298static __always_inline __attribute_const__ int variable_ffs(int x)
299{
300 int r;
301
302#ifdef CONFIG_X86_64
303 /*
304 * AMD64 says BSFL won't clobber the dest reg if x==0; Intel64 says the
305 * dest reg is undefined if x==0, but their CPU architect says its
306 * value is written to set it to the same as before, except that the
307 * top 32 bits will be cleared.
308 *
309 * We cannot do this on 32 bits because at the very least some
310 * 486 CPUs did not behave this way.
311 */
312 asm("bsfl %1,%0"
313 : "=r" (r)
314 : ASM_INPUT_RM (x), "0" (-1));
315#elif defined(CONFIG_X86_CMOV)
316 asm("bsfl %1,%0\n\t"
317 "cmovzl %2,%0"
318 : "=&r" (r) : "rm" (x), "r" (-1));
319#else
320 asm("bsfl %1,%0\n\t"
321 "jnz 1f\n\t"
322 "movl $-1,%0\n"
323 "1:" : "=r" (r) : "rm" (x));
324#endif
325 return r + 1;
326}
327
328/**
329 * ffs - find first set bit in word
330 * @x: the word to search
331 *
332 * This is defined the same way as the libc and compiler builtin ffs
333 * routines, therefore differs in spirit from the other bitops.
334 *
335 * ffs(value) returns 0 if value is 0 or the position of the first
336 * set bit if value is nonzero. The first (least significant) bit
337 * is at position 1.
338 */
339#define ffs(x) (__builtin_constant_p(x) ? __builtin_ffs(x) : variable_ffs(x))
340
341/**
342 * fls - find last set bit in word
343 * @x: the word to search
344 *
345 * This is defined in a similar way as the libc and compiler builtin
346 * ffs, but returns the position of the most significant set bit.
347 *
348 * fls(value) returns 0 if value is 0 or the position of the last
349 * set bit if value is nonzero. The last (most significant) bit is
350 * at position 32.
351 */
352static __always_inline __attribute_const__ int fls(unsigned int x)
353{
354 int r;
355
356 if (__builtin_constant_p(x))
357 return x ? 32 - __builtin_clz(x) : 0;
358
359#ifdef CONFIG_X86_64
360 /*
361 * AMD64 says BSRL won't clobber the dest reg if x==0; Intel64 says the
362 * dest reg is undefined if x==0, but their CPU architect says its
363 * value is written to set it to the same as before, except that the
364 * top 32 bits will be cleared.
365 *
366 * We cannot do this on 32 bits because at the very least some
367 * 486 CPUs did not behave this way.
368 */
369 asm("bsrl %1,%0"
370 : "=r" (r)
371 : ASM_INPUT_RM (x), "0" (-1));
372#elif defined(CONFIG_X86_CMOV)
373 asm("bsrl %1,%0\n\t"
374 "cmovzl %2,%0"
375 : "=&r" (r) : "rm" (x), "rm" (-1));
376#else
377 asm("bsrl %1,%0\n\t"
378 "jnz 1f\n\t"
379 "movl $-1,%0\n"
380 "1:" : "=r" (r) : "rm" (x));
381#endif
382 return r + 1;
383}
384
385/**
386 * fls64 - find last set bit in a 64-bit word
387 * @x: the word to search
388 *
389 * This is defined in a similar way as the libc and compiler builtin
390 * ffsll, but returns the position of the most significant set bit.
391 *
392 * fls64(value) returns 0 if value is 0 or the position of the last
393 * set bit if value is nonzero. The last (most significant) bit is
394 * at position 64.
395 */
396#ifdef CONFIG_X86_64
397static __always_inline __attribute_const__ int fls64(__u64 x)
398{
399 int bitpos = -1;
400
401 if (__builtin_constant_p(x))
402 return x ? 64 - __builtin_clzll(x) : 0;
403 /*
404 * AMD64 says BSRQ won't clobber the dest reg if x==0; Intel64 says the
405 * dest reg is undefined if x==0, but their CPU architect says its
406 * value is written to set it to the same as before.
407 */
408 asm("bsrq %1,%q0"
409 : "+r" (bitpos)
410 : ASM_INPUT_RM (x));
411 return bitpos + 1;
412}
413#else
414#include <asm-generic/bitops/fls64.h>
415#endif
416
417#include <asm-generic/bitops/sched.h>
418
419#include <asm/arch_hweight.h>
420
421#include <asm-generic/bitops/const_hweight.h>
422
423#include <asm-generic/bitops/instrumented-atomic.h>
424#include <asm-generic/bitops/instrumented-non-atomic.h>
425#include <asm-generic/bitops/instrumented-lock.h>
426
427#include <asm-generic/bitops/le.h>
428
429#include <asm-generic/bitops/ext2-atomic-setbit.h>
430
431#endif /* __KERNEL__ */
432#endif /* _ASM_X86_BITOPS_H */
433