xref: /linux/include/linux/compiler.h (revision 132db93572821ec2fdf81e354cc40f558faf7e4f)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_COMPILER_H
3 #define __LINUX_COMPILER_H
4 
5 #include <linux/compiler_types.h>
6 
7 #ifndef __ASSEMBLY__
8 
9 #ifdef __KERNEL__
10 
11 /*
12  * Note: DISABLE_BRANCH_PROFILING can be used by special lowlevel code
13  * to disable branch tracing on a per file basis.
14  */
15 #if defined(CONFIG_TRACE_BRANCH_PROFILING) \
16     && !defined(DISABLE_BRANCH_PROFILING) && !defined(__CHECKER__)
17 void ftrace_likely_update(struct ftrace_likely_data *f, int val,
18 			  int expect, int is_constant);
19 
20 #define likely_notrace(x)	__builtin_expect(!!(x), 1)
21 #define unlikely_notrace(x)	__builtin_expect(!!(x), 0)
22 
23 #define __branch_check__(x, expect, is_constant) ({			\
24 			long ______r;					\
25 			static struct ftrace_likely_data		\
26 				__aligned(4)				\
27 				__section(_ftrace_annotated_branch)	\
28 				______f = {				\
29 				.data.func = __func__,			\
30 				.data.file = __FILE__,			\
31 				.data.line = __LINE__,			\
32 			};						\
33 			______r = __builtin_expect(!!(x), expect);	\
34 			ftrace_likely_update(&______f, ______r,		\
35 					     expect, is_constant);	\
36 			______r;					\
37 		})
38 
39 /*
40  * Using __builtin_constant_p(x) to ignore cases where the return
41  * value is always the same.  This idea is taken from a similar patch
42  * written by Daniel Walker.
43  */
44 # ifndef likely
45 #  define likely(x)	(__branch_check__(x, 1, __builtin_constant_p(x)))
46 # endif
47 # ifndef unlikely
48 #  define unlikely(x)	(__branch_check__(x, 0, __builtin_constant_p(x)))
49 # endif
50 
51 #ifdef CONFIG_PROFILE_ALL_BRANCHES
52 /*
53  * "Define 'is'", Bill Clinton
54  * "Define 'if'", Steven Rostedt
55  */
56 #define if(cond, ...) if ( __trace_if_var( !!(cond , ## __VA_ARGS__) ) )
57 
58 #define __trace_if_var(cond) (__builtin_constant_p(cond) ? (cond) : __trace_if_value(cond))
59 
60 #define __trace_if_value(cond) ({			\
61 	static struct ftrace_branch_data		\
62 		__aligned(4)				\
63 		__section(_ftrace_branch)		\
64 		__if_trace = {				\
65 			.func = __func__,		\
66 			.file = __FILE__,		\
67 			.line = __LINE__,		\
68 		};					\
69 	(cond) ?					\
70 		(__if_trace.miss_hit[1]++,1) :		\
71 		(__if_trace.miss_hit[0]++,0);		\
72 })
73 
74 #endif /* CONFIG_PROFILE_ALL_BRANCHES */
75 
76 #else
77 # define likely(x)	__builtin_expect(!!(x), 1)
78 # define unlikely(x)	__builtin_expect(!!(x), 0)
79 #endif
80 
81 /* Optimization barrier */
82 #ifndef barrier
83 # define barrier() __memory_barrier()
84 #endif
85 
86 #ifndef barrier_data
87 # define barrier_data(ptr) barrier()
88 #endif
89 
90 /* workaround for GCC PR82365 if needed */
91 #ifndef barrier_before_unreachable
92 # define barrier_before_unreachable() do { } while (0)
93 #endif
94 
95 /* Unreachable code */
96 #ifdef CONFIG_STACK_VALIDATION
97 /*
98  * These macros help objtool understand GCC code flow for unreachable code.
99  * The __COUNTER__ based labels are a hack to make each instance of the macros
100  * unique, to convince GCC not to merge duplicate inline asm statements.
101  */
102 #define annotate_reachable() ({						\
103 	asm volatile("%c0:\n\t"						\
104 		     ".pushsection .discard.reachable\n\t"		\
105 		     ".long %c0b - .\n\t"				\
106 		     ".popsection\n\t" : : "i" (__COUNTER__));		\
107 })
108 #define annotate_unreachable() ({					\
109 	asm volatile("%c0:\n\t"						\
110 		     ".pushsection .discard.unreachable\n\t"		\
111 		     ".long %c0b - .\n\t"				\
112 		     ".popsection\n\t" : : "i" (__COUNTER__));		\
113 })
114 #define ASM_UNREACHABLE							\
115 	"999:\n\t"							\
116 	".pushsection .discard.unreachable\n\t"				\
117 	".long 999b - .\n\t"						\
118 	".popsection\n\t"
119 
120 /* Annotate a C jump table to allow objtool to follow the code flow */
121 #define __annotate_jump_table __section(.rodata..c_jump_table)
122 
123 #ifdef CONFIG_DEBUG_ENTRY
124 /* Begin/end of an instrumentation safe region */
125 #define instrumentation_begin() ({					\
126 	asm volatile("%c0:\n\t"						\
127 		     ".pushsection .discard.instr_begin\n\t"		\
128 		     ".long %c0b - .\n\t"				\
129 		     ".popsection\n\t" : : "i" (__COUNTER__));		\
130 })
131 
132 /*
133  * Because instrumentation_{begin,end}() can nest, objtool validation considers
134  * _begin() a +1 and _end() a -1 and computes a sum over the instructions.
135  * When the value is greater than 0, we consider instrumentation allowed.
136  *
137  * There is a problem with code like:
138  *
139  * noinstr void foo()
140  * {
141  *	instrumentation_begin();
142  *	...
143  *	if (cond) {
144  *		instrumentation_begin();
145  *		...
146  *		instrumentation_end();
147  *	}
148  *	bar();
149  *	instrumentation_end();
150  * }
151  *
152  * If instrumentation_end() would be an empty label, like all the other
153  * annotations, the inner _end(), which is at the end of a conditional block,
154  * would land on the instruction after the block.
155  *
156  * If we then consider the sum of the !cond path, we'll see that the call to
157  * bar() is with a 0-value, even though, we meant it to happen with a positive
158  * value.
159  *
160  * To avoid this, have _end() be a NOP instruction, this ensures it will be
161  * part of the condition block and does not escape.
162  */
163 #define instrumentation_end() ({					\
164 	asm volatile("%c0: nop\n\t"					\
165 		     ".pushsection .discard.instr_end\n\t"		\
166 		     ".long %c0b - .\n\t"				\
167 		     ".popsection\n\t" : : "i" (__COUNTER__));		\
168 })
169 #endif /* CONFIG_DEBUG_ENTRY */
170 
171 #else
172 #define annotate_reachable()
173 #define annotate_unreachable()
174 #define __annotate_jump_table
175 #endif
176 
177 #ifndef instrumentation_begin
178 #define instrumentation_begin()		do { } while(0)
179 #define instrumentation_end()		do { } while(0)
180 #endif
181 
182 #ifndef ASM_UNREACHABLE
183 # define ASM_UNREACHABLE
184 #endif
185 #ifndef unreachable
186 # define unreachable() do {		\
187 	annotate_unreachable();		\
188 	__builtin_unreachable();	\
189 } while (0)
190 #endif
191 
192 /*
193  * KENTRY - kernel entry point
194  * This can be used to annotate symbols (functions or data) that are used
195  * without their linker symbol being referenced explicitly. For example,
196  * interrupt vector handlers, or functions in the kernel image that are found
197  * programatically.
198  *
199  * Not required for symbols exported with EXPORT_SYMBOL, or initcalls. Those
200  * are handled in their own way (with KEEP() in linker scripts).
201  *
202  * KENTRY can be avoided if the symbols in question are marked as KEEP() in the
203  * linker script. For example an architecture could KEEP() its entire
204  * boot/exception vector code rather than annotate each function and data.
205  */
206 #ifndef KENTRY
207 # define KENTRY(sym)						\
208 	extern typeof(sym) sym;					\
209 	static const unsigned long __kentry_##sym		\
210 	__used							\
211 	__section("___kentry" "+" #sym )			\
212 	= (unsigned long)&sym;
213 #endif
214 
215 #ifndef RELOC_HIDE
216 # define RELOC_HIDE(ptr, off)					\
217   ({ unsigned long __ptr;					\
218      __ptr = (unsigned long) (ptr);				\
219     (typeof(ptr)) (__ptr + (off)); })
220 #endif
221 
222 #ifndef OPTIMIZER_HIDE_VAR
223 /* Make the optimizer believe the variable can be manipulated arbitrarily. */
224 #define OPTIMIZER_HIDE_VAR(var)						\
225 	__asm__ ("" : "=r" (var) : "0" (var))
226 #endif
227 
228 /* Not-quite-unique ID. */
229 #ifndef __UNIQUE_ID
230 # define __UNIQUE_ID(prefix) __PASTE(__PASTE(__UNIQUE_ID_, prefix), __LINE__)
231 #endif
232 
233 /*
234  * Prevent the compiler from merging or refetching reads or writes. The
235  * compiler is also forbidden from reordering successive instances of
236  * READ_ONCE and WRITE_ONCE, but only when the compiler is aware of some
237  * particular ordering. One way to make the compiler aware of ordering is to
238  * put the two invocations of READ_ONCE or WRITE_ONCE in different C
239  * statements.
240  *
241  * These two macros will also work on aggregate data types like structs or
242  * unions.
243  *
244  * Their two major use cases are: (1) Mediating communication between
245  * process-level code and irq/NMI handlers, all running on the same CPU,
246  * and (2) Ensuring that the compiler does not fold, spindle, or otherwise
247  * mutilate accesses that either do not require ordering or that interact
248  * with an explicit memory barrier or atomic instruction that provides the
249  * required ordering.
250  */
251 #include <asm/barrier.h>
252 #include <linux/kasan-checks.h>
253 #include <linux/kcsan-checks.h>
254 
255 /**
256  * data_race - mark an expression as containing intentional data races
257  *
258  * This data_race() macro is useful for situations in which data races
259  * should be forgiven.  One example is diagnostic code that accesses
260  * shared variables but is not a part of the core synchronization design.
261  *
262  * This macro *does not* affect normal code generation, but is a hint
263  * to tooling that data races here are to be ignored.
264  */
265 #define data_race(expr)							\
266 ({									\
267 	__unqual_scalar_typeof(({ expr; })) __v = ({			\
268 		__kcsan_disable_current();				\
269 		expr;							\
270 	});								\
271 	__kcsan_enable_current();					\
272 	__v;								\
273 })
274 
275 /*
276  * Use __READ_ONCE() instead of READ_ONCE() if you do not require any
277  * atomicity or dependency ordering guarantees. Note that this may result
278  * in tears!
279  */
280 #define __READ_ONCE(x)	(*(const volatile __unqual_scalar_typeof(x) *)&(x))
281 
282 #define __READ_ONCE_SCALAR(x)						\
283 ({									\
284 	__unqual_scalar_typeof(x) __x = __READ_ONCE(x);			\
285 	smp_read_barrier_depends();					\
286 	(typeof(x))__x;							\
287 })
288 
289 #define READ_ONCE(x)							\
290 ({									\
291 	compiletime_assert_rwonce_type(x);				\
292 	__READ_ONCE_SCALAR(x);						\
293 })
294 
295 #define __WRITE_ONCE(x, val)						\
296 do {									\
297 	*(volatile typeof(x) *)&(x) = (val);				\
298 } while (0)
299 
300 #define WRITE_ONCE(x, val)						\
301 do {									\
302 	compiletime_assert_rwonce_type(x);				\
303 	__WRITE_ONCE(x, val);						\
304 } while (0)
305 
306 static __no_sanitize_or_inline
307 unsigned long __read_once_word_nocheck(const void *addr)
308 {
309 	return __READ_ONCE(*(unsigned long *)addr);
310 }
311 
312 /*
313  * Use READ_ONCE_NOCHECK() instead of READ_ONCE() if you need to load a
314  * word from memory atomically but without telling KASAN/KCSAN. This is
315  * usually used by unwinding code when walking the stack of a running process.
316  */
317 #define READ_ONCE_NOCHECK(x)						\
318 ({									\
319 	unsigned long __x;						\
320 	compiletime_assert(sizeof(x) == sizeof(__x),			\
321 		"Unsupported access size for READ_ONCE_NOCHECK().");	\
322 	__x = __read_once_word_nocheck(&(x));				\
323 	smp_read_barrier_depends();					\
324 	(typeof(x))__x;							\
325 })
326 
327 static __no_kasan_or_inline
328 unsigned long read_word_at_a_time(const void *addr)
329 {
330 	kasan_check_read(addr, 1);
331 	return *(unsigned long *)addr;
332 }
333 
334 #endif /* __KERNEL__ */
335 
336 /*
337  * Force the compiler to emit 'sym' as a symbol, so that we can reference
338  * it from inline assembler. Necessary in case 'sym' could be inlined
339  * otherwise, or eliminated entirely due to lack of references that are
340  * visible to the compiler.
341  */
342 #define __ADDRESSABLE(sym) \
343 	static void * __section(.discard.addressable) __used \
344 		__PASTE(__addressable_##sym, __LINE__) = (void *)&sym;
345 
346 /**
347  * offset_to_ptr - convert a relative memory offset to an absolute pointer
348  * @off:	the address of the 32-bit offset value
349  */
350 static inline void *offset_to_ptr(const int *off)
351 {
352 	return (void *)((unsigned long)off + *off);
353 }
354 
355 #endif /* __ASSEMBLY__ */
356 
357 /* Compile time object size, -1 for unknown */
358 #ifndef __compiletime_object_size
359 # define __compiletime_object_size(obj) -1
360 #endif
361 #ifndef __compiletime_warning
362 # define __compiletime_warning(message)
363 #endif
364 #ifndef __compiletime_error
365 # define __compiletime_error(message)
366 #endif
367 
368 #ifdef __OPTIMIZE__
369 # define __compiletime_assert(condition, msg, prefix, suffix)		\
370 	do {								\
371 		extern void prefix ## suffix(void) __compiletime_error(msg); \
372 		if (!(condition))					\
373 			prefix ## suffix();				\
374 	} while (0)
375 #else
376 # define __compiletime_assert(condition, msg, prefix, suffix) do { } while (0)
377 #endif
378 
379 #define _compiletime_assert(condition, msg, prefix, suffix) \
380 	__compiletime_assert(condition, msg, prefix, suffix)
381 
382 /**
383  * compiletime_assert - break build and emit msg if condition is false
384  * @condition: a compile-time constant condition to check
385  * @msg:       a message to emit if condition is false
386  *
387  * In tradition of POSIX assert, this macro will break the build if the
388  * supplied condition is *false*, emitting the supplied error message if the
389  * compiler has support to do so.
390  */
391 #define compiletime_assert(condition, msg) \
392 	_compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
393 
394 #define compiletime_assert_atomic_type(t)				\
395 	compiletime_assert(__native_word(t),				\
396 		"Need native word sized stores/loads for atomicity.")
397 
398 /*
399  * Yes, this permits 64-bit accesses on 32-bit architectures. These will
400  * actually be atomic in some cases (namely Armv7 + LPAE), but for others we
401  * rely on the access being split into 2x32-bit accesses for a 32-bit quantity
402  * (e.g. a virtual address) and a strong prevailing wind.
403  */
404 #define compiletime_assert_rwonce_type(t)					\
405 	compiletime_assert(__native_word(t) || sizeof(t) == sizeof(long long),	\
406 		"Unsupported access size for {READ,WRITE}_ONCE().")
407 
408 /* &a[0] degrades to a pointer: a different type from an array */
409 #define __must_be_array(a)	BUILD_BUG_ON_ZERO(__same_type((a), &(a)[0]))
410 
411 /*
412  * This is needed in functions which generate the stack canary, see
413  * arch/x86/kernel/smpboot.c::start_secondary() for an example.
414  */
415 #define prevent_tail_call_optimization()	mb()
416 
417 #endif /* __LINUX_COMPILER_H */
418