cpu.h

 1/* SPDX-License-Identifier: GPL-2.0-only */
 2/*
 3 * Copyright 2023 Red Hat
 4 */
 5
 6#ifndef UDS_CPU_H
 7#define UDS_CPU_H
 8
 9#include <linux/cache.h>
10
11/**
12 * uds_prefetch_address() - Minimize cache-miss latency by attempting to move data into a CPU cache
13 *                          before it is accessed.
14 *
15 * @address: the address to fetch (may be invalid)
16 * @for_write: must be constant at compile time--false if for reading, true if for writing
17 */
18static inline void uds_prefetch_address(const void *address, bool for_write)
19{
20	/*
21	 * for_write won't be a constant if we are compiled with optimization turned off, in which
22	 * case prefetching really doesn't matter. clang can't figure out that if for_write is a
23	 * constant, it can be passed as the second, mandatorily constant argument to prefetch(),
24	 * at least currently on llvm 12.
25	 */
26	if (__builtin_constant_p(for_write)) {
27		if (for_write)
28			__builtin_prefetch(address, true);
29		else
30			__builtin_prefetch(address, false);
31	}
32}
33
34/**
35 * uds_prefetch_range() - Minimize cache-miss latency by attempting to move a range of addresses
36 *                        into a CPU cache before they are accessed.
37 *
38 * @start: the starting address to fetch (may be invalid)
39 * @size: the number of bytes in the address range
40 * @for_write: must be constant at compile time--false if for reading, true if for writing
41 */
42static inline void uds_prefetch_range(const void *start, unsigned int size,
43				      bool for_write)
44{
45	/*
46	 * Count the number of cache lines to fetch, allowing for the address range to span an
47	 * extra cache line boundary due to address alignment.
48	 */
49	const char *address = (const char *) start;
50	unsigned int offset = ((uintptr_t) address % L1_CACHE_BYTES);
51	unsigned int cache_lines = (1 + ((size + offset) / L1_CACHE_BYTES));
52
53	while (cache_lines-- > 0) {
54		uds_prefetch_address(address, for_write);
55		address += L1_CACHE_BYTES;
56	}
57}
58
59#endif /* UDS_CPU_H */