1/* SPDX-License-Identifier: GPL-2.0-only */ 2/* 3 * Copyright (C) 2013 ARM Ltd. 4 * Copyright (C) 2013 Linaro. 5 * 6 * This code is based on glibc cortex strings work originally authored by Linaro 7 * be found @ 8 * 9 * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/ 10 * files/head:/src/aarch64/ 11 */ 12 13#include <linux/linkage.h> 14#include <asm/assembler.h> 15 16/* 17 * determine the length of a fixed-size string 18 * 19 * Parameters: 20 * x0 - const string pointer 21 * x1 - maximal string length 22 * Returns: 23 * x0 - the return length of specific string 24 */ 25 26/* Arguments and results. */ 27srcin .req x0 28len .req x0 29limit .req x1 30 31/* Locals and temporaries. */ 32src .req x2 33data1 .req x3 34data2 .req x4 35data2a .req x5 36has_nul1 .req x6 37has_nul2 .req x7 38tmp1 .req x8 39tmp2 .req x9 40tmp3 .req x10 41tmp4 .req x11 42zeroones .req x12 43pos .req x13 44limit_wd .req x14 45 46#define REP8_01 0x0101010101010101 47#define REP8_7f 0x7f7f7f7f7f7f7f7f 48#define REP8_80 0x8080808080808080 49 50SYM_FUNC_START(__pi_strnlen) 51 cbz limit, .Lhit_limit 52 mov zeroones, #REP8_01 53 bic src, srcin, #15 54 ands tmp1, srcin, #15 55 b.ne .Lmisaligned 56 /* Calculate the number of full and partial words -1. */ 57 sub limit_wd, limit, #1 /* Limit != 0, so no underflow. */ 58 lsr limit_wd, limit_wd, #4 /* Convert to Qwords. */ 59 60 /* 61 * NUL detection works on the principle that (X - 1) & (~X) & 0x80 62 * (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and 63 * can be done in parallel across the entire word. 64 */ 65 /* 66 * The inner loop deals with two Dwords at a time. This has a 67 * slightly higher start-up cost, but we should win quite quickly, 68 * especially on cores with a high number of issue slots per 69 * cycle, as we get much better parallelism out of the operations. 70 */ 71.Lloop: 72 ldp data1, data2, [src], #16 73.Lrealigned: 74 sub tmp1, data1, zeroones 75 orr tmp2, data1, #REP8_7f 76 sub tmp3, data2, zeroones 77 orr tmp4, data2, #REP8_7f 78 bic has_nul1, tmp1, tmp2 79 bic has_nul2, tmp3, tmp4 80 subs limit_wd, limit_wd, #1 81 orr tmp1, has_nul1, has_nul2 82 ccmp tmp1, #0, #0, pl /* NZCV = 0000 */ 83 b.eq .Lloop 84 85 cbz tmp1, .Lhit_limit /* No null in final Qword. */ 86 87 /* 88 * We know there's a null in the final Qword. The easiest thing 89 * to do now is work out the length of the string and return 90 * MIN (len, limit). 91 */ 92 sub len, src, srcin 93 cbz has_nul1, .Lnul_in_data2 94 95 sub len, len, #8 96 mov has_nul2, has_nul1 97.Lnul_in_data2: 98 sub len, len, #8 99 rev has_nul2, has_nul2 100 clz pos, has_nul2 101 add len, len, pos, lsr #3 /* Bits to bytes. */ 102 cmp len, limit 103 csel len, len, limit, ls /* Return the lower value. */ 104 ret 105 106.Lmisaligned: 107 /* 108 * Deal with a partial first word. 109 * We're doing two things in parallel here; 110 * 1) Calculate the number of words (but avoiding overflow if 111 * limit is near ULONG_MAX) - to do this we need to work out 112 * limit + tmp1 - 1 as a 65-bit value before shifting it; 113 * 2) Load and mask the initial data words - we force the bytes 114 * before the ones we are interested in to 0xff - this ensures 115 * early bytes will not hit any zero detection. 116 */ 117 ldp data1, data2, [src], #16 118 119 sub limit_wd, limit, #1 120 and tmp3, limit_wd, #15 121 lsr limit_wd, limit_wd, #4 122 123 add tmp3, tmp3, tmp1 124 add limit_wd, limit_wd, tmp3, lsr #4 125 126 neg tmp4, tmp1 127 lsl tmp4, tmp4, #3 /* Bytes beyond alignment -> bits. */ 128 129 mov tmp2, #~0 130 /* Little-endian. Early bytes are at LSB. */ 131 lsr tmp2, tmp2, tmp4 /* Shift (tmp1 & 63). */ 132 133 cmp tmp1, #8 134 135 orr data1, data1, tmp2 136 orr data2a, data2, tmp2 137 138 csinv data1, data1, xzr, le 139 csel data2, data2, data2a, le 140 b .Lrealigned 141 142.Lhit_limit: 143 mov len, limit 144 ret 145SYM_FUNC_END(__pi_strnlen) 146 147SYM_FUNC_ALIAS_WEAK(strnlen, __pi_strnlen) 148EXPORT_SYMBOL_NOKASAN(strnlen) 149