]>
Commit | Line | Data |
---|---|---|
8fefde90 AB |
1 | /* |
2 | * Accelerated CRC32(C) using arm64 NEON and Crypto Extensions instructions | |
3 | * | |
15c7d8f8 | 4 | * Copyright (C) 2016 - 2017 Linaro Ltd <ard.biesheuvel@linaro.org> |
8fefde90 AB |
5 | * |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 as | |
8 | * published by the Free Software Foundation. | |
9 | */ | |
10 | ||
11 | #include <linux/cpufeature.h> | |
12 | #include <linux/crc32.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/kernel.h> | |
15 | #include <linux/module.h> | |
16 | #include <linux/string.h> | |
17 | ||
18 | #include <crypto/internal/hash.h> | |
19 | ||
20 | #include <asm/hwcap.h> | |
21 | #include <asm/neon.h> | |
15c7d8f8 | 22 | #include <asm/simd.h> |
8fefde90 AB |
23 | #include <asm/unaligned.h> |
24 | ||
25 | #define PMULL_MIN_LEN 64L /* minimum size of buffer | |
26 | * for crc32_pmull_le_16 */ | |
27 | #define SCALE_F 16L /* size of NEON register */ | |
28 | ||
29 | asmlinkage u32 crc32_pmull_le(const u8 buf[], u64 len, u32 init_crc); | |
30 | asmlinkage u32 crc32_armv8_le(u32 init_crc, const u8 buf[], size_t len); | |
31 | ||
32 | asmlinkage u32 crc32c_pmull_le(const u8 buf[], u64 len, u32 init_crc); | |
33 | asmlinkage u32 crc32c_armv8_le(u32 init_crc, const u8 buf[], size_t len); | |
34 | ||
35 | static u32 (*fallback_crc32)(u32 init_crc, const u8 buf[], size_t len); | |
36 | static u32 (*fallback_crc32c)(u32 init_crc, const u8 buf[], size_t len); | |
37 | ||
38 | static int crc32_pmull_cra_init(struct crypto_tfm *tfm) | |
39 | { | |
40 | u32 *key = crypto_tfm_ctx(tfm); | |
41 | ||
42 | *key = 0; | |
43 | return 0; | |
44 | } | |
45 | ||
46 | static int crc32c_pmull_cra_init(struct crypto_tfm *tfm) | |
47 | { | |
48 | u32 *key = crypto_tfm_ctx(tfm); | |
49 | ||
50 | *key = ~0; | |
51 | return 0; | |
52 | } | |
53 | ||
54 | static int crc32_pmull_setkey(struct crypto_shash *hash, const u8 *key, | |
55 | unsigned int keylen) | |
56 | { | |
57 | u32 *mctx = crypto_shash_ctx(hash); | |
58 | ||
59 | if (keylen != sizeof(u32)) { | |
60 | crypto_shash_set_flags(hash, CRYPTO_TFM_RES_BAD_KEY_LEN); | |
61 | return -EINVAL; | |
62 | } | |
63 | *mctx = le32_to_cpup((__le32 *)key); | |
64 | return 0; | |
65 | } | |
66 | ||
67 | static int crc32_pmull_init(struct shash_desc *desc) | |
68 | { | |
69 | u32 *mctx = crypto_shash_ctx(desc->tfm); | |
70 | u32 *crc = shash_desc_ctx(desc); | |
71 | ||
72 | *crc = *mctx; | |
73 | return 0; | |
74 | } | |
75 | ||
5d3d9c8b AB |
76 | static int crc32_update(struct shash_desc *desc, const u8 *data, |
77 | unsigned int length) | |
78 | { | |
79 | u32 *crc = shash_desc_ctx(desc); | |
80 | ||
81 | *crc = crc32_armv8_le(*crc, data, length); | |
82 | return 0; | |
83 | } | |
84 | ||
85 | static int crc32c_update(struct shash_desc *desc, const u8 *data, | |
86 | unsigned int length) | |
87 | { | |
88 | u32 *crc = shash_desc_ctx(desc); | |
89 | ||
90 | *crc = crc32c_armv8_le(*crc, data, length); | |
91 | return 0; | |
92 | } | |
93 | ||
8fefde90 AB |
94 | static int crc32_pmull_update(struct shash_desc *desc, const u8 *data, |
95 | unsigned int length) | |
96 | { | |
97 | u32 *crc = shash_desc_ctx(desc); | |
98 | unsigned int l; | |
99 | ||
100 | if ((u64)data % SCALE_F) { | |
101 | l = min_t(u32, length, SCALE_F - ((u64)data % SCALE_F)); | |
102 | ||
103 | *crc = fallback_crc32(*crc, data, l); | |
104 | ||
105 | data += l; | |
106 | length -= l; | |
107 | } | |
108 | ||
15c7d8f8 | 109 | if (length >= PMULL_MIN_LEN && may_use_simd()) { |
8fefde90 AB |
110 | l = round_down(length, SCALE_F); |
111 | ||
15c7d8f8 | 112 | kernel_neon_begin(); |
8fefde90 AB |
113 | *crc = crc32_pmull_le(data, l, *crc); |
114 | kernel_neon_end(); | |
115 | ||
116 | data += l; | |
117 | length -= l; | |
118 | } | |
119 | ||
120 | if (length > 0) | |
121 | *crc = fallback_crc32(*crc, data, length); | |
122 | ||
123 | return 0; | |
124 | } | |
125 | ||
126 | static int crc32c_pmull_update(struct shash_desc *desc, const u8 *data, | |
127 | unsigned int length) | |
128 | { | |
129 | u32 *crc = shash_desc_ctx(desc); | |
130 | unsigned int l; | |
131 | ||
132 | if ((u64)data % SCALE_F) { | |
133 | l = min_t(u32, length, SCALE_F - ((u64)data % SCALE_F)); | |
134 | ||
135 | *crc = fallback_crc32c(*crc, data, l); | |
136 | ||
137 | data += l; | |
138 | length -= l; | |
139 | } | |
140 | ||
15c7d8f8 | 141 | if (length >= PMULL_MIN_LEN && may_use_simd()) { |
8fefde90 AB |
142 | l = round_down(length, SCALE_F); |
143 | ||
15c7d8f8 | 144 | kernel_neon_begin(); |
8fefde90 AB |
145 | *crc = crc32c_pmull_le(data, l, *crc); |
146 | kernel_neon_end(); | |
147 | ||
148 | data += l; | |
149 | length -= l; | |
150 | } | |
151 | ||
152 | if (length > 0) { | |
153 | *crc = fallback_crc32c(*crc, data, length); | |
154 | } | |
155 | ||
156 | return 0; | |
157 | } | |
158 | ||
159 | static int crc32_pmull_final(struct shash_desc *desc, u8 *out) | |
160 | { | |
161 | u32 *crc = shash_desc_ctx(desc); | |
162 | ||
163 | put_unaligned_le32(*crc, out); | |
164 | return 0; | |
165 | } | |
166 | ||
167 | static int crc32c_pmull_final(struct shash_desc *desc, u8 *out) | |
168 | { | |
169 | u32 *crc = shash_desc_ctx(desc); | |
170 | ||
171 | put_unaligned_le32(~*crc, out); | |
172 | return 0; | |
173 | } | |
174 | ||
175 | static struct shash_alg crc32_pmull_algs[] = { { | |
176 | .setkey = crc32_pmull_setkey, | |
177 | .init = crc32_pmull_init, | |
5d3d9c8b | 178 | .update = crc32_update, |
8fefde90 AB |
179 | .final = crc32_pmull_final, |
180 | .descsize = sizeof(u32), | |
181 | .digestsize = sizeof(u32), | |
182 | ||
183 | .base.cra_ctxsize = sizeof(u32), | |
184 | .base.cra_init = crc32_pmull_cra_init, | |
185 | .base.cra_name = "crc32", | |
186 | .base.cra_driver_name = "crc32-arm64-ce", | |
187 | .base.cra_priority = 200, | |
b5c690bf | 188 | .base.cra_flags = CRYPTO_ALG_OPTIONAL_KEY, |
8fefde90 AB |
189 | .base.cra_blocksize = 1, |
190 | .base.cra_module = THIS_MODULE, | |
191 | }, { | |
192 | .setkey = crc32_pmull_setkey, | |
193 | .init = crc32_pmull_init, | |
5d3d9c8b | 194 | .update = crc32c_update, |
8fefde90 AB |
195 | .final = crc32c_pmull_final, |
196 | .descsize = sizeof(u32), | |
197 | .digestsize = sizeof(u32), | |
198 | ||
199 | .base.cra_ctxsize = sizeof(u32), | |
200 | .base.cra_init = crc32c_pmull_cra_init, | |
201 | .base.cra_name = "crc32c", | |
202 | .base.cra_driver_name = "crc32c-arm64-ce", | |
203 | .base.cra_priority = 200, | |
b5c690bf | 204 | .base.cra_flags = CRYPTO_ALG_OPTIONAL_KEY, |
8fefde90 AB |
205 | .base.cra_blocksize = 1, |
206 | .base.cra_module = THIS_MODULE, | |
207 | } }; | |
208 | ||
209 | static int __init crc32_pmull_mod_init(void) | |
210 | { | |
5d3d9c8b AB |
211 | if (IS_ENABLED(CONFIG_KERNEL_MODE_NEON) && (elf_hwcap & HWCAP_PMULL)) { |
212 | crc32_pmull_algs[0].update = crc32_pmull_update; | |
213 | crc32_pmull_algs[1].update = crc32c_pmull_update; | |
214 | ||
215 | if (elf_hwcap & HWCAP_CRC32) { | |
216 | fallback_crc32 = crc32_armv8_le; | |
217 | fallback_crc32c = crc32c_armv8_le; | |
218 | } else { | |
219 | fallback_crc32 = crc32_le; | |
220 | fallback_crc32c = __crc32c_le; | |
221 | } | |
222 | } else if (!(elf_hwcap & HWCAP_CRC32)) { | |
223 | return -ENODEV; | |
8fefde90 | 224 | } |
8fefde90 AB |
225 | return crypto_register_shashes(crc32_pmull_algs, |
226 | ARRAY_SIZE(crc32_pmull_algs)); | |
227 | } | |
228 | ||
229 | static void __exit crc32_pmull_mod_exit(void) | |
230 | { | |
231 | crypto_unregister_shashes(crc32_pmull_algs, | |
232 | ARRAY_SIZE(crc32_pmull_algs)); | |
233 | } | |
234 | ||
5d3d9c8b AB |
235 | static const struct cpu_feature crc32_cpu_feature[] = { |
236 | { cpu_feature(CRC32) }, { cpu_feature(PMULL) }, { } | |
237 | }; | |
238 | MODULE_DEVICE_TABLE(cpu, crc32_cpu_feature); | |
239 | ||
240 | module_init(crc32_pmull_mod_init); | |
8fefde90 AB |
241 | module_exit(crc32_pmull_mod_exit); |
242 | ||
243 | MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>"); | |
244 | MODULE_LICENSE("GPL v2"); |