1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* |
3 | * Crypto-API module for CRC-32 algorithms implemented with the |
4 | * z/Architecture Vector Extension Facility. |
5 | * |
6 | * Copyright IBM Corp. 2015 |
7 | * Author(s): Hendrik Brueckner <brueckner@linux.vnet.ibm.com> |
8 | */ |
9 | #define KMSG_COMPONENT "crc32-vx" |
10 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt |
11 | |
12 | #include <linux/module.h> |
13 | #include <linux/cpufeature.h> |
14 | #include <linux/crc32.h> |
15 | #include <crypto/internal/hash.h> |
16 | #include <asm/fpu.h> |
17 | #include "crc32-vx.h" |
18 | |
19 | #define CRC32_BLOCK_SIZE 1 |
20 | #define CRC32_DIGEST_SIZE 4 |
21 | |
22 | #define VX_MIN_LEN 64 |
23 | #define VX_ALIGNMENT 16L |
24 | #define VX_ALIGN_MASK (VX_ALIGNMENT - 1) |
25 | |
26 | struct crc_ctx { |
27 | u32 key; |
28 | }; |
29 | |
30 | struct crc_desc_ctx { |
31 | u32 crc; |
32 | }; |
33 | |
34 | /* |
35 | * DEFINE_CRC32_VX() - Define a CRC-32 function using the vector extension |
36 | * |
37 | * Creates a function to perform a particular CRC-32 computation. Depending |
38 | * on the message buffer, the hardware-accelerated or software implementation |
39 | * is used. Note that the message buffer is aligned to improve fetch |
40 | * operations of VECTOR LOAD MULTIPLE instructions. |
41 | * |
42 | */ |
43 | #define DEFINE_CRC32_VX(___fname, ___crc32_vx, ___crc32_sw) \ |
44 | static u32 __pure ___fname(u32 crc, \ |
45 | unsigned char const *data, size_t datalen) \ |
46 | { \ |
47 | unsigned long prealign, aligned, remaining; \ |
48 | DECLARE_KERNEL_FPU_ONSTACK16(vxstate); \ |
49 | \ |
50 | if (datalen < VX_MIN_LEN + VX_ALIGN_MASK) \ |
51 | return ___crc32_sw(crc, data, datalen); \ |
52 | \ |
53 | if ((unsigned long)data & VX_ALIGN_MASK) { \ |
54 | prealign = VX_ALIGNMENT - \ |
55 | ((unsigned long)data & VX_ALIGN_MASK); \ |
56 | datalen -= prealign; \ |
57 | crc = ___crc32_sw(crc, data, prealign); \ |
58 | data = (void *)((unsigned long)data + prealign); \ |
59 | } \ |
60 | \ |
61 | aligned = datalen & ~VX_ALIGN_MASK; \ |
62 | remaining = datalen & VX_ALIGN_MASK; \ |
63 | \ |
64 | kernel_fpu_begin(&vxstate, KERNEL_VXR_LOW); \ |
65 | crc = ___crc32_vx(crc, data, aligned); \ |
66 | kernel_fpu_end(&vxstate, KERNEL_VXR_LOW); \ |
67 | \ |
68 | if (remaining) \ |
69 | crc = ___crc32_sw(crc, data + aligned, remaining); \ |
70 | \ |
71 | return crc; \ |
72 | } |
73 | |
74 | DEFINE_CRC32_VX(crc32_le_vx, crc32_le_vgfm_16, crc32_le) |
75 | DEFINE_CRC32_VX(crc32_be_vx, crc32_be_vgfm_16, crc32_be) |
76 | DEFINE_CRC32_VX(crc32c_le_vx, crc32c_le_vgfm_16, __crc32c_le) |
77 | |
78 | |
79 | static int crc32_vx_cra_init_zero(struct crypto_tfm *tfm) |
80 | { |
81 | struct crc_ctx *mctx = crypto_tfm_ctx(tfm); |
82 | |
83 | mctx->key = 0; |
84 | return 0; |
85 | } |
86 | |
87 | static int crc32_vx_cra_init_invert(struct crypto_tfm *tfm) |
88 | { |
89 | struct crc_ctx *mctx = crypto_tfm_ctx(tfm); |
90 | |
91 | mctx->key = ~0; |
92 | return 0; |
93 | } |
94 | |
95 | static int crc32_vx_init(struct shash_desc *desc) |
96 | { |
97 | struct crc_ctx *mctx = crypto_shash_ctx(tfm: desc->tfm); |
98 | struct crc_desc_ctx *ctx = shash_desc_ctx(desc); |
99 | |
100 | ctx->crc = mctx->key; |
101 | return 0; |
102 | } |
103 | |
104 | static int crc32_vx_setkey(struct crypto_shash *tfm, const u8 *newkey, |
105 | unsigned int newkeylen) |
106 | { |
107 | struct crc_ctx *mctx = crypto_shash_ctx(tfm); |
108 | |
109 | if (newkeylen != sizeof(mctx->key)) |
110 | return -EINVAL; |
111 | mctx->key = le32_to_cpu(*(__le32 *)newkey); |
112 | return 0; |
113 | } |
114 | |
115 | static int crc32be_vx_setkey(struct crypto_shash *tfm, const u8 *newkey, |
116 | unsigned int newkeylen) |
117 | { |
118 | struct crc_ctx *mctx = crypto_shash_ctx(tfm); |
119 | |
120 | if (newkeylen != sizeof(mctx->key)) |
121 | return -EINVAL; |
122 | mctx->key = be32_to_cpu(*(__be32 *)newkey); |
123 | return 0; |
124 | } |
125 | |
126 | static int crc32le_vx_final(struct shash_desc *desc, u8 *out) |
127 | { |
128 | struct crc_desc_ctx *ctx = shash_desc_ctx(desc); |
129 | |
130 | *(__le32 *)out = cpu_to_le32p(p: &ctx->crc); |
131 | return 0; |
132 | } |
133 | |
134 | static int crc32be_vx_final(struct shash_desc *desc, u8 *out) |
135 | { |
136 | struct crc_desc_ctx *ctx = shash_desc_ctx(desc); |
137 | |
138 | *(__be32 *)out = cpu_to_be32p(p: &ctx->crc); |
139 | return 0; |
140 | } |
141 | |
142 | static int crc32c_vx_final(struct shash_desc *desc, u8 *out) |
143 | { |
144 | struct crc_desc_ctx *ctx = shash_desc_ctx(desc); |
145 | |
146 | /* |
147 | * Perform a final XOR with 0xFFFFFFFF to be in sync |
148 | * with the generic crc32c shash implementation. |
149 | */ |
150 | *(__le32 *)out = ~cpu_to_le32p(p: &ctx->crc); |
151 | return 0; |
152 | } |
153 | |
154 | static int __crc32le_vx_finup(u32 *crc, const u8 *data, unsigned int len, |
155 | u8 *out) |
156 | { |
157 | *(__le32 *)out = cpu_to_le32(crc32_le_vx(*crc, data, len)); |
158 | return 0; |
159 | } |
160 | |
161 | static int __crc32be_vx_finup(u32 *crc, const u8 *data, unsigned int len, |
162 | u8 *out) |
163 | { |
164 | *(__be32 *)out = cpu_to_be32(crc32_be_vx(*crc, data, len)); |
165 | return 0; |
166 | } |
167 | |
168 | static int __crc32c_vx_finup(u32 *crc, const u8 *data, unsigned int len, |
169 | u8 *out) |
170 | { |
171 | /* |
172 | * Perform a final XOR with 0xFFFFFFFF to be in sync |
173 | * with the generic crc32c shash implementation. |
174 | */ |
175 | *(__le32 *)out = ~cpu_to_le32(crc32c_le_vx(*crc, data, len)); |
176 | return 0; |
177 | } |
178 | |
179 | |
180 | #define CRC32_VX_FINUP(alg, func) \ |
181 | static int alg ## _vx_finup(struct shash_desc *desc, const u8 *data, \ |
182 | unsigned int datalen, u8 *out) \ |
183 | { \ |
184 | return __ ## alg ## _vx_finup(shash_desc_ctx(desc), \ |
185 | data, datalen, out); \ |
186 | } |
187 | |
188 | CRC32_VX_FINUP(crc32le, crc32_le_vx) |
189 | CRC32_VX_FINUP(crc32be, crc32_be_vx) |
190 | CRC32_VX_FINUP(crc32c, crc32c_le_vx) |
191 | |
192 | #define CRC32_VX_DIGEST(alg, func) \ |
193 | static int alg ## _vx_digest(struct shash_desc *desc, const u8 *data, \ |
194 | unsigned int len, u8 *out) \ |
195 | { \ |
196 | return __ ## alg ## _vx_finup(crypto_shash_ctx(desc->tfm), \ |
197 | data, len, out); \ |
198 | } |
199 | |
200 | CRC32_VX_DIGEST(crc32le, crc32_le_vx) |
201 | CRC32_VX_DIGEST(crc32be, crc32_be_vx) |
202 | CRC32_VX_DIGEST(crc32c, crc32c_le_vx) |
203 | |
204 | #define CRC32_VX_UPDATE(alg, func) \ |
205 | static int alg ## _vx_update(struct shash_desc *desc, const u8 *data, \ |
206 | unsigned int datalen) \ |
207 | { \ |
208 | struct crc_desc_ctx *ctx = shash_desc_ctx(desc); \ |
209 | ctx->crc = func(ctx->crc, data, datalen); \ |
210 | return 0; \ |
211 | } |
212 | |
213 | CRC32_VX_UPDATE(crc32le, crc32_le_vx) |
214 | CRC32_VX_UPDATE(crc32be, crc32_be_vx) |
215 | CRC32_VX_UPDATE(crc32c, crc32c_le_vx) |
216 | |
217 | |
218 | static struct shash_alg crc32_vx_algs[] = { |
219 | /* CRC-32 LE */ |
220 | { |
221 | .init = crc32_vx_init, |
222 | .setkey = crc32_vx_setkey, |
223 | .update = crc32le_vx_update, |
224 | .final = crc32le_vx_final, |
225 | .finup = crc32le_vx_finup, |
226 | .digest = crc32le_vx_digest, |
227 | .descsize = sizeof(struct crc_desc_ctx), |
228 | .digestsize = CRC32_DIGEST_SIZE, |
229 | .base = { |
230 | .cra_name = "crc32" , |
231 | .cra_driver_name = "crc32-vx" , |
232 | .cra_priority = 200, |
233 | .cra_flags = CRYPTO_ALG_OPTIONAL_KEY, |
234 | .cra_blocksize = CRC32_BLOCK_SIZE, |
235 | .cra_ctxsize = sizeof(struct crc_ctx), |
236 | .cra_module = THIS_MODULE, |
237 | .cra_init = crc32_vx_cra_init_zero, |
238 | }, |
239 | }, |
240 | /* CRC-32 BE */ |
241 | { |
242 | .init = crc32_vx_init, |
243 | .setkey = crc32be_vx_setkey, |
244 | .update = crc32be_vx_update, |
245 | .final = crc32be_vx_final, |
246 | .finup = crc32be_vx_finup, |
247 | .digest = crc32be_vx_digest, |
248 | .descsize = sizeof(struct crc_desc_ctx), |
249 | .digestsize = CRC32_DIGEST_SIZE, |
250 | .base = { |
251 | .cra_name = "crc32be" , |
252 | .cra_driver_name = "crc32be-vx" , |
253 | .cra_priority = 200, |
254 | .cra_flags = CRYPTO_ALG_OPTIONAL_KEY, |
255 | .cra_blocksize = CRC32_BLOCK_SIZE, |
256 | .cra_ctxsize = sizeof(struct crc_ctx), |
257 | .cra_module = THIS_MODULE, |
258 | .cra_init = crc32_vx_cra_init_zero, |
259 | }, |
260 | }, |
261 | /* CRC-32C LE */ |
262 | { |
263 | .init = crc32_vx_init, |
264 | .setkey = crc32_vx_setkey, |
265 | .update = crc32c_vx_update, |
266 | .final = crc32c_vx_final, |
267 | .finup = crc32c_vx_finup, |
268 | .digest = crc32c_vx_digest, |
269 | .descsize = sizeof(struct crc_desc_ctx), |
270 | .digestsize = CRC32_DIGEST_SIZE, |
271 | .base = { |
272 | .cra_name = "crc32c" , |
273 | .cra_driver_name = "crc32c-vx" , |
274 | .cra_priority = 200, |
275 | .cra_flags = CRYPTO_ALG_OPTIONAL_KEY, |
276 | .cra_blocksize = CRC32_BLOCK_SIZE, |
277 | .cra_ctxsize = sizeof(struct crc_ctx), |
278 | .cra_module = THIS_MODULE, |
279 | .cra_init = crc32_vx_cra_init_invert, |
280 | }, |
281 | }, |
282 | }; |
283 | |
284 | |
285 | static int __init crc_vx_mod_init(void) |
286 | { |
287 | return crypto_register_shashes(algs: crc32_vx_algs, |
288 | ARRAY_SIZE(crc32_vx_algs)); |
289 | } |
290 | |
291 | static void __exit crc_vx_mod_exit(void) |
292 | { |
293 | crypto_unregister_shashes(algs: crc32_vx_algs, ARRAY_SIZE(crc32_vx_algs)); |
294 | } |
295 | |
296 | module_cpu_feature_match(S390_CPU_FEATURE_VXRS, crc_vx_mod_init); |
297 | module_exit(crc_vx_mod_exit); |
298 | |
299 | MODULE_AUTHOR("Hendrik Brueckner <brueckner@linux.vnet.ibm.com>" ); |
300 | MODULE_LICENSE("GPL" ); |
301 | |
302 | MODULE_ALIAS_CRYPTO("crc32" ); |
303 | MODULE_ALIAS_CRYPTO("crc32-vx" ); |
304 | MODULE_ALIAS_CRYPTO("crc32c" ); |
305 | MODULE_ALIAS_CRYPTO("crc32c-vx" ); |
306 | |