Commit | Line | Data |
---|---|---|
63b94509 TL |
1 | /* |
2 | * AMD Cryptographic Coprocessor (CCP) driver | |
3 | * | |
4 | * Copyright (C) 2013 Advanced Micro Devices, Inc. | |
5 | * | |
6 | * Author: Tom Lendacky <thomas.lendacky@amd.com> | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or modify | |
9 | * it under the terms of the GNU General Public License version 2 as | |
10 | * published by the Free Software Foundation. | |
11 | */ | |
12 | ||
13 | #include <linux/module.h> | |
14 | #include <linux/kernel.h> | |
15 | #include <linux/pci.h> | |
16 | #include <linux/pci_ids.h> | |
17 | #include <linux/kthread.h> | |
18 | #include <linux/sched.h> | |
19 | #include <linux/interrupt.h> | |
20 | #include <linux/spinlock.h> | |
21 | #include <linux/mutex.h> | |
22 | #include <linux/delay.h> | |
23 | #include <linux/ccp.h> | |
24 | #include <linux/scatterlist.h> | |
25 | #include <crypto/scatterwalk.h> | |
26 | ||
27 | #include "ccp-dev.h" | |
28 | ||
29 | ||
30 | enum ccp_memtype { | |
31 | CCP_MEMTYPE_SYSTEM = 0, | |
32 | CCP_MEMTYPE_KSB, | |
33 | CCP_MEMTYPE_LOCAL, | |
34 | CCP_MEMTYPE__LAST, | |
35 | }; | |
36 | ||
37 | struct ccp_dma_info { | |
38 | dma_addr_t address; | |
39 | unsigned int offset; | |
40 | unsigned int length; | |
41 | enum dma_data_direction dir; | |
42 | }; | |
43 | ||
44 | struct ccp_dm_workarea { | |
45 | struct device *dev; | |
46 | struct dma_pool *dma_pool; | |
47 | unsigned int length; | |
48 | ||
49 | u8 *address; | |
50 | struct ccp_dma_info dma; | |
51 | }; | |
52 | ||
53 | struct ccp_sg_workarea { | |
54 | struct scatterlist *sg; | |
55 | unsigned int nents; | |
56 | unsigned int length; | |
57 | ||
58 | struct scatterlist *dma_sg; | |
59 | struct device *dma_dev; | |
60 | unsigned int dma_count; | |
61 | enum dma_data_direction dma_dir; | |
62 | ||
63 | u32 sg_used; | |
64 | ||
65 | u32 bytes_left; | |
66 | }; | |
67 | ||
68 | struct ccp_data { | |
69 | struct ccp_sg_workarea sg_wa; | |
70 | struct ccp_dm_workarea dm_wa; | |
71 | }; | |
72 | ||
73 | struct ccp_mem { | |
74 | enum ccp_memtype type; | |
75 | union { | |
76 | struct ccp_dma_info dma; | |
77 | u32 ksb; | |
78 | } u; | |
79 | }; | |
80 | ||
81 | struct ccp_aes_op { | |
82 | enum ccp_aes_type type; | |
83 | enum ccp_aes_mode mode; | |
84 | enum ccp_aes_action action; | |
85 | }; | |
86 | ||
87 | struct ccp_xts_aes_op { | |
88 | enum ccp_aes_action action; | |
89 | enum ccp_xts_aes_unit_size unit_size; | |
90 | }; | |
91 | ||
92 | struct ccp_sha_op { | |
93 | enum ccp_sha_type type; | |
94 | u64 msg_bits; | |
95 | }; | |
96 | ||
97 | struct ccp_rsa_op { | |
98 | u32 mod_size; | |
99 | u32 input_len; | |
100 | }; | |
101 | ||
102 | struct ccp_passthru_op { | |
103 | enum ccp_passthru_bitwise bit_mod; | |
104 | enum ccp_passthru_byteswap byte_swap; | |
105 | }; | |
106 | ||
107 | struct ccp_ecc_op { | |
108 | enum ccp_ecc_function function; | |
109 | }; | |
110 | ||
111 | struct ccp_op { | |
112 | struct ccp_cmd_queue *cmd_q; | |
113 | ||
114 | u32 jobid; | |
115 | u32 ioc; | |
116 | u32 soc; | |
117 | u32 ksb_key; | |
118 | u32 ksb_ctx; | |
119 | u32 init; | |
120 | u32 eom; | |
121 | ||
122 | struct ccp_mem src; | |
123 | struct ccp_mem dst; | |
124 | ||
125 | union { | |
126 | struct ccp_aes_op aes; | |
127 | struct ccp_xts_aes_op xts; | |
128 | struct ccp_sha_op sha; | |
129 | struct ccp_rsa_op rsa; | |
130 | struct ccp_passthru_op passthru; | |
131 | struct ccp_ecc_op ecc; | |
132 | } u; | |
133 | }; | |
134 | ||
135 | /* The CCP cannot perform zero-length sha operations so the caller | |
136 | * is required to buffer data for the final operation. However, a | |
137 | * sha operation for a message with a total length of zero is valid | |
138 | * so known values are required to supply the result. | |
139 | */ | |
140 | static const u8 ccp_sha1_zero[CCP_SHA_CTXSIZE] = { | |
141 | 0xda, 0x39, 0xa3, 0xee, 0x5e, 0x6b, 0x4b, 0x0d, | |
142 | 0x32, 0x55, 0xbf, 0xef, 0x95, 0x60, 0x18, 0x90, | |
143 | 0xaf, 0xd8, 0x07, 0x09, 0x00, 0x00, 0x00, 0x00, | |
144 | 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, | |
145 | }; | |
146 | ||
147 | static const u8 ccp_sha224_zero[CCP_SHA_CTXSIZE] = { | |
148 | 0xd1, 0x4a, 0x02, 0x8c, 0x2a, 0x3a, 0x2b, 0xc9, | |
149 | 0x47, 0x61, 0x02, 0xbb, 0x28, 0x82, 0x34, 0xc4, | |
150 | 0x15, 0xa2, 0xb0, 0x1f, 0x82, 0x8e, 0xa6, 0x2a, | |
151 | 0xc5, 0xb3, 0xe4, 0x2f, 0x00, 0x00, 0x00, 0x00, | |
152 | }; | |
153 | ||
154 | static const u8 ccp_sha256_zero[CCP_SHA_CTXSIZE] = { | |
155 | 0xe3, 0xb0, 0xc4, 0x42, 0x98, 0xfc, 0x1c, 0x14, | |
156 | 0x9a, 0xfb, 0xf4, 0xc8, 0x99, 0x6f, 0xb9, 0x24, | |
157 | 0x27, 0xae, 0x41, 0xe4, 0x64, 0x9b, 0x93, 0x4c, | |
158 | 0xa4, 0x95, 0x99, 0x1b, 0x78, 0x52, 0xb8, 0x55, | |
159 | }; | |
160 | ||
161 | static u32 ccp_addr_lo(struct ccp_dma_info *info) | |
162 | { | |
163 | return lower_32_bits(info->address + info->offset); | |
164 | } | |
165 | ||
166 | static u32 ccp_addr_hi(struct ccp_dma_info *info) | |
167 | { | |
168 | return upper_32_bits(info->address + info->offset) & 0x0000ffff; | |
169 | } | |
170 | ||
171 | static int ccp_do_cmd(struct ccp_op *op, u32 *cr, unsigned int cr_count) | |
172 | { | |
173 | struct ccp_cmd_queue *cmd_q = op->cmd_q; | |
174 | struct ccp_device *ccp = cmd_q->ccp; | |
175 | void __iomem *cr_addr; | |
176 | u32 cr0, cmd; | |
177 | unsigned int i; | |
178 | int ret = 0; | |
179 | ||
180 | /* We could read a status register to see how many free slots | |
181 | * are actually available, but reading that register resets it | |
182 | * and you could lose some error information. | |
183 | */ | |
184 | cmd_q->free_slots--; | |
185 | ||
186 | cr0 = (cmd_q->id << REQ0_CMD_Q_SHIFT) | |
187 | | (op->jobid << REQ0_JOBID_SHIFT) | |
188 | | REQ0_WAIT_FOR_WRITE; | |
189 | ||
190 | if (op->soc) | |
191 | cr0 |= REQ0_STOP_ON_COMPLETE | |
192 | | REQ0_INT_ON_COMPLETE; | |
193 | ||
194 | if (op->ioc || !cmd_q->free_slots) | |
195 | cr0 |= REQ0_INT_ON_COMPLETE; | |
196 | ||
197 | /* Start at CMD_REQ1 */ | |
198 | cr_addr = ccp->io_regs + CMD_REQ0 + CMD_REQ_INCR; | |
199 | ||
200 | mutex_lock(&ccp->req_mutex); | |
201 | ||
202 | /* Write CMD_REQ1 through CMD_REQx first */ | |
203 | for (i = 0; i < cr_count; i++, cr_addr += CMD_REQ_INCR) | |
204 | iowrite32(*(cr + i), cr_addr); | |
205 | ||
206 | /* Tell the CCP to start */ | |
207 | wmb(); | |
208 | iowrite32(cr0, ccp->io_regs + CMD_REQ0); | |
209 | ||
210 | mutex_unlock(&ccp->req_mutex); | |
211 | ||
212 | if (cr0 & REQ0_INT_ON_COMPLETE) { | |
213 | /* Wait for the job to complete */ | |
214 | ret = wait_event_interruptible(cmd_q->int_queue, | |
215 | cmd_q->int_rcvd); | |
216 | if (ret || cmd_q->cmd_error) { | |
217 | /* On error delete all related jobs from the queue */ | |
218 | cmd = (cmd_q->id << DEL_Q_ID_SHIFT) | |
219 | | op->jobid; | |
220 | ||
221 | iowrite32(cmd, ccp->io_regs + DEL_CMD_Q_JOB); | |
222 | ||
223 | if (!ret) | |
224 | ret = -EIO; | |
225 | } else if (op->soc) { | |
226 | /* Delete just head job from the queue on SoC */ | |
227 | cmd = DEL_Q_ACTIVE | |
228 | | (cmd_q->id << DEL_Q_ID_SHIFT) | |
229 | | op->jobid; | |
230 | ||
231 | iowrite32(cmd, ccp->io_regs + DEL_CMD_Q_JOB); | |
232 | } | |
233 | ||
234 | cmd_q->free_slots = CMD_Q_DEPTH(cmd_q->q_status); | |
235 | ||
236 | cmd_q->int_rcvd = 0; | |
237 | } | |
238 | ||
239 | return ret; | |
240 | } | |
241 | ||
242 | static int ccp_perform_aes(struct ccp_op *op) | |
243 | { | |
244 | u32 cr[6]; | |
245 | ||
246 | /* Fill out the register contents for REQ1 through REQ6 */ | |
247 | cr[0] = (CCP_ENGINE_AES << REQ1_ENGINE_SHIFT) | |
248 | | (op->u.aes.type << REQ1_AES_TYPE_SHIFT) | |
249 | | (op->u.aes.mode << REQ1_AES_MODE_SHIFT) | |
250 | | (op->u.aes.action << REQ1_AES_ACTION_SHIFT) | |
251 | | (op->ksb_key << REQ1_KEY_KSB_SHIFT); | |
252 | cr[1] = op->src.u.dma.length - 1; | |
253 | cr[2] = ccp_addr_lo(&op->src.u.dma); | |
254 | cr[3] = (op->ksb_ctx << REQ4_KSB_SHIFT) | |
255 | | (CCP_MEMTYPE_SYSTEM << REQ4_MEMTYPE_SHIFT) | |
256 | | ccp_addr_hi(&op->src.u.dma); | |
257 | cr[4] = ccp_addr_lo(&op->dst.u.dma); | |
258 | cr[5] = (CCP_MEMTYPE_SYSTEM << REQ6_MEMTYPE_SHIFT) | |
259 | | ccp_addr_hi(&op->dst.u.dma); | |
260 | ||
261 | if (op->u.aes.mode == CCP_AES_MODE_CFB) | |
262 | cr[0] |= ((0x7f) << REQ1_AES_CFB_SIZE_SHIFT); | |
263 | ||
264 | if (op->eom) | |
265 | cr[0] |= REQ1_EOM; | |
266 | ||
267 | if (op->init) | |
268 | cr[0] |= REQ1_INIT; | |
269 | ||
270 | return ccp_do_cmd(op, cr, ARRAY_SIZE(cr)); | |
271 | } | |
272 | ||
273 | static int ccp_perform_xts_aes(struct ccp_op *op) | |
274 | { | |
275 | u32 cr[6]; | |
276 | ||
277 | /* Fill out the register contents for REQ1 through REQ6 */ | |
278 | cr[0] = (CCP_ENGINE_XTS_AES_128 << REQ1_ENGINE_SHIFT) | |
279 | | (op->u.xts.action << REQ1_AES_ACTION_SHIFT) | |
280 | | (op->u.xts.unit_size << REQ1_XTS_AES_SIZE_SHIFT) | |
281 | | (op->ksb_key << REQ1_KEY_KSB_SHIFT); | |
282 | cr[1] = op->src.u.dma.length - 1; | |
283 | cr[2] = ccp_addr_lo(&op->src.u.dma); | |
284 | cr[3] = (op->ksb_ctx << REQ4_KSB_SHIFT) | |
285 | | (CCP_MEMTYPE_SYSTEM << REQ4_MEMTYPE_SHIFT) | |
286 | | ccp_addr_hi(&op->src.u.dma); | |
287 | cr[4] = ccp_addr_lo(&op->dst.u.dma); | |
288 | cr[5] = (CCP_MEMTYPE_SYSTEM << REQ6_MEMTYPE_SHIFT) | |
289 | | ccp_addr_hi(&op->dst.u.dma); | |
290 | ||
291 | if (op->eom) | |
292 | cr[0] |= REQ1_EOM; | |
293 | ||
294 | if (op->init) | |
295 | cr[0] |= REQ1_INIT; | |
296 | ||
297 | return ccp_do_cmd(op, cr, ARRAY_SIZE(cr)); | |
298 | } | |
299 | ||
300 | static int ccp_perform_sha(struct ccp_op *op) | |
301 | { | |
302 | u32 cr[6]; | |
303 | ||
304 | /* Fill out the register contents for REQ1 through REQ6 */ | |
305 | cr[0] = (CCP_ENGINE_SHA << REQ1_ENGINE_SHIFT) | |
306 | | (op->u.sha.type << REQ1_SHA_TYPE_SHIFT) | |
307 | | REQ1_INIT; | |
308 | cr[1] = op->src.u.dma.length - 1; | |
309 | cr[2] = ccp_addr_lo(&op->src.u.dma); | |
310 | cr[3] = (op->ksb_ctx << REQ4_KSB_SHIFT) | |
311 | | (CCP_MEMTYPE_SYSTEM << REQ4_MEMTYPE_SHIFT) | |
312 | | ccp_addr_hi(&op->src.u.dma); | |
313 | ||
314 | if (op->eom) { | |
315 | cr[0] |= REQ1_EOM; | |
316 | cr[4] = lower_32_bits(op->u.sha.msg_bits); | |
317 | cr[5] = upper_32_bits(op->u.sha.msg_bits); | |
318 | } else { | |
319 | cr[4] = 0; | |
320 | cr[5] = 0; | |
321 | } | |
322 | ||
323 | return ccp_do_cmd(op, cr, ARRAY_SIZE(cr)); | |
324 | } | |
325 | ||
326 | static int ccp_perform_rsa(struct ccp_op *op) | |
327 | { | |
328 | u32 cr[6]; | |
329 | ||
330 | /* Fill out the register contents for REQ1 through REQ6 */ | |
331 | cr[0] = (CCP_ENGINE_RSA << REQ1_ENGINE_SHIFT) | |
332 | | (op->u.rsa.mod_size << REQ1_RSA_MOD_SIZE_SHIFT) | |
333 | | (op->ksb_key << REQ1_KEY_KSB_SHIFT) | |
334 | | REQ1_EOM; | |
335 | cr[1] = op->u.rsa.input_len - 1; | |
336 | cr[2] = ccp_addr_lo(&op->src.u.dma); | |
337 | cr[3] = (op->ksb_ctx << REQ4_KSB_SHIFT) | |
338 | | (CCP_MEMTYPE_SYSTEM << REQ4_MEMTYPE_SHIFT) | |
339 | | ccp_addr_hi(&op->src.u.dma); | |
340 | cr[4] = ccp_addr_lo(&op->dst.u.dma); | |
341 | cr[5] = (CCP_MEMTYPE_SYSTEM << REQ6_MEMTYPE_SHIFT) | |
342 | | ccp_addr_hi(&op->dst.u.dma); | |
343 | ||
344 | return ccp_do_cmd(op, cr, ARRAY_SIZE(cr)); | |
345 | } | |
346 | ||
347 | static int ccp_perform_passthru(struct ccp_op *op) | |
348 | { | |
349 | u32 cr[6]; | |
350 | ||
351 | /* Fill out the register contents for REQ1 through REQ6 */ | |
352 | cr[0] = (CCP_ENGINE_PASSTHRU << REQ1_ENGINE_SHIFT) | |
353 | | (op->u.passthru.bit_mod << REQ1_PT_BW_SHIFT) | |
354 | | (op->u.passthru.byte_swap << REQ1_PT_BS_SHIFT); | |
355 | ||
356 | if (op->src.type == CCP_MEMTYPE_SYSTEM) | |
357 | cr[1] = op->src.u.dma.length - 1; | |
358 | else | |
359 | cr[1] = op->dst.u.dma.length - 1; | |
360 | ||
361 | if (op->src.type == CCP_MEMTYPE_SYSTEM) { | |
362 | cr[2] = ccp_addr_lo(&op->src.u.dma); | |
363 | cr[3] = (CCP_MEMTYPE_SYSTEM << REQ4_MEMTYPE_SHIFT) | |
364 | | ccp_addr_hi(&op->src.u.dma); | |
365 | ||
366 | if (op->u.passthru.bit_mod != CCP_PASSTHRU_BITWISE_NOOP) | |
367 | cr[3] |= (op->ksb_key << REQ4_KSB_SHIFT); | |
368 | } else { | |
369 | cr[2] = op->src.u.ksb * CCP_KSB_BYTES; | |
370 | cr[3] = (CCP_MEMTYPE_KSB << REQ4_MEMTYPE_SHIFT); | |
371 | } | |
372 | ||
373 | if (op->dst.type == CCP_MEMTYPE_SYSTEM) { | |
374 | cr[4] = ccp_addr_lo(&op->dst.u.dma); | |
375 | cr[5] = (CCP_MEMTYPE_SYSTEM << REQ6_MEMTYPE_SHIFT) | |
376 | | ccp_addr_hi(&op->dst.u.dma); | |
377 | } else { | |
378 | cr[4] = op->dst.u.ksb * CCP_KSB_BYTES; | |
379 | cr[5] = (CCP_MEMTYPE_KSB << REQ6_MEMTYPE_SHIFT); | |
380 | } | |
381 | ||
382 | if (op->eom) | |
383 | cr[0] |= REQ1_EOM; | |
384 | ||
385 | return ccp_do_cmd(op, cr, ARRAY_SIZE(cr)); | |
386 | } | |
387 | ||
388 | static int ccp_perform_ecc(struct ccp_op *op) | |
389 | { | |
390 | u32 cr[6]; | |
391 | ||
392 | /* Fill out the register contents for REQ1 through REQ6 */ | |
393 | cr[0] = REQ1_ECC_AFFINE_CONVERT | |
394 | | (CCP_ENGINE_ECC << REQ1_ENGINE_SHIFT) | |
395 | | (op->u.ecc.function << REQ1_ECC_FUNCTION_SHIFT) | |
396 | | REQ1_EOM; | |
397 | cr[1] = op->src.u.dma.length - 1; | |
398 | cr[2] = ccp_addr_lo(&op->src.u.dma); | |
399 | cr[3] = (CCP_MEMTYPE_SYSTEM << REQ4_MEMTYPE_SHIFT) | |
400 | | ccp_addr_hi(&op->src.u.dma); | |
401 | cr[4] = ccp_addr_lo(&op->dst.u.dma); | |
402 | cr[5] = (CCP_MEMTYPE_SYSTEM << REQ6_MEMTYPE_SHIFT) | |
403 | | ccp_addr_hi(&op->dst.u.dma); | |
404 | ||
405 | return ccp_do_cmd(op, cr, ARRAY_SIZE(cr)); | |
406 | } | |
407 | ||
408 | static u32 ccp_alloc_ksb(struct ccp_device *ccp, unsigned int count) | |
409 | { | |
410 | int start; | |
411 | ||
412 | for (;;) { | |
413 | mutex_lock(&ccp->ksb_mutex); | |
414 | ||
415 | start = (u32)bitmap_find_next_zero_area(ccp->ksb, | |
416 | ccp->ksb_count, | |
417 | ccp->ksb_start, | |
418 | count, 0); | |
419 | if (start <= ccp->ksb_count) { | |
420 | bitmap_set(ccp->ksb, start, count); | |
421 | ||
422 | mutex_unlock(&ccp->ksb_mutex); | |
423 | break; | |
424 | } | |
425 | ||
426 | ccp->ksb_avail = 0; | |
427 | ||
428 | mutex_unlock(&ccp->ksb_mutex); | |
429 | ||
430 | /* Wait for KSB entries to become available */ | |
431 | if (wait_event_interruptible(ccp->ksb_queue, ccp->ksb_avail)) | |
432 | return 0; | |
433 | } | |
434 | ||
435 | return KSB_START + start; | |
436 | } | |
437 | ||
438 | static void ccp_free_ksb(struct ccp_device *ccp, unsigned int start, | |
439 | unsigned int count) | |
440 | { | |
441 | if (!start) | |
442 | return; | |
443 | ||
444 | mutex_lock(&ccp->ksb_mutex); | |
445 | ||
446 | bitmap_clear(ccp->ksb, start - KSB_START, count); | |
447 | ||
448 | ccp->ksb_avail = 1; | |
449 | ||
450 | mutex_unlock(&ccp->ksb_mutex); | |
451 | ||
452 | wake_up_interruptible_all(&ccp->ksb_queue); | |
453 | } | |
454 | ||
455 | static u32 ccp_gen_jobid(struct ccp_device *ccp) | |
456 | { | |
457 | return atomic_inc_return(&ccp->current_id) & CCP_JOBID_MASK; | |
458 | } | |
459 | ||
460 | static void ccp_sg_free(struct ccp_sg_workarea *wa) | |
461 | { | |
462 | if (wa->dma_count) | |
463 | dma_unmap_sg(wa->dma_dev, wa->dma_sg, wa->nents, wa->dma_dir); | |
464 | ||
465 | wa->dma_count = 0; | |
466 | } | |
467 | ||
468 | static int ccp_init_sg_workarea(struct ccp_sg_workarea *wa, struct device *dev, | |
469 | struct scatterlist *sg, unsigned int len, | |
470 | enum dma_data_direction dma_dir) | |
471 | { | |
472 | memset(wa, 0, sizeof(*wa)); | |
473 | ||
474 | wa->sg = sg; | |
475 | if (!sg) | |
476 | return 0; | |
477 | ||
478 | wa->nents = sg_nents(sg); | |
479 | wa->length = sg->length; | |
480 | wa->bytes_left = len; | |
481 | wa->sg_used = 0; | |
482 | ||
483 | if (len == 0) | |
484 | return 0; | |
485 | ||
486 | if (dma_dir == DMA_NONE) | |
487 | return 0; | |
488 | ||
489 | wa->dma_sg = sg; | |
490 | wa->dma_dev = dev; | |
491 | wa->dma_dir = dma_dir; | |
492 | wa->dma_count = dma_map_sg(dev, sg, wa->nents, dma_dir); | |
493 | if (!wa->dma_count) | |
494 | return -ENOMEM; | |
495 | ||
496 | ||
497 | return 0; | |
498 | } | |
499 | ||
500 | static void ccp_update_sg_workarea(struct ccp_sg_workarea *wa, unsigned int len) | |
501 | { | |
502 | unsigned int nbytes = min(len, wa->bytes_left); | |
503 | ||
504 | if (!wa->sg) | |
505 | return; | |
506 | ||
507 | wa->sg_used += nbytes; | |
508 | wa->bytes_left -= nbytes; | |
509 | if (wa->sg_used == wa->sg->length) { | |
510 | wa->sg = sg_next(wa->sg); | |
511 | wa->sg_used = 0; | |
512 | } | |
513 | } | |
514 | ||
515 | static void ccp_dm_free(struct ccp_dm_workarea *wa) | |
516 | { | |
517 | if (wa->length <= CCP_DMAPOOL_MAX_SIZE) { | |
518 | if (wa->address) | |
519 | dma_pool_free(wa->dma_pool, wa->address, | |
520 | wa->dma.address); | |
521 | } else { | |
522 | if (wa->dma.address) | |
523 | dma_unmap_single(wa->dev, wa->dma.address, wa->length, | |
524 | wa->dma.dir); | |
525 | kfree(wa->address); | |
526 | } | |
527 | ||
528 | wa->address = NULL; | |
529 | wa->dma.address = 0; | |
530 | } | |
531 | ||
532 | static int ccp_init_dm_workarea(struct ccp_dm_workarea *wa, | |
533 | struct ccp_cmd_queue *cmd_q, | |
534 | unsigned int len, | |
535 | enum dma_data_direction dir) | |
536 | { | |
537 | memset(wa, 0, sizeof(*wa)); | |
538 | ||
539 | if (!len) | |
540 | return 0; | |
541 | ||
542 | wa->dev = cmd_q->ccp->dev; | |
543 | wa->length = len; | |
544 | ||
545 | if (len <= CCP_DMAPOOL_MAX_SIZE) { | |
546 | wa->dma_pool = cmd_q->dma_pool; | |
547 | ||
548 | wa->address = dma_pool_alloc(wa->dma_pool, GFP_KERNEL, | |
549 | &wa->dma.address); | |
550 | if (!wa->address) | |
551 | return -ENOMEM; | |
552 | ||
553 | wa->dma.length = CCP_DMAPOOL_MAX_SIZE; | |
554 | ||
555 | memset(wa->address, 0, CCP_DMAPOOL_MAX_SIZE); | |
556 | } else { | |
557 | wa->address = kzalloc(len, GFP_KERNEL); | |
558 | if (!wa->address) | |
559 | return -ENOMEM; | |
560 | ||
561 | wa->dma.address = dma_map_single(wa->dev, wa->address, len, | |
562 | dir); | |
563 | if (!wa->dma.address) | |
564 | return -ENOMEM; | |
565 | ||
566 | wa->dma.length = len; | |
567 | } | |
568 | wa->dma.dir = dir; | |
569 | ||
570 | return 0; | |
571 | } | |
572 | ||
573 | static void ccp_set_dm_area(struct ccp_dm_workarea *wa, unsigned int wa_offset, | |
574 | struct scatterlist *sg, unsigned int sg_offset, | |
575 | unsigned int len) | |
576 | { | |
577 | WARN_ON(!wa->address); | |
578 | ||
579 | scatterwalk_map_and_copy(wa->address + wa_offset, sg, sg_offset, len, | |
580 | 0); | |
581 | } | |
582 | ||
583 | static void ccp_get_dm_area(struct ccp_dm_workarea *wa, unsigned int wa_offset, | |
584 | struct scatterlist *sg, unsigned int sg_offset, | |
585 | unsigned int len) | |
586 | { | |
587 | WARN_ON(!wa->address); | |
588 | ||
589 | scatterwalk_map_and_copy(wa->address + wa_offset, sg, sg_offset, len, | |
590 | 1); | |
591 | } | |
592 | ||
593 | static void ccp_reverse_set_dm_area(struct ccp_dm_workarea *wa, | |
594 | struct scatterlist *sg, | |
595 | unsigned int len, unsigned int se_len, | |
596 | bool sign_extend) | |
597 | { | |
598 | unsigned int nbytes, sg_offset, dm_offset, ksb_len, i; | |
599 | u8 buffer[CCP_REVERSE_BUF_SIZE]; | |
600 | ||
601 | BUG_ON(se_len > sizeof(buffer)); | |
602 | ||
603 | sg_offset = len; | |
604 | dm_offset = 0; | |
605 | nbytes = len; | |
606 | while (nbytes) { | |
607 | ksb_len = min_t(unsigned int, nbytes, se_len); | |
608 | sg_offset -= ksb_len; | |
609 | ||
610 | scatterwalk_map_and_copy(buffer, sg, sg_offset, ksb_len, 0); | |
611 | for (i = 0; i < ksb_len; i++) | |
612 | wa->address[dm_offset + i] = buffer[ksb_len - i - 1]; | |
613 | ||
614 | dm_offset += ksb_len; | |
615 | nbytes -= ksb_len; | |
616 | ||
617 | if ((ksb_len != se_len) && sign_extend) { | |
618 | /* Must sign-extend to nearest sign-extend length */ | |
619 | if (wa->address[dm_offset - 1] & 0x80) | |
620 | memset(wa->address + dm_offset, 0xff, | |
621 | se_len - ksb_len); | |
622 | } | |
623 | } | |
624 | } | |
625 | ||
626 | static void ccp_reverse_get_dm_area(struct ccp_dm_workarea *wa, | |
627 | struct scatterlist *sg, | |
628 | unsigned int len) | |
629 | { | |
630 | unsigned int nbytes, sg_offset, dm_offset, ksb_len, i; | |
631 | u8 buffer[CCP_REVERSE_BUF_SIZE]; | |
632 | ||
633 | sg_offset = 0; | |
634 | dm_offset = len; | |
635 | nbytes = len; | |
636 | while (nbytes) { | |
637 | ksb_len = min_t(unsigned int, nbytes, sizeof(buffer)); | |
638 | dm_offset -= ksb_len; | |
639 | ||
640 | for (i = 0; i < ksb_len; i++) | |
641 | buffer[ksb_len - i - 1] = wa->address[dm_offset + i]; | |
642 | scatterwalk_map_and_copy(buffer, sg, sg_offset, ksb_len, 1); | |
643 | ||
644 | sg_offset += ksb_len; | |
645 | nbytes -= ksb_len; | |
646 | } | |
647 | } | |
648 | ||
649 | static void ccp_free_data(struct ccp_data *data, struct ccp_cmd_queue *cmd_q) | |
650 | { | |
651 | ccp_dm_free(&data->dm_wa); | |
652 | ccp_sg_free(&data->sg_wa); | |
653 | } | |
654 | ||
655 | static int ccp_init_data(struct ccp_data *data, struct ccp_cmd_queue *cmd_q, | |
656 | struct scatterlist *sg, unsigned int sg_len, | |
657 | unsigned int dm_len, | |
658 | enum dma_data_direction dir) | |
659 | { | |
660 | int ret; | |
661 | ||
662 | memset(data, 0, sizeof(*data)); | |
663 | ||
664 | ret = ccp_init_sg_workarea(&data->sg_wa, cmd_q->ccp->dev, sg, sg_len, | |
665 | dir); | |
666 | if (ret) | |
667 | goto e_err; | |
668 | ||
669 | ret = ccp_init_dm_workarea(&data->dm_wa, cmd_q, dm_len, dir); | |
670 | if (ret) | |
671 | goto e_err; | |
672 | ||
673 | return 0; | |
674 | ||
675 | e_err: | |
676 | ccp_free_data(data, cmd_q); | |
677 | ||
678 | return ret; | |
679 | } | |
680 | ||
681 | static unsigned int ccp_queue_buf(struct ccp_data *data, unsigned int from) | |
682 | { | |
683 | struct ccp_sg_workarea *sg_wa = &data->sg_wa; | |
684 | struct ccp_dm_workarea *dm_wa = &data->dm_wa; | |
685 | unsigned int buf_count, nbytes; | |
686 | ||
687 | /* Clear the buffer if setting it */ | |
688 | if (!from) | |
689 | memset(dm_wa->address, 0, dm_wa->length); | |
690 | ||
691 | if (!sg_wa->sg) | |
692 | return 0; | |
693 | ||
694 | /* Perform the copy operation */ | |
695 | nbytes = min(sg_wa->bytes_left, dm_wa->length); | |
696 | scatterwalk_map_and_copy(dm_wa->address, sg_wa->sg, sg_wa->sg_used, | |
697 | nbytes, from); | |
698 | ||
699 | /* Update the structures and generate the count */ | |
700 | buf_count = 0; | |
701 | while (sg_wa->bytes_left && (buf_count < dm_wa->length)) { | |
702 | nbytes = min3(sg_wa->sg->length - sg_wa->sg_used, | |
703 | dm_wa->length - buf_count, | |
704 | sg_wa->bytes_left); | |
705 | ||
706 | buf_count += nbytes; | |
707 | ccp_update_sg_workarea(sg_wa, nbytes); | |
708 | } | |
709 | ||
710 | return buf_count; | |
711 | } | |
712 | ||
713 | static unsigned int ccp_fill_queue_buf(struct ccp_data *data) | |
714 | { | |
715 | return ccp_queue_buf(data, 0); | |
716 | } | |
717 | ||
718 | static unsigned int ccp_empty_queue_buf(struct ccp_data *data) | |
719 | { | |
720 | return ccp_queue_buf(data, 1); | |
721 | } | |
722 | ||
723 | static void ccp_prepare_data(struct ccp_data *src, struct ccp_data *dst, | |
724 | struct ccp_op *op, unsigned int block_size, | |
725 | bool blocksize_op) | |
726 | { | |
727 | unsigned int sg_src_len, sg_dst_len, op_len; | |
728 | ||
729 | /* The CCP can only DMA from/to one address each per operation. This | |
730 | * requires that we find the smallest DMA area between the source | |
731 | * and destination. | |
732 | */ | |
733 | sg_src_len = min(sg_dma_len(src->sg_wa.sg) - src->sg_wa.sg_used, | |
734 | src->sg_wa.bytes_left); | |
735 | ||
736 | if (dst) { | |
737 | sg_dst_len = min(sg_dma_len(dst->sg_wa.sg) - dst->sg_wa.sg_used, | |
738 | src->sg_wa.bytes_left); | |
739 | op_len = min(sg_src_len, sg_dst_len); | |
740 | } else | |
741 | op_len = sg_src_len; | |
742 | ||
743 | /* The data operation length will be at least block_size in length | |
744 | * or the smaller of available sg room remaining for the source or | |
745 | * the destination | |
746 | */ | |
747 | op_len = max(op_len, block_size); | |
748 | ||
749 | /* Unless we have to buffer data, there's no reason to wait */ | |
750 | op->soc = 0; | |
751 | ||
752 | if (sg_src_len < block_size) { | |
753 | /* Not enough data in the sg element, so it | |
754 | * needs to be buffered into a blocksize chunk | |
755 | */ | |
756 | int cp_len = ccp_fill_queue_buf(src); | |
757 | ||
758 | op->soc = 1; | |
759 | op->src.u.dma.address = src->dm_wa.dma.address; | |
760 | op->src.u.dma.offset = 0; | |
761 | op->src.u.dma.length = (blocksize_op) ? block_size : cp_len; | |
762 | } else { | |
763 | /* Enough data in the sg element, but we need to | |
764 | * adjust for any previously copied data | |
765 | */ | |
766 | op->src.u.dma.address = sg_dma_address(src->sg_wa.sg); | |
767 | op->src.u.dma.offset = src->sg_wa.sg_used; | |
768 | op->src.u.dma.length = op_len & ~(block_size - 1); | |
769 | ||
770 | ccp_update_sg_workarea(&src->sg_wa, op->src.u.dma.length); | |
771 | } | |
772 | ||
773 | if (dst) { | |
774 | if (sg_dst_len < block_size) { | |
775 | /* Not enough room in the sg element or we're on the | |
776 | * last piece of data (when using padding), so the | |
777 | * output needs to be buffered into a blocksize chunk | |
778 | */ | |
779 | op->soc = 1; | |
780 | op->dst.u.dma.address = dst->dm_wa.dma.address; | |
781 | op->dst.u.dma.offset = 0; | |
782 | op->dst.u.dma.length = op->src.u.dma.length; | |
783 | } else { | |
784 | /* Enough room in the sg element, but we need to | |
785 | * adjust for any previously used area | |
786 | */ | |
787 | op->dst.u.dma.address = sg_dma_address(dst->sg_wa.sg); | |
788 | op->dst.u.dma.offset = dst->sg_wa.sg_used; | |
789 | op->dst.u.dma.length = op->src.u.dma.length; | |
790 | } | |
791 | } | |
792 | } | |
793 | ||
794 | static void ccp_process_data(struct ccp_data *src, struct ccp_data *dst, | |
795 | struct ccp_op *op) | |
796 | { | |
797 | op->init = 0; | |
798 | ||
799 | if (dst) { | |
800 | if (op->dst.u.dma.address == dst->dm_wa.dma.address) | |
801 | ccp_empty_queue_buf(dst); | |
802 | else | |
803 | ccp_update_sg_workarea(&dst->sg_wa, | |
804 | op->dst.u.dma.length); | |
805 | } | |
806 | } | |
807 | ||
808 | static int ccp_copy_to_from_ksb(struct ccp_cmd_queue *cmd_q, | |
809 | struct ccp_dm_workarea *wa, u32 jobid, u32 ksb, | |
810 | u32 byte_swap, bool from) | |
811 | { | |
812 | struct ccp_op op; | |
813 | ||
814 | memset(&op, 0, sizeof(op)); | |
815 | ||
816 | op.cmd_q = cmd_q; | |
817 | op.jobid = jobid; | |
818 | op.eom = 1; | |
819 | ||
820 | if (from) { | |
821 | op.soc = 1; | |
822 | op.src.type = CCP_MEMTYPE_KSB; | |
823 | op.src.u.ksb = ksb; | |
824 | op.dst.type = CCP_MEMTYPE_SYSTEM; | |
825 | op.dst.u.dma.address = wa->dma.address; | |
826 | op.dst.u.dma.length = wa->length; | |
827 | } else { | |
828 | op.src.type = CCP_MEMTYPE_SYSTEM; | |
829 | op.src.u.dma.address = wa->dma.address; | |
830 | op.src.u.dma.length = wa->length; | |
831 | op.dst.type = CCP_MEMTYPE_KSB; | |
832 | op.dst.u.ksb = ksb; | |
833 | } | |
834 | ||
835 | op.u.passthru.byte_swap = byte_swap; | |
836 | ||
837 | return ccp_perform_passthru(&op); | |
838 | } | |
839 | ||
840 | static int ccp_copy_to_ksb(struct ccp_cmd_queue *cmd_q, | |
841 | struct ccp_dm_workarea *wa, u32 jobid, u32 ksb, | |
842 | u32 byte_swap) | |
843 | { | |
844 | return ccp_copy_to_from_ksb(cmd_q, wa, jobid, ksb, byte_swap, false); | |
845 | } | |
846 | ||
847 | static int ccp_copy_from_ksb(struct ccp_cmd_queue *cmd_q, | |
848 | struct ccp_dm_workarea *wa, u32 jobid, u32 ksb, | |
849 | u32 byte_swap) | |
850 | { | |
851 | return ccp_copy_to_from_ksb(cmd_q, wa, jobid, ksb, byte_swap, true); | |
852 | } | |
853 | ||
854 | static int ccp_run_aes_cmac_cmd(struct ccp_cmd_queue *cmd_q, | |
855 | struct ccp_cmd *cmd) | |
856 | { | |
857 | struct ccp_aes_engine *aes = &cmd->u.aes; | |
858 | struct ccp_dm_workarea key, ctx; | |
859 | struct ccp_data src; | |
860 | struct ccp_op op; | |
861 | unsigned int dm_offset; | |
862 | int ret; | |
863 | ||
864 | if (!((aes->key_len == AES_KEYSIZE_128) || | |
865 | (aes->key_len == AES_KEYSIZE_192) || | |
866 | (aes->key_len == AES_KEYSIZE_256))) | |
867 | return -EINVAL; | |
868 | ||
869 | if (aes->src_len & (AES_BLOCK_SIZE - 1)) | |
870 | return -EINVAL; | |
871 | ||
872 | if (aes->iv_len != AES_BLOCK_SIZE) | |
873 | return -EINVAL; | |
874 | ||
875 | if (!aes->key || !aes->iv || !aes->src) | |
876 | return -EINVAL; | |
877 | ||
878 | if (aes->cmac_final) { | |
879 | if (aes->cmac_key_len != AES_BLOCK_SIZE) | |
880 | return -EINVAL; | |
881 | ||
882 | if (!aes->cmac_key) | |
883 | return -EINVAL; | |
884 | } | |
885 | ||
886 | BUILD_BUG_ON(CCP_AES_KEY_KSB_COUNT != 1); | |
887 | BUILD_BUG_ON(CCP_AES_CTX_KSB_COUNT != 1); | |
888 | ||
889 | ret = -EIO; | |
890 | memset(&op, 0, sizeof(op)); | |
891 | op.cmd_q = cmd_q; | |
892 | op.jobid = ccp_gen_jobid(cmd_q->ccp); | |
893 | op.ksb_key = cmd_q->ksb_key; | |
894 | op.ksb_ctx = cmd_q->ksb_ctx; | |
895 | op.init = 1; | |
896 | op.u.aes.type = aes->type; | |
897 | op.u.aes.mode = aes->mode; | |
898 | op.u.aes.action = aes->action; | |
899 | ||
900 | /* All supported key sizes fit in a single (32-byte) KSB entry | |
901 | * and must be in little endian format. Use the 256-bit byte | |
902 | * swap passthru option to convert from big endian to little | |
903 | * endian. | |
904 | */ | |
905 | ret = ccp_init_dm_workarea(&key, cmd_q, | |
906 | CCP_AES_KEY_KSB_COUNT * CCP_KSB_BYTES, | |
907 | DMA_TO_DEVICE); | |
908 | if (ret) | |
909 | return ret; | |
910 | ||
911 | dm_offset = CCP_KSB_BYTES - aes->key_len; | |
912 | ccp_set_dm_area(&key, dm_offset, aes->key, 0, aes->key_len); | |
913 | ret = ccp_copy_to_ksb(cmd_q, &key, op.jobid, op.ksb_key, | |
914 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
915 | if (ret) { | |
916 | cmd->engine_error = cmd_q->cmd_error; | |
917 | goto e_key; | |
918 | } | |
919 | ||
920 | /* The AES context fits in a single (32-byte) KSB entry and | |
921 | * must be in little endian format. Use the 256-bit byte swap | |
922 | * passthru option to convert from big endian to little endian. | |
923 | */ | |
924 | ret = ccp_init_dm_workarea(&ctx, cmd_q, | |
925 | CCP_AES_CTX_KSB_COUNT * CCP_KSB_BYTES, | |
926 | DMA_BIDIRECTIONAL); | |
927 | if (ret) | |
928 | goto e_key; | |
929 | ||
930 | dm_offset = CCP_KSB_BYTES - AES_BLOCK_SIZE; | |
931 | ccp_set_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len); | |
932 | ret = ccp_copy_to_ksb(cmd_q, &ctx, op.jobid, op.ksb_ctx, | |
933 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
934 | if (ret) { | |
935 | cmd->engine_error = cmd_q->cmd_error; | |
936 | goto e_ctx; | |
937 | } | |
938 | ||
939 | /* Send data to the CCP AES engine */ | |
940 | ret = ccp_init_data(&src, cmd_q, aes->src, aes->src_len, | |
941 | AES_BLOCK_SIZE, DMA_TO_DEVICE); | |
942 | if (ret) | |
943 | goto e_ctx; | |
944 | ||
945 | while (src.sg_wa.bytes_left) { | |
946 | ccp_prepare_data(&src, NULL, &op, AES_BLOCK_SIZE, true); | |
947 | if (aes->cmac_final && !src.sg_wa.bytes_left) { | |
948 | op.eom = 1; | |
949 | ||
950 | /* Push the K1/K2 key to the CCP now */ | |
951 | ret = ccp_copy_from_ksb(cmd_q, &ctx, op.jobid, | |
952 | op.ksb_ctx, | |
953 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
954 | if (ret) { | |
955 | cmd->engine_error = cmd_q->cmd_error; | |
956 | goto e_src; | |
957 | } | |
958 | ||
959 | ccp_set_dm_area(&ctx, 0, aes->cmac_key, 0, | |
960 | aes->cmac_key_len); | |
961 | ret = ccp_copy_to_ksb(cmd_q, &ctx, op.jobid, op.ksb_ctx, | |
962 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
963 | if (ret) { | |
964 | cmd->engine_error = cmd_q->cmd_error; | |
965 | goto e_src; | |
966 | } | |
967 | } | |
968 | ||
969 | ret = ccp_perform_aes(&op); | |
970 | if (ret) { | |
971 | cmd->engine_error = cmd_q->cmd_error; | |
972 | goto e_src; | |
973 | } | |
974 | ||
975 | ccp_process_data(&src, NULL, &op); | |
976 | } | |
977 | ||
978 | /* Retrieve the AES context - convert from LE to BE using | |
979 | * 32-byte (256-bit) byteswapping | |
980 | */ | |
981 | ret = ccp_copy_from_ksb(cmd_q, &ctx, op.jobid, op.ksb_ctx, | |
982 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
983 | if (ret) { | |
984 | cmd->engine_error = cmd_q->cmd_error; | |
985 | goto e_src; | |
986 | } | |
987 | ||
988 | /* ...but we only need AES_BLOCK_SIZE bytes */ | |
989 | dm_offset = CCP_KSB_BYTES - AES_BLOCK_SIZE; | |
990 | ccp_get_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len); | |
991 | ||
992 | e_src: | |
993 | ccp_free_data(&src, cmd_q); | |
994 | ||
995 | e_ctx: | |
996 | ccp_dm_free(&ctx); | |
997 | ||
998 | e_key: | |
999 | ccp_dm_free(&key); | |
1000 | ||
1001 | return ret; | |
1002 | } | |
1003 | ||
1004 | static int ccp_run_aes_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) | |
1005 | { | |
1006 | struct ccp_aes_engine *aes = &cmd->u.aes; | |
1007 | struct ccp_dm_workarea key, ctx; | |
1008 | struct ccp_data src, dst; | |
1009 | struct ccp_op op; | |
1010 | unsigned int dm_offset; | |
1011 | bool in_place = false; | |
1012 | int ret; | |
1013 | ||
1014 | if (aes->mode == CCP_AES_MODE_CMAC) | |
1015 | return ccp_run_aes_cmac_cmd(cmd_q, cmd); | |
1016 | ||
1017 | if (!((aes->key_len == AES_KEYSIZE_128) || | |
1018 | (aes->key_len == AES_KEYSIZE_192) || | |
1019 | (aes->key_len == AES_KEYSIZE_256))) | |
1020 | return -EINVAL; | |
1021 | ||
1022 | if (((aes->mode == CCP_AES_MODE_ECB) || | |
1023 | (aes->mode == CCP_AES_MODE_CBC) || | |
1024 | (aes->mode == CCP_AES_MODE_CFB)) && | |
1025 | (aes->src_len & (AES_BLOCK_SIZE - 1))) | |
1026 | return -EINVAL; | |
1027 | ||
1028 | if (!aes->key || !aes->src || !aes->dst) | |
1029 | return -EINVAL; | |
1030 | ||
1031 | if (aes->mode != CCP_AES_MODE_ECB) { | |
1032 | if (aes->iv_len != AES_BLOCK_SIZE) | |
1033 | return -EINVAL; | |
1034 | ||
1035 | if (!aes->iv) | |
1036 | return -EINVAL; | |
1037 | } | |
1038 | ||
1039 | BUILD_BUG_ON(CCP_AES_KEY_KSB_COUNT != 1); | |
1040 | BUILD_BUG_ON(CCP_AES_CTX_KSB_COUNT != 1); | |
1041 | ||
1042 | ret = -EIO; | |
1043 | memset(&op, 0, sizeof(op)); | |
1044 | op.cmd_q = cmd_q; | |
1045 | op.jobid = ccp_gen_jobid(cmd_q->ccp); | |
1046 | op.ksb_key = cmd_q->ksb_key; | |
1047 | op.ksb_ctx = cmd_q->ksb_ctx; | |
1048 | op.init = (aes->mode == CCP_AES_MODE_ECB) ? 0 : 1; | |
1049 | op.u.aes.type = aes->type; | |
1050 | op.u.aes.mode = aes->mode; | |
1051 | op.u.aes.action = aes->action; | |
1052 | ||
1053 | /* All supported key sizes fit in a single (32-byte) KSB entry | |
1054 | * and must be in little endian format. Use the 256-bit byte | |
1055 | * swap passthru option to convert from big endian to little | |
1056 | * endian. | |
1057 | */ | |
1058 | ret = ccp_init_dm_workarea(&key, cmd_q, | |
1059 | CCP_AES_KEY_KSB_COUNT * CCP_KSB_BYTES, | |
1060 | DMA_TO_DEVICE); | |
1061 | if (ret) | |
1062 | return ret; | |
1063 | ||
1064 | dm_offset = CCP_KSB_BYTES - aes->key_len; | |
1065 | ccp_set_dm_area(&key, dm_offset, aes->key, 0, aes->key_len); | |
1066 | ret = ccp_copy_to_ksb(cmd_q, &key, op.jobid, op.ksb_key, | |
1067 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
1068 | if (ret) { | |
1069 | cmd->engine_error = cmd_q->cmd_error; | |
1070 | goto e_key; | |
1071 | } | |
1072 | ||
1073 | /* The AES context fits in a single (32-byte) KSB entry and | |
1074 | * must be in little endian format. Use the 256-bit byte swap | |
1075 | * passthru option to convert from big endian to little endian. | |
1076 | */ | |
1077 | ret = ccp_init_dm_workarea(&ctx, cmd_q, | |
1078 | CCP_AES_CTX_KSB_COUNT * CCP_KSB_BYTES, | |
1079 | DMA_BIDIRECTIONAL); | |
1080 | if (ret) | |
1081 | goto e_key; | |
1082 | ||
1083 | if (aes->mode != CCP_AES_MODE_ECB) { | |
1084 | /* Load the AES context - conver to LE */ | |
1085 | dm_offset = CCP_KSB_BYTES - AES_BLOCK_SIZE; | |
1086 | ccp_set_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len); | |
1087 | ret = ccp_copy_to_ksb(cmd_q, &ctx, op.jobid, op.ksb_ctx, | |
1088 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
1089 | if (ret) { | |
1090 | cmd->engine_error = cmd_q->cmd_error; | |
1091 | goto e_ctx; | |
1092 | } | |
1093 | } | |
1094 | ||
1095 | /* Prepare the input and output data workareas. For in-place | |
1096 | * operations we need to set the dma direction to BIDIRECTIONAL | |
1097 | * and copy the src workarea to the dst workarea. | |
1098 | */ | |
1099 | if (sg_virt(aes->src) == sg_virt(aes->dst)) | |
1100 | in_place = true; | |
1101 | ||
1102 | ret = ccp_init_data(&src, cmd_q, aes->src, aes->src_len, | |
1103 | AES_BLOCK_SIZE, | |
1104 | in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE); | |
1105 | if (ret) | |
1106 | goto e_ctx; | |
1107 | ||
1108 | if (in_place) | |
1109 | dst = src; | |
1110 | else { | |
1111 | ret = ccp_init_data(&dst, cmd_q, aes->dst, aes->src_len, | |
1112 | AES_BLOCK_SIZE, DMA_FROM_DEVICE); | |
1113 | if (ret) | |
1114 | goto e_src; | |
1115 | } | |
1116 | ||
1117 | /* Send data to the CCP AES engine */ | |
1118 | while (src.sg_wa.bytes_left) { | |
1119 | ccp_prepare_data(&src, &dst, &op, AES_BLOCK_SIZE, true); | |
1120 | if (!src.sg_wa.bytes_left) { | |
1121 | op.eom = 1; | |
1122 | ||
1123 | /* Since we don't retrieve the AES context in ECB | |
1124 | * mode we have to wait for the operation to complete | |
1125 | * on the last piece of data | |
1126 | */ | |
1127 | if (aes->mode == CCP_AES_MODE_ECB) | |
1128 | op.soc = 1; | |
1129 | } | |
1130 | ||
1131 | ret = ccp_perform_aes(&op); | |
1132 | if (ret) { | |
1133 | cmd->engine_error = cmd_q->cmd_error; | |
1134 | goto e_dst; | |
1135 | } | |
1136 | ||
1137 | ccp_process_data(&src, &dst, &op); | |
1138 | } | |
1139 | ||
1140 | if (aes->mode != CCP_AES_MODE_ECB) { | |
1141 | /* Retrieve the AES context - convert from LE to BE using | |
1142 | * 32-byte (256-bit) byteswapping | |
1143 | */ | |
1144 | ret = ccp_copy_from_ksb(cmd_q, &ctx, op.jobid, op.ksb_ctx, | |
1145 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
1146 | if (ret) { | |
1147 | cmd->engine_error = cmd_q->cmd_error; | |
1148 | goto e_dst; | |
1149 | } | |
1150 | ||
1151 | /* ...but we only need AES_BLOCK_SIZE bytes */ | |
1152 | dm_offset = CCP_KSB_BYTES - AES_BLOCK_SIZE; | |
1153 | ccp_get_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len); | |
1154 | } | |
1155 | ||
1156 | e_dst: | |
1157 | if (!in_place) | |
1158 | ccp_free_data(&dst, cmd_q); | |
1159 | ||
1160 | e_src: | |
1161 | ccp_free_data(&src, cmd_q); | |
1162 | ||
1163 | e_ctx: | |
1164 | ccp_dm_free(&ctx); | |
1165 | ||
1166 | e_key: | |
1167 | ccp_dm_free(&key); | |
1168 | ||
1169 | return ret; | |
1170 | } | |
1171 | ||
1172 | static int ccp_run_xts_aes_cmd(struct ccp_cmd_queue *cmd_q, | |
1173 | struct ccp_cmd *cmd) | |
1174 | { | |
1175 | struct ccp_xts_aes_engine *xts = &cmd->u.xts; | |
1176 | struct ccp_dm_workarea key, ctx; | |
1177 | struct ccp_data src, dst; | |
1178 | struct ccp_op op; | |
1179 | unsigned int unit_size, dm_offset; | |
1180 | bool in_place = false; | |
1181 | int ret; | |
1182 | ||
1183 | switch (xts->unit_size) { | |
1184 | case CCP_XTS_AES_UNIT_SIZE_16: | |
1185 | unit_size = 16; | |
1186 | break; | |
1187 | case CCP_XTS_AES_UNIT_SIZE_512: | |
1188 | unit_size = 512; | |
1189 | break; | |
1190 | case CCP_XTS_AES_UNIT_SIZE_1024: | |
1191 | unit_size = 1024; | |
1192 | break; | |
1193 | case CCP_XTS_AES_UNIT_SIZE_2048: | |
1194 | unit_size = 2048; | |
1195 | break; | |
1196 | case CCP_XTS_AES_UNIT_SIZE_4096: | |
1197 | unit_size = 4096; | |
1198 | break; | |
1199 | ||
1200 | default: | |
1201 | return -EINVAL; | |
1202 | } | |
1203 | ||
1204 | if (xts->key_len != AES_KEYSIZE_128) | |
1205 | return -EINVAL; | |
1206 | ||
1207 | if (!xts->final && (xts->src_len & (AES_BLOCK_SIZE - 1))) | |
1208 | return -EINVAL; | |
1209 | ||
1210 | if (xts->iv_len != AES_BLOCK_SIZE) | |
1211 | return -EINVAL; | |
1212 | ||
1213 | if (!xts->key || !xts->iv || !xts->src || !xts->dst) | |
1214 | return -EINVAL; | |
1215 | ||
1216 | BUILD_BUG_ON(CCP_XTS_AES_KEY_KSB_COUNT != 1); | |
1217 | BUILD_BUG_ON(CCP_XTS_AES_CTX_KSB_COUNT != 1); | |
1218 | ||
1219 | ret = -EIO; | |
1220 | memset(&op, 0, sizeof(op)); | |
1221 | op.cmd_q = cmd_q; | |
1222 | op.jobid = ccp_gen_jobid(cmd_q->ccp); | |
1223 | op.ksb_key = cmd_q->ksb_key; | |
1224 | op.ksb_ctx = cmd_q->ksb_ctx; | |
1225 | op.init = 1; | |
1226 | op.u.xts.action = xts->action; | |
1227 | op.u.xts.unit_size = xts->unit_size; | |
1228 | ||
1229 | /* All supported key sizes fit in a single (32-byte) KSB entry | |
1230 | * and must be in little endian format. Use the 256-bit byte | |
1231 | * swap passthru option to convert from big endian to little | |
1232 | * endian. | |
1233 | */ | |
1234 | ret = ccp_init_dm_workarea(&key, cmd_q, | |
1235 | CCP_XTS_AES_KEY_KSB_COUNT * CCP_KSB_BYTES, | |
1236 | DMA_TO_DEVICE); | |
1237 | if (ret) | |
1238 | return ret; | |
1239 | ||
1240 | dm_offset = CCP_KSB_BYTES - AES_KEYSIZE_128; | |
1241 | ccp_set_dm_area(&key, dm_offset, xts->key, 0, xts->key_len); | |
1242 | ccp_set_dm_area(&key, 0, xts->key, dm_offset, xts->key_len); | |
1243 | ret = ccp_copy_to_ksb(cmd_q, &key, op.jobid, op.ksb_key, | |
1244 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
1245 | if (ret) { | |
1246 | cmd->engine_error = cmd_q->cmd_error; | |
1247 | goto e_key; | |
1248 | } | |
1249 | ||
1250 | /* The AES context fits in a single (32-byte) KSB entry and | |
1251 | * for XTS is already in little endian format so no byte swapping | |
1252 | * is needed. | |
1253 | */ | |
1254 | ret = ccp_init_dm_workarea(&ctx, cmd_q, | |
1255 | CCP_XTS_AES_CTX_KSB_COUNT * CCP_KSB_BYTES, | |
1256 | DMA_BIDIRECTIONAL); | |
1257 | if (ret) | |
1258 | goto e_key; | |
1259 | ||
1260 | ccp_set_dm_area(&ctx, 0, xts->iv, 0, xts->iv_len); | |
1261 | ret = ccp_copy_to_ksb(cmd_q, &ctx, op.jobid, op.ksb_ctx, | |
1262 | CCP_PASSTHRU_BYTESWAP_NOOP); | |
1263 | if (ret) { | |
1264 | cmd->engine_error = cmd_q->cmd_error; | |
1265 | goto e_ctx; | |
1266 | } | |
1267 | ||
1268 | /* Prepare the input and output data workareas. For in-place | |
1269 | * operations we need to set the dma direction to BIDIRECTIONAL | |
1270 | * and copy the src workarea to the dst workarea. | |
1271 | */ | |
1272 | if (sg_virt(xts->src) == sg_virt(xts->dst)) | |
1273 | in_place = true; | |
1274 | ||
1275 | ret = ccp_init_data(&src, cmd_q, xts->src, xts->src_len, | |
1276 | unit_size, | |
1277 | in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE); | |
1278 | if (ret) | |
1279 | goto e_ctx; | |
1280 | ||
1281 | if (in_place) | |
1282 | dst = src; | |
1283 | else { | |
1284 | ret = ccp_init_data(&dst, cmd_q, xts->dst, xts->src_len, | |
1285 | unit_size, DMA_FROM_DEVICE); | |
1286 | if (ret) | |
1287 | goto e_src; | |
1288 | } | |
1289 | ||
1290 | /* Send data to the CCP AES engine */ | |
1291 | while (src.sg_wa.bytes_left) { | |
1292 | ccp_prepare_data(&src, &dst, &op, unit_size, true); | |
1293 | if (!src.sg_wa.bytes_left) | |
1294 | op.eom = 1; | |
1295 | ||
1296 | ret = ccp_perform_xts_aes(&op); | |
1297 | if (ret) { | |
1298 | cmd->engine_error = cmd_q->cmd_error; | |
1299 | goto e_dst; | |
1300 | } | |
1301 | ||
1302 | ccp_process_data(&src, &dst, &op); | |
1303 | } | |
1304 | ||
1305 | /* Retrieve the AES context - convert from LE to BE using | |
1306 | * 32-byte (256-bit) byteswapping | |
1307 | */ | |
1308 | ret = ccp_copy_from_ksb(cmd_q, &ctx, op.jobid, op.ksb_ctx, | |
1309 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
1310 | if (ret) { | |
1311 | cmd->engine_error = cmd_q->cmd_error; | |
1312 | goto e_dst; | |
1313 | } | |
1314 | ||
1315 | /* ...but we only need AES_BLOCK_SIZE bytes */ | |
1316 | dm_offset = CCP_KSB_BYTES - AES_BLOCK_SIZE; | |
1317 | ccp_get_dm_area(&ctx, dm_offset, xts->iv, 0, xts->iv_len); | |
1318 | ||
1319 | e_dst: | |
1320 | if (!in_place) | |
1321 | ccp_free_data(&dst, cmd_q); | |
1322 | ||
1323 | e_src: | |
1324 | ccp_free_data(&src, cmd_q); | |
1325 | ||
1326 | e_ctx: | |
1327 | ccp_dm_free(&ctx); | |
1328 | ||
1329 | e_key: | |
1330 | ccp_dm_free(&key); | |
1331 | ||
1332 | return ret; | |
1333 | } | |
1334 | ||
1335 | static int ccp_run_sha_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) | |
1336 | { | |
1337 | struct ccp_sha_engine *sha = &cmd->u.sha; | |
1338 | struct ccp_dm_workarea ctx; | |
1339 | struct ccp_data src; | |
1340 | struct ccp_op op; | |
1341 | int ret; | |
1342 | ||
1343 | if (sha->ctx_len != CCP_SHA_CTXSIZE) | |
1344 | return -EINVAL; | |
1345 | ||
1346 | if (!sha->ctx) | |
1347 | return -EINVAL; | |
1348 | ||
1349 | if (!sha->final && (sha->src_len & (CCP_SHA_BLOCKSIZE - 1))) | |
1350 | return -EINVAL; | |
1351 | ||
1352 | if (!sha->src_len) { | |
1353 | const u8 *sha_zero; | |
1354 | ||
1355 | /* Not final, just return */ | |
1356 | if (!sha->final) | |
1357 | return 0; | |
1358 | ||
1359 | /* CCP can't do a zero length sha operation so the caller | |
1360 | * must buffer the data. | |
1361 | */ | |
1362 | if (sha->msg_bits) | |
1363 | return -EINVAL; | |
1364 | ||
1365 | /* A sha operation for a message with a total length of zero, | |
1366 | * return known result. | |
1367 | */ | |
1368 | switch (sha->type) { | |
1369 | case CCP_SHA_TYPE_1: | |
1370 | sha_zero = ccp_sha1_zero; | |
1371 | break; | |
1372 | case CCP_SHA_TYPE_224: | |
1373 | sha_zero = ccp_sha224_zero; | |
1374 | break; | |
1375 | case CCP_SHA_TYPE_256: | |
1376 | sha_zero = ccp_sha256_zero; | |
1377 | break; | |
1378 | default: | |
1379 | return -EINVAL; | |
1380 | } | |
1381 | ||
1382 | scatterwalk_map_and_copy((void *)sha_zero, sha->ctx, 0, | |
1383 | sha->ctx_len, 1); | |
1384 | ||
1385 | return 0; | |
1386 | } | |
1387 | ||
1388 | if (!sha->src) | |
1389 | return -EINVAL; | |
1390 | ||
1391 | BUILD_BUG_ON(CCP_SHA_KSB_COUNT != 1); | |
1392 | ||
1393 | memset(&op, 0, sizeof(op)); | |
1394 | op.cmd_q = cmd_q; | |
1395 | op.jobid = ccp_gen_jobid(cmd_q->ccp); | |
1396 | op.ksb_ctx = cmd_q->ksb_ctx; | |
1397 | op.u.sha.type = sha->type; | |
1398 | op.u.sha.msg_bits = sha->msg_bits; | |
1399 | ||
1400 | /* The SHA context fits in a single (32-byte) KSB entry and | |
1401 | * must be in little endian format. Use the 256-bit byte swap | |
1402 | * passthru option to convert from big endian to little endian. | |
1403 | */ | |
1404 | ret = ccp_init_dm_workarea(&ctx, cmd_q, | |
1405 | CCP_SHA_KSB_COUNT * CCP_KSB_BYTES, | |
1406 | DMA_BIDIRECTIONAL); | |
1407 | if (ret) | |
1408 | return ret; | |
1409 | ||
1410 | ccp_set_dm_area(&ctx, 0, sha->ctx, 0, sha->ctx_len); | |
1411 | ret = ccp_copy_to_ksb(cmd_q, &ctx, op.jobid, op.ksb_ctx, | |
1412 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
1413 | if (ret) { | |
1414 | cmd->engine_error = cmd_q->cmd_error; | |
1415 | goto e_ctx; | |
1416 | } | |
1417 | ||
1418 | /* Send data to the CCP SHA engine */ | |
1419 | ret = ccp_init_data(&src, cmd_q, sha->src, sha->src_len, | |
1420 | CCP_SHA_BLOCKSIZE, DMA_TO_DEVICE); | |
1421 | if (ret) | |
1422 | goto e_ctx; | |
1423 | ||
1424 | while (src.sg_wa.bytes_left) { | |
1425 | ccp_prepare_data(&src, NULL, &op, CCP_SHA_BLOCKSIZE, false); | |
1426 | if (sha->final && !src.sg_wa.bytes_left) | |
1427 | op.eom = 1; | |
1428 | ||
1429 | ret = ccp_perform_sha(&op); | |
1430 | if (ret) { | |
1431 | cmd->engine_error = cmd_q->cmd_error; | |
1432 | goto e_data; | |
1433 | } | |
1434 | ||
1435 | ccp_process_data(&src, NULL, &op); | |
1436 | } | |
1437 | ||
1438 | /* Retrieve the SHA context - convert from LE to BE using | |
1439 | * 32-byte (256-bit) byteswapping to BE | |
1440 | */ | |
1441 | ret = ccp_copy_from_ksb(cmd_q, &ctx, op.jobid, op.ksb_ctx, | |
1442 | CCP_PASSTHRU_BYTESWAP_256BIT); | |
1443 | if (ret) { | |
1444 | cmd->engine_error = cmd_q->cmd_error; | |
1445 | goto e_data; | |
1446 | } | |
1447 | ||
1448 | ccp_get_dm_area(&ctx, 0, sha->ctx, 0, sha->ctx_len); | |
1449 | ||
1450 | e_data: | |
1451 | ccp_free_data(&src, cmd_q); | |
1452 | ||
1453 | e_ctx: | |
1454 | ccp_dm_free(&ctx); | |
1455 | ||
1456 | return ret; | |
1457 | } | |
1458 | ||
1459 | static int ccp_run_rsa_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) | |
1460 | { | |
1461 | struct ccp_rsa_engine *rsa = &cmd->u.rsa; | |
1462 | struct ccp_dm_workarea exp, src; | |
1463 | struct ccp_data dst; | |
1464 | struct ccp_op op; | |
1465 | unsigned int ksb_count, i_len, o_len; | |
1466 | int ret; | |
1467 | ||
1468 | if (rsa->key_size > CCP_RSA_MAX_WIDTH) | |
1469 | return -EINVAL; | |
1470 | ||
1471 | if (!rsa->exp || !rsa->mod || !rsa->src || !rsa->dst) | |
1472 | return -EINVAL; | |
1473 | ||
1474 | /* The RSA modulus must precede the message being acted upon, so | |
1475 | * it must be copied to a DMA area where the message and the | |
1476 | * modulus can be concatenated. Therefore the input buffer | |
1477 | * length required is twice the output buffer length (which | |
1478 | * must be a multiple of 256-bits). | |
1479 | */ | |
1480 | o_len = ((rsa->key_size + 255) / 256) * 32; | |
1481 | i_len = o_len * 2; | |
1482 | ||
1483 | ksb_count = o_len / CCP_KSB_BYTES; | |
1484 | ||
1485 | memset(&op, 0, sizeof(op)); | |
1486 | op.cmd_q = cmd_q; | |
1487 | op.jobid = ccp_gen_jobid(cmd_q->ccp); | |
1488 | op.ksb_key = ccp_alloc_ksb(cmd_q->ccp, ksb_count); | |
1489 | if (!op.ksb_key) | |
1490 | return -EIO; | |
1491 | ||
1492 | /* The RSA exponent may span multiple (32-byte) KSB entries and must | |
1493 | * be in little endian format. Reverse copy each 32-byte chunk | |
1494 | * of the exponent (En chunk to E0 chunk, E(n-1) chunk to E1 chunk) | |
1495 | * and each byte within that chunk and do not perform any byte swap | |
1496 | * operations on the passthru operation. | |
1497 | */ | |
1498 | ret = ccp_init_dm_workarea(&exp, cmd_q, o_len, DMA_TO_DEVICE); | |
1499 | if (ret) | |
1500 | goto e_ksb; | |
1501 | ||
1502 | ccp_reverse_set_dm_area(&exp, rsa->exp, rsa->exp_len, CCP_KSB_BYTES, | |
1503 | true); | |
1504 | ret = ccp_copy_to_ksb(cmd_q, &exp, op.jobid, op.ksb_key, | |
1505 | CCP_PASSTHRU_BYTESWAP_NOOP); | |
1506 | if (ret) { | |
1507 | cmd->engine_error = cmd_q->cmd_error; | |
1508 | goto e_exp; | |
1509 | } | |
1510 | ||
1511 | /* Concatenate the modulus and the message. Both the modulus and | |
1512 | * the operands must be in little endian format. Since the input | |
1513 | * is in big endian format it must be converted. | |
1514 | */ | |
1515 | ret = ccp_init_dm_workarea(&src, cmd_q, i_len, DMA_TO_DEVICE); | |
1516 | if (ret) | |
1517 | goto e_exp; | |
1518 | ||
1519 | ccp_reverse_set_dm_area(&src, rsa->mod, rsa->mod_len, CCP_KSB_BYTES, | |
1520 | true); | |
1521 | src.address += o_len; /* Adjust the address for the copy operation */ | |
1522 | ccp_reverse_set_dm_area(&src, rsa->src, rsa->src_len, CCP_KSB_BYTES, | |
1523 | true); | |
1524 | src.address -= o_len; /* Reset the address to original value */ | |
1525 | ||
1526 | /* Prepare the output area for the operation */ | |
1527 | ret = ccp_init_data(&dst, cmd_q, rsa->dst, rsa->mod_len, | |
1528 | o_len, DMA_FROM_DEVICE); | |
1529 | if (ret) | |
1530 | goto e_src; | |
1531 | ||
1532 | op.soc = 1; | |
1533 | op.src.u.dma.address = src.dma.address; | |
1534 | op.src.u.dma.offset = 0; | |
1535 | op.src.u.dma.length = i_len; | |
1536 | op.dst.u.dma.address = dst.dm_wa.dma.address; | |
1537 | op.dst.u.dma.offset = 0; | |
1538 | op.dst.u.dma.length = o_len; | |
1539 | ||
1540 | op.u.rsa.mod_size = rsa->key_size; | |
1541 | op.u.rsa.input_len = i_len; | |
1542 | ||
1543 | ret = ccp_perform_rsa(&op); | |
1544 | if (ret) { | |
1545 | cmd->engine_error = cmd_q->cmd_error; | |
1546 | goto e_dst; | |
1547 | } | |
1548 | ||
1549 | ccp_reverse_get_dm_area(&dst.dm_wa, rsa->dst, rsa->mod_len); | |
1550 | ||
1551 | e_dst: | |
1552 | ccp_free_data(&dst, cmd_q); | |
1553 | ||
1554 | e_src: | |
1555 | ccp_dm_free(&src); | |
1556 | ||
1557 | e_exp: | |
1558 | ccp_dm_free(&exp); | |
1559 | ||
1560 | e_ksb: | |
1561 | ccp_free_ksb(cmd_q->ccp, op.ksb_key, ksb_count); | |
1562 | ||
1563 | return ret; | |
1564 | } | |
1565 | ||
1566 | static int ccp_run_passthru_cmd(struct ccp_cmd_queue *cmd_q, | |
1567 | struct ccp_cmd *cmd) | |
1568 | { | |
1569 | struct ccp_passthru_engine *pt = &cmd->u.passthru; | |
1570 | struct ccp_dm_workarea mask; | |
1571 | struct ccp_data src, dst; | |
1572 | struct ccp_op op; | |
1573 | bool in_place = false; | |
1574 | unsigned int i; | |
1575 | int ret; | |
1576 | ||
1577 | if (!pt->final && (pt->src_len & (CCP_PASSTHRU_BLOCKSIZE - 1))) | |
1578 | return -EINVAL; | |
1579 | ||
1580 | if (!pt->src || !pt->dst) | |
1581 | return -EINVAL; | |
1582 | ||
1583 | if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP) { | |
1584 | if (pt->mask_len != CCP_PASSTHRU_MASKSIZE) | |
1585 | return -EINVAL; | |
1586 | if (!pt->mask) | |
1587 | return -EINVAL; | |
1588 | } | |
1589 | ||
1590 | BUILD_BUG_ON(CCP_PASSTHRU_KSB_COUNT != 1); | |
1591 | ||
1592 | memset(&op, 0, sizeof(op)); | |
1593 | op.cmd_q = cmd_q; | |
1594 | op.jobid = ccp_gen_jobid(cmd_q->ccp); | |
1595 | ||
1596 | if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP) { | |
1597 | /* Load the mask */ | |
1598 | op.ksb_key = cmd_q->ksb_key; | |
1599 | ||
1600 | ret = ccp_init_dm_workarea(&mask, cmd_q, | |
1601 | CCP_PASSTHRU_KSB_COUNT * | |
1602 | CCP_KSB_BYTES, | |
1603 | DMA_TO_DEVICE); | |
1604 | if (ret) | |
1605 | return ret; | |
1606 | ||
1607 | ccp_set_dm_area(&mask, 0, pt->mask, 0, pt->mask_len); | |
1608 | ret = ccp_copy_to_ksb(cmd_q, &mask, op.jobid, op.ksb_key, | |
1609 | CCP_PASSTHRU_BYTESWAP_NOOP); | |
1610 | if (ret) { | |
1611 | cmd->engine_error = cmd_q->cmd_error; | |
1612 | goto e_mask; | |
1613 | } | |
1614 | } | |
1615 | ||
1616 | /* Prepare the input and output data workareas. For in-place | |
1617 | * operations we need to set the dma direction to BIDIRECTIONAL | |
1618 | * and copy the src workarea to the dst workarea. | |
1619 | */ | |
1620 | if (sg_virt(pt->src) == sg_virt(pt->dst)) | |
1621 | in_place = true; | |
1622 | ||
1623 | ret = ccp_init_data(&src, cmd_q, pt->src, pt->src_len, | |
1624 | CCP_PASSTHRU_MASKSIZE, | |
1625 | in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE); | |
1626 | if (ret) | |
1627 | goto e_mask; | |
1628 | ||
1629 | if (in_place) | |
1630 | dst = src; | |
1631 | else { | |
1632 | ret = ccp_init_data(&dst, cmd_q, pt->dst, pt->src_len, | |
1633 | CCP_PASSTHRU_MASKSIZE, DMA_FROM_DEVICE); | |
1634 | if (ret) | |
1635 | goto e_src; | |
1636 | } | |
1637 | ||
1638 | /* Send data to the CCP Passthru engine | |
1639 | * Because the CCP engine works on a single source and destination | |
1640 | * dma address at a time, each entry in the source scatterlist | |
1641 | * (after the dma_map_sg call) must be less than or equal to the | |
1642 | * (remaining) length in the destination scatterlist entry and the | |
1643 | * length must be a multiple of CCP_PASSTHRU_BLOCKSIZE | |
1644 | */ | |
1645 | dst.sg_wa.sg_used = 0; | |
1646 | for (i = 1; i <= src.sg_wa.dma_count; i++) { | |
1647 | if (!dst.sg_wa.sg || | |
1648 | (dst.sg_wa.sg->length < src.sg_wa.sg->length)) { | |
1649 | ret = -EINVAL; | |
1650 | goto e_dst; | |
1651 | } | |
1652 | ||
1653 | if (i == src.sg_wa.dma_count) { | |
1654 | op.eom = 1; | |
1655 | op.soc = 1; | |
1656 | } | |
1657 | ||
1658 | op.src.type = CCP_MEMTYPE_SYSTEM; | |
1659 | op.src.u.dma.address = sg_dma_address(src.sg_wa.sg); | |
1660 | op.src.u.dma.offset = 0; | |
1661 | op.src.u.dma.length = sg_dma_len(src.sg_wa.sg); | |
1662 | ||
1663 | op.dst.type = CCP_MEMTYPE_SYSTEM; | |
1664 | op.dst.u.dma.address = sg_dma_address(dst.sg_wa.sg); | |
1665 | op.src.u.dma.offset = dst.sg_wa.sg_used; | |
1666 | op.src.u.dma.length = op.src.u.dma.length; | |
1667 | ||
1668 | ret = ccp_perform_passthru(&op); | |
1669 | if (ret) { | |
1670 | cmd->engine_error = cmd_q->cmd_error; | |
1671 | goto e_dst; | |
1672 | } | |
1673 | ||
1674 | dst.sg_wa.sg_used += src.sg_wa.sg->length; | |
1675 | if (dst.sg_wa.sg_used == dst.sg_wa.sg->length) { | |
1676 | dst.sg_wa.sg = sg_next(dst.sg_wa.sg); | |
1677 | dst.sg_wa.sg_used = 0; | |
1678 | } | |
1679 | src.sg_wa.sg = sg_next(src.sg_wa.sg); | |
1680 | } | |
1681 | ||
1682 | e_dst: | |
1683 | if (!in_place) | |
1684 | ccp_free_data(&dst, cmd_q); | |
1685 | ||
1686 | e_src: | |
1687 | ccp_free_data(&src, cmd_q); | |
1688 | ||
1689 | e_mask: | |
1690 | if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP) | |
1691 | ccp_dm_free(&mask); | |
1692 | ||
1693 | return ret; | |
1694 | } | |
1695 | ||
1696 | static int ccp_run_ecc_mm_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) | |
1697 | { | |
1698 | struct ccp_ecc_engine *ecc = &cmd->u.ecc; | |
1699 | struct ccp_dm_workarea src, dst; | |
1700 | struct ccp_op op; | |
1701 | int ret; | |
1702 | u8 *save; | |
1703 | ||
1704 | if (!ecc->u.mm.operand_1 || | |
1705 | (ecc->u.mm.operand_1_len > CCP_ECC_MODULUS_BYTES)) | |
1706 | return -EINVAL; | |
1707 | ||
1708 | if (ecc->function != CCP_ECC_FUNCTION_MINV_384BIT) | |
1709 | if (!ecc->u.mm.operand_2 || | |
1710 | (ecc->u.mm.operand_2_len > CCP_ECC_MODULUS_BYTES)) | |
1711 | return -EINVAL; | |
1712 | ||
1713 | if (!ecc->u.mm.result || | |
1714 | (ecc->u.mm.result_len < CCP_ECC_MODULUS_BYTES)) | |
1715 | return -EINVAL; | |
1716 | ||
1717 | memset(&op, 0, sizeof(op)); | |
1718 | op.cmd_q = cmd_q; | |
1719 | op.jobid = ccp_gen_jobid(cmd_q->ccp); | |
1720 | ||
1721 | /* Concatenate the modulus and the operands. Both the modulus and | |
1722 | * the operands must be in little endian format. Since the input | |
1723 | * is in big endian format it must be converted and placed in a | |
1724 | * fixed length buffer. | |
1725 | */ | |
1726 | ret = ccp_init_dm_workarea(&src, cmd_q, CCP_ECC_SRC_BUF_SIZE, | |
1727 | DMA_TO_DEVICE); | |
1728 | if (ret) | |
1729 | return ret; | |
1730 | ||
1731 | /* Save the workarea address since it is updated in order to perform | |
1732 | * the concatenation | |
1733 | */ | |
1734 | save = src.address; | |
1735 | ||
1736 | /* Copy the ECC modulus */ | |
1737 | ccp_reverse_set_dm_area(&src, ecc->mod, ecc->mod_len, | |
1738 | CCP_ECC_OPERAND_SIZE, true); | |
1739 | src.address += CCP_ECC_OPERAND_SIZE; | |
1740 | ||
1741 | /* Copy the first operand */ | |
1742 | ccp_reverse_set_dm_area(&src, ecc->u.mm.operand_1, | |
1743 | ecc->u.mm.operand_1_len, | |
1744 | CCP_ECC_OPERAND_SIZE, true); | |
1745 | src.address += CCP_ECC_OPERAND_SIZE; | |
1746 | ||
1747 | if (ecc->function != CCP_ECC_FUNCTION_MINV_384BIT) { | |
1748 | /* Copy the second operand */ | |
1749 | ccp_reverse_set_dm_area(&src, ecc->u.mm.operand_2, | |
1750 | ecc->u.mm.operand_2_len, | |
1751 | CCP_ECC_OPERAND_SIZE, true); | |
1752 | src.address += CCP_ECC_OPERAND_SIZE; | |
1753 | } | |
1754 | ||
1755 | /* Restore the workarea address */ | |
1756 | src.address = save; | |
1757 | ||
1758 | /* Prepare the output area for the operation */ | |
1759 | ret = ccp_init_dm_workarea(&dst, cmd_q, CCP_ECC_DST_BUF_SIZE, | |
1760 | DMA_FROM_DEVICE); | |
1761 | if (ret) | |
1762 | goto e_src; | |
1763 | ||
1764 | op.soc = 1; | |
1765 | op.src.u.dma.address = src.dma.address; | |
1766 | op.src.u.dma.offset = 0; | |
1767 | op.src.u.dma.length = src.length; | |
1768 | op.dst.u.dma.address = dst.dma.address; | |
1769 | op.dst.u.dma.offset = 0; | |
1770 | op.dst.u.dma.length = dst.length; | |
1771 | ||
1772 | op.u.ecc.function = cmd->u.ecc.function; | |
1773 | ||
1774 | ret = ccp_perform_ecc(&op); | |
1775 | if (ret) { | |
1776 | cmd->engine_error = cmd_q->cmd_error; | |
1777 | goto e_dst; | |
1778 | } | |
1779 | ||
1780 | ecc->ecc_result = le16_to_cpup( | |
1781 | (const __le16 *)(dst.address + CCP_ECC_RESULT_OFFSET)); | |
1782 | if (!(ecc->ecc_result & CCP_ECC_RESULT_SUCCESS)) { | |
1783 | ret = -EIO; | |
1784 | goto e_dst; | |
1785 | } | |
1786 | ||
1787 | /* Save the ECC result */ | |
1788 | ccp_reverse_get_dm_area(&dst, ecc->u.mm.result, CCP_ECC_MODULUS_BYTES); | |
1789 | ||
1790 | e_dst: | |
1791 | ccp_dm_free(&dst); | |
1792 | ||
1793 | e_src: | |
1794 | ccp_dm_free(&src); | |
1795 | ||
1796 | return ret; | |
1797 | } | |
1798 | ||
1799 | static int ccp_run_ecc_pm_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) | |
1800 | { | |
1801 | struct ccp_ecc_engine *ecc = &cmd->u.ecc; | |
1802 | struct ccp_dm_workarea src, dst; | |
1803 | struct ccp_op op; | |
1804 | int ret; | |
1805 | u8 *save; | |
1806 | ||
1807 | if (!ecc->u.pm.point_1.x || | |
1808 | (ecc->u.pm.point_1.x_len > CCP_ECC_MODULUS_BYTES) || | |
1809 | !ecc->u.pm.point_1.y || | |
1810 | (ecc->u.pm.point_1.y_len > CCP_ECC_MODULUS_BYTES)) | |
1811 | return -EINVAL; | |
1812 | ||
1813 | if (ecc->function == CCP_ECC_FUNCTION_PADD_384BIT) { | |
1814 | if (!ecc->u.pm.point_2.x || | |
1815 | (ecc->u.pm.point_2.x_len > CCP_ECC_MODULUS_BYTES) || | |
1816 | !ecc->u.pm.point_2.y || | |
1817 | (ecc->u.pm.point_2.y_len > CCP_ECC_MODULUS_BYTES)) | |
1818 | return -EINVAL; | |
1819 | } else { | |
1820 | if (!ecc->u.pm.domain_a || | |
1821 | (ecc->u.pm.domain_a_len > CCP_ECC_MODULUS_BYTES)) | |
1822 | return -EINVAL; | |
1823 | ||
1824 | if (ecc->function == CCP_ECC_FUNCTION_PMUL_384BIT) | |
1825 | if (!ecc->u.pm.scalar || | |
1826 | (ecc->u.pm.scalar_len > CCP_ECC_MODULUS_BYTES)) | |
1827 | return -EINVAL; | |
1828 | } | |
1829 | ||
1830 | if (!ecc->u.pm.result.x || | |
1831 | (ecc->u.pm.result.x_len < CCP_ECC_MODULUS_BYTES) || | |
1832 | !ecc->u.pm.result.y || | |
1833 | (ecc->u.pm.result.y_len < CCP_ECC_MODULUS_BYTES)) | |
1834 | return -EINVAL; | |
1835 | ||
1836 | memset(&op, 0, sizeof(op)); | |
1837 | op.cmd_q = cmd_q; | |
1838 | op.jobid = ccp_gen_jobid(cmd_q->ccp); | |
1839 | ||
1840 | /* Concatenate the modulus and the operands. Both the modulus and | |
1841 | * the operands must be in little endian format. Since the input | |
1842 | * is in big endian format it must be converted and placed in a | |
1843 | * fixed length buffer. | |
1844 | */ | |
1845 | ret = ccp_init_dm_workarea(&src, cmd_q, CCP_ECC_SRC_BUF_SIZE, | |
1846 | DMA_TO_DEVICE); | |
1847 | if (ret) | |
1848 | return ret; | |
1849 | ||
1850 | /* Save the workarea address since it is updated in order to perform | |
1851 | * the concatenation | |
1852 | */ | |
1853 | save = src.address; | |
1854 | ||
1855 | /* Copy the ECC modulus */ | |
1856 | ccp_reverse_set_dm_area(&src, ecc->mod, ecc->mod_len, | |
1857 | CCP_ECC_OPERAND_SIZE, true); | |
1858 | src.address += CCP_ECC_OPERAND_SIZE; | |
1859 | ||
1860 | /* Copy the first point X and Y coordinate */ | |
1861 | ccp_reverse_set_dm_area(&src, ecc->u.pm.point_1.x, | |
1862 | ecc->u.pm.point_1.x_len, | |
1863 | CCP_ECC_OPERAND_SIZE, true); | |
1864 | src.address += CCP_ECC_OPERAND_SIZE; | |
1865 | ccp_reverse_set_dm_area(&src, ecc->u.pm.point_1.y, | |
1866 | ecc->u.pm.point_1.y_len, | |
1867 | CCP_ECC_OPERAND_SIZE, true); | |
1868 | src.address += CCP_ECC_OPERAND_SIZE; | |
1869 | ||
1870 | /* Set the first point Z coordianate to 1 */ | |
1871 | *(src.address) = 0x01; | |
1872 | src.address += CCP_ECC_OPERAND_SIZE; | |
1873 | ||
1874 | if (ecc->function == CCP_ECC_FUNCTION_PADD_384BIT) { | |
1875 | /* Copy the second point X and Y coordinate */ | |
1876 | ccp_reverse_set_dm_area(&src, ecc->u.pm.point_2.x, | |
1877 | ecc->u.pm.point_2.x_len, | |
1878 | CCP_ECC_OPERAND_SIZE, true); | |
1879 | src.address += CCP_ECC_OPERAND_SIZE; | |
1880 | ccp_reverse_set_dm_area(&src, ecc->u.pm.point_2.y, | |
1881 | ecc->u.pm.point_2.y_len, | |
1882 | CCP_ECC_OPERAND_SIZE, true); | |
1883 | src.address += CCP_ECC_OPERAND_SIZE; | |
1884 | ||
1885 | /* Set the second point Z coordianate to 1 */ | |
1886 | *(src.address) = 0x01; | |
1887 | src.address += CCP_ECC_OPERAND_SIZE; | |
1888 | } else { | |
1889 | /* Copy the Domain "a" parameter */ | |
1890 | ccp_reverse_set_dm_area(&src, ecc->u.pm.domain_a, | |
1891 | ecc->u.pm.domain_a_len, | |
1892 | CCP_ECC_OPERAND_SIZE, true); | |
1893 | src.address += CCP_ECC_OPERAND_SIZE; | |
1894 | ||
1895 | if (ecc->function == CCP_ECC_FUNCTION_PMUL_384BIT) { | |
1896 | /* Copy the scalar value */ | |
1897 | ccp_reverse_set_dm_area(&src, ecc->u.pm.scalar, | |
1898 | ecc->u.pm.scalar_len, | |
1899 | CCP_ECC_OPERAND_SIZE, true); | |
1900 | src.address += CCP_ECC_OPERAND_SIZE; | |
1901 | } | |
1902 | } | |
1903 | ||
1904 | /* Restore the workarea address */ | |
1905 | src.address = save; | |
1906 | ||
1907 | /* Prepare the output area for the operation */ | |
1908 | ret = ccp_init_dm_workarea(&dst, cmd_q, CCP_ECC_DST_BUF_SIZE, | |
1909 | DMA_FROM_DEVICE); | |
1910 | if (ret) | |
1911 | goto e_src; | |
1912 | ||
1913 | op.soc = 1; | |
1914 | op.src.u.dma.address = src.dma.address; | |
1915 | op.src.u.dma.offset = 0; | |
1916 | op.src.u.dma.length = src.length; | |
1917 | op.dst.u.dma.address = dst.dma.address; | |
1918 | op.dst.u.dma.offset = 0; | |
1919 | op.dst.u.dma.length = dst.length; | |
1920 | ||
1921 | op.u.ecc.function = cmd->u.ecc.function; | |
1922 | ||
1923 | ret = ccp_perform_ecc(&op); | |
1924 | if (ret) { | |
1925 | cmd->engine_error = cmd_q->cmd_error; | |
1926 | goto e_dst; | |
1927 | } | |
1928 | ||
1929 | ecc->ecc_result = le16_to_cpup( | |
1930 | (const __le16 *)(dst.address + CCP_ECC_RESULT_OFFSET)); | |
1931 | if (!(ecc->ecc_result & CCP_ECC_RESULT_SUCCESS)) { | |
1932 | ret = -EIO; | |
1933 | goto e_dst; | |
1934 | } | |
1935 | ||
1936 | /* Save the workarea address since it is updated as we walk through | |
1937 | * to copy the point math result | |
1938 | */ | |
1939 | save = dst.address; | |
1940 | ||
1941 | /* Save the ECC result X and Y coordinates */ | |
1942 | ccp_reverse_get_dm_area(&dst, ecc->u.pm.result.x, | |
1943 | CCP_ECC_MODULUS_BYTES); | |
1944 | dst.address += CCP_ECC_OUTPUT_SIZE; | |
1945 | ccp_reverse_get_dm_area(&dst, ecc->u.pm.result.y, | |
1946 | CCP_ECC_MODULUS_BYTES); | |
1947 | dst.address += CCP_ECC_OUTPUT_SIZE; | |
1948 | ||
1949 | /* Restore the workarea address */ | |
1950 | dst.address = save; | |
1951 | ||
1952 | e_dst: | |
1953 | ccp_dm_free(&dst); | |
1954 | ||
1955 | e_src: | |
1956 | ccp_dm_free(&src); | |
1957 | ||
1958 | return ret; | |
1959 | } | |
1960 | ||
1961 | static int ccp_run_ecc_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) | |
1962 | { | |
1963 | struct ccp_ecc_engine *ecc = &cmd->u.ecc; | |
1964 | ||
1965 | ecc->ecc_result = 0; | |
1966 | ||
1967 | if (!ecc->mod || | |
1968 | (ecc->mod_len > CCP_ECC_MODULUS_BYTES)) | |
1969 | return -EINVAL; | |
1970 | ||
1971 | switch (ecc->function) { | |
1972 | case CCP_ECC_FUNCTION_MMUL_384BIT: | |
1973 | case CCP_ECC_FUNCTION_MADD_384BIT: | |
1974 | case CCP_ECC_FUNCTION_MINV_384BIT: | |
1975 | return ccp_run_ecc_mm_cmd(cmd_q, cmd); | |
1976 | ||
1977 | case CCP_ECC_FUNCTION_PADD_384BIT: | |
1978 | case CCP_ECC_FUNCTION_PMUL_384BIT: | |
1979 | case CCP_ECC_FUNCTION_PDBL_384BIT: | |
1980 | return ccp_run_ecc_pm_cmd(cmd_q, cmd); | |
1981 | ||
1982 | default: | |
1983 | return -EINVAL; | |
1984 | } | |
1985 | } | |
1986 | ||
1987 | int ccp_run_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) | |
1988 | { | |
1989 | int ret; | |
1990 | ||
1991 | cmd->engine_error = 0; | |
1992 | cmd_q->cmd_error = 0; | |
1993 | cmd_q->int_rcvd = 0; | |
1994 | cmd_q->free_slots = CMD_Q_DEPTH(ioread32(cmd_q->reg_status)); | |
1995 | ||
1996 | switch (cmd->engine) { | |
1997 | case CCP_ENGINE_AES: | |
1998 | ret = ccp_run_aes_cmd(cmd_q, cmd); | |
1999 | break; | |
2000 | case CCP_ENGINE_XTS_AES_128: | |
2001 | ret = ccp_run_xts_aes_cmd(cmd_q, cmd); | |
2002 | break; | |
2003 | case CCP_ENGINE_SHA: | |
2004 | ret = ccp_run_sha_cmd(cmd_q, cmd); | |
2005 | break; | |
2006 | case CCP_ENGINE_RSA: | |
2007 | ret = ccp_run_rsa_cmd(cmd_q, cmd); | |
2008 | break; | |
2009 | case CCP_ENGINE_PASSTHRU: | |
2010 | ret = ccp_run_passthru_cmd(cmd_q, cmd); | |
2011 | break; | |
2012 | case CCP_ENGINE_ECC: | |
2013 | ret = ccp_run_ecc_cmd(cmd_q, cmd); | |
2014 | break; | |
2015 | default: | |
2016 | ret = -EINVAL; | |
2017 | } | |
2018 | ||
2019 | return ret; | |
2020 | } |