4 * Support for ATMEL SHA1/SHA256 HW acceleration.
6 * Copyright (c) 2012 Eukréa Electromatique - ATMEL
7 * Author: Nicolas Royer <nicolas@eukrea.com>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as published
11 * by the Free Software Foundation.
13 * Some ideas are from omap-sham.c drivers.
17 #include <linux/kernel.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
20 #include <linux/err.h>
21 #include <linux/clk.h>
23 #include <linux/hw_random.h>
24 #include <linux/platform_device.h>
26 #include <linux/device.h>
27 #include <linux/init.h>
28 #include <linux/errno.h>
29 #include <linux/interrupt.h>
30 #include <linux/irq.h>
31 #include <linux/scatterlist.h>
32 #include <linux/dma-mapping.h>
33 #include <linux/of_device.h>
34 #include <linux/delay.h>
35 #include <linux/crypto.h>
36 #include <linux/cryptohash.h>
37 #include <crypto/scatterwalk.h>
38 #include <crypto/algapi.h>
39 #include <crypto/sha.h>
40 #include <crypto/hash.h>
41 #include <crypto/internal/hash.h>
42 #include <linux/platform_data/crypto-atmel.h>
43 #include "atmel-sha-regs.h"
46 #define SHA_FLAGS_BUSY BIT(0)
47 #define SHA_FLAGS_FINAL BIT(1)
48 #define SHA_FLAGS_DMA_ACTIVE BIT(2)
49 #define SHA_FLAGS_OUTPUT_READY BIT(3)
50 #define SHA_FLAGS_INIT BIT(4)
51 #define SHA_FLAGS_CPU BIT(5)
52 #define SHA_FLAGS_DMA_READY BIT(6)
54 #define SHA_FLAGS_FINUP BIT(16)
55 #define SHA_FLAGS_SG BIT(17)
56 #define SHA_FLAGS_SHA1 BIT(18)
57 #define SHA_FLAGS_SHA224 BIT(19)
58 #define SHA_FLAGS_SHA256 BIT(20)
59 #define SHA_FLAGS_SHA384 BIT(21)
60 #define SHA_FLAGS_SHA512 BIT(22)
61 #define SHA_FLAGS_ERROR BIT(23)
62 #define SHA_FLAGS_PAD BIT(24)
64 #define SHA_OP_UPDATE 1
65 #define SHA_OP_FINAL 2
67 #define SHA_BUFFER_LEN PAGE_SIZE
69 #define ATMEL_SHA_DMA_THRESHOLD 56
71 struct atmel_sha_caps {
80 struct atmel_sha_reqctx {
81 struct atmel_sha_dev *dd;
85 u8 digest[SHA512_DIGEST_SIZE] __aligned(sizeof(u32));
92 struct scatterlist *sg;
93 unsigned int offset; /* offset in current sg */
94 unsigned int total; /* total request */
98 u8 buffer[0] __aligned(sizeof(u32));
101 struct atmel_sha_ctx {
102 struct atmel_sha_dev *dd;
107 #define ATMEL_SHA_QUEUE_LENGTH 50
109 struct atmel_sha_dma {
110 struct dma_chan *chan;
111 struct dma_slave_config dma_conf;
114 struct atmel_sha_dev {
115 struct list_head list;
116 unsigned long phys_base;
120 void __iomem *io_base;
124 struct tasklet_struct done_task;
127 struct crypto_queue queue;
128 struct ahash_request *req;
130 struct atmel_sha_dma dma_lch_in;
132 struct atmel_sha_caps caps;
137 struct atmel_sha_drv {
138 struct list_head dev_list;
142 static struct atmel_sha_drv atmel_sha = {
143 .dev_list = LIST_HEAD_INIT(atmel_sha.dev_list),
144 .lock = __SPIN_LOCK_UNLOCKED(atmel_sha.lock),
147 static inline u32 atmel_sha_read(struct atmel_sha_dev *dd, u32 offset)
149 return readl_relaxed(dd->io_base + offset);
152 static inline void atmel_sha_write(struct atmel_sha_dev *dd,
153 u32 offset, u32 value)
155 writel_relaxed(value, dd->io_base + offset);
158 static size_t atmel_sha_append_sg(struct atmel_sha_reqctx *ctx)
162 while ((ctx->bufcnt < ctx->buflen) && ctx->total) {
163 count = min(ctx->sg->length - ctx->offset, ctx->total);
164 count = min(count, ctx->buflen - ctx->bufcnt);
168 * Check if count <= 0 because the buffer is full or
169 * because the sg length is 0. In the latest case,
170 * check if there is another sg in the list, a 0 length
171 * sg doesn't necessarily mean the end of the sg list.
173 if ((ctx->sg->length == 0) && !sg_is_last(ctx->sg)) {
174 ctx->sg = sg_next(ctx->sg);
181 scatterwalk_map_and_copy(ctx->buffer + ctx->bufcnt, ctx->sg,
182 ctx->offset, count, 0);
184 ctx->bufcnt += count;
185 ctx->offset += count;
188 if (ctx->offset == ctx->sg->length) {
189 ctx->sg = sg_next(ctx->sg);
201 * The purpose of this padding is to ensure that the padded message is a
202 * multiple of 512 bits (SHA1/SHA224/SHA256) or 1024 bits (SHA384/SHA512).
203 * The bit "1" is appended at the end of the message followed by
204 * "padlen-1" zero bits. Then a 64 bits block (SHA1/SHA224/SHA256) or
205 * 128 bits block (SHA384/SHA512) equals to the message length in bits
208 * For SHA1/SHA224/SHA256, padlen is calculated as followed:
209 * - if message length < 56 bytes then padlen = 56 - message length
210 * - else padlen = 64 + 56 - message length
212 * For SHA384/SHA512, padlen is calculated as followed:
213 * - if message length < 112 bytes then padlen = 112 - message length
214 * - else padlen = 128 + 112 - message length
216 static void atmel_sha_fill_padding(struct atmel_sha_reqctx *ctx, int length)
218 unsigned int index, padlen;
222 size[0] = ctx->digcnt[0];
223 size[1] = ctx->digcnt[1];
225 size[0] += ctx->bufcnt;
226 if (size[0] < ctx->bufcnt)
230 if (size[0] < length)
233 bits[1] = cpu_to_be64(size[0] << 3);
234 bits[0] = cpu_to_be64(size[1] << 3 | size[0] >> 61);
236 if (ctx->flags & (SHA_FLAGS_SHA384 | SHA_FLAGS_SHA512)) {
237 index = ctx->bufcnt & 0x7f;
238 padlen = (index < 112) ? (112 - index) : ((128+112) - index);
239 *(ctx->buffer + ctx->bufcnt) = 0x80;
240 memset(ctx->buffer + ctx->bufcnt + 1, 0, padlen-1);
241 memcpy(ctx->buffer + ctx->bufcnt + padlen, bits, 16);
242 ctx->bufcnt += padlen + 16;
243 ctx->flags |= SHA_FLAGS_PAD;
245 index = ctx->bufcnt & 0x3f;
246 padlen = (index < 56) ? (56 - index) : ((64+56) - index);
247 *(ctx->buffer + ctx->bufcnt) = 0x80;
248 memset(ctx->buffer + ctx->bufcnt + 1, 0, padlen-1);
249 memcpy(ctx->buffer + ctx->bufcnt + padlen, &bits[1], 8);
250 ctx->bufcnt += padlen + 8;
251 ctx->flags |= SHA_FLAGS_PAD;
255 static int atmel_sha_init(struct ahash_request *req)
257 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
258 struct atmel_sha_ctx *tctx = crypto_ahash_ctx(tfm);
259 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
260 struct atmel_sha_dev *dd = NULL;
261 struct atmel_sha_dev *tmp;
263 spin_lock_bh(&atmel_sha.lock);
265 list_for_each_entry(tmp, &atmel_sha.dev_list, list) {
274 spin_unlock_bh(&atmel_sha.lock);
280 dev_dbg(dd->dev, "init: digest size: %d\n",
281 crypto_ahash_digestsize(tfm));
283 switch (crypto_ahash_digestsize(tfm)) {
284 case SHA1_DIGEST_SIZE:
285 ctx->flags |= SHA_FLAGS_SHA1;
286 ctx->block_size = SHA1_BLOCK_SIZE;
288 case SHA224_DIGEST_SIZE:
289 ctx->flags |= SHA_FLAGS_SHA224;
290 ctx->block_size = SHA224_BLOCK_SIZE;
292 case SHA256_DIGEST_SIZE:
293 ctx->flags |= SHA_FLAGS_SHA256;
294 ctx->block_size = SHA256_BLOCK_SIZE;
296 case SHA384_DIGEST_SIZE:
297 ctx->flags |= SHA_FLAGS_SHA384;
298 ctx->block_size = SHA384_BLOCK_SIZE;
300 case SHA512_DIGEST_SIZE:
301 ctx->flags |= SHA_FLAGS_SHA512;
302 ctx->block_size = SHA512_BLOCK_SIZE;
312 ctx->buflen = SHA_BUFFER_LEN;
317 static void atmel_sha_write_ctrl(struct atmel_sha_dev *dd, int dma)
319 struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
320 u32 valcr = 0, valmr = SHA_MR_MODE_AUTO;
323 if (!dd->caps.has_dma)
324 atmel_sha_write(dd, SHA_IER, SHA_INT_TXBUFE);
325 valmr = SHA_MR_MODE_PDC;
326 if (dd->caps.has_dualbuff)
327 valmr |= SHA_MR_DUALBUFF;
329 atmel_sha_write(dd, SHA_IER, SHA_INT_DATARDY);
332 if (ctx->flags & SHA_FLAGS_SHA1)
333 valmr |= SHA_MR_ALGO_SHA1;
334 else if (ctx->flags & SHA_FLAGS_SHA224)
335 valmr |= SHA_MR_ALGO_SHA224;
336 else if (ctx->flags & SHA_FLAGS_SHA256)
337 valmr |= SHA_MR_ALGO_SHA256;
338 else if (ctx->flags & SHA_FLAGS_SHA384)
339 valmr |= SHA_MR_ALGO_SHA384;
340 else if (ctx->flags & SHA_FLAGS_SHA512)
341 valmr |= SHA_MR_ALGO_SHA512;
343 /* Setting CR_FIRST only for the first iteration */
344 if (!(ctx->digcnt[0] || ctx->digcnt[1]))
345 valcr = SHA_CR_FIRST;
347 atmel_sha_write(dd, SHA_CR, valcr);
348 atmel_sha_write(dd, SHA_MR, valmr);
351 static int atmel_sha_xmit_cpu(struct atmel_sha_dev *dd, const u8 *buf,
352 size_t length, int final)
354 struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
356 const u32 *buffer = (const u32 *)buf;
358 dev_dbg(dd->dev, "xmit_cpu: digcnt: 0x%llx 0x%llx, length: %d, final: %d\n",
359 ctx->digcnt[1], ctx->digcnt[0], length, final);
361 atmel_sha_write_ctrl(dd, 0);
363 /* should be non-zero before next lines to disable clocks later */
364 ctx->digcnt[0] += length;
365 if (ctx->digcnt[0] < length)
369 dd->flags |= SHA_FLAGS_FINAL; /* catch last interrupt */
371 len32 = DIV_ROUND_UP(length, sizeof(u32));
373 dd->flags |= SHA_FLAGS_CPU;
375 for (count = 0; count < len32; count++)
376 atmel_sha_write(dd, SHA_REG_DIN(count), buffer[count]);
381 static int atmel_sha_xmit_pdc(struct atmel_sha_dev *dd, dma_addr_t dma_addr1,
382 size_t length1, dma_addr_t dma_addr2, size_t length2, int final)
384 struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
387 dev_dbg(dd->dev, "xmit_pdc: digcnt: 0x%llx 0x%llx, length: %d, final: %d\n",
388 ctx->digcnt[1], ctx->digcnt[0], length1, final);
390 len32 = DIV_ROUND_UP(length1, sizeof(u32));
391 atmel_sha_write(dd, SHA_PTCR, SHA_PTCR_TXTDIS);
392 atmel_sha_write(dd, SHA_TPR, dma_addr1);
393 atmel_sha_write(dd, SHA_TCR, len32);
395 len32 = DIV_ROUND_UP(length2, sizeof(u32));
396 atmel_sha_write(dd, SHA_TNPR, dma_addr2);
397 atmel_sha_write(dd, SHA_TNCR, len32);
399 atmel_sha_write_ctrl(dd, 1);
401 /* should be non-zero before next lines to disable clocks later */
402 ctx->digcnt[0] += length1;
403 if (ctx->digcnt[0] < length1)
407 dd->flags |= SHA_FLAGS_FINAL; /* catch last interrupt */
409 dd->flags |= SHA_FLAGS_DMA_ACTIVE;
411 /* Start DMA transfer */
412 atmel_sha_write(dd, SHA_PTCR, SHA_PTCR_TXTEN);
417 static void atmel_sha_dma_callback(void *data)
419 struct atmel_sha_dev *dd = data;
421 /* dma_lch_in - completed - wait DATRDY */
422 atmel_sha_write(dd, SHA_IER, SHA_INT_DATARDY);
425 static int atmel_sha_xmit_dma(struct atmel_sha_dev *dd, dma_addr_t dma_addr1,
426 size_t length1, dma_addr_t dma_addr2, size_t length2, int final)
428 struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
429 struct dma_async_tx_descriptor *in_desc;
430 struct scatterlist sg[2];
432 dev_dbg(dd->dev, "xmit_dma: digcnt: 0x%llx 0x%llx, length: %d, final: %d\n",
433 ctx->digcnt[1], ctx->digcnt[0], length1, final);
435 dd->dma_lch_in.dma_conf.src_maxburst = 16;
436 dd->dma_lch_in.dma_conf.dst_maxburst = 16;
438 dmaengine_slave_config(dd->dma_lch_in.chan, &dd->dma_lch_in.dma_conf);
441 sg_init_table(sg, 2);
442 sg_dma_address(&sg[0]) = dma_addr1;
443 sg_dma_len(&sg[0]) = length1;
444 sg_dma_address(&sg[1]) = dma_addr2;
445 sg_dma_len(&sg[1]) = length2;
446 in_desc = dmaengine_prep_slave_sg(dd->dma_lch_in.chan, sg, 2,
447 DMA_MEM_TO_DEV, DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
449 sg_init_table(sg, 1);
450 sg_dma_address(&sg[0]) = dma_addr1;
451 sg_dma_len(&sg[0]) = length1;
452 in_desc = dmaengine_prep_slave_sg(dd->dma_lch_in.chan, sg, 1,
453 DMA_MEM_TO_DEV, DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
458 in_desc->callback = atmel_sha_dma_callback;
459 in_desc->callback_param = dd;
461 atmel_sha_write_ctrl(dd, 1);
463 /* should be non-zero before next lines to disable clocks later */
464 ctx->digcnt[0] += length1;
465 if (ctx->digcnt[0] < length1)
469 dd->flags |= SHA_FLAGS_FINAL; /* catch last interrupt */
471 dd->flags |= SHA_FLAGS_DMA_ACTIVE;
473 /* Start DMA transfer */
474 dmaengine_submit(in_desc);
475 dma_async_issue_pending(dd->dma_lch_in.chan);
480 static int atmel_sha_xmit_start(struct atmel_sha_dev *dd, dma_addr_t dma_addr1,
481 size_t length1, dma_addr_t dma_addr2, size_t length2, int final)
483 if (dd->caps.has_dma)
484 return atmel_sha_xmit_dma(dd, dma_addr1, length1,
485 dma_addr2, length2, final);
487 return atmel_sha_xmit_pdc(dd, dma_addr1, length1,
488 dma_addr2, length2, final);
491 static int atmel_sha_update_cpu(struct atmel_sha_dev *dd)
493 struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
496 atmel_sha_append_sg(ctx);
497 atmel_sha_fill_padding(ctx, 0);
498 bufcnt = ctx->bufcnt;
501 return atmel_sha_xmit_cpu(dd, ctx->buffer, bufcnt, 1);
504 static int atmel_sha_xmit_dma_map(struct atmel_sha_dev *dd,
505 struct atmel_sha_reqctx *ctx,
506 size_t length, int final)
508 ctx->dma_addr = dma_map_single(dd->dev, ctx->buffer,
509 ctx->buflen + ctx->block_size, DMA_TO_DEVICE);
510 if (dma_mapping_error(dd->dev, ctx->dma_addr)) {
511 dev_err(dd->dev, "dma %u bytes error\n", ctx->buflen +
516 ctx->flags &= ~SHA_FLAGS_SG;
518 /* next call does not fail... so no unmap in the case of error */
519 return atmel_sha_xmit_start(dd, ctx->dma_addr, length, 0, 0, final);
522 static int atmel_sha_update_dma_slow(struct atmel_sha_dev *dd)
524 struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
528 atmel_sha_append_sg(ctx);
530 final = (ctx->flags & SHA_FLAGS_FINUP) && !ctx->total;
532 dev_dbg(dd->dev, "slow: bufcnt: %u, digcnt: 0x%llx 0x%llx, final: %d\n",
533 ctx->bufcnt, ctx->digcnt[1], ctx->digcnt[0], final);
536 atmel_sha_fill_padding(ctx, 0);
538 if (final || (ctx->bufcnt == ctx->buflen)) {
541 return atmel_sha_xmit_dma_map(dd, ctx, count, final);
547 static int atmel_sha_update_dma_start(struct atmel_sha_dev *dd)
549 struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
550 unsigned int length, final, tail;
551 struct scatterlist *sg;
557 if (ctx->bufcnt || ctx->offset)
558 return atmel_sha_update_dma_slow(dd);
560 dev_dbg(dd->dev, "fast: digcnt: 0x%llx 0x%llx, bufcnt: %u, total: %u\n",
561 ctx->digcnt[1], ctx->digcnt[0], ctx->bufcnt, ctx->total);
565 if (!IS_ALIGNED(sg->offset, sizeof(u32)))
566 return atmel_sha_update_dma_slow(dd);
568 if (!sg_is_last(sg) && !IS_ALIGNED(sg->length, ctx->block_size))
569 /* size is not ctx->block_size aligned */
570 return atmel_sha_update_dma_slow(dd);
572 length = min(ctx->total, sg->length);
574 if (sg_is_last(sg)) {
575 if (!(ctx->flags & SHA_FLAGS_FINUP)) {
576 /* not last sg must be ctx->block_size aligned */
577 tail = length & (ctx->block_size - 1);
582 ctx->total -= length;
583 ctx->offset = length; /* offset where to start slow */
585 final = (ctx->flags & SHA_FLAGS_FINUP) && !ctx->total;
589 tail = length & (ctx->block_size - 1);
592 ctx->offset = length; /* offset where to start slow */
595 atmel_sha_append_sg(ctx);
597 atmel_sha_fill_padding(ctx, length);
599 ctx->dma_addr = dma_map_single(dd->dev, ctx->buffer,
600 ctx->buflen + ctx->block_size, DMA_TO_DEVICE);
601 if (dma_mapping_error(dd->dev, ctx->dma_addr)) {
602 dev_err(dd->dev, "dma %u bytes error\n",
603 ctx->buflen + ctx->block_size);
608 ctx->flags &= ~SHA_FLAGS_SG;
611 return atmel_sha_xmit_start(dd, ctx->dma_addr, count, 0,
615 if (!dma_map_sg(dd->dev, ctx->sg, 1,
617 dev_err(dd->dev, "dma_map_sg error\n");
621 ctx->flags |= SHA_FLAGS_SG;
625 return atmel_sha_xmit_start(dd, sg_dma_address(ctx->sg),
626 length, ctx->dma_addr, count, final);
630 if (!dma_map_sg(dd->dev, ctx->sg, 1, DMA_TO_DEVICE)) {
631 dev_err(dd->dev, "dma_map_sg error\n");
635 ctx->flags |= SHA_FLAGS_SG;
637 /* next call does not fail... so no unmap in the case of error */
638 return atmel_sha_xmit_start(dd, sg_dma_address(ctx->sg), length, 0,
642 static int atmel_sha_update_dma_stop(struct atmel_sha_dev *dd)
644 struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
646 if (ctx->flags & SHA_FLAGS_SG) {
647 dma_unmap_sg(dd->dev, ctx->sg, 1, DMA_TO_DEVICE);
648 if (ctx->sg->length == ctx->offset) {
649 ctx->sg = sg_next(ctx->sg);
653 if (ctx->flags & SHA_FLAGS_PAD) {
654 dma_unmap_single(dd->dev, ctx->dma_addr,
655 ctx->buflen + ctx->block_size, DMA_TO_DEVICE);
658 dma_unmap_single(dd->dev, ctx->dma_addr, ctx->buflen +
659 ctx->block_size, DMA_TO_DEVICE);
665 static int atmel_sha_update_req(struct atmel_sha_dev *dd)
667 struct ahash_request *req = dd->req;
668 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
671 dev_dbg(dd->dev, "update_req: total: %u, digcnt: 0x%llx 0x%llx\n",
672 ctx->total, ctx->digcnt[1], ctx->digcnt[0]);
674 if (ctx->flags & SHA_FLAGS_CPU)
675 err = atmel_sha_update_cpu(dd);
677 err = atmel_sha_update_dma_start(dd);
679 /* wait for dma completion before can take more data */
680 dev_dbg(dd->dev, "update: err: %d, digcnt: 0x%llx 0%llx\n",
681 err, ctx->digcnt[1], ctx->digcnt[0]);
686 static int atmel_sha_final_req(struct atmel_sha_dev *dd)
688 struct ahash_request *req = dd->req;
689 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
693 if (ctx->bufcnt >= ATMEL_SHA_DMA_THRESHOLD) {
694 atmel_sha_fill_padding(ctx, 0);
697 err = atmel_sha_xmit_dma_map(dd, ctx, count, 1);
699 /* faster to handle last block with cpu */
701 atmel_sha_fill_padding(ctx, 0);
704 err = atmel_sha_xmit_cpu(dd, ctx->buffer, count, 1);
707 dev_dbg(dd->dev, "final_req: err: %d\n", err);
712 static void atmel_sha_copy_hash(struct ahash_request *req)
714 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
715 u32 *hash = (u32 *)ctx->digest;
718 if (ctx->flags & SHA_FLAGS_SHA1)
719 for (i = 0; i < SHA1_DIGEST_SIZE / sizeof(u32); i++)
720 hash[i] = atmel_sha_read(ctx->dd, SHA_REG_DIGEST(i));
721 else if (ctx->flags & SHA_FLAGS_SHA224)
722 for (i = 0; i < SHA224_DIGEST_SIZE / sizeof(u32); i++)
723 hash[i] = atmel_sha_read(ctx->dd, SHA_REG_DIGEST(i));
724 else if (ctx->flags & SHA_FLAGS_SHA256)
725 for (i = 0; i < SHA256_DIGEST_SIZE / sizeof(u32); i++)
726 hash[i] = atmel_sha_read(ctx->dd, SHA_REG_DIGEST(i));
727 else if (ctx->flags & SHA_FLAGS_SHA384)
728 for (i = 0; i < SHA384_DIGEST_SIZE / sizeof(u32); i++)
729 hash[i] = atmel_sha_read(ctx->dd, SHA_REG_DIGEST(i));
731 for (i = 0; i < SHA512_DIGEST_SIZE / sizeof(u32); i++)
732 hash[i] = atmel_sha_read(ctx->dd, SHA_REG_DIGEST(i));
735 static void atmel_sha_copy_ready_hash(struct ahash_request *req)
737 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
742 if (ctx->flags & SHA_FLAGS_SHA1)
743 memcpy(req->result, ctx->digest, SHA1_DIGEST_SIZE);
744 else if (ctx->flags & SHA_FLAGS_SHA224)
745 memcpy(req->result, ctx->digest, SHA224_DIGEST_SIZE);
746 else if (ctx->flags & SHA_FLAGS_SHA256)
747 memcpy(req->result, ctx->digest, SHA256_DIGEST_SIZE);
748 else if (ctx->flags & SHA_FLAGS_SHA384)
749 memcpy(req->result, ctx->digest, SHA384_DIGEST_SIZE);
751 memcpy(req->result, ctx->digest, SHA512_DIGEST_SIZE);
754 static int atmel_sha_finish(struct ahash_request *req)
756 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
757 struct atmel_sha_dev *dd = ctx->dd;
760 if (ctx->digcnt[0] || ctx->digcnt[1])
761 atmel_sha_copy_ready_hash(req);
763 dev_dbg(dd->dev, "digcnt: 0x%llx 0x%llx, bufcnt: %d\n", ctx->digcnt[1],
764 ctx->digcnt[0], ctx->bufcnt);
769 static void atmel_sha_finish_req(struct ahash_request *req, int err)
771 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
772 struct atmel_sha_dev *dd = ctx->dd;
775 atmel_sha_copy_hash(req);
776 if (SHA_FLAGS_FINAL & dd->flags)
777 err = atmel_sha_finish(req);
779 ctx->flags |= SHA_FLAGS_ERROR;
782 /* atomic operation is not needed here */
783 dd->flags &= ~(SHA_FLAGS_BUSY | SHA_FLAGS_FINAL | SHA_FLAGS_CPU |
784 SHA_FLAGS_DMA_READY | SHA_FLAGS_OUTPUT_READY);
786 clk_disable_unprepare(dd->iclk);
788 if (req->base.complete)
789 req->base.complete(&req->base, err);
791 /* handle new request */
792 tasklet_schedule(&dd->done_task);
795 static int atmel_sha_hw_init(struct atmel_sha_dev *dd)
799 err = clk_prepare_enable(dd->iclk);
803 if (!(SHA_FLAGS_INIT & dd->flags)) {
804 atmel_sha_write(dd, SHA_CR, SHA_CR_SWRST);
805 dd->flags |= SHA_FLAGS_INIT;
812 static inline unsigned int atmel_sha_get_version(struct atmel_sha_dev *dd)
814 return atmel_sha_read(dd, SHA_HW_VERSION) & 0x00000fff;
817 static void atmel_sha_hw_version_init(struct atmel_sha_dev *dd)
819 atmel_sha_hw_init(dd);
821 dd->hw_version = atmel_sha_get_version(dd);
824 "version: 0x%x\n", dd->hw_version);
826 clk_disable_unprepare(dd->iclk);
829 static int atmel_sha_handle_queue(struct atmel_sha_dev *dd,
830 struct ahash_request *req)
832 struct crypto_async_request *async_req, *backlog;
833 struct atmel_sha_reqctx *ctx;
835 int err = 0, ret = 0;
837 spin_lock_irqsave(&dd->lock, flags);
839 ret = ahash_enqueue_request(&dd->queue, req);
841 if (SHA_FLAGS_BUSY & dd->flags) {
842 spin_unlock_irqrestore(&dd->lock, flags);
846 backlog = crypto_get_backlog(&dd->queue);
847 async_req = crypto_dequeue_request(&dd->queue);
849 dd->flags |= SHA_FLAGS_BUSY;
851 spin_unlock_irqrestore(&dd->lock, flags);
857 backlog->complete(backlog, -EINPROGRESS);
859 req = ahash_request_cast(async_req);
861 ctx = ahash_request_ctx(req);
863 dev_dbg(dd->dev, "handling new req, op: %lu, nbytes: %d\n",
864 ctx->op, req->nbytes);
866 err = atmel_sha_hw_init(dd);
871 if (ctx->op == SHA_OP_UPDATE) {
872 err = atmel_sha_update_req(dd);
873 if (err != -EINPROGRESS && (ctx->flags & SHA_FLAGS_FINUP))
874 /* no final() after finup() */
875 err = atmel_sha_final_req(dd);
876 } else if (ctx->op == SHA_OP_FINAL) {
877 err = atmel_sha_final_req(dd);
881 if (err != -EINPROGRESS)
882 /* done_task will not finish it, so do it here */
883 atmel_sha_finish_req(req, err);
885 dev_dbg(dd->dev, "exit, err: %d\n", err);
890 static int atmel_sha_enqueue(struct ahash_request *req, unsigned int op)
892 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
893 struct atmel_sha_ctx *tctx = crypto_tfm_ctx(req->base.tfm);
894 struct atmel_sha_dev *dd = tctx->dd;
898 return atmel_sha_handle_queue(dd, req);
901 static int atmel_sha_update(struct ahash_request *req)
903 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
908 ctx->total = req->nbytes;
912 if (ctx->flags & SHA_FLAGS_FINUP) {
913 if (ctx->bufcnt + ctx->total < ATMEL_SHA_DMA_THRESHOLD)
914 /* faster to use CPU for short transfers */
915 ctx->flags |= SHA_FLAGS_CPU;
916 } else if (ctx->bufcnt + ctx->total < ctx->buflen) {
917 atmel_sha_append_sg(ctx);
920 return atmel_sha_enqueue(req, SHA_OP_UPDATE);
923 static int atmel_sha_final(struct ahash_request *req)
925 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
926 struct atmel_sha_ctx *tctx = crypto_tfm_ctx(req->base.tfm);
927 struct atmel_sha_dev *dd = tctx->dd;
931 ctx->flags |= SHA_FLAGS_FINUP;
933 if (ctx->flags & SHA_FLAGS_ERROR)
934 return 0; /* uncompleted hash is not needed */
937 return atmel_sha_enqueue(req, SHA_OP_FINAL);
938 } else if (!(ctx->flags & SHA_FLAGS_PAD)) { /* add padding */
939 err = atmel_sha_hw_init(dd);
943 dd->flags |= SHA_FLAGS_BUSY;
944 err = atmel_sha_final_req(dd);
946 /* copy ready hash (+ finalize hmac) */
947 return atmel_sha_finish(req);
951 if (err != -EINPROGRESS)
952 /* done_task will not finish it, so do it here */
953 atmel_sha_finish_req(req, err);
958 static int atmel_sha_finup(struct ahash_request *req)
960 struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
963 ctx->flags |= SHA_FLAGS_FINUP;
965 err1 = atmel_sha_update(req);
966 if (err1 == -EINPROGRESS || err1 == -EBUSY)
970 * final() has to be always called to cleanup resources
971 * even if udpate() failed, except EINPROGRESS
973 err2 = atmel_sha_final(req);
978 static int atmel_sha_digest(struct ahash_request *req)
980 return atmel_sha_init(req) ?: atmel_sha_finup(req);
983 static int atmel_sha_cra_init(struct crypto_tfm *tfm)
985 crypto_ahash_set_reqsize(__crypto_ahash_cast(tfm),
986 sizeof(struct atmel_sha_reqctx) +
987 SHA_BUFFER_LEN + SHA512_BLOCK_SIZE);
992 static struct ahash_alg sha_1_256_algs[] = {
994 .init = atmel_sha_init,
995 .update = atmel_sha_update,
996 .final = atmel_sha_final,
997 .finup = atmel_sha_finup,
998 .digest = atmel_sha_digest,
1000 .digestsize = SHA1_DIGEST_SIZE,
1003 .cra_driver_name = "atmel-sha1",
1004 .cra_priority = 100,
1005 .cra_flags = CRYPTO_ALG_ASYNC,
1006 .cra_blocksize = SHA1_BLOCK_SIZE,
1007 .cra_ctxsize = sizeof(struct atmel_sha_ctx),
1009 .cra_module = THIS_MODULE,
1010 .cra_init = atmel_sha_cra_init,
1015 .init = atmel_sha_init,
1016 .update = atmel_sha_update,
1017 .final = atmel_sha_final,
1018 .finup = atmel_sha_finup,
1019 .digest = atmel_sha_digest,
1021 .digestsize = SHA256_DIGEST_SIZE,
1023 .cra_name = "sha256",
1024 .cra_driver_name = "atmel-sha256",
1025 .cra_priority = 100,
1026 .cra_flags = CRYPTO_ALG_ASYNC,
1027 .cra_blocksize = SHA256_BLOCK_SIZE,
1028 .cra_ctxsize = sizeof(struct atmel_sha_ctx),
1030 .cra_module = THIS_MODULE,
1031 .cra_init = atmel_sha_cra_init,
1037 static struct ahash_alg sha_224_alg = {
1038 .init = atmel_sha_init,
1039 .update = atmel_sha_update,
1040 .final = atmel_sha_final,
1041 .finup = atmel_sha_finup,
1042 .digest = atmel_sha_digest,
1044 .digestsize = SHA224_DIGEST_SIZE,
1046 .cra_name = "sha224",
1047 .cra_driver_name = "atmel-sha224",
1048 .cra_priority = 100,
1049 .cra_flags = CRYPTO_ALG_ASYNC,
1050 .cra_blocksize = SHA224_BLOCK_SIZE,
1051 .cra_ctxsize = sizeof(struct atmel_sha_ctx),
1053 .cra_module = THIS_MODULE,
1054 .cra_init = atmel_sha_cra_init,
1059 static struct ahash_alg sha_384_512_algs[] = {
1061 .init = atmel_sha_init,
1062 .update = atmel_sha_update,
1063 .final = atmel_sha_final,
1064 .finup = atmel_sha_finup,
1065 .digest = atmel_sha_digest,
1067 .digestsize = SHA384_DIGEST_SIZE,
1069 .cra_name = "sha384",
1070 .cra_driver_name = "atmel-sha384",
1071 .cra_priority = 100,
1072 .cra_flags = CRYPTO_ALG_ASYNC,
1073 .cra_blocksize = SHA384_BLOCK_SIZE,
1074 .cra_ctxsize = sizeof(struct atmel_sha_ctx),
1075 .cra_alignmask = 0x3,
1076 .cra_module = THIS_MODULE,
1077 .cra_init = atmel_sha_cra_init,
1082 .init = atmel_sha_init,
1083 .update = atmel_sha_update,
1084 .final = atmel_sha_final,
1085 .finup = atmel_sha_finup,
1086 .digest = atmel_sha_digest,
1088 .digestsize = SHA512_DIGEST_SIZE,
1090 .cra_name = "sha512",
1091 .cra_driver_name = "atmel-sha512",
1092 .cra_priority = 100,
1093 .cra_flags = CRYPTO_ALG_ASYNC,
1094 .cra_blocksize = SHA512_BLOCK_SIZE,
1095 .cra_ctxsize = sizeof(struct atmel_sha_ctx),
1096 .cra_alignmask = 0x3,
1097 .cra_module = THIS_MODULE,
1098 .cra_init = atmel_sha_cra_init,
1104 static void atmel_sha_done_task(unsigned long data)
1106 struct atmel_sha_dev *dd = (struct atmel_sha_dev *)data;
1109 if (!(SHA_FLAGS_BUSY & dd->flags)) {
1110 atmel_sha_handle_queue(dd, NULL);
1114 if (SHA_FLAGS_CPU & dd->flags) {
1115 if (SHA_FLAGS_OUTPUT_READY & dd->flags) {
1116 dd->flags &= ~SHA_FLAGS_OUTPUT_READY;
1119 } else if (SHA_FLAGS_DMA_READY & dd->flags) {
1120 if (SHA_FLAGS_DMA_ACTIVE & dd->flags) {
1121 dd->flags &= ~SHA_FLAGS_DMA_ACTIVE;
1122 atmel_sha_update_dma_stop(dd);
1128 if (SHA_FLAGS_OUTPUT_READY & dd->flags) {
1129 /* hash or semi-hash ready */
1130 dd->flags &= ~(SHA_FLAGS_DMA_READY |
1131 SHA_FLAGS_OUTPUT_READY);
1132 err = atmel_sha_update_dma_start(dd);
1133 if (err != -EINPROGRESS)
1140 /* finish curent request */
1141 atmel_sha_finish_req(dd->req, err);
1144 static irqreturn_t atmel_sha_irq(int irq, void *dev_id)
1146 struct atmel_sha_dev *sha_dd = dev_id;
1149 reg = atmel_sha_read(sha_dd, SHA_ISR);
1150 if (reg & atmel_sha_read(sha_dd, SHA_IMR)) {
1151 atmel_sha_write(sha_dd, SHA_IDR, reg);
1152 if (SHA_FLAGS_BUSY & sha_dd->flags) {
1153 sha_dd->flags |= SHA_FLAGS_OUTPUT_READY;
1154 if (!(SHA_FLAGS_CPU & sha_dd->flags))
1155 sha_dd->flags |= SHA_FLAGS_DMA_READY;
1156 tasklet_schedule(&sha_dd->done_task);
1158 dev_warn(sha_dd->dev, "SHA interrupt when no active requests.\n");
1166 static void atmel_sha_unregister_algs(struct atmel_sha_dev *dd)
1170 for (i = 0; i < ARRAY_SIZE(sha_1_256_algs); i++)
1171 crypto_unregister_ahash(&sha_1_256_algs[i]);
1173 if (dd->caps.has_sha224)
1174 crypto_unregister_ahash(&sha_224_alg);
1176 if (dd->caps.has_sha_384_512) {
1177 for (i = 0; i < ARRAY_SIZE(sha_384_512_algs); i++)
1178 crypto_unregister_ahash(&sha_384_512_algs[i]);
1182 static int atmel_sha_register_algs(struct atmel_sha_dev *dd)
1186 for (i = 0; i < ARRAY_SIZE(sha_1_256_algs); i++) {
1187 err = crypto_register_ahash(&sha_1_256_algs[i]);
1189 goto err_sha_1_256_algs;
1192 if (dd->caps.has_sha224) {
1193 err = crypto_register_ahash(&sha_224_alg);
1195 goto err_sha_224_algs;
1198 if (dd->caps.has_sha_384_512) {
1199 for (i = 0; i < ARRAY_SIZE(sha_384_512_algs); i++) {
1200 err = crypto_register_ahash(&sha_384_512_algs[i]);
1202 goto err_sha_384_512_algs;
1208 err_sha_384_512_algs:
1209 for (j = 0; j < i; j++)
1210 crypto_unregister_ahash(&sha_384_512_algs[j]);
1211 crypto_unregister_ahash(&sha_224_alg);
1213 i = ARRAY_SIZE(sha_1_256_algs);
1215 for (j = 0; j < i; j++)
1216 crypto_unregister_ahash(&sha_1_256_algs[j]);
1221 static bool atmel_sha_filter(struct dma_chan *chan, void *slave)
1223 struct at_dma_slave *sl = slave;
1225 if (sl && sl->dma_dev == chan->device->dev) {
1233 static int atmel_sha_dma_init(struct atmel_sha_dev *dd,
1234 struct crypto_platform_data *pdata)
1237 dma_cap_mask_t mask_in;
1239 /* Try to grab DMA channel */
1240 dma_cap_zero(mask_in);
1241 dma_cap_set(DMA_SLAVE, mask_in);
1243 dd->dma_lch_in.chan = dma_request_slave_channel_compat(mask_in,
1244 atmel_sha_filter, &pdata->dma_slave->rxdata, dd->dev, "tx");
1245 if (!dd->dma_lch_in.chan) {
1246 dev_warn(dd->dev, "no DMA channel available\n");
1250 dd->dma_lch_in.dma_conf.direction = DMA_MEM_TO_DEV;
1251 dd->dma_lch_in.dma_conf.dst_addr = dd->phys_base +
1253 dd->dma_lch_in.dma_conf.src_maxburst = 1;
1254 dd->dma_lch_in.dma_conf.src_addr_width =
1255 DMA_SLAVE_BUSWIDTH_4_BYTES;
1256 dd->dma_lch_in.dma_conf.dst_maxburst = 1;
1257 dd->dma_lch_in.dma_conf.dst_addr_width =
1258 DMA_SLAVE_BUSWIDTH_4_BYTES;
1259 dd->dma_lch_in.dma_conf.device_fc = false;
1264 static void atmel_sha_dma_cleanup(struct atmel_sha_dev *dd)
1266 dma_release_channel(dd->dma_lch_in.chan);
1269 static void atmel_sha_get_cap(struct atmel_sha_dev *dd)
1272 dd->caps.has_dma = 0;
1273 dd->caps.has_dualbuff = 0;
1274 dd->caps.has_sha224 = 0;
1275 dd->caps.has_sha_384_512 = 0;
1277 /* keep only major version number */
1278 switch (dd->hw_version & 0xff0) {
1280 dd->caps.has_dma = 1;
1281 dd->caps.has_dualbuff = 1;
1282 dd->caps.has_sha224 = 1;
1283 dd->caps.has_sha_384_512 = 1;
1286 dd->caps.has_dma = 1;
1287 dd->caps.has_dualbuff = 1;
1288 dd->caps.has_sha224 = 1;
1289 dd->caps.has_sha_384_512 = 1;
1292 dd->caps.has_dma = 1;
1293 dd->caps.has_dualbuff = 1;
1294 dd->caps.has_sha224 = 1;
1300 "Unmanaged sha version, set minimum capabilities\n");
1305 #if defined(CONFIG_OF)
1306 static const struct of_device_id atmel_sha_dt_ids[] = {
1307 { .compatible = "atmel,at91sam9g46-sha" },
1311 MODULE_DEVICE_TABLE(of, atmel_sha_dt_ids);
1313 static struct crypto_platform_data *atmel_sha_of_init(struct platform_device *pdev)
1315 struct device_node *np = pdev->dev.of_node;
1316 struct crypto_platform_data *pdata;
1319 dev_err(&pdev->dev, "device node not found\n");
1320 return ERR_PTR(-EINVAL);
1323 pdata = devm_kzalloc(&pdev->dev, sizeof(*pdata), GFP_KERNEL);
1325 dev_err(&pdev->dev, "could not allocate memory for pdata\n");
1326 return ERR_PTR(-ENOMEM);
1329 pdata->dma_slave = devm_kzalloc(&pdev->dev,
1330 sizeof(*(pdata->dma_slave)),
1332 if (!pdata->dma_slave) {
1333 dev_err(&pdev->dev, "could not allocate memory for dma_slave\n");
1334 return ERR_PTR(-ENOMEM);
1339 #else /* CONFIG_OF */
1340 static inline struct crypto_platform_data *atmel_sha_of_init(struct platform_device *dev)
1342 return ERR_PTR(-EINVAL);
1346 static int atmel_sha_probe(struct platform_device *pdev)
1348 struct atmel_sha_dev *sha_dd;
1349 struct crypto_platform_data *pdata;
1350 struct device *dev = &pdev->dev;
1351 struct resource *sha_res;
1354 sha_dd = devm_kzalloc(&pdev->dev, sizeof(*sha_dd), GFP_KERNEL);
1355 if (sha_dd == NULL) {
1356 dev_err(dev, "unable to alloc data struct.\n");
1363 platform_set_drvdata(pdev, sha_dd);
1365 INIT_LIST_HEAD(&sha_dd->list);
1366 spin_lock_init(&sha_dd->lock);
1368 tasklet_init(&sha_dd->done_task, atmel_sha_done_task,
1369 (unsigned long)sha_dd);
1371 crypto_init_queue(&sha_dd->queue, ATMEL_SHA_QUEUE_LENGTH);
1375 /* Get the base address */
1376 sha_res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1378 dev_err(dev, "no MEM resource info\n");
1382 sha_dd->phys_base = sha_res->start;
1385 sha_dd->irq = platform_get_irq(pdev, 0);
1386 if (sha_dd->irq < 0) {
1387 dev_err(dev, "no IRQ resource info\n");
1392 err = devm_request_irq(&pdev->dev, sha_dd->irq, atmel_sha_irq,
1393 IRQF_SHARED, "atmel-sha", sha_dd);
1395 dev_err(dev, "unable to request sha irq.\n");
1399 /* Initializing the clock */
1400 sha_dd->iclk = devm_clk_get(&pdev->dev, "sha_clk");
1401 if (IS_ERR(sha_dd->iclk)) {
1402 dev_err(dev, "clock initialization failed.\n");
1403 err = PTR_ERR(sha_dd->iclk);
1407 sha_dd->io_base = devm_ioremap_resource(&pdev->dev, sha_res);
1408 if (!sha_dd->io_base) {
1409 dev_err(dev, "can't ioremap\n");
1414 atmel_sha_hw_version_init(sha_dd);
1416 atmel_sha_get_cap(sha_dd);
1418 if (sha_dd->caps.has_dma) {
1419 pdata = pdev->dev.platform_data;
1421 pdata = atmel_sha_of_init(pdev);
1422 if (IS_ERR(pdata)) {
1423 dev_err(&pdev->dev, "platform data not available\n");
1424 err = PTR_ERR(pdata);
1428 if (!pdata->dma_slave) {
1432 err = atmel_sha_dma_init(sha_dd, pdata);
1436 dev_info(dev, "using %s for DMA transfers\n",
1437 dma_chan_name(sha_dd->dma_lch_in.chan));
1440 spin_lock(&atmel_sha.lock);
1441 list_add_tail(&sha_dd->list, &atmel_sha.dev_list);
1442 spin_unlock(&atmel_sha.lock);
1444 err = atmel_sha_register_algs(sha_dd);
1448 dev_info(dev, "Atmel SHA1/SHA256%s%s\n",
1449 sha_dd->caps.has_sha224 ? "/SHA224" : "",
1450 sha_dd->caps.has_sha_384_512 ? "/SHA384/SHA512" : "");
1455 spin_lock(&atmel_sha.lock);
1456 list_del(&sha_dd->list);
1457 spin_unlock(&atmel_sha.lock);
1458 if (sha_dd->caps.has_dma)
1459 atmel_sha_dma_cleanup(sha_dd);
1462 tasklet_kill(&sha_dd->done_task);
1464 dev_err(dev, "initialization failed.\n");
1469 static int atmel_sha_remove(struct platform_device *pdev)
1471 static struct atmel_sha_dev *sha_dd;
1473 sha_dd = platform_get_drvdata(pdev);
1476 spin_lock(&atmel_sha.lock);
1477 list_del(&sha_dd->list);
1478 spin_unlock(&atmel_sha.lock);
1480 atmel_sha_unregister_algs(sha_dd);
1482 tasklet_kill(&sha_dd->done_task);
1484 if (sha_dd->caps.has_dma)
1485 atmel_sha_dma_cleanup(sha_dd);
1487 iounmap(sha_dd->io_base);
1489 clk_put(sha_dd->iclk);
1491 if (sha_dd->irq >= 0)
1492 free_irq(sha_dd->irq, sha_dd);
1497 static struct platform_driver atmel_sha_driver = {
1498 .probe = atmel_sha_probe,
1499 .remove = atmel_sha_remove,
1501 .name = "atmel_sha",
1502 .of_match_table = of_match_ptr(atmel_sha_dt_ids),
1506 module_platform_driver(atmel_sha_driver);
1508 MODULE_DESCRIPTION("Atmel SHA (1/256/224/384/512) hw acceleration support.");
1509 MODULE_LICENSE("GPL v2");
1510 MODULE_AUTHOR("Nicolas Royer - Eukréa Electromatique");