crypto: atmel - add Atmel SHA1/SHA256 driver
authorNicolas Royer <nicolas@eukrea.com>
Sun, 1 Jul 2012 17:19:46 +0000 (19:19 +0200)
committerHerbert Xu <herbert@gondor.apana.org.au>
Wed, 11 Jul 2012 03:08:28 +0000 (11:08 +0800)
Signed-off-by: Nicolas Royer <nicolas@eukrea.com>
Acked-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Acked-by: Eric Bénard <eric@eukrea.com>
Tested-by: Eric Bénard <eric@eukrea.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
drivers/crypto/Kconfig
drivers/crypto/Makefile
drivers/crypto/atmel-sha-regs.h [new file with mode: 0644]
drivers/crypto/atmel-sha.c [new file with mode: 0644]

index 76d489b..7d74d09 100644 (file)
@@ -365,4 +365,18 @@ config CRYPTO_DEV_ATMEL_TDES
          To compile this driver as a module, choose M here: the module
          will be called atmel-tdes.
 
+config CRYPTO_DEV_ATMEL_SHA
+       tristate "Support for Atmel SHA1/SHA256 hw accelerator"
+       depends on ARCH_AT91
+       select CRYPTO_SHA1
+       select CRYPTO_SHA256
+       select CRYPTO_ALGAPI
+       help
+         Some Atmel processors have SHA1/SHA256 hw accelerator.
+         Select this if you want to use the Atmel module for
+         SHA1/SHA256 algorithms.
+
+         To compile this driver as a module, choose M here: the module
+         will be called atmel-sha.
+
 endif # CRYPTO_HW
index 6b8b75f..880a47b 100644 (file)
@@ -19,3 +19,4 @@ obj-$(CONFIG_CRYPTO_DEV_BFIN_CRC) += bfin_crc.o
 obj-$(CONFIG_CRYPTO_DEV_NX) += nx/
 obj-$(CONFIG_CRYPTO_DEV_ATMEL_AES) += atmel-aes.o
 obj-$(CONFIG_CRYPTO_DEV_ATMEL_TDES) += atmel-tdes.o
+obj-$(CONFIG_CRYPTO_DEV_ATMEL_SHA) += atmel-sha.o
diff --git a/drivers/crypto/atmel-sha-regs.h b/drivers/crypto/atmel-sha-regs.h
new file mode 100644 (file)
index 0000000..dc53a20
--- /dev/null
@@ -0,0 +1,46 @@
+#ifndef __ATMEL_SHA_REGS_H__
+#define __ATMEL_SHA_REGS_H__
+
+#define SHA_REG_DIGEST(x)              (0x80 + ((x) * 0x04))
+#define SHA_REG_DIN(x)                 (0x40 + ((x) * 0x04))
+
+#define SHA_CR                         0x00
+#define SHA_CR_START                   (1 << 0)
+#define SHA_CR_FIRST                   (1 << 4)
+#define SHA_CR_SWRST                   (1 << 8)
+
+#define SHA_MR                         0x04
+#define SHA_MR_MODE_MASK               (0x3 << 0)
+#define SHA_MR_MODE_MANUAL             0x0
+#define SHA_MR_MODE_AUTO               0x1
+#define SHA_MR_MODE_PDC                        0x2
+#define        SHA_MR_DUALBUFF                 (1 << 3)
+#define SHA_MR_PROCDLY                 (1 << 4)
+#define SHA_MR_ALGO_SHA1               (0 << 8)
+#define SHA_MR_ALGO_SHA256             (1 << 8)
+
+#define SHA_IER                                0x10
+#define SHA_IDR                                0x14
+#define SHA_IMR                                0x18
+#define SHA_ISR                                0x1C
+#define SHA_INT_DATARDY                        (1 << 0)
+#define SHA_INT_ENDTX                  (1 << 1)
+#define SHA_INT_TXBUFE                 (1 << 2)
+#define SHA_INT_URAD                   (1 << 8)
+#define SHA_ISR_URAT_MASK              (0x7 << 12)
+#define SHA_ISR_URAT_IDR               (0x0 << 12)
+#define SHA_ISR_URAT_ODR               (0x1 << 12)
+#define SHA_ISR_URAT_MR                        (0x2 << 12)
+#define SHA_ISR_URAT_WO                        (0x5 << 12)
+
+#define SHA_TPR                                0x108
+#define SHA_TCR                                0x10C
+#define SHA_TNPR                       0x118
+#define SHA_TNCR                       0x11C
+#define SHA_PTCR                       0x120
+#define SHA_PTCR_TXTEN         (1 << 8)
+#define SHA_PTCR_TXTDIS                (1 << 9)
+#define SHA_PTSR                       0x124
+#define SHA_PTSR_TXTEN         (1 << 8)
+
+#endif /* __ATMEL_SHA_REGS_H__ */
diff --git a/drivers/crypto/atmel-sha.c b/drivers/crypto/atmel-sha.c
new file mode 100644 (file)
index 0000000..f938b9d
--- /dev/null
@@ -0,0 +1,1112 @@
+/*
+ * Cryptographic API.
+ *
+ * Support for ATMEL SHA1/SHA256 HW acceleration.
+ *
+ * Copyright (c) 2012 Eukréa Electromatique - ATMEL
+ * Author: Nicolas Royer <nicolas@eukrea.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation.
+ *
+ * Some ideas are from omap-sham.c drivers.
+ */
+
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/err.h>
+#include <linux/clk.h>
+#include <linux/io.h>
+#include <linux/hw_random.h>
+#include <linux/platform_device.h>
+
+#include <linux/device.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <linux/clk.h>
+#include <linux/irq.h>
+#include <linux/io.h>
+#include <linux/platform_device.h>
+#include <linux/scatterlist.h>
+#include <linux/dma-mapping.h>
+#include <linux/delay.h>
+#include <linux/crypto.h>
+#include <linux/cryptohash.h>
+#include <crypto/scatterwalk.h>
+#include <crypto/algapi.h>
+#include <crypto/sha.h>
+#include <crypto/hash.h>
+#include <crypto/internal/hash.h>
+#include "atmel-sha-regs.h"
+
+/* SHA flags */
+#define SHA_FLAGS_BUSY                 BIT(0)
+#define        SHA_FLAGS_FINAL                 BIT(1)
+#define SHA_FLAGS_DMA_ACTIVE   BIT(2)
+#define SHA_FLAGS_OUTPUT_READY BIT(3)
+#define SHA_FLAGS_INIT                 BIT(4)
+#define SHA_FLAGS_CPU                  BIT(5)
+#define SHA_FLAGS_DMA_READY            BIT(6)
+
+#define SHA_FLAGS_FINUP                BIT(16)
+#define SHA_FLAGS_SG           BIT(17)
+#define SHA_FLAGS_SHA1         BIT(18)
+#define SHA_FLAGS_SHA256       BIT(19)
+#define SHA_FLAGS_ERROR                BIT(20)
+#define SHA_FLAGS_PAD          BIT(21)
+
+#define SHA_FLAGS_DUALBUFF     BIT(24)
+
+#define SHA_OP_UPDATE  1
+#define SHA_OP_FINAL   2
+
+#define SHA_BUFFER_LEN         PAGE_SIZE
+
+#define ATMEL_SHA_DMA_THRESHOLD                56
+
+
+struct atmel_sha_dev;
+
+struct atmel_sha_reqctx {
+       struct atmel_sha_dev    *dd;
+       unsigned long   flags;
+       unsigned long   op;
+
+       u8      digest[SHA256_DIGEST_SIZE] __aligned(sizeof(u32));
+       size_t  digcnt;
+       size_t  bufcnt;
+       size_t  buflen;
+       dma_addr_t      dma_addr;
+
+       /* walk state */
+       struct scatterlist      *sg;
+       unsigned int    offset; /* offset in current sg */
+       unsigned int    total;  /* total request */
+
+       u8      buffer[0] __aligned(sizeof(u32));
+};
+
+struct atmel_sha_ctx {
+       struct atmel_sha_dev    *dd;
+
+       unsigned long           flags;
+
+       /* fallback stuff */
+       struct crypto_shash     *fallback;
+
+};
+
+#define ATMEL_SHA_QUEUE_LENGTH 1
+
+struct atmel_sha_dev {
+       struct list_head        list;
+       unsigned long           phys_base;
+       struct device           *dev;
+       struct clk                      *iclk;
+       int                                     irq;
+       void __iomem            *io_base;
+
+       spinlock_t              lock;
+       int                     err;
+       struct tasklet_struct   done_task;
+
+       unsigned long           flags;
+       struct crypto_queue     queue;
+       struct ahash_request    *req;
+};
+
+struct atmel_sha_drv {
+       struct list_head        dev_list;
+       spinlock_t              lock;
+};
+
+static struct atmel_sha_drv atmel_sha = {
+       .dev_list = LIST_HEAD_INIT(atmel_sha.dev_list),
+       .lock = __SPIN_LOCK_UNLOCKED(atmel_sha.lock),
+};
+
+static inline u32 atmel_sha_read(struct atmel_sha_dev *dd, u32 offset)
+{
+       return readl_relaxed(dd->io_base + offset);
+}
+
+static inline void atmel_sha_write(struct atmel_sha_dev *dd,
+                                       u32 offset, u32 value)
+{
+       writel_relaxed(value, dd->io_base + offset);
+}
+
+static void atmel_sha_dualbuff_test(struct atmel_sha_dev *dd)
+{
+       atmel_sha_write(dd, SHA_MR, SHA_MR_DUALBUFF);
+
+       if (atmel_sha_read(dd, SHA_MR) & SHA_MR_DUALBUFF)
+               dd->flags |= SHA_FLAGS_DUALBUFF;
+}
+
+static size_t atmel_sha_append_sg(struct atmel_sha_reqctx *ctx)
+{
+       size_t count;
+
+       while ((ctx->bufcnt < ctx->buflen) && ctx->total) {
+               count = min(ctx->sg->length - ctx->offset, ctx->total);
+               count = min(count, ctx->buflen - ctx->bufcnt);
+
+               if (count <= 0)
+                       break;
+
+               scatterwalk_map_and_copy(ctx->buffer + ctx->bufcnt, ctx->sg,
+                       ctx->offset, count, 0);
+
+               ctx->bufcnt += count;
+               ctx->offset += count;
+               ctx->total -= count;
+
+               if (ctx->offset == ctx->sg->length) {
+                       ctx->sg = sg_next(ctx->sg);
+                       if (ctx->sg)
+                               ctx->offset = 0;
+                       else
+                               ctx->total = 0;
+               }
+       }
+
+       return 0;
+}
+
+/*
+ * The purpose of this padding is to ensure that the padded message
+ * is a multiple of 512 bits. The bit "1" is appended at the end of
+ * the message followed by "padlen-1" zero bits. Then a 64 bits block
+ * equals to the message length in bits is appended.
+ *
+ * padlen is calculated as followed:
+ *  - if message length < 56 bytes then padlen = 56 - message length
+ *  - else padlen = 64 + 56 - message length
+ */
+static void atmel_sha_fill_padding(struct atmel_sha_reqctx *ctx, int length)
+{
+       unsigned int index, padlen;
+       u64 bits;
+       u64 size;
+
+       bits = (ctx->bufcnt + ctx->digcnt + length) << 3;
+       size = cpu_to_be64(bits);
+
+       index = ctx->bufcnt & 0x3f;
+       padlen = (index < 56) ? (56 - index) : ((64+56) - index);
+       *(ctx->buffer + ctx->bufcnt) = 0x80;
+       memset(ctx->buffer + ctx->bufcnt + 1, 0, padlen-1);
+       memcpy(ctx->buffer + ctx->bufcnt + padlen, &size, 8);
+       ctx->bufcnt += padlen + 8;
+       ctx->flags |= SHA_FLAGS_PAD;
+}
+
+static int atmel_sha_init(struct ahash_request *req)
+{
+       struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
+       struct atmel_sha_ctx *tctx = crypto_ahash_ctx(tfm);
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+       struct atmel_sha_dev *dd = NULL;
+       struct atmel_sha_dev *tmp;
+
+       spin_lock_bh(&atmel_sha.lock);
+       if (!tctx->dd) {
+               list_for_each_entry(tmp, &atmel_sha.dev_list, list) {
+                       dd = tmp;
+                       break;
+               }
+               tctx->dd = dd;
+       } else {
+               dd = tctx->dd;
+       }
+
+       spin_unlock_bh(&atmel_sha.lock);
+
+       ctx->dd = dd;
+
+       ctx->flags = 0;
+
+       dev_dbg(dd->dev, "init: digest size: %d\n",
+               crypto_ahash_digestsize(tfm));
+
+       if (crypto_ahash_digestsize(tfm) == SHA1_DIGEST_SIZE)
+               ctx->flags |= SHA_FLAGS_SHA1;
+       else if (crypto_ahash_digestsize(tfm) == SHA256_DIGEST_SIZE)
+               ctx->flags |= SHA_FLAGS_SHA256;
+
+       ctx->bufcnt = 0;
+       ctx->digcnt = 0;
+       ctx->buflen = SHA_BUFFER_LEN;
+
+       return 0;
+}
+
+static void atmel_sha_write_ctrl(struct atmel_sha_dev *dd, int dma)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
+       u32 valcr = 0, valmr = SHA_MR_MODE_AUTO;
+
+       if (likely(dma)) {
+               atmel_sha_write(dd, SHA_IER, SHA_INT_TXBUFE);
+               valmr = SHA_MR_MODE_PDC;
+               if (dd->flags & SHA_FLAGS_DUALBUFF)
+                       valmr = SHA_MR_DUALBUFF;
+       } else {
+               atmel_sha_write(dd, SHA_IER, SHA_INT_DATARDY);
+       }
+
+       if (ctx->flags & SHA_FLAGS_SHA256)
+               valmr |= SHA_MR_ALGO_SHA256;
+
+       /* Setting CR_FIRST only for the first iteration */
+       if (!ctx->digcnt)
+               valcr = SHA_CR_FIRST;
+
+       atmel_sha_write(dd, SHA_CR, valcr);
+       atmel_sha_write(dd, SHA_MR, valmr);
+}
+
+static int atmel_sha_xmit_cpu(struct atmel_sha_dev *dd, const u8 *buf,
+                             size_t length, int final)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
+       int count, len32;
+       const u32 *buffer = (const u32 *)buf;
+
+       dev_dbg(dd->dev, "xmit_cpu: digcnt: %d, length: %d, final: %d\n",
+                                               ctx->digcnt, length, final);
+
+       atmel_sha_write_ctrl(dd, 0);
+
+       /* should be non-zero before next lines to disable clocks later */
+       ctx->digcnt += length;
+
+       if (final)
+               dd->flags |= SHA_FLAGS_FINAL; /* catch last interrupt */
+
+       len32 = DIV_ROUND_UP(length, sizeof(u32));
+
+       dd->flags |= SHA_FLAGS_CPU;
+
+       for (count = 0; count < len32; count++)
+               atmel_sha_write(dd, SHA_REG_DIN(count), buffer[count]);
+
+       return -EINPROGRESS;
+}
+
+static int atmel_sha_xmit_pdc(struct atmel_sha_dev *dd, dma_addr_t dma_addr1,
+               size_t length1, dma_addr_t dma_addr2, size_t length2, int final)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
+       int len32;
+
+       dev_dbg(dd->dev, "xmit_pdc: digcnt: %d, length: %d, final: %d\n",
+                                               ctx->digcnt, length1, final);
+
+       len32 = DIV_ROUND_UP(length1, sizeof(u32));
+       atmel_sha_write(dd, SHA_PTCR, SHA_PTCR_TXTDIS);
+       atmel_sha_write(dd, SHA_TPR, dma_addr1);
+       atmel_sha_write(dd, SHA_TCR, len32);
+
+       len32 = DIV_ROUND_UP(length2, sizeof(u32));
+       atmel_sha_write(dd, SHA_TNPR, dma_addr2);
+       atmel_sha_write(dd, SHA_TNCR, len32);
+
+       atmel_sha_write_ctrl(dd, 1);
+
+       /* should be non-zero before next lines to disable clocks later */
+       ctx->digcnt += length1;
+
+       if (final)
+               dd->flags |= SHA_FLAGS_FINAL; /* catch last interrupt */
+
+       dd->flags |=  SHA_FLAGS_DMA_ACTIVE;
+
+       /* Start DMA transfer */
+       atmel_sha_write(dd, SHA_PTCR, SHA_PTCR_TXTEN);
+
+       return -EINPROGRESS;
+}
+
+static int atmel_sha_update_cpu(struct atmel_sha_dev *dd)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
+       int bufcnt;
+
+       atmel_sha_append_sg(ctx);
+       atmel_sha_fill_padding(ctx, 0);
+
+       bufcnt = ctx->bufcnt;
+       ctx->bufcnt = 0;
+
+       return atmel_sha_xmit_cpu(dd, ctx->buffer, bufcnt, 1);
+}
+
+static int atmel_sha_xmit_dma_map(struct atmel_sha_dev *dd,
+                                       struct atmel_sha_reqctx *ctx,
+                                       size_t length, int final)
+{
+       ctx->dma_addr = dma_map_single(dd->dev, ctx->buffer,
+                               ctx->buflen + SHA1_BLOCK_SIZE, DMA_TO_DEVICE);
+       if (dma_mapping_error(dd->dev, ctx->dma_addr)) {
+               dev_err(dd->dev, "dma %u bytes error\n", ctx->buflen +
+                               SHA1_BLOCK_SIZE);
+               return -EINVAL;
+       }
+
+       ctx->flags &= ~SHA_FLAGS_SG;
+
+       /* next call does not fail... so no unmap in the case of error */
+       return atmel_sha_xmit_pdc(dd, ctx->dma_addr, length, 0, 0, final);
+}
+
+static int atmel_sha_update_dma_slow(struct atmel_sha_dev *dd)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
+       unsigned int final;
+       size_t count;
+
+       atmel_sha_append_sg(ctx);
+
+       final = (ctx->flags & SHA_FLAGS_FINUP) && !ctx->total;
+
+       dev_dbg(dd->dev, "slow: bufcnt: %u, digcnt: %d, final: %d\n",
+                                        ctx->bufcnt, ctx->digcnt, final);
+
+       if (final)
+               atmel_sha_fill_padding(ctx, 0);
+
+       if (final || (ctx->bufcnt == ctx->buflen && ctx->total)) {
+               count = ctx->bufcnt;
+               ctx->bufcnt = 0;
+               return atmel_sha_xmit_dma_map(dd, ctx, count, final);
+       }
+
+       return 0;
+}
+
+static int atmel_sha_update_dma_start(struct atmel_sha_dev *dd)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
+       unsigned int length, final, tail;
+       struct scatterlist *sg;
+       unsigned int count;
+
+       if (!ctx->total)
+               return 0;
+
+       if (ctx->bufcnt || ctx->offset)
+               return atmel_sha_update_dma_slow(dd);
+
+       dev_dbg(dd->dev, "fast: digcnt: %d, bufcnt: %u, total: %u\n",
+                       ctx->digcnt, ctx->bufcnt, ctx->total);
+
+       sg = ctx->sg;
+
+       if (!IS_ALIGNED(sg->offset, sizeof(u32)))
+               return atmel_sha_update_dma_slow(dd);
+
+       if (!sg_is_last(sg) && !IS_ALIGNED(sg->length, SHA1_BLOCK_SIZE))
+               /* size is not SHA1_BLOCK_SIZE aligned */
+               return atmel_sha_update_dma_slow(dd);
+
+       length = min(ctx->total, sg->length);
+
+       if (sg_is_last(sg)) {
+               if (!(ctx->flags & SHA_FLAGS_FINUP)) {
+                       /* not last sg must be SHA1_BLOCK_SIZE aligned */
+                       tail = length & (SHA1_BLOCK_SIZE - 1);
+                       length -= tail;
+                       if (length == 0) {
+                               /* offset where to start slow */
+                               ctx->offset = length;
+                               return atmel_sha_update_dma_slow(dd);
+                       }
+               }
+       }
+
+       ctx->total -= length;
+       ctx->offset = length; /* offset where to start slow */
+
+       final = (ctx->flags & SHA_FLAGS_FINUP) && !ctx->total;
+
+       /* Add padding */
+       if (final) {
+               tail = length & (SHA1_BLOCK_SIZE - 1);
+               length -= tail;
+               ctx->total += tail;
+               ctx->offset = length; /* offset where to start slow */
+
+               sg = ctx->sg;
+               atmel_sha_append_sg(ctx);
+
+               atmel_sha_fill_padding(ctx, length);
+
+               ctx->dma_addr = dma_map_single(dd->dev, ctx->buffer,
+                       ctx->buflen + SHA1_BLOCK_SIZE, DMA_TO_DEVICE);
+               if (dma_mapping_error(dd->dev, ctx->dma_addr)) {
+                       dev_err(dd->dev, "dma %u bytes error\n",
+                               ctx->buflen + SHA1_BLOCK_SIZE);
+                       return -EINVAL;
+               }
+
+               if (length == 0) {
+                       ctx->flags &= ~SHA_FLAGS_SG;
+                       count = ctx->bufcnt;
+                       ctx->bufcnt = 0;
+                       return atmel_sha_xmit_pdc(dd, ctx->dma_addr, count, 0,
+                                       0, final);
+               } else {
+                       ctx->sg = sg;
+                       if (!dma_map_sg(dd->dev, ctx->sg, 1,
+                               DMA_TO_DEVICE)) {
+                                       dev_err(dd->dev, "dma_map_sg  error\n");
+                                       return -EINVAL;
+                       }
+
+                       ctx->flags |= SHA_FLAGS_SG;
+
+                       count = ctx->bufcnt;
+                       ctx->bufcnt = 0;
+                       return atmel_sha_xmit_pdc(dd, sg_dma_address(ctx->sg),
+                                       length, ctx->dma_addr, count, final);
+               }
+       }
+
+       if (!dma_map_sg(dd->dev, ctx->sg, 1, DMA_TO_DEVICE)) {
+               dev_err(dd->dev, "dma_map_sg  error\n");
+               return -EINVAL;
+       }
+
+       ctx->flags |= SHA_FLAGS_SG;
+
+       /* next call does not fail... so no unmap in the case of error */
+       return atmel_sha_xmit_pdc(dd, sg_dma_address(ctx->sg), length, 0,
+                                                               0, final);
+}
+
+static int atmel_sha_update_dma_stop(struct atmel_sha_dev *dd)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(dd->req);
+
+       if (ctx->flags & SHA_FLAGS_SG) {
+               dma_unmap_sg(dd->dev, ctx->sg, 1, DMA_TO_DEVICE);
+               if (ctx->sg->length == ctx->offset) {
+                       ctx->sg = sg_next(ctx->sg);
+                       if (ctx->sg)
+                               ctx->offset = 0;
+               }
+               if (ctx->flags & SHA_FLAGS_PAD)
+                       dma_unmap_single(dd->dev, ctx->dma_addr,
+                               ctx->buflen + SHA1_BLOCK_SIZE, DMA_TO_DEVICE);
+       } else {
+               dma_unmap_single(dd->dev, ctx->dma_addr, ctx->buflen +
+                                               SHA1_BLOCK_SIZE, DMA_TO_DEVICE);
+       }
+
+       return 0;
+}
+
+static int atmel_sha_update_req(struct atmel_sha_dev *dd)
+{
+       struct ahash_request *req = dd->req;
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+       int err;
+
+       dev_dbg(dd->dev, "update_req: total: %u, digcnt: %d, finup: %d\n",
+                ctx->total, ctx->digcnt, (ctx->flags & SHA_FLAGS_FINUP) != 0);
+
+       if (ctx->flags & SHA_FLAGS_CPU)
+               err = atmel_sha_update_cpu(dd);
+       else
+               err = atmel_sha_update_dma_start(dd);
+
+       /* wait for dma completion before can take more data */
+       dev_dbg(dd->dev, "update: err: %d, digcnt: %d\n",
+                       err, ctx->digcnt);
+
+       return err;
+}
+
+static int atmel_sha_final_req(struct atmel_sha_dev *dd)
+{
+       struct ahash_request *req = dd->req;
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+       int err = 0;
+       int count;
+
+       if (ctx->bufcnt >= ATMEL_SHA_DMA_THRESHOLD) {
+               atmel_sha_fill_padding(ctx, 0);
+               count = ctx->bufcnt;
+               ctx->bufcnt = 0;
+               err = atmel_sha_xmit_dma_map(dd, ctx, count, 1);
+       }
+       /* faster to handle last block with cpu */
+       else {
+               atmel_sha_fill_padding(ctx, 0);
+               count = ctx->bufcnt;
+               ctx->bufcnt = 0;
+               err = atmel_sha_xmit_cpu(dd, ctx->buffer, count, 1);
+       }
+
+       dev_dbg(dd->dev, "final_req: err: %d\n", err);
+
+       return err;
+}
+
+static void atmel_sha_copy_hash(struct ahash_request *req)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+       u32 *hash = (u32 *)ctx->digest;
+       int i;
+
+       if (likely(ctx->flags & SHA_FLAGS_SHA1))
+               for (i = 0; i < SHA1_DIGEST_SIZE / sizeof(u32); i++)
+                       hash[i] = atmel_sha_read(ctx->dd, SHA_REG_DIGEST(i));
+       else
+               for (i = 0; i < SHA256_DIGEST_SIZE / sizeof(u32); i++)
+                       hash[i] = atmel_sha_read(ctx->dd, SHA_REG_DIGEST(i));
+}
+
+static void atmel_sha_copy_ready_hash(struct ahash_request *req)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+
+       if (!req->result)
+               return;
+
+       if (likely(ctx->flags & SHA_FLAGS_SHA1))
+               memcpy(req->result, ctx->digest, SHA1_DIGEST_SIZE);
+       else
+               memcpy(req->result, ctx->digest, SHA256_DIGEST_SIZE);
+}
+
+static int atmel_sha_finish(struct ahash_request *req)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+       struct atmel_sha_dev *dd = ctx->dd;
+       int err = 0;
+
+       if (ctx->digcnt)
+               atmel_sha_copy_ready_hash(req);
+
+       dev_dbg(dd->dev, "digcnt: %d, bufcnt: %d\n", ctx->digcnt,
+               ctx->bufcnt);
+
+       return err;
+}
+
+static void atmel_sha_finish_req(struct ahash_request *req, int err)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+       struct atmel_sha_dev *dd = ctx->dd;
+
+       if (!err) {
+               atmel_sha_copy_hash(req);
+               if (SHA_FLAGS_FINAL & dd->flags)
+                       err = atmel_sha_finish(req);
+       } else {
+               ctx->flags |= SHA_FLAGS_ERROR;
+       }
+
+       /* atomic operation is not needed here */
+       dd->flags &= ~(SHA_FLAGS_BUSY | SHA_FLAGS_FINAL | SHA_FLAGS_CPU |
+                       SHA_FLAGS_DMA_READY | SHA_FLAGS_OUTPUT_READY);
+
+       clk_disable_unprepare(dd->iclk);
+
+       if (req->base.complete)
+               req->base.complete(&req->base, err);
+
+       /* handle new request */
+       tasklet_schedule(&dd->done_task);
+}
+
+static int atmel_sha_hw_init(struct atmel_sha_dev *dd)
+{
+       clk_prepare_enable(dd->iclk);
+
+       if (SHA_FLAGS_INIT & dd->flags) {
+               atmel_sha_write(dd, SHA_CR, SHA_CR_SWRST);
+               atmel_sha_dualbuff_test(dd);
+               dd->flags |= SHA_FLAGS_INIT;
+               dd->err = 0;
+       }
+
+       return 0;
+}
+
+static int atmel_sha_handle_queue(struct atmel_sha_dev *dd,
+                                 struct ahash_request *req)
+{
+       struct crypto_async_request *async_req, *backlog;
+       struct atmel_sha_reqctx *ctx;
+       unsigned long flags;
+       int err = 0, ret = 0;
+
+       spin_lock_irqsave(&dd->lock, flags);
+       if (req)
+               ret = ahash_enqueue_request(&dd->queue, req);
+
+       if (SHA_FLAGS_BUSY & dd->flags) {
+               spin_unlock_irqrestore(&dd->lock, flags);
+               return ret;
+       }
+
+       backlog = crypto_get_backlog(&dd->queue);
+       async_req = crypto_dequeue_request(&dd->queue);
+       if (async_req)
+               dd->flags |= SHA_FLAGS_BUSY;
+
+       spin_unlock_irqrestore(&dd->lock, flags);
+
+       if (!async_req)
+               return ret;
+
+       if (backlog)
+               backlog->complete(backlog, -EINPROGRESS);
+
+       req = ahash_request_cast(async_req);
+       dd->req = req;
+       ctx = ahash_request_ctx(req);
+
+       dev_dbg(dd->dev, "handling new req, op: %lu, nbytes: %d\n",
+                                               ctx->op, req->nbytes);
+
+       err = atmel_sha_hw_init(dd);
+
+       if (err)
+               goto err1;
+
+       if (ctx->op == SHA_OP_UPDATE) {
+               err = atmel_sha_update_req(dd);
+               if (err != -EINPROGRESS && (ctx->flags & SHA_FLAGS_FINUP)) {
+                       /* no final() after finup() */
+                       err = atmel_sha_final_req(dd);
+               }
+       } else if (ctx->op == SHA_OP_FINAL) {
+               err = atmel_sha_final_req(dd);
+       }
+
+err1:
+       if (err != -EINPROGRESS)
+               /* done_task will not finish it, so do it here */
+               atmel_sha_finish_req(req, err);
+
+       dev_dbg(dd->dev, "exit, err: %d\n", err);
+
+       return ret;
+}
+
+static int atmel_sha_enqueue(struct ahash_request *req, unsigned int op)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+       struct atmel_sha_ctx *tctx = crypto_tfm_ctx(req->base.tfm);
+       struct atmel_sha_dev *dd = tctx->dd;
+
+       ctx->op = op;
+
+       return atmel_sha_handle_queue(dd, req);
+}
+
+static int atmel_sha_update(struct ahash_request *req)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+
+       if (!req->nbytes)
+               return 0;
+
+       ctx->total = req->nbytes;
+       ctx->sg = req->src;
+       ctx->offset = 0;
+
+       if (ctx->flags & SHA_FLAGS_FINUP) {
+               if (ctx->bufcnt + ctx->total < ATMEL_SHA_DMA_THRESHOLD)
+                       /* faster to use CPU for short transfers */
+                       ctx->flags |= SHA_FLAGS_CPU;
+       } else if (ctx->bufcnt + ctx->total < ctx->buflen) {
+               atmel_sha_append_sg(ctx);
+               return 0;
+       }
+       return atmel_sha_enqueue(req, SHA_OP_UPDATE);
+}
+
+static int atmel_sha_final(struct ahash_request *req)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+       struct atmel_sha_ctx *tctx = crypto_tfm_ctx(req->base.tfm);
+       struct atmel_sha_dev *dd = tctx->dd;
+
+       int err = 0;
+
+       ctx->flags |= SHA_FLAGS_FINUP;
+
+       if (ctx->flags & SHA_FLAGS_ERROR)
+               return 0; /* uncompleted hash is not needed */
+
+       if (ctx->bufcnt) {
+               return atmel_sha_enqueue(req, SHA_OP_FINAL);
+       } else if (!(ctx->flags & SHA_FLAGS_PAD)) { /* add padding */
+               err = atmel_sha_hw_init(dd);
+               if (err)
+                       goto err1;
+
+               dd->flags |= SHA_FLAGS_BUSY;
+               err = atmel_sha_final_req(dd);
+       } else {
+               /* copy ready hash (+ finalize hmac) */
+               return atmel_sha_finish(req);
+       }
+
+err1:
+       if (err != -EINPROGRESS)
+               /* done_task will not finish it, so do it here */
+               atmel_sha_finish_req(req, err);
+
+       return err;
+}
+
+static int atmel_sha_finup(struct ahash_request *req)
+{
+       struct atmel_sha_reqctx *ctx = ahash_request_ctx(req);
+       int err1, err2;
+
+       ctx->flags |= SHA_FLAGS_FINUP;
+
+       err1 = atmel_sha_update(req);
+       if (err1 == -EINPROGRESS || err1 == -EBUSY)
+               return err1;
+
+       /*
+        * final() has to be always called to cleanup resources
+        * even if udpate() failed, except EINPROGRESS
+        */
+       err2 = atmel_sha_final(req);
+
+       return err1 ?: err2;
+}
+
+static int atmel_sha_digest(struct ahash_request *req)
+{
+       return atmel_sha_init(req) ?: atmel_sha_finup(req);
+}
+
+static int atmel_sha_cra_init_alg(struct crypto_tfm *tfm, const char *alg_base)
+{
+       struct atmel_sha_ctx *tctx = crypto_tfm_ctx(tfm);
+       const char *alg_name = crypto_tfm_alg_name(tfm);
+
+       /* Allocate a fallback and abort if it failed. */
+       tctx->fallback = crypto_alloc_shash(alg_name, 0,
+                                           CRYPTO_ALG_NEED_FALLBACK);
+       if (IS_ERR(tctx->fallback)) {
+               pr_err("atmel-sha: fallback driver '%s' could not be loaded.\n",
+                               alg_name);
+               return PTR_ERR(tctx->fallback);
+       }
+       crypto_ahash_set_reqsize(__crypto_ahash_cast(tfm),
+                                sizeof(struct atmel_sha_reqctx) +
+                                SHA_BUFFER_LEN + SHA256_BLOCK_SIZE);
+
+       return 0;
+}
+
+static int atmel_sha_cra_init(struct crypto_tfm *tfm)
+{
+       return atmel_sha_cra_init_alg(tfm, NULL);
+}
+
+static void atmel_sha_cra_exit(struct crypto_tfm *tfm)
+{
+       struct atmel_sha_ctx *tctx = crypto_tfm_ctx(tfm);
+
+       crypto_free_shash(tctx->fallback);
+       tctx->fallback = NULL;
+}
+
+static struct ahash_alg sha_algs[] = {
+{
+       .init           = atmel_sha_init,
+       .update         = atmel_sha_update,
+       .final          = atmel_sha_final,
+       .finup          = atmel_sha_finup,
+       .digest         = atmel_sha_digest,
+       .halg = {
+               .digestsize     = SHA1_DIGEST_SIZE,
+               .base   = {
+                       .cra_name               = "sha1",
+                       .cra_driver_name        = "atmel-sha1",
+                       .cra_priority           = 100,
+                       .cra_flags              = CRYPTO_ALG_ASYNC |
+                                               CRYPTO_ALG_NEED_FALLBACK,
+                       .cra_blocksize          = SHA1_BLOCK_SIZE,
+                       .cra_ctxsize            = sizeof(struct atmel_sha_ctx),
+                       .cra_alignmask          = 0,
+                       .cra_module             = THIS_MODULE,
+                       .cra_init               = atmel_sha_cra_init,
+                       .cra_exit               = atmel_sha_cra_exit,
+               }
+       }
+},
+{
+       .init           = atmel_sha_init,
+       .update         = atmel_sha_update,
+       .final          = atmel_sha_final,
+       .finup          = atmel_sha_finup,
+       .digest         = atmel_sha_digest,
+       .halg = {
+               .digestsize     = SHA256_DIGEST_SIZE,
+               .base   = {
+                       .cra_name               = "sha256",
+                       .cra_driver_name        = "atmel-sha256",
+                       .cra_priority           = 100,
+                       .cra_flags              = CRYPTO_ALG_ASYNC |
+                                               CRYPTO_ALG_NEED_FALLBACK,
+                       .cra_blocksize          = SHA256_BLOCK_SIZE,
+                       .cra_ctxsize            = sizeof(struct atmel_sha_ctx),
+                       .cra_alignmask          = 0,
+                       .cra_module             = THIS_MODULE,
+                       .cra_init               = atmel_sha_cra_init,
+                       .cra_exit               = atmel_sha_cra_exit,
+               }
+       }
+},
+};
+
+static void atmel_sha_done_task(unsigned long data)
+{
+       struct atmel_sha_dev *dd = (struct atmel_sha_dev *)data;
+       int err = 0;
+
+       if (!(SHA_FLAGS_BUSY & dd->flags)) {
+               atmel_sha_handle_queue(dd, NULL);
+               return;
+       }
+
+       if (SHA_FLAGS_CPU & dd->flags) {
+               if (SHA_FLAGS_OUTPUT_READY & dd->flags) {
+                       dd->flags &= ~SHA_FLAGS_OUTPUT_READY;
+                       goto finish;
+               }
+       } else if (SHA_FLAGS_DMA_READY & dd->flags) {
+               if (SHA_FLAGS_DMA_ACTIVE & dd->flags) {
+                       dd->flags &= ~SHA_FLAGS_DMA_ACTIVE;
+                       atmel_sha_update_dma_stop(dd);
+                       if (dd->err) {
+                               err = dd->err;
+                               goto finish;
+                       }
+               }
+               if (SHA_FLAGS_OUTPUT_READY & dd->flags) {
+                       /* hash or semi-hash ready */
+                       dd->flags &= ~(SHA_FLAGS_DMA_READY |
+                                               SHA_FLAGS_OUTPUT_READY);
+                       err = atmel_sha_update_dma_start(dd);
+                       if (err != -EINPROGRESS)
+                               goto finish;
+               }
+       }
+       return;
+
+finish:
+       /* finish curent request */
+       atmel_sha_finish_req(dd->req, err);
+}
+
+static irqreturn_t atmel_sha_irq(int irq, void *dev_id)
+{
+       struct atmel_sha_dev *sha_dd = dev_id;
+       u32 reg;
+
+       reg = atmel_sha_read(sha_dd, SHA_ISR);
+       if (reg & atmel_sha_read(sha_dd, SHA_IMR)) {
+               atmel_sha_write(sha_dd, SHA_IDR, reg);
+               if (SHA_FLAGS_BUSY & sha_dd->flags) {
+                       sha_dd->flags |= SHA_FLAGS_OUTPUT_READY;
+                       if (!(SHA_FLAGS_CPU & sha_dd->flags))
+                               sha_dd->flags |= SHA_FLAGS_DMA_READY;
+                       tasklet_schedule(&sha_dd->done_task);
+               } else {
+                       dev_warn(sha_dd->dev, "SHA interrupt when no active requests.\n");
+               }
+               return IRQ_HANDLED;
+       }
+
+       return IRQ_NONE;
+}
+
+static void atmel_sha_unregister_algs(struct atmel_sha_dev *dd)
+{
+       int i;
+
+       for (i = 0; i < ARRAY_SIZE(sha_algs); i++)
+               crypto_unregister_ahash(&sha_algs[i]);
+}
+
+static int atmel_sha_register_algs(struct atmel_sha_dev *dd)
+{
+       int err, i, j;
+
+       for (i = 0; i < ARRAY_SIZE(sha_algs); i++) {
+               err = crypto_register_ahash(&sha_algs[i]);
+               if (err)
+                       goto err_sha_algs;
+       }
+
+       return 0;
+
+err_sha_algs:
+       for (j = 0; j < i; j++)
+               crypto_unregister_ahash(&sha_algs[j]);
+
+       return err;
+}
+
+static int __devinit atmel_sha_probe(struct platform_device *pdev)
+{
+       struct atmel_sha_dev *sha_dd;
+       struct device *dev = &pdev->dev;
+       struct resource *sha_res;
+       unsigned long sha_phys_size;
+       int err;
+
+       sha_dd = kzalloc(sizeof(struct atmel_sha_dev), GFP_KERNEL);
+       if (sha_dd == NULL) {
+               dev_err(dev, "unable to alloc data struct.\n");
+               err = -ENOMEM;
+               goto sha_dd_err;
+       }
+
+       sha_dd->dev = dev;
+
+       platform_set_drvdata(pdev, sha_dd);
+
+       INIT_LIST_HEAD(&sha_dd->list);
+
+       tasklet_init(&sha_dd->done_task, atmel_sha_done_task,
+                                       (unsigned long)sha_dd);
+
+       crypto_init_queue(&sha_dd->queue, ATMEL_SHA_QUEUE_LENGTH);
+
+       sha_dd->irq = -1;
+
+       /* Get the base address */
+       sha_res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!sha_res) {
+               dev_err(dev, "no MEM resource info\n");
+               err = -ENODEV;
+               goto res_err;
+       }
+       sha_dd->phys_base = sha_res->start;
+       sha_phys_size = resource_size(sha_res);
+
+       /* Get the IRQ */
+       sha_dd->irq = platform_get_irq(pdev,  0);
+       if (sha_dd->irq < 0) {
+               dev_err(dev, "no IRQ resource info\n");
+               err = sha_dd->irq;
+               goto res_err;
+       }
+
+       err = request_irq(sha_dd->irq, atmel_sha_irq, IRQF_SHARED, "atmel-sha",
+                                               sha_dd);
+       if (err) {
+               dev_err(dev, "unable to request sha irq.\n");
+               goto res_err;
+       }
+
+       /* Initializing the clock */
+       sha_dd->iclk = clk_get(&pdev->dev, NULL);
+       if (IS_ERR(sha_dd->iclk)) {
+               dev_err(dev, "clock intialization failed.\n");
+               err = PTR_ERR(sha_dd->iclk);
+               goto clk_err;
+       }
+
+       sha_dd->io_base = ioremap(sha_dd->phys_base, sha_phys_size);
+       if (!sha_dd->io_base) {
+               dev_err(dev, "can't ioremap\n");
+               err = -ENOMEM;
+               goto sha_io_err;
+       }
+
+       spin_lock(&atmel_sha.lock);
+       list_add_tail(&sha_dd->list, &atmel_sha.dev_list);
+       spin_unlock(&atmel_sha.lock);
+
+       err = atmel_sha_register_algs(sha_dd);
+       if (err)
+               goto err_algs;
+
+       dev_info(dev, "Atmel SHA1/SHA256\n");
+
+       return 0;
+
+err_algs:
+       spin_lock(&atmel_sha.lock);
+       list_del(&sha_dd->list);
+       spin_unlock(&atmel_sha.lock);
+       iounmap(sha_dd->io_base);
+sha_io_err:
+       clk_put(sha_dd->iclk);
+clk_err:
+       free_irq(sha_dd->irq, sha_dd);
+res_err:
+       tasklet_kill(&sha_dd->done_task);
+       kfree(sha_dd);
+       sha_dd = NULL;
+sha_dd_err:
+       dev_err(dev, "initialization failed.\n");
+
+       return err;
+}
+
+static int __devexit atmel_sha_remove(struct platform_device *pdev)
+{
+       static struct atmel_sha_dev *sha_dd;
+
+       sha_dd = platform_get_drvdata(pdev);
+       if (!sha_dd)
+               return -ENODEV;
+       spin_lock(&atmel_sha.lock);
+       list_del(&sha_dd->list);
+       spin_unlock(&atmel_sha.lock);
+
+       atmel_sha_unregister_algs(sha_dd);
+
+       tasklet_kill(&sha_dd->done_task);
+
+       iounmap(sha_dd->io_base);
+
+       clk_put(sha_dd->iclk);
+
+       if (sha_dd->irq >= 0)
+               free_irq(sha_dd->irq, sha_dd);
+
+       kfree(sha_dd);
+       sha_dd = NULL;
+
+       return 0;
+}
+
+static struct platform_driver atmel_sha_driver = {
+       .probe          = atmel_sha_probe,
+       .remove         = __devexit_p(atmel_sha_remove),
+       .driver         = {
+               .name   = "atmel_sha",
+               .owner  = THIS_MODULE,
+       },
+};
+
+module_platform_driver(atmel_sha_driver);
+
+MODULE_DESCRIPTION("Atmel SHA1/SHA256 hw acceleration support.");
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Nicolas Royer - Eukréa Electromatique");