crypto: caam - add support for xcbc(aes)

Add xcbc(aes) offloading support.

Due to xcbc algorithm design and HW implementation in CAAM,
driver must still have some bytes to send to the crypto engine when
ahash_final() is called - such that HW correctly uses either K2 or K3
for the last block.
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

drivers/crypto/caam/caamhash.c

@@ -98,13 +98,14 @@ struct caam_hash_ctx {
 	u32 sh_desc_update_first[DESC_HASH_MAX_USED_LEN] ____cacheline_aligned;
 	u32 sh_desc_fin[DESC_HASH_MAX_USED_LEN] ____cacheline_aligned;
 	u32 sh_desc_digest[DESC_HASH_MAX_USED_LEN] ____cacheline_aligned;
+	u8 key[CAAM_MAX_HASH_KEY_SIZE] ____cacheline_aligned;
 	dma_addr_t sh_desc_update_dma ____cacheline_aligned;
 	dma_addr_t sh_desc_update_first_dma;
 	dma_addr_t sh_desc_fin_dma;
 	dma_addr_t sh_desc_digest_dma;
+	dma_addr_t key_dma;
 	enum dma_data_direction dir;
 	struct device *jrdev;
-	u8 key[CAAM_MAX_HASH_KEY_SIZE];
 	int ctx_len;
 	struct alginfo adata;
 };
@@ -158,6 +159,12 @@ static inline int *alt_buflen(struct caam_hash_state *state)
 	return state->current_buf ? &state->buflen_0 : &state->buflen_1;
 }
 
+static inline bool is_xcbc_aes(u32 algtype)
+{
+	return (algtype & (OP_ALG_ALGSEL_MASK | OP_ALG_AAI_MASK)) ==
+	       (OP_ALG_ALGSEL_AES | OP_ALG_AAI_XCBC_MAC);
+}
+
 /* Common job descriptor seq in/out ptr routines */
 
 /* Map state->caam_ctx, and append seq_out_ptr command that points to it */
@@ -292,6 +299,62 @@ static int ahash_set_sh_desc(struct crypto_ahash *ahash)
 	return 0;
 }
 
+static int axcbc_set_sh_desc(struct crypto_ahash *ahash)
+{
+	struct caam_hash_ctx *ctx = crypto_ahash_ctx(ahash);
+	int digestsize = crypto_ahash_digestsize(ahash);
+	struct device *jrdev = ctx->jrdev;
+	u32 *desc;
+
+	/* key is loaded from memory for UPDATE and FINALIZE states */
+	ctx->adata.key_dma = ctx->key_dma;
+
+	/* shared descriptor for ahash_update */
+	desc = ctx->sh_desc_update;
+	cnstr_shdsc_axcbc(desc, &ctx->adata, OP_ALG_AS_UPDATE, ctx->ctx_len,
+			  ctx->ctx_len, 0);
+	dma_sync_single_for_device(jrdev, ctx->sh_desc_update_dma,
+				   desc_bytes(desc), ctx->dir);
+	print_hex_dump_debug("axcbc update shdesc@" __stringify(__LINE__)" : ",
+			     DUMP_PREFIX_ADDRESS, 16, 4, desc, desc_bytes(desc),
+			     1);
+
+	/* shared descriptor for ahash_{final,finup} */
+	desc = ctx->sh_desc_fin;
+	cnstr_shdsc_axcbc(desc, &ctx->adata, OP_ALG_AS_FINALIZE, digestsize,
+			  ctx->ctx_len, 0);
+	dma_sync_single_for_device(jrdev, ctx->sh_desc_fin_dma,
+				   desc_bytes(desc), ctx->dir);
+	print_hex_dump_debug("axcbc finup shdesc@" __stringify(__LINE__)" : ",
+			     DUMP_PREFIX_ADDRESS, 16, 4, desc, desc_bytes(desc),
+			     1);
+
+	/* key is immediate data for INIT and INITFINAL states */
+	ctx->adata.key_virt = ctx->key;
+
+	/* shared descriptor for first invocation of ahash_update */
+	desc = ctx->sh_desc_update_first;
+	cnstr_shdsc_axcbc(desc, &ctx->adata, OP_ALG_AS_INIT, ctx->ctx_len,
+			  ctx->ctx_len, ctx->key_dma);
+	dma_sync_single_for_device(jrdev, ctx->sh_desc_update_first_dma,
+				   desc_bytes(desc), ctx->dir);
+	print_hex_dump_debug("axcbc update first shdesc@" __stringify(__LINE__)" : ",
+			     DUMP_PREFIX_ADDRESS, 16, 4, desc, desc_bytes(desc),
+			     1);
+
+	/* shared descriptor for ahash_digest */
+	desc = ctx->sh_desc_digest;
+	cnstr_shdsc_axcbc(desc, &ctx->adata, OP_ALG_AS_INITFINAL, digestsize,
+			  ctx->ctx_len, 0);
+	dma_sync_single_for_device(jrdev, ctx->sh_desc_digest_dma,
+				   desc_bytes(desc), ctx->dir);
+	print_hex_dump_debug("axcbc digest shdesc@" __stringify(__LINE__)" : ",
+			     DUMP_PREFIX_ADDRESS, 16, 4, desc, desc_bytes(desc),
+			     1);
+
+	return 0;
+}
+
 /* Digest hash size if it is too large */
 static int hash_digest_key(struct caam_hash_ctx *ctx, const u8 *key_in,
 			   u32 *keylen, u8 *key_out, u32 digestsize)
@@ -424,6 +487,21 @@ static int ahash_setkey(struct crypto_ahash *ahash,
 	return -EINVAL;
 }
 
+static int axcbc_setkey(struct crypto_ahash *ahash, const u8 *key,
+			unsigned int keylen)
+{
+	struct caam_hash_ctx *ctx = crypto_ahash_ctx(ahash);
+	struct device *jrdev = ctx->jrdev;
+
+	memcpy(ctx->key, key, keylen);
+	dma_sync_single_for_device(jrdev, ctx->key_dma, keylen, DMA_TO_DEVICE);
+	ctx->adata.keylen = keylen;
+
+	print_hex_dump_debug("axcbc ctx.key@" __stringify(__LINE__)" : ",
+			     DUMP_PREFIX_ADDRESS, 16, 4, ctx->key, keylen, 1);
+
+	return axcbc_set_sh_desc(ahash);
+}
 /*
  * ahash_edesc - s/w-extended ahash descriptor
  * @dst_dma: physical mapped address of req->result
@@ -688,6 +766,7 @@ static int ahash_update_ctx(struct ahash_request *req)
 	u8 *buf = current_buf(state);
 	int *buflen = current_buflen(state);
 	u8 *next_buf = alt_buf(state);
+	int blocksize = crypto_ahash_blocksize(ahash);
 	int *next_buflen = alt_buflen(state), last_buflen;
 	int in_len = *buflen + req->nbytes, to_hash;
 	u32 *desc;
@@ -696,9 +775,19 @@ static int ahash_update_ctx(struct ahash_request *req)
 	int ret = 0;
 
 	last_buflen = *next_buflen;
-	*next_buflen = in_len & (crypto_tfm_alg_blocksize(&ahash->base) - 1);
+	*next_buflen = in_len & (blocksize - 1);
 	to_hash = in_len - *next_buflen;
 
+	/*
+	 * For XCBC, if to_hash is multiple of block size,
+	 * keep last block in internal buffer
+	 */
+	if (is_xcbc_aes(ctx->adata.algtype) && to_hash >= blocksize &&
+	    (*next_buflen == 0)) {
+		*next_buflen = blocksize;
+		to_hash -= blocksize;
+	}
+
 	if (to_hash) {
 		src_nents = sg_nents_for_len(req->src,
 					     req->nbytes - (*next_buflen));
@@ -1122,6 +1211,7 @@ static int ahash_update_no_ctx(struct ahash_request *req)
 		       GFP_KERNEL : GFP_ATOMIC;
 	u8 *buf = current_buf(state);
 	int *buflen = current_buflen(state);
+	int blocksize = crypto_ahash_blocksize(ahash);
 	u8 *next_buf = alt_buf(state);
 	int *next_buflen = alt_buflen(state);
 	int in_len = *buflen + req->nbytes, to_hash;
@@ -1130,9 +1220,19 @@ static int ahash_update_no_ctx(struct ahash_request *req)
 	u32 *desc;
 	int ret = 0;
 
-	*next_buflen = in_len & (crypto_tfm_alg_blocksize(&ahash->base) - 1);
+	*next_buflen = in_len & (blocksize - 1);
 	to_hash = in_len - *next_buflen;
 
+	/*
+	 * For XCBC, if to_hash is multiple of block size,
+	 * keep last block in internal buffer
+	 */
+	if (is_xcbc_aes(ctx->adata.algtype) && to_hash >= blocksize &&
+	    (*next_buflen == 0)) {
+		*next_buflen = blocksize;
+		to_hash -= blocksize;
+	}
+
 	if (to_hash) {
 		src_nents = sg_nents_for_len(req->src,
 					     req->nbytes - *next_buflen);
@@ -1338,15 +1438,25 @@ static int ahash_update_first(struct ahash_request *req)
 	u8 *next_buf = alt_buf(state);
 	int *next_buflen = alt_buflen(state);
 	int to_hash;
+	int blocksize = crypto_ahash_blocksize(ahash);
 	u32 *desc;
 	int src_nents, mapped_nents;
 	struct ahash_edesc *edesc;
 	int ret = 0;
 
-	*next_buflen = req->nbytes & (crypto_tfm_alg_blocksize(&ahash->base) -
-				      1);
+	*next_buflen = req->nbytes & (blocksize - 1);
 	to_hash = req->nbytes - *next_buflen;
 
+	/*
+	 * For XCBC, if to_hash is multiple of block size,
+	 * keep last block in internal buffer
+	 */
+	if (is_xcbc_aes(ctx->adata.algtype) && to_hash >= blocksize &&
+	    (*next_buflen == 0)) {
+		*next_buflen = blocksize;
+		to_hash -= blocksize;
+	}
+
 	if (to_hash) {
 		src_nents = sg_nents_for_len(req->src,
 					     req->nbytes - *next_buflen);
@@ -1654,6 +1764,25 @@ static struct caam_hash_template driver_hash[] = {
 			},
 		},
 		.alg_type = OP_ALG_ALGSEL_MD5,
+	}, {
+		.hmac_name = "xcbc(aes)",
+		.hmac_driver_name = "xcbc-aes-caam",
+		.blocksize = AES_BLOCK_SIZE,
+		.template_ahash = {
+			.init = ahash_init,
+			.update = ahash_update,
+			.final = ahash_final,
+			.finup = ahash_finup,
+			.digest = ahash_digest,
+			.export = ahash_export,
+			.import = ahash_import,
+			.setkey = axcbc_setkey,
+			.halg = {
+				.digestsize = AES_BLOCK_SIZE,
+				.statesize = sizeof(struct caam_export_state),
+			},
+		 },
+		.alg_type = OP_ALG_ALGSEL_AES | OP_ALG_AAI_XCBC_MAC,
 	},
 };
 
@@ -1695,7 +1824,28 @@ static int caam_hash_cra_init(struct crypto_tfm *tfm)
 	}
 
 	priv = dev_get_drvdata(ctx->jrdev->parent);
-	ctx->dir = priv->era >= 6 ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+
+	if (is_xcbc_aes(caam_hash->alg_type)) {
+		ctx->dir = DMA_TO_DEVICE;
+		ctx->adata.algtype = OP_TYPE_CLASS1_ALG | caam_hash->alg_type;
+		ctx->ctx_len = 48;
+
+		ctx->key_dma = dma_map_single_attrs(ctx->jrdev, ctx->key,
+						    ARRAY_SIZE(ctx->key),
+						    DMA_BIDIRECTIONAL,
+						    DMA_ATTR_SKIP_CPU_SYNC);
+		if (dma_mapping_error(ctx->jrdev, ctx->key_dma)) {
+			dev_err(ctx->jrdev, "unable to map key\n");
+			caam_jr_free(ctx->jrdev);
+			return -ENOMEM;
+		}
+	} else {
+		ctx->dir = priv->era >= 6 ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+		ctx->adata.algtype = OP_TYPE_CLASS2_ALG | caam_hash->alg_type;
+		ctx->ctx_len = runninglen[(ctx->adata.algtype &
+					   OP_ALG_ALGSEL_SUBMASK) >>
+					  OP_ALG_ALGSEL_SHIFT];
+	}
 
 	dma_addr = dma_map_single_attrs(ctx->jrdev, ctx->sh_desc_update,
 					offsetof(struct caam_hash_ctx,
@@ -1703,6 +1853,13 @@ static int caam_hash_cra_init(struct crypto_tfm *tfm)
 					ctx->dir, DMA_ATTR_SKIP_CPU_SYNC);
 	if (dma_mapping_error(ctx->jrdev, dma_addr)) {
 		dev_err(ctx->jrdev, "unable to map shared descriptors\n");
+
+		if (is_xcbc_aes(caam_hash->alg_type))
+			dma_unmap_single_attrs(ctx->jrdev, ctx->key_dma,
+					       ARRAY_SIZE(ctx->key),
+					       DMA_BIDIRECTIONAL,
+					       DMA_ATTR_SKIP_CPU_SYNC);
+
 		caam_jr_free(ctx->jrdev);
 		return -ENOMEM;
 	}
@@ -1716,13 +1873,6 @@ static int caam_hash_cra_init(struct crypto_tfm *tfm)
 	ctx->sh_desc_digest_dma = dma_addr + offsetof(struct caam_hash_ctx,
 						      sh_desc_digest);
 
-	/* copy descriptor header template value */
-	ctx->adata.algtype = OP_TYPE_CLASS2_ALG | caam_hash->alg_type;
-
-	ctx->ctx_len = runninglen[(ctx->adata.algtype &
-				   OP_ALG_ALGSEL_SUBMASK) >>
-				  OP_ALG_ALGSEL_SHIFT];
-
 	crypto_ahash_set_reqsize(__crypto_ahash_cast(tfm),
 				 sizeof(struct caam_hash_state));
 
@@ -1738,9 +1888,12 @@ static void caam_hash_cra_exit(struct crypto_tfm *tfm)
 	struct caam_hash_ctx *ctx = crypto_tfm_ctx(tfm);
 
 	dma_unmap_single_attrs(ctx->jrdev, ctx->sh_desc_update_dma,
-			       offsetof(struct caam_hash_ctx,
-					sh_desc_update_dma),
+			       offsetof(struct caam_hash_ctx, key),
 			       ctx->dir, DMA_ATTR_SKIP_CPU_SYNC);
+	if (is_xcbc_aes(ctx->adata.algtype))
+		dma_unmap_single_attrs(ctx->jrdev, ctx->key_dma,
+				       ARRAY_SIZE(ctx->key), DMA_BIDIRECTIONAL,
+				       DMA_ATTR_SKIP_CPU_SYNC);
 	caam_jr_free(ctx->jrdev);
 }
 
@@ -1871,7 +2024,8 @@ static int __init caam_algapi_hash_init(void)
 		struct caam_hash_template *alg = driver_hash + i;
 
 		/* If MD size is not supported by device, skip registration */
-		if (alg->template_ahash.halg.digestsize > md_limit)
+		if (is_mdha(alg->alg_type) &&
+		    alg->template_ahash.halg.digestsize > md_limit)
 			continue;
 
 		/* register hmac version */
@@ -1892,6 +2046,9 @@ static int __init caam_algapi_hash_init(void)
 		} else
 			list_add_tail(&t_alg->entry, &hash_list);
 
+		if ((alg->alg_type & OP_ALG_ALGSEL_MASK) == OP_ALG_ALGSEL_AES)
+			continue;
+
 		/* register unkeyed version */
 		t_alg = caam_hash_alloc(alg, false);
 		if (IS_ERR(t_alg)) {

drivers/crypto/caam/caamhash_desc.c

@@ -2,7 +2,7 @@
 /*
  * Shared descriptors for ahash algorithms
  *
- * Copyright 2017 NXP
+ * Copyright 2017-2018 NXP
  */
 
 #include "compat.h"
@@ -75,6 +75,62 @@ void cnstr_shdsc_ahash(u32 * const desc, struct alginfo *adata, u32 state,
 }
 EXPORT_SYMBOL(cnstr_shdsc_ahash);
 
+/**
+ * cnstr_shdsc_axcbc - axcbc shared descriptor
+ * @desc: pointer to buffer used for descriptor construction
+ * @adata: pointer to authentication transform definitions.
+ * @state: algorithm state OP_ALG_AS_{INIT, FINALIZE, INITFINALIZE, UPDATE}
+ * @digestsize: algorithm's digest size
+ * @ctx_len: size of Context Register
+ * @key_dma: I/O Virtual Address of the key
+ */
+void cnstr_shdsc_axcbc(u32 * const desc, struct alginfo *adata, u32 state,
+		       int digestsize, int ctx_len, dma_addr_t key_dma)
+{
+	u32 *skip_key_load;
+
+	init_sh_desc(desc, HDR_SHARE_SERIAL | HDR_SAVECTX);
+
+	/* Skip loading of key, context if already shared */
+	skip_key_load = append_jump(desc, JUMP_TEST_ALL | JUMP_COND_SHRD);
+
+	if (state == OP_ALG_AS_INIT || state == OP_ALG_AS_INITFINAL) {
+		append_key_as_imm(desc, adata->key_virt, adata->keylen,
+				  adata->keylen, CLASS_1 | KEY_DEST_CLASS_REG);
+	} else { /* UPDATE, FINALIZE */
+		/* Load K1 */
+		append_key(desc, adata->key_dma, adata->keylen,
+			   CLASS_1 | KEY_DEST_CLASS_REG | KEY_ENC);
+		/* Restore context */
+		append_seq_load(desc, ctx_len, LDST_CLASS_1_CCB |
+				LDST_SRCDST_BYTE_CONTEXT);
+	}
+
+	set_jump_tgt_here(desc, skip_key_load);
+
+	/* Class 1 operation */
+	append_operation(desc, adata->algtype | state | OP_ALG_ENCRYPT);
+
+	/*
+	 * Load from buf and/or src and write to req->result or state->context
+	 * Calculate remaining bytes to read
+	 */
+	append_math_add(desc, VARSEQINLEN, SEQINLEN, REG0, CAAM_CMD_SZ);
+
+	/* Read remaining bytes */
+	append_seq_fifo_load(desc, 0, FIFOLD_CLASS_CLASS1 | FIFOLD_TYPE_LAST1 |
+			     FIFOLD_TYPE_MSG | FIFOLDST_VLF);
+
+	/* Save context (partial hash, K2, K3) */
+	append_seq_store(desc, digestsize, LDST_CLASS_1_CCB |
+			 LDST_SRCDST_BYTE_CONTEXT);
+	if (state == OP_ALG_AS_INIT)
+		/* Save K1 */
+		append_fifo_store(desc, key_dma, adata->keylen,
+				  LDST_CLASS_1_CCB | FIFOST_TYPE_KEY_KEK);
+}
+EXPORT_SYMBOL(cnstr_shdsc_axcbc);
+
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_DESCRIPTION("FSL CAAM ahash descriptors support");
 MODULE_AUTHOR("NXP Semiconductors");

drivers/crypto/caam/caamhash_desc.h

@@ -18,4 +18,6 @@
 void cnstr_shdsc_ahash(u32 * const desc, struct alginfo *adata, u32 state,
 		       int digestsize, int ctx_len, bool import_ctx, int era);
 
+void cnstr_shdsc_axcbc(u32 * const desc, struct alginfo *adata, u32 state,
+		       int digestsize, int ctx_len, dma_addr_t key_dma);
 #endif /* _CAAMHASH_DESC_H_ */