DPDK patches and discussions
 help / color / mirror / Atom feed
From: Anoob Joseph <anoobj@marvell.com>
To: Akhil Goyal <gakhil@marvell.com>, Thomas Monjalon <thomas@monjalon.net>
Cc: Archana Muniganti <marchana@marvell.com>,
	Jerin Jacob <jerinj@marvell.com>,
	Ankur Dwivedi <adwivedi@marvell.com>,
	Tejasree Kondoj <ktejasree@marvell.com>, <dev@dpdk.org>,
	Anoob Joseph <anoobj@marvell.com>
Subject: [dpdk-dev] [PATCH 13/20] crypto/cnxk: add flexi crypto cipher encrypt
Date: Wed, 2 Jun 2021 22:13:34 +0530	[thread overview]
Message-ID: <1622652221-22732-14-git-send-email-anoobj@marvell.com> (raw)
In-Reply-To: <1622652221-22732-1-git-send-email-anoobj@marvell.com>

From: Archana Muniganti <marchana@marvell.com>

Add flexi crypto cipher encrypt in enqueue API. Flexi crypto
opcode covers a broad set of ciphers including variants of AES.

Signed-off-by: Ankur Dwivedi <adwivedi@marvell.com>
Signed-off-by: Anoob Joseph <anoobj@marvell.com>
Signed-off-by: Archana Muniganti <marchana@marvell.com>
Signed-off-by: Tejasree Kondoj <ktejasree@marvell.com>
---
 doc/guides/cryptodevs/features/cn10k.ini  |  16 +
 doc/guides/cryptodevs/features/cn9k.ini   |  20 +
 drivers/crypto/cnxk/cn10k_cryptodev.c     |   4 +
 drivers/crypto/cnxk/cn10k_cryptodev_ops.c |  14 +-
 drivers/crypto/cnxk/cn9k_cryptodev.c      |   4 +
 drivers/crypto/cnxk/cn9k_cryptodev_ops.c  |  14 +-
 drivers/crypto/cnxk/cnxk_se.h             | 815 ++++++++++++++++++++++++++++++
 7 files changed, 875 insertions(+), 12 deletions(-)

diff --git a/doc/guides/cryptodevs/features/cn10k.ini b/doc/guides/cryptodevs/features/cn10k.ini
index 175fbf7..f097d8e 100644
--- a/doc/guides/cryptodevs/features/cn10k.ini
+++ b/doc/guides/cryptodevs/features/cn10k.ini
@@ -7,6 +7,10 @@
 Symmetric crypto       = Y
 Sym operation chaining = Y
 HW Accelerated         = Y
+In Place SGL           = Y
+OOP SGL In LB  Out     = Y
+OOP SGL In SGL Out     = Y
+OOP LB  In LB  Out     = Y
 Symmetric sessionless  = Y
 Digest encrypted       = Y
 
@@ -14,6 +18,18 @@ Digest encrypted       = Y
 ; Supported crypto algorithms of 'cn10k' crypto driver.
 ;
 [Cipher]
+NULL           = Y
+3DES CBC       = Y
+3DES ECB       = Y
+AES CBC (128)  = Y
+AES CBC (192)  = Y
+AES CBC (256)  = Y
+AES CTR (128)  = Y
+AES CTR (192)  = Y
+AES CTR (256)  = Y
+AES XTS (128)  = Y
+AES XTS (256)  = Y
+DES CBC        = Y
 
 ;
 ; Supported authentication algorithms of 'cn10k' crypto driver.
diff --git a/doc/guides/cryptodevs/features/cn9k.ini b/doc/guides/cryptodevs/features/cn9k.ini
index c22b25c..7007d11 100644
--- a/doc/guides/cryptodevs/features/cn9k.ini
+++ b/doc/guides/cryptodevs/features/cn9k.ini
@@ -7,6 +7,10 @@
 Symmetric crypto       = Y
 Sym operation chaining = Y
 HW Accelerated         = Y
+In Place SGL           = Y
+OOP SGL In LB  Out     = Y
+OOP SGL In SGL Out     = Y
+OOP LB  In LB  Out     = Y
 Symmetric sessionless  = Y
 Digest encrypted       = Y
 
@@ -14,6 +18,18 @@ Digest encrypted       = Y
 ; Supported crypto algorithms of 'cn9k' crypto driver.
 ;
 [Cipher]
+NULL           = Y
+3DES CBC       = Y
+3DES ECB       = Y
+AES CBC (128)  = Y
+AES CBC (192)  = Y
+AES CBC (256)  = Y
+AES CTR (128)  = Y
+AES CTR (192)  = Y
+AES CTR (256)  = Y
+AES XTS (128)  = Y
+AES XTS (256)  = Y
+DES CBC        = Y
 
 ;
 ; Supported authentication algorithms of 'cn9k' crypto driver.
@@ -24,3 +40,7 @@ Digest encrypted       = Y
 ; Supported AEAD algorithms of 'cn9k' crypto driver.
 ;
 [AEAD]
+AES GCM (128)     = Y
+AES GCM (192)     = Y
+AES GCM (256)     = Y
+CHACHA20-POLY1305 = Y
diff --git a/drivers/crypto/cnxk/cn10k_cryptodev.c b/drivers/crypto/cnxk/cn10k_cryptodev.c
index 4d20409..ca3adea 100644
--- a/drivers/crypto/cnxk/cn10k_cryptodev.c
+++ b/drivers/crypto/cnxk/cn10k_cryptodev.c
@@ -82,6 +82,10 @@ cn10k_cpt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 	dev->feature_flags = RTE_CRYPTODEV_FF_SYMMETRIC_CRYPTO |
 			     RTE_CRYPTODEV_FF_HW_ACCELERATED |
 			     RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING |
+			     RTE_CRYPTODEV_FF_IN_PLACE_SGL |
+			     RTE_CRYPTODEV_FF_OOP_LB_IN_LB_OUT |
+			     RTE_CRYPTODEV_FF_OOP_SGL_IN_LB_OUT |
+			     RTE_CRYPTODEV_FF_OOP_SGL_IN_SGL_OUT |
 			     RTE_CRYPTODEV_FF_SYM_SESSIONLESS |
 			     RTE_CRYPTODEV_FF_DIGEST_ENCRYPTED;
 
diff --git a/drivers/crypto/cnxk/cn10k_cryptodev_ops.c b/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
index 83b24c9..b0faebc 100644
--- a/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
+++ b/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
@@ -46,13 +46,15 @@ cpt_sym_inst_fill(struct cnxk_cpt_qp *qp, struct rte_crypto_op *op,
 		  struct cnxk_se_sess *sess, struct cpt_inflight_req *infl_req,
 		  struct cpt_inst_s *inst)
 {
-	RTE_SET_USED(qp);
-	RTE_SET_USED(op);
-	RTE_SET_USED(sess);
-	RTE_SET_USED(infl_req);
-	RTE_SET_USED(inst);
+	uint64_t cpt_op;
+	int ret;
+
+	cpt_op = sess->cpt_op;
+
+	if (cpt_op & ROC_SE_OP_CIPHER_MASK)
+		ret = fill_fc_params(op, sess, &qp->meta_info, infl_req, inst);
 
-	return -ENOTSUP;
+	return ret;
 }
 
 static inline int
diff --git a/drivers/crypto/cnxk/cn9k_cryptodev.c b/drivers/crypto/cnxk/cn9k_cryptodev.c
index f629dac..ffa01a2 100644
--- a/drivers/crypto/cnxk/cn9k_cryptodev.c
+++ b/drivers/crypto/cnxk/cn9k_cryptodev.c
@@ -80,6 +80,10 @@ cn9k_cpt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 	dev->feature_flags = RTE_CRYPTODEV_FF_SYMMETRIC_CRYPTO |
 			     RTE_CRYPTODEV_FF_HW_ACCELERATED |
 			     RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING |
+			     RTE_CRYPTODEV_FF_IN_PLACE_SGL |
+			     RTE_CRYPTODEV_FF_OOP_LB_IN_LB_OUT |
+			     RTE_CRYPTODEV_FF_OOP_SGL_IN_LB_OUT |
+			     RTE_CRYPTODEV_FF_OOP_SGL_IN_SGL_OUT |
 			     RTE_CRYPTODEV_FF_SYM_SESSIONLESS |
 			     RTE_CRYPTODEV_FF_DIGEST_ENCRYPTED;
 
diff --git a/drivers/crypto/cnxk/cn9k_cryptodev_ops.c b/drivers/crypto/cnxk/cn9k_cryptodev_ops.c
index 41c411b..fed67c9 100644
--- a/drivers/crypto/cnxk/cn9k_cryptodev_ops.c
+++ b/drivers/crypto/cnxk/cn9k_cryptodev_ops.c
@@ -17,13 +17,15 @@ cn9k_cpt_sym_inst_fill(struct cnxk_cpt_qp *qp, struct rte_crypto_op *op,
 		       struct cpt_inflight_req *infl_req,
 		       struct cpt_inst_s *inst)
 {
-	RTE_SET_USED(qp);
-	RTE_SET_USED(op);
-	RTE_SET_USED(sess);
-	RTE_SET_USED(infl_req);
-	RTE_SET_USED(inst);
+	uint64_t cpt_op;
+	int ret;
+
+	cpt_op = sess->cpt_op;
+
+	if (cpt_op & ROC_SE_OP_CIPHER_MASK)
+		ret = fill_fc_params(op, sess, &qp->meta_info, infl_req, inst);
 
-	return -ENOTSUP;
+	return ret;
 }
 
 static inline struct cnxk_se_sess *
diff --git a/drivers/crypto/cnxk/cnxk_se.h b/drivers/crypto/cnxk/cnxk_se.h
index c522d2e..34ed75a 100644
--- a/drivers/crypto/cnxk/cnxk_se.h
+++ b/drivers/crypto/cnxk/cnxk_se.h
@@ -6,8 +6,18 @@
 #define _CNXK_SE_H_
 #include <stdbool.h>
 
+#include "cnxk_cryptodev.h"
+#include "cnxk_cryptodev_ops.h"
+
 #include "roc_se.h"
 
+#define SRC_IOV_SIZE                                                           \
+	(sizeof(struct roc_se_iov_ptr) +                                       \
+	 (sizeof(struct roc_se_buf_ptr) * ROC_SE_MAX_SG_CNT))
+#define DST_IOV_SIZE                                                           \
+	(sizeof(struct roc_se_iov_ptr) +                                       \
+	 (sizeof(struct roc_se_buf_ptr) * ROC_SE_MAX_SG_CNT))
+
 struct cnxk_se_sess {
 	uint16_t cpt_op : 4;
 	uint16_t zsk_flag : 4;
@@ -348,6 +358,453 @@ cpt_fc_ciph_set_key(struct roc_se_ctx *se_ctx, roc_se_cipher_type type,
 	return 0;
 }
 
+static __rte_always_inline uint32_t
+fill_sg_comp(struct roc_se_sglist_comp *list, uint32_t i, phys_addr_t dma_addr,
+	     uint32_t size)
+{
+	struct roc_se_sglist_comp *to = &list[i >> 2];
+
+	to->u.s.len[i % 4] = rte_cpu_to_be_16(size);
+	to->ptr[i % 4] = rte_cpu_to_be_64(dma_addr);
+	i++;
+	return i;
+}
+
+static __rte_always_inline uint32_t
+fill_sg_comp_from_buf(struct roc_se_sglist_comp *list, uint32_t i,
+		      struct roc_se_buf_ptr *from)
+{
+	struct roc_se_sglist_comp *to = &list[i >> 2];
+
+	to->u.s.len[i % 4] = rte_cpu_to_be_16(from->size);
+	to->ptr[i % 4] = rte_cpu_to_be_64((uint64_t)from->vaddr);
+	i++;
+	return i;
+}
+
+static __rte_always_inline uint32_t
+fill_sg_comp_from_buf_min(struct roc_se_sglist_comp *list, uint32_t i,
+			  struct roc_se_buf_ptr *from, uint32_t *psize)
+{
+	struct roc_se_sglist_comp *to = &list[i >> 2];
+	uint32_t size = *psize;
+	uint32_t e_len;
+
+	e_len = (size > from->size) ? from->size : size;
+	to->u.s.len[i % 4] = rte_cpu_to_be_16(e_len);
+	to->ptr[i % 4] = rte_cpu_to_be_64((uint64_t)from->vaddr);
+	*psize -= e_len;
+	i++;
+	return i;
+}
+
+/*
+ * This fills the MC expected SGIO list
+ * from IOV given by user.
+ */
+static __rte_always_inline uint32_t
+fill_sg_comp_from_iov(struct roc_se_sglist_comp *list, uint32_t i,
+		      struct roc_se_iov_ptr *from, uint32_t from_offset,
+		      uint32_t *psize, struct roc_se_buf_ptr *extra_buf,
+		      uint32_t extra_offset)
+{
+	int32_t j;
+	uint32_t extra_len = extra_buf ? extra_buf->size : 0;
+	uint32_t size = *psize;
+	struct roc_se_buf_ptr *bufs;
+
+	bufs = from->bufs;
+	for (j = 0; (j < from->buf_cnt) && size; j++) {
+		uint64_t e_vaddr;
+		uint32_t e_len;
+		struct roc_se_sglist_comp *to = &list[i >> 2];
+
+		if (unlikely(from_offset)) {
+			if (from_offset >= bufs[j].size) {
+				from_offset -= bufs[j].size;
+				continue;
+			}
+			e_vaddr = (uint64_t)bufs[j].vaddr + from_offset;
+			e_len = (size > (bufs[j].size - from_offset)) ?
+					(bufs[j].size - from_offset) :
+					size;
+			from_offset = 0;
+		} else {
+			e_vaddr = (uint64_t)bufs[j].vaddr;
+			e_len = (size > bufs[j].size) ? bufs[j].size : size;
+		}
+
+		to->u.s.len[i % 4] = rte_cpu_to_be_16(e_len);
+		to->ptr[i % 4] = rte_cpu_to_be_64(e_vaddr);
+
+		if (extra_len && (e_len >= extra_offset)) {
+			/* Break the data at given offset */
+			uint32_t next_len = e_len - extra_offset;
+			uint64_t next_vaddr = e_vaddr + extra_offset;
+
+			if (!extra_offset) {
+				i--;
+			} else {
+				e_len = extra_offset;
+				size -= e_len;
+				to->u.s.len[i % 4] = rte_cpu_to_be_16(e_len);
+			}
+
+			extra_len = RTE_MIN(extra_len, size);
+			/* Insert extra data ptr */
+			if (extra_len) {
+				i++;
+				to = &list[i >> 2];
+				to->u.s.len[i % 4] =
+					rte_cpu_to_be_16(extra_len);
+				to->ptr[i % 4] = rte_cpu_to_be_64(
+					(uint64_t)extra_buf->vaddr);
+				size -= extra_len;
+			}
+
+			next_len = RTE_MIN(next_len, size);
+			/* insert the rest of the data */
+			if (next_len) {
+				i++;
+				to = &list[i >> 2];
+				to->u.s.len[i % 4] = rte_cpu_to_be_16(next_len);
+				to->ptr[i % 4] = rte_cpu_to_be_64(next_vaddr);
+				size -= next_len;
+			}
+			extra_len = 0;
+
+		} else {
+			size -= e_len;
+		}
+		if (extra_offset)
+			extra_offset -= size;
+		i++;
+	}
+
+	*psize = size;
+	return (uint32_t)i;
+}
+
+static __rte_always_inline int
+cpt_enc_hmac_prep(uint32_t flags, uint64_t d_offs, uint64_t d_lens,
+		  struct roc_se_fc_params *fc_params, struct cpt_inst_s *inst)
+{
+	uint32_t iv_offset = 0;
+	int32_t inputlen, outputlen, enc_dlen, auth_dlen;
+	struct roc_se_ctx *se_ctx;
+	uint32_t cipher_type, hash_type;
+	uint32_t mac_len, size;
+	uint8_t iv_len = 16;
+	struct roc_se_buf_ptr *aad_buf = NULL;
+	uint32_t encr_offset, auth_offset;
+	uint32_t encr_data_len, auth_data_len, aad_len = 0;
+	uint32_t passthrough_len = 0;
+	union cpt_inst_w4 cpt_inst_w4;
+	void *offset_vaddr;
+	uint8_t op_minor;
+
+	encr_offset = ROC_SE_ENCR_OFFSET(d_offs);
+	auth_offset = ROC_SE_AUTH_OFFSET(d_offs);
+	encr_data_len = ROC_SE_ENCR_DLEN(d_lens);
+	auth_data_len = ROC_SE_AUTH_DLEN(d_lens);
+	if (unlikely(flags & ROC_SE_VALID_AAD_BUF)) {
+		/*
+		 * We dont support both aad
+		 * and auth data separately
+		 */
+		auth_data_len = 0;
+		auth_offset = 0;
+		aad_len = fc_params->aad_buf.size;
+		aad_buf = &fc_params->aad_buf;
+	}
+	se_ctx = fc_params->ctx_buf.vaddr;
+	cipher_type = se_ctx->enc_cipher;
+	hash_type = se_ctx->hash_type;
+	mac_len = se_ctx->mac_len;
+	op_minor = se_ctx->template_w4.s.opcode_minor;
+
+	if (unlikely(!(flags & ROC_SE_VALID_IV_BUF))) {
+		iv_len = 0;
+		iv_offset = ROC_SE_ENCR_IV_OFFSET(d_offs);
+	}
+
+	if (unlikely(flags & ROC_SE_VALID_AAD_BUF)) {
+		/*
+		 * When AAD is given, data above encr_offset is pass through
+		 * Since AAD is given as separate pointer and not as offset,
+		 * this is a special case as we need to fragment input data
+		 * into passthrough + encr_data and then insert AAD in between.
+		 */
+		if (hash_type != ROC_SE_GMAC_TYPE) {
+			passthrough_len = encr_offset;
+			auth_offset = passthrough_len + iv_len;
+			encr_offset = passthrough_len + aad_len + iv_len;
+			auth_data_len = aad_len + encr_data_len;
+		} else {
+			passthrough_len = 16 + aad_len;
+			auth_offset = passthrough_len + iv_len;
+			auth_data_len = aad_len;
+		}
+	} else {
+		encr_offset += iv_len;
+		auth_offset += iv_len;
+	}
+
+	/* Encryption */
+	cpt_inst_w4.s.opcode_major = ROC_SE_MAJOR_OP_FC;
+	cpt_inst_w4.s.opcode_minor = ROC_SE_FC_MINOR_OP_ENCRYPT;
+	cpt_inst_w4.s.opcode_minor |= (uint64_t)op_minor;
+
+	if (hash_type == ROC_SE_GMAC_TYPE) {
+		encr_offset = 0;
+		encr_data_len = 0;
+	}
+
+	auth_dlen = auth_offset + auth_data_len;
+	enc_dlen = encr_data_len + encr_offset;
+	if (unlikely(encr_data_len & 0xf)) {
+		if ((cipher_type == ROC_SE_DES3_CBC) ||
+		    (cipher_type == ROC_SE_DES3_ECB))
+			enc_dlen =
+				RTE_ALIGN_CEIL(encr_data_len, 8) + encr_offset;
+		else if (likely((cipher_type == ROC_SE_AES_CBC) ||
+				(cipher_type == ROC_SE_AES_ECB)))
+			enc_dlen =
+				RTE_ALIGN_CEIL(encr_data_len, 8) + encr_offset;
+	}
+
+	if (unlikely(auth_dlen > enc_dlen)) {
+		inputlen = auth_dlen;
+		outputlen = auth_dlen + mac_len;
+	} else {
+		inputlen = enc_dlen;
+		outputlen = enc_dlen + mac_len;
+	}
+
+	if (op_minor & ROC_SE_FC_MINOR_OP_HMAC_FIRST)
+		outputlen = enc_dlen;
+
+	/* GP op header */
+	cpt_inst_w4.s.param1 = encr_data_len;
+	cpt_inst_w4.s.param2 = auth_data_len;
+
+	/*
+	 * In cn9k, cn10k since we have a limitation of
+	 * IV & Offset control word not part of instruction
+	 * and need to be part of Data Buffer, we check if
+	 * head room is there and then only do the Direct mode processing
+	 */
+	if (likely((flags & ROC_SE_SINGLE_BUF_INPLACE) &&
+		   (flags & ROC_SE_SINGLE_BUF_HEADROOM))) {
+		void *dm_vaddr = fc_params->bufs[0].vaddr;
+
+		/* Use Direct mode */
+
+		offset_vaddr =
+			(uint8_t *)dm_vaddr - ROC_SE_OFF_CTRL_LEN - iv_len;
+
+		/* DPTR */
+		inst->dptr = (uint64_t)offset_vaddr;
+
+		/* RPTR should just exclude offset control word */
+		inst->rptr = (uint64_t)dm_vaddr - iv_len;
+
+		cpt_inst_w4.s.dlen = inputlen + ROC_SE_OFF_CTRL_LEN;
+
+		if (likely(iv_len)) {
+			uint64_t *dest = (uint64_t *)((uint8_t *)offset_vaddr +
+						      ROC_SE_OFF_CTRL_LEN);
+			uint64_t *src = fc_params->iv_buf;
+			dest[0] = src[0];
+			dest[1] = src[1];
+		}
+
+	} else {
+		void *m_vaddr = fc_params->meta_buf.vaddr;
+		uint32_t i, g_size_bytes, s_size_bytes;
+		struct roc_se_sglist_comp *gather_comp;
+		struct roc_se_sglist_comp *scatter_comp;
+		uint8_t *in_buffer;
+
+		/* This falls under strict SG mode */
+		offset_vaddr = m_vaddr;
+		size = ROC_SE_OFF_CTRL_LEN + iv_len;
+
+		m_vaddr = (uint8_t *)m_vaddr + size;
+
+		cpt_inst_w4.s.opcode_major |= (uint64_t)ROC_SE_DMA_MODE;
+
+		if (likely(iv_len)) {
+			uint64_t *dest = (uint64_t *)((uint8_t *)offset_vaddr +
+						      ROC_SE_OFF_CTRL_LEN);
+			uint64_t *src = fc_params->iv_buf;
+			dest[0] = src[0];
+			dest[1] = src[1];
+		}
+
+		/* DPTR has SG list */
+		in_buffer = m_vaddr;
+
+		((uint16_t *)in_buffer)[0] = 0;
+		((uint16_t *)in_buffer)[1] = 0;
+
+		/* TODO Add error check if space will be sufficient */
+		gather_comp =
+			(struct roc_se_sglist_comp *)((uint8_t *)m_vaddr + 8);
+
+		/*
+		 * Input Gather List
+		 */
+
+		i = 0;
+
+		/* Offset control word that includes iv */
+		i = fill_sg_comp(gather_comp, i, (uint64_t)offset_vaddr,
+				 ROC_SE_OFF_CTRL_LEN + iv_len);
+
+		/* Add input data */
+		size = inputlen - iv_len;
+		if (likely(size)) {
+			uint32_t aad_offset = aad_len ? passthrough_len : 0;
+
+			if (unlikely(flags & ROC_SE_SINGLE_BUF_INPLACE)) {
+				i = fill_sg_comp_from_buf_min(
+					gather_comp, i, fc_params->bufs, &size);
+			} else {
+				i = fill_sg_comp_from_iov(
+					gather_comp, i, fc_params->src_iov, 0,
+					&size, aad_buf, aad_offset);
+			}
+
+			if (unlikely(size)) {
+				CPT_LOG_DP_ERR("Insufficient buffer space,"
+					       " size %d needed",
+					       size);
+				return -1;
+			}
+		}
+		((uint16_t *)in_buffer)[2] = rte_cpu_to_be_16(i);
+		g_size_bytes =
+			((i + 3) / 4) * sizeof(struct roc_se_sglist_comp);
+
+		/*
+		 * Output Scatter list
+		 */
+		i = 0;
+		scatter_comp =
+			(struct roc_se_sglist_comp *)((uint8_t *)gather_comp +
+						      g_size_bytes);
+
+		/* Add IV */
+		if (likely(iv_len)) {
+			i = fill_sg_comp(scatter_comp, i,
+					 (uint64_t)offset_vaddr +
+						 ROC_SE_OFF_CTRL_LEN,
+					 iv_len);
+		}
+
+		/* output data or output data + digest*/
+		if (unlikely(flags & ROC_SE_VALID_MAC_BUF)) {
+			size = outputlen - iv_len - mac_len;
+			if (size) {
+				uint32_t aad_offset =
+					aad_len ? passthrough_len : 0;
+
+				if (unlikely(flags &
+					     ROC_SE_SINGLE_BUF_INPLACE)) {
+					i = fill_sg_comp_from_buf_min(
+						scatter_comp, i,
+						fc_params->bufs, &size);
+				} else {
+					i = fill_sg_comp_from_iov(
+						scatter_comp, i,
+						fc_params->dst_iov, 0, &size,
+						aad_buf, aad_offset);
+				}
+				if (unlikely(size)) {
+					CPT_LOG_DP_ERR("Insufficient buffer"
+						       " space, size %d needed",
+						       size);
+					return -1;
+				}
+			}
+			/* mac_data */
+			if (mac_len) {
+				i = fill_sg_comp_from_buf(scatter_comp, i,
+							  &fc_params->mac_buf);
+			}
+		} else {
+			/* Output including mac */
+			size = outputlen - iv_len;
+			if (likely(size)) {
+				uint32_t aad_offset =
+					aad_len ? passthrough_len : 0;
+
+				if (unlikely(flags &
+					     ROC_SE_SINGLE_BUF_INPLACE)) {
+					i = fill_sg_comp_from_buf_min(
+						scatter_comp, i,
+						fc_params->bufs, &size);
+				} else {
+					i = fill_sg_comp_from_iov(
+						scatter_comp, i,
+						fc_params->dst_iov, 0, &size,
+						aad_buf, aad_offset);
+				}
+				if (unlikely(size)) {
+					CPT_LOG_DP_ERR("Insufficient buffer"
+						       " space, size %d needed",
+						       size);
+					return -1;
+				}
+			}
+		}
+		((uint16_t *)in_buffer)[3] = rte_cpu_to_be_16(i);
+		s_size_bytes =
+			((i + 3) / 4) * sizeof(struct roc_se_sglist_comp);
+
+		size = g_size_bytes + s_size_bytes + ROC_SE_SG_LIST_HDR_SIZE;
+
+		/* This is DPTR len in case of SG mode */
+		cpt_inst_w4.s.dlen = size;
+
+		inst->dptr = (uint64_t)in_buffer;
+	}
+
+	if (unlikely((encr_offset >> 16) || (iv_offset >> 8) ||
+		     (auth_offset >> 8))) {
+		CPT_LOG_DP_ERR("Offset not supported");
+		CPT_LOG_DP_ERR("enc_offset: %d", encr_offset);
+		CPT_LOG_DP_ERR("iv_offset : %d", iv_offset);
+		CPT_LOG_DP_ERR("auth_offset: %d", auth_offset);
+		return -1;
+	}
+
+	*(uint64_t *)offset_vaddr = rte_cpu_to_be_64(
+		((uint64_t)encr_offset << 16) | ((uint64_t)iv_offset << 8) |
+		((uint64_t)auth_offset));
+
+	inst->w4.u64 = cpt_inst_w4.u64;
+	return 0;
+}
+
+static __rte_always_inline int
+cpt_fc_enc_hmac_prep(uint32_t flags, uint64_t d_offs, uint64_t d_lens,
+		     struct roc_se_fc_params *fc_params,
+		     struct cpt_inst_s *inst)
+{
+	struct roc_se_ctx *ctx = fc_params->ctx_buf.vaddr;
+	uint8_t fc_type;
+	int ret = -1;
+
+	fc_type = ctx->fc_type;
+
+	if (likely(fc_type == ROC_SE_FC_GEN))
+		ret = cpt_enc_hmac_prep(flags, d_offs, d_lens, fc_params, inst);
+
+	return ret;
+}
+
 static __rte_always_inline int
 cpt_fc_auth_set_key(struct roc_se_ctx *se_ctx, roc_se_auth_type type,
 		    const uint8_t *key, uint16_t key_len, uint16_t mac_len)
@@ -762,4 +1219,362 @@ fill_sess_gmac(struct rte_crypto_sym_xform *xform, struct cnxk_se_sess *sess)
 	return 0;
 }
 
+static __rte_always_inline void *
+alloc_op_meta(struct roc_se_buf_ptr *buf, int32_t len,
+	      struct rte_mempool *cpt_meta_pool,
+	      struct cpt_inflight_req *infl_req)
+{
+	uint8_t *mdata;
+
+	if (unlikely(rte_mempool_get(cpt_meta_pool, (void **)&mdata) < 0))
+		return NULL;
+
+	buf->vaddr = mdata;
+	buf->size = len;
+
+	infl_req->mdata = mdata;
+	infl_req->op_flags |= CPT_OP_FLAGS_METABUF;
+
+	return mdata;
+}
+
+static __rte_always_inline uint32_t
+prepare_iov_from_pkt(struct rte_mbuf *pkt, struct roc_se_iov_ptr *iovec,
+		     uint32_t start_offset)
+{
+	uint16_t index = 0;
+	void *seg_data = NULL;
+	int32_t seg_size = 0;
+
+	if (!pkt) {
+		iovec->buf_cnt = 0;
+		return 0;
+	}
+
+	if (!start_offset) {
+		seg_data = rte_pktmbuf_mtod(pkt, void *);
+		seg_size = pkt->data_len;
+	} else {
+		while (start_offset >= pkt->data_len) {
+			start_offset -= pkt->data_len;
+			pkt = pkt->next;
+		}
+
+		seg_data = rte_pktmbuf_mtod_offset(pkt, void *, start_offset);
+		seg_size = pkt->data_len - start_offset;
+		if (!seg_size)
+			return 1;
+	}
+
+	/* first seg */
+	iovec->bufs[index].vaddr = seg_data;
+	iovec->bufs[index].size = seg_size;
+	index++;
+	pkt = pkt->next;
+
+	while (unlikely(pkt != NULL)) {
+		seg_data = rte_pktmbuf_mtod(pkt, void *);
+		seg_size = pkt->data_len;
+		if (!seg_size)
+			break;
+
+		iovec->bufs[index].vaddr = seg_data;
+		iovec->bufs[index].size = seg_size;
+
+		index++;
+
+		pkt = pkt->next;
+	}
+
+	iovec->buf_cnt = index;
+	return 0;
+}
+
+static __rte_always_inline uint32_t
+prepare_iov_from_pkt_inplace(struct rte_mbuf *pkt,
+			     struct roc_se_fc_params *param, uint32_t *flags)
+{
+	uint16_t index = 0;
+	void *seg_data = NULL;
+	uint32_t seg_size = 0;
+	struct roc_se_iov_ptr *iovec;
+
+	seg_data = rte_pktmbuf_mtod(pkt, void *);
+	seg_size = pkt->data_len;
+
+	/* first seg */
+	if (likely(!pkt->next)) {
+		uint32_t headroom;
+
+		*flags |= ROC_SE_SINGLE_BUF_INPLACE;
+		headroom = rte_pktmbuf_headroom(pkt);
+		if (likely(headroom >= 24))
+			*flags |= ROC_SE_SINGLE_BUF_HEADROOM;
+
+		param->bufs[0].vaddr = seg_data;
+		param->bufs[0].size = seg_size;
+		return 0;
+	}
+	iovec = param->src_iov;
+	iovec->bufs[index].vaddr = seg_data;
+	iovec->bufs[index].size = seg_size;
+	index++;
+	pkt = pkt->next;
+
+	while (unlikely(pkt != NULL)) {
+		seg_data = rte_pktmbuf_mtod(pkt, void *);
+		seg_size = pkt->data_len;
+
+		if (!seg_size)
+			break;
+
+		iovec->bufs[index].vaddr = seg_data;
+		iovec->bufs[index].size = seg_size;
+
+		index++;
+
+		pkt = pkt->next;
+	}
+
+	iovec->buf_cnt = index;
+	return 0;
+}
+
+static __rte_always_inline int
+fill_fc_params(struct rte_crypto_op *cop, struct cnxk_se_sess *sess,
+	       struct cpt_qp_meta_info *m_info,
+	       struct cpt_inflight_req *infl_req, struct cpt_inst_s *inst)
+{
+	struct roc_se_ctx *ctx = &sess->roc_se_ctx;
+	uint8_t op_minor = ctx->template_w4.s.opcode_minor;
+	struct rte_crypto_sym_op *sym_op = cop->sym;
+	void *mdata = NULL;
+	uint32_t mc_hash_off;
+	uint32_t flags = 0;
+	uint64_t d_offs, d_lens;
+	struct rte_mbuf *m_src, *m_dst;
+	uint8_t cpt_op = sess->cpt_op;
+#ifdef CPT_ALWAYS_USE_SG_MODE
+	uint8_t inplace = 0;
+#else
+	uint8_t inplace = 1;
+#endif
+	struct roc_se_fc_params fc_params;
+	char src[SRC_IOV_SIZE];
+	char dst[SRC_IOV_SIZE];
+	uint32_t iv_buf[4];
+	int ret;
+
+	if (likely(sess->iv_length)) {
+		flags |= ROC_SE_VALID_IV_BUF;
+		fc_params.iv_buf = rte_crypto_op_ctod_offset(cop, uint8_t *,
+							     sess->iv_offset);
+		if (sess->aes_ctr && unlikely(sess->iv_length != 16)) {
+			memcpy((uint8_t *)iv_buf,
+			       rte_crypto_op_ctod_offset(cop, uint8_t *,
+							 sess->iv_offset),
+			       12);
+			iv_buf[3] = rte_cpu_to_be_32(0x1);
+			fc_params.iv_buf = iv_buf;
+		}
+	}
+
+	if (sess->zsk_flag) {
+		fc_params.auth_iv_buf = rte_crypto_op_ctod_offset(
+			cop, uint8_t *, sess->auth_iv_offset);
+		if (sess->zsk_flag != ROC_SE_ZS_EA)
+			inplace = 0;
+	}
+	m_src = sym_op->m_src;
+	m_dst = sym_op->m_dst;
+
+	if (sess->aes_gcm || sess->chacha_poly) {
+		uint8_t *salt;
+		uint8_t *aad_data;
+		uint16_t aad_len;
+
+		d_offs = sym_op->aead.data.offset;
+		d_lens = sym_op->aead.data.length;
+		mc_hash_off =
+			sym_op->aead.data.offset + sym_op->aead.data.length;
+
+		aad_data = sym_op->aead.aad.data;
+		aad_len = sess->aad_length;
+		if (likely((aad_data + aad_len) ==
+			   rte_pktmbuf_mtod_offset(m_src, uint8_t *,
+						   sym_op->aead.data.offset))) {
+			d_offs = (d_offs - aad_len) | (d_offs << 16);
+			d_lens = (d_lens + aad_len) | (d_lens << 32);
+		} else {
+			fc_params.aad_buf.vaddr = sym_op->aead.aad.data;
+			fc_params.aad_buf.size = aad_len;
+			flags |= ROC_SE_VALID_AAD_BUF;
+			inplace = 0;
+			d_offs = d_offs << 16;
+			d_lens = d_lens << 32;
+		}
+
+		salt = fc_params.iv_buf;
+		if (unlikely(*(uint32_t *)salt != sess->salt)) {
+			cpt_fc_salt_update(&sess->roc_se_ctx, salt);
+			sess->salt = *(uint32_t *)salt;
+		}
+		fc_params.iv_buf = salt + 4;
+		if (likely(sess->mac_len)) {
+			struct rte_mbuf *m =
+				(cpt_op & ROC_SE_OP_ENCODE) ? m_dst : m_src;
+
+			if (!m)
+				m = m_src;
+
+			/* hmac immediately following data is best case */
+			if (unlikely(rte_pktmbuf_mtod(m, uint8_t *) +
+					     mc_hash_off !=
+				     (uint8_t *)sym_op->aead.digest.data)) {
+				flags |= ROC_SE_VALID_MAC_BUF;
+				fc_params.mac_buf.size = sess->mac_len;
+				fc_params.mac_buf.vaddr =
+					sym_op->aead.digest.data;
+				inplace = 0;
+			}
+		}
+	} else {
+		d_offs = sym_op->cipher.data.offset;
+		d_lens = sym_op->cipher.data.length;
+		mc_hash_off =
+			sym_op->cipher.data.offset + sym_op->cipher.data.length;
+		d_offs = (d_offs << 16) | sym_op->auth.data.offset;
+		d_lens = (d_lens << 32) | sym_op->auth.data.length;
+
+		if (mc_hash_off <
+		    (sym_op->auth.data.offset + sym_op->auth.data.length)) {
+			mc_hash_off = (sym_op->auth.data.offset +
+				       sym_op->auth.data.length);
+		}
+		/* for gmac, salt should be updated like in gcm */
+		if (unlikely(sess->is_gmac)) {
+			uint8_t *salt;
+			salt = fc_params.iv_buf;
+			if (unlikely(*(uint32_t *)salt != sess->salt)) {
+				cpt_fc_salt_update(&sess->roc_se_ctx, salt);
+				sess->salt = *(uint32_t *)salt;
+			}
+			fc_params.iv_buf = salt + 4;
+		}
+		if (likely(sess->mac_len)) {
+			struct rte_mbuf *m;
+
+			m = (cpt_op & ROC_SE_OP_ENCODE) ? m_dst : m_src;
+			if (!m)
+				m = m_src;
+
+			/* hmac immediately following data is best case */
+			if (!(op_minor & ROC_SE_FC_MINOR_OP_HMAC_FIRST) &&
+			    (unlikely(rte_pktmbuf_mtod(m, uint8_t *) +
+					      mc_hash_off !=
+				      (uint8_t *)sym_op->auth.digest.data))) {
+				flags |= ROC_SE_VALID_MAC_BUF;
+				fc_params.mac_buf.size = sess->mac_len;
+				fc_params.mac_buf.vaddr =
+					sym_op->auth.digest.data;
+				inplace = 0;
+			}
+		}
+	}
+	fc_params.ctx_buf.vaddr = &sess->roc_se_ctx;
+
+	if (!(op_minor & ROC_SE_FC_MINOR_OP_HMAC_FIRST) &&
+	    unlikely(sess->is_null || sess->cpt_op == ROC_SE_OP_DECODE))
+		inplace = 0;
+
+	if (likely(!m_dst && inplace)) {
+		/* Case of single buffer without AAD buf or
+		 * separate mac buf in place and
+		 * not air crypto
+		 */
+		fc_params.dst_iov = fc_params.src_iov = (void *)src;
+
+		if (unlikely(prepare_iov_from_pkt_inplace(m_src, &fc_params,
+							  &flags))) {
+			CPT_LOG_DP_ERR("Prepare inplace src iov failed");
+			ret = -EINVAL;
+			goto err_exit;
+		}
+
+	} else {
+		/* Out of place processing */
+		fc_params.src_iov = (void *)src;
+		fc_params.dst_iov = (void *)dst;
+
+		/* Store SG I/O in the api for reuse */
+		if (prepare_iov_from_pkt(m_src, fc_params.src_iov, 0)) {
+			CPT_LOG_DP_ERR("Prepare src iov failed");
+			ret = -EINVAL;
+			goto err_exit;
+		}
+
+		if (unlikely(m_dst != NULL)) {
+			uint32_t pkt_len;
+
+			/* Try to make room as much as src has */
+			pkt_len = rte_pktmbuf_pkt_len(m_dst);
+
+			if (unlikely(pkt_len < rte_pktmbuf_pkt_len(m_src))) {
+				pkt_len = rte_pktmbuf_pkt_len(m_src) - pkt_len;
+				if (!rte_pktmbuf_append(m_dst, pkt_len)) {
+					CPT_LOG_DP_ERR("Not enough space in "
+						       "m_dst %p, need %u"
+						       " more",
+						       m_dst, pkt_len);
+					ret = -EINVAL;
+					goto err_exit;
+				}
+			}
+
+			if (prepare_iov_from_pkt(m_dst, fc_params.dst_iov, 0)) {
+				CPT_LOG_DP_ERR("Prepare dst iov failed for "
+					       "m_dst %p",
+					       m_dst);
+				ret = -EINVAL;
+				goto err_exit;
+			}
+		} else {
+			fc_params.dst_iov = (void *)src;
+		}
+	}
+
+	if (unlikely(!((flags & ROC_SE_SINGLE_BUF_INPLACE) &&
+		       (flags & ROC_SE_SINGLE_BUF_HEADROOM) &&
+		       ((ctx->fc_type == ROC_SE_FC_GEN) ||
+			(ctx->fc_type == ROC_SE_PDCP))))) {
+		mdata = alloc_op_meta(&fc_params.meta_buf, m_info->mlen,
+				      m_info->pool, infl_req);
+		if (mdata == NULL) {
+			CPT_LOG_DP_ERR(
+				"Error allocating meta buffer for request");
+			return -ENOMEM;
+		}
+	}
+
+	/* Finally prepare the instruction */
+	if (cpt_op & ROC_SE_OP_ENCODE)
+		ret = cpt_fc_enc_hmac_prep(flags, d_offs, d_lens, &fc_params,
+					   inst);
+	else
+		ret = ENOTSUP;
+
+	if (unlikely(ret)) {
+		CPT_LOG_DP_ERR("Preparing request failed due to bad input arg");
+		goto free_mdata_and_exit;
+	}
+
+	return 0;
+
+free_mdata_and_exit:
+	if (infl_req->op_flags & CPT_OP_FLAGS_METABUF)
+		rte_mempool_put(m_info->pool, infl_req->mdata);
+err_exit:
+	return ret;
+}
+
 #endif /*_CNXK_SE_H_ */
-- 
2.7.4


  parent reply	other threads:[~2021-06-02 16:46 UTC|newest]

Thread overview: 34+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-06-02 16:43 [dpdk-dev] [PATCH 00/20] Add Marvell CNXK crypto PMDs Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 01/20] crypto/cnxk: add driver skeleton Anoob Joseph
2021-06-16  7:28   ` Akhil Goyal
2021-06-16  7:37     ` Anoob Joseph
2021-06-16  7:47       ` Akhil Goyal
2021-06-16 19:58   ` Akhil Goyal
2021-06-02 16:43 ` [dpdk-dev] [PATCH 02/20] crypto/cnxk: add probe and remove Anoob Joseph
2021-06-16 10:51   ` Akhil Goyal
2021-06-02 16:43 ` [dpdk-dev] [PATCH 03/20] crypto/cnxk: add device control ops Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 04/20] crypto/cnxk: add symmetric crypto capabilities Anoob Joseph
2021-06-16  9:47   ` Akhil Goyal
2021-06-02 16:43 ` [dpdk-dev] [PATCH 05/20] crypto/cnxk: add queue pair ops Anoob Joseph
2021-06-16 11:05   ` Akhil Goyal
2021-06-17  7:13     ` Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 06/20] crypto/cnxk: add session ops framework Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 07/20] crypto/cnxk: add enqueue burst op Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 08/20] crypto/cnxk: add dequeue " Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 09/20] crypto/cnxk: add cipher operation in session Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 10/20] crypto/cnxk: add auth " Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 11/20] crypto/cnxk: add aead " Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 12/20] crypto/cnxk: add chained " Anoob Joseph
2021-06-02 16:43 ` Anoob Joseph [this message]
2021-06-16 19:45   ` [dpdk-dev] [PATCH 13/20] crypto/cnxk: add flexi crypto cipher encrypt Akhil Goyal
2021-06-02 16:43 ` [dpdk-dev] [PATCH 14/20] crypto/cnxk: add flexi crypto cipher decrypt Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 15/20] crypto/cnxk: add ZUC and SNOW3G encrypt Anoob Joseph
2021-06-16 19:51   ` Akhil Goyal
2021-06-02 16:43 ` [dpdk-dev] [PATCH 16/20] crypto/cnxk: add ZUC and SNOW3G decrypt Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 17/20] crypto/cnxk: add KASUMI encrypt Anoob Joseph
2021-06-16 19:51   ` Akhil Goyal
2021-06-02 16:43 ` [dpdk-dev] [PATCH 18/20] crypto/cnxk: add KASUMI decrypt Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 19/20] crypto/cnxk: add digest support Anoob Joseph
2021-06-02 16:43 ` [dpdk-dev] [PATCH 20/20] test/crypto: enable cnxk crypto PMDs Anoob Joseph
2021-06-16  7:23 ` [dpdk-dev] [PATCH 00/20] Add Marvell CNXK " Akhil Goyal
2021-06-16 19:56 ` Akhil Goyal

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1622652221-22732-14-git-send-email-anoobj@marvell.com \
    --to=anoobj@marvell.com \
    --cc=adwivedi@marvell.com \
    --cc=dev@dpdk.org \
    --cc=gakhil@marvell.com \
    --cc=jerinj@marvell.com \
    --cc=ktejasree@marvell.com \
    --cc=marchana@marvell.com \
    --cc=thomas@monjalon.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).