[dpdk-dev] [PATCH v3 8/8] crypto/octeontx2: add cryptodev sec enqueue and dequeue routines

[Tejasree Kondoj <ktejasree@marvell.com>]

From: Vamsi Attunuru <vattunuru@marvell.com>

This patch adds lookaside IPsec enqueue and dequeue routines.

Signed-off-by: Vamsi Attunuru <vattunuru@marvell.com>
Signed-off-by: Tejasree Kondoj <ktejasree@marvell.com>
---
 doc/guides/cryptodevs/octeontx2.rst           |  21 +++
 doc/guides/rel_notes/release_20_08.rst        |   5 +
 drivers/crypto/octeontx2/otx2_cryptodev.h     |   8 +
 drivers/crypto/octeontx2/otx2_cryptodev_ops.c |  73 +++++++-
 drivers/crypto/octeontx2/otx2_ipsec_po.h      |   8 +
 drivers/crypto/octeontx2/otx2_ipsec_po_ops.h  | 175 ++++++++++++++++++
 6 files changed, 289 insertions(+), 1 deletion(-)
 create mode 100644 drivers/crypto/octeontx2/otx2_ipsec_po_ops.h

diff --git a/doc/guides/cryptodevs/octeontx2.rst b/doc/guides/cryptodevs/octeontx2.rst
index 085d669e49..432146db04 100644
--- a/doc/guides/cryptodevs/octeontx2.rst
+++ b/doc/guides/cryptodevs/octeontx2.rst
@@ -158,3 +158,24 @@ application:
 
     ./test
     RTE>>cryptodev_octeontx2_asym_autotest
+
+
+Lookaside IPsec Support
+-----------------------
+
+The OCTEON TX2 SoC can accelerate IPsec traffic in lookaside protocol mode,
+with its **cryptographic accelerator (CPT)**. ``OCTEON TX2 crypto PMD`` implements
+this as an ``RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL`` offload.
+
+Refer to :doc:`../prog_guide/rte_security` for more details on protocol offloads.
+
+This feature can be tested with ipsec-secgw sample application.
+
+
+Features supported
+~~~~~~~~~~~~~~~~~~
+
+* IPv4
+* ESP
+* Tunnel mode
+* AES-128/192/256-GCM
diff --git a/doc/guides/rel_notes/release_20_08.rst b/doc/guides/rel_notes/release_20_08.rst
index f19b748728..53cd13455f 100644
--- a/doc/guides/rel_notes/release_20_08.rst
+++ b/doc/guides/rel_notes/release_20_08.rst
@@ -196,6 +196,11 @@ New Features
 
   Added Chacha20-Poly1305 AEAD algorithm support in OCTEON TX2 crypto PMD.
 
+* **Updated the OCTEON TX2 crypto PMD to support rte_security.**
+
+  Updated the OCTEON TX2 crypto PMD to support ``rte_security`` lookaside
+  protocol offload for IPsec.
+
 * **Added support for BPF_ABS/BPF_IND load instructions.**
 
   Added support for two BPF non-generic instructions:
diff --git a/drivers/crypto/octeontx2/otx2_cryptodev.h b/drivers/crypto/octeontx2/otx2_cryptodev.h
index e7a1730b22..f329741b38 100644
--- a/drivers/crypto/octeontx2/otx2_cryptodev.h
+++ b/drivers/crypto/octeontx2/otx2_cryptodev.h
@@ -6,6 +6,7 @@
 #define _OTX2_CRYPTODEV_H_
 
 #include "cpt_common.h"
+#include "cpt_hw_types.h"
 
 #include "otx2_dev.h"
 
@@ -33,6 +34,13 @@ struct otx2_cpt_vf {
 	/**< CPT device capabilities */
 };
 
+struct cpt_meta_info {
+	uint64_t deq_op_info[4];
+	uint64_t comp_code_sz;
+	union cpt_res_s cpt_res __rte_aligned(16);
+	struct cpt_request_info cpt_req __rte_aligned(8);
+};
+
 #define CPT_LOGTYPE otx2_cpt_logtype
 
 extern int otx2_cpt_logtype;
diff --git a/drivers/crypto/octeontx2/otx2_cryptodev_ops.c b/drivers/crypto/octeontx2/otx2_cryptodev_ops.c
index 229b719b42..9d51b17ddd 100644
--- a/drivers/crypto/octeontx2/otx2_cryptodev_ops.c
+++ b/drivers/crypto/octeontx2/otx2_cryptodev_ops.c
@@ -13,8 +13,10 @@
 #include "otx2_cryptodev_hw_access.h"
 #include "otx2_cryptodev_mbox.h"
 #include "otx2_cryptodev_ops.h"
+#include "otx2_ipsec_po_ops.h"
 #include "otx2_mbox.h"
 #include "otx2_sec_idev.h"
+#include "otx2_security.h"
 
 #include "cpt_hw_types.h"
 #include "cpt_pmd_logs.h"
@@ -606,6 +608,36 @@ otx2_cpt_enqueue_sym(struct otx2_cpt_qp *qp, struct rte_crypto_op *op,
 	return ret;
 }
 
+static __rte_always_inline int __rte_hot
+otx2_cpt_enqueue_sec(struct otx2_cpt_qp *qp, struct rte_crypto_op *op,
+		     struct pending_queue *pend_q)
+{
+	struct otx2_sec_session_ipsec_lp *sess;
+	struct otx2_ipsec_po_sa_ctl *ctl_wrd;
+	struct otx2_sec_session *priv;
+	struct cpt_request_info *req;
+	int ret;
+
+	priv = get_sec_session_private_data(op->sym->sec_session);
+	sess = &priv->ipsec.lp;
+
+	ctl_wrd = &sess->in_sa.ctl;
+
+	if (ctl_wrd->direction == OTX2_IPSEC_PO_SA_DIRECTION_OUTBOUND)
+		ret = process_outb_sa(op, sess, &qp->meta_info, (void **)&req);
+	else
+		ret = process_inb_sa(op, sess, &qp->meta_info, (void **)&req);
+
+	if (unlikely(ret)) {
+		otx2_err("Crypto req : op %p, ret 0x%x", op, ret);
+		return ret;
+	}
+
+	ret = otx2_cpt_enqueue_req(qp, pend_q, req);
+
+	return ret;
+}
+
 static __rte_always_inline int __rte_hot
 otx2_cpt_enqueue_sym_sessless(struct otx2_cpt_qp *qp, struct rte_crypto_op *op,
 			      struct pending_queue *pend_q)
@@ -659,7 +691,9 @@ otx2_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops)
 	for (count = 0; count < nb_ops; count++) {
 		op = ops[count];
 		if (op->type == RTE_CRYPTO_OP_TYPE_SYMMETRIC) {
-			if (op->sess_type == RTE_CRYPTO_OP_WITH_SESSION)
+			if (op->sess_type == RTE_CRYPTO_OP_SECURITY_SESSION)
+				ret = otx2_cpt_enqueue_sec(qp, op, pend_q);
+			else if (op->sess_type == RTE_CRYPTO_OP_WITH_SESSION)
 				ret = otx2_cpt_enqueue_sym(qp, op, pend_q);
 			else
 				ret = otx2_cpt_enqueue_sym_sessless(qp, op,
@@ -801,11 +835,48 @@ otx2_cpt_asym_post_process(struct rte_crypto_op *cop,
 	}
 }
 
+static void
+otx2_cpt_sec_post_process(struct rte_crypto_op *cop, uintptr_t *rsp)
+{
+	struct cpt_request_info *req = (struct cpt_request_info *)rsp[2];
+	vq_cmd_word0_t *word0 = (vq_cmd_word0_t *)&req->ist.ei0;
+	struct rte_crypto_sym_op *sym_op = cop->sym;
+	struct rte_mbuf *m = sym_op->m_src;
+	struct rte_ipv4_hdr *ip;
+	uint16_t m_len;
+	int mdata_len;
+	char *data;
+
+	mdata_len = (int)rsp[3];
+	rte_pktmbuf_trim(m, mdata_len);
+
+	if ((word0->s.opcode & 0xff) == OTX2_IPSEC_PO_PROCESS_IPSEC_INB) {
+		data = rte_pktmbuf_mtod(m, char *);
+		ip = (struct rte_ipv4_hdr *)(data + OTX2_IPSEC_PO_INB_RPTR_HDR);
+
+		m_len = rte_be_to_cpu_16(ip->total_length);
+
+		m->data_len = m_len;
+		m->pkt_len = m_len;
+		m->data_off += OTX2_IPSEC_PO_INB_RPTR_HDR;
+	}
+}
+
 static inline void
 otx2_cpt_dequeue_post_process(struct otx2_cpt_qp *qp, struct rte_crypto_op *cop,
 			      uintptr_t *rsp, uint8_t cc)
 {
 	if (cop->type == RTE_CRYPTO_OP_TYPE_SYMMETRIC) {
+		if (cop->sess_type == RTE_CRYPTO_OP_SECURITY_SESSION) {
+			if (likely(cc == OTX2_IPSEC_PO_CC_SUCCESS)) {
+				otx2_cpt_sec_post_process(cop, rsp);
+				cop->status = RTE_CRYPTO_OP_STATUS_SUCCESS;
+			} else
+				cop->status = RTE_CRYPTO_OP_STATUS_ERROR;
+
+			return;
+		}
+
 		if (likely(cc == NO_ERR)) {
 			/* Verify authentication data if required */
 			if (unlikely(rsp[2]))
diff --git a/drivers/crypto/octeontx2/otx2_ipsec_po.h b/drivers/crypto/octeontx2/otx2_ipsec_po.h
index f2167f220a..020748609e 100644
--- a/drivers/crypto/octeontx2/otx2_ipsec_po.h
+++ b/drivers/crypto/octeontx2/otx2_ipsec_po.h
@@ -22,6 +22,8 @@
 #define OTX2_IPSEC_PO_PROCESS_IPSEC_OUTB   0x23
 #define OTX2_IPSEC_PO_PROCESS_IPSEC_INB    0x24
 
+#define OTX2_IPSEC_PO_INB_RPTR_HDR         0x8
+
 enum otx2_ipsec_po_comp_e {
 	OTX2_IPSEC_PO_CC_SUCCESS = 0x00,
 	OTX2_IPSEC_PO_CC_AUTH_UNSUPPORTED = 0xB0,
@@ -86,6 +88,12 @@ enum {
 	OTX2_IPSEC_PO_SA_ENCAP_UDP = 1,
 };
 
+struct otx2_ipsec_po_out_hdr {
+	uint32_t ip_id;
+	uint32_t seq;
+	uint8_t iv[16];
+};
+
 union otx2_ipsec_po_bit_perfect_iv {
 	uint8_t aes_iv[16];
 	uint8_t des_iv[8];
diff --git a/drivers/crypto/octeontx2/otx2_ipsec_po_ops.h b/drivers/crypto/octeontx2/otx2_ipsec_po_ops.h
new file mode 100644
index 0000000000..dd29c413d3
--- /dev/null
+++ b/drivers/crypto/octeontx2/otx2_ipsec_po_ops.h
@@ -0,0 +1,175 @@
+
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(C) 2019 Marvell International Ltd.
+ */
+
+#ifndef __OTX2_IPSEC_PO_OPS_H__
+#define __OTX2_IPSEC_PO_OPS_H__
+
+#include <rte_crypto_sym.h>
+#include <rte_security.h>
+
+#include "otx2_cryptodev.h"
+#include "otx2_security.h"
+
+static __rte_always_inline int32_t
+otx2_ipsec_po_out_rlen_get(struct otx2_sec_session_ipsec_lp *sess,
+			   uint32_t plen)
+{
+	uint32_t enc_payload_len;
+
+	enc_payload_len = RTE_ALIGN_CEIL(plen + sess->roundup_len,
+			sess->roundup_byte);
+
+	return sess->partial_len + enc_payload_len;
+}
+
+static __rte_always_inline struct cpt_request_info *
+alloc_request_struct(char *maddr, void *cop, int mdata_len)
+{
+	struct cpt_request_info *req;
+	struct cpt_meta_info *meta;
+	uint8_t *resp_addr;
+	uintptr_t *op;
+
+	meta = (void *)RTE_PTR_ALIGN((uint8_t *)maddr, 16);
+
+	op = (uintptr_t *)meta->deq_op_info;
+	req = &meta->cpt_req;
+	resp_addr = (uint8_t *)&meta->cpt_res;
+
+	req->completion_addr = (uint64_t *)((uint8_t *)resp_addr);
+	*req->completion_addr = COMPLETION_CODE_INIT;
+	req->comp_baddr = rte_mem_virt2iova(resp_addr);
+	req->op = op;
+
+	op[0] = (uintptr_t)((uint64_t)meta | 1ull);
+	op[1] = (uintptr_t)cop;
+	op[2] = (uintptr_t)req;
+	op[3] = mdata_len;
+
+	return req;
+}
+
+static __rte_always_inline int
+process_outb_sa(struct rte_crypto_op *cop,
+	       struct otx2_sec_session_ipsec_lp *sess,
+	       struct cpt_qp_meta_info *m_info, void **prep_req)
+{
+	uint32_t dlen, rlen, extend_head, extend_tail;
+	struct rte_crypto_sym_op *sym_op = cop->sym;
+	struct rte_mbuf *m_src = sym_op->m_src;
+	struct otx2_ipsec_po_sa_ctl *ctl_wrd;
+	struct cpt_request_info *req = NULL;
+	struct otx2_ipsec_po_out_hdr *hdr;
+	struct otx2_ipsec_po_out_sa *sa;
+	int hdr_len, mdata_len, ret = 0;
+	vq_cmd_word0_t word0;
+	char *mdata, *data;
+
+	sa = &sess->out_sa;
+	ctl_wrd = &sa->ctl;
+	hdr_len = sizeof(*hdr);
+
+	dlen = rte_pktmbuf_pkt_len(m_src) + hdr_len;
+	rlen = otx2_ipsec_po_out_rlen_get(sess, dlen - hdr_len);
+
+	extend_head = hdr_len + RTE_ETHER_HDR_LEN;
+	extend_tail = rlen - dlen;
+	mdata_len = m_info->lb_mlen + 8;
+
+	mdata = rte_pktmbuf_append(m_src, extend_tail + mdata_len);
+	if (unlikely(mdata == NULL)) {
+		otx2_err("Not enough tail room\n");
+		ret = -ENOMEM;
+		goto exit;
+	}
+
+	mdata += extend_tail; /* mdata follows encrypted data */
+	req = alloc_request_struct(mdata, (void *)cop, mdata_len);
+
+	data = rte_pktmbuf_prepend(m_src, extend_head);
+	if (unlikely(data == NULL)) {
+		otx2_err("Not enough head room\n");
+		ret = -ENOMEM;
+		goto exit;
+	}
+
+	/*
+	 * Move the Ethernet header, to insert otx2_ipsec_po_out_hdr prior
+	 * to the IP header
+	 */
+	memcpy(data, data + hdr_len, RTE_ETHER_HDR_LEN);
+
+	hdr = (struct otx2_ipsec_po_out_hdr *)rte_pktmbuf_adj(m_src,
+							RTE_ETHER_HDR_LEN);
+
+	if (ctl_wrd->enc_type == OTX2_IPSEC_FP_SA_ENC_AES_GCM) {
+		memcpy(&hdr->iv[0], &sa->iv.gcm.nonce, 4);
+		memcpy(&hdr->iv[4], rte_crypto_op_ctod_offset(cop, uint8_t *,
+			sess->iv_offset), sess->iv_length);
+	} else if (ctl_wrd->auth_type == OTX2_IPSEC_FP_SA_ENC_AES_CBC) {
+		memcpy(&hdr->iv[0], rte_crypto_op_ctod_offset(cop, uint8_t *,
+			sess->iv_offset), sess->iv_length);
+	}
+
+	/* Prepare CPT instruction */
+	word0.u64 = sess->ucmd_w0;
+	word0.s.dlen = dlen;
+
+	req->ist.ei0 = word0.u64;
+	req->ist.ei1 = rte_pktmbuf_iova(m_src);
+	req->ist.ei2 = req->ist.ei1;
+	req->ist.ei3 = sess->ucmd_w3;
+
+	hdr->seq = rte_cpu_to_be_32(sess->seq_lo);
+	hdr->ip_id = rte_cpu_to_be_32(sess->ip_id);
+
+	sess->ip_id++;
+	sess->esn++;
+
+exit:
+	*prep_req = req;
+
+	return ret;
+}
+
+static __rte_always_inline int
+process_inb_sa(struct rte_crypto_op *cop,
+	      struct otx2_sec_session_ipsec_lp *sess,
+	      struct cpt_qp_meta_info *m_info, void **prep_req)
+{
+	struct rte_crypto_sym_op *sym_op = cop->sym;
+	struct rte_mbuf *m_src = sym_op->m_src;
+	struct cpt_request_info *req = NULL;
+	int mdata_len, ret = 0;
+	vq_cmd_word0_t word0;
+	uint32_t dlen;
+	char *mdata;
+
+	dlen = rte_pktmbuf_pkt_len(m_src);
+	mdata_len = m_info->lb_mlen + 8;
+
+	mdata = rte_pktmbuf_append(m_src, mdata_len);
+	if (unlikely(mdata == NULL)) {
+		otx2_err("Not enough tail room\n");
+		ret = -ENOMEM;
+		goto exit;
+	}
+
+	req = alloc_request_struct(mdata, (void *)cop, mdata_len);
+
+	/* Prepare CPT instruction */
+	word0.u64 = sess->ucmd_w0;
+	word0.s.dlen   = dlen;
+
+	req->ist.ei0 = word0.u64;
+	req->ist.ei1 = rte_pktmbuf_iova(m_src);
+	req->ist.ei2 = req->ist.ei1;
+	req->ist.ei3 = sess->ucmd_w3;
+
+exit:
+	*prep_req = req;
+	return ret;
+}
+#endif /* __OTX2_IPSEC_PO_OPS_H__ */
-- 
2.27.0