[v2,09/11] crypto/cnxk: support cn10k IPsec SG mode

Message ID 20230224094014.3246764-10-ktejasree@marvell.com (mailing list archive)
State Accepted, archived
Delegated to: akhil goyal
Headers
Series fixes and improvements to cnxk crypto PMD |

Checks

Context Check Description
ci/checkpatch success coding style OK

Commit Message

Tejasree Kondoj Feb. 24, 2023, 9:40 a.m. UTC
  Adding support for scatter-gather mode in 103XX and
106XX lookaside IPsec.

Signed-off-by: Tejasree Kondoj <ktejasree@marvell.com>
---
 drivers/crypto/cnxk/cn10k_cryptodev_ops.c |  21 +-
 drivers/crypto/cnxk/cn10k_ipsec_la_ops.h  | 222 ++++++++++++++++++++--
 drivers/crypto/cnxk/cnxk_sg.h             |  23 +++
 3 files changed, 239 insertions(+), 27 deletions(-)
  

Patch

diff --git a/drivers/crypto/cnxk/cn10k_cryptodev_ops.c b/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
index 9f6fd4e411..e405a2ad9f 100644
--- a/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
+++ b/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
@@ -77,8 +77,8 @@  cn10k_cpt_sym_temp_sess_create(struct cnxk_cpt_qp *qp, struct rte_crypto_op *op)
 }
 
 static __rte_always_inline int __rte_hot
-cpt_sec_inst_fill(struct cnxk_cpt_qp *qp, struct rte_crypto_op *op,
-		  struct cn10k_sec_session *sess, struct cpt_inst_s *inst)
+cpt_sec_inst_fill(struct cnxk_cpt_qp *qp, struct rte_crypto_op *op, struct cn10k_sec_session *sess,
+		  struct cpt_inst_s *inst, struct cpt_inflight_req *infl_req, const bool is_sg_ver2)
 {
 	struct rte_crypto_sym_op *sym_op = op->sym;
 	int ret;
@@ -88,15 +88,11 @@  cpt_sec_inst_fill(struct cnxk_cpt_qp *qp, struct rte_crypto_op *op,
 		return -ENOTSUP;
 	}
 
-	if (unlikely(!rte_pktmbuf_is_contiguous(sym_op->m_src))) {
-		plt_dp_err("Scatter Gather mode is not supported");
-		return -ENOTSUP;
-	}
-
 	if (sess->is_outbound)
-		ret = process_outb_sa(&qp->lf, op, sess, inst);
+		ret = process_outb_sa(&qp->lf, op, sess, &qp->meta_info, infl_req, inst,
+				      is_sg_ver2);
 	else
-		ret = process_inb_sa(op, sess, inst);
+		ret = process_inb_sa(op, sess, inst, &qp->meta_info, infl_req, is_sg_ver2);
 
 	return ret;
 }
@@ -129,7 +125,7 @@  cn10k_cpt_fill_inst(struct cnxk_cpt_qp *qp, struct rte_crypto_op *ops[], struct
 	if (op->type == RTE_CRYPTO_OP_TYPE_SYMMETRIC) {
 		if (op->sess_type == RTE_CRYPTO_OP_SECURITY_SESSION) {
 			sec_sess = (struct cn10k_sec_session *)sym_op->session;
-			ret = cpt_sec_inst_fill(qp, op, sec_sess, &inst[0]);
+			ret = cpt_sec_inst_fill(qp, op, sec_sess, &inst[0], infl_req, is_sg_ver2);
 			if (unlikely(ret))
 				return 0;
 			w7 = sec_sess->inst.w7;
@@ -827,7 +823,10 @@  cn10k_cpt_sec_post_process(struct rte_crypto_op *cop, struct cpt_cn10k_res_s *re
 		cop->status = RTE_CRYPTO_OP_STATUS_ERROR;
 		return;
 	}
-	mbuf->data_len = m_len;
+
+	if (mbuf->next == NULL)
+		mbuf->data_len = m_len;
+
 	mbuf->pkt_len = m_len;
 }
 
diff --git a/drivers/crypto/cnxk/cn10k_ipsec_la_ops.h b/drivers/crypto/cnxk/cn10k_ipsec_la_ops.h
index f2761a55a5..8e208eb2ca 100644
--- a/drivers/crypto/cnxk/cn10k_ipsec_la_ops.h
+++ b/drivers/crypto/cnxk/cn10k_ipsec_la_ops.h
@@ -8,9 +8,13 @@ 
 #include <rte_crypto_sym.h>
 #include <rte_security.h>
 
+#include "roc_ie.h"
+
 #include "cn10k_cryptodev.h"
 #include "cn10k_ipsec.h"
 #include "cnxk_cryptodev.h"
+#include "cnxk_cryptodev_ops.h"
+#include "cnxk_sg.h"
 
 static inline void
 ipsec_po_sa_iv_set(struct cn10k_sec_session *sess, struct rte_crypto_op *cop)
@@ -44,18 +48,14 @@  ipsec_po_sa_aes_gcm_iv_set(struct cn10k_sec_session *sess, struct rte_crypto_op
 
 static __rte_always_inline int
 process_outb_sa(struct roc_cpt_lf *lf, struct rte_crypto_op *cop, struct cn10k_sec_session *sess,
-		struct cpt_inst_s *inst)
+		struct cpt_qp_meta_info *m_info, struct cpt_inflight_req *infl_req,
+		struct cpt_inst_s *inst, const bool is_sg_ver2)
 {
 	struct rte_crypto_sym_op *sym_op = cop->sym;
 	struct rte_mbuf *m_src = sym_op->m_src;
 	uint64_t inst_w4_u64 = sess->inst.w4;
 	uint64_t dptr;
 
-	if (unlikely(rte_pktmbuf_tailroom(m_src) < sess->max_extended_len)) {
-		plt_dp_err("Not enough tail room");
-		return -ENOMEM;
-	}
-
 	RTE_SET_USED(lf);
 
 #ifdef LA_IPSEC_DEBUG
@@ -79,27 +79,217 @@  process_outb_sa(struct roc_cpt_lf *lf, struct rte_crypto_op *cop, struct cn10k_s
 	if (m_src->ol_flags & RTE_MBUF_F_TX_L4_MASK)
 		inst_w4_u64 &= ~BIT_ULL(32);
 
-	/* Prepare CPT instruction */
-	inst->w4.u64 = inst_w4_u64 | rte_pktmbuf_pkt_len(m_src);
-	dptr = rte_pktmbuf_mtod(m_src, uint64_t);
-	inst->dptr = dptr;
+	if (likely(m_src->next == NULL)) {
+		if (unlikely(rte_pktmbuf_tailroom(m_src) < sess->max_extended_len)) {
+			plt_dp_err("Not enough tail room");
+			return -ENOMEM;
+		}
+
+		/* Prepare CPT instruction */
+		inst->w4.u64 = inst_w4_u64 | rte_pktmbuf_pkt_len(m_src);
+		dptr = rte_pktmbuf_mtod(m_src, uint64_t);
+		inst->dptr = dptr;
+	} else if (is_sg_ver2 == false) {
+		struct roc_sglist_comp *scatter_comp, *gather_comp;
+		uint32_t g_size_bytes, s_size_bytes;
+		struct rte_mbuf *last_seg;
+		uint8_t *in_buffer;
+		uint32_t dlen;
+		void *m_data;
+		int i;
+
+		last_seg = rte_pktmbuf_lastseg(m_src);
+
+		if (unlikely(rte_pktmbuf_tailroom(last_seg) < sess->max_extended_len)) {
+			plt_dp_err("Not enough tail room (required: %d, available: %d)",
+				   sess->max_extended_len, rte_pktmbuf_tailroom(last_seg));
+			return -ENOMEM;
+		}
+
+		m_data = alloc_op_meta(NULL, m_info->mlen, m_info->pool, infl_req);
+		if (unlikely(m_data == NULL)) {
+			plt_dp_err("Error allocating meta buffer for request");
+			return -ENOMEM;
+		}
+
+		in_buffer = m_data;
+
+		((uint16_t *)in_buffer)[0] = 0;
+		((uint16_t *)in_buffer)[1] = 0;
+
+		/* Input Gather List */
+		i = 0;
+		gather_comp = (struct roc_sglist_comp *)((uint8_t *)m_data + 8);
+
+		i = fill_ipsec_sg_comp_from_pkt(gather_comp, i, m_src);
+		((uint16_t *)in_buffer)[2] = rte_cpu_to_be_16(i);
+
+		g_size_bytes = ((i + 3) / 4) * sizeof(struct roc_sglist_comp);
+
+		/* Output Scatter List */
+		last_seg->data_len += sess->max_extended_len;
+
+		i = 0;
+		scatter_comp = (struct roc_sglist_comp *)((uint8_t *)gather_comp + g_size_bytes);
+
+		i = fill_ipsec_sg_comp_from_pkt(scatter_comp, i, m_src);
+		((uint16_t *)in_buffer)[3] = rte_cpu_to_be_16(i);
+
+		s_size_bytes = ((i + 3) / 4) * sizeof(struct roc_sglist_comp);
+
+		dlen = g_size_bytes + s_size_bytes + ROC_SG_LIST_HDR_SIZE;
+
+		inst->dptr = (uint64_t)in_buffer;
+
+		inst->w4.u64 = sess->inst.w4 | dlen;
+		inst->w4.s.opcode_major |= (uint64_t)ROC_DMA_MODE_SG;
+	} else {
+		struct roc_sg2list_comp *scatter_comp, *gather_comp;
+		union cpt_inst_w5 cpt_inst_w5;
+		union cpt_inst_w6 cpt_inst_w6;
+		struct rte_mbuf *last_seg;
+		uint32_t g_size_bytes;
+		void *m_data;
+		int i;
+
+		last_seg = rte_pktmbuf_lastseg(m_src);
+
+		if (unlikely(rte_pktmbuf_tailroom(last_seg) < sess->max_extended_len)) {
+			plt_dp_err("Not enough tail room (required: %d, available: %d)",
+				   sess->max_extended_len, rte_pktmbuf_tailroom(last_seg));
+			return -ENOMEM;
+		}
+
+		m_data = alloc_op_meta(NULL, m_info->mlen, m_info->pool, infl_req);
+		if (unlikely(m_data == NULL)) {
+			plt_dp_err("Error allocating meta buffer for request");
+			return -ENOMEM;
+		}
+
+		/* Input Gather List */
+		i = 0;
+		gather_comp = (struct roc_sg2list_comp *)((uint8_t *)m_data);
+
+		i = fill_ipsec_sg2_comp_from_pkt(gather_comp, i, m_src);
+
+		cpt_inst_w5.s.gather_sz = ((i + 2) / 3);
+		g_size_bytes = ((i + 2) / 3) * sizeof(struct roc_sg2list_comp);
+
+		/* Output Scatter List */
+		last_seg->data_len += sess->max_extended_len;
+
+		i = 0;
+		scatter_comp = (struct roc_sg2list_comp *)((uint8_t *)gather_comp + g_size_bytes);
+
+		i = fill_ipsec_sg2_comp_from_pkt(scatter_comp, i, m_src);
+
+		cpt_inst_w6.s.scatter_sz = ((i + 2) / 3);
+
+		cpt_inst_w5.s.dptr = (uint64_t)gather_comp;
+		cpt_inst_w6.s.rptr = (uint64_t)scatter_comp;
+
+		inst->w5.u64 = cpt_inst_w5.u64;
+		inst->w6.u64 = cpt_inst_w6.u64;
+		inst->w4.u64 = sess->inst.w4 | rte_pktmbuf_pkt_len(m_src);
+		inst->w4.s.opcode_major &= (~(ROC_IE_OT_INPLACE_BIT));
+	}
 
 	return 0;
 }
 
 static __rte_always_inline int
-process_inb_sa(struct rte_crypto_op *cop, struct cn10k_sec_session *sess, struct cpt_inst_s *inst)
+process_inb_sa(struct rte_crypto_op *cop, struct cn10k_sec_session *sess, struct cpt_inst_s *inst,
+	       struct cpt_qp_meta_info *m_info, struct cpt_inflight_req *infl_req,
+	       const bool is_sg_ver2)
 {
 	struct rte_crypto_sym_op *sym_op = cop->sym;
 	struct rte_mbuf *m_src = sym_op->m_src;
 	uint64_t dptr;
 
-	/* Prepare CPT instruction */
-	inst->w4.u64 = sess->inst.w4 | rte_pktmbuf_pkt_len(m_src);
-	dptr = rte_pktmbuf_mtod(m_src, uint64_t);
-	inst->dptr = dptr;
-	m_src->ol_flags |= (uint64_t)sess->ip_csum;
+	if (likely(m_src->next == NULL)) {
+		/* Prepare CPT instruction */
+		inst->w4.u64 = sess->inst.w4 | rte_pktmbuf_pkt_len(m_src);
+		dptr = rte_pktmbuf_mtod(m_src, uint64_t);
+		inst->dptr = dptr;
+		m_src->ol_flags |= (uint64_t)sess->ip_csum;
+	} else if (is_sg_ver2 == false) {
+		struct roc_sglist_comp *scatter_comp, *gather_comp;
+		uint32_t g_size_bytes, s_size_bytes;
+		uint8_t *in_buffer;
+		uint32_t dlen;
+		void *m_data;
+		int i;
+
+		m_data = alloc_op_meta(NULL, m_info->mlen, m_info->pool, infl_req);
+		if (unlikely(m_data == NULL)) {
+			plt_dp_err("Error allocating meta buffer for request");
+			return -ENOMEM;
+		}
+
+		in_buffer = m_data;
+
+		((uint16_t *)in_buffer)[0] = 0;
+		((uint16_t *)in_buffer)[1] = 0;
+
+		/* Input Gather List */
+		i = 0;
+		gather_comp = (struct roc_sglist_comp *)((uint8_t *)m_data + 8);
+		i = fill_ipsec_sg_comp_from_pkt(gather_comp, i, m_src);
+		((uint16_t *)in_buffer)[2] = rte_cpu_to_be_16(i);
+
+		g_size_bytes = ((i + 3) / 4) * sizeof(struct roc_sglist_comp);
 
+		/* Output Scatter List */
+		i = 0;
+		scatter_comp = (struct roc_sglist_comp *)((uint8_t *)gather_comp + g_size_bytes);
+		i = fill_ipsec_sg_comp_from_pkt(scatter_comp, i, m_src);
+		((uint16_t *)in_buffer)[3] = rte_cpu_to_be_16(i);
+
+		s_size_bytes = ((i + 3) / 4) * sizeof(struct roc_sglist_comp);
+
+		dlen = g_size_bytes + s_size_bytes + ROC_SG_LIST_HDR_SIZE;
+
+		inst->dptr = (uint64_t)in_buffer;
+		inst->w4.u64 = sess->inst.w4 | dlen;
+		inst->w4.s.opcode_major |= (uint64_t)ROC_DMA_MODE_SG;
+	} else {
+		struct roc_sg2list_comp *scatter_comp, *gather_comp;
+		union cpt_inst_w5 cpt_inst_w5;
+		union cpt_inst_w6 cpt_inst_w6;
+		uint32_t g_size_bytes;
+		void *m_data;
+		int i;
+
+		m_data = alloc_op_meta(NULL, m_info->mlen, m_info->pool, infl_req);
+		if (unlikely(m_data == NULL)) {
+			plt_dp_err("Error allocating meta buffer for request");
+			return -ENOMEM;
+		}
+
+		/* Input Gather List */
+		i = 0;
+		gather_comp = (struct roc_sg2list_comp *)((uint8_t *)m_data);
+
+		i = fill_ipsec_sg2_comp_from_pkt(gather_comp, i, m_src);
+
+		cpt_inst_w5.s.gather_sz = ((i + 2) / 3);
+		g_size_bytes = ((i + 2) / 3) * sizeof(struct roc_sg2list_comp);
+
+		/* Output Scatter List */
+		i = 0;
+		scatter_comp = (struct roc_sg2list_comp *)((uint8_t *)gather_comp + g_size_bytes);
+		i = fill_ipsec_sg2_comp_from_pkt(scatter_comp, i, m_src);
+
+		cpt_inst_w6.s.scatter_sz = ((i + 2) / 3);
+
+		cpt_inst_w5.s.dptr = (uint64_t)gather_comp;
+		cpt_inst_w6.s.rptr = (uint64_t)scatter_comp;
+
+		inst->w5.u64 = cpt_inst_w5.u64;
+		inst->w6.u64 = cpt_inst_w6.u64;
+		inst->w4.u64 = sess->inst.w4 | rte_pktmbuf_pkt_len(m_src);
+		inst->w4.s.opcode_major &= (~(ROC_IE_OT_INPLACE_BIT));
+	}
 	return 0;
 }
 
diff --git a/drivers/crypto/cnxk/cnxk_sg.h b/drivers/crypto/cnxk/cnxk_sg.h
index ead2886e99..65244199bd 100644
--- a/drivers/crypto/cnxk/cnxk_sg.h
+++ b/drivers/crypto/cnxk/cnxk_sg.h
@@ -6,6 +6,7 @@ 
 #define _CNXK_SG_H_
 
 #include "roc_cpt_sg.h"
+#include "roc_se.h"
 
 static __rte_always_inline uint32_t
 fill_sg_comp(struct roc_sglist_comp *list, uint32_t i, phys_addr_t dma_addr, uint32_t size)
@@ -148,6 +149,28 @@  fill_ipsec_sg_comp_from_pkt(struct roc_sglist_comp *list, uint32_t i, struct rte
 	return i;
 }
 
+static __rte_always_inline uint32_t
+fill_ipsec_sg2_comp_from_pkt(struct roc_sg2list_comp *list, uint32_t i, struct rte_mbuf *pkt)
+{
+	uint32_t buf_sz;
+	void *vaddr;
+
+	while (unlikely(pkt != NULL)) {
+		struct roc_sg2list_comp *to = &list[i / 3];
+		buf_sz = pkt->data_len;
+		vaddr = rte_pktmbuf_mtod(pkt, void *);
+
+		to->u.s.len[i % 3] = buf_sz;
+		to->ptr[i % 3] = (uint64_t)vaddr;
+		to->u.s.valid_segs = (i % 3) + 1;
+
+		pkt = pkt->next;
+		i++;
+	}
+
+	return i;
+}
+
 static __rte_always_inline uint32_t
 fill_sg2_comp(struct roc_sg2list_comp *list, uint32_t i, phys_addr_t dma_addr, uint32_t size)
 {