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
next prev 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).