octeon: add crypto framework

Configure crypto device.
Add crypto support in control plane and data plane.
Control plane
 - Handle vnet crypto key add and delete
 - Register crypto async enqueue and dequeue handlers
Data plane
 - Add encryption and decryption support for
    - AES-GCM
    - AES-CBC hmac sha1/256/384/512
    - AES-CTR sha1
    - 3DES-CBC md5 sha1/256/384/512

Type: feature

Signed-off-by: Nithinsen Kaithakadan <nkaithakadan@marvell.com>
Signed-off-by: Monendra Singh Kushwaha <kmonendra@marvell.com>
Change-Id: Ia9e16c61ed84800a59e0c932a4ba6aa1423c1ec8
This commit is contained in:
Nithinsen Kaithakadan
2024-02-16 11:58:04 +05:30
committed by Damjan Marion
parent 17a918133b
commit 6937c0b2df
6 changed files with 2040 additions and 13 deletions

View File

@ -36,6 +36,7 @@ add_vpp_plugin(dev_octeon
tx_node.c
flow.c
counter.c
crypto.c
MULTIARCH_SOURCES
rx_node.c

File diff suppressed because it is too large Load Diff

View File

@ -0,0 +1,174 @@
/*
* Copyright (c) 2024 Marvell.
* SPDX-License-Identifier: Apache-2.0
* https://spdx.org/licenses/Apache-2.0.html
*/
#ifndef _CRYPTO_H_
#define _CRYPTO_H_
#include <vnet/crypto/crypto.h>
#include <vnet/ip/ip.h>
#define OCT_MAX_N_CPT_DEV 2
#define OCT_CPT_LF_MAX_NB_DESC 128000
/* CRYPTO_ID, KEY_LENGTH_IN_BYTES, TAG_LEN, AAD_LEN */
#define foreach_oct_crypto_aead_async_alg \
_ (AES_128_GCM, 16, 16, 8) \
_ (AES_128_GCM, 16, 16, 12) \
_ (AES_192_GCM, 24, 16, 8) \
_ (AES_192_GCM, 24, 16, 12) \
_ (AES_256_GCM, 32, 16, 8) \
_ (AES_256_GCM, 32, 16, 12)
/* CRYPTO_ID, INTEG_ID, KEY_LENGTH_IN_BYTES, DIGEST_LEN */
#define foreach_oct_crypto_link_async_alg \
_ (AES_128_CBC, SHA1, 16, 12) \
_ (AES_192_CBC, SHA1, 24, 12) \
_ (AES_256_CBC, SHA1, 32, 12) \
_ (AES_128_CBC, SHA256, 16, 16) \
_ (AES_192_CBC, SHA256, 24, 16) \
_ (AES_256_CBC, SHA256, 32, 16) \
_ (AES_128_CBC, SHA384, 16, 24) \
_ (AES_192_CBC, SHA384, 24, 24) \
_ (AES_256_CBC, SHA384, 32, 24) \
_ (AES_128_CBC, SHA512, 16, 32) \
_ (AES_192_CBC, SHA512, 24, 32) \
_ (AES_256_CBC, SHA512, 32, 32) \
_ (3DES_CBC, MD5, 24, 12) \
_ (3DES_CBC, SHA1, 24, 12) \
_ (3DES_CBC, SHA256, 24, 16) \
_ (3DES_CBC, SHA384, 24, 24) \
_ (3DES_CBC, SHA512, 24, 32) \
_ (AES_128_CTR, SHA1, 16, 12) \
_ (AES_192_CTR, SHA1, 24, 12) \
_ (AES_256_CTR, SHA1, 32, 12)
#define OCT_MOD_INC(i, l) ((i) == (l - 1) ? (i) = 0 : (i)++)
#define OCT_SCATTER_GATHER_BUFFER_SIZE 1024
#define CPT_LMT_SIZE_COPY (sizeof (struct cpt_inst_s) / 16)
#define OCT_MAX_LMT_SZ 16
#define SRC_IOV_SIZE \
(sizeof (struct roc_se_iov_ptr) + \
(sizeof (struct roc_se_buf_ptr) * ROC_MAX_SG_CNT))
#define OCT_CPT_LMT_GET_LINE_ADDR(lmt_addr, lmt_num) \
(void *) ((u64) (lmt_addr) + ((u64) (lmt_num) << ROC_LMT_LINE_SIZE_LOG2))
typedef struct
{
CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
struct roc_cpt *roc_cpt;
struct roc_cpt_lmtline lmtline;
struct roc_cpt_lf lf;
vnet_dev_t *dev;
} oct_crypto_dev_t;
typedef struct
{
CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
/** CPT opcode */
u16 cpt_op : 4;
/** Flag for AES GCM */
u16 aes_gcm : 1;
/** IV length in bytes */
u8 iv_length;
/** Auth IV length in bytes */
u8 auth_iv_length;
/** IV offset in bytes */
u16 iv_offset;
/** Auth IV offset in bytes */
u16 auth_iv_offset;
/** CPT inst word 7 */
u64 cpt_inst_w7;
/* initialise as part of first packet */
u8 initialised;
/* store link key index in case of linked algo */
vnet_crypto_key_index_t key_index;
oct_crypto_dev_t *crypto_dev;
struct roc_se_ctx cpt_ctx;
} oct_crypto_sess_t;
typedef struct
{
CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
oct_crypto_sess_t *sess;
oct_crypto_dev_t *crypto_dev;
} oct_crypto_key_t;
typedef struct oct_crypto_scatter_gather
{
u8 buf[OCT_SCATTER_GATHER_BUFFER_SIZE];
} oct_crypto_scatter_gather_t;
typedef struct
{
CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
/** Result data of all entries in the frame */
volatile union cpt_res_s res[VNET_CRYPTO_FRAME_SIZE];
/** Scatter gather data */
void *sg_data;
/** Frame pointer */
vnet_crypto_async_frame_t *frame;
/** Number of async elements in frame */
u16 elts;
/** Next read entry in frame, when dequeue */
u16 deq_elts;
} oct_crypto_inflight_req_t;
typedef struct
{
/** Array of pending request */
oct_crypto_inflight_req_t *req_queue;
/** Number of inflight operations in queue */
u32 n_crypto_inflight;
/** Tail of queue to be used for enqueue */
u16 enq_tail;
/** Head of queue to be used for dequeue */
u16 deq_head;
/** Number of descriptors */
u16 n_desc;
} oct_crypto_pending_queue_t;
typedef struct
{
oct_crypto_dev_t *crypto_dev[OCT_MAX_N_CPT_DEV];
oct_crypto_key_t *keys[VNET_CRYPTO_ASYNC_OP_N_TYPES];
oct_crypto_pending_queue_t *pend_q;
int n_cpt;
u8 started;
} oct_crypto_main_t;
extern oct_crypto_main_t oct_crypto_main;
void oct_crypto_key_del_handler (vlib_main_t *vm,
vnet_crypto_key_index_t key_index);
void oct_crypto_key_add_handler (vlib_main_t *vm,
vnet_crypto_key_index_t key_index);
void oct_crypto_key_handler (vlib_main_t *vm, vnet_crypto_key_op_t kop,
vnet_crypto_key_index_t idx);
int oct_crypto_enqueue_linked_alg_enc (vlib_main_t *vm,
vnet_crypto_async_frame_t *frame);
int oct_crypto_enqueue_linked_alg_dec (vlib_main_t *vm,
vnet_crypto_async_frame_t *frame);
int oct_crypto_enqueue_aead_aad_8_enc (vlib_main_t *vm,
vnet_crypto_async_frame_t *frame);
int oct_crypto_enqueue_aead_aad_12_enc (vlib_main_t *vm,
vnet_crypto_async_frame_t *frame);
int oct_crypto_enqueue_aead_aad_8_dec (vlib_main_t *vm,
vnet_crypto_async_frame_t *frame);
int oct_crypto_enqueue_aead_aad_12_dec (vlib_main_t *vm,
vnet_crypto_async_frame_t *frame);
vnet_crypto_async_frame_t *oct_crypto_frame_dequeue (vlib_main_t *vm,
u32 *nb_elts_processed,
u32 *enqueue_thread_idx);
int oct_init_crypto_engine_handlers (vlib_main_t *vm, vnet_dev_t *dev);
int oct_conf_sw_queue (vlib_main_t *vm, vnet_dev_t *dev);
#endif /* _CRYPTO_H_ */

View File

@ -10,6 +10,7 @@
#include <vnet/plugin/plugin.h>
#include <vpp/app/version.h>
#include <dev_octeon/octeon.h>
#include <dev_octeon/crypto.h>
#include <base/roc_api.h>
#include <common.h>
@ -54,7 +55,9 @@ static struct
_ (0xa064, RVU_VF, "Marvell Octeon Resource Virtualization Unit VF"),
_ (0xa0f8, LBK_VF, "Marvell Octeon Loopback Unit VF"),
_ (0xa0f7, SDP_VF, "Marvell Octeon System DPI Packet Interface Unit VF"),
_ (0xa0f3, CPT_VF, "Marvell Octeon Cryptographic Accelerator Unit VF"),
_ (0xa0f3, O10K_CPT_VF,
"Marvell Octeon-10 Cryptographic Accelerator Unit VF"),
_ (0xa0fe, O9K_CPT_VF, "Marvell Octeon-9 Cryptographic Accelerator Unit VF"),
#undef _
};
@ -191,17 +194,113 @@ oct_init_nix (vlib_main_t *vm, vnet_dev_t *dev)
return vnet_dev_port_add (vm, dev, 0, &port_add_args);
}
static int
oct_conf_cpt (vlib_main_t *vm, vnet_dev_t *dev, oct_crypto_dev_t *ocd,
int nb_lf)
{
struct roc_cpt *roc_cpt = ocd->roc_cpt;
int rrv;
if ((rrv = roc_cpt_eng_grp_add (roc_cpt, CPT_ENG_TYPE_SE)) < 0)
{
log_err (dev, "Could not add CPT SE engines");
return cnx_return_roc_err (dev, rrv, "roc_cpt_eng_grp_add");
}
if ((rrv = roc_cpt_eng_grp_add (roc_cpt, CPT_ENG_TYPE_IE)) < 0)
{
log_err (dev, "Could not add CPT IE engines");
return cnx_return_roc_err (dev, rrv, "roc_cpt_eng_grp_add");
}
if (roc_cpt->eng_grp[CPT_ENG_TYPE_IE] != ROC_CPT_DFLT_ENG_GRP_SE_IE)
{
log_err (dev, "Invalid CPT IE engine group configuration");
return -1;
}
if (roc_cpt->eng_grp[CPT_ENG_TYPE_SE] != ROC_CPT_DFLT_ENG_GRP_SE)
{
log_err (dev, "Invalid CPT SE engine group configuration");
return -1;
}
if ((rrv = roc_cpt_dev_configure (roc_cpt, nb_lf, false, 0)) < 0)
{
log_err (dev, "could not configure crypto device %U",
format_vlib_pci_addr, roc_cpt->pci_dev->addr);
return cnx_return_roc_err (dev, rrv, "roc_cpt_dev_configure");
}
return 0;
}
static vnet_dev_rv_t
oct_conf_cpt_queue (vlib_main_t *vm, vnet_dev_t *dev, oct_crypto_dev_t *ocd)
{
struct roc_cpt *roc_cpt = ocd->roc_cpt;
struct roc_cpt_lmtline *cpt_lmtline;
struct roc_cpt_lf *cpt_lf;
int rrv;
cpt_lf = &ocd->lf;
cpt_lmtline = &ocd->lmtline;
cpt_lf->nb_desc = OCT_CPT_LF_MAX_NB_DESC;
cpt_lf->lf_id = 0;
if ((rrv = roc_cpt_lf_init (roc_cpt, cpt_lf)) < 0)
return cnx_return_roc_err (dev, rrv, "roc_cpt_lf_init");
roc_cpt_iq_enable (cpt_lf);
if ((rrv = roc_cpt_lmtline_init (roc_cpt, cpt_lmtline, 0) < 0))
return cnx_return_roc_err (dev, rrv, "roc_cpt_lmtline_init");
return 0;
}
static vnet_dev_rv_t
oct_init_cpt (vlib_main_t *vm, vnet_dev_t *dev)
{
oct_crypto_main_t *ocm = &oct_crypto_main;
extern oct_plt_init_param_t oct_plt_init_param;
oct_device_t *cd = vnet_dev_get_data (dev);
oct_crypto_dev_t *ocd = NULL;
int rrv;
struct roc_cpt cpt = {
.pci_dev = &cd->plt_pci_dev,
};
if ((rrv = roc_cpt_dev_init (&cpt)))
if (ocm->n_cpt == OCT_MAX_N_CPT_DEV || ocm->started)
return VNET_DEV_ERR_NOT_SUPPORTED;
ocd = oct_plt_init_param.oct_plt_zmalloc (sizeof (oct_crypto_dev_t),
CLIB_CACHE_LINE_BYTES);
ocd->roc_cpt = oct_plt_init_param.oct_plt_zmalloc (sizeof (struct roc_cpt),
CLIB_CACHE_LINE_BYTES);
ocd->roc_cpt->pci_dev = &cd->plt_pci_dev;
ocd->dev = dev;
if ((rrv = roc_cpt_dev_init (ocd->roc_cpt)))
return cnx_return_roc_err (dev, rrv, "roc_cpt_dev_init");
if ((rrv = oct_conf_cpt (vm, dev, ocd, 1)))
return rrv;
if ((rrv = oct_conf_cpt_queue (vm, dev, ocd)))
return rrv;
if (!ocm->n_cpt)
{
/*
* Initialize s/w queues, which are common across multiple
* crypto devices
*/
oct_conf_sw_queue (vm, dev);
ocm->crypto_dev[0] = ocd;
}
ocm->crypto_dev[1] = ocd;
oct_init_crypto_engine_handlers (vm, dev);
ocm->n_cpt++;
return VNET_DEV_OK;
}
@ -256,7 +355,8 @@ oct_init (vlib_main_t *vm, vnet_dev_t *dev)
case OCT_DEVICE_TYPE_SDP_VF:
return oct_init_nix (vm, dev);
case OCT_DEVICE_TYPE_CPT_VF:
case OCT_DEVICE_TYPE_O10K_CPT_VF:
case OCT_DEVICE_TYPE_O9K_CPT_VF:
return oct_init_cpt (vm, dev);
default:

View File

@ -30,7 +30,8 @@ typedef enum
OCT_DEVICE_TYPE_RVU_VF,
OCT_DEVICE_TYPE_LBK_VF,
OCT_DEVICE_TYPE_SDP_VF,
OCT_DEVICE_TYPE_CPT_VF,
OCT_DEVICE_TYPE_O10K_CPT_VF,
OCT_DEVICE_TYPE_O9K_CPT_VF,
} __clib_packed oct_device_type_t;
typedef struct
@ -41,7 +42,6 @@ typedef struct
u8 full_duplex : 1;
u32 speed;
struct plt_pci_device plt_pci_dev;
struct roc_cpt cpt;
struct roc_nix *nix;
} oct_device_t;
@ -102,7 +102,6 @@ typedef struct
u64 aura_handle;
u64 io_addr;
void *lmt_addr;
oct_npa_batch_alloc_cl128_t *ba_buffer;
u8 ba_first_cl;
u8 ba_num_cl;

View File

@ -75,13 +75,12 @@ oct_drv_physmem_alloc (vlib_main_t *vm, u32 size, u32 align)
if (align)
{
/* Force cache line alloc in case alignment is less than cache line */
align = align < CLIB_CACHE_LINE_BYTES ? CLIB_CACHE_LINE_BYTES : align;
/* Force ROC align alloc in case alignment is less than ROC align */
align = align < ROC_ALIGN ? ROC_ALIGN : align;
mem = vlib_physmem_alloc_aligned_on_numa (vm, size, align, 0);
}
else
mem =
vlib_physmem_alloc_aligned_on_numa (vm, size, CLIB_CACHE_LINE_BYTES, 0);
mem = vlib_physmem_alloc_aligned_on_numa (vm, size, ROC_ALIGN, 0);
if (!mem)
return NULL;