From: Hemant Agrawal <hemant.agrawal@nxp.com>
To: Akhil Goyal <akhil.goyal@nxp.com>, <dev@dpdk.org>
Cc: <pablo.de.lara.guarch@intel.com>, <shreyansh.jain@nxp.com>,
Nipun Gupta <nipun.gupta@nxp.com>
Subject: Re: [dpdk-dev] [PATCH 5/5] crypto/dpaa_sec: rewrite Rx/Tx path
Date: Tue, 19 Dec 2017 18:15:55 +0530 [thread overview]
Message-ID: <c1259a3f-b0b9-d9b6-63cc-606c4f247e06@nxp.com> (raw)
In-Reply-To: <20171213135659.32648-6-akhil.goyal@nxp.com>
Hi Akhil,
On 12/13/2017 7:26 PM, Akhil Goyal wrote:
> Rx and Tx patch are rewritten with improved internal APIs
> to improve performance.
>
> Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com>
> Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
> ---
> drivers/crypto/dpaa_sec/dpaa_sec.c | 260 ++++++++++++++++++++++---------------
> drivers/crypto/dpaa_sec/dpaa_sec.h | 2 +-
> 2 files changed, 153 insertions(+), 109 deletions(-)
>
> diff --git a/drivers/crypto/dpaa_sec/dpaa_sec.c b/drivers/crypto/dpaa_sec/dpaa_sec.c
> index ea744e6..b650d5c 100644
> --- a/drivers/crypto/dpaa_sec/dpaa_sec.c
> +++ b/drivers/crypto/dpaa_sec/dpaa_sec.c
> @@ -563,46 +563,67 @@ dpaa_sec_prep_cdb(dpaa_sec_session *ses)
> return 0;
> }
>
..<snip>
> -
> +#define DPAA_MAX_DEQUEUE_NUM_FRAMES 32
It will be better, if you define it in dpaa_sec.h
> /* qp is lockless, should be accessed by only one thread */
> static int
> dpaa_sec_deq(struct dpaa_sec_qp *qp, struct rte_crypto_op **ops, int nb_ops)
> {
> struct qman_fq *fq;
> + unsigned int pkts = 0;
> + int ret;
> + struct qm_dqrr_entry *dq;
>
> fq = &qp->outq;
> - dpaa_sec_op_nb = 0;
> - dpaa_sec_ops = ops;
> + ret = qman_set_vdq(fq, (nb_ops > DPAA_MAX_DEQUEUE_NUM_FRAMES) ?
> + DPAA_MAX_DEQUEUE_NUM_FRAMES : nb_ops);
Any particular reason for keeping the limit as 32 for SEC.
The dpaa eth PMD is using it as 63 i.e 6 bits
Also, you have a option to use '0'. NUM_FRAMES is zero—indicates that
the volatile command is not terminate until the specified FQ becomes
empty.
>
.. <snip>
> -
> static uint16_t
> dpaa_sec_enqueue_burst(void *qp, struct rte_crypto_op **ops,
> uint16_t nb_ops)
> {
> /* Function to transmit the frames to given device and queuepair */
> uint32_t loop;
> - int32_t ret;
> struct dpaa_sec_qp *dpaa_qp = (struct dpaa_sec_qp *)qp;
you can avoid this explicit typecasting
> uint16_t num_tx = 0;
> + struct qm_fd fds[DPAA_SEC_BURST], *fd;
> + uint32_t frames_to_send;
> + struct rte_crypto_op *op;
> + struct dpaa_sec_job *cf;
> + dpaa_sec_session *ses;
> + struct dpaa_sec_op_ctx *ctx;
> + uint32_t auth_only_len;
> + struct qman_fq *inq[DPAA_SEC_BURST];
> +
> + while (nb_ops) {
> + frames_to_send = (nb_ops > DPAA_SEC_BURST) ?
> + DPAA_SEC_BURST : nb_ops;
> + for (loop = 0; loop < frames_to_send; loop++) {
> + op = *(ops++);
> + switch (op->sess_type) {
> + case RTE_CRYPTO_OP_WITH_SESSION:
> + ses = (dpaa_sec_session *)
here and other places as well
> + get_session_private_data(
> + op->sym->session,
> + cryptodev_driver_id);
> + break;
> + case RTE_CRYPTO_OP_SECURITY_SESSION:
> + ses = (dpaa_sec_session *)
> + get_sec_session_private_data(
> + op->sym->sec_session);
> + break;
> + default:
> + PMD_TX_LOG(ERR,
> + "sessionless crypto op not supported");
> + frames_to_send = loop;
> + nb_ops = loop;
> + goto send_pkts;
> + }
> + if (unlikely(!ses->qp || ses->qp != qp)) {
> + PMD_INIT_LOG(DEBUG, "sess->qp - %p qp %p",
> + ses->qp, qp);
> + if (dpaa_sec_attach_sess_q(qp, ses)) {
> + frames_to_send = loop;
> + nb_ops = loop;
> + goto send_pkts;
> + }
> + }
>
> - if (unlikely(nb_ops == 0))
> - return 0;
> + /*
> + * Segmented buffer is not supported.
> + */
> + if (!rte_pktmbuf_is_contiguous(op->sym->m_src)) {
> + op->status = RTE_CRYPTO_OP_STATUS_ERROR;
> + frames_to_send = loop;
> + nb_ops = loop;
> + goto send_pkts;
> + }
> + auth_only_len = op->sym->auth.data.length -
> + op->sym->cipher.data.length;
> +
> + if (is_auth_only(ses)) {
> + cf = build_auth_only(op, ses);
> + } else if (is_cipher_only(ses)) {
> + cf = build_cipher_only(op, ses);
> + } else if (is_aead(ses)) {
> + cf = build_cipher_auth_gcm(op, ses);
> + auth_only_len = ses->auth_only_len;
> + } else if (is_auth_cipher(ses)) {
> + cf = build_cipher_auth(op, ses);
> + } else if (is_proto_ipsec(ses)) {
> + cf = build_proto(op, ses);
> + } else {
> + PMD_TX_LOG(ERR, "not supported sec op");
> + frames_to_send = loop;
> + nb_ops = loop;
> + goto send_pkts;
> + }
next prev parent reply other threads:[~2017-12-19 12:46 UTC|newest]
Thread overview: 24+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-12-13 13:56 [dpdk-dev] [PATCH 0/5] crypto/dpaa_sec: performance optimizations Akhil Goyal
2017-12-13 13:56 ` [dpdk-dev] [PATCH 1/5] crypto/dpaa_sec: optimize virt to phy conversion Akhil Goyal
2017-12-13 13:56 ` [dpdk-dev] [PATCH 2/5] crypto/dpaa_sec: support multiple sessions per qp Akhil Goyal
2017-12-13 13:56 ` [dpdk-dev] [PATCH 3/5] crypto/dpaa_sec: support ipsec protocol offload Akhil Goyal
2017-12-19 12:59 ` Hemant Agrawal
2017-12-13 13:56 ` [dpdk-dev] [PATCH 4/5] bus/dpaa: support for enqueue frames of multiple queues Akhil Goyal
2017-12-19 11:32 ` Hemant Agrawal
2017-12-13 13:56 ` [dpdk-dev] [PATCH 5/5] crypto/dpaa_sec: rewrite Rx/Tx path Akhil Goyal
2017-12-19 12:45 ` Hemant Agrawal [this message]
2018-01-08 11:13 ` De Lara Guarch, Pablo
2018-01-08 11:16 ` Akhil Goyal
2018-01-08 11:24 ` De Lara Guarch, Pablo
2018-01-11 11:44 ` [dpdk-dev] [PATCH v2] " Akhil Goyal
2018-01-17 16:54 ` De Lara Guarch, Pablo
2018-01-11 11:33 ` [dpdk-dev] [PATCH v2 0/3] crypto/dpaa_sec: performance optimizations Akhil Goyal
2018-01-11 11:33 ` [dpdk-dev] [PATCH v2 1/3] crypto/dpaa_sec: optimize virt to phy conversion Akhil Goyal
2018-01-11 11:33 ` [dpdk-dev] [PATCH v2 2/3] crypto/dpaa_sec: support multiple sessions per qp Akhil Goyal
2018-01-11 11:33 ` [dpdk-dev] [PATCH v2 3/3] crypto/dpaa_sec: support ipsec protocol offload Akhil Goyal
2018-01-11 14:13 ` De Lara Guarch, Pablo
2018-01-15 6:35 ` [dpdk-dev] [PATCH v3 0/3] crypto/dpaa_sec: performance optimizations Akhil Goyal
2018-01-15 6:35 ` [dpdk-dev] [PATCH v3 1/3] crypto/dpaa_sec: optimize virt to phy conversion Akhil Goyal
2018-01-15 6:35 ` [dpdk-dev] [PATCH v3 2/3] crypto/dpaa_sec: support multiple sessions per qp Akhil Goyal
2018-01-15 6:35 ` [dpdk-dev] [PATCH v3 3/3] crypto/dpaa_sec: support ipsec protocol offload Akhil Goyal
2018-01-15 14:46 ` [dpdk-dev] [PATCH v3 0/3] crypto/dpaa_sec: performance optimizations De Lara Guarch, Pablo
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=c1259a3f-b0b9-d9b6-63cc-606c4f247e06@nxp.com \
--to=hemant.agrawal@nxp.com \
--cc=akhil.goyal@nxp.com \
--cc=dev@dpdk.org \
--cc=nipun.gupta@nxp.com \
--cc=pablo.de.lara.guarch@intel.com \
--cc=shreyansh.jain@nxp.com \
/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).