[dpdk-stable] [19.11 PATCH] crypto/dpaa_sec: affine the thread portal affinity

[dpdk-stable] [19.11 PATCH] crypto/dpaa_sec: affine the thread portal affinity

[Hemant Agrawal]

DPAA requires the I/O shall be done in a HW portal context only.
The portal affinity is currently only being done in session create
and config APIs with the assumption that same thread will be used
for IO. This is causing issue.
This patch add support during I/O to check the HW portal affinity
and affine portal- if not affined already.

Fixes: 9a984458f755 ("crypto/dpaa_sec: rewrite Rx/Tx path")
Cc: stable@dpdk.org

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/crypto/dpaa_sec/dpaa_sec.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/drivers/crypto/dpaa_sec/dpaa_sec.c b/drivers/crypto/dpaa_sec/dpaa_sec.c
index a650313cdb..6e454273ee 100644
--- a/drivers/crypto/dpaa_sec/dpaa_sec.c
+++ b/drivers/crypto/dpaa_sec/dpaa_sec.c
@@ -1690,6 +1690,13 @@ dpaa_sec_enqueue_burst(void *qp, struct rte_crypto_op **ops,
 	uint32_t index, flags[DPAA_SEC_BURST] = {0};
 	struct qman_fq *inq[DPAA_SEC_BURST];
 
+	if (unlikely(!RTE_PER_LCORE(dpaa_io))) {
+		if (rte_dpaa_portal_init((void *)0)) {
+			DPAA_SEC_ERR("Failure in affining portal");
+			return 0;
+		}
+	}
+
 	while (nb_ops) {
 		frames_to_send = (nb_ops > DPAA_SEC_BURST) ?
 				DPAA_SEC_BURST : nb_ops;
@@ -1890,6 +1897,13 @@ dpaa_sec_dequeue_burst(void *qp, struct rte_crypto_op **ops,
 	uint16_t num_rx;
 	struct dpaa_sec_qp *dpaa_qp = (struct dpaa_sec_qp *)qp;
 
+	if (unlikely(!RTE_PER_LCORE(dpaa_io))) {
+		if (rte_dpaa_portal_init((void *)0)) {
+			DPAA_SEC_ERR("Failure in affining portal");
+			return 0;
+		}
+	}
+
 	num_rx = dpaa_sec_deq(dpaa_qp, ops, nb_ops);
 
 	dpaa_qp->rx_pkts += num_rx;
-- 
2.17.1

Re: [dpdk-stable] [19.11 PATCH] crypto/dpaa_sec: affine the thread portal affinity

[Christian Ehrhardt]

On Thu, Jun 3, 2021 at 2:19 PM Hemant Agrawal <hemant.agrawal@nxp.com> wrote:
>
> DPAA requires the I/O shall be done in a HW portal context only.
> The portal affinity is currently only being done in session create
> and config APIs with the assumption that same thread will be used
> for IO. This is causing issue.
> This patch add support during I/O to check the HW portal affinity
> and affine portal- if not affined already.
>
> Fixes: 9a984458f755 ("crypto/dpaa_sec: rewrite Rx/Tx path")
> Cc: stable@dpdk.org
>
> Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

Applied to 19.11.x - thanks!

> ---
>  drivers/crypto/dpaa_sec/dpaa_sec.c | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
>
> diff --git a/drivers/crypto/dpaa_sec/dpaa_sec.c b/drivers/crypto/dpaa_sec/dpaa_sec.c
> index a650313cdb..6e454273ee 100644
> --- a/drivers/crypto/dpaa_sec/dpaa_sec.c
> +++ b/drivers/crypto/dpaa_sec/dpaa_sec.c
> @@ -1690,6 +1690,13 @@ dpaa_sec_enqueue_burst(void *qp, struct rte_crypto_op **ops,
>         uint32_t index, flags[DPAA_SEC_BURST] = {0};
>         struct qman_fq *inq[DPAA_SEC_BURST];
>
> +       if (unlikely(!RTE_PER_LCORE(dpaa_io))) {
> +               if (rte_dpaa_portal_init((void *)0)) {
> +                       DPAA_SEC_ERR("Failure in affining portal");
> +                       return 0;
> +               }
> +       }
> +
>         while (nb_ops) {
>                 frames_to_send = (nb_ops > DPAA_SEC_BURST) ?
>                                 DPAA_SEC_BURST : nb_ops;
> @@ -1890,6 +1897,13 @@ dpaa_sec_dequeue_burst(void *qp, struct rte_crypto_op **ops,
>         uint16_t num_rx;
>         struct dpaa_sec_qp *dpaa_qp = (struct dpaa_sec_qp *)qp;
>
> +       if (unlikely(!RTE_PER_LCORE(dpaa_io))) {
> +               if (rte_dpaa_portal_init((void *)0)) {
> +                       DPAA_SEC_ERR("Failure in affining portal");
> +                       return 0;
> +               }
> +       }
> +
>         num_rx = dpaa_sec_deq(dpaa_qp, ops, nb_ops);
>
>         dpaa_qp->rx_pkts += num_rx;
> --
> 2.17.1
>


-- 
Christian Ehrhardt
Staff Engineer, Ubuntu Server
Canonical Ltd