[v4] net/octeontx2: add devargs to lock Rx/Tx ctx
diff mbox series

Message ID 20200327095359.1934-1-pbhagavatula@marvell.com
State Superseded, archived
Delegated to: Jerin Jacob
Headers show
Series
  • [v4] net/octeontx2: add devargs to lock Rx/Tx ctx
Related show

Checks

Context Check Description
ci/checkpatch success coding style OK

Commit Message

Pavan Nikhilesh Bhagavatula March 27, 2020, 9:53 a.m. UTC
From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add device arguments to lock Rx/Tx contexts.
Application can either choose to lock Rx or Tx contexts by using
'lock_rx_ctx' or 'lock_tx_ctx' respectively per each port.

Example:
	-w 0002:02:00.0,lock_rx_ctx=1 -w 0002:03:00.0,lock_tx_ctx=1

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Reviewed-by: Andrzej Ostruszka <aostruszka@marvell.com>
---
 Depends on http://patches.dpdk.org/patch/67178/
 v4 Changes:
 - Fix return path using unnecessary goto.(Andrzej)
 - Fix datatype of values passed to devargs parser.(Andrzej)

 v3 Changes:
 - Split series into individual patches as targets are different.

 doc/guides/nics/octeontx2.rst               |  16 ++
 drivers/net/octeontx2/otx2_ethdev.c         | 187 +++++++++++++++++++-
 drivers/net/octeontx2/otx2_ethdev.h         |   2 +
 drivers/net/octeontx2/otx2_ethdev_devargs.c |  16 +-
 drivers/net/octeontx2/otx2_rss.c            |  23 +++
 5 files changed, 241 insertions(+), 3 deletions(-)

--
2.17.1

Comments

Andrzej Ostruszka March 27, 2020, 4:19 p.m. UTC | #1
On 3/27/20 10:53 AM, pbhagavatula@marvell.com wrote:
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Add device arguments to lock Rx/Tx contexts.
> Application can either choose to lock Rx or Tx contexts by using
> 'lock_rx_ctx' or 'lock_tx_ctx' respectively per each port.
> 
> Example:
> 	-w 0002:02:00.0,lock_rx_ctx=1 -w 0002:03:00.0,lock_tx_ctx=1
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Reviewed-by: Andrzej Ostruszka <aostruszka@marvell.com>
> ---
>  Depends on http://patches.dpdk.org/patch/67178/
>  v4 Changes:
>  - Fix return path using unnecessary goto.(Andrzej)
>  - Fix datatype of values passed to devargs parser.(Andrzej)
> 
>  v3 Changes:
>  - Split series into individual patches as targets are different.
> 
>  doc/guides/nics/octeontx2.rst               |  16 ++
>  drivers/net/octeontx2/otx2_ethdev.c         | 187 +++++++++++++++++++-
>  drivers/net/octeontx2/otx2_ethdev.h         |   2 +
>  drivers/net/octeontx2/otx2_ethdev_devargs.c |  16 +-
>  drivers/net/octeontx2/otx2_rss.c            |  23 +++
>  5 files changed, 241 insertions(+), 3 deletions(-)
[...]
> diff --git a/drivers/net/octeontx2/otx2_ethdev.c b/drivers/net/octeontx2/otx2_ethdev.c
> index e60f4901c..a6f2c0f42 100644
> --- a/drivers/net/octeontx2/otx2_ethdev.c
> +++ b/drivers/net/octeontx2/otx2_ethdev.c
> @@ -381,6 +381,40 @@ nix_cq_rq_init(struct rte_eth_dev *eth_dev, struct otx2_eth_dev *dev,
>  		goto fail;
>  	}
> 
> +	if (dev->lock_rx_ctx) {
> +		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
> +		aq->qidx = qid;
> +		aq->ctype = NIX_AQ_CTYPE_CQ;
> +		aq->op = NIX_AQ_INSTOP_LOCK;
> +
> +		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
> +		if (!aq) {
> +			/* The shared memory buffer can be full.
> +			 * Flush it and retry
> +			 */
> +			otx2_mbox_msg_send(mbox, 0);
> +			rc = otx2_mbox_wait_for_rsp(mbox, 0);
> +			if (rc < 0) {
> +				otx2_err("Failed to LOCK cq context");
> +				return rc;
> +			}
> +
> +			aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
> +			if (!aq) {
> +				otx2_err("Failed to LOCK rq context");
> +				return -ENOMEM;
> +			}
> +		}
> +		aq->qidx = qid;
> +		aq->ctype = NIX_AQ_CTYPE_RQ;
> +		aq->op = NIX_AQ_INSTOP_LOCK;
> +		rc = otx2_mbox_process(mbox);
> +		if (rc < 0) {
> +			otx2_err("Failed to LOCK rq context");
> +			return rc;
> +		}
> +	}
> +
>  	return 0;
>  fail:
>  	return rc;

Pavan - sorry for being so ... finicky :)
I've said 'replace all "goto fail" with "return rc"' and I meant that.
So not only the "goto fail" in you changes but all "goto fail" in that
function.

Apart from that:
Reviewed-by: Andrzej Ostruszka <aostruszka@marvell.com>

With regards
Andrzej

PS. Thanks for the patience ;)
Pavan Nikhilesh Bhagavatula March 27, 2020, 5:49 p.m. UTC | #2
<Snip>
>>  fail:
>>  	return rc;
>
>Pavan - sorry for being so ... finicky :)
>I've said 'replace all "goto fail" with "return rc"' and I meant that.
>So not only the "goto fail" in you changes but all "goto fail" in that
>function.

Ah, sure I will send a v5.

>
>Apart from that:
>Reviewed-by: Andrzej Ostruszka <aostruszka@marvell.com>
>
>With regards
>Andrzej
>
>PS. Thanks for the patience ;)

😊

Patch
diff mbox series

diff --git a/doc/guides/nics/octeontx2.rst b/doc/guides/nics/octeontx2.rst
index c2d87c9d0..df19443e3 100644
--- a/doc/guides/nics/octeontx2.rst
+++ b/doc/guides/nics/octeontx2.rst
@@ -209,6 +209,22 @@  Runtime Config Options
    With the above configuration, application can enable inline IPsec processing
    on 128 SAs (SPI 0-127).

+- ``Lock Rx contexts in NDC cache``
+
+   Lock Rx contexts in NDC cache by using ``lock_rx_ctx`` parameter.
+
+   For example::
+
+      -w 0002:02:00.0,lock_rx_ctx=1
+
+- ``Lock Tx contexts in NDC cache``
+
+   Lock Tx contexts in NDC cache by using ``lock_tx_ctx`` parameter.
+
+   For example::
+
+      -w 0002:02:00.0,lock_tx_ctx=1
+
 .. note::

    Above devarg parameters are configurable per device, user needs to pass the
diff --git a/drivers/net/octeontx2/otx2_ethdev.c b/drivers/net/octeontx2/otx2_ethdev.c
index e60f4901c..a6f2c0f42 100644
--- a/drivers/net/octeontx2/otx2_ethdev.c
+++ b/drivers/net/octeontx2/otx2_ethdev.c
@@ -381,6 +381,40 @@  nix_cq_rq_init(struct rte_eth_dev *eth_dev, struct otx2_eth_dev *dev,
 		goto fail;
 	}

+	if (dev->lock_rx_ctx) {
+		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
+		aq->qidx = qid;
+		aq->ctype = NIX_AQ_CTYPE_CQ;
+		aq->op = NIX_AQ_INSTOP_LOCK;
+
+		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
+		if (!aq) {
+			/* The shared memory buffer can be full.
+			 * Flush it and retry
+			 */
+			otx2_mbox_msg_send(mbox, 0);
+			rc = otx2_mbox_wait_for_rsp(mbox, 0);
+			if (rc < 0) {
+				otx2_err("Failed to LOCK cq context");
+				return rc;
+			}
+
+			aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
+			if (!aq) {
+				otx2_err("Failed to LOCK rq context");
+				return -ENOMEM;
+			}
+		}
+		aq->qidx = qid;
+		aq->ctype = NIX_AQ_CTYPE_RQ;
+		aq->op = NIX_AQ_INSTOP_LOCK;
+		rc = otx2_mbox_process(mbox);
+		if (rc < 0) {
+			otx2_err("Failed to LOCK rq context");
+			return rc;
+		}
+	}
+
 	return 0;
 fail:
 	return rc;
@@ -430,6 +464,40 @@  nix_cq_rq_uninit(struct rte_eth_dev *eth_dev, struct otx2_eth_rxq *rxq)
 		return rc;
 	}

+	if (dev->lock_rx_ctx) {
+		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
+		aq->qidx = rxq->rq;
+		aq->ctype = NIX_AQ_CTYPE_CQ;
+		aq->op = NIX_AQ_INSTOP_UNLOCK;
+
+		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
+		if (!aq) {
+			/* The shared memory buffer can be full.
+			 * Flush it and retry
+			 */
+			otx2_mbox_msg_send(mbox, 0);
+			rc = otx2_mbox_wait_for_rsp(mbox, 0);
+			if (rc < 0) {
+				otx2_err("Failed to UNLOCK cq context");
+				return rc;
+			}
+
+			aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
+			if (!aq) {
+				otx2_err("Failed to UNLOCK rq context");
+				return -ENOMEM;
+			}
+		}
+		aq->qidx = rxq->rq;
+		aq->ctype = NIX_AQ_CTYPE_RQ;
+		aq->op = NIX_AQ_INSTOP_UNLOCK;
+		rc = otx2_mbox_process(mbox);
+		if (rc < 0) {
+			otx2_err("Failed to UNLOCK rq context");
+			return rc;
+		}
+	}
+
 	return 0;
 }

@@ -715,6 +783,94 @@  nix_tx_offload_flags(struct rte_eth_dev *eth_dev)
 	return flags;
 }

+static int
+nix_sqb_lock(struct rte_mempool *mp)
+{
+	struct otx2_npa_lf *npa_lf = otx2_intra_dev_get_cfg()->npa_lf;
+	struct npa_aq_enq_req *req;
+	int rc;
+
+	req = otx2_mbox_alloc_msg_npa_aq_enq(npa_lf->mbox);
+	req->aura_id = npa_lf_aura_handle_to_aura(mp->pool_id);
+	req->ctype = NPA_AQ_CTYPE_AURA;
+	req->op = NPA_AQ_INSTOP_LOCK;
+
+	req = otx2_mbox_alloc_msg_npa_aq_enq(npa_lf->mbox);
+	if (!req) {
+		/* The shared memory buffer can be full.
+		 * Flush it and retry
+		 */
+		otx2_mbox_msg_send(npa_lf->mbox, 0);
+		rc = otx2_mbox_wait_for_rsp(npa_lf->mbox, 0);
+		if (rc < 0) {
+			otx2_err("Failed to LOCK AURA context");
+			return rc;
+		}
+
+		req = otx2_mbox_alloc_msg_npa_aq_enq(npa_lf->mbox);
+		if (!req) {
+			otx2_err("Failed to LOCK POOL context");
+			return -ENOMEM;
+		}
+	}
+
+	req->aura_id = npa_lf_aura_handle_to_aura(mp->pool_id);
+	req->ctype = NPA_AQ_CTYPE_POOL;
+	req->op = NPA_AQ_INSTOP_LOCK;
+
+	rc = otx2_mbox_process(npa_lf->mbox);
+	if (rc < 0) {
+		otx2_err("Unable to lock POOL in NDC");
+		return rc;
+	}
+
+	return 0;
+}
+
+static int
+nix_sqb_unlock(struct rte_mempool *mp)
+{
+	struct otx2_npa_lf *npa_lf = otx2_intra_dev_get_cfg()->npa_lf;
+	struct npa_aq_enq_req *req;
+	int rc;
+
+	req = otx2_mbox_alloc_msg_npa_aq_enq(npa_lf->mbox);
+	req->aura_id = npa_lf_aura_handle_to_aura(mp->pool_id);
+	req->ctype = NPA_AQ_CTYPE_AURA;
+	req->op = NPA_AQ_INSTOP_UNLOCK;
+
+	req = otx2_mbox_alloc_msg_npa_aq_enq(npa_lf->mbox);
+	if (!req) {
+		/* The shared memory buffer can be full.
+		 * Flush it and retry
+		 */
+		otx2_mbox_msg_send(npa_lf->mbox, 0);
+		rc = otx2_mbox_wait_for_rsp(npa_lf->mbox, 0);
+		if (rc < 0) {
+			otx2_err("Failed to UNLOCK AURA context");
+			return rc;
+		}
+
+		req = otx2_mbox_alloc_msg_npa_aq_enq(npa_lf->mbox);
+		if (!req) {
+			otx2_err("Failed to UNLOCK POOL context");
+			return -ENOMEM;
+		}
+	}
+	req = otx2_mbox_alloc_msg_npa_aq_enq(npa_lf->mbox);
+	req->aura_id = npa_lf_aura_handle_to_aura(mp->pool_id);
+	req->ctype = NPA_AQ_CTYPE_POOL;
+	req->op = NPA_AQ_INSTOP_UNLOCK;
+
+	rc = otx2_mbox_process(npa_lf->mbox);
+	if (rc < 0) {
+		otx2_err("Unable to UNLOCK AURA in NDC");
+		return rc;
+	}
+
+	return 0;
+}
+
 static int
 nix_sq_init(struct otx2_eth_txq *txq)
 {
@@ -757,7 +913,20 @@  nix_sq_init(struct otx2_eth_txq *txq)
 	/* Many to one reduction */
 	sq->sq.qint_idx = txq->sq % dev->qints;

-	return otx2_mbox_process(mbox);
+	rc = otx2_mbox_process(mbox);
+	if (rc < 0)
+		return rc;
+
+	if (dev->lock_tx_ctx) {
+		sq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
+		sq->qidx = txq->sq;
+		sq->ctype = NIX_AQ_CTYPE_SQ;
+		sq->op = NIX_AQ_INSTOP_LOCK;
+
+		rc = otx2_mbox_process(mbox);
+	}
+
+	return rc;
 }

 static int
@@ -800,6 +969,20 @@  nix_sq_uninit(struct otx2_eth_txq *txq)
 	if (rc)
 		return rc;

+	if (dev->lock_tx_ctx) {
+		/* Unlock sq */
+		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
+		aq->qidx = txq->sq;
+		aq->ctype = NIX_AQ_CTYPE_SQ;
+		aq->op = NIX_AQ_INSTOP_UNLOCK;
+
+		rc = otx2_mbox_process(mbox);
+		if (rc < 0)
+			return rc;
+
+		nix_sqb_unlock(txq->sqb_pool);
+	}
+
 	/* Read SQ and free sqb's */
 	aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
 	aq->qidx = txq->sq;
@@ -921,6 +1104,8 @@  nix_alloc_sqb_pool(int port, struct otx2_eth_txq *txq, uint16_t nb_desc)
 	}

 	nix_sqb_aura_limit_cfg(txq->sqb_pool, txq->nb_sqb_bufs);
+	if (dev->lock_tx_ctx)
+		nix_sqb_lock(txq->sqb_pool);

 	return 0;
 fail:
diff --git a/drivers/net/octeontx2/otx2_ethdev.h b/drivers/net/octeontx2/otx2_ethdev.h
index e5684f9f0..90ca8cbed 100644
--- a/drivers/net/octeontx2/otx2_ethdev.h
+++ b/drivers/net/octeontx2/otx2_ethdev.h
@@ -272,6 +272,8 @@  struct otx2_eth_dev {
 	uint8_t max_mac_entries;
 	uint8_t lf_tx_stats;
 	uint8_t lf_rx_stats;
+	uint8_t lock_rx_ctx;
+	uint8_t lock_tx_ctx;
 	uint16_t flags;
 	uint16_t cints;
 	uint16_t qints;
diff --git a/drivers/net/octeontx2/otx2_ethdev_devargs.c b/drivers/net/octeontx2/otx2_ethdev_devargs.c
index 5390eb217..b0480504c 100644
--- a/drivers/net/octeontx2/otx2_ethdev_devargs.c
+++ b/drivers/net/octeontx2/otx2_ethdev_devargs.c
@@ -124,6 +124,8 @@  parse_switch_header_type(const char *key, const char *value, void *extra_args)
 #define OTX2_FLOW_MAX_PRIORITY "flow_max_priority"
 #define OTX2_SWITCH_HEADER_TYPE "switch_header"
 #define OTX2_RSS_TAG_AS_XOR "tag_as_xor"
+#define OTX2_LOCK_RX_CTX "lock_rx_ctx"
+#define OTX2_LOCK_TX_CTX "lock_tx_ctx"

 int
 otx2_ethdev_parse_devargs(struct rte_devargs *devargs, struct otx2_eth_dev *dev)
@@ -134,9 +136,11 @@  otx2_ethdev_parse_devargs(struct rte_devargs *devargs, struct otx2_eth_dev *dev)
 	uint16_t switch_header_type = 0;
 	uint16_t flow_max_priority = 3;
 	uint16_t ipsec_in_max_spi = 1;
-	uint16_t scalar_enable = 0;
 	uint16_t rss_tag_as_xor = 0;
+	uint16_t scalar_enable = 0;
 	struct rte_kvargs *kvlist;
+	uint16_t lock_rx_ctx = 0;
+	uint16_t lock_tx_ctx = 0;

 	if (devargs == NULL)
 		goto null_devargs;
@@ -161,6 +165,10 @@  otx2_ethdev_parse_devargs(struct rte_devargs *devargs, struct otx2_eth_dev *dev)
 			   &parse_switch_header_type, &switch_header_type);
 	rte_kvargs_process(kvlist, OTX2_RSS_TAG_AS_XOR,
 			   &parse_flag, &rss_tag_as_xor);
+	rte_kvargs_process(kvlist, OTX2_LOCK_RX_CTX,
+			   &parse_flag, &lock_rx_ctx);
+	rte_kvargs_process(kvlist, OTX2_LOCK_TX_CTX,
+			   &parse_flag, &lock_tx_ctx);
 	otx2_parse_common_devargs(kvlist);
 	rte_kvargs_free(kvlist);

@@ -169,6 +177,8 @@  otx2_ethdev_parse_devargs(struct rte_devargs *devargs, struct otx2_eth_dev *dev)
 	dev->scalar_ena = scalar_enable;
 	dev->rss_tag_as_xor = rss_tag_as_xor;
 	dev->max_sqb_count = sqb_count;
+	dev->lock_rx_ctx = lock_rx_ctx;
+	dev->lock_tx_ctx = lock_tx_ctx;
 	dev->rss_info.rss_size = rss_size;
 	dev->npc_flow.flow_prealloc_size = flow_prealloc_size;
 	dev->npc_flow.flow_max_priority = flow_max_priority;
@@ -188,4 +198,6 @@  RTE_PMD_REGISTER_PARAM_STRING(net_octeontx2,
 			      OTX2_FLOW_MAX_PRIORITY "=<1-32>"
 			      OTX2_SWITCH_HEADER_TYPE "=<higig2|dsa>"
 			      OTX2_RSS_TAG_AS_XOR "=1"
-			      OTX2_NPA_LOCK_MASK "=<1-65535>");
+			      OTX2_NPA_LOCK_MASK "=<1-65535>"
+			      OTX2_LOCK_RX_CTX "=1"
+			      OTX2_LOCK_TX_CTX "=1");
diff --git a/drivers/net/octeontx2/otx2_rss.c b/drivers/net/octeontx2/otx2_rss.c
index 7a8c8f3de..34005ef02 100644
--- a/drivers/net/octeontx2/otx2_rss.c
+++ b/drivers/net/octeontx2/otx2_rss.c
@@ -33,6 +33,29 @@  otx2_nix_rss_tbl_init(struct otx2_eth_dev *dev,
 		req->qidx = (group * rss->rss_size) + idx;
 		req->ctype = NIX_AQ_CTYPE_RSS;
 		req->op = NIX_AQ_INSTOP_INIT;
+
+		if (!dev->lock_rx_ctx)
+			continue;
+
+		req = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
+		if (!req) {
+			/* The shared memory buffer can be full.
+			 * Flush it and retry
+			 */
+			otx2_mbox_msg_send(mbox, 0);
+			rc = otx2_mbox_wait_for_rsp(mbox, 0);
+			if (rc < 0)
+				return rc;
+
+			req = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
+			if (!req)
+				return -ENOMEM;
+		}
+		req->rss.rq = ind_tbl[idx];
+		/* Fill AQ info */
+		req->qidx = (group * rss->rss_size) + idx;
+		req->ctype = NIX_AQ_CTYPE_RSS;
+		req->op = NIX_AQ_INSTOP_LOCK;
 	}

 	otx2_mbox_msg_send(mbox, 0);