[PATCH v3 01/32] net/cnxk: add eth port specific PTP enable

[Nithin Dabilpuram <ndabilpuram@marvell.com>]

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add support to enable PTP per ethernet device when that
specific ethernet device is connected to event device via
Rx adapter.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Signed-off-by: Nithin Dabilpuram <ndabilpuram@marvell.com>
---
v3:
- Included this patch from series 23650 tp resolve compilation
  dependency.
- Fixed commit header for patch 29/32
- Fixed commit message for patch 15/32

v2: 
- Included this patch from series 24029 as suggested by Jerin to resolve
  compilation dependency with event dev.
- Fixed one-liner commit messages
- Added few more patches for upstream.

 drivers/common/cnxk/roc_io.h             |  5 ++-
 drivers/event/cnxk/cn10k_eventdev.c      |  9 ++---
 drivers/event/cnxk/cn10k_worker.h        | 48 +++++++++++++++---------
 drivers/event/cnxk/cn9k_eventdev.c       | 13 +++----
 drivers/event/cnxk/cn9k_worker.h         | 32 +++++++++++-----
 drivers/event/cnxk/cnxk_eventdev.h       | 14 ++++---
 drivers/event/cnxk/cnxk_eventdev_adptr.c |  9 +++++
 drivers/net/cnxk/cn10k_rx.h              |  3 +-
 8 files changed, 82 insertions(+), 51 deletions(-)

diff --git a/drivers/common/cnxk/roc_io.h b/drivers/common/cnxk/roc_io.h
index 9d73e263f7..13f98ed549 100644
--- a/drivers/common/cnxk/roc_io.h
+++ b/drivers/common/cnxk/roc_io.h
@@ -161,14 +161,15 @@ roc_lmt_mov(void *out, const void *in, const uint32_t lmtext)
 {
 	volatile const __uint128_t *src128 = (const __uint128_t *)in;
 	volatile __uint128_t *dst128 = (__uint128_t *)out;
+	uint32_t i;
 
 	dst128[0] = src128[0];
 	dst128[1] = src128[1];
 	/* lmtext receives following value:
 	 * 1: NIX_SUBDC_EXT needed i.e. tx vlan case
 	 */
-	if (lmtext)
-		dst128[2] = src128[2];
+	for (i = 0; i < lmtext; i++)
+		dst128[2 + i] = src128[2 + i];
 }
 
 static __plt_always_inline void
diff --git a/drivers/event/cnxk/cn10k_eventdev.c b/drivers/event/cnxk/cn10k_eventdev.c
index 5a0cab40a9..0be7ebfe29 100644
--- a/drivers/event/cnxk/cn10k_eventdev.c
+++ b/drivers/event/cnxk/cn10k_eventdev.c
@@ -694,8 +694,7 @@ cn10k_sso_rx_adapter_caps_get(const struct rte_eventdev *event_dev,
 }
 
 static void
-cn10k_sso_set_priv_mem(const struct rte_eventdev *event_dev, void *lookup_mem,
-		       void *tstmp_info)
+cn10k_sso_set_priv_mem(const struct rte_eventdev *event_dev, void *lookup_mem)
 {
 	struct cnxk_sso_evdev *dev = cnxk_sso_pmd_priv(event_dev);
 	int i;
@@ -703,7 +702,7 @@ cn10k_sso_set_priv_mem(const struct rte_eventdev *event_dev, void *lookup_mem,
 	for (i = 0; i < dev->nb_event_ports; i++) {
 		struct cn10k_sso_hws *ws = event_dev->data->ports[i];
 		ws->lookup_mem = lookup_mem;
-		ws->tstamp = tstmp_info;
+		ws->tstamp = dev->tstamp;
 	}
 }
 
@@ -715,7 +714,6 @@ cn10k_sso_rx_adapter_queue_add(
 {
 	struct cn10k_eth_rxq *rxq;
 	void *lookup_mem;
-	void *tstmp_info;
 	int rc;
 
 	rc = strncmp(eth_dev->device->driver->name, "net_cn10k", 8);
@@ -728,8 +726,7 @@ cn10k_sso_rx_adapter_queue_add(
 		return -EINVAL;
 	rxq = eth_dev->data->rx_queues[0];
 	lookup_mem = rxq->lookup_mem;
-	tstmp_info = rxq->tstamp;
-	cn10k_sso_set_priv_mem(event_dev, lookup_mem, tstmp_info);
+	cn10k_sso_set_priv_mem(event_dev, lookup_mem);
 	cn10k_sso_fp_fns_set((struct rte_eventdev *)(uintptr_t)event_dev);
 
 	return 0;
diff --git a/drivers/event/cnxk/cn10k_worker.h b/drivers/event/cnxk/cn10k_worker.h
index 0915f404e0..db56d96404 100644
--- a/drivers/event/cnxk/cn10k_worker.h
+++ b/drivers/event/cnxk/cn10k_worker.h
@@ -108,12 +108,29 @@ cn10k_wqe_to_mbuf(uint64_t wqe, const uint64_t __mbuf, uint8_t port_id,
 			      mbuf_init | ((uint64_t)port_id) << 48, flags);
 }
 
+static void
+cn10k_sso_process_tstamp(uint64_t u64, uint64_t mbuf,
+			 struct cnxk_timesync_info *tstamp)
+{
+	uint64_t tstamp_ptr;
+	uint8_t laptr;
+
+	laptr = (uint8_t) *
+		(uint64_t *)(u64 + (CNXK_SSO_WQE_LAYR_PTR * sizeof(uint64_t)));
+	if (laptr == sizeof(uint64_t)) {
+		/* Extracting tstamp, if PTP enabled*/
+		tstamp_ptr = *(uint64_t *)(((struct nix_wqe_hdr_s *)u64) +
+					   CNXK_SSO_WQE_SG_PTR);
+		cn10k_nix_mbuf_to_tstamp((struct rte_mbuf *)mbuf, tstamp, true,
+					 (uint64_t *)tstamp_ptr);
+	}
+}
+
 static __rte_always_inline void
 cn10k_process_vwqe(uintptr_t vwqe, uint16_t port_id, const uint32_t flags,
 		   void *lookup_mem, void *tstamp, uintptr_t lbase)
 {
-	uint64_t mbuf_init = 0x100010000ULL | RTE_PKTMBUF_HEADROOM |
-			     (flags & NIX_RX_OFFLOAD_TSTAMP_F ? 8 : 0);
+	uint64_t mbuf_init = 0x100010000ULL | RTE_PKTMBUF_HEADROOM;
 	struct rte_event_vector *vec;
 	uint64_t aura_handle, laddr;
 	uint16_t nb_mbufs, non_vec;
@@ -133,6 +150,9 @@ cn10k_process_vwqe(uintptr_t vwqe, uint16_t port_id, const uint32_t flags,
 	for (i = OBJS_PER_CLINE; i < vec->nb_elem; i += OBJS_PER_CLINE)
 		rte_prefetch0(&vec->ptrs[i]);
 
+	if (flags & NIX_RX_OFFLOAD_TSTAMP_F && tstamp)
+		mbuf_init |= 8;
+
 	nb_mbufs = RTE_ALIGN_FLOOR(vec->nb_elem, NIX_DESCS_PER_LOOP);
 	nb_mbufs = cn10k_nix_recv_pkts_vector(&mbuf_init, wqe, nb_mbufs,
 					      flags | NIX_RX_VWQE_F, lookup_mem,
@@ -158,7 +178,6 @@ cn10k_process_vwqe(uintptr_t vwqe, uint16_t port_id, const uint32_t flags,
 
 	while (non_vec) {
 		struct nix_cqe_hdr_s *cqe = (struct nix_cqe_hdr_s *)wqe[0];
-		uint64_t tstamp_ptr;
 
 		mbuf = (struct rte_mbuf *)((char *)cqe -
 					   sizeof(struct rte_mbuf));
@@ -178,12 +197,10 @@ cn10k_process_vwqe(uintptr_t vwqe, uint16_t port_id, const uint32_t flags,
 
 		cn10k_nix_cqe_to_mbuf(cqe, cqe->tag, mbuf, lookup_mem,
 				      mbuf_init, flags);
-		/* Extracting tstamp, if PTP enabled*/
-		tstamp_ptr = *(uint64_t *)(((struct nix_wqe_hdr_s *)cqe) +
-					   CNXK_SSO_WQE_SG_PTR);
-		cn10k_nix_mbuf_to_tstamp((struct rte_mbuf *)mbuf, tstamp,
-					flags & NIX_RX_OFFLOAD_TSTAMP_F,
-					(uint64_t *)tstamp_ptr);
+
+		if (flags & NIX_RX_OFFLOAD_TSTAMP_F)
+			cn10k_sso_process_tstamp((uint64_t)wqe[0],
+						 (uint64_t)mbuf, tstamp);
 		wqe[0] = (struct rte_mbuf *)mbuf;
 		non_vec--;
 		wqe++;
@@ -200,8 +217,6 @@ static __rte_always_inline void
 cn10k_sso_hws_post_process(struct cn10k_sso_hws *ws, uint64_t *u64,
 			   const uint32_t flags)
 {
-	uint64_t tstamp_ptr;
-
 	u64[0] = (u64[0] & (0x3ull << 32)) << 6 |
 		 (u64[0] & (0x3FFull << 36)) << 4 | (u64[0] & 0xffffffff);
 	if ((flags & CPT_RX_WQE_F) &&
@@ -246,12 +261,9 @@ cn10k_sso_hws_post_process(struct cn10k_sso_hws *ws, uint64_t *u64,
 		u64[0] = CNXK_CLR_SUB_EVENT(u64[0]);
 		cn10k_wqe_to_mbuf(u64[1], mbuf, port, u64[0] & 0xFFFFF, flags,
 				  ws->lookup_mem);
-		/* Extracting tstamp, if PTP enabled*/
-		tstamp_ptr = *(uint64_t *)(((struct nix_wqe_hdr_s *)u64[1]) +
-					   CNXK_SSO_WQE_SG_PTR);
-		cn10k_nix_mbuf_to_tstamp((struct rte_mbuf *)mbuf, ws->tstamp,
-					 flags & NIX_RX_OFFLOAD_TSTAMP_F,
-					 (uint64_t *)tstamp_ptr);
+		if (flags & NIX_RX_OFFLOAD_TSTAMP_F)
+			cn10k_sso_process_tstamp(u64[1], mbuf,
+						 ws->tstamp[port]);
 		u64[1] = mbuf;
 	} else if (CNXK_EVENT_TYPE_FROM_TAG(u64[0]) ==
 		   RTE_EVENT_TYPE_ETHDEV_VECTOR) {
@@ -262,7 +274,7 @@ cn10k_sso_hws_post_process(struct cn10k_sso_hws *ws, uint64_t *u64,
 			   ((vwqe_hdr & 0xFFFF) << 48) | ((uint64_t)port << 32);
 		*(uint64_t *)u64[1] = (uint64_t)vwqe_hdr;
 		cn10k_process_vwqe(u64[1], port, flags, ws->lookup_mem,
-				   ws->tstamp, ws->lmt_base);
+				   ws->tstamp[port], ws->lmt_base);
 		/* Mark vector mempool object as get */
 		RTE_MEMPOOL_CHECK_COOKIES(rte_mempool_from_obj((void *)u64[1]),
 					  (void **)&u64[1], 1, 1);
diff --git a/drivers/event/cnxk/cn9k_eventdev.c b/drivers/event/cnxk/cn9k_eventdev.c
index 2e27030049..8ade30f84b 100644
--- a/drivers/event/cnxk/cn9k_eventdev.c
+++ b/drivers/event/cnxk/cn9k_eventdev.c
@@ -123,7 +123,7 @@ cn9k_sso_hws_flush_events(void *hws, uint8_t queue_id, uintptr_t base,
 {
 	struct cnxk_sso_evdev *dev = cnxk_sso_pmd_priv(arg);
 	uint64_t retry = CNXK_SSO_FLUSH_RETRY_MAX;
-	struct cnxk_timesync_info *tstamp;
+	struct cnxk_timesync_info **tstamp;
 	struct cn9k_sso_hws_dual *dws;
 	struct cn9k_sso_hws *ws;
 	uint64_t cq_ds_cnt = 1;
@@ -942,8 +942,7 @@ cn9k_sso_rx_adapter_caps_get(const struct rte_eventdev *event_dev,
 }
 
 static void
-cn9k_sso_set_priv_mem(const struct rte_eventdev *event_dev, void *lookup_mem,
-		      void *tstmp_info)
+cn9k_sso_set_priv_mem(const struct rte_eventdev *event_dev, void *lookup_mem)
 {
 	struct cnxk_sso_evdev *dev = cnxk_sso_pmd_priv(event_dev);
 	int i;
@@ -953,11 +952,11 @@ cn9k_sso_set_priv_mem(const struct rte_eventdev *event_dev, void *lookup_mem,
 			struct cn9k_sso_hws_dual *dws =
 				event_dev->data->ports[i];
 			dws->lookup_mem = lookup_mem;
-			dws->tstamp = tstmp_info;
+			dws->tstamp = dev->tstamp;
 		} else {
 			struct cn9k_sso_hws *ws = event_dev->data->ports[i];
 			ws->lookup_mem = lookup_mem;
-			ws->tstamp = tstmp_info;
+			ws->tstamp = dev->tstamp;
 		}
 	}
 }
@@ -970,7 +969,6 @@ cn9k_sso_rx_adapter_queue_add(
 {
 	struct cn9k_eth_rxq *rxq;
 	void *lookup_mem;
-	void *tstmp_info;
 	int rc;
 
 	rc = strncmp(eth_dev->device->driver->name, "net_cn9k", 8);
@@ -984,8 +982,7 @@ cn9k_sso_rx_adapter_queue_add(
 
 	rxq = eth_dev->data->rx_queues[0];
 	lookup_mem = rxq->lookup_mem;
-	tstmp_info = rxq->tstamp;
-	cn9k_sso_set_priv_mem(event_dev, lookup_mem, tstmp_info);
+	cn9k_sso_set_priv_mem(event_dev, lookup_mem);
 	cn9k_sso_fp_fns_set((struct rte_eventdev *)(uintptr_t)event_dev);
 
 	return 0;
diff --git a/drivers/event/cnxk/cn9k_worker.h b/drivers/event/cnxk/cn9k_worker.h
index 64e97e321a..54b3545022 100644
--- a/drivers/event/cnxk/cn9k_worker.h
+++ b/drivers/event/cnxk/cn9k_worker.h
@@ -169,13 +169,29 @@ cn9k_wqe_to_mbuf(uint64_t wqe, const uint64_t mbuf, uint8_t port_id,
 			     mbuf_init | ((uint64_t)port_id) << 48, flags);
 }
 
+static void
+cn9k_sso_process_tstamp(uint64_t u64, uint64_t mbuf,
+			struct cnxk_timesync_info *tstamp)
+{
+	uint64_t tstamp_ptr;
+	uint8_t laptr;
+
+	laptr = (uint8_t) *
+		(uint64_t *)(u64 + (CNXK_SSO_WQE_LAYR_PTR * sizeof(uint64_t)));
+	if (laptr == sizeof(uint64_t)) {
+		/* Extracting tstamp, if PTP enabled*/
+		tstamp_ptr = *(uint64_t *)(((struct nix_wqe_hdr_s *)u64) +
+					   CNXK_SSO_WQE_SG_PTR);
+		cn9k_nix_mbuf_to_tstamp((struct rte_mbuf *)mbuf, tstamp, true,
+					(uint64_t *)tstamp_ptr);
+	}
+}
+
 static __rte_always_inline void
 cn9k_sso_hws_post_process(uint64_t *u64, uint64_t mbuf, const uint32_t flags,
 			  const void *const lookup_mem,
-			  struct cnxk_timesync_info *tstamp)
+			  struct cnxk_timesync_info **tstamp)
 {
-	uint64_t tstamp_ptr;
-
 	u64[0] = (u64[0] & (0x3ull << 32)) << 6 |
 		 (u64[0] & (0x3FFull << 36)) << 4 | (u64[0] & 0xffffffff);
 	if ((flags & CPT_RX_WQE_F) &&
@@ -187,12 +203,8 @@ cn9k_sso_hws_post_process(uint64_t *u64, uint64_t mbuf, const uint32_t flags,
 		u64[0] = CNXK_CLR_SUB_EVENT(u64[0]);
 		cn9k_wqe_to_mbuf(u64[1], mbuf, port, u64[0] & 0xFFFFF, flags,
 				 lookup_mem);
-		/* Extracting tstamp, if PTP enabled*/
-		tstamp_ptr = *(uint64_t *)(((struct nix_wqe_hdr_s *)u64[1]) +
-					   CNXK_SSO_WQE_SG_PTR);
-		cn9k_nix_mbuf_to_tstamp((struct rte_mbuf *)mbuf, tstamp,
-					flags & NIX_RX_OFFLOAD_TSTAMP_F,
-					(uint64_t *)tstamp_ptr);
+		if (flags & NIX_RX_OFFLOAD_TSTAMP_F)
+			cn9k_sso_process_tstamp(u64[1], mbuf, tstamp[port]);
 		u64[1] = mbuf;
 	}
 }
@@ -298,7 +310,7 @@ cn9k_sso_hws_get_work(struct cn9k_sso_hws *ws, struct rte_event *ev,
 static __rte_always_inline uint16_t
 cn9k_sso_hws_get_work_empty(uint64_t base, struct rte_event *ev,
 			    const uint32_t flags, void *lookup_mem,
-			    struct cnxk_timesync_info *tstamp)
+			    struct cnxk_timesync_info **tstamp)
 {
 	union {
 		__uint128_t get_work;
diff --git a/drivers/event/cnxk/cnxk_eventdev.h b/drivers/event/cnxk/cnxk_eventdev.h
index bfd0c5627e..fae4484758 100644
--- a/drivers/event/cnxk/cnxk_eventdev.h
+++ b/drivers/event/cnxk/cnxk_eventdev.h
@@ -38,6 +38,7 @@
 #define CNXK_SSO_XAQ_CACHE_CNT (0x7)
 #define CNXK_SSO_XAQ_SLACK     (8)
 #define CNXK_SSO_WQE_SG_PTR    (9)
+#define CNXK_SSO_WQE_LAYR_PTR  (5)
 #define CNXK_SSO_PRIORITY_CNT  (0x8)
 #define CNXK_SSO_WEIGHT_MAX    (0x3f)
 #define CNXK_SSO_WEIGHT_MIN    (0x3)
@@ -123,6 +124,7 @@ struct cnxk_sso_evdev {
 	uint64_t *timer_adptr_sz;
 	uint16_t vec_pool_cnt;
 	uint64_t *vec_pools;
+	struct cnxk_timesync_info *tstamp[RTE_MAX_ETHPORTS];
 	struct cnxk_sso_mlt_prio mlt_prio[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	/* Dev args */
 	uint32_t xae_cnt;
@@ -140,12 +142,12 @@ struct cnxk_sso_evdev {
 struct cn10k_sso_hws {
 	uint64_t base;
 	uint64_t gw_rdata;
-	/* PTP timestamp */
-	struct cnxk_timesync_info *tstamp;
 	void *lookup_mem;
 	uint32_t gw_wdata;
 	uint8_t swtag_req;
 	uint8_t hws_id;
+	/* PTP timestamp */
+	struct cnxk_timesync_info **tstamp;
 	/* Add Work Fastpath data */
 	uint64_t xaq_lmt __rte_cache_aligned;
 	uint64_t *fc_mem;
@@ -160,11 +162,11 @@ struct cn10k_sso_hws {
 struct cn9k_sso_hws {
 	uint64_t base;
 	uint64_t gw_wdata;
-	/* PTP timestamp */
-	struct cnxk_timesync_info *tstamp;
 	void *lookup_mem;
 	uint8_t swtag_req;
 	uint8_t hws_id;
+	/* PTP timestamp */
+	struct cnxk_timesync_info **tstamp;
 	/* Add Work Fastpath data */
 	uint64_t xaq_lmt __rte_cache_aligned;
 	uint64_t *fc_mem;
@@ -177,12 +179,12 @@ struct cn9k_sso_hws {
 struct cn9k_sso_hws_dual {
 	uint64_t base[2]; /* Ping and Pong */
 	uint64_t gw_wdata;
-	/* PTP timestamp */
-	struct cnxk_timesync_info *tstamp;
 	void *lookup_mem;
 	uint8_t swtag_req;
 	uint8_t vws; /* Ping pong bit */
 	uint8_t hws_id;
+	/* PTP timestamp */
+	struct cnxk_timesync_info **tstamp;
 	/* Add Work Fastpath data */
 	uint64_t xaq_lmt __rte_cache_aligned;
 	uint64_t *fc_mem;
diff --git a/drivers/event/cnxk/cnxk_eventdev_adptr.c b/drivers/event/cnxk/cnxk_eventdev_adptr.c
index 1f2e1b4b5d..b4fd821912 100644
--- a/drivers/event/cnxk/cnxk_eventdev_adptr.c
+++ b/drivers/event/cnxk/cnxk_eventdev_adptr.c
@@ -207,6 +207,14 @@ cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
 	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
 }
 
+static void
+cnxk_sso_tstamp_cfg(uint16_t port_id, struct cnxk_eth_dev *cnxk_eth_dev,
+		    struct cnxk_sso_evdev *dev)
+{
+	if (cnxk_eth_dev->rx_offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP)
+		dev->tstamp[port_id] = &cnxk_eth_dev->tstamp;
+}
+
 int
 cnxk_sso_rx_adapter_queue_add(
 	const struct rte_eventdev *event_dev, const struct rte_eth_dev *eth_dev,
@@ -255,6 +263,7 @@ cnxk_sso_rx_adapter_queue_add(
 			roc_nix_fc_npa_bp_cfg(&cnxk_eth_dev->nix,
 					      rxq_sp->qconf.mp->pool_id, true,
 					      dev->force_ena_bp, rxq_sp->tc);
+		cnxk_sso_tstamp_cfg(eth_dev->data->port_id, cnxk_eth_dev, dev);
 		cnxk_eth_dev->nb_rxq_sso++;
 	}
 
diff --git a/drivers/net/cnxk/cn10k_rx.h b/drivers/net/cnxk/cn10k_rx.h
index 5ecb20f038..0f8790b8c7 100644
--- a/drivers/net/cnxk/cn10k_rx.h
+++ b/drivers/net/cnxk/cn10k_rx.h
@@ -1567,7 +1567,8 @@ cn10k_nix_recv_pkts_vector(void *args, struct rte_mbuf **mbufs, uint16_t pkts,
 				ol_flags3, mbuf3);
 		}
 
-		if (flags & NIX_RX_OFFLOAD_TSTAMP_F) {
+		if ((flags & NIX_RX_OFFLOAD_TSTAMP_F) &&
+		    ((flags & NIX_RX_VWQE_F) && tstamp)) {
 			const uint16x8_t len_off = {
 				0,			     /* ptype   0:15 */
 				0,			     /* ptype  16:32 */
-- 
2.25.1