[dpdk-dev] [PATCH v2 4/4] event/octeontx: support Rx Tx checksum offload

Harman Kalra hkalra at marvell.com
Tue Apr 28 14:40:12 CEST 2020


Adding support for rx checksum offload. In case of wrong
checksum received (inner/outer l3/l4) it reports the
corresponding layer which has bad checksum. It also adds
rx burst function pointer hook for rx checksum offload to
event PMD.

Signed-off-by: Harman Kalra <hkalra at marvell.com>
---
 drivers/event/octeontx/ssovf_evdev.c  |   1 +
 drivers/event/octeontx/ssovf_evdev.h  |   2 +
 drivers/event/octeontx/ssovf_worker.c | 127 ++++++++++++++++++++++----
 drivers/event/octeontx/ssovf_worker.h |  44 ++++++++-
 drivers/net/octeontx/octeontx_rxtx.h  |  17 +++-
 5 files changed, 168 insertions(+), 23 deletions(-)

diff --git a/drivers/event/octeontx/ssovf_evdev.c b/drivers/event/octeontx/ssovf_evdev.c
index 5d074bcbc..1b1a5d939 100644
--- a/drivers/event/octeontx/ssovf_evdev.c
+++ b/drivers/event/octeontx/ssovf_evdev.c
@@ -272,6 +272,7 @@ ssovf_port_setup(struct rte_eventdev *dev, uint8_t port_id,
 	reg_off |= 1 << 16; /* Wait */
 	ws->getwork = ws->base + reg_off;
 	ws->port = port_id;
+	ws->lookup_mem = octeontx_fastpath_lookup_mem_get();
 
 	for (q = 0; q < edev->nb_event_queues; q++) {
 		ws->grps[q] = ssovf_bar(OCTEONTX_SSO_GROUP, q, 2);
diff --git a/drivers/event/octeontx/ssovf_evdev.h b/drivers/event/octeontx/ssovf_evdev.h
index 1f5066c9a..aa5acf246 100644
--- a/drivers/event/octeontx/ssovf_evdev.h
+++ b/drivers/event/octeontx/ssovf_evdev.h
@@ -157,6 +157,7 @@ struct ssows {
 	uint8_t *getwork;
 	uint8_t *grps[SSO_MAX_VHGRP];
 	uint8_t port;
+	void *lookup_mem;
 } __rte_cache_aligned;
 
 static inline struct ssovf_evdev *
@@ -182,5 +183,6 @@ int ssovf_info(struct ssovf_info *info);
 void *ssovf_bar(enum ssovf_type, uint8_t id, uint8_t bar);
 int test_eventdev_octeontx(void);
 void ssovf_fastpath_fns_set(struct rte_eventdev *dev);
+void *octeontx_fastpath_lookup_mem_get(void);
 
 #endif /* __SSOVF_EVDEV_H__ */
diff --git a/drivers/event/octeontx/ssovf_worker.c b/drivers/event/octeontx/ssovf_worker.c
index a276269d7..d2d5eea8f 100644
--- a/drivers/event/octeontx/ssovf_worker.c
+++ b/drivers/event/octeontx/ssovf_worker.c
@@ -91,7 +91,7 @@ ssows_release_event(struct ssows *ws)
 		ssows_swtag_untag(ws);
 }
 
-#define R(name, f1, f0, flags)						     \
+#define R(name, f2, f1, f0, flags)					     \
 static uint16_t __rte_noinline	__rte_hot				     \
 ssows_deq_ ##name(void *port, struct rte_event *ev, uint64_t timeout_ticks)  \
 {									     \
@@ -238,7 +238,8 @@ ssows_flush_events(struct ssows *ws, uint8_t queue_id,
 			ev.mbuf = ssovf_octeontx_wqe_to_pkt(get_work1,
 					(ev.event >> 20) & 0x7F,
 					OCCTX_RX_OFFLOAD_NONE |
-					OCCTX_RX_MULTI_SEG_F);
+					OCCTX_RX_MULTI_SEG_F,
+					ws->lookup_mem);
 		else
 			ev.u64 = get_work1;
 
@@ -340,16 +341,16 @@ SSO_TX_ADPTR_ENQ_FASTPATH_FUNC
 
 	dev->txa_enqueue = ssow_txa_enqueue
 		[!!(edev->tx_offload_flags & OCCTX_TX_OFFLOAD_MBUF_NOFF_F)]
-		[0]
-		[0]
+		[!!(edev->tx_offload_flags & OCCTX_TX_OFFLOAD_OL3_OL4_CSUM_F)]
+		[!!(edev->tx_offload_flags & OCCTX_TX_OFFLOAD_L3_L4_CSUM_F)]
 		[!!(edev->tx_offload_flags & OCCTX_TX_MULTI_SEG_F)];
 
 	dev->txa_enqueue_same_dest = dev->txa_enqueue;
 
 	/* Assigning dequeue func pointers */
-	const event_dequeue_t ssow_deq[2][2] = {
-#define R(name, f1, f0, flags)					\
-	[f1][f0] =  ssows_deq_ ##name,
+	const event_dequeue_t ssow_deq[2][2][2] = {
+#define R(name, f2, f1, f0, flags)					\
+	[f2][f1][f0] =  ssows_deq_ ##name,
 
 SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
 #undef R
@@ -357,11 +358,12 @@ SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
 
 	dev->dequeue = ssow_deq
 		[!!(edev->rx_offload_flags & OCCTX_RX_VLAN_FLTR_F)]
+		[!!(edev->rx_offload_flags & OCCTX_RX_OFFLOAD_CSUM_F)]
 		[!!(edev->rx_offload_flags & OCCTX_RX_MULTI_SEG_F)];
 
-	const event_dequeue_burst_t ssow_deq_burst[2][2] = {
-#define R(name, f1, f0, flags)						\
-	[f1][f0] =  ssows_deq_burst_ ##name,
+	const event_dequeue_burst_t ssow_deq_burst[2][2][2] = {
+#define R(name, f2, f1, f0, flags)					\
+	[f2][f1][f0] =  ssows_deq_burst_ ##name,
 
 SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
 #undef R
@@ -369,12 +371,13 @@ SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
 
 	dev->dequeue_burst = ssow_deq_burst
 		[!!(edev->rx_offload_flags & OCCTX_RX_VLAN_FLTR_F)]
+		[!!(edev->rx_offload_flags & OCCTX_RX_OFFLOAD_CSUM_F)]
 		[!!(edev->rx_offload_flags & OCCTX_RX_MULTI_SEG_F)];
 
 	if (edev->is_timeout_deq) {
-		const event_dequeue_t ssow_deq_timeout[2][2] = {
-#define R(name, f1, f0, flags)						\
-	[f1][f0] =  ssows_deq_timeout_ ##name,
+		const event_dequeue_t ssow_deq_timeout[2][2][2] = {
+#define R(name, f2, f1, f0, flags)					\
+	[f2][f1][f0] =  ssows_deq_timeout_ ##name,
 
 SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
 #undef R
@@ -382,11 +385,12 @@ SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
 
 	dev->dequeue = ssow_deq_timeout
 		[!!(edev->rx_offload_flags & OCCTX_RX_VLAN_FLTR_F)]
+		[!!(edev->rx_offload_flags & OCCTX_RX_OFFLOAD_CSUM_F)]
 		[!!(edev->rx_offload_flags & OCCTX_RX_MULTI_SEG_F)];
 
-	const event_dequeue_burst_t ssow_deq_timeout_burst[2][2] = {
-#define R(name, f1, f0, flags)						\
-	[f1][f0] =  ssows_deq_timeout_burst_ ##name,
+	const event_dequeue_burst_t ssow_deq_timeout_burst[2][2][2] = {
+#define R(name, f2, f1, f0, flags)					\
+	[f2][f1][f0] =  ssows_deq_timeout_burst_ ##name,
 
 SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
 #undef R
@@ -394,6 +398,97 @@ SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
 
 	dev->dequeue_burst = ssow_deq_timeout_burst
 		[!!(edev->rx_offload_flags & OCCTX_RX_VLAN_FLTR_F)]
+		[!!(edev->rx_offload_flags & OCCTX_RX_OFFLOAD_CSUM_F)]
 		[!!(edev->rx_offload_flags & OCCTX_RX_MULTI_SEG_F)];
 	}
 }
+
+static void
+octeontx_create_rx_ol_flags_array(void *mem)
+{
+	uint16_t idx, errcode, errlev;
+	uint32_t val, *ol_flags;
+
+	/* Skip ptype array memory */
+	ol_flags = (uint32_t *)mem;
+
+	for (idx = 0; idx < BIT(ERRCODE_ERRLEN_WIDTH); idx++) {
+		errcode = idx & 0xff;
+		errlev = (idx & 0x700) >> 8;
+
+		val = PKT_RX_IP_CKSUM_UNKNOWN;
+		val |= PKT_RX_L4_CKSUM_UNKNOWN;
+		val |= PKT_RX_OUTER_L4_CKSUM_UNKNOWN;
+
+		switch (errlev) {
+		case OCCTX_ERRLEV_RE:
+			if (errcode) {
+				val |= PKT_RX_IP_CKSUM_BAD;
+				val |= PKT_RX_L4_CKSUM_BAD;
+			} else {
+				val |= PKT_RX_IP_CKSUM_GOOD;
+				val |= PKT_RX_L4_CKSUM_GOOD;
+			}
+			break;
+		case OCCTX_ERRLEV_LC:
+			if (errcode == OCCTX_EC_IP4_CSUM) {
+				val |= PKT_RX_IP_CKSUM_BAD;
+				val |= PKT_RX_EIP_CKSUM_BAD;
+			} else {
+				val |= PKT_RX_IP_CKSUM_GOOD;
+			}
+			break;
+		case OCCTX_ERRLEV_LD:
+			/* Check if parsed packet is neither IPv4 or IPV6 */
+			if (errcode == OCCTX_EC_IP4_NOT)
+				break;
+			val |= PKT_RX_IP_CKSUM_GOOD;
+			if (errcode == OCCTX_EC_L4_CSUM)
+				val |= PKT_RX_OUTER_L4_CKSUM_BAD;
+			else
+				val |= PKT_RX_L4_CKSUM_GOOD;
+			break;
+		case OCCTX_ERRLEV_LE:
+			if (errcode == OCCTX_EC_IP4_CSUM)
+				val |= PKT_RX_IP_CKSUM_BAD;
+			else
+				val |= PKT_RX_IP_CKSUM_GOOD;
+			break;
+		case OCCTX_ERRLEV_LF:
+			/* Check if parsed packet is neither IPv4 or IPV6 */
+			if (errcode == OCCTX_EC_IP4_NOT)
+				break;
+			val |= PKT_RX_IP_CKSUM_GOOD;
+			if (errcode == OCCTX_EC_L4_CSUM)
+				val |= PKT_RX_L4_CKSUM_BAD;
+			else
+				val |= PKT_RX_L4_CKSUM_GOOD;
+			break;
+		}
+
+		ol_flags[idx] = val;
+	}
+}
+
+void *
+octeontx_fastpath_lookup_mem_get(void)
+{
+	const char name[] = OCCTX_FASTPATH_LOOKUP_MEM;
+	const struct rte_memzone *mz;
+	void *mem;
+
+	mz = rte_memzone_lookup(name);
+	if (mz != NULL)
+		return mz->addr;
+
+	/* Request for the first time */
+	mz = rte_memzone_reserve_aligned(name, LOOKUP_ARRAY_SZ,
+					 SOCKET_ID_ANY, 0, OCCTX_ALIGN);
+	if (mz != NULL) {
+		mem = mz->addr;
+		/* Form the rx ol_flags based on errcode */
+		octeontx_create_rx_ol_flags_array(mem);
+		return mem;
+	}
+	return NULL;
+}
diff --git a/drivers/event/octeontx/ssovf_worker.h b/drivers/event/octeontx/ssovf_worker.h
index b2bcc27c3..99a0726c2 100644
--- a/drivers/event/octeontx/ssovf_worker.h
+++ b/drivers/event/octeontx/ssovf_worker.h
@@ -10,6 +10,34 @@
 #include "ssovf_evdev.h"
 #include "octeontx_rxtx.h"
 
+/* Alignment */
+#define OCCTX_ALIGN  128
+
+/* Fastpath lookup */
+#define OCCTX_FASTPATH_LOOKUP_MEM	"octeontx_fastpath_lookup_mem"
+
+/* WQE's ERRCODE + ERRLEV (11 bits) */
+#define ERRCODE_ERRLEN_WIDTH		11
+#define ERR_ARRAY_SZ			((BIT(ERRCODE_ERRLEN_WIDTH)) *\
+					sizeof(uint32_t))
+
+#define LOOKUP_ARRAY_SZ			(ERR_ARRAY_SZ)
+
+#define OCCTX_EC_IP4_NOT		0x41
+#define OCCTX_EC_IP4_CSUM		0x42
+#define OCCTX_EC_L4_CSUM		0x62
+
+enum OCCTX_ERRLEV_E {
+	OCCTX_ERRLEV_RE = 0,
+	OCCTX_ERRLEV_LA = 1,
+	OCCTX_ERRLEV_LB = 2,
+	OCCTX_ERRLEV_LC = 3,
+	OCCTX_ERRLEV_LD = 4,
+	OCCTX_ERRLEV_LE = 5,
+	OCCTX_ERRLEV_LF = 6,
+	OCCTX_ERRLEV_LG = 7,
+};
+
 enum {
 	SSO_SYNC_ORDERED,
 	SSO_SYNC_ATOMIC,
@@ -19,6 +47,14 @@ enum {
 
 /* SSO Operations */
 
+static __rte_always_inline uint32_t
+ssovf_octeontx_rx_olflags_get(const void * const lookup_mem, const uint64_t in)
+{
+	const uint32_t * const ol_flags = (const uint32_t *)lookup_mem;
+
+	return ol_flags[(in & 0x7ff)];
+}
+
 static __rte_always_inline void
 ssovf_octeontx_wqe_xtract_mseg(octtx_wqe_t *wqe,
 			       struct rte_mbuf *mbuf)
@@ -57,7 +93,7 @@ ssovf_octeontx_wqe_xtract_mseg(octtx_wqe_t *wqe,
 
 static __rte_always_inline struct rte_mbuf *
 ssovf_octeontx_wqe_to_pkt(uint64_t work, uint16_t port_info,
-			  const uint16_t flag)
+			  const uint16_t flag, const void *lookup_mem)
 {
 	struct rte_mbuf *mbuf;
 	octtx_wqe_t *wqe = (octtx_wqe_t *)(uintptr_t)work;
@@ -71,6 +107,10 @@ ssovf_octeontx_wqe_to_pkt(uint64_t work, uint16_t port_info,
 	mbuf->ol_flags = 0;
 	mbuf->pkt_len = wqe->s.w1.len;
 
+	if (!!(flag & OCCTX_RX_OFFLOAD_CSUM_F))
+		mbuf->ol_flags = ssovf_octeontx_rx_olflags_get(lookup_mem,
+							       wqe->w[2]);
+
 	if (!!(flag & OCCTX_RX_MULTI_SEG_F)) {
 		mbuf->nb_segs = wqe->s.w0.bufs;
 		mbuf->data_len = wqe->s.w5.size;
@@ -136,7 +176,7 @@ ssows_get_work(struct ssows *ws, struct rte_event *ev, const uint16_t flag)
 
 	if (get_work1 && ev->event_type == RTE_EVENT_TYPE_ETHDEV) {
 		ev->mbuf = ssovf_octeontx_wqe_to_pkt(get_work1,
-				(ev->event >> 20) & 0x7F, flag);
+				(ev->event >> 20) & 0x7F, flag, ws->lookup_mem);
 	} else if (unlikely((get_work0 & 0xFFFFFFFF) == 0xFFFFFFFF)) {
 		ssovf_octeontx_wqe_free(get_work1);
 		return 0;
diff --git a/drivers/net/octeontx/octeontx_rxtx.h b/drivers/net/octeontx/octeontx_rxtx.h
index 3f2147a02..8b46105b6 100644
--- a/drivers/net/octeontx/octeontx_rxtx.h
+++ b/drivers/net/octeontx/octeontx_rxtx.h
@@ -486,12 +486,19 @@ T(noff_ol3ol4csum_l3l4csum_mseg,	1, 1, 1, 1,	14,		       \
 
 /* RX offload macros */
 #define VLAN_FLTR_F     OCCTX_RX_VLAN_FLTR_F
+#define CSUM_F		OCCTX_RX_OFFLOAD_CSUM_F
 #define MULT_RX_F       OCCTX_RX_MULTI_SEG_F
-/* [VLAN_FLTR][MULTI_SEG] */
+
+/* [VLAN_FLTR] [CSUM_F] [MULTI_SEG] */
 #define OCCTX_RX_FASTPATH_MODES						       \
-R(no_offload,				0, 0,  OCCTX_RX_OFFLOAD_NONE)	       \
-R(mseg,					0, 1,  MULT_RX_F)		       \
-R(vlan,					1, 0,  VLAN_FLTR_F)		       \
-R(vlan_mseg,				1, 1,  VLAN_FLTR_F | MULT_RX_F)
+R(no_offload,				0, 0, 0,  OCCTX_RX_OFFLOAD_NONE)       \
+R(mseg,					0, 0, 1,  MULT_RX_F)		       \
+R(csum,					0, 1, 0,  CSUM_F)		       \
+R(csum_mseg,				0, 1, 1,  CSUM_F | MULT_RX_F)	       \
+R(vlan,					1, 0, 0,  VLAN_FLTR_F)		       \
+R(vlan_mseg,				1, 0, 1,  VLAN_FLTR_F | MULT_RX_F)     \
+R(vlan_csum,				1, 1, 0,  VLAN_FLTR_F | CSUM_F)	       \
+R(vlan_csum_mseg,			1, 1, 1,  CSUM_F | VLAN_FLTR_F |       \
+					MULT_RX_F)
 
  #endif /* __OCTEONTX_RXTX_H__ */
-- 
2.18.0



More information about the dev mailing list