Add Tx offloading support:
- support TSO
Signed-off-by: Beilei Xing <beilei.x...@intel.com>
Signed-off-by: Xiaoyun Li <xiaoyun...@intel.com>
Signed-off-by: Junfeng Guo <junfeng....@intel.com>
---
doc/guides/nics/features/idpf.ini | 1 +
drivers/net/idpf/idpf_ethdev.c | 4 +-
drivers/net/idpf/idpf_rxtx.c | 134 +++++++++++++++++++++++++++++-
drivers/net/idpf/idpf_rxtx.h | 22 +++++
4 files changed, 158 insertions(+), 3 deletions(-)
diff --git a/doc/guides/nics/features/idpf.ini
b/doc/guides/nics/features/idpf.ini
index c86d9378ea..47c686762d 100644
--- a/doc/guides/nics/features/idpf.ini
+++ b/doc/guides/nics/features/idpf.ini
@@ -9,6 +9,7 @@
[Features]
Queue start/stop = Y
MTU update = Y
+TSO = P
L3 checksum offload = P
L4 checksum offload = P
Packet type parsing = Y
diff --git a/drivers/net/idpf/idpf_ethdev.c b/drivers/net/idpf/idpf_ethdev.c
index d8cc423a23..21315866bf 100644
--- a/drivers/net/idpf/idpf_ethdev.c
+++ b/drivers/net/idpf/idpf_ethdev.c
@@ -102,7 +102,9 @@ idpf_dev_info_get(struct rte_eth_dev *dev, struct
rte_eth_dev_info *dev_info)
RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
RTE_ETH_RX_OFFLOAD_RSS_HASH;
- dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
+ dev_info->tx_offload_capa =
+ RTE_ETH_TX_OFFLOAD_TCP_TSO |
+ RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
dev_info->default_rxconf = (struct rte_eth_rxconf) {
.rx_free_thresh = IDPF_DEFAULT_RX_FREE_THRESH,
diff --git a/drivers/net/idpf/idpf_rxtx.c b/drivers/net/idpf/idpf_rxtx.c
index 143c8b69f3..8f82cf1b59 100644
--- a/drivers/net/idpf/idpf_rxtx.c
+++ b/drivers/net/idpf/idpf_rxtx.c
@@ -1549,6 +1549,49 @@ idpf_split_tx_free(struct idpf_tx_queue *cq)
cq->tx_tail = next;
}
+/* Check if the context descriptor is needed for TX offloading */
+static inline uint16_t
+idpf_calc_context_desc(uint64_t flags)
+{
+ if ((flags & RTE_MBUF_F_TX_TCP_SEG) != 0)
+ return 1;
+
+ return 0;
+}
+
+/* set TSO context descriptor
+ */
+static inline void
+idpf_set_splitq_tso_ctx(struct rte_mbuf *mbuf,
+ union idpf_tx_offload tx_offload,
+ volatile union idpf_flex_tx_ctx_desc *ctx_desc)
+{
+ uint16_t cmd_dtype;
+ uint32_t tso_len;
+ uint8_t hdr_len;
+
+ if (tx_offload.l4_len == 0) {
+ PMD_TX_LOG(DEBUG, "L4 length set to 0");
+ return;
+ }
+
+ hdr_len = tx_offload.l2_len +
+ tx_offload.l3_len +
+ tx_offload.l4_len;
+ cmd_dtype = IDPF_TX_DESC_DTYPE_FLEX_TSO_CTX |
+ IDPF_TX_FLEX_CTX_DESC_CMD_TSO;
+ tso_len = mbuf->pkt_len - hdr_len;
+
+ ctx_desc->tso.qw1.cmd_dtype = rte_cpu_to_le_16(cmd_dtype);
+ ctx_desc->tso.qw0.hdr_len = hdr_len;
+ ctx_desc->tso.qw0.mss_rt =
+ rte_cpu_to_le_16((uint16_t)mbuf->tso_segsz &
+ IDPF_TXD_FLEX_CTX_MSS_RT_M);
+ ctx_desc->tso.qw0.flex_tlen =
+ rte_cpu_to_le_32(tso_len &
+ IDPF_TXD_FLEX_CTX_MSS_RT_M);
+}
+
uint16_t
idpf_splitq_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
uint16_t nb_pkts)
@@ -1557,11 +1600,14 @@ idpf_splitq_xmit_pkts(void *tx_queue, struct rte_mbuf
**tx_pkts,
volatile struct idpf_flex_tx_sched_desc *txr;
volatile struct idpf_flex_tx_sched_desc *txd;
struct idpf_tx_entry *sw_ring;
+ union idpf_tx_offload tx_offload = {0};
struct idpf_tx_entry *txe, *txn;
uint16_t nb_used, tx_id, sw_id;
struct rte_mbuf *tx_pkt;
uint16_t nb_to_clean;
uint16_t nb_tx = 0;
+ uint64_t ol_flags;
+ uint16_t nb_ctx;
if (unlikely(txq == NULL) || unlikely(!txq->q_started))
return nb_tx;
@@ -1591,7 +1637,29 @@ idpf_splitq_xmit_pkts(void *tx_queue, struct rte_mbuf
**tx_pkts,
if (txq->nb_free < tx_pkt->nb_segs)
break;
- nb_used = tx_pkt->nb_segs;
+
+ ol_flags = tx_pkt->ol_flags;
+ tx_offload.l2_len = tx_pkt->l2_len;
+ tx_offload.l3_len = tx_pkt->l3_len;
+ tx_offload.l4_len = tx_pkt->l4_len;
+ tx_offload.tso_segsz = tx_pkt->tso_segsz;
+ /* Calculate the number of context descriptors needed. */
+ nb_ctx = idpf_calc_context_desc(ol_flags);
+ nb_used = tx_pkt->nb_segs + nb_ctx;
+
+ /* context descriptor */
+ if (nb_ctx != 0) {
+ volatile union idpf_flex_tx_ctx_desc *ctx_desc =
+ (volatile union idpf_flex_tx_ctx_desc *)&txr[tx_id];
+
+ if ((ol_flags & RTE_MBUF_F_TX_TCP_SEG) != 0)
+ idpf_set_splitq_tso_ctx(tx_pkt, tx_offload,
+ ctx_desc);
+
+ tx_id++;
+ if (tx_id == txq->nb_tx_desc)
+ tx_id = 0;
+ }
do {
txd = &txr[tx_id];
@@ -1842,14 +1910,17 @@ idpf_singleq_xmit_pkts(void *tx_queue, struct rte_mbuf
**tx_pkts,
{
volatile struct idpf_flex_tx_desc *txd;
volatile struct idpf_flex_tx_desc *txr;
+ union idpf_tx_offload tx_offload = {0};
struct idpf_tx_entry *txe, *txn;
struct idpf_tx_entry *sw_ring;
struct idpf_tx_queue *txq;
struct rte_mbuf *tx_pkt;
struct rte_mbuf *m_seg;
uint64_t buf_dma_addr;
+ uint64_t ol_flags;
uint16_t tx_last;
uint16_t nb_used;
+ uint16_t nb_ctx;
uint16_t td_cmd;
uint16_t tx_id;
uint16_t nb_tx;
@@ -1876,11 +1947,19 @@ idpf_singleq_xmit_pkts(void *tx_queue, struct rte_mbuf
**tx_pkts,
tx_pkt = *tx_pkts++;
RTE_MBUF_PREFETCH_TO_FREE(txe->mbuf);
+ ol_flags = tx_pkt->ol_flags;
+ tx_offload.l2_len = tx_pkt->l2_len;
+ tx_offload.l3_len = tx_pkt->l3_len;
+ tx_offload.l4_len = tx_pkt->l4_len;
+ tx_offload.tso_segsz = tx_pkt->tso_segsz;
+ /* Calculate the number of context descriptors needed. */
+ nb_ctx = idpf_calc_context_desc(ol_flags);
+
/* The number of descriptors that must be allocated for
* a packet equals to the number of the segments of that
* packet plus 1 context descriptor if needed.
*/
- nb_used = (uint16_t)(tx_pkt->nb_segs);
+ nb_used = (uint16_t)(tx_pkt->nb_segs + nb_ctx);
tx_last = (uint16_t)(tx_id + nb_used - 1);
/* Circular ring */
@@ -1908,6 +1987,29 @@ idpf_singleq_xmit_pkts(void *tx_queue, struct rte_mbuf
**tx_pkts,
}
}
+ if (nb_ctx != 0) {
+ /* Setup TX context descriptor if required */
+ volatile union idpf_flex_tx_ctx_desc *ctx_txd =
+ (volatile union idpf_flex_tx_ctx_desc *)
+ &txr[tx_id];
+
+ txn = &sw_ring[txe->next_id];
+ RTE_MBUF_PREFETCH_TO_FREE(txn->mbuf);
+ if (txe->mbuf != NULL) {
+ rte_pktmbuf_free_seg(txe->mbuf);
+ txe->mbuf = NULL;
+ }
+
+ /* TSO enabled */
+ if ((ol_flags & RTE_MBUF_F_TX_TCP_SEG) != 0)
+ idpf_set_splitq_tso_ctx(tx_pkt, tx_offload,
+ ctx_txd);
+
+ txe->last_id = tx_last;
+ tx_id = txe->next_id;
+ txe = txn;
+ }
+
m_seg = tx_pkt;
do {
txd = &txr[tx_id];
@@ -1968,16 +2070,44 @@ idpf_prep_pkts(__rte_unused void *tx_queue, struct
rte_mbuf **tx_pkts,
uint16_t nb_pkts)
{
int i, ret;
+ uint64_t ol_flags;
struct rte_mbuf *m;
for (i = 0; i < nb_pkts; i++) {
m = tx_pkts[i];
+ ol_flags = m->ol_flags;
+
+ /* Check condition for nb_segs > IDPF_TX_MAX_MTU_SEG. */
+ if ((ol_flags & RTE_MBUF_F_TX_TCP_SEG) == 0) {
+ if (m->nb_segs > IDPF_TX_MAX_MTU_SEG) {
+ rte_errno = EINVAL;
+ return i;
+ }
+ } else if ((m->tso_segsz < IDPF_MIN_TSO_MSS) ||
+ (m->tso_segsz > IDPF_MAX_TSO_MSS) ||
+ (m->pkt_len > IDPF_MAX_TSO_FRAME_SIZE)) {
+ /* MSS outside the range are considered malicious */
+ rte_errno = EINVAL;
+ return i;
+ }
+
+ if ((ol_flags & IDPF_TX_OFFLOAD_NOTSUP_MASK) != 0) {
+ rte_errno = ENOTSUP;
+ return i;
+ }
if (m->pkt_len < IDPF_MIN_FRAME_SIZE) {
rte_errno = EINVAL;
return i;
}
+#ifdef RTE_LIBRTE_ETHDEV_DEBUG
+ ret = rte_validate_tx_offload(m);
+ if (ret != 0) {
+ rte_errno = -ret;
+ return i;
+ }
+#endif
ret = rte_net_intel_cksum_prepare(m);
if (ret != 0) {
rte_errno = -ret;
diff --git a/drivers/net/idpf/idpf_rxtx.h b/drivers/net/idpf/idpf_rxtx.h
index eb0b230d3a..efb2734d85 100644
--- a/drivers/net/idpf/idpf_rxtx.h
+++ b/drivers/net/idpf/idpf_rxtx.h
@@ -21,6 +21,16 @@
#define IDPF_DEFAULT_TX_RS_THRESH 32
#define IDPF_DEFAULT_TX_FREE_THRESH 32
+#define IDPF_MIN_TSO_MSS 88
+#define IDPF_MAX_TSO_MSS 9728
+#define IDPF_MAX_TSO_FRAME_SIZE 262143
+#define IDPF_TX_MAX_MTU_SEG 10
+
+#define IDPF_TX_OFFLOAD_MASK RTE_MBUF_F_TX_TCP_SEG
+
+#define IDPF_TX_OFFLOAD_NOTSUP_MASK \
+ (RTE_MBUF_F_TX_OFFLOAD_MASK ^ IDPF_TX_OFFLOAD_MASK)
+
#define IDPF_GET_PTYPE_SIZE(p) \
(sizeof(struct virtchnl2_ptype) + \
(((p)->proto_id_count ? ((p)->proto_id_count - 1) : 0) *
sizeof((p)->proto_id[0])))
@@ -115,6 +125,18 @@ struct idpf_tx_queue {
struct idpf_tx_queue *complq;
};
+/* Offload features */
+union idpf_tx_offload {
+ uint64_t data;
+ struct {
+ uint64_t l2_len:7; /* L2 (MAC) Header Length. */
+ uint64_t l3_len:9; /* L3 (IP) Header Length. */
+ uint64_t l4_len:8; /* L4 Header Length. */
+ uint64_t tso_segsz:16; /* TCP TSO segment size */
+ /* uint64_t unused : 24; */
+ };
+};
+
struct idpf_rxq_ops {
void (*release_mbufs)(struct idpf_rx_queue *rxq);
};
--
2.34.1
