From e4891e4687c8dd136d80d6c1b857a02931ed6fc8 Mon Sep 17 00:00:00 2001 From: Alexander Lobakin Date: Thu, 20 Jun 2024 15:53:38 +0200 Subject: idpf: split &idpf_queue into 4 strictly-typed queue structures Currently, sizeof(struct idpf_queue) is 32 Kb. This is due to the 12-bit hashtable declaration at the end of the queue. This HT is needed only for Tx queues when the flow scheduling mode is enabled. But &idpf_queue is unified for all of the queue types, provoking excessive memory usage. The unified structure in general makes the code less effective via suboptimal fields placement. You can't avoid that unless you make unions each 2 fields. Even then, different field alignment etc., doesn't allow you to optimize things to the limit. Split &idpf_queue into 4 structures corresponding to the queue types: RQ (Rx queue), SQ (Tx queue), FQ (buffer queue), and CQ (completion queue). Place only needed fields there and shortcuts handy for hotpath. Allocate the abovementioned hashtable dynamically and only when needed, keeping &idpf_tx_queue relatively short (192 bytes, same as Rx). This HT is used only for OOO completions, which aren't really hotpath anyway. Note that this change must be done atomically, otherwise it's really easy to get lost and miss something. Signed-off-by: Alexander Lobakin Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/idpf/idpf_virtchnl.c | 73 +++++++++++++------------ 1 file changed, 38 insertions(+), 35 deletions(-) (limited to 'drivers/net/ethernet/intel/idpf/idpf_virtchnl.c') diff --git a/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c b/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c index a5f9b7a5effe..44602b87cd41 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c +++ b/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c @@ -750,7 +750,7 @@ static int idpf_wait_for_marker_event(struct idpf_vport *vport) int i; for (i = 0; i < vport->num_txq; i++) - set_bit(__IDPF_Q_SW_MARKER, vport->txqs[i]->flags); + idpf_queue_set(SW_MARKER, vport->txqs[i]); event = wait_event_timeout(vport->sw_marker_wq, test_and_clear_bit(IDPF_VPORT_SW_MARKER, @@ -758,7 +758,7 @@ static int idpf_wait_for_marker_event(struct idpf_vport *vport) msecs_to_jiffies(500)); for (i = 0; i < vport->num_txq; i++) - clear_bit(__IDPF_Q_POLL_MODE, vport->txqs[i]->flags); + idpf_queue_clear(POLL_MODE, vport->txqs[i]); if (event) return 0; @@ -1092,7 +1092,6 @@ static int __idpf_queue_reg_init(struct idpf_vport *vport, u32 *reg_vals, int num_regs, u32 q_type) { struct idpf_adapter *adapter = vport->adapter; - struct idpf_queue *q; int i, j, k = 0; switch (q_type) { @@ -1111,6 +1110,8 @@ static int __idpf_queue_reg_init(struct idpf_vport *vport, u32 *reg_vals, u16 num_rxq = rx_qgrp->singleq.num_rxq; for (j = 0; j < num_rxq && k < num_regs; j++, k++) { + struct idpf_rx_queue *q; + q = rx_qgrp->singleq.rxqs[j]; q->tail = idpf_get_reg_addr(adapter, reg_vals[k]); @@ -1123,6 +1124,8 @@ static int __idpf_queue_reg_init(struct idpf_vport *vport, u32 *reg_vals, u8 num_bufqs = vport->num_bufqs_per_qgrp; for (j = 0; j < num_bufqs && k < num_regs; j++, k++) { + struct idpf_buf_queue *q; + q = &rx_qgrp->splitq.bufq_sets[j].bufq; q->tail = idpf_get_reg_addr(adapter, reg_vals[k]); @@ -1449,19 +1452,19 @@ static int idpf_send_config_tx_queues_msg(struct idpf_vport *vport) qi[k].model = cpu_to_le16(vport->txq_model); qi[k].type = - cpu_to_le32(tx_qgrp->txqs[j]->q_type); + cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_TX); qi[k].ring_len = cpu_to_le16(tx_qgrp->txqs[j]->desc_count); qi[k].dma_ring_addr = cpu_to_le64(tx_qgrp->txqs[j]->dma); if (idpf_is_queue_model_split(vport->txq_model)) { - struct idpf_queue *q = tx_qgrp->txqs[j]; + struct idpf_tx_queue *q = tx_qgrp->txqs[j]; qi[k].tx_compl_queue_id = cpu_to_le16(tx_qgrp->complq->q_id); qi[k].relative_queue_id = cpu_to_le16(j); - if (test_bit(__IDPF_Q_FLOW_SCH_EN, q->flags)) + if (idpf_queue_has(FLOW_SCH_EN, q)) qi[k].sched_mode = cpu_to_le16(VIRTCHNL2_TXQ_SCHED_MODE_FLOW); else @@ -1478,11 +1481,11 @@ static int idpf_send_config_tx_queues_msg(struct idpf_vport *vport) qi[k].queue_id = cpu_to_le32(tx_qgrp->complq->q_id); qi[k].model = cpu_to_le16(vport->txq_model); - qi[k].type = cpu_to_le32(tx_qgrp->complq->q_type); + qi[k].type = cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_TX_COMPLETION); qi[k].ring_len = cpu_to_le16(tx_qgrp->complq->desc_count); qi[k].dma_ring_addr = cpu_to_le64(tx_qgrp->complq->dma); - if (test_bit(__IDPF_Q_FLOW_SCH_EN, tx_qgrp->complq->flags)) + if (idpf_queue_has(FLOW_SCH_EN, tx_qgrp->complq)) sched_mode = VIRTCHNL2_TXQ_SCHED_MODE_FLOW; else sched_mode = VIRTCHNL2_TXQ_SCHED_MODE_QUEUE; @@ -1567,17 +1570,18 @@ static int idpf_send_config_rx_queues_msg(struct idpf_vport *vport) goto setup_rxqs; for (j = 0; j < vport->num_bufqs_per_qgrp; j++, k++) { - struct idpf_queue *bufq = + struct idpf_buf_queue *bufq = &rx_qgrp->splitq.bufq_sets[j].bufq; qi[k].queue_id = cpu_to_le32(bufq->q_id); qi[k].model = cpu_to_le16(vport->rxq_model); - qi[k].type = cpu_to_le32(bufq->q_type); + qi[k].type = + cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_RX_BUFFER); qi[k].desc_ids = cpu_to_le64(VIRTCHNL2_RXDID_2_FLEX_SPLITQ_M); qi[k].ring_len = cpu_to_le16(bufq->desc_count); qi[k].dma_ring_addr = cpu_to_le64(bufq->dma); qi[k].data_buffer_size = cpu_to_le32(bufq->rx_buf_size); - qi[k].buffer_notif_stride = bufq->rx_buf_stride; + qi[k].buffer_notif_stride = IDPF_RX_BUF_STRIDE; qi[k].rx_buffer_low_watermark = cpu_to_le16(bufq->rx_buffer_low_watermark); if (idpf_is_feature_ena(vport, NETIF_F_GRO_HW)) @@ -1591,7 +1595,7 @@ setup_rxqs: num_rxq = rx_qgrp->singleq.num_rxq; for (j = 0; j < num_rxq; j++, k++) { - struct idpf_queue *rxq; + struct idpf_rx_queue *rxq; if (!idpf_is_queue_model_split(vport->rxq_model)) { rxq = rx_qgrp->singleq.rxqs[j]; @@ -1599,11 +1603,11 @@ setup_rxqs: } rxq = &rx_qgrp->splitq.rxq_sets[j]->rxq; qi[k].rx_bufq1_id = - cpu_to_le16(rxq->rxq_grp->splitq.bufq_sets[0].bufq.q_id); + cpu_to_le16(rxq->bufq_sets[0].bufq.q_id); if (vport->num_bufqs_per_qgrp > IDPF_SINGLE_BUFQ_PER_RXQ_GRP) { qi[k].bufq2_ena = IDPF_BUFQ2_ENA; qi[k].rx_bufq2_id = - cpu_to_le16(rxq->rxq_grp->splitq.bufq_sets[1].bufq.q_id); + cpu_to_le16(rxq->bufq_sets[1].bufq.q_id); } qi[k].rx_buffer_low_watermark = cpu_to_le16(rxq->rx_buffer_low_watermark); @@ -1611,7 +1615,7 @@ setup_rxqs: qi[k].qflags |= cpu_to_le16(VIRTCHNL2_RXQ_RSC); common_qi_fields: - if (rxq->rx_hsplit_en) { + if (idpf_queue_has(HSPLIT_EN, rxq)) { qi[k].qflags |= cpu_to_le16(VIRTCHNL2_RXQ_HDR_SPLIT); qi[k].hdr_buffer_size = @@ -1619,7 +1623,7 @@ common_qi_fields: } qi[k].queue_id = cpu_to_le32(rxq->q_id); qi[k].model = cpu_to_le16(vport->rxq_model); - qi[k].type = cpu_to_le32(rxq->q_type); + qi[k].type = cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_RX); qi[k].ring_len = cpu_to_le16(rxq->desc_count); qi[k].dma_ring_addr = cpu_to_le64(rxq->dma); qi[k].max_pkt_size = cpu_to_le32(rxq->rx_max_pkt_size); @@ -1706,7 +1710,7 @@ static int idpf_send_ena_dis_queues_msg(struct idpf_vport *vport, bool ena) struct idpf_txq_group *tx_qgrp = &vport->txq_grps[i]; for (j = 0; j < tx_qgrp->num_txq; j++, k++) { - qc[k].type = cpu_to_le32(tx_qgrp->txqs[j]->q_type); + qc[k].type = cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_TX); qc[k].start_queue_id = cpu_to_le32(tx_qgrp->txqs[j]->q_id); qc[k].num_queues = cpu_to_le32(IDPF_NUMQ_PER_CHUNK); } @@ -1720,7 +1724,7 @@ static int idpf_send_ena_dis_queues_msg(struct idpf_vport *vport, bool ena) for (i = 0; i < vport->num_txq_grp; i++, k++) { struct idpf_txq_group *tx_qgrp = &vport->txq_grps[i]; - qc[k].type = cpu_to_le32(tx_qgrp->complq->q_type); + qc[k].type = cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_TX_COMPLETION); qc[k].start_queue_id = cpu_to_le32(tx_qgrp->complq->q_id); qc[k].num_queues = cpu_to_le32(IDPF_NUMQ_PER_CHUNK); } @@ -1741,12 +1745,12 @@ setup_rx: qc[k].start_queue_id = cpu_to_le32(rx_qgrp->splitq.rxq_sets[j]->rxq.q_id); qc[k].type = - cpu_to_le32(rx_qgrp->splitq.rxq_sets[j]->rxq.q_type); + cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_RX); } else { qc[k].start_queue_id = cpu_to_le32(rx_qgrp->singleq.rxqs[j]->q_id); qc[k].type = - cpu_to_le32(rx_qgrp->singleq.rxqs[j]->q_type); + cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_RX); } qc[k].num_queues = cpu_to_le32(IDPF_NUMQ_PER_CHUNK); } @@ -1761,10 +1765,11 @@ setup_rx: struct idpf_rxq_group *rx_qgrp = &vport->rxq_grps[i]; for (j = 0; j < vport->num_bufqs_per_qgrp; j++, k++) { - struct idpf_queue *q; + const struct idpf_buf_queue *q; q = &rx_qgrp->splitq.bufq_sets[j].bufq; - qc[k].type = cpu_to_le32(q->q_type); + qc[k].type = + cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_RX_BUFFER); qc[k].start_queue_id = cpu_to_le32(q->q_id); qc[k].num_queues = cpu_to_le32(IDPF_NUMQ_PER_CHUNK); } @@ -1849,7 +1854,8 @@ int idpf_send_map_unmap_queue_vector_msg(struct idpf_vport *vport, bool map) struct idpf_txq_group *tx_qgrp = &vport->txq_grps[i]; for (j = 0; j < tx_qgrp->num_txq; j++, k++) { - vqv[k].queue_type = cpu_to_le32(tx_qgrp->txqs[j]->q_type); + vqv[k].queue_type = + cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_TX); vqv[k].queue_id = cpu_to_le32(tx_qgrp->txqs[j]->q_id); if (idpf_is_queue_model_split(vport->txq_model)) { @@ -1879,14 +1885,15 @@ int idpf_send_map_unmap_queue_vector_msg(struct idpf_vport *vport, bool map) num_rxq = rx_qgrp->singleq.num_rxq; for (j = 0; j < num_rxq; j++, k++) { - struct idpf_queue *rxq; + struct idpf_rx_queue *rxq; if (idpf_is_queue_model_split(vport->rxq_model)) rxq = &rx_qgrp->splitq.rxq_sets[j]->rxq; else rxq = rx_qgrp->singleq.rxqs[j]; - vqv[k].queue_type = cpu_to_le32(rxq->q_type); + vqv[k].queue_type = + cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_RX); vqv[k].queue_id = cpu_to_le32(rxq->q_id); vqv[k].vector_id = cpu_to_le16(rxq->q_vector->v_idx); vqv[k].itr_idx = cpu_to_le32(rxq->q_vector->rx_itr_idx); @@ -1975,7 +1982,7 @@ int idpf_send_disable_queues_msg(struct idpf_vport *vport) * queues virtchnl message is sent */ for (i = 0; i < vport->num_txq; i++) - set_bit(__IDPF_Q_POLL_MODE, vport->txqs[i]->flags); + idpf_queue_set(POLL_MODE, vport->txqs[i]); /* schedule the napi to receive all the marker packets */ local_bh_disable(); @@ -3242,7 +3249,6 @@ static int __idpf_vport_queue_ids_init(struct idpf_vport *vport, int num_qids, u32 q_type) { - struct idpf_queue *q; int i, j, k = 0; switch (q_type) { @@ -3250,11 +3256,8 @@ static int __idpf_vport_queue_ids_init(struct idpf_vport *vport, for (i = 0; i < vport->num_txq_grp; i++) { struct idpf_txq_group *tx_qgrp = &vport->txq_grps[i]; - for (j = 0; j < tx_qgrp->num_txq && k < num_qids; j++, k++) { + for (j = 0; j < tx_qgrp->num_txq && k < num_qids; j++, k++) tx_qgrp->txqs[j]->q_id = qids[k]; - tx_qgrp->txqs[j]->q_type = - VIRTCHNL2_QUEUE_TYPE_TX; - } } break; case VIRTCHNL2_QUEUE_TYPE_RX: @@ -3268,12 +3271,13 @@ static int __idpf_vport_queue_ids_init(struct idpf_vport *vport, num_rxq = rx_qgrp->singleq.num_rxq; for (j = 0; j < num_rxq && k < num_qids; j++, k++) { + struct idpf_rx_queue *q; + if (idpf_is_queue_model_split(vport->rxq_model)) q = &rx_qgrp->splitq.rxq_sets[j]->rxq; else q = rx_qgrp->singleq.rxqs[j]; q->q_id = qids[k]; - q->q_type = VIRTCHNL2_QUEUE_TYPE_RX; } } break; @@ -3282,8 +3286,6 @@ static int __idpf_vport_queue_ids_init(struct idpf_vport *vport, struct idpf_txq_group *tx_qgrp = &vport->txq_grps[i]; tx_qgrp->complq->q_id = qids[k]; - tx_qgrp->complq->q_type = - VIRTCHNL2_QUEUE_TYPE_TX_COMPLETION; } break; case VIRTCHNL2_QUEUE_TYPE_RX_BUFFER: @@ -3292,9 +3294,10 @@ static int __idpf_vport_queue_ids_init(struct idpf_vport *vport, u8 num_bufqs = vport->num_bufqs_per_qgrp; for (j = 0; j < num_bufqs && k < num_qids; j++, k++) { + struct idpf_buf_queue *q; + q = &rx_qgrp->splitq.bufq_sets[j].bufq; q->q_id = qids[k]; - q->q_type = VIRTCHNL2_QUEUE_TYPE_RX_BUFFER; } } break; -- cgit v1.2.3-58-ga151 From f771314d6b75181de7079c3c7d666293e4ed2b22 Mon Sep 17 00:00:00 2001 From: Alexander Lobakin Date: Thu, 20 Jun 2024 15:53:42 +0200 Subject: idpf: compile singleq code only under default-n CONFIG_IDPF_SINGLEQ Currently, all HW supporting idpf supports the singleq model, but none of it advertises it by default, as splitq is supported and preferred for multiple reasons. Still, this almost dead code often times adds hotpath branches and redundant cacheline accesses. While it can't currently be removed, add CONFIG_IDPF_SINGLEQ and build the singleq code only when it's enabled manually. This corresponds to -10 Kb of object code size and a good bunch of hotpath checks. idpf_is_queue_model_split() works as a gate and compiles out to `true` when the config option is disabled. Reviewed-by: Przemek Kitszel Signed-off-by: Alexander Lobakin Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/Kconfig | 13 +----------- drivers/net/ethernet/intel/idpf/Kconfig | 27 +++++++++++++++++++++++++ drivers/net/ethernet/intel/idpf/Makefile | 3 ++- drivers/net/ethernet/intel/idpf/idpf.h | 3 ++- drivers/net/ethernet/intel/idpf/idpf_txrx.c | 2 +- drivers/net/ethernet/intel/idpf/idpf_virtchnl.c | 15 ++++++++++---- 6 files changed, 44 insertions(+), 19 deletions(-) create mode 100644 drivers/net/ethernet/intel/idpf/Kconfig (limited to 'drivers/net/ethernet/intel/idpf/idpf_virtchnl.c') diff --git a/drivers/net/ethernet/intel/Kconfig b/drivers/net/ethernet/intel/Kconfig index e0287fbd501d..0375c7448a57 100644 --- a/drivers/net/ethernet/intel/Kconfig +++ b/drivers/net/ethernet/intel/Kconfig @@ -384,17 +384,6 @@ config IGC_LEDS Optional support for controlling the NIC LED's with the netdev LED trigger. -config IDPF - tristate "Intel(R) Infrastructure Data Path Function Support" - depends on PCI_MSI - select DIMLIB - select PAGE_POOL - select PAGE_POOL_STATS - help - This driver supports Intel(R) Infrastructure Data Path Function - devices. - - To compile this driver as a module, choose M here. The module - will be called idpf. +source "drivers/net/ethernet/intel/idpf/Kconfig" endif # NET_VENDOR_INTEL diff --git a/drivers/net/ethernet/intel/idpf/Kconfig b/drivers/net/ethernet/intel/idpf/Kconfig new file mode 100644 index 000000000000..9082c16edb7e --- /dev/null +++ b/drivers/net/ethernet/intel/idpf/Kconfig @@ -0,0 +1,27 @@ +# SPDX-License-Identifier: GPL-2.0-only +# Copyright (C) 2024 Intel Corporation + +config IDPF + tristate "Intel(R) Infrastructure Data Path Function Support" + depends on PCI_MSI + select DIMLIB + select PAGE_POOL + select PAGE_POOL_STATS + help + This driver supports Intel(R) Infrastructure Data Path Function + devices. + + To compile this driver as a module, choose M here. The module + will be called idpf. + +if IDPF + +config IDPF_SINGLEQ + bool "idpf singleq support" + help + This option enables support for legacy single Rx/Tx queues w/no + completion and fill queues. Only enable if you have hardware which + wants to work in this mode as it increases the driver size and adds + runtme checks on hotpath. + +endif # IDPF diff --git a/drivers/net/ethernet/intel/idpf/Makefile b/drivers/net/ethernet/intel/idpf/Makefile index 6844ead2f3ac..2ce01a0b5898 100644 --- a/drivers/net/ethernet/intel/idpf/Makefile +++ b/drivers/net/ethernet/intel/idpf/Makefile @@ -12,7 +12,8 @@ idpf-y := \ idpf_ethtool.o \ idpf_lib.o \ idpf_main.o \ - idpf_singleq_txrx.o \ idpf_txrx.o \ idpf_virtchnl.o \ idpf_vf_dev.o + +idpf-$(CONFIG_IDPF_SINGLEQ) += idpf_singleq_txrx.o diff --git a/drivers/net/ethernet/intel/idpf/idpf.h b/drivers/net/ethernet/intel/idpf/idpf.h index f9e43d171f17..5d9529f5b41b 100644 --- a/drivers/net/ethernet/intel/idpf/idpf.h +++ b/drivers/net/ethernet/intel/idpf/idpf.h @@ -599,7 +599,8 @@ struct idpf_adapter { */ static inline int idpf_is_queue_model_split(u16 q_model) { - return q_model == VIRTCHNL2_QUEUE_MODEL_SPLIT; + return !IS_ENABLED(CONFIG_IDPF_SINGLEQ) || + q_model == VIRTCHNL2_QUEUE_MODEL_SPLIT; } #define idpf_is_cap_ena(adapter, field, flag) \ diff --git a/drivers/net/ethernet/intel/idpf/idpf_txrx.c b/drivers/net/ethernet/intel/idpf/idpf_txrx.c index 5dd1b1a9e624..e79a8f9dfc40 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_txrx.c +++ b/drivers/net/ethernet/intel/idpf/idpf_txrx.c @@ -1309,7 +1309,7 @@ static void idpf_vport_calc_numq_per_grp(struct idpf_vport *vport, static void idpf_rxq_set_descids(const struct idpf_vport *vport, struct idpf_rx_queue *q) { - if (vport->rxq_model == VIRTCHNL2_QUEUE_MODEL_SPLIT) { + if (idpf_is_queue_model_split(vport->rxq_model)) { q->rxdids = VIRTCHNL2_RXDID_2_FLEX_SPLITQ_M; } else { if (vport->base_rxd) diff --git a/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c b/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c index 44602b87cd41..1aa4770dfe18 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c +++ b/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c @@ -1256,12 +1256,12 @@ int idpf_send_create_vport_msg(struct idpf_adapter *adapter, vport_msg->vport_type = cpu_to_le16(VIRTCHNL2_VPORT_TYPE_DEFAULT); vport_msg->vport_index = cpu_to_le16(idx); - if (adapter->req_tx_splitq) + if (adapter->req_tx_splitq || !IS_ENABLED(CONFIG_IDPF_SINGLEQ)) vport_msg->txq_model = cpu_to_le16(VIRTCHNL2_QUEUE_MODEL_SPLIT); else vport_msg->txq_model = cpu_to_le16(VIRTCHNL2_QUEUE_MODEL_SINGLE); - if (adapter->req_rx_splitq) + if (adapter->req_rx_splitq || !IS_ENABLED(CONFIG_IDPF_SINGLEQ)) vport_msg->rxq_model = cpu_to_le16(VIRTCHNL2_QUEUE_MODEL_SPLIT); else vport_msg->rxq_model = cpu_to_le16(VIRTCHNL2_QUEUE_MODEL_SINGLE); @@ -1323,10 +1323,17 @@ int idpf_check_supported_desc_ids(struct idpf_vport *vport) vport_msg = adapter->vport_params_recvd[vport->idx]; + if (!IS_ENABLED(CONFIG_IDPF_SINGLEQ) && + (vport_msg->rxq_model == VIRTCHNL2_QUEUE_MODEL_SINGLE || + vport_msg->txq_model == VIRTCHNL2_QUEUE_MODEL_SINGLE)) { + pci_err(adapter->pdev, "singleq mode requested, but not compiled-in\n"); + return -EOPNOTSUPP; + } + rx_desc_ids = le64_to_cpu(vport_msg->rx_desc_ids); tx_desc_ids = le64_to_cpu(vport_msg->tx_desc_ids); - if (vport->rxq_model == VIRTCHNL2_QUEUE_MODEL_SPLIT) { + if (idpf_is_queue_model_split(vport->rxq_model)) { if (!(rx_desc_ids & VIRTCHNL2_RXDID_2_FLEX_SPLITQ_M)) { dev_info(&adapter->pdev->dev, "Minimum RX descriptor support not provided, using the default\n"); vport_msg->rx_desc_ids = cpu_to_le64(VIRTCHNL2_RXDID_2_FLEX_SPLITQ_M); @@ -1336,7 +1343,7 @@ int idpf_check_supported_desc_ids(struct idpf_vport *vport) vport->base_rxd = true; } - if (vport->txq_model != VIRTCHNL2_QUEUE_MODEL_SPLIT) + if (!idpf_is_queue_model_split(vport->txq_model)) return 0; if ((tx_desc_ids & MIN_SUPPORT_TXDID) != MIN_SUPPORT_TXDID) { -- cgit v1.2.3-58-ga151 From 1b1b26208515482d8bc3d875267076f61f3355f7 Mon Sep 17 00:00:00 2001 From: Alexander Lobakin Date: Thu, 20 Jun 2024 15:53:43 +0200 Subject: idpf: reuse libeth's definitions of parsed ptype structures idpf's in-kernel parsed ptype structure is almost identical to the one used in the previous Intel drivers, which means it can be converted to use libeth's definitions and even helpers. The only difference is that it doesn't use a constant table (libie), rather than one obtained from the device. Remove the driver counterpart and use libeth's helpers for hashes and checksums. This slightly optimizes skb fields processing due to faster checks. Also don't define big static array of ptypes in &idpf_vport -- allocate them dynamically. The pointer to it is anyway cached in &idpf_rx_queue. Reviewed-by: Przemek Kitszel Reviewed-by: Jacob Keller Signed-off-by: Alexander Lobakin Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/idpf/Kconfig | 1 + drivers/net/ethernet/intel/idpf/idpf.h | 2 +- drivers/net/ethernet/intel/idpf/idpf_lib.c | 3 + drivers/net/ethernet/intel/idpf/idpf_main.c | 1 + .../net/ethernet/intel/idpf/idpf_singleq_txrx.c | 113 ++++++++----------- drivers/net/ethernet/intel/idpf/idpf_txrx.c | 125 ++++++++------------- drivers/net/ethernet/intel/idpf/idpf_txrx.h | 88 +-------------- drivers/net/ethernet/intel/idpf/idpf_virtchnl.c | 72 +++++++----- 8 files changed, 153 insertions(+), 252 deletions(-) (limited to 'drivers/net/ethernet/intel/idpf/idpf_virtchnl.c') diff --git a/drivers/net/ethernet/intel/idpf/Kconfig b/drivers/net/ethernet/intel/idpf/Kconfig index 9082c16edb7e..638484c5723c 100644 --- a/drivers/net/ethernet/intel/idpf/Kconfig +++ b/drivers/net/ethernet/intel/idpf/Kconfig @@ -5,6 +5,7 @@ config IDPF tristate "Intel(R) Infrastructure Data Path Function Support" depends on PCI_MSI select DIMLIB + select LIBETH select PAGE_POOL select PAGE_POOL_STATS help diff --git a/drivers/net/ethernet/intel/idpf/idpf.h b/drivers/net/ethernet/intel/idpf/idpf.h index 5d9529f5b41b..078340a01757 100644 --- a/drivers/net/ethernet/intel/idpf/idpf.h +++ b/drivers/net/ethernet/intel/idpf/idpf.h @@ -312,7 +312,7 @@ struct idpf_vport { u16 num_rxq_grp; struct idpf_rxq_group *rxq_grps; u32 rxq_model; - struct idpf_rx_ptype_decoded rx_ptype_lkup[IDPF_RX_MAX_PTYPE]; + struct libeth_rx_pt *rx_ptype_lkup; struct idpf_adapter *adapter; struct net_device *netdev; diff --git a/drivers/net/ethernet/intel/idpf/idpf_lib.c b/drivers/net/ethernet/intel/idpf/idpf_lib.c index 56fdbb2b6fbc..5dbf2b4ba1b0 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_lib.c +++ b/drivers/net/ethernet/intel/idpf/idpf_lib.c @@ -941,6 +941,9 @@ static void idpf_decfg_netdev(struct idpf_vport *vport) { struct idpf_adapter *adapter = vport->adapter; + kfree(vport->rx_ptype_lkup); + vport->rx_ptype_lkup = NULL; + unregister_netdev(vport->netdev); free_netdev(vport->netdev); vport->netdev = NULL; diff --git a/drivers/net/ethernet/intel/idpf/idpf_main.c b/drivers/net/ethernet/intel/idpf/idpf_main.c index f784eea044bd..db476b3314c8 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_main.c +++ b/drivers/net/ethernet/intel/idpf/idpf_main.c @@ -8,6 +8,7 @@ #define DRV_SUMMARY "Intel(R) Infrastructure Data Path Function Linux Driver" MODULE_DESCRIPTION(DRV_SUMMARY); +MODULE_IMPORT_NS(LIBETH); MODULE_LICENSE("GPL"); /** diff --git a/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c b/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c index 8630db24f63a..6c0794e01922 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c +++ b/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c @@ -1,6 +1,8 @@ // SPDX-License-Identifier: GPL-2.0-only /* Copyright (C) 2023 Intel Corporation */ +#include + #include "idpf.h" /** @@ -601,75 +603,62 @@ static bool idpf_rx_singleq_is_non_eop(const union virtchnl2_rx_desc *rx_desc) * @rxq: Rx ring being processed * @skb: skb currently being received and modified * @csum_bits: checksum bits from descriptor - * @ptype: the packet type decoded by hardware + * @decoded: the packet type decoded by hardware * * skb->protocol must be set before this function is called */ -static void idpf_rx_singleq_csum(struct idpf_rx_queue *rxq, struct sk_buff *skb, - struct idpf_rx_csum_decoded *csum_bits, - u16 ptype) +static void idpf_rx_singleq_csum(struct idpf_rx_queue *rxq, + struct sk_buff *skb, + struct idpf_rx_csum_decoded csum_bits, + struct libeth_rx_pt decoded) { - struct idpf_rx_ptype_decoded decoded; bool ipv4, ipv6; /* check if Rx checksum is enabled */ - if (unlikely(!(rxq->netdev->features & NETIF_F_RXCSUM))) + if (!libeth_rx_pt_has_checksum(rxq->netdev, decoded)) return; /* check if HW has decoded the packet and checksum */ - if (unlikely(!(csum_bits->l3l4p))) - return; - - decoded = rxq->rx_ptype_lkup[ptype]; - if (unlikely(!(decoded.known && decoded.outer_ip))) + if (unlikely(!csum_bits.l3l4p)) return; - ipv4 = IDPF_RX_PTYPE_TO_IPV(&decoded, IDPF_RX_PTYPE_OUTER_IPV4); - ipv6 = IDPF_RX_PTYPE_TO_IPV(&decoded, IDPF_RX_PTYPE_OUTER_IPV6); + ipv4 = libeth_rx_pt_get_ip_ver(decoded) == LIBETH_RX_PT_OUTER_IPV4; + ipv6 = libeth_rx_pt_get_ip_ver(decoded) == LIBETH_RX_PT_OUTER_IPV6; /* Check if there were any checksum errors */ - if (unlikely(ipv4 && (csum_bits->ipe || csum_bits->eipe))) + if (unlikely(ipv4 && (csum_bits.ipe || csum_bits.eipe))) goto checksum_fail; /* Device could not do any checksum offload for certain extension * headers as indicated by setting IPV6EXADD bit */ - if (unlikely(ipv6 && csum_bits->ipv6exadd)) + if (unlikely(ipv6 && csum_bits.ipv6exadd)) return; /* check for L4 errors and handle packets that were not able to be * checksummed due to arrival speed */ - if (unlikely(csum_bits->l4e)) + if (unlikely(csum_bits.l4e)) goto checksum_fail; - if (unlikely(csum_bits->nat && csum_bits->eudpe)) + if (unlikely(csum_bits.nat && csum_bits.eudpe)) goto checksum_fail; /* Handle packets that were not able to be checksummed due to arrival * speed, in this case the stack can compute the csum. */ - if (unlikely(csum_bits->pprs)) + if (unlikely(csum_bits.pprs)) return; /* If there is an outer header present that might contain a checksum * we need to bump the checksum level by 1 to reflect the fact that * we are indicating we validated the inner checksum. */ - if (decoded.tunnel_type >= IDPF_RX_PTYPE_TUNNEL_IP_GRENAT) + if (decoded.tunnel_type >= LIBETH_RX_PT_TUNNEL_IP_GRENAT) skb->csum_level = 1; - /* Only report checksum unnecessary for ICMP, TCP, UDP, or SCTP */ - switch (decoded.inner_prot) { - case IDPF_RX_PTYPE_INNER_PROT_ICMP: - case IDPF_RX_PTYPE_INNER_PROT_TCP: - case IDPF_RX_PTYPE_INNER_PROT_UDP: - case IDPF_RX_PTYPE_INNER_PROT_SCTP: - skb->ip_summed = CHECKSUM_UNNECESSARY; - return; - default: - return; - } + skb->ip_summed = CHECKSUM_UNNECESSARY; + return; checksum_fail: u64_stats_update_begin(&rxq->stats_sync); @@ -679,20 +668,17 @@ checksum_fail: /** * idpf_rx_singleq_base_csum - Indicate in skb if hw indicated a good cksum - * @rx_q: Rx completion queue - * @skb: skb currently being received and modified * @rx_desc: the receive descriptor - * @ptype: Rx packet type * * This function only operates on the VIRTCHNL2_RXDID_1_32B_BASE_M base 32byte * descriptor writeback format. + * + * Return: parsed checksum status. **/ -static void idpf_rx_singleq_base_csum(struct idpf_rx_queue *rx_q, - struct sk_buff *skb, - const union virtchnl2_rx_desc *rx_desc, - u16 ptype) +static struct idpf_rx_csum_decoded +idpf_rx_singleq_base_csum(const union virtchnl2_rx_desc *rx_desc) { - struct idpf_rx_csum_decoded csum_bits; + struct idpf_rx_csum_decoded csum_bits = { }; u32 rx_error, rx_status; u64 qword; @@ -711,28 +697,23 @@ static void idpf_rx_singleq_base_csum(struct idpf_rx_queue *rx_q, rx_status); csum_bits.ipv6exadd = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_STATUS_IPV6EXADD_M, rx_status); - csum_bits.nat = 0; - csum_bits.eudpe = 0; - idpf_rx_singleq_csum(rx_q, skb, &csum_bits, ptype); + return csum_bits; } /** * idpf_rx_singleq_flex_csum - Indicate in skb if hw indicated a good cksum - * @rx_q: Rx completion queue - * @skb: skb currently being received and modified * @rx_desc: the receive descriptor - * @ptype: Rx packet type * * This function only operates on the VIRTCHNL2_RXDID_2_FLEX_SQ_NIC flexible * descriptor writeback format. + * + * Return: parsed checksum status. **/ -static void idpf_rx_singleq_flex_csum(struct idpf_rx_queue *rx_q, - struct sk_buff *skb, - const union virtchnl2_rx_desc *rx_desc, - u16 ptype) +static struct idpf_rx_csum_decoded +idpf_rx_singleq_flex_csum(const union virtchnl2_rx_desc *rx_desc) { - struct idpf_rx_csum_decoded csum_bits; + struct idpf_rx_csum_decoded csum_bits = { }; u16 rx_status0, rx_status1; rx_status0 = le16_to_cpu(rx_desc->flex_nic_wb.status_error0); @@ -752,9 +733,8 @@ static void idpf_rx_singleq_flex_csum(struct idpf_rx_queue *rx_q, rx_status0); csum_bits.nat = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_STATUS1_NAT_M, rx_status1); - csum_bits.pprs = 0; - idpf_rx_singleq_csum(rx_q, skb, &csum_bits, ptype); + return csum_bits; } /** @@ -770,11 +750,11 @@ static void idpf_rx_singleq_flex_csum(struct idpf_rx_queue *rx_q, static void idpf_rx_singleq_base_hash(struct idpf_rx_queue *rx_q, struct sk_buff *skb, const union virtchnl2_rx_desc *rx_desc, - struct idpf_rx_ptype_decoded *decoded) + struct libeth_rx_pt decoded) { u64 mask, qw1; - if (unlikely(!(rx_q->netdev->features & NETIF_F_RXHASH))) + if (!libeth_rx_pt_has_hash(rx_q->netdev, decoded)) return; mask = VIRTCHNL2_RX_BASE_DESC_FLTSTAT_RSS_HASH_M; @@ -783,7 +763,7 @@ static void idpf_rx_singleq_base_hash(struct idpf_rx_queue *rx_q, if (FIELD_GET(mask, qw1) == mask) { u32 hash = le32_to_cpu(rx_desc->base_wb.qword0.hi_dword.rss); - skb_set_hash(skb, hash, idpf_ptype_to_htype(decoded)); + libeth_rx_pt_set_hash(skb, hash, decoded); } } @@ -800,15 +780,17 @@ static void idpf_rx_singleq_base_hash(struct idpf_rx_queue *rx_q, static void idpf_rx_singleq_flex_hash(struct idpf_rx_queue *rx_q, struct sk_buff *skb, const union virtchnl2_rx_desc *rx_desc, - struct idpf_rx_ptype_decoded *decoded) + struct libeth_rx_pt decoded) { - if (unlikely(!(rx_q->netdev->features & NETIF_F_RXHASH))) + if (!libeth_rx_pt_has_hash(rx_q->netdev, decoded)) return; if (FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_STATUS0_RSS_VALID_M, - le16_to_cpu(rx_desc->flex_nic_wb.status_error0))) - skb_set_hash(skb, le32_to_cpu(rx_desc->flex_nic_wb.rss_hash), - idpf_ptype_to_htype(decoded)); + le16_to_cpu(rx_desc->flex_nic_wb.status_error0))) { + u32 hash = le32_to_cpu(rx_desc->flex_nic_wb.rss_hash); + + libeth_rx_pt_set_hash(skb, hash, decoded); + } } /** @@ -829,19 +811,22 @@ idpf_rx_singleq_process_skb_fields(struct idpf_rx_queue *rx_q, const union virtchnl2_rx_desc *rx_desc, u16 ptype) { - struct idpf_rx_ptype_decoded decoded = rx_q->rx_ptype_lkup[ptype]; + struct libeth_rx_pt decoded = rx_q->rx_ptype_lkup[ptype]; + struct idpf_rx_csum_decoded csum_bits; /* modifies the skb - consumes the enet header */ skb->protocol = eth_type_trans(skb, rx_q->netdev); /* Check if we're using base mode descriptor IDs */ if (rx_q->rxdids == VIRTCHNL2_RXDID_1_32B_BASE_M) { - idpf_rx_singleq_base_hash(rx_q, skb, rx_desc, &decoded); - idpf_rx_singleq_base_csum(rx_q, skb, rx_desc, ptype); + idpf_rx_singleq_base_hash(rx_q, skb, rx_desc, decoded); + csum_bits = idpf_rx_singleq_base_csum(rx_desc); } else { - idpf_rx_singleq_flex_hash(rx_q, skb, rx_desc, &decoded); - idpf_rx_singleq_flex_csum(rx_q, skb, rx_desc, ptype); + idpf_rx_singleq_flex_hash(rx_q, skb, rx_desc, decoded); + csum_bits = idpf_rx_singleq_flex_csum(rx_desc); } + + idpf_rx_singleq_csum(rx_q, skb, csum_bits, decoded); } /** diff --git a/drivers/net/ethernet/intel/idpf/idpf_txrx.c b/drivers/net/ethernet/intel/idpf/idpf_txrx.c index e79a8f9dfc40..883e58815f92 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_txrx.c +++ b/drivers/net/ethernet/intel/idpf/idpf_txrx.c @@ -1,6 +1,8 @@ // SPDX-License-Identifier: GPL-2.0-only /* Copyright (C) 2023 Intel Corporation */ +#include + #include "idpf.h" #include "idpf_virtchnl.h" @@ -2886,30 +2888,6 @@ netdev_tx_t idpf_tx_start(struct sk_buff *skb, struct net_device *netdev) return idpf_tx_singleq_frame(skb, tx_q); } -/** - * idpf_ptype_to_htype - get a hash type - * @decoded: Decoded Rx packet type related fields - * - * Returns appropriate hash type (such as PKT_HASH_TYPE_L2/L3/L4) to be used by - * skb_set_hash based on PTYPE as parsed by HW Rx pipeline and is part of - * Rx desc. - */ -enum pkt_hash_types idpf_ptype_to_htype(const struct idpf_rx_ptype_decoded *decoded) -{ - if (!decoded->known) - return PKT_HASH_TYPE_NONE; - if (decoded->payload_layer == IDPF_RX_PTYPE_PAYLOAD_LAYER_PAY2 && - decoded->inner_prot) - return PKT_HASH_TYPE_L4; - if (decoded->payload_layer == IDPF_RX_PTYPE_PAYLOAD_LAYER_PAY2 && - decoded->outer_ip) - return PKT_HASH_TYPE_L3; - if (decoded->outer_ip == IDPF_RX_PTYPE_OUTER_L2) - return PKT_HASH_TYPE_L2; - - return PKT_HASH_TYPE_NONE; -} - /** * idpf_rx_hash - set the hash value in the skb * @rxq: Rx descriptor ring packet is being transacted on @@ -2920,18 +2898,18 @@ enum pkt_hash_types idpf_ptype_to_htype(const struct idpf_rx_ptype_decoded *deco static void idpf_rx_hash(const struct idpf_rx_queue *rxq, struct sk_buff *skb, const struct virtchnl2_rx_flex_desc_adv_nic_3 *rx_desc, - struct idpf_rx_ptype_decoded *decoded) + struct libeth_rx_pt decoded) { u32 hash; - if (unlikely(!(rxq->netdev->features & NETIF_F_RXHASH))) + if (!libeth_rx_pt_has_hash(rxq->netdev, decoded)) return; hash = le16_to_cpu(rx_desc->hash1) | (rx_desc->ff2_mirrid_hash2.hash2 << 16) | (rx_desc->hash3 << 24); - skb_set_hash(skb, hash, idpf_ptype_to_htype(decoded)); + libeth_rx_pt_set_hash(skb, hash, decoded); } /** @@ -2944,55 +2922,43 @@ idpf_rx_hash(const struct idpf_rx_queue *rxq, struct sk_buff *skb, * skb->protocol must be set before this function is called */ static void idpf_rx_csum(struct idpf_rx_queue *rxq, struct sk_buff *skb, - struct idpf_rx_csum_decoded *csum_bits, - struct idpf_rx_ptype_decoded *decoded) + struct idpf_rx_csum_decoded csum_bits, + struct libeth_rx_pt decoded) { bool ipv4, ipv6; /* check if Rx checksum is enabled */ - if (unlikely(!(rxq->netdev->features & NETIF_F_RXCSUM))) + if (!libeth_rx_pt_has_checksum(rxq->netdev, decoded)) return; /* check if HW has decoded the packet and checksum */ - if (!(csum_bits->l3l4p)) + if (unlikely(!csum_bits.l3l4p)) return; - ipv4 = IDPF_RX_PTYPE_TO_IPV(decoded, IDPF_RX_PTYPE_OUTER_IPV4); - ipv6 = IDPF_RX_PTYPE_TO_IPV(decoded, IDPF_RX_PTYPE_OUTER_IPV6); + ipv4 = libeth_rx_pt_get_ip_ver(decoded) == LIBETH_RX_PT_OUTER_IPV4; + ipv6 = libeth_rx_pt_get_ip_ver(decoded) == LIBETH_RX_PT_OUTER_IPV6; - if (ipv4 && (csum_bits->ipe || csum_bits->eipe)) + if (unlikely(ipv4 && (csum_bits.ipe || csum_bits.eipe))) goto checksum_fail; - if (ipv6 && csum_bits->ipv6exadd) + if (unlikely(ipv6 && csum_bits.ipv6exadd)) return; /* check for L4 errors and handle packets that were not able to be * checksummed */ - if (csum_bits->l4e) + if (unlikely(csum_bits.l4e)) goto checksum_fail; - /* Only report checksum unnecessary for ICMP, TCP, UDP, or SCTP */ - switch (decoded->inner_prot) { - case IDPF_RX_PTYPE_INNER_PROT_ICMP: - case IDPF_RX_PTYPE_INNER_PROT_TCP: - case IDPF_RX_PTYPE_INNER_PROT_UDP: - if (!csum_bits->raw_csum_inv) { - u16 csum = csum_bits->raw_csum; - - skb->csum = csum_unfold((__force __sum16)~swab16(csum)); - skb->ip_summed = CHECKSUM_COMPLETE; - } else { - skb->ip_summed = CHECKSUM_UNNECESSARY; - } - break; - case IDPF_RX_PTYPE_INNER_PROT_SCTP: + if (csum_bits.raw_csum_inv || + decoded.inner_prot == LIBETH_RX_PT_INNER_SCTP) { skb->ip_summed = CHECKSUM_UNNECESSARY; - break; - default: - break; + return; } + skb->csum = csum_unfold((__force __sum16)~swab16(csum_bits.raw_csum)); + skb->ip_summed = CHECKSUM_COMPLETE; + return; checksum_fail: @@ -3004,32 +2970,34 @@ checksum_fail: /** * idpf_rx_splitq_extract_csum_bits - Extract checksum bits from descriptor * @rx_desc: receive descriptor - * @csum: structure to extract checksum fields * + * Return: parsed checksum status. **/ -static void -idpf_rx_splitq_extract_csum_bits(const struct virtchnl2_rx_flex_desc_adv_nic_3 *rx_desc, - struct idpf_rx_csum_decoded *csum) +static struct idpf_rx_csum_decoded +idpf_rx_splitq_extract_csum_bits(const struct virtchnl2_rx_flex_desc_adv_nic_3 *rx_desc) { + struct idpf_rx_csum_decoded csum = { }; u8 qword0, qword1; qword0 = rx_desc->status_err0_qw0; qword1 = rx_desc->status_err0_qw1; - csum->ipe = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_XSUM_IPE_M, + csum.ipe = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_XSUM_IPE_M, + qword1); + csum.eipe = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_XSUM_EIPE_M, qword1); - csum->eipe = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_XSUM_EIPE_M, + csum.l4e = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_XSUM_L4E_M, + qword1); + csum.l3l4p = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_L3L4P_M, qword1); - csum->l4e = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_XSUM_L4E_M, - qword1); - csum->l3l4p = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_L3L4P_M, - qword1); - csum->ipv6exadd = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_IPV6EXADD_M, - qword0); - csum->raw_csum_inv = + csum.ipv6exadd = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_IPV6EXADD_M, + qword0); + csum.raw_csum_inv = le16_get_bits(rx_desc->ptype_err_fflags0, VIRTCHNL2_RX_FLEX_DESC_ADV_RAW_CSUM_INV_M); - csum->raw_csum = le16_to_cpu(rx_desc->misc.raw_cs); + csum.raw_csum = le16_to_cpu(rx_desc->misc.raw_cs); + + return csum; } /** @@ -3046,21 +3014,22 @@ idpf_rx_splitq_extract_csum_bits(const struct virtchnl2_rx_flex_desc_adv_nic_3 * */ static int idpf_rx_rsc(struct idpf_rx_queue *rxq, struct sk_buff *skb, const struct virtchnl2_rx_flex_desc_adv_nic_3 *rx_desc, - struct idpf_rx_ptype_decoded *decoded) + struct libeth_rx_pt decoded) { u16 rsc_segments, rsc_seg_len; bool ipv4, ipv6; int len; - if (unlikely(!decoded->outer_ip)) + if (unlikely(libeth_rx_pt_get_ip_ver(decoded) == + LIBETH_RX_PT_OUTER_L2)) return -EINVAL; rsc_seg_len = le16_to_cpu(rx_desc->misc.rscseglen); if (unlikely(!rsc_seg_len)) return -EINVAL; - ipv4 = IDPF_RX_PTYPE_TO_IPV(decoded, IDPF_RX_PTYPE_OUTER_IPV4); - ipv6 = IDPF_RX_PTYPE_TO_IPV(decoded, IDPF_RX_PTYPE_OUTER_IPV6); + ipv4 = libeth_rx_pt_get_ip_ver(decoded) == LIBETH_RX_PT_OUTER_IPV4; + ipv6 = libeth_rx_pt_get_ip_ver(decoded) == LIBETH_RX_PT_OUTER_IPV6; if (unlikely(!(ipv4 ^ ipv6))) return -EINVAL; @@ -3118,8 +3087,8 @@ static int idpf_rx_process_skb_fields(struct idpf_rx_queue *rxq, struct sk_buff *skb, const struct virtchnl2_rx_flex_desc_adv_nic_3 *rx_desc) { - struct idpf_rx_csum_decoded csum_bits = { }; - struct idpf_rx_ptype_decoded decoded; + struct idpf_rx_csum_decoded csum_bits; + struct libeth_rx_pt decoded; u16 rx_ptype; rx_ptype = le16_get_bits(rx_desc->ptype_err_fflags0, @@ -3127,16 +3096,16 @@ idpf_rx_process_skb_fields(struct idpf_rx_queue *rxq, struct sk_buff *skb, decoded = rxq->rx_ptype_lkup[rx_ptype]; /* process RSS/hash */ - idpf_rx_hash(rxq, skb, rx_desc, &decoded); + idpf_rx_hash(rxq, skb, rx_desc, decoded); skb->protocol = eth_type_trans(skb, rxq->netdev); if (le16_get_bits(rx_desc->hdrlen_flags, VIRTCHNL2_RX_FLEX_DESC_ADV_RSC_M)) - return idpf_rx_rsc(rxq, skb, rx_desc, &decoded); + return idpf_rx_rsc(rxq, skb, rx_desc, decoded); - idpf_rx_splitq_extract_csum_bits(rx_desc, &csum_bits); - idpf_rx_csum(rxq, skb, &csum_bits, &decoded); + csum_bits = idpf_rx_splitq_extract_csum_bits(rx_desc); + idpf_rx_csum(rxq, skb, csum_bits, decoded); return 0; } diff --git a/drivers/net/ethernet/intel/idpf/idpf_txrx.h b/drivers/net/ethernet/intel/idpf/idpf_txrx.h index b2bf58146484..98d47728cc8c 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_txrx.h +++ b/drivers/net/ethernet/intel/idpf/idpf_txrx.h @@ -331,72 +331,6 @@ struct idpf_rx_buf { #define IDPF_RX_MAX_BASE_PTYPE 256 #define IDPF_INVALID_PTYPE_ID 0xFFFF -/* Packet type non-ip values */ -enum idpf_rx_ptype_l2 { - IDPF_RX_PTYPE_L2_RESERVED = 0, - IDPF_RX_PTYPE_L2_MAC_PAY2 = 1, - IDPF_RX_PTYPE_L2_TIMESYNC_PAY2 = 2, - IDPF_RX_PTYPE_L2_FIP_PAY2 = 3, - IDPF_RX_PTYPE_L2_OUI_PAY2 = 4, - IDPF_RX_PTYPE_L2_MACCNTRL_PAY2 = 5, - IDPF_RX_PTYPE_L2_LLDP_PAY2 = 6, - IDPF_RX_PTYPE_L2_ECP_PAY2 = 7, - IDPF_RX_PTYPE_L2_EVB_PAY2 = 8, - IDPF_RX_PTYPE_L2_QCN_PAY2 = 9, - IDPF_RX_PTYPE_L2_EAPOL_PAY2 = 10, - IDPF_RX_PTYPE_L2_ARP = 11, -}; - -enum idpf_rx_ptype_outer_ip { - IDPF_RX_PTYPE_OUTER_L2 = 0, - IDPF_RX_PTYPE_OUTER_IP = 1, -}; - -#define IDPF_RX_PTYPE_TO_IPV(ptype, ipv) \ - (((ptype)->outer_ip == IDPF_RX_PTYPE_OUTER_IP) && \ - ((ptype)->outer_ip_ver == (ipv))) - -enum idpf_rx_ptype_outer_ip_ver { - IDPF_RX_PTYPE_OUTER_NONE = 0, - IDPF_RX_PTYPE_OUTER_IPV4 = 1, - IDPF_RX_PTYPE_OUTER_IPV6 = 2, -}; - -enum idpf_rx_ptype_outer_fragmented { - IDPF_RX_PTYPE_NOT_FRAG = 0, - IDPF_RX_PTYPE_FRAG = 1, -}; - -enum idpf_rx_ptype_tunnel_type { - IDPF_RX_PTYPE_TUNNEL_NONE = 0, - IDPF_RX_PTYPE_TUNNEL_IP_IP = 1, - IDPF_RX_PTYPE_TUNNEL_IP_GRENAT = 2, - IDPF_RX_PTYPE_TUNNEL_IP_GRENAT_MAC = 3, - IDPF_RX_PTYPE_TUNNEL_IP_GRENAT_MAC_VLAN = 4, -}; - -enum idpf_rx_ptype_tunnel_end_prot { - IDPF_RX_PTYPE_TUNNEL_END_NONE = 0, - IDPF_RX_PTYPE_TUNNEL_END_IPV4 = 1, - IDPF_RX_PTYPE_TUNNEL_END_IPV6 = 2, -}; - -enum idpf_rx_ptype_inner_prot { - IDPF_RX_PTYPE_INNER_PROT_NONE = 0, - IDPF_RX_PTYPE_INNER_PROT_UDP = 1, - IDPF_RX_PTYPE_INNER_PROT_TCP = 2, - IDPF_RX_PTYPE_INNER_PROT_SCTP = 3, - IDPF_RX_PTYPE_INNER_PROT_ICMP = 4, - IDPF_RX_PTYPE_INNER_PROT_TIMESYNC = 5, -}; - -enum idpf_rx_ptype_payload_layer { - IDPF_RX_PTYPE_PAYLOAD_LAYER_NONE = 0, - IDPF_RX_PTYPE_PAYLOAD_LAYER_PAY2 = 1, - IDPF_RX_PTYPE_PAYLOAD_LAYER_PAY3 = 2, - IDPF_RX_PTYPE_PAYLOAD_LAYER_PAY4 = 3, -}; - enum idpf_tunnel_state { IDPF_PTYPE_TUNNEL_IP = BIT(0), IDPF_PTYPE_TUNNEL_IP_GRENAT = BIT(1), @@ -404,22 +338,9 @@ enum idpf_tunnel_state { }; struct idpf_ptype_state { - bool outer_ip; - bool outer_frag; - u8 tunnel_state; -}; - -struct idpf_rx_ptype_decoded { - u32 ptype:10; - u32 known:1; - u32 outer_ip:1; - u32 outer_ip_ver:2; - u32 outer_frag:1; - u32 tunnel_type:3; - u32 tunnel_end_prot:2; - u32 tunnel_end_frag:1; - u32 inner_prot:4; - u32 payload_layer:3; + bool outer_ip:1; + bool outer_frag:1; + u8 tunnel_state:6; }; /** @@ -677,7 +598,7 @@ struct idpf_rx_queue { u16 desc_count; u32 rxdids; - const struct idpf_rx_ptype_decoded *rx_ptype_lkup; + const struct libeth_rx_pt *rx_ptype_lkup; __cacheline_group_end_aligned(read_mostly); __cacheline_group_begin_aligned(read_write); @@ -1180,7 +1101,6 @@ void idpf_vport_intr_update_itr_ena_irq(struct idpf_q_vector *q_vector); void idpf_vport_intr_deinit(struct idpf_vport *vport); int idpf_vport_intr_init(struct idpf_vport *vport); void idpf_vport_intr_ena(struct idpf_vport *vport); -enum pkt_hash_types idpf_ptype_to_htype(const struct idpf_rx_ptype_decoded *decoded); int idpf_config_rss(struct idpf_vport *vport); int idpf_init_rss(struct idpf_vport *vport); void idpf_deinit_rss(struct idpf_vport *vport); diff --git a/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c b/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c index 1aa4770dfe18..83f3543a30e1 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c +++ b/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c @@ -1,6 +1,8 @@ // SPDX-License-Identifier: GPL-2.0-only /* Copyright (C) 2023 Intel Corporation */ +#include + #include "idpf.h" #include "idpf_virtchnl.h" @@ -2483,39 +2485,52 @@ do_memcpy: * @frag: fragmentation allowed * */ -static void idpf_fill_ptype_lookup(struct idpf_rx_ptype_decoded *ptype, +static void idpf_fill_ptype_lookup(struct libeth_rx_pt *ptype, struct idpf_ptype_state *pstate, bool ipv4, bool frag) { if (!pstate->outer_ip || !pstate->outer_frag) { - ptype->outer_ip = IDPF_RX_PTYPE_OUTER_IP; pstate->outer_ip = true; if (ipv4) - ptype->outer_ip_ver = IDPF_RX_PTYPE_OUTER_IPV4; + ptype->outer_ip = LIBETH_RX_PT_OUTER_IPV4; else - ptype->outer_ip_ver = IDPF_RX_PTYPE_OUTER_IPV6; + ptype->outer_ip = LIBETH_RX_PT_OUTER_IPV6; if (frag) { - ptype->outer_frag = IDPF_RX_PTYPE_FRAG; + ptype->outer_frag = LIBETH_RX_PT_FRAG; pstate->outer_frag = true; } } else { - ptype->tunnel_type = IDPF_RX_PTYPE_TUNNEL_IP_IP; + ptype->tunnel_type = LIBETH_RX_PT_TUNNEL_IP_IP; pstate->tunnel_state = IDPF_PTYPE_TUNNEL_IP; if (ipv4) - ptype->tunnel_end_prot = - IDPF_RX_PTYPE_TUNNEL_END_IPV4; + ptype->tunnel_end_prot = LIBETH_RX_PT_TUNNEL_END_IPV4; else - ptype->tunnel_end_prot = - IDPF_RX_PTYPE_TUNNEL_END_IPV6; + ptype->tunnel_end_prot = LIBETH_RX_PT_TUNNEL_END_IPV6; if (frag) - ptype->tunnel_end_frag = IDPF_RX_PTYPE_FRAG; + ptype->tunnel_end_frag = LIBETH_RX_PT_FRAG; } } +static void idpf_finalize_ptype_lookup(struct libeth_rx_pt *ptype) +{ + if (ptype->payload_layer == LIBETH_RX_PT_PAYLOAD_L2 && + ptype->inner_prot) + ptype->payload_layer = LIBETH_RX_PT_PAYLOAD_L4; + else if (ptype->payload_layer == LIBETH_RX_PT_PAYLOAD_L2 && + ptype->outer_ip) + ptype->payload_layer = LIBETH_RX_PT_PAYLOAD_L3; + else if (ptype->outer_ip == LIBETH_RX_PT_OUTER_L2) + ptype->payload_layer = LIBETH_RX_PT_PAYLOAD_L2; + else + ptype->payload_layer = LIBETH_RX_PT_PAYLOAD_NONE; + + libeth_rx_pt_gen_hash_type(ptype); +} + /** * idpf_send_get_rx_ptype_msg - Send virtchnl for ptype info * @vport: virtual port data structure @@ -2526,7 +2541,7 @@ int idpf_send_get_rx_ptype_msg(struct idpf_vport *vport) { struct virtchnl2_get_ptype_info *get_ptype_info __free(kfree) = NULL; struct virtchnl2_get_ptype_info *ptype_info __free(kfree) = NULL; - struct idpf_rx_ptype_decoded *ptype_lkup = vport->rx_ptype_lkup; + struct libeth_rx_pt *ptype_lkup __free(kfree) = NULL; int max_ptype, ptypes_recvd = 0, ptype_offset; struct idpf_adapter *adapter = vport->adapter; struct idpf_vc_xn_params xn_params = {}; @@ -2534,12 +2549,17 @@ int idpf_send_get_rx_ptype_msg(struct idpf_vport *vport) ssize_t reply_sz; int i, j, k; + if (vport->rx_ptype_lkup) + return 0; + if (idpf_is_queue_model_split(vport->rxq_model)) max_ptype = IDPF_RX_MAX_PTYPE; else max_ptype = IDPF_RX_MAX_BASE_PTYPE; - memset(vport->rx_ptype_lkup, 0, sizeof(vport->rx_ptype_lkup)); + ptype_lkup = kcalloc(max_ptype, sizeof(*ptype_lkup), GFP_KERNEL); + if (!ptype_lkup) + return -ENOMEM; get_ptype_info = kzalloc(sizeof(*get_ptype_info), GFP_KERNEL); if (!get_ptype_info) @@ -2597,16 +2617,13 @@ int idpf_send_get_rx_ptype_msg(struct idpf_vport *vport) /* 0xFFFF indicates end of ptypes */ if (le16_to_cpu(ptype->ptype_id_10) == IDPF_INVALID_PTYPE_ID) - return 0; + goto out; if (idpf_is_queue_model_split(vport->rxq_model)) k = le16_to_cpu(ptype->ptype_id_10); else k = ptype->ptype_id_8; - if (ptype->proto_id_count) - ptype_lkup[k].known = 1; - for (j = 0; j < ptype->proto_id_count; j++) { id = le16_to_cpu(ptype->proto_id[j]); switch (id) { @@ -2614,18 +2631,18 @@ int idpf_send_get_rx_ptype_msg(struct idpf_vport *vport) if (pstate.tunnel_state == IDPF_PTYPE_TUNNEL_IP) { ptype_lkup[k].tunnel_type = - IDPF_RX_PTYPE_TUNNEL_IP_GRENAT; + LIBETH_RX_PT_TUNNEL_IP_GRENAT; pstate.tunnel_state |= IDPF_PTYPE_TUNNEL_IP_GRENAT; } break; case VIRTCHNL2_PROTO_HDR_MAC: ptype_lkup[k].outer_ip = - IDPF_RX_PTYPE_OUTER_L2; + LIBETH_RX_PT_OUTER_L2; if (pstate.tunnel_state == IDPF_TUN_IP_GRE) { ptype_lkup[k].tunnel_type = - IDPF_RX_PTYPE_TUNNEL_IP_GRENAT_MAC; + LIBETH_RX_PT_TUNNEL_IP_GRENAT_MAC; pstate.tunnel_state |= IDPF_PTYPE_TUNNEL_IP_GRENAT_MAC; } @@ -2652,23 +2669,23 @@ int idpf_send_get_rx_ptype_msg(struct idpf_vport *vport) break; case VIRTCHNL2_PROTO_HDR_UDP: ptype_lkup[k].inner_prot = - IDPF_RX_PTYPE_INNER_PROT_UDP; + LIBETH_RX_PT_INNER_UDP; break; case VIRTCHNL2_PROTO_HDR_TCP: ptype_lkup[k].inner_prot = - IDPF_RX_PTYPE_INNER_PROT_TCP; + LIBETH_RX_PT_INNER_TCP; break; case VIRTCHNL2_PROTO_HDR_SCTP: ptype_lkup[k].inner_prot = - IDPF_RX_PTYPE_INNER_PROT_SCTP; + LIBETH_RX_PT_INNER_SCTP; break; case VIRTCHNL2_PROTO_HDR_ICMP: ptype_lkup[k].inner_prot = - IDPF_RX_PTYPE_INNER_PROT_ICMP; + LIBETH_RX_PT_INNER_ICMP; break; case VIRTCHNL2_PROTO_HDR_PAY: ptype_lkup[k].payload_layer = - IDPF_RX_PTYPE_PAYLOAD_LAYER_PAY2; + LIBETH_RX_PT_PAYLOAD_L2; break; case VIRTCHNL2_PROTO_HDR_ICMPV6: case VIRTCHNL2_PROTO_HDR_IPV6_EH: @@ -2722,9 +2739,14 @@ int idpf_send_get_rx_ptype_msg(struct idpf_vport *vport) break; } } + + idpf_finalize_ptype_lookup(&ptype_lkup[k]); } } +out: + vport->rx_ptype_lkup = no_free_ptr(ptype_lkup); + return 0; } -- cgit v1.2.3-58-ga151 From 90912f9f4f2d159e93cf9a4c732506342489a1d5 Mon Sep 17 00:00:00 2001 From: Alexander Lobakin Date: Thu, 20 Jun 2024 15:53:46 +0200 Subject: idpf: convert header split mode to libeth + napi_build_skb() Currently, idpf uses the following model for the header buffers: * buffers are allocated via dma_alloc_coherent(); * when receiving, napi_alloc_skb() is called and then the header is copied to the newly allocated linear part. This is far from optimal as DMA coherent zone is slow on many systems and memcpy() neutralizes the idea and benefits of the header split. Not speaking of that XDP can't be run on DMA coherent buffers, but at the same time the idea of allocating an skb to run XDP program is ill. Instead, use libeth to create page_pools for the header buffers, allocate them dynamically and then build an skb via napi_build_skb() around them with no memory copy. With one exception... When you enable header split, you expect you'll always have a separate header buffer, so that you could reserve headroom and tailroom only there and then use full buffers for the data. For example, this is how TCP zerocopy works -- you have to have the payload aligned to PAGE_SIZE. The current hardware running idpf does *not* guarantee that you'll always have headers placed separately. For example, on my setup, even ICMP packets are written as one piece to the data buffers. You can't build a valid skb around a data buffer in this case. To not complicate things and not lose TCP zerocopy etc., when such thing happens, use the empty header buffer and pull either full frame (if it's short) or the Ethernet header there and build an skb around it. GRO layer will pull more from the data buffer later. This W/A will hopefully be removed one day. Signed-off-by: Alexander Lobakin Signed-off-by: Tony Nguyen --- .../net/ethernet/intel/idpf/idpf_singleq_txrx.c | 1 + drivers/net/ethernet/intel/idpf/idpf_txrx.c | 253 ++++++++++++++------- drivers/net/ethernet/intel/idpf/idpf_txrx.h | 52 +++-- drivers/net/ethernet/intel/idpf/idpf_virtchnl.c | 14 +- 4 files changed, 204 insertions(+), 116 deletions(-) (limited to 'drivers/net/ethernet/intel/idpf/idpf_virtchnl.c') diff --git a/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c b/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c index 6c0794e01922..ced1fdf7a1fb 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c +++ b/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c @@ -827,6 +827,7 @@ idpf_rx_singleq_process_skb_fields(struct idpf_rx_queue *rx_q, } idpf_rx_singleq_csum(rx_q, skb, csum_bits, decoded); + skb_record_rx_queue(skb, rx_q->idx); } /** diff --git a/drivers/net/ethernet/intel/idpf/idpf_txrx.c b/drivers/net/ethernet/intel/idpf/idpf_txrx.c index 883e58815f92..ec4a9501d19f 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_txrx.c +++ b/drivers/net/ethernet/intel/idpf/idpf_txrx.c @@ -381,7 +381,7 @@ err_out: * idpf_rx_page_rel - Release an rx buffer page * @rx_buf: the buffer to free */ -static void idpf_rx_page_rel(struct idpf_rx_buf *rx_buf) +static void idpf_rx_page_rel(struct libeth_fqe *rx_buf) { if (unlikely(!rx_buf->page)) return; @@ -389,46 +389,50 @@ static void idpf_rx_page_rel(struct idpf_rx_buf *rx_buf) page_pool_put_full_page(rx_buf->page->pp, rx_buf->page, false); rx_buf->page = NULL; - rx_buf->page_offset = 0; + rx_buf->offset = 0; } /** * idpf_rx_hdr_buf_rel_all - Release header buffer memory * @bufq: queue to use - * @dev: device to free DMA memory */ -static void idpf_rx_hdr_buf_rel_all(struct idpf_buf_queue *bufq, - struct device *dev) +static void idpf_rx_hdr_buf_rel_all(struct idpf_buf_queue *bufq) { - dma_free_coherent(dev, bufq->desc_count * IDPF_HDR_BUF_SIZE, - bufq->rx_buf.hdr_buf_va, bufq->rx_buf.hdr_buf_pa); - bufq->rx_buf.hdr_buf_va = NULL; + struct libeth_fq fq = { + .fqes = bufq->hdr_buf, + .pp = bufq->hdr_pp, + }; + + for (u32 i = 0; i < bufq->desc_count; i++) + idpf_rx_page_rel(&bufq->hdr_buf[i]); + + libeth_rx_fq_destroy(&fq); + bufq->hdr_buf = NULL; + bufq->hdr_pp = NULL; } /** * idpf_rx_buf_rel_bufq - Free all Rx buffer resources for a buffer queue * @bufq: queue to be cleaned - * @dev: device to free DMA memory */ -static void idpf_rx_buf_rel_bufq(struct idpf_buf_queue *bufq, - struct device *dev) +static void idpf_rx_buf_rel_bufq(struct idpf_buf_queue *bufq) { /* queue already cleared, nothing to do */ - if (!bufq->rx_buf.buf) + if (!bufq->buf) return; /* Free all the bufs allocated and given to hw on Rx queue */ for (u32 i = 0; i < bufq->desc_count; i++) - idpf_rx_page_rel(&bufq->rx_buf.buf[i]); + idpf_rx_page_rel(&bufq->buf[i]); if (idpf_queue_has(HSPLIT_EN, bufq)) - idpf_rx_hdr_buf_rel_all(bufq, dev); + idpf_rx_hdr_buf_rel_all(bufq); page_pool_destroy(bufq->pp); bufq->pp = NULL; - kfree(bufq->rx_buf.buf); - bufq->rx_buf.buf = NULL; + kfree(bufq->buf); + bufq->buf = NULL; } /** @@ -493,7 +497,7 @@ static void idpf_rx_desc_rel_bufq(struct idpf_buf_queue *bufq, if (!bufq) return; - idpf_rx_buf_rel_bufq(bufq, dev); + idpf_rx_buf_rel_bufq(bufq); bufq->next_to_alloc = 0; bufq->next_to_clean = 0; @@ -573,12 +577,21 @@ static void idpf_rx_buf_hw_update(struct idpf_buf_queue *bufq, u32 val) */ static int idpf_rx_hdr_buf_alloc_all(struct idpf_buf_queue *bufq) { - bufq->rx_buf.hdr_buf_va = - dma_alloc_coherent(bufq->q_vector->vport->netdev->dev.parent, - IDPF_HDR_BUF_SIZE * bufq->desc_count, - &bufq->rx_buf.hdr_buf_pa, GFP_KERNEL); - if (!bufq->rx_buf.hdr_buf_va) - return -ENOMEM; + struct libeth_fq fq = { + .count = bufq->desc_count, + .type = LIBETH_FQE_HDR, + .nid = idpf_q_vector_to_mem(bufq->q_vector), + }; + int ret; + + ret = libeth_rx_fq_create(&fq, &bufq->q_vector->napi); + if (ret) + return ret; + + bufq->hdr_pp = fq.pp; + bufq->hdr_buf = fq.fqes; + bufq->hdr_truesize = fq.truesize; + bufq->rx_hbuf_size = fq.buf_len; return 0; } @@ -616,17 +629,27 @@ static void idpf_rx_post_buf_refill(struct idpf_sw_queue *refillq, u16 buf_id) static bool idpf_rx_post_buf_desc(struct idpf_buf_queue *bufq, u16 buf_id) { struct virtchnl2_splitq_rx_buf_desc *splitq_rx_desc = NULL; + struct libeth_fq_fp fq = { + .count = bufq->desc_count, + }; u16 nta = bufq->next_to_alloc; struct idpf_rx_buf *buf; dma_addr_t addr; splitq_rx_desc = &bufq->split_buf[nta]; - buf = &bufq->rx_buf.buf[buf_id]; + buf = &bufq->buf[buf_id]; - if (idpf_queue_has(HSPLIT_EN, bufq)) - splitq_rx_desc->hdr_addr = - cpu_to_le64(bufq->rx_buf.hdr_buf_pa + - (u32)buf_id * IDPF_HDR_BUF_SIZE); + if (idpf_queue_has(HSPLIT_EN, bufq)) { + fq.pp = bufq->hdr_pp; + fq.fqes = bufq->hdr_buf; + fq.truesize = bufq->hdr_truesize; + + addr = libeth_rx_alloc(&fq, buf_id); + if (addr == DMA_MAPPING_ERROR) + return false; + + splitq_rx_desc->hdr_addr = cpu_to_le64(addr); + } addr = idpf_alloc_page(bufq->pp, buf, bufq->rx_buf_size); if (unlikely(addr == DMA_MAPPING_ERROR)) @@ -741,13 +764,12 @@ static int idpf_rx_bufs_init_singleq(struct idpf_rx_queue *rxq) */ static int idpf_rx_buf_alloc_all(struct idpf_buf_queue *rxbufq) { - struct device *dev = rxbufq->q_vector->vport->netdev->dev.parent; int err = 0; /* Allocate book keeping buffers */ - rxbufq->rx_buf.buf = kcalloc(rxbufq->desc_count, - sizeof(struct idpf_rx_buf), GFP_KERNEL); - if (!rxbufq->rx_buf.buf) { + rxbufq->buf = kcalloc(rxbufq->desc_count, sizeof(*rxbufq->buf), + GFP_KERNEL); + if (!rxbufq->buf) { err = -ENOMEM; goto rx_buf_alloc_all_out; } @@ -764,7 +786,7 @@ static int idpf_rx_buf_alloc_all(struct idpf_buf_queue *rxbufq) rx_buf_alloc_all_out: if (err) - idpf_rx_buf_rel_bufq(rxbufq, dev); + idpf_rx_buf_rel_bufq(rxbufq); return err; } @@ -1489,7 +1511,6 @@ static int idpf_rxq_group_alloc(struct idpf_vport *vport, u16 num_rxq) q->rx_buffer_low_watermark = IDPF_LOW_WATERMARK; idpf_queue_assign(HSPLIT_EN, q, hs); - q->rx_hbuf_size = hs ? IDPF_HDR_BUF_SIZE : 0; bufq_set->num_refillqs = num_rxq; bufq_set->refillqs = kcalloc(num_rxq, swq_size, @@ -1532,7 +1553,6 @@ skip_splitq_rx_init: &rx_qgrp->splitq.bufq_sets[1].refillqs[j]; idpf_queue_assign(HSPLIT_EN, q, hs); - q->rx_hbuf_size = hs ? IDPF_HDR_BUF_SIZE : 0; setup_rxq: q->desc_count = vport->rxq_desc_count; @@ -3107,6 +3127,8 @@ idpf_rx_process_skb_fields(struct idpf_rx_queue *rxq, struct sk_buff *skb, csum_bits = idpf_rx_splitq_extract_csum_bits(rx_desc); idpf_rx_csum(rxq, skb, csum_bits, decoded); + skb_record_rx_queue(skb, rxq->idx); + return 0; } @@ -3124,7 +3146,7 @@ void idpf_rx_add_frag(struct idpf_rx_buf *rx_buf, struct sk_buff *skb, unsigned int size) { skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buf->page, - rx_buf->page_offset, size, rx_buf->truesize); + rx_buf->offset, size, rx_buf->truesize); rx_buf->page = NULL; } @@ -3147,7 +3169,7 @@ struct sk_buff *idpf_rx_construct_skb(const struct idpf_rx_queue *rxq, struct sk_buff *skb; void *va; - va = page_address(rx_buf->page) + rx_buf->page_offset; + va = page_address(rx_buf->page) + rx_buf->offset; /* prefetch first cache line of first page */ net_prefetch(va); @@ -3159,7 +3181,6 @@ struct sk_buff *idpf_rx_construct_skb(const struct idpf_rx_queue *rxq, return NULL; } - skb_record_rx_queue(skb, rxq->idx); skb_mark_for_recycle(skb); /* Determine available headroom for copy */ @@ -3178,7 +3199,7 @@ struct sk_buff *idpf_rx_construct_skb(const struct idpf_rx_queue *rxq, return skb; } - skb_add_rx_frag(skb, 0, rx_buf->page, rx_buf->page_offset + headlen, + skb_add_rx_frag(skb, 0, rx_buf->page, rx_buf->offset + headlen, size, rx_buf->truesize); /* Since we're giving the page to the stack, clear our reference to it. @@ -3190,36 +3211,66 @@ struct sk_buff *idpf_rx_construct_skb(const struct idpf_rx_queue *rxq, } /** - * idpf_rx_hdr_construct_skb - Allocate skb and populate it from header buffer - * @rxq: Rx descriptor queue - * @va: Rx buffer to pull data from + * idpf_rx_hsplit_wa - handle header buffer overflows and split errors + * @hdr: Rx buffer for the headers + * @buf: Rx buffer for the payload + * @data_len: number of bytes received to the payload buffer + * + * When a header buffer overflow occurs or the HW was unable do parse the + * packet type to perform header split, the whole frame gets placed to the + * payload buffer. We can't build a valid skb around a payload buffer when + * the header split is active since it doesn't reserve any head- or tailroom. + * In that case, copy either the whole frame when it's short or just the + * Ethernet header to the header buffer to be able to build an skb and adjust + * the data offset in the payload buffer, IOW emulate the header split. + * + * Return: number of bytes copied to the header buffer. + */ +static u32 idpf_rx_hsplit_wa(const struct libeth_fqe *hdr, + struct libeth_fqe *buf, u32 data_len) +{ + u32 copy = data_len <= L1_CACHE_BYTES ? data_len : ETH_HLEN; + const void *src; + void *dst; + + if (!libeth_rx_sync_for_cpu(buf, copy)) + return 0; + + dst = page_address(hdr->page) + hdr->offset + hdr->page->pp->p.offset; + src = page_address(buf->page) + buf->offset + buf->page->pp->p.offset; + memcpy(dst, src, LARGEST_ALIGN(copy)); + + buf->offset += copy; + + return copy; +} + +/** + * idpf_rx_build_skb - Allocate skb and populate it from header buffer + * @buf: Rx buffer to pull data from * @size: the length of the packet * * This function allocates an skb. It then populates it with the page data from * the current receive descriptor, taking care to set up the skb correctly. - * This specifically uses a header buffer to start building the skb. */ -static struct sk_buff * -idpf_rx_hdr_construct_skb(const struct idpf_rx_queue *rxq, const void *va, - unsigned int size) +struct sk_buff *idpf_rx_build_skb(const struct libeth_fqe *buf, u32 size) { + u32 hr = buf->page->pp->p.offset; struct sk_buff *skb; + void *va; - /* allocate a skb to store the frags */ - skb = napi_alloc_skb(rxq->napi, size); + va = page_address(buf->page) + buf->offset; + prefetch(va + hr); + + skb = napi_build_skb(va, buf->truesize); if (unlikely(!skb)) return NULL; - skb_record_rx_queue(skb, rxq->idx); - - memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long))); - - /* More than likely, a payload fragment, which will use a page from - * page_pool will be added to the SKB so mark it for recycle - * preemptively. And if not, it's inconsequential. - */ skb_mark_for_recycle(skb); + skb_reserve(skb, hr); + __skb_put(skb, size); + return skb; } @@ -3272,14 +3323,12 @@ static int idpf_rx_splitq_clean(struct idpf_rx_queue *rxq, int budget) /* Process Rx packets bounded by budget */ while (likely(total_rx_pkts < budget)) { struct virtchnl2_rx_flex_desc_adv_nic_3 *rx_desc; + struct libeth_fqe *hdr, *rx_buf = NULL; struct idpf_sw_queue *refillq = NULL; struct idpf_rxq_set *rxq_set = NULL; - struct idpf_rx_buf *rx_buf = NULL; unsigned int pkt_len = 0; unsigned int hdr_len = 0; u16 gen_id, buf_id = 0; - /* Header buffer overflow only valid for header split */ - bool hbo = false; int bufq_id; u8 rxdid; @@ -3311,25 +3360,6 @@ static int idpf_rx_splitq_clean(struct idpf_rx_queue *rxq, int budget) pkt_len = le16_get_bits(rx_desc->pktlen_gen_bufq_id, VIRTCHNL2_RX_FLEX_DESC_ADV_LEN_PBUF_M); - hbo = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_HBO_M, - rx_desc->status_err0_qw1); - - if (unlikely(hbo)) { - /* If a header buffer overflow, occurs, i.e. header is - * too large to fit in the header split buffer, HW will - * put the entire packet, including headers, in the - * data/payload buffer. - */ - u64_stats_update_begin(&rxq->stats_sync); - u64_stats_inc(&rxq->q_stats.hsplit_buf_ovf); - u64_stats_update_end(&rxq->stats_sync); - goto bypass_hsplit; - } - - hdr_len = le16_get_bits(rx_desc->hdrlen_flags, - VIRTCHNL2_RX_FLEX_DESC_ADV_LEN_HDR_M); - -bypass_hsplit: bufq_id = le16_get_bits(rx_desc->pktlen_gen_bufq_id, VIRTCHNL2_RX_FLEX_DESC_ADV_BUFQ_ID_M); @@ -3341,18 +3371,48 @@ bypass_hsplit: buf_id = le16_to_cpu(rx_desc->buf_id); - rx_buf = &rx_bufq->rx_buf.buf[buf_id]; + rx_buf = &rx_bufq->buf[buf_id]; - if (hdr_len) { - const void *va = (u8 *)rx_bufq->rx_buf.hdr_buf_va + - (u32)buf_id * IDPF_HDR_BUF_SIZE; + if (!rx_bufq->hdr_pp) + goto payload; + +#define __HBO_BIT VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_HBO_M +#define __HDR_LEN_MASK VIRTCHNL2_RX_FLEX_DESC_ADV_LEN_HDR_M + if (likely(!(rx_desc->status_err0_qw1 & __HBO_BIT))) + /* If a header buffer overflow, occurs, i.e. header is + * too large to fit in the header split buffer, HW will + * put the entire packet, including headers, in the + * data/payload buffer. + */ + hdr_len = le16_get_bits(rx_desc->hdrlen_flags, + __HDR_LEN_MASK); +#undef __HDR_LEN_MASK +#undef __HBO_BIT + + hdr = &rx_bufq->hdr_buf[buf_id]; + + if (unlikely(!hdr_len && !skb)) { + hdr_len = idpf_rx_hsplit_wa(hdr, rx_buf, pkt_len); + pkt_len -= hdr_len; + + u64_stats_update_begin(&rxq->stats_sync); + u64_stats_inc(&rxq->q_stats.hsplit_buf_ovf); + u64_stats_update_end(&rxq->stats_sync); + } + + if (libeth_rx_sync_for_cpu(hdr, hdr_len)) { + skb = idpf_rx_build_skb(hdr, hdr_len); + if (!skb) + break; - skb = idpf_rx_hdr_construct_skb(rxq, va, hdr_len); u64_stats_update_begin(&rxq->stats_sync); u64_stats_inc(&rxq->q_stats.hsplit_pkts); u64_stats_update_end(&rxq->stats_sync); } + hdr->page = NULL; + +payload: if (pkt_len) { idpf_rx_sync_for_cpu(rx_buf, pkt_len); if (skb) @@ -3422,10 +3482,13 @@ bypass_hsplit: static int idpf_rx_update_bufq_desc(struct idpf_buf_queue *bufq, u32 buf_id, struct virtchnl2_splitq_rx_buf_desc *buf_desc) { + struct libeth_fq_fp fq = { + .count = bufq->desc_count, + }; struct idpf_rx_buf *buf; dma_addr_t addr; - buf = &bufq->rx_buf.buf[buf_id]; + buf = &bufq->buf[buf_id]; addr = idpf_alloc_page(bufq->pp, buf, bufq->rx_buf_size); if (unlikely(addr == DMA_MAPPING_ERROR)) @@ -3437,8 +3500,15 @@ static int idpf_rx_update_bufq_desc(struct idpf_buf_queue *bufq, u32 buf_id, if (!idpf_queue_has(HSPLIT_EN, bufq)) return 0; - buf_desc->hdr_addr = cpu_to_le64(bufq->rx_buf.hdr_buf_pa + - (u32)buf_id * IDPF_HDR_BUF_SIZE); + fq.pp = bufq->hdr_pp; + fq.fqes = bufq->hdr_buf; + fq.truesize = bufq->hdr_truesize; + + addr = libeth_rx_alloc(&fq, buf_id); + if (addr == DMA_MAPPING_ERROR) + return -ENOMEM; + + buf_desc->hdr_addr = cpu_to_le64(addr); return 0; } @@ -3509,16 +3579,20 @@ static void idpf_rx_clean_refillq(struct idpf_buf_queue *bufq, /** * idpf_rx_clean_refillq_all - Clean all refill queues * @bufq: buffer queue with refill queues + * @nid: ID of the closest NUMA node with memory * * Iterates through all refill queues assigned to the buffer queue assigned to * this vector. Returns true if clean is complete within budget, false * otherwise. */ -static void idpf_rx_clean_refillq_all(struct idpf_buf_queue *bufq) +static void idpf_rx_clean_refillq_all(struct idpf_buf_queue *bufq, int nid) { struct idpf_bufq_set *bufq_set; int i; + if (bufq->hdr_pp) + page_pool_nid_changed(bufq->hdr_pp, nid); + bufq_set = container_of(bufq, struct idpf_bufq_set, bufq); for (i = 0; i < bufq_set->num_refillqs; i++) idpf_rx_clean_refillq(bufq, &bufq_set->refillqs[i]); @@ -4020,6 +4094,7 @@ static bool idpf_rx_splitq_clean_all(struct idpf_q_vector *q_vec, int budget, bool clean_complete = true; int pkts_cleaned = 0; int i, budget_per_q; + int nid; /* We attempt to distribute budget to each Rx queue fairly, but don't * allow the budget to go below 1 because that would exit polling early. @@ -4037,8 +4112,10 @@ static bool idpf_rx_splitq_clean_all(struct idpf_q_vector *q_vec, int budget, } *cleaned = pkts_cleaned; + nid = numa_mem_id(); + for (i = 0; i < q_vec->num_bufq; i++) - idpf_rx_clean_refillq_all(q_vec->bufq[i]); + idpf_rx_clean_refillq_all(q_vec->bufq[i], nid); return clean_complete; } diff --git a/drivers/net/ethernet/intel/idpf/idpf_txrx.h b/drivers/net/ethernet/intel/idpf/idpf_txrx.h index 98d47728cc8c..47d70ad0d6b4 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_txrx.h +++ b/drivers/net/ethernet/intel/idpf/idpf_txrx.h @@ -7,7 +7,7 @@ #include #include -#include +#include #include #include @@ -103,8 +103,6 @@ do { \ #define IDPF_RX_BUF_STRIDE 32 #define IDPF_RX_BUF_POST_STRIDE 16 #define IDPF_LOW_WATERMARK 64 -/* Size of header buffer specifically for header split */ -#define IDPF_HDR_BUF_SIZE 256 #define IDPF_PACKET_HDR_PAD \ (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN * 2) #define IDPF_TX_TSO_MIN_MSS 88 @@ -300,14 +298,7 @@ struct idpf_rx_extracted { #define IDPF_TX_MAX_DESC_DATA_ALIGNED \ ALIGN_DOWN(IDPF_TX_MAX_DESC_DATA, IDPF_TX_MAX_READ_REQ_SIZE) -#define IDPF_RX_DMA_ATTR \ - (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING) - -struct idpf_rx_buf { - struct page *page; - unsigned int page_offset; - u16 truesize; -}; +#define idpf_rx_buf libeth_fqe #define IDPF_RX_MAX_PTYPE_PROTO_IDS 32 #define IDPF_RX_MAX_PTYPE_SZ (sizeof(struct virtchnl2_ptype) + \ @@ -751,14 +742,14 @@ libeth_cacheline_set_assert(struct idpf_tx_queue, 64, /** * struct idpf_buf_queue - software structure representing a buffer queue * @split_buf: buffer descriptor array - * @rx_buf: Struct with RX buffer related members - * @rx_buf.buf: See struct idpf_rx_buf - * @rx_buf.hdr_buf_pa: DMA handle - * @rx_buf.hdr_buf_va: Virtual address - * @pp: Page pool pointer + * @hdr_buf: &libeth_fqe for header buffers + * @hdr_pp: &page_pool for header buffers + * @buf: &idpf_rx_buf for data buffers + * @pp: &page_pool for data buffers * @tail: Tail offset * @flags: See enum idpf_queue_flags_t * @desc_count: Number of descriptors + * @hdr_truesize: truesize for buffer headers * @next_to_use: Next descriptor to use * @next_to_clean: Next descriptor to clean * @next_to_alloc: RX buffer to allocate at @@ -773,16 +764,16 @@ libeth_cacheline_set_assert(struct idpf_tx_queue, 64, struct idpf_buf_queue { __cacheline_group_begin_aligned(read_mostly); struct virtchnl2_splitq_rx_buf_desc *split_buf; - struct { - struct idpf_rx_buf *buf; - dma_addr_t hdr_buf_pa; - void *hdr_buf_va; - } rx_buf; + struct libeth_fqe *hdr_buf; + struct page_pool *hdr_pp; + struct idpf_rx_buf *buf; struct page_pool *pp; void __iomem *tail; DECLARE_BITMAP(flags, __IDPF_Q_FLAGS_NBITS); u32 desc_count; + + u32 hdr_truesize; __cacheline_group_end_aligned(read_mostly); __cacheline_group_begin_aligned(read_write); @@ -976,6 +967,18 @@ struct idpf_txq_group { u32 num_completions_pending; }; +static inline int idpf_q_vector_to_mem(const struct idpf_q_vector *q_vector) +{ + u32 cpu; + + if (!q_vector) + return NUMA_NO_NODE; + + cpu = cpumask_first(q_vector->affinity_mask); + + return cpu < nr_cpu_ids ? cpu_to_mem(cpu) : NUMA_NO_NODE; +} + /** * idpf_size_to_txd_count - Get number of descriptors needed for large Tx frag * @size: transmit request size in bytes @@ -1044,7 +1047,7 @@ static inline dma_addr_t idpf_alloc_page(struct page_pool *pool, unsigned int buf_size) { if (buf_size == IDPF_RX_BUF_2048) - buf->page = page_pool_dev_alloc_frag(pool, &buf->page_offset, + buf->page = page_pool_dev_alloc_frag(pool, &buf->offset, buf_size); else buf->page = page_pool_dev_alloc_pages(pool); @@ -1054,7 +1057,7 @@ static inline dma_addr_t idpf_alloc_page(struct page_pool *pool, buf->truesize = buf_size; - return page_pool_get_dma_addr(buf->page) + buf->page_offset + + return page_pool_get_dma_addr(buf->page) + buf->offset + pool->p.offset; } @@ -1081,7 +1084,7 @@ static inline void idpf_rx_sync_for_cpu(struct idpf_rx_buf *rx_buf, u32 len) dma_sync_single_range_for_cpu(pp->p.dev, page_pool_get_dma_addr(page), - rx_buf->page_offset + pp->p.offset, len, + rx_buf->offset + pp->p.offset, len, page_pool_get_dma_dir(pp)); } @@ -1110,6 +1113,7 @@ void idpf_rx_add_frag(struct idpf_rx_buf *rx_buf, struct sk_buff *skb, struct sk_buff *idpf_rx_construct_skb(const struct idpf_rx_queue *rxq, struct idpf_rx_buf *rx_buf, unsigned int size); +struct sk_buff *idpf_rx_build_skb(const struct libeth_fqe *buf, u32 size); void idpf_tx_buf_hw_update(struct idpf_tx_queue *tx_q, u32 val, bool xmit_more); unsigned int idpf_size_to_txd_count(unsigned int size); diff --git a/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c b/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c index 83f3543a30e1..50dcb3ab02b1 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c +++ b/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c @@ -1604,32 +1604,38 @@ setup_rxqs: num_rxq = rx_qgrp->singleq.num_rxq; for (j = 0; j < num_rxq; j++, k++) { + const struct idpf_bufq_set *sets; struct idpf_rx_queue *rxq; if (!idpf_is_queue_model_split(vport->rxq_model)) { rxq = rx_qgrp->singleq.rxqs[j]; goto common_qi_fields; } + rxq = &rx_qgrp->splitq.rxq_sets[j]->rxq; - qi[k].rx_bufq1_id = - cpu_to_le16(rxq->bufq_sets[0].bufq.q_id); + sets = rxq->bufq_sets; + + qi[k].rx_bufq1_id = cpu_to_le16(sets[0].bufq.q_id); if (vport->num_bufqs_per_qgrp > IDPF_SINGLE_BUFQ_PER_RXQ_GRP) { qi[k].bufq2_ena = IDPF_BUFQ2_ENA; qi[k].rx_bufq2_id = - cpu_to_le16(rxq->bufq_sets[1].bufq.q_id); + cpu_to_le16(sets[1].bufq.q_id); } qi[k].rx_buffer_low_watermark = cpu_to_le16(rxq->rx_buffer_low_watermark); if (idpf_is_feature_ena(vport, NETIF_F_GRO_HW)) qi[k].qflags |= cpu_to_le16(VIRTCHNL2_RXQ_RSC); -common_qi_fields: + rxq->rx_hbuf_size = sets[0].bufq.rx_hbuf_size; + if (idpf_queue_has(HSPLIT_EN, rxq)) { qi[k].qflags |= cpu_to_le16(VIRTCHNL2_RXQ_HDR_SPLIT); qi[k].hdr_buffer_size = cpu_to_le16(rxq->rx_hbuf_size); } + +common_qi_fields: qi[k].queue_id = cpu_to_le32(rxq->q_id); qi[k].model = cpu_to_le16(vport->rxq_model); qi[k].type = cpu_to_le32(VIRTCHNL2_QUEUE_TYPE_RX); -- cgit v1.2.3-58-ga151 From 74d1412ac8f3719bc8dd08b5775276dd1cbc3997 Mon Sep 17 00:00:00 2001 From: Alexander Lobakin Date: Thu, 20 Jun 2024 15:53:47 +0200 Subject: idpf: use libeth Rx buffer management for payload buffer idpf uses Page Pool for data buffers with hardcoded buffer lengths of 4k for "classic" buffers and 2k for "short" ones. This is not flexible and does not ensure optimal memory usage. Why would you need 4k buffers when the MTU is 1500? Use libeth for the data buffers and don't hardcode any buffer sizes. Let them be calculated from the MTU for "classics" and then divide the truesize by 2 for "short" ones. The memory usage is now greatly reduced and 2 buffer queues starts make sense: on frames <= 1024, you'll recycle (and resync) a page only after 4 HW writes rather than two. Signed-off-by: Alexander Lobakin Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/idpf/Kconfig | 1 - drivers/net/ethernet/intel/idpf/idpf.h | 2 - .../net/ethernet/intel/idpf/idpf_singleq_txrx.c | 27 +-- drivers/net/ethernet/intel/idpf/idpf_txrx.c | 242 ++++++++------------- drivers/net/ethernet/intel/idpf/idpf_txrx.h | 86 ++------ drivers/net/ethernet/intel/idpf/idpf_virtchnl.c | 8 +- 6 files changed, 120 insertions(+), 246 deletions(-) (limited to 'drivers/net/ethernet/intel/idpf/idpf_virtchnl.c') diff --git a/drivers/net/ethernet/intel/idpf/Kconfig b/drivers/net/ethernet/intel/idpf/Kconfig index 1f071143d992..1addd663acad 100644 --- a/drivers/net/ethernet/intel/idpf/Kconfig +++ b/drivers/net/ethernet/intel/idpf/Kconfig @@ -6,7 +6,6 @@ config IDPF depends on PCI_MSI select DIMLIB select LIBETH - select PAGE_POOL help This driver supports Intel(R) Infrastructure Data Path Function devices. diff --git a/drivers/net/ethernet/intel/idpf/idpf.h b/drivers/net/ethernet/intel/idpf/idpf.h index 078340a01757..2c31ad87587a 100644 --- a/drivers/net/ethernet/intel/idpf/idpf.h +++ b/drivers/net/ethernet/intel/idpf/idpf.h @@ -264,7 +264,6 @@ struct idpf_port_stats { * the worst case. * @num_bufqs_per_qgrp: Buffer queues per RX queue in a given grouping * @bufq_desc_count: Buffer queue descriptor count - * @bufq_size: Size of buffers in ring (e.g. 2K, 4K, etc) * @num_rxq_grp: Number of RX queues in a group * @rxq_grps: Total number of RX groups. Number of groups * number of RX per * group will yield total number of RX queues. @@ -308,7 +307,6 @@ struct idpf_vport { u32 rxq_desc_count; u8 num_bufqs_per_qgrp; u32 bufq_desc_count[IDPF_MAX_BUFQS_PER_RXQ_GRP]; - u32 bufq_size[IDPF_MAX_BUFQS_PER_RXQ_GRP]; u16 num_rxq_grp; struct idpf_rxq_group *rxq_grps; u32 rxq_model; diff --git a/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c b/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c index ced1fdf7a1fb..fe64febf7436 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c +++ b/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c @@ -857,20 +857,24 @@ bool idpf_rx_singleq_buf_hw_alloc_all(struct idpf_rx_queue *rx_q, u16 cleaned_count) { struct virtchnl2_singleq_rx_buf_desc *desc; + const struct libeth_fq_fp fq = { + .pp = rx_q->pp, + .fqes = rx_q->rx_buf, + .truesize = rx_q->truesize, + .count = rx_q->desc_count, + }; u16 nta = rx_q->next_to_alloc; - struct idpf_rx_buf *buf; if (!cleaned_count) return false; desc = &rx_q->single_buf[nta]; - buf = &rx_q->rx_buf[nta]; do { dma_addr_t addr; - addr = idpf_alloc_page(rx_q->pp, buf, rx_q->rx_buf_size); - if (unlikely(addr == DMA_MAPPING_ERROR)) + addr = libeth_rx_alloc(&fq, nta); + if (addr == DMA_MAPPING_ERROR) break; /* Refresh the desc even if buffer_addrs didn't change @@ -880,11 +884,9 @@ bool idpf_rx_singleq_buf_hw_alloc_all(struct idpf_rx_queue *rx_q, desc->hdr_addr = 0; desc++; - buf++; nta++; if (unlikely(nta == rx_q->desc_count)) { desc = &rx_q->single_buf[0]; - buf = rx_q->rx_buf; nta = 0; } @@ -1004,28 +1006,26 @@ static int idpf_rx_singleq_clean(struct idpf_rx_queue *rx_q, int budget) idpf_rx_singleq_extract_fields(rx_q, rx_desc, &fields); rx_buf = &rx_q->rx_buf[ntc]; - if (!fields.size) { - idpf_rx_put_page(rx_buf); + if (!libeth_rx_sync_for_cpu(rx_buf, fields.size)) goto skip_data; - } - idpf_rx_sync_for_cpu(rx_buf, fields.size); if (skb) idpf_rx_add_frag(rx_buf, skb, fields.size); else - skb = idpf_rx_construct_skb(rx_q, rx_buf, fields.size); + skb = idpf_rx_build_skb(rx_buf, fields.size); /* exit if we failed to retrieve a buffer */ if (!skb) break; skip_data: - IDPF_SINGLEQ_BUMP_RING_IDX(rx_q, ntc); + rx_buf->page = NULL; + IDPF_SINGLEQ_BUMP_RING_IDX(rx_q, ntc); cleaned_count++; /* skip if it is non EOP desc */ - if (idpf_rx_singleq_is_non_eop(rx_desc)) + if (idpf_rx_singleq_is_non_eop(rx_desc) || unlikely(!skb)) continue; #define IDPF_RXD_ERR_S FIELD_PREP(VIRTCHNL2_RX_BASE_DESC_QW1_ERROR_M, \ @@ -1062,6 +1062,7 @@ skip_data: rx_q->next_to_clean = ntc; + page_pool_nid_changed(rx_q->pp, numa_mem_id()); if (cleaned_count) failure = idpf_rx_singleq_buf_hw_alloc_all(rx_q, cleaned_count); diff --git a/drivers/net/ethernet/intel/idpf/idpf_txrx.c b/drivers/net/ethernet/intel/idpf/idpf_txrx.c index ec4a9501d19f..af2879f03b8d 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_txrx.c +++ b/drivers/net/ethernet/intel/idpf/idpf_txrx.c @@ -417,6 +417,11 @@ static void idpf_rx_hdr_buf_rel_all(struct idpf_buf_queue *bufq) */ static void idpf_rx_buf_rel_bufq(struct idpf_buf_queue *bufq) { + struct libeth_fq fq = { + .fqes = bufq->buf, + .pp = bufq->pp, + }; + /* queue already cleared, nothing to do */ if (!bufq->buf) return; @@ -428,11 +433,9 @@ static void idpf_rx_buf_rel_bufq(struct idpf_buf_queue *bufq) if (idpf_queue_has(HSPLIT_EN, bufq)) idpf_rx_hdr_buf_rel_all(bufq); - page_pool_destroy(bufq->pp); - bufq->pp = NULL; - - kfree(bufq->buf); + libeth_rx_fq_destroy(&fq); bufq->buf = NULL; + bufq->pp = NULL; } /** @@ -441,17 +444,20 @@ static void idpf_rx_buf_rel_bufq(struct idpf_buf_queue *bufq) */ static void idpf_rx_buf_rel_all(struct idpf_rx_queue *rxq) { + struct libeth_fq fq = { + .fqes = rxq->rx_buf, + .pp = rxq->pp, + }; + if (!rxq->rx_buf) return; for (u32 i = 0; i < rxq->desc_count; i++) idpf_rx_page_rel(&rxq->rx_buf[i]); - page_pool_destroy(rxq->pp); - rxq->pp = NULL; - - kfree(rxq->rx_buf); + libeth_rx_fq_destroy(&fq); rxq->rx_buf = NULL; + rxq->pp = NULL; } /** @@ -633,11 +639,9 @@ static bool idpf_rx_post_buf_desc(struct idpf_buf_queue *bufq, u16 buf_id) .count = bufq->desc_count, }; u16 nta = bufq->next_to_alloc; - struct idpf_rx_buf *buf; dma_addr_t addr; splitq_rx_desc = &bufq->split_buf[nta]; - buf = &bufq->buf[buf_id]; if (idpf_queue_has(HSPLIT_EN, bufq)) { fq.pp = bufq->hdr_pp; @@ -651,8 +655,12 @@ static bool idpf_rx_post_buf_desc(struct idpf_buf_queue *bufq, u16 buf_id) splitq_rx_desc->hdr_addr = cpu_to_le64(addr); } - addr = idpf_alloc_page(bufq->pp, buf, bufq->rx_buf_size); - if (unlikely(addr == DMA_MAPPING_ERROR)) + fq.pp = bufq->pp; + fq.fqes = bufq->buf; + fq.truesize = bufq->truesize; + + addr = libeth_rx_alloc(&fq, buf_id); + if (addr == DMA_MAPPING_ERROR) return false; splitq_rx_desc->pkt_addr = cpu_to_le64(addr); @@ -689,30 +697,6 @@ static bool idpf_rx_post_init_bufs(struct idpf_buf_queue *bufq, return true; } -/** - * idpf_rx_create_page_pool - Create a page pool - * @napi: NAPI of the associated queue vector - * @count: queue descriptor count - * - * Returns &page_pool on success, casted -errno on failure - */ -static struct page_pool *idpf_rx_create_page_pool(struct napi_struct *napi, - u32 count) -{ - struct page_pool_params pp = { - .flags = PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV, - .order = 0, - .pool_size = count, - .nid = NUMA_NO_NODE, - .dev = napi->dev->dev.parent, - .max_len = PAGE_SIZE, - .dma_dir = DMA_FROM_DEVICE, - .offset = 0, - }; - - return page_pool_create(&pp); -} - /** * idpf_rx_buf_alloc_singleq - Allocate memory for all buffer resources * @rxq: queue for which the buffers are allocated @@ -721,11 +705,6 @@ static struct page_pool *idpf_rx_create_page_pool(struct napi_struct *napi, */ static int idpf_rx_buf_alloc_singleq(struct idpf_rx_queue *rxq) { - rxq->rx_buf = kcalloc(rxq->desc_count, sizeof(*rxq->rx_buf), - GFP_KERNEL); - if (!rxq->rx_buf) - return -ENOMEM; - if (idpf_rx_singleq_buf_hw_alloc_all(rxq, rxq->desc_count - 1)) goto err; @@ -745,13 +724,21 @@ err: */ static int idpf_rx_bufs_init_singleq(struct idpf_rx_queue *rxq) { - struct page_pool *pool; + struct libeth_fq fq = { + .count = rxq->desc_count, + .type = LIBETH_FQE_MTU, + .nid = idpf_q_vector_to_mem(rxq->q_vector), + }; + int ret; - pool = idpf_rx_create_page_pool(&rxq->q_vector->napi, rxq->desc_count); - if (IS_ERR(pool)) - return PTR_ERR(pool); + ret = libeth_rx_fq_create(&fq, &rxq->q_vector->napi); + if (ret) + return ret; - rxq->pp = pool; + rxq->pp = fq.pp; + rxq->rx_buf = fq.fqes; + rxq->truesize = fq.truesize; + rxq->rx_buf_size = fq.buf_len; return idpf_rx_buf_alloc_singleq(rxq); } @@ -766,14 +753,6 @@ static int idpf_rx_buf_alloc_all(struct idpf_buf_queue *rxbufq) { int err = 0; - /* Allocate book keeping buffers */ - rxbufq->buf = kcalloc(rxbufq->desc_count, sizeof(*rxbufq->buf), - GFP_KERNEL); - if (!rxbufq->buf) { - err = -ENOMEM; - goto rx_buf_alloc_all_out; - } - if (idpf_queue_has(HSPLIT_EN, rxbufq)) { err = idpf_rx_hdr_buf_alloc_all(rxbufq); if (err) @@ -794,19 +773,30 @@ rx_buf_alloc_all_out: /** * idpf_rx_bufs_init - Initialize page pool, allocate rx bufs, and post to HW * @bufq: buffer queue to create page pool for + * @type: type of Rx buffers to allocate * * Returns 0 on success, negative on failure */ -static int idpf_rx_bufs_init(struct idpf_buf_queue *bufq) +static int idpf_rx_bufs_init(struct idpf_buf_queue *bufq, + enum libeth_fqe_type type) { - struct page_pool *pool; + struct libeth_fq fq = { + .truesize = bufq->truesize, + .count = bufq->desc_count, + .type = type, + .hsplit = idpf_queue_has(HSPLIT_EN, bufq), + .nid = idpf_q_vector_to_mem(bufq->q_vector), + }; + int ret; - pool = idpf_rx_create_page_pool(&bufq->q_vector->napi, - bufq->desc_count); - if (IS_ERR(pool)) - return PTR_ERR(pool); + ret = libeth_rx_fq_create(&fq, &bufq->q_vector->napi); + if (ret) + return ret; - bufq->pp = pool; + bufq->pp = fq.pp; + bufq->buf = fq.fqes; + bufq->truesize = fq.truesize; + bufq->rx_buf_size = fq.buf_len; return idpf_rx_buf_alloc_all(bufq); } @@ -819,14 +809,15 @@ static int idpf_rx_bufs_init(struct idpf_buf_queue *bufq) */ int idpf_rx_bufs_init_all(struct idpf_vport *vport) { - struct idpf_rxq_group *rx_qgrp; + bool split = idpf_is_queue_model_split(vport->rxq_model); int i, j, err; for (i = 0; i < vport->num_rxq_grp; i++) { - rx_qgrp = &vport->rxq_grps[i]; + struct idpf_rxq_group *rx_qgrp = &vport->rxq_grps[i]; + u32 truesize = 0; /* Allocate bufs for the rxq itself in singleq */ - if (!idpf_is_queue_model_split(vport->rxq_model)) { + if (!split) { int num_rxq = rx_qgrp->singleq.num_rxq; for (j = 0; j < num_rxq; j++) { @@ -843,12 +834,19 @@ int idpf_rx_bufs_init_all(struct idpf_vport *vport) /* Otherwise, allocate bufs for the buffer queues */ for (j = 0; j < vport->num_bufqs_per_qgrp; j++) { + enum libeth_fqe_type type; struct idpf_buf_queue *q; q = &rx_qgrp->splitq.bufq_sets[j].bufq; - err = idpf_rx_bufs_init(q); + q->truesize = truesize; + + type = truesize ? LIBETH_FQE_SHORT : LIBETH_FQE_MTU; + + err = idpf_rx_bufs_init(q, type); if (err) return err; + + truesize = q->truesize >> 1; } } @@ -1160,17 +1158,11 @@ void idpf_vport_init_num_qs(struct idpf_vport *vport, /* Adjust number of buffer queues per Rx queue group. */ if (!idpf_is_queue_model_split(vport->rxq_model)) { vport->num_bufqs_per_qgrp = 0; - vport->bufq_size[0] = IDPF_RX_BUF_2048; return; } vport->num_bufqs_per_qgrp = IDPF_MAX_BUFQS_PER_RXQ_GRP; - /* Bufq[0] default buffer size is 4K - * Bufq[1] default buffer size is 2K - */ - vport->bufq_size[0] = IDPF_RX_BUF_4096; - vport->bufq_size[1] = IDPF_RX_BUF_2048; } /** @@ -1507,7 +1499,6 @@ static int idpf_rxq_group_alloc(struct idpf_vport *vport, u16 num_rxq) q = &rx_qgrp->splitq.bufq_sets[j].bufq; q->desc_count = vport->bufq_desc_count[j]; - q->rx_buf_size = vport->bufq_size[j]; q->rx_buffer_low_watermark = IDPF_LOW_WATERMARK; idpf_queue_assign(HSPLIT_EN, q, hs); @@ -1560,14 +1551,9 @@ setup_rxq: q->netdev = vport->netdev; q->bufq_sets = rx_qgrp->splitq.bufq_sets; q->idx = (i * num_rxq) + j; - /* In splitq mode, RXQ buffer size should be - * set to that of the first buffer queue - * associated with this RXQ - */ - q->rx_buf_size = vport->bufq_size[0]; q->rx_buffer_low_watermark = IDPF_LOW_WATERMARK; q->rx_max_pkt_size = vport->netdev->mtu + - IDPF_PACKET_HDR_PAD; + LIBETH_RX_LL_LEN; idpf_rxq_set_descids(vport, q); } } @@ -3145,69 +3131,10 @@ idpf_rx_process_skb_fields(struct idpf_rx_queue *rxq, struct sk_buff *skb, void idpf_rx_add_frag(struct idpf_rx_buf *rx_buf, struct sk_buff *skb, unsigned int size) { - skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buf->page, - rx_buf->offset, size, rx_buf->truesize); - - rx_buf->page = NULL; -} - -/** - * idpf_rx_construct_skb - Allocate skb and populate it - * @rxq: Rx descriptor queue - * @rx_buf: Rx buffer to pull data from - * @size: the length of the packet - * - * This function allocates an skb. It then populates it with the page - * data from the current receive descriptor, taking care to set up the - * skb correctly. - */ -struct sk_buff *idpf_rx_construct_skb(const struct idpf_rx_queue *rxq, - struct idpf_rx_buf *rx_buf, - unsigned int size) -{ - unsigned int headlen; - struct sk_buff *skb; - void *va; - - va = page_address(rx_buf->page) + rx_buf->offset; - - /* prefetch first cache line of first page */ - net_prefetch(va); - /* allocate a skb to store the frags */ - skb = napi_alloc_skb(rxq->napi, IDPF_RX_HDR_SIZE); - if (unlikely(!skb)) { - idpf_rx_put_page(rx_buf); - - return NULL; - } + u32 hr = rx_buf->page->pp->p.offset; - skb_mark_for_recycle(skb); - - /* Determine available headroom for copy */ - headlen = size; - if (headlen > IDPF_RX_HDR_SIZE) - headlen = eth_get_headlen(skb->dev, va, IDPF_RX_HDR_SIZE); - - /* align pull length to size of long to optimize memcpy performance */ - memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long))); - - /* if we exhaust the linear part then add what is left as a frag */ - size -= headlen; - if (!size) { - idpf_rx_put_page(rx_buf); - - return skb; - } - - skb_add_rx_frag(skb, 0, rx_buf->page, rx_buf->offset + headlen, - size, rx_buf->truesize); - - /* Since we're giving the page to the stack, clear our reference to it. - * We'll get a new one during buffer posting. - */ - rx_buf->page = NULL; - - return skb; + skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buf->page, + rx_buf->offset + hr, size, rx_buf->truesize); } /** @@ -3413,26 +3340,26 @@ static int idpf_rx_splitq_clean(struct idpf_rx_queue *rxq, int budget) hdr->page = NULL; payload: - if (pkt_len) { - idpf_rx_sync_for_cpu(rx_buf, pkt_len); - if (skb) - idpf_rx_add_frag(rx_buf, skb, pkt_len); - else - skb = idpf_rx_construct_skb(rxq, rx_buf, - pkt_len); - } else { - idpf_rx_put_page(rx_buf); - } + if (!libeth_rx_sync_for_cpu(rx_buf, pkt_len)) + goto skip_data; + + if (skb) + idpf_rx_add_frag(rx_buf, skb, pkt_len); + else + skb = idpf_rx_build_skb(rx_buf, pkt_len); /* exit if we failed to retrieve a buffer */ if (!skb) break; - idpf_rx_post_buf_refill(refillq, buf_id); +skip_data: + rx_buf->page = NULL; + idpf_rx_post_buf_refill(refillq, buf_id); IDPF_RX_BUMP_NTC(rxq, ntc); + /* skip if it is non EOP desc */ - if (!idpf_rx_splitq_is_eop(rx_desc)) + if (!idpf_rx_splitq_is_eop(rx_desc) || unlikely(!skb)) continue; /* pad skb if needed (to make valid ethernet frame) */ @@ -3483,15 +3410,15 @@ static int idpf_rx_update_bufq_desc(struct idpf_buf_queue *bufq, u32 buf_id, struct virtchnl2_splitq_rx_buf_desc *buf_desc) { struct libeth_fq_fp fq = { + .pp = bufq->pp, + .fqes = bufq->buf, + .truesize = bufq->truesize, .count = bufq->desc_count, }; - struct idpf_rx_buf *buf; dma_addr_t addr; - buf = &bufq->buf[buf_id]; - - addr = idpf_alloc_page(bufq->pp, buf, bufq->rx_buf_size); - if (unlikely(addr == DMA_MAPPING_ERROR)) + addr = libeth_rx_alloc(&fq, buf_id); + if (addr == DMA_MAPPING_ERROR) return -ENOMEM; buf_desc->pkt_addr = cpu_to_le64(addr); @@ -3590,6 +3517,7 @@ static void idpf_rx_clean_refillq_all(struct idpf_buf_queue *bufq, int nid) struct idpf_bufq_set *bufq_set; int i; + page_pool_nid_changed(bufq->pp, nid); if (bufq->hdr_pp) page_pool_nid_changed(bufq->hdr_pp, nid); diff --git a/drivers/net/ethernet/intel/idpf/idpf_txrx.h b/drivers/net/ethernet/intel/idpf/idpf_txrx.h index 47d70ad0d6b4..6215dbee5546 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_txrx.h +++ b/drivers/net/ethernet/intel/idpf/idpf_txrx.h @@ -7,7 +7,6 @@ #include #include -#include #include #include @@ -97,14 +96,10 @@ do { \ idx = 0; \ } while (0) -#define IDPF_RX_HDR_SIZE 256 -#define IDPF_RX_BUF_2048 2048 -#define IDPF_RX_BUF_4096 4096 #define IDPF_RX_BUF_STRIDE 32 #define IDPF_RX_BUF_POST_STRIDE 16 #define IDPF_LOW_WATERMARK 64 -#define IDPF_PACKET_HDR_PAD \ - (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN * 2) + #define IDPF_TX_TSO_MIN_MSS 88 /* Minimum number of descriptors between 2 descriptors with the RE bit set; @@ -538,7 +533,7 @@ struct idpf_txq_stash { * @desc_ring: virtual descriptor ring address * @bufq_sets: Pointer to the array of buffer queues in splitq mode * @napi: NAPI instance corresponding to this queue (splitq) - * @rx_buf: See struct idpf_rx_buf + * @rx_buf: See struct &libeth_fqe * @pp: Page pool pointer in singleq mode * @netdev: &net_device corresponding to this queue * @tail: Tail offset. Used for both queue models single and split. @@ -552,6 +547,7 @@ struct idpf_txq_stash { * @next_to_clean: Next descriptor to clean * @next_to_alloc: RX buffer to allocate at * @skb: Pointer to the skb + * @truesize: data buffer truesize in singleq * @stats_sync: See struct u64_stats_sync * @q_stats: See union idpf_rx_queue_stats * @q_id: Queue id @@ -577,7 +573,7 @@ struct idpf_rx_queue { struct napi_struct *napi; }; struct { - struct idpf_rx_buf *rx_buf; + struct libeth_fqe *rx_buf; struct page_pool *pp; }; }; @@ -598,6 +594,7 @@ struct idpf_rx_queue { u16 next_to_alloc; struct sk_buff *skb; + u32 truesize; struct u64_stats_sync stats_sync; struct idpf_rx_queue_stats q_stats; @@ -617,7 +614,7 @@ struct idpf_rx_queue { __cacheline_group_end_aligned(cold); }; libeth_cacheline_set_assert(struct idpf_rx_queue, 64, - 72 + sizeof(struct u64_stats_sync), + 80 + sizeof(struct u64_stats_sync), 32); /** @@ -744,15 +741,16 @@ libeth_cacheline_set_assert(struct idpf_tx_queue, 64, * @split_buf: buffer descriptor array * @hdr_buf: &libeth_fqe for header buffers * @hdr_pp: &page_pool for header buffers - * @buf: &idpf_rx_buf for data buffers + * @buf: &libeth_fqe for data buffers * @pp: &page_pool for data buffers * @tail: Tail offset * @flags: See enum idpf_queue_flags_t * @desc_count: Number of descriptors - * @hdr_truesize: truesize for buffer headers * @next_to_use: Next descriptor to use * @next_to_clean: Next descriptor to clean * @next_to_alloc: RX buffer to allocate at + * @hdr_truesize: truesize for buffer headers + * @truesize: truesize for data buffers * @q_id: Queue id * @size: Length of descriptor ring in bytes * @dma: Physical address of ring @@ -766,20 +764,21 @@ struct idpf_buf_queue { struct virtchnl2_splitq_rx_buf_desc *split_buf; struct libeth_fqe *hdr_buf; struct page_pool *hdr_pp; - struct idpf_rx_buf *buf; + struct libeth_fqe *buf; struct page_pool *pp; void __iomem *tail; DECLARE_BITMAP(flags, __IDPF_Q_FLAGS_NBITS); u32 desc_count; - - u32 hdr_truesize; __cacheline_group_end_aligned(read_mostly); __cacheline_group_begin_aligned(read_write); u32 next_to_use; u32 next_to_clean; u32 next_to_alloc; + + u32 hdr_truesize; + u32 truesize; __cacheline_group_end_aligned(read_write); __cacheline_group_begin_aligned(cold); @@ -794,7 +793,7 @@ struct idpf_buf_queue { u16 rx_buf_size; __cacheline_group_end_aligned(cold); }; -libeth_cacheline_set_assert(struct idpf_buf_queue, 64, 16, 32); +libeth_cacheline_set_assert(struct idpf_buf_queue, 64, 24, 32); /** * struct idpf_compl_queue - software structure representing a completion queue @@ -1034,60 +1033,6 @@ static inline void idpf_tx_splitq_build_desc(union idpf_tx_flex_desc *desc, idpf_tx_splitq_build_flow_desc(desc, params, td_cmd, size); } -/** - * idpf_alloc_page - Allocate a new RX buffer from the page pool - * @pool: page_pool to allocate from - * @buf: metadata struct to populate with page info - * @buf_size: 2K or 4K - * - * Returns &dma_addr_t to be passed to HW for Rx, %DMA_MAPPING_ERROR otherwise. - */ -static inline dma_addr_t idpf_alloc_page(struct page_pool *pool, - struct idpf_rx_buf *buf, - unsigned int buf_size) -{ - if (buf_size == IDPF_RX_BUF_2048) - buf->page = page_pool_dev_alloc_frag(pool, &buf->offset, - buf_size); - else - buf->page = page_pool_dev_alloc_pages(pool); - - if (!buf->page) - return DMA_MAPPING_ERROR; - - buf->truesize = buf_size; - - return page_pool_get_dma_addr(buf->page) + buf->offset + - pool->p.offset; -} - -/** - * idpf_rx_put_page - Return RX buffer page to pool - * @rx_buf: RX buffer metadata struct - */ -static inline void idpf_rx_put_page(struct idpf_rx_buf *rx_buf) -{ - page_pool_put_page(rx_buf->page->pp, rx_buf->page, - rx_buf->truesize, true); - rx_buf->page = NULL; -} - -/** - * idpf_rx_sync_for_cpu - Synchronize DMA buffer - * @rx_buf: RX buffer metadata struct - * @len: frame length from descriptor - */ -static inline void idpf_rx_sync_for_cpu(struct idpf_rx_buf *rx_buf, u32 len) -{ - struct page *page = rx_buf->page; - struct page_pool *pp = page->pp; - - dma_sync_single_range_for_cpu(pp->p.dev, - page_pool_get_dma_addr(page), - rx_buf->offset + pp->p.offset, len, - page_pool_get_dma_dir(pp)); -} - int idpf_vport_singleq_napi_poll(struct napi_struct *napi, int budget); void idpf_vport_init_num_qs(struct idpf_vport *vport, struct virtchnl2_create_vport *vport_msg); @@ -1110,9 +1055,6 @@ void idpf_deinit_rss(struct idpf_vport *vport); int idpf_rx_bufs_init_all(struct idpf_vport *vport); void idpf_rx_add_frag(struct idpf_rx_buf *rx_buf, struct sk_buff *skb, unsigned int size); -struct sk_buff *idpf_rx_construct_skb(const struct idpf_rx_queue *rxq, - struct idpf_rx_buf *rx_buf, - unsigned int size); struct sk_buff *idpf_rx_build_skb(const struct libeth_fqe *buf, u32 size); void idpf_tx_buf_hw_update(struct idpf_tx_queue *tx_q, u32 val, bool xmit_more); diff --git a/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c b/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c index 50dcb3ab02b1..70986e12da28 100644 --- a/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c +++ b/drivers/net/ethernet/intel/idpf/idpf_virtchnl.c @@ -1615,6 +1615,12 @@ setup_rxqs: rxq = &rx_qgrp->splitq.rxq_sets[j]->rxq; sets = rxq->bufq_sets; + /* In splitq mode, RXQ buffer size should be + * set to that of the first buffer queue + * associated with this RXQ. + */ + rxq->rx_buf_size = sets[0].bufq.rx_buf_size; + qi[k].rx_bufq1_id = cpu_to_le16(sets[0].bufq.q_id); if (vport->num_bufqs_per_qgrp > IDPF_SINGLE_BUFQ_PER_RXQ_GRP) { qi[k].bufq2_ena = IDPF_BUFQ2_ENA; @@ -3167,7 +3173,7 @@ void idpf_vport_init(struct idpf_vport *vport, struct idpf_vport_max_q *max_q) rss_data->rss_lut_size = le16_to_cpu(vport_msg->rss_lut_size); ether_addr_copy(vport->default_mac_addr, vport_msg->default_mac_addr); - vport->max_mtu = le16_to_cpu(vport_msg->max_mtu) - IDPF_PACKET_HDR_PAD; + vport->max_mtu = le16_to_cpu(vport_msg->max_mtu) - LIBETH_RX_LL_LEN; /* Initialize Tx and Rx profiles for Dynamic Interrupt Moderation */ memcpy(vport->rx_itr_profile, rx_itr, IDPF_DIM_PROFILE_SLOTS); -- cgit v1.2.3-58-ga151