diff options
author | Shinas Rasheed <srasheed@marvell.com> | 2024-02-08 02:18:38 -0800 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2024-02-12 09:01:52 +0000 |
commit | 1cd3b407977c3ab1d2ddc26cb7113e7fb1509cd1 (patch) | |
tree | e94ac3d9d5db4c0a570badb9b63aca3c36b671cb /drivers/net/ethernet/marvell/octeon_ep_vf/octep_vf_main.c | |
parent | c3fad23cdc069aeff8f1b22a27e8a471c1d9d517 (diff) |
octeon_ep_vf: add Tx/Rx processing and interrupt support
Add support to enable MSI-x and register interrupts.
Add support to process Tx and Rx traffic. Includes processing
Tx completions and Rx refill.
Signed-off-by: Shinas Rasheed <srasheed@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'drivers/net/ethernet/marvell/octeon_ep_vf/octep_vf_main.c')
-rw-r--r-- | drivers/net/ethernet/marvell/octeon_ep_vf/octep_vf_main.c | 550 |
1 files changed, 550 insertions, 0 deletions
diff --git a/drivers/net/ethernet/marvell/octeon_ep_vf/octep_vf_main.c b/drivers/net/ethernet/marvell/octeon_ep_vf/octep_vf_main.c index ca823933b9fd..e5d258980547 100644 --- a/drivers/net/ethernet/marvell/octeon_ep_vf/octep_vf_main.c +++ b/drivers/net/ethernet/marvell/octeon_ep_vf/octep_vf_main.c @@ -13,6 +13,7 @@ #include <linux/etherdevice.h> #include <linux/rtnetlink.h> #include <linux/vmalloc.h> +#include <net/netdev_queues.h> #include "octep_vf_config.h" #include "octep_vf_main.h" @@ -36,6 +37,372 @@ MODULE_AUTHOR("Veerasenareddy Burru <vburru@marvell.com>"); MODULE_DESCRIPTION(OCTEP_VF_DRV_STRING); MODULE_LICENSE("GPL"); +/** + * octep_vf_alloc_ioq_vectors() - Allocate Tx/Rx Queue interrupt info. + * + * @oct: Octeon device private data structure. + * + * Allocate resources to hold per Tx/Rx queue interrupt info. + * This is the information passed to interrupt handler, from which napi poll + * is scheduled and includes quick access to private data of Tx/Rx queue + * corresponding to the interrupt being handled. + * + * Return: 0, on successful allocation of resources for all queue interrupts. + * -1, if failed to allocate any resource. + */ +static int octep_vf_alloc_ioq_vectors(struct octep_vf_device *oct) +{ + struct octep_vf_ioq_vector *ioq_vector; + int i; + + for (i = 0; i < oct->num_oqs; i++) { + oct->ioq_vector[i] = vzalloc(sizeof(*oct->ioq_vector[i])); + if (!oct->ioq_vector[i]) + goto free_ioq_vector; + + ioq_vector = oct->ioq_vector[i]; + ioq_vector->iq = oct->iq[i]; + ioq_vector->oq = oct->oq[i]; + ioq_vector->octep_vf_dev = oct; + } + + dev_info(&oct->pdev->dev, "Allocated %d IOQ vectors\n", oct->num_oqs); + return 0; + +free_ioq_vector: + while (i) { + i--; + vfree(oct->ioq_vector[i]); + oct->ioq_vector[i] = NULL; + } + return -1; +} + +/** + * octep_vf_free_ioq_vectors() - Free Tx/Rx Queue interrupt vector info. + * + * @oct: Octeon device private data structure. + */ +static void octep_vf_free_ioq_vectors(struct octep_vf_device *oct) +{ + int i; + + for (i = 0; i < oct->num_oqs; i++) { + if (oct->ioq_vector[i]) { + vfree(oct->ioq_vector[i]); + oct->ioq_vector[i] = NULL; + } + } + netdev_info(oct->netdev, "Freed IOQ Vectors\n"); +} + +/** + * octep_vf_enable_msix_range() - enable MSI-x interrupts. + * + * @oct: Octeon device private data structure. + * + * Allocate and enable all MSI-x interrupts (queue and non-queue interrupts) + * for the Octeon device. + * + * Return: 0, on successfully enabling all MSI-x interrupts. + * -1, if failed to enable any MSI-x interrupt. + */ +static int octep_vf_enable_msix_range(struct octep_vf_device *oct) +{ + int num_msix, msix_allocated; + int i; + + /* Generic interrupts apart from input/output queues */ + //num_msix = oct->num_oqs + CFG_GET_NON_IOQ_MSIX(oct->conf); + num_msix = oct->num_oqs; + oct->msix_entries = kcalloc(num_msix, sizeof(struct msix_entry), GFP_KERNEL); + if (!oct->msix_entries) + goto msix_alloc_err; + + for (i = 0; i < num_msix; i++) + oct->msix_entries[i].entry = i; + + msix_allocated = pci_enable_msix_range(oct->pdev, oct->msix_entries, + num_msix, num_msix); + if (msix_allocated != num_msix) { + dev_err(&oct->pdev->dev, + "Failed to enable %d msix irqs; got only %d\n", + num_msix, msix_allocated); + goto enable_msix_err; + } + oct->num_irqs = msix_allocated; + dev_info(&oct->pdev->dev, "MSI-X enabled successfully\n"); + + return 0; + +enable_msix_err: + if (msix_allocated > 0) + pci_disable_msix(oct->pdev); + kfree(oct->msix_entries); + oct->msix_entries = NULL; +msix_alloc_err: + return -1; +} + +/** + * octep_vf_disable_msix() - disable MSI-x interrupts. + * + * @oct: Octeon device private data structure. + * + * Disable MSI-x on the Octeon device. + */ +static void octep_vf_disable_msix(struct octep_vf_device *oct) +{ + pci_disable_msix(oct->pdev); + kfree(oct->msix_entries); + oct->msix_entries = NULL; + dev_info(&oct->pdev->dev, "Disabled MSI-X\n"); +} + +/** + * octep_vf_ioq_intr_handler() - handler for all Tx/Rx queue interrupts. + * + * @irq: Interrupt number. + * @data: interrupt data contains pointers to Tx/Rx queue private data + * and correspong NAPI context. + * + * this is common handler for all non-queue (generic) interrupts. + */ +static irqreturn_t octep_vf_ioq_intr_handler(int irq, void *data) +{ + struct octep_vf_ioq_vector *ioq_vector = data; + struct octep_vf_device *oct = ioq_vector->octep_vf_dev; + + return oct->hw_ops.ioq_intr_handler(ioq_vector); +} + +/** + * octep_vf_request_irqs() - Register interrupt handlers. + * + * @oct: Octeon device private data structure. + * + * Register handlers for all queue and non-queue interrupts. + * + * Return: 0, on successful registration of all interrupt handlers. + * -1, on any error. + */ +static int octep_vf_request_irqs(struct octep_vf_device *oct) +{ + struct net_device *netdev = oct->netdev; + struct octep_vf_ioq_vector *ioq_vector; + struct msix_entry *msix_entry; + int ret, i; + + /* Request IRQs for Tx/Rx queues */ + for (i = 0; i < oct->num_oqs; i++) { + ioq_vector = oct->ioq_vector[i]; + msix_entry = &oct->msix_entries[i]; + + snprintf(ioq_vector->name, sizeof(ioq_vector->name), + "%s-q%d", netdev->name, i); + ret = request_irq(msix_entry->vector, + octep_vf_ioq_intr_handler, 0, + ioq_vector->name, ioq_vector); + if (ret) { + netdev_err(netdev, + "request_irq failed for Q-%d; err=%d", + i, ret); + goto ioq_irq_err; + } + + cpumask_set_cpu(i % num_online_cpus(), + &ioq_vector->affinity_mask); + irq_set_affinity_hint(msix_entry->vector, + &ioq_vector->affinity_mask); + } + + return 0; +ioq_irq_err: + while (i) { + --i; + free_irq(oct->msix_entries[i].vector, oct); + } + return -1; +} + +/** + * octep_vf_free_irqs() - free all registered interrupts. + * + * @oct: Octeon device private data structure. + * + * Free all queue and non-queue interrupts of the Octeon device. + */ +static void octep_vf_free_irqs(struct octep_vf_device *oct) +{ + int i; + + for (i = 0; i < oct->num_irqs; i++) { + irq_set_affinity_hint(oct->msix_entries[i].vector, NULL); + free_irq(oct->msix_entries[i].vector, oct->ioq_vector[i]); + } + netdev_info(oct->netdev, "IRQs freed\n"); +} + +/** + * octep_vf_setup_irqs() - setup interrupts for the Octeon device. + * + * @oct: Octeon device private data structure. + * + * Allocate data structures to hold per interrupt information, allocate/enable + * MSI-x interrupt and register interrupt handlers. + * + * Return: 0, on successful allocation and registration of all interrupts. + * -1, on any error. + */ +static int octep_vf_setup_irqs(struct octep_vf_device *oct) +{ + if (octep_vf_alloc_ioq_vectors(oct)) + goto ioq_vector_err; + + if (octep_vf_enable_msix_range(oct)) + goto enable_msix_err; + + if (octep_vf_request_irqs(oct)) + goto request_irq_err; + + return 0; + +request_irq_err: + octep_vf_disable_msix(oct); +enable_msix_err: + octep_vf_free_ioq_vectors(oct); +ioq_vector_err: + return -1; +} + +/** + * octep_vf_clean_irqs() - free all interrupts and its resources. + * + * @oct: Octeon device private data structure. + */ +static void octep_vf_clean_irqs(struct octep_vf_device *oct) +{ + octep_vf_free_irqs(oct); + octep_vf_disable_msix(oct); + octep_vf_free_ioq_vectors(oct); +} + +/** + * octep_vf_enable_ioq_irq() - Enable MSI-x interrupt of a Tx/Rx queue. + * + * @iq: Octeon Tx queue data structure. + * @oq: Octeon Rx queue data structure. + */ +static void octep_vf_enable_ioq_irq(struct octep_vf_iq *iq, struct octep_vf_oq *oq) +{ + u32 pkts_pend = oq->pkts_pending; + + netdev_dbg(iq->netdev, "enabling intr for Q-%u\n", iq->q_no); + if (iq->pkts_processed) { + writel(iq->pkts_processed, iq->inst_cnt_reg); + iq->pkt_in_done -= iq->pkts_processed; + iq->pkts_processed = 0; + } + if (oq->last_pkt_count - pkts_pend) { + writel(oq->last_pkt_count - pkts_pend, oq->pkts_sent_reg); + oq->last_pkt_count = pkts_pend; + } + + /* Flush the previous wrties before writing to RESEND bit */ + smp_wmb(); + writeq(1UL << OCTEP_VF_OQ_INTR_RESEND_BIT, oq->pkts_sent_reg); + writeq(1UL << OCTEP_VF_IQ_INTR_RESEND_BIT, iq->inst_cnt_reg); +} + +/** + * octep_vf_napi_poll() - NAPI poll function for Tx/Rx. + * + * @napi: pointer to napi context. + * @budget: max number of packets to be processed in single invocation. + */ +static int octep_vf_napi_poll(struct napi_struct *napi, int budget) +{ + struct octep_vf_ioq_vector *ioq_vector = + container_of(napi, struct octep_vf_ioq_vector, napi); + u32 tx_pending, rx_done; + + tx_pending = octep_vf_iq_process_completions(ioq_vector->iq, 64); + rx_done = octep_vf_oq_process_rx(ioq_vector->oq, budget); + + /* need more polling if tx completion processing is still pending or + * processed at least 'budget' number of rx packets. + */ + if (tx_pending || rx_done >= budget) + return budget; + + if (likely(napi_complete_done(napi, rx_done))) + octep_vf_enable_ioq_irq(ioq_vector->iq, ioq_vector->oq); + + return rx_done; +} + +/** + * octep_vf_napi_add() - Add NAPI poll for all Tx/Rx queues. + * + * @oct: Octeon device private data structure. + */ +static void octep_vf_napi_add(struct octep_vf_device *oct) +{ + int i; + + for (i = 0; i < oct->num_oqs; i++) { + netdev_dbg(oct->netdev, "Adding NAPI on Q-%d\n", i); + netif_napi_add(oct->netdev, &oct->ioq_vector[i]->napi, octep_vf_napi_poll); + oct->oq[i]->napi = &oct->ioq_vector[i]->napi; + } +} + +/** + * octep_vf_napi_delete() - delete NAPI poll callback for all Tx/Rx queues. + * + * @oct: Octeon device private data structure. + */ +static void octep_vf_napi_delete(struct octep_vf_device *oct) +{ + int i; + + for (i = 0; i < oct->num_oqs; i++) { + netdev_dbg(oct->netdev, "Deleting NAPI on Q-%d\n", i); + netif_napi_del(&oct->ioq_vector[i]->napi); + oct->oq[i]->napi = NULL; + } +} + +/** + * octep_vf_napi_enable() - enable NAPI for all Tx/Rx queues. + * + * @oct: Octeon device private data structure. + */ +static void octep_vf_napi_enable(struct octep_vf_device *oct) +{ + int i; + + for (i = 0; i < oct->num_oqs; i++) { + netdev_dbg(oct->netdev, "Enabling NAPI on Q-%d\n", i); + napi_enable(&oct->ioq_vector[i]->napi); + } +} + +/** + * octep_vf_napi_disable() - disable NAPI for all Tx/Rx queues. + * + * @oct: Octeon device private data structure. + */ +static void octep_vf_napi_disable(struct octep_vf_device *oct) +{ + int i; + + for (i = 0; i < oct->num_oqs; i++) { + netdev_dbg(oct->netdev, "Disabling NAPI on Q-%d\n", i); + napi_disable(&oct->ioq_vector[i]->napi); + } +} + static void octep_vf_link_up(struct net_device *netdev) { netif_carrier_on(netdev); @@ -98,6 +465,8 @@ static int octep_vf_open(struct net_device *netdev) goto setup_iq_err; if (octep_vf_setup_oqs(oct)) goto setup_oq_err; + if (octep_vf_setup_irqs(oct)) + goto setup_irq_err; err = netif_set_real_num_tx_queues(netdev, oct->num_oqs); if (err) @@ -106,6 +475,9 @@ static int octep_vf_open(struct net_device *netdev) if (err) goto set_queues_err; + octep_vf_napi_add(oct); + octep_vf_napi_enable(oct); + oct->link_info.admin_up = 1; octep_vf_set_rx_state(oct, true); @@ -128,6 +500,10 @@ static int octep_vf_open(struct net_device *netdev) return 0; set_queues_err: + octep_vf_napi_disable(oct); + octep_vf_napi_delete(oct); + octep_vf_clean_irqs(oct); +setup_irq_err: octep_vf_free_oqs(oct); setup_oq_err: octep_vf_free_iqs(oct); @@ -160,7 +536,10 @@ static int octep_vf_stop(struct net_device *netdev) oct->link_info.oper_up = 0; oct->hw_ops.disable_interrupts(oct); + octep_vf_napi_disable(oct); + octep_vf_napi_delete(oct); + octep_vf_clean_irqs(oct); octep_vf_clean_iqs(oct); oct->hw_ops.disable_io_queues(oct); @@ -172,6 +551,39 @@ static int octep_vf_stop(struct net_device *netdev) } /** + * octep_vf_iq_full_check() - check if a Tx queue is full. + * + * @iq: Octeon Tx queue data structure. + * + * Return: 0, if the Tx queue is not full. + * 1, if the Tx queue is full. + */ +static int octep_vf_iq_full_check(struct octep_vf_iq *iq) +{ + int ret; + + ret = netif_subqueue_maybe_stop(iq->netdev, iq->q_no, IQ_INSTR_SPACE(iq), + OCTEP_VF_WAKE_QUEUE_THRESHOLD, + OCTEP_VF_WAKE_QUEUE_THRESHOLD); + switch (ret) { + case 0: /* Stopped the queue, since IQ is full */ + return 1; + case -1: /* + * Pending updates in write index from + * iq_process_completion in other cpus + * caused queues to get re-enabled after + * being stopped + */ + iq->stats.restart_cnt++; + fallthrough; + case 1: /* Queue left enabled, since IQ is not yet full*/ + return 0; + } + + return 1; +} + +/** * octep_vf_start_xmit() - Enqueue packet to Octoen hardware Tx Queue. * * @skb: packet skbuff pointer. @@ -183,6 +595,144 @@ static int octep_vf_stop(struct net_device *netdev) static netdev_tx_t octep_vf_start_xmit(struct sk_buff *skb, struct net_device *netdev) { + struct octep_vf_device *oct = netdev_priv(netdev); + netdev_features_t feat = netdev->features; + struct octep_vf_tx_sglist_desc *sglist; + struct octep_vf_tx_buffer *tx_buffer; + struct octep_vf_tx_desc_hw *hw_desc; + struct skb_shared_info *shinfo; + struct octep_vf_instr_hdr *ih; + struct octep_vf_iq *iq; + skb_frag_t *frag; + u16 nr_frags, si; + int xmit_more; + u16 q_no, wi; + + if (skb_put_padto(skb, ETH_ZLEN)) + return NETDEV_TX_OK; + + q_no = skb_get_queue_mapping(skb); + if (q_no >= oct->num_iqs) { + netdev_err(netdev, "Invalid Tx skb->queue_mapping=%d\n", q_no); + q_no = q_no % oct->num_iqs; + } + + iq = oct->iq[q_no]; + + shinfo = skb_shinfo(skb); + nr_frags = shinfo->nr_frags; + + wi = iq->host_write_index; + hw_desc = &iq->desc_ring[wi]; + hw_desc->ih64 = 0; + + tx_buffer = iq->buff_info + wi; + tx_buffer->skb = skb; + + ih = &hw_desc->ih; + ih->tlen = skb->len; + ih->pkind = oct->fw_info.pkind; + ih->fsz = oct->fw_info.fsz; + ih->tlen = skb->len + ih->fsz; + + if (!nr_frags) { + tx_buffer->gather = 0; + tx_buffer->dma = dma_map_single(iq->dev, skb->data, + skb->len, DMA_TO_DEVICE); + if (dma_mapping_error(iq->dev, tx_buffer->dma)) + goto dma_map_err; + hw_desc->dptr = tx_buffer->dma; + } else { + /* Scatter/Gather */ + dma_addr_t dma; + u16 len; + + sglist = tx_buffer->sglist; + + ih->gsz = nr_frags + 1; + ih->gather = 1; + tx_buffer->gather = 1; + + len = skb_headlen(skb); + dma = dma_map_single(iq->dev, skb->data, len, DMA_TO_DEVICE); + if (dma_mapping_error(iq->dev, dma)) + goto dma_map_err; + + memset(sglist, 0, OCTEP_VF_SGLIST_SIZE_PER_PKT); + sglist[0].len[3] = len; + sglist[0].dma_ptr[0] = dma; + + si = 1; /* entry 0 is main skb, mapped above */ + frag = &shinfo->frags[0]; + while (nr_frags--) { + len = skb_frag_size(frag); + dma = skb_frag_dma_map(iq->dev, frag, 0, + len, DMA_TO_DEVICE); + if (dma_mapping_error(iq->dev, dma)) + goto dma_map_sg_err; + + sglist[si >> 2].len[3 - (si & 3)] = len; + sglist[si >> 2].dma_ptr[si & 3] = dma; + + frag++; + si++; + } + hw_desc->dptr = tx_buffer->sglist_dma; + } + if (oct->fw_info.tx_ol_flags) { + if ((feat & (NETIF_F_TSO)) && (skb_is_gso(skb))) { + hw_desc->txm.ol_flags = OCTEP_VF_TX_OFFLOAD_CKSUM; + hw_desc->txm.ol_flags |= OCTEP_VF_TX_OFFLOAD_TSO; + hw_desc->txm.gso_size = skb_shinfo(skb)->gso_size; + hw_desc->txm.gso_segs = skb_shinfo(skb)->gso_segs; + } else if (feat & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { + hw_desc->txm.ol_flags = OCTEP_VF_TX_OFFLOAD_CKSUM; + } + /* due to ESR txm will be swapped by hw */ + hw_desc->txm64[0] = (__force u64)cpu_to_be64(hw_desc->txm64[0]); + } + + xmit_more = netdev_xmit_more(); + + netdev_tx_sent_queue(iq->netdev_q, skb->len); + + skb_tx_timestamp(skb); + iq->fill_cnt++; + wi++; + iq->host_write_index = wi & iq->ring_size_mask; + + /* octep_iq_full_check stops the queue and returns + * true if so, in case the queue has become full + * by inserting current packet. If so, we can + * go ahead and ring doorbell. + */ + if (!octep_vf_iq_full_check(iq) && xmit_more && + iq->fill_cnt < iq->fill_threshold) + return NETDEV_TX_OK; + + goto ring_dbell; + +dma_map_sg_err: + if (si > 0) { + dma_unmap_single(iq->dev, sglist[0].dma_ptr[0], + sglist[0].len[0], DMA_TO_DEVICE); + sglist[0].len[0] = 0; + } + while (si > 1) { + dma_unmap_page(iq->dev, sglist[si >> 2].dma_ptr[si & 3], + sglist[si >> 2].len[si & 3], DMA_TO_DEVICE); + sglist[si >> 2].len[si & 3] = 0; + si--; + } + tx_buffer->gather = 0; +dma_map_err: + dev_kfree_skb_any(skb); +ring_dbell: + /* Flush the hw descriptors before writing to doorbell */ + smp_wmb(); + writel(iq->fill_cnt, iq->doorbell_reg); + iq->stats.instr_posted += iq->fill_cnt; + iq->fill_cnt = 0; return NETDEV_TX_OK; } |