Add XDP support. Separate stack's and XDP's TX rings logically. Add functions for handling XDP_TX and cleanup of XDP's TX rings. For XDP allocate all RX buffers as separate pages and map them with DMA_BIDIRECTIONAL.
Signed-off-by: Jakub Kicinski <jakub.kicin...@netronome.com> --- drivers/net/ethernet/netronome/nfp/nfp_net.h | 17 +- .../net/ethernet/netronome/nfp/nfp_net_common.c | 474 +++++++++++++++++---- .../net/ethernet/netronome/nfp/nfp_net_debugfs.c | 37 +- .../net/ethernet/netronome/nfp/nfp_net_ethtool.c | 22 +- 4 files changed, 449 insertions(+), 101 deletions(-) diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net.h b/drivers/net/ethernet/netronome/nfp/nfp_net.h index 486e7c6453bc..abc9e56e93b8 100644 --- a/drivers/net/ethernet/netronome/nfp/nfp_net.h +++ b/drivers/net/ethernet/netronome/nfp/nfp_net.h @@ -171,7 +171,10 @@ struct nfp_net_tx_desc { * on the head's buffer). Equal to skb->len for non-TSO packets. */ struct nfp_net_tx_buf { - struct sk_buff *skb; + union { + struct sk_buff *skb; + void *frag; + }; dma_addr_t dma_addr; short int fidx; u16 pkt_cnt; @@ -341,6 +344,7 @@ struct nfp_net_rx_ring { * @napi: NAPI structure for this ring vec * @tx_ring: Pointer to TX ring * @rx_ring: Pointer to RX ring + * @xdp_ring: Pointer to an extra TX ring for XDP * @irq_idx: Index into MSI-X table * @rx_sync: Seqlock for atomic updates of RX stats * @rx_pkts: Number of received packets @@ -384,6 +388,8 @@ struct nfp_net_r_vector { u64 hw_csum_rx_inner_ok; u64 hw_csum_rx_error; + struct nfp_net_tx_ring *xdp_ring; + struct u64_stats_sync tx_sync; u64 tx_pkts; u64 tx_bytes; @@ -432,6 +438,7 @@ struct nfp_stat_pair { * @ctrl: Local copy of the control register/word. * @fl_bufsz: Currently configured size of the freelist buffers * @rx_offset: Offset in the RX buffers where packet data starts + * @xdp_prog: Installed XDP program * @cpp: Pointer to the CPP handle * @nfp_dev_cpp: Pointer to the NFP Device handle * @ctrl_area: Pointer to the CPP area for the control BAR @@ -451,6 +458,7 @@ struct nfp_stat_pair { * @max_tx_rings: Maximum number of TX rings supported by the Firmware * @max_rx_rings: Maximum number of RX rings supported by the Firmware * @num_tx_rings: Currently configured number of TX rings + * @num_stack_tx_rings: Number of TX rings used by the stack (not XDP) * @num_rx_rings: Currently configured number of RX rings * @txd_cnt: Size of the TX ring in number of descriptors * @rxd_cnt: Size of the RX ring in number of descriptors @@ -500,6 +508,8 @@ struct nfp_net { u32 rx_offset; + struct bpf_prog *xdp_prog; + struct nfp_net_tx_ring *tx_rings; struct nfp_net_rx_ring *rx_rings; @@ -532,6 +542,7 @@ struct nfp_net { unsigned int max_rx_rings; unsigned int num_tx_rings; + unsigned int num_stack_tx_rings; unsigned int num_rx_rings; int stride_tx; @@ -779,8 +790,8 @@ struct nfp_net * int nfp_net_irqs_alloc(struct nfp_net *nn); void nfp_net_irqs_disable(struct nfp_net *nn); int -nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_ring_set *rx, - struct nfp_net_ring_set *tx); +nfp_net_ring_reconfig(struct nfp_net *nn, struct bpf_prog **xdp_prog, + struct nfp_net_ring_set *rx, struct nfp_net_ring_set *tx); #ifdef CONFIG_NFP_NET_DEBUG void nfp_net_debugfs_create(void); diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c index 2ab63661a6fd..fa43dbcecc4f 100644 --- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c +++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c @@ -41,6 +41,7 @@ * Chris Telfer <chris.tel...@netronome.com> */ +#include <linux/bpf.h> #include <linux/module.h> #include <linux/kernel.h> #include <linux/init.h> @@ -490,6 +491,7 @@ static void nfp_net_irqs_assign(struct net_device *netdev) nn->num_rx_rings = min(nn->num_r_vecs, nn->num_rx_rings); nn->num_tx_rings = min(nn->num_r_vecs, nn->num_tx_rings); + nn->num_stack_tx_rings = nn->num_tx_rings; nn->lsc_handler = nfp_net_irq_lsc; nn->exn_handler = nfp_net_irq_exn; @@ -713,6 +715,13 @@ static void nfp_net_tx_csum(struct nfp_net *nn, struct nfp_net_r_vector *r_vec, u64_stats_update_end(&r_vec->tx_sync); } +static void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring) +{ + wmb(); + nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add); + tx_ring->wr_ptr_add = 0; +} + /** * nfp_net_tx() - Main transmit entry point * @skb: SKB to transmit @@ -827,12 +836,8 @@ static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev) nfp_net_tx_ring_stop(nd_q, tx_ring); tx_ring->wr_ptr_add += nr_frags + 1; - if (!skb->xmit_more || netif_xmit_stopped(nd_q)) { - /* force memory write before we let HW know */ - wmb(); - nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add); - tx_ring->wr_ptr_add = 0; - } + if (!skb->xmit_more || netif_xmit_stopped(nd_q)) + nfp_net_tx_xmit_more_flush(tx_ring); skb_tx_timestamp(skb); @@ -954,6 +959,56 @@ static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring) tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt); } +static void nfp_net_xdp_complete(struct nfp_net_tx_ring *tx_ring) +{ + struct nfp_net_r_vector *r_vec = tx_ring->r_vec; + struct nfp_net *nn = r_vec->nfp_net; + u32 done_pkts = 0, done_bytes = 0; + int idx, todo; + u32 qcp_rd_p; + + /* Work out how many descriptors have been transmitted */ + qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q); + + if (qcp_rd_p == tx_ring->qcp_rd_p) + return; + + if (qcp_rd_p > tx_ring->qcp_rd_p) + todo = qcp_rd_p - tx_ring->qcp_rd_p; + else + todo = qcp_rd_p + tx_ring->cnt - tx_ring->qcp_rd_p; + + while (todo--) { + idx = tx_ring->rd_p & (tx_ring->cnt - 1); + tx_ring->rd_p++; + + if (!tx_ring->txbufs[idx].frag) + continue; + + nfp_net_dma_unmap_rx(nn, tx_ring->txbufs[idx].dma_addr, + nn->fl_bufsz, DMA_BIDIRECTIONAL); + __free_page(virt_to_page(tx_ring->txbufs[idx].frag)); + + done_pkts++; + done_bytes += tx_ring->txbufs[idx].real_len; + + tx_ring->txbufs[idx].dma_addr = 0; + tx_ring->txbufs[idx].frag = NULL; + tx_ring->txbufs[idx].fidx = -2; + } + + tx_ring->qcp_rd_p = qcp_rd_p; + + u64_stats_update_begin(&r_vec->tx_sync); + r_vec->tx_bytes += done_bytes; + r_vec->tx_pkts += done_pkts; + u64_stats_update_end(&r_vec->tx_sync); + + WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt, + "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n", + tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt); +} + /** * nfp_net_tx_ring_reset() - Free any untransmitted buffers and reset pointers * @nn: NFP Net device @@ -964,39 +1019,47 @@ static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring) static void nfp_net_tx_ring_reset(struct nfp_net *nn, struct nfp_net_tx_ring *tx_ring) { + struct nfp_net_r_vector *r_vec = tx_ring->r_vec; const struct skb_frag_struct *frag; - struct netdev_queue *nd_q; struct pci_dev *pdev = nn->pdev; + struct netdev_queue *nd_q; while (tx_ring->rd_p != tx_ring->wr_p) { - int nr_frags, fidx, idx; - struct sk_buff *skb; + struct nfp_net_tx_buf *tx_buf; + int idx; idx = tx_ring->rd_p & (tx_ring->cnt - 1); - skb = tx_ring->txbufs[idx].skb; - nr_frags = skb_shinfo(skb)->nr_frags; - fidx = tx_ring->txbufs[idx].fidx; + tx_buf = &tx_ring->txbufs[idx]; - if (fidx == -1) { - /* unmap head */ - dma_unmap_single(&pdev->dev, - tx_ring->txbufs[idx].dma_addr, - skb_headlen(skb), DMA_TO_DEVICE); + if (tx_ring == r_vec->xdp_ring) { + nfp_net_dma_unmap_rx(nn, tx_buf->dma_addr, + nn->fl_bufsz, DMA_BIDIRECTIONAL); + __free_page(virt_to_page(tx_ring->txbufs[idx].frag)); } else { - /* unmap fragment */ - frag = &skb_shinfo(skb)->frags[fidx]; - dma_unmap_page(&pdev->dev, - tx_ring->txbufs[idx].dma_addr, - skb_frag_size(frag), DMA_TO_DEVICE); - } + struct sk_buff *skb = tx_ring->txbufs[idx].skb; + int nr_frags = skb_shinfo(skb)->nr_frags; + + if (tx_buf->fidx == -1) { + /* unmap head */ + dma_unmap_single(&pdev->dev, tx_buf->dma_addr, + skb_headlen(skb), + DMA_TO_DEVICE); + } else { + /* unmap fragment */ + frag = &skb_shinfo(skb)->frags[tx_buf->fidx]; + dma_unmap_page(&pdev->dev, tx_buf->dma_addr, + skb_frag_size(frag), + DMA_TO_DEVICE); + } - /* check for last gather fragment */ - if (fidx == nr_frags - 1) - dev_kfree_skb_any(skb); + /* check for last gather fragment */ + if (tx_buf->fidx == nr_frags - 1) + dev_kfree_skb_any(skb); + } - tx_ring->txbufs[idx].dma_addr = 0; - tx_ring->txbufs[idx].skb = NULL; - tx_ring->txbufs[idx].fidx = -2; + tx_buf->dma_addr = 0; + tx_buf->skb = NULL; + tx_buf->fidx = -2; tx_ring->qcp_rd_p++; tx_ring->rd_p++; @@ -1008,6 +1071,9 @@ static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring) tx_ring->qcp_rd_p = 0; tx_ring->wr_ptr_add = 0; + if (tx_ring == r_vec->xdp_ring) + return; + nd_q = netdev_get_tx_queue(nn->netdev, tx_ring->idx); netdev_tx_reset_queue(nd_q); } @@ -1017,7 +1083,7 @@ static void nfp_net_tx_timeout(struct net_device *netdev) struct nfp_net *nn = netdev_priv(netdev); int i; - for (i = 0; i < nn->num_tx_rings; i++) { + for (i = 0; i < nn->netdev->real_num_tx_queues; i++) { if (!netif_tx_queue_stopped(netdev_get_tx_queue(netdev, i))) continue; nn_warn(nn, "TX timeout on ring: %d\n", i); @@ -1045,11 +1111,21 @@ static void nfp_net_tx_timeout(struct net_device *netdev) return fl_bufsz; } +static void +nfp_net_free_frag(void *frag, bool xdp) +{ + if (!xdp) + skb_free_frag(frag); + else + __free_page(virt_to_page(frag)); +} + /** * nfp_net_rx_alloc_one() - Allocate and map page frag for RX * @rx_ring: RX ring structure of the skb * @dma_addr: Pointer to storage for DMA address (output param) * @fl_bufsz: size of freelist buffers + * @xdp: Whether XDP is enabled * * This function will allcate a new page frag, map it for DMA. * @@ -1057,20 +1133,26 @@ static void nfp_net_tx_timeout(struct net_device *netdev) */ static void * nfp_net_rx_alloc_one(struct nfp_net_rx_ring *rx_ring, dma_addr_t *dma_addr, - unsigned int fl_bufsz) + unsigned int fl_bufsz, bool xdp) { struct nfp_net *nn = rx_ring->r_vec->nfp_net; + int direction; void *frag; - frag = netdev_alloc_frag(fl_bufsz); + if (!xdp) + frag = netdev_alloc_frag(fl_bufsz); + else + frag = page_address(alloc_page(GFP_KERNEL | __GFP_COLD)); if (!frag) { nn_warn_ratelimit(nn, "Failed to alloc receive page frag\n"); return NULL; } - *dma_addr = nfp_net_dma_map_rx(nn, frag, fl_bufsz, DMA_FROM_DEVICE); + direction = xdp ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE; + + *dma_addr = nfp_net_dma_map_rx(nn, frag, fl_bufsz, direction); if (dma_mapping_error(&nn->pdev->dev, *dma_addr)) { - skb_free_frag(frag); + nfp_net_free_frag(frag, xdp); nn_warn_ratelimit(nn, "Failed to map DMA RX buffer\n"); return NULL; } @@ -1078,19 +1160,23 @@ static void nfp_net_tx_timeout(struct net_device *netdev) return frag; } -static void *nfp_net_napi_alloc_one(struct nfp_net *nn, dma_addr_t *dma_addr) +static void * +nfp_net_napi_alloc_one(struct nfp_net *nn, int direction, dma_addr_t *dma_addr) { void *frag; - frag = napi_alloc_frag(nn->fl_bufsz); + if (!nn->xdp_prog) + frag = napi_alloc_frag(nn->fl_bufsz); + else + frag = page_address(alloc_page(GFP_ATOMIC | __GFP_COLD)); if (!frag) { nn_warn_ratelimit(nn, "Failed to alloc receive page frag\n"); return NULL; } - *dma_addr = nfp_net_dma_map_rx(nn, frag, nn->fl_bufsz, DMA_FROM_DEVICE); + *dma_addr = nfp_net_dma_map_rx(nn, frag, nn->fl_bufsz, direction); if (dma_mapping_error(&nn->pdev->dev, *dma_addr)) { - skb_free_frag(frag); + nfp_net_free_frag(frag, nn->xdp_prog); nn_warn_ratelimit(nn, "Failed to map DMA RX buffer\n"); return NULL; } @@ -1161,14 +1247,17 @@ static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring) * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring * @nn: NFP Net device * @rx_ring: RX ring to remove buffers from + * @xdp: Whether XDP is enabled * * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1) * entries. After device is disabled nfp_net_rx_ring_reset() must be called * to restore required ring geometry. */ static void -nfp_net_rx_ring_bufs_free(struct nfp_net *nn, struct nfp_net_rx_ring *rx_ring) +nfp_net_rx_ring_bufs_free(struct nfp_net *nn, struct nfp_net_rx_ring *rx_ring, + bool xdp) { + int direction = xdp ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE; unsigned int i; for (i = 0; i < rx_ring->cnt - 1; i++) { @@ -1180,8 +1269,8 @@ static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring) continue; nfp_net_dma_unmap_rx(nn, rx_ring->rxbufs[i].dma_addr, - rx_ring->bufsz, DMA_FROM_DEVICE); - skb_free_frag(rx_ring->rxbufs[i].frag); + rx_ring->bufsz, direction); + nfp_net_free_frag(rx_ring->rxbufs[i].frag, xdp); rx_ring->rxbufs[i].dma_addr = 0; rx_ring->rxbufs[i].frag = NULL; } @@ -1191,9 +1280,11 @@ static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring) * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW) * @nn: NFP Net device * @rx_ring: RX ring to remove buffers from + * @xdp: Whether XDP is enabled */ static int -nfp_net_rx_ring_bufs_alloc(struct nfp_net *nn, struct nfp_net_rx_ring *rx_ring) +nfp_net_rx_ring_bufs_alloc(struct nfp_net *nn, struct nfp_net_rx_ring *rx_ring, + bool xdp) { struct nfp_net_rx_buf *rxbufs; unsigned int i; @@ -1203,9 +1294,9 @@ static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring) for (i = 0; i < rx_ring->cnt - 1; i++) { rxbufs[i].frag = nfp_net_rx_alloc_one(rx_ring, &rxbufs[i].dma_addr, - rx_ring->bufsz); + rx_ring->bufsz, xdp); if (!rxbufs[i].frag) { - nfp_net_rx_ring_bufs_free(nn, rx_ring); + nfp_net_rx_ring_bufs_free(nn, rx_ring, xdp); return -ENOMEM; } } @@ -1368,6 +1459,68 @@ static void nfp_net_set_hash(struct net_device *netdev, struct sk_buff *skb, dev_kfree_skb_any(skb); } +static void +nfp_net_tx_xdp_buf(struct nfp_net *nn, struct nfp_net_rx_ring *rx_ring, + struct nfp_net_tx_ring *tx_ring, + struct nfp_net_rx_buf *rxbuf, unsigned int pkt_off, + unsigned int pkt_len) +{ + struct nfp_net_tx_buf *txbuf; + struct nfp_net_tx_desc *txd; + dma_addr_t new_dma_addr; + void *new_frag; + int wr_idx; + + if (unlikely(nfp_net_tx_full(tx_ring, 1))) { + nfp_net_rx_drop(rx_ring->r_vec, rx_ring, rxbuf, NULL); + return; + } + + new_frag = nfp_net_napi_alloc_one(nn, DMA_BIDIRECTIONAL, &new_dma_addr); + if (unlikely(!new_frag)) { + nfp_net_rx_drop(rx_ring->r_vec, rx_ring, rxbuf, NULL); + return; + } + nfp_net_rx_give_one(rx_ring, new_frag, new_dma_addr); + + wr_idx = tx_ring->wr_p & (tx_ring->cnt - 1); + + /* Stash the soft descriptor of the head then initialize it */ + txbuf = &tx_ring->txbufs[wr_idx]; + txbuf->frag = rxbuf->frag; + txbuf->dma_addr = rxbuf->dma_addr; + txbuf->fidx = -1; + txbuf->pkt_cnt = 1; + txbuf->real_len = pkt_len; + + dma_sync_single_for_device(&nn->pdev->dev, rxbuf->dma_addr + pkt_off, + pkt_len, DMA_TO_DEVICE); + + /* Build TX descriptor */ + txd = &tx_ring->txds[wr_idx]; + txd->offset_eop = PCIE_DESC_TX_EOP; + txd->dma_len = cpu_to_le16(pkt_len); + nfp_desc_set_dma_addr(txd, rxbuf->dma_addr + pkt_off); + txd->data_len = cpu_to_le16(pkt_len); + + txd->flags = 0; + txd->mss = 0; + txd->l4_offset = 0; + + tx_ring->wr_p++; + tx_ring->wr_ptr_add++; +} + +static int nfp_net_run_xdp(struct bpf_prog *prog, void *data, unsigned int len) +{ + struct xdp_buff xdp; + + xdp.data = data; + xdp.data_end = data + len; + + return BPF_PROG_RUN(prog, (void *)&xdp); +} + /** * nfp_net_rx() - receive up to @budget packets on @rx_ring * @rx_ring: RX ring to receive from @@ -1383,10 +1536,20 @@ static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget) { struct nfp_net_r_vector *r_vec = rx_ring->r_vec; struct nfp_net *nn = r_vec->nfp_net; + struct nfp_net_tx_ring *tx_ring; + struct bpf_prog *xdp_prog; + unsigned int true_bufsz; struct sk_buff *skb; int pkts_polled = 0; + int rx_dma_map_dir; int idx; + rcu_read_lock(); + xdp_prog = READ_ONCE(nn->xdp_prog); + rx_dma_map_dir = xdp_prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE; + true_bufsz = xdp_prog ? PAGE_SIZE : nn->fl_bufsz; + tx_ring = r_vec->xdp_ring; + while (pkts_polled < budget) { unsigned int meta_len, data_len, data_off, pkt_len, pkt_off; struct nfp_net_rx_buf *rxbuf; @@ -1437,19 +1600,45 @@ static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget) r_vec->rx_bytes += pkt_len; u64_stats_update_end(&r_vec->rx_sync); - skb = build_skb(rxbuf->frag, nn->fl_bufsz); + if (xdp_prog) { + int act; + + dma_sync_single_for_cpu(&nn->pdev->dev, + rxbuf->dma_addr + pkt_off, + pkt_len, DMA_FROM_DEVICE); + act = nfp_net_run_xdp(xdp_prog, rxbuf->frag + data_off, + pkt_len); + switch (act) { + case XDP_PASS: + break; + case XDP_TX: + nfp_net_tx_xdp_buf(nn, rx_ring, tx_ring, rxbuf, + pkt_off, pkt_len); + continue; + default: + bpf_warn_invalid_xdp_action(act); + case XDP_ABORTED: + case XDP_DROP: + nfp_net_rx_give_one(rx_ring, rxbuf->frag, + rxbuf->dma_addr); + continue; + } + } + + skb = build_skb(rxbuf->frag, true_bufsz); if (unlikely(!skb)) { nfp_net_rx_drop(r_vec, rx_ring, rxbuf, NULL); continue; } - new_frag = nfp_net_napi_alloc_one(nn, &new_dma_addr); + new_frag = nfp_net_napi_alloc_one(nn, rx_dma_map_dir, + &new_dma_addr); if (unlikely(!new_frag)) { nfp_net_rx_drop(r_vec, rx_ring, rxbuf, skb); continue; } - nfp_net_dma_unmap_rx(nn, rx_ring->rxbufs[idx].dma_addr, - nn->fl_bufsz, DMA_FROM_DEVICE); + nfp_net_dma_unmap_rx(nn, rxbuf->dma_addr, nn->fl_bufsz, + rx_dma_map_dir); nfp_net_rx_give_one(rx_ring, new_frag, new_dma_addr); @@ -1481,6 +1670,10 @@ static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget) napi_gro_receive(&rx_ring->r_vec->napi, skb); } + if (xdp_prog && tx_ring->wr_ptr_add) + nfp_net_tx_xmit_more_flush(tx_ring); + rcu_read_unlock(); + return pkts_polled; } @@ -1499,8 +1692,11 @@ static int nfp_net_poll(struct napi_struct *napi, int budget) if (r_vec->tx_ring) nfp_net_tx_complete(r_vec->tx_ring); - if (r_vec->rx_ring) + if (r_vec->rx_ring) { pkts_polled = nfp_net_rx(r_vec->rx_ring, budget); + if (r_vec->xdp_ring) + nfp_net_xdp_complete(r_vec->xdp_ring); + } if (pkts_polled < budget) { napi_complete_done(napi, pkts_polled); @@ -1540,10 +1736,12 @@ static void nfp_net_tx_ring_free(struct nfp_net_tx_ring *tx_ring) * nfp_net_tx_ring_alloc() - Allocate resource for a TX ring * @tx_ring: TX Ring structure to allocate * @cnt: Ring buffer count + * @is_xdp: True if ring will be used for XDP * * Return: 0 on success, negative errno otherwise. */ -static int nfp_net_tx_ring_alloc(struct nfp_net_tx_ring *tx_ring, u32 cnt) +static int +nfp_net_tx_ring_alloc(struct nfp_net_tx_ring *tx_ring, u32 cnt, bool is_xdp) { struct nfp_net_r_vector *r_vec = tx_ring->r_vec; struct nfp_net *nn = r_vec->nfp_net; @@ -1563,11 +1761,14 @@ static int nfp_net_tx_ring_alloc(struct nfp_net_tx_ring *tx_ring, u32 cnt) if (!tx_ring->txbufs) goto err_alloc; - netif_set_xps_queue(nn->netdev, &r_vec->affinity_mask, tx_ring->idx); + if (!is_xdp) + netif_set_xps_queue(nn->netdev, &r_vec->affinity_mask, + tx_ring->idx); - nn_dbg(nn, "TxQ%02d: QCidx=%02d cnt=%d dma=%#llx host=%p\n", + nn_dbg(nn, "TxQ%02d: QCidx=%02d cnt=%d dma=%#llx host=%p %s\n", tx_ring->idx, tx_ring->qcidx, - tx_ring->cnt, (unsigned long long)tx_ring->dma, tx_ring->txds); + tx_ring->cnt, (unsigned long long)tx_ring->dma, tx_ring->txds, + is_xdp ? "XDP" : ""); return 0; @@ -1577,7 +1778,8 @@ static int nfp_net_tx_ring_alloc(struct nfp_net_tx_ring *tx_ring, u32 cnt) } static struct nfp_net_tx_ring * -nfp_net_tx_ring_set_prepare(struct nfp_net *nn, struct nfp_net_ring_set *s) +nfp_net_tx_ring_set_prepare(struct nfp_net *nn, struct nfp_net_ring_set *s, + unsigned int num_stack_tx_rings) { struct nfp_net_tx_ring *rings; unsigned int r; @@ -1587,9 +1789,14 @@ static int nfp_net_tx_ring_alloc(struct nfp_net_tx_ring *tx_ring, u32 cnt) return NULL; for (r = 0; r < s->n_rings; r++) { - nfp_net_tx_ring_init(&rings[r], &nn->r_vecs[r], r); + int bias = 0; + + if (r >= num_stack_tx_rings) + bias = num_stack_tx_rings; - if (nfp_net_tx_ring_alloc(&rings[r], s->dcnt)) + nfp_net_tx_ring_init(&rings[r], &nn->r_vecs[r - bias], r); + + if (nfp_net_tx_ring_alloc(&rings[r], s->dcnt, bias)) goto err_free_prev; } @@ -1694,7 +1901,8 @@ static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring) } static struct nfp_net_rx_ring * -nfp_net_rx_ring_set_prepare(struct nfp_net *nn, struct nfp_net_ring_set *s) +nfp_net_rx_ring_set_prepare(struct nfp_net *nn, struct nfp_net_ring_set *s, + bool xdp) { unsigned int fl_bufsz = nfp_net_calc_fl_bufsz(nn, s->mtu); struct nfp_net_rx_ring *rings; @@ -1710,7 +1918,7 @@ static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring) if (nfp_net_rx_ring_alloc(&rings[r], fl_bufsz, s->dcnt)) goto err_free_prev; - if (nfp_net_rx_ring_bufs_alloc(nn, &rings[r])) + if (nfp_net_rx_ring_bufs_alloc(nn, &rings[r], xdp)) goto err_free_ring; } @@ -1718,7 +1926,7 @@ static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring) err_free_prev: while (r--) { - nfp_net_rx_ring_bufs_free(nn, &rings[r]); + nfp_net_rx_ring_bufs_free(nn, &rings[r], xdp); err_free_ring: nfp_net_rx_ring_free(&rings[r]); } @@ -1744,13 +1952,14 @@ static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring) } static void -nfp_net_rx_ring_set_free(struct nfp_net *nn, struct nfp_net_ring_set *s) +nfp_net_rx_ring_set_free(struct nfp_net *nn, struct nfp_net_ring_set *s, + bool xdp) { struct nfp_net_rx_ring *rings = s->rings; unsigned int r; for (r = 0; r < s->n_rings; r++) { - nfp_net_rx_ring_bufs_free(nn, &rings[r]); + nfp_net_rx_ring_bufs_free(nn, &rings[r], xdp); nfp_net_rx_ring_free(&rings[r]); } @@ -1762,7 +1971,11 @@ static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring) int idx) { r_vec->rx_ring = idx < nn->num_rx_rings ? &nn->rx_rings[idx] : NULL; - r_vec->tx_ring = idx < nn->num_tx_rings ? &nn->tx_rings[idx] : NULL; + r_vec->tx_ring = + idx < nn->num_stack_tx_rings ? &nn->tx_rings[idx] : NULL; + + r_vec->xdp_ring = idx < nn->num_tx_rings - nn->num_stack_tx_rings ? + &nn->tx_rings[nn->num_stack_tx_rings + idx] : NULL; } static int @@ -2083,13 +2296,14 @@ static int nfp_net_netdev_open(struct net_device *netdev) goto err_cleanup_vec_p; } - nn->rx_rings = nfp_net_rx_ring_set_prepare(nn, &rx); + nn->rx_rings = nfp_net_rx_ring_set_prepare(nn, &rx, nn->xdp_prog); if (!nn->rx_rings) { err = -ENOMEM; goto err_cleanup_vec; } - nn->tx_rings = nfp_net_tx_ring_set_prepare(nn, &tx); + nn->tx_rings = nfp_net_tx_ring_set_prepare(nn, &tx, + nn->num_stack_tx_rings); if (!nn->tx_rings) { err = -ENOMEM; goto err_free_rx_rings; @@ -2098,7 +2312,7 @@ static int nfp_net_netdev_open(struct net_device *netdev) for (r = 0; r < nn->max_r_vecs; r++) nfp_net_vector_assign_rings(nn, &nn->r_vecs[r], r); - err = netif_set_real_num_tx_queues(netdev, nn->num_tx_rings); + err = netif_set_real_num_tx_queues(netdev, nn->num_stack_tx_rings); if (err) goto err_free_rings; @@ -2130,7 +2344,7 @@ static int nfp_net_netdev_open(struct net_device *netdev) err_free_rings: nfp_net_tx_ring_set_free(nn, &tx); err_free_rx_rings: - nfp_net_rx_ring_set_free(nn, &rx); + nfp_net_rx_ring_set_free(nn, &rx, nn->xdp_prog); err_cleanup_vec: r = nn->num_r_vecs; err_cleanup_vec_p: @@ -2171,7 +2385,7 @@ static void nfp_net_close_free_all(struct nfp_net *nn) unsigned int r; for (r = 0; r < nn->num_rx_rings; r++) { - nfp_net_rx_ring_bufs_free(nn, &nn->rx_rings[r]); + nfp_net_rx_ring_bufs_free(nn, &nn->rx_rings[r], nn->xdp_prog); nfp_net_rx_ring_free(&nn->rx_rings[r]); } for (r = 0; r < nn->num_tx_rings; r++) @@ -2251,6 +2465,8 @@ static void nfp_net_rss_init_itbl(struct nfp_net *nn) static int nfp_net_ring_swap_enable(struct nfp_net *nn, unsigned int *num_vecs, + unsigned int *stack_tx_rings, + struct bpf_prog **xdp_prog, struct nfp_net_ring_set *rx, struct nfp_net_ring_set *tx) { @@ -2263,6 +2479,8 @@ static void nfp_net_rss_init_itbl(struct nfp_net *nn) nfp_net_tx_ring_set_swap(nn, tx); swap(*num_vecs, nn->num_r_vecs); + swap(*stack_tx_rings, nn->num_stack_tx_rings); + *xdp_prog = xchg(&nn->xdp_prog, *xdp_prog); for (r = 0; r < nn->max_r_vecs; r++) nfp_net_vector_assign_rings(nn, &nn->r_vecs[r], r); @@ -2277,9 +2495,9 @@ static void nfp_net_rss_init_itbl(struct nfp_net *nn) return err; } - if (nn->netdev->real_num_tx_queues != nn->num_tx_rings) { + if (nn->netdev->real_num_tx_queues != nn->num_stack_tx_rings) { err = netif_set_real_num_tx_queues(nn->netdev, - nn->num_tx_rings); + nn->num_stack_tx_rings); if (err) return err; } @@ -2287,11 +2505,30 @@ static void nfp_net_rss_init_itbl(struct nfp_net *nn) return __nfp_net_set_config_and_enable(nn); } +static int +nfp_net_check_config(struct nfp_net *nn, struct bpf_prog *xdp_prog, + struct nfp_net_ring_set *rx, struct nfp_net_ring_set *tx) +{ + /* XDP-enabled tests */ + if (!xdp_prog) + return 0; + if (rx && nfp_net_calc_fl_bufsz(nn, rx->mtu) > PAGE_SIZE) { + nn_warn(nn, "MTU too large w/ XDP enabled\n"); + return -EINVAL; + } + if (tx && tx->n_rings > nn->max_tx_rings) { + nn_warn(nn, "Insufficient number of TX rings w/ XDP enabled\n"); + return -EINVAL; + } + + return 0; +} + static void -nfp_net_ring_reconfig_down(struct nfp_net *nn, +nfp_net_ring_reconfig_down(struct nfp_net *nn, struct bpf_prog **xdp_prog, struct nfp_net_ring_set *rx, struct nfp_net_ring_set *tx, - unsigned int num_vecs) + unsigned int stack_tx_rings, unsigned int num_vecs) { nn->netdev->mtu = rx ? rx->mtu : nn->netdev->mtu; nn->fl_bufsz = nfp_net_calc_fl_bufsz(nn, nn->netdev->mtu); @@ -2299,24 +2536,34 @@ static void nfp_net_rss_init_itbl(struct nfp_net *nn) nn->txd_cnt = tx ? tx->dcnt : nn->txd_cnt; nn->num_rx_rings = rx ? rx->n_rings : nn->num_rx_rings; nn->num_tx_rings = tx ? tx->n_rings : nn->num_tx_rings; + nn->num_stack_tx_rings = stack_tx_rings; nn->num_r_vecs = num_vecs; + *xdp_prog = xchg(&nn->xdp_prog, *xdp_prog); if (!netif_is_rxfh_configured(nn->netdev)) nfp_net_rss_init_itbl(nn); } int -nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_ring_set *rx, - struct nfp_net_ring_set *tx) +nfp_net_ring_reconfig(struct nfp_net *nn, struct bpf_prog **xdp_prog, + struct nfp_net_ring_set *rx, struct nfp_net_ring_set *tx) { - unsigned int num_vecs, r; + unsigned int stack_tx_rings, num_vecs, r; int err; - num_vecs = max(rx ? rx->n_rings : nn->num_rx_rings, - tx ? tx->n_rings : nn->num_tx_rings); + stack_tx_rings = tx ? tx->n_rings : nn->num_tx_rings; + if (*xdp_prog) + stack_tx_rings -= rx ? rx->n_rings : nn->num_rx_rings; + + num_vecs = max(rx ? rx->n_rings : nn->num_rx_rings, stack_tx_rings); + + err = nfp_net_check_config(nn, *xdp_prog, rx, tx); + if (err) + return err; if (!netif_running(nn->netdev)) { - nfp_net_ring_reconfig_down(nn, rx, tx, num_vecs); + nfp_net_ring_reconfig_down(nn, xdp_prog, rx, tx, + stack_tx_rings, num_vecs); return 0; } @@ -2329,13 +2576,13 @@ static void nfp_net_rss_init_itbl(struct nfp_net *nn) } } if (rx) { - if (!nfp_net_rx_ring_set_prepare(nn, rx)) { + if (!nfp_net_rx_ring_set_prepare(nn, rx, *xdp_prog)) { err = -ENOMEM; goto err_cleanup_vecs; } } if (tx) { - if (!nfp_net_tx_ring_set_prepare(nn, tx)) { + if (!nfp_net_tx_ring_set_prepare(nn, tx, stack_tx_rings)) { err = -ENOMEM; goto err_free_rx; } @@ -2345,14 +2592,16 @@ static void nfp_net_rss_init_itbl(struct nfp_net *nn) nfp_net_close_stack(nn); nfp_net_clear_config_and_disable(nn); - err = nfp_net_ring_swap_enable(nn, &num_vecs, rx, tx); + err = nfp_net_ring_swap_enable(nn, &num_vecs, &stack_tx_rings, + xdp_prog, rx, tx); if (err) { int err2; nfp_net_clear_config_and_disable(nn); /* Try with old configuration and old rings */ - err2 = nfp_net_ring_swap_enable(nn, &num_vecs, rx, tx); + err2 = nfp_net_ring_swap_enable(nn, &num_vecs, &stack_tx_rings, + xdp_prog, rx, tx); if (err2) nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n", err, err2); @@ -2361,7 +2610,7 @@ static void nfp_net_rss_init_itbl(struct nfp_net *nn) nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); if (rx) - nfp_net_rx_ring_set_free(nn, rx); + nfp_net_rx_ring_set_free(nn, rx, *xdp_prog); if (tx) nfp_net_tx_ring_set_free(nn, tx); @@ -2371,7 +2620,7 @@ static void nfp_net_rss_init_itbl(struct nfp_net *nn) err_free_rx: if (rx) - nfp_net_rx_ring_set_free(nn, rx); + nfp_net_rx_ring_set_free(nn, rx, *xdp_prog); err_cleanup_vecs: for (r = num_vecs - 1; r >= nn->num_r_vecs; r--) nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); @@ -2387,7 +2636,7 @@ static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu) .dcnt = nn->rxd_cnt, }; - return nfp_net_ring_reconfig(nn, &rx, NULL); + return nfp_net_ring_reconfig(nn, &nn->xdp_prog, &rx, NULL); } static struct rtnl_link_stats64 *nfp_net_stat64(struct net_device *netdev, @@ -2653,6 +2902,56 @@ static void nfp_net_del_vxlan_port(struct net_device *netdev, nfp_net_set_vxlan_port(nn, idx, 0); } +static int nfp_net_xdp_setup(struct nfp_net *nn, struct bpf_prog *prog) +{ + struct nfp_net_ring_set rx = { + .n_rings = nn->num_rx_rings, + .mtu = nn->netdev->mtu, + .dcnt = nn->rxd_cnt, + }; + struct nfp_net_ring_set tx = { + .n_rings = nn->num_tx_rings, + .dcnt = nn->txd_cnt, + }; + int err; + + if (!prog && !nn->xdp_prog) + return 0; + if (prog && nn->xdp_prog) { + prog = xchg(&nn->xdp_prog, prog); + bpf_prog_put(prog); + return 0; + } + + tx.n_rings += prog ? nn->num_rx_rings : -nn->num_rx_rings; + + /* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */ + err = nfp_net_ring_reconfig(nn, &prog, &rx, &tx); + if (err) + return err; + + /* @prog got swapped and is now the old one */ + if (prog) + bpf_prog_put(prog); + + return 0; +} + +static int nfp_net_xdp(struct net_device *netdev, struct netdev_xdp *xdp) +{ + struct nfp_net *nn = netdev_priv(netdev); + + switch (xdp->command) { + case XDP_SETUP_PROG: + return nfp_net_xdp_setup(nn, xdp->prog); + case XDP_QUERY_PROG: + xdp->prog_attached = !!nn->xdp_prog; + return 0; + default: + return -EINVAL; + } +} + static const struct net_device_ops nfp_net_netdev_ops = { .ndo_open = nfp_net_netdev_open, .ndo_stop = nfp_net_netdev_close, @@ -2667,6 +2966,7 @@ static void nfp_net_del_vxlan_port(struct net_device *netdev, .ndo_features_check = nfp_net_features_check, .ndo_udp_tunnel_add = nfp_net_add_vxlan_port, .ndo_udp_tunnel_del = nfp_net_del_vxlan_port, + .ndo_xdp = nfp_net_xdp, }; /** @@ -2929,5 +3229,9 @@ int nfp_net_netdev_init(struct net_device *netdev) */ void nfp_net_netdev_clean(struct net_device *netdev) { - unregister_netdev(netdev); + struct nfp_net *nn = netdev_priv(netdev); + + if (nn->xdp_prog) + bpf_prog_put(nn->xdp_prog); + unregister_netdev(nn->netdev); } diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c b/drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c index e90df8145323..c66f3f954aa8 100644 --- a/drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c +++ b/drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c @@ -114,6 +114,16 @@ static int nfp_net_debugfs_rx_q_open(struct inode *inode, struct file *f) .llseek = seq_lseek }; +static int nfp_net_debugfs_tx_q_open(struct inode *inode, struct file *f); + +static const struct file_operations nfp_tx_q_fops = { + .owner = THIS_MODULE, + .open = nfp_net_debugfs_tx_q_open, + .release = single_release, + .read = seq_read, + .llseek = seq_lseek +}; + static int nfp_net_debugfs_tx_q_read(struct seq_file *file, void *data) { struct nfp_net_r_vector *r_vec = file->private; @@ -126,10 +136,13 @@ static int nfp_net_debugfs_tx_q_read(struct seq_file *file, void *data) rtnl_lock(); - if (!r_vec->nfp_net || !r_vec->tx_ring) + if (debugfs_real_fops(file->file) == &nfp_tx_q_fops) + tx_ring = r_vec->tx_ring; + else + tx_ring = r_vec->xdp_ring; + if (!r_vec->nfp_net || !tx_ring) goto out; nn = r_vec->nfp_net; - tx_ring = r_vec->tx_ring; if (!netif_running(nn->netdev)) goto out; @@ -148,9 +161,14 @@ static int nfp_net_debugfs_tx_q_read(struct seq_file *file, void *data) txd->vals[2], txd->vals[3]); skb = READ_ONCE(tx_ring->txbufs[i].skb); - if (skb) - seq_printf(file, " skb->head=%p skb->data=%p", - skb->head, skb->data); + if (skb) { + if (tx_ring == r_vec->tx_ring) + seq_printf(file, " skb->head=%p skb->data=%p", + skb->head, skb->data); + else + seq_printf(file, " frag=%p", skb); + } + if (tx_ring->txbufs[i].dma_addr) seq_printf(file, " dma_addr=%pad", &tx_ring->txbufs[i].dma_addr); @@ -176,7 +194,7 @@ static int nfp_net_debugfs_tx_q_open(struct inode *inode, struct file *f) return single_open(f, nfp_net_debugfs_tx_q_read, inode->i_private); } -static const struct file_operations nfp_tx_q_fops = { +static const struct file_operations nfp_xdp_q_fops = { .owner = THIS_MODULE, .open = nfp_net_debugfs_tx_q_open, .release = single_release, @@ -186,7 +204,7 @@ static int nfp_net_debugfs_tx_q_open(struct inode *inode, struct file *f) void nfp_net_debugfs_adapter_add(struct nfp_net *nn) { - struct dentry *queues, *tx, *rx; + struct dentry *queues, *tx, *rx, *xdp; char int_name[16]; int i; @@ -204,13 +222,16 @@ void nfp_net_debugfs_adapter_add(struct nfp_net *nn) rx = debugfs_create_dir("rx", queues); tx = debugfs_create_dir("tx", queues); - if (IS_ERR_OR_NULL(rx) || IS_ERR_OR_NULL(tx)) + xdp = debugfs_create_dir("xdp", queues); + if (IS_ERR_OR_NULL(rx) || IS_ERR_OR_NULL(tx) || IS_ERR_OR_NULL(xdp)) return; for (i = 0; i < min(nn->max_rx_rings, nn->max_r_vecs); i++) { sprintf(int_name, "%d", i); debugfs_create_file(int_name, S_IRUSR, rx, &nn->r_vecs[i], &nfp_rx_q_fops); + debugfs_create_file(int_name, S_IRUSR, xdp, + &nn->r_vecs[i], &nfp_xdp_q_fops); } for (i = 0; i < min(nn->max_tx_rings, nn->max_r_vecs); i++) { diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c b/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c index b87f1b73f200..1b26e9646574 100644 --- a/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c +++ b/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c @@ -176,7 +176,8 @@ static int nfp_net_set_ring_size(struct nfp_net *nn, u32 rxd_cnt, u32 txd_cnt) if (nn->txd_cnt != txd_cnt) reconfig_tx = &tx; - return nfp_net_ring_reconfig(nn, reconfig_rx, reconfig_tx); + return nfp_net_ring_reconfig(nn, &nn->xdp_prog, + reconfig_rx, reconfig_tx); } static int nfp_net_set_ringparam(struct net_device *netdev, @@ -639,14 +640,19 @@ static void nfp_net_get_channels(struct net_device *netdev, struct ethtool_channels *channel) { struct nfp_net *nn = netdev_priv(netdev); + unsigned int num_tx_rings; + + num_tx_rings = nn->num_tx_rings; + if (nn->xdp_prog) + num_tx_rings -= nn->num_rx_rings; channel->max_rx = min(nn->max_rx_rings, nn->max_r_vecs); channel->max_tx = min(nn->max_tx_rings, nn->max_r_vecs); channel->max_combined = min(channel->max_rx, channel->max_tx); channel->max_other = NFP_NET_NON_Q_VECTORS; - channel->combined_count = min(nn->num_rx_rings, nn->num_tx_rings); + channel->combined_count = min(nn->num_rx_rings, num_tx_rings); channel->rx_count = nn->num_rx_rings - channel->combined_count; - channel->tx_count = nn->num_tx_rings - channel->combined_count; + channel->tx_count = num_tx_rings - channel->combined_count; channel->other_count = NFP_NET_NON_Q_VECTORS; } @@ -666,10 +672,16 @@ static int nfp_net_set_num_rings(struct nfp_net *nn, unsigned int total_rx, if (nn->num_rx_rings != total_rx) reconfig_rx = ℞ - if (nn->num_tx_rings != total_tx) + if (nn->num_stack_tx_rings != total_tx || + (nn->xdp_prog && reconfig_rx)) reconfig_tx = &tx; - return nfp_net_ring_reconfig(nn, reconfig_rx, reconfig_tx); + /* nfp_net_check_config() will catch tx.n_rings > nn->max_tx_rings */ + if (nn->xdp_prog) + tx.n_rings += total_rx; + + return nfp_net_ring_reconfig(nn, &nn->xdp_prog, + reconfig_rx, reconfig_tx); } static int nfp_net_set_channels(struct net_device *netdev, -- 1.9.1