From fb9afd961c7628d9971fb9ca4096b6164bfe2cba Mon Sep 17 00:00:00 2001 From: Camelia Groza Date: Wed, 25 Nov 2020 18:53:30 +0200 Subject: [PATCH 1/7] dpaa_eth: add struct for software backpointers We maintain an skb backpointer in the software annotations area of Tx frames. Introduce a structure for explicit handling. Acked-by: Madalin Bucur Signed-off-by: Camelia Groza Reviewed-by: Maciej Fijalkowski Signed-off-by: Jakub Kicinski --- drivers/net/ethernet/freescale/dpaa/dpaa_eth.c | 16 +++++++++------- drivers/net/ethernet/freescale/dpaa/dpaa_eth.h | 8 ++++++++ 2 files changed, 17 insertions(+), 7 deletions(-) diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c index 8867693e0fb3..88533a2aea1d 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c @@ -1633,6 +1633,7 @@ static struct sk_buff *dpaa_cleanup_tx_fd(const struct dpaa_priv *priv, dma_addr_t addr = qm_fd_addr(fd); void *vaddr = phys_to_virt(addr); const struct qm_sg_entry *sgt; + struct dpaa_eth_swbp *swbp; struct sk_buff *skb; u64 ns; int i; @@ -1665,7 +1666,8 @@ static struct sk_buff *dpaa_cleanup_tx_fd(const struct dpaa_priv *priv, dma_dir); } - skb = *(struct sk_buff **)vaddr; + swbp = (struct dpaa_eth_swbp *)vaddr; + skb = swbp->skb; /* DMA unmapping is required before accessing the HW provided info */ if (ts && priv->tx_tstamp && @@ -1879,8 +1881,8 @@ static int skb_to_contig_fd(struct dpaa_priv *priv, { struct net_device *net_dev = priv->net_dev; enum dma_data_direction dma_dir; + struct dpaa_eth_swbp *swbp; unsigned char *buff_start; - struct sk_buff **skbh; dma_addr_t addr; int err; @@ -1891,8 +1893,8 @@ static int skb_to_contig_fd(struct dpaa_priv *priv, buff_start = skb->data - priv->tx_headroom; dma_dir = DMA_TO_DEVICE; - skbh = (struct sk_buff **)buff_start; - *skbh = skb; + swbp = (struct dpaa_eth_swbp *)buff_start; + swbp->skb = skb; /* Enable L3/L4 hardware checksum computation. * @@ -1931,8 +1933,8 @@ static int skb_to_sg_fd(struct dpaa_priv *priv, const enum dma_data_direction dma_dir = DMA_TO_DEVICE; const int nr_frags = skb_shinfo(skb)->nr_frags; struct net_device *net_dev = priv->net_dev; + struct dpaa_eth_swbp *swbp; struct qm_sg_entry *sgt; - struct sk_buff **skbh; void *buff_start; skb_frag_t *frag; dma_addr_t addr; @@ -2005,8 +2007,8 @@ static int skb_to_sg_fd(struct dpaa_priv *priv, qm_fd_set_sg(fd, priv->tx_headroom, skb->len); /* DMA map the SGT page */ - skbh = (struct sk_buff **)buff_start; - *skbh = skb; + swbp = (struct dpaa_eth_swbp *)buff_start; + swbp->skb = skb; addr = dma_map_page(priv->tx_dma_dev, p, 0, priv->tx_headroom + DPAA_SGT_SIZE, dma_dir); diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h index fc2cc4c48e06..da30e5dc55b3 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h @@ -144,6 +144,14 @@ struct dpaa_buffer_layout { u16 priv_data_size; }; +/* Information to be used on the Tx confirmation path. Stored just + * before the start of the transmit buffer. Maximum size allowed + * is DPAA_TX_PRIV_DATA_SIZE bytes. + */ +struct dpaa_eth_swbp { + struct sk_buff *skb; +}; + struct dpaa_priv { struct dpaa_percpu_priv __percpu *percpu_priv; struct dpaa_bp *dpaa_bp; From 86c0c196cbe48f844721783d9162e46bc35c0c5a Mon Sep 17 00:00:00 2001 From: Camelia Groza Date: Wed, 25 Nov 2020 18:53:31 +0200 Subject: [PATCH 2/7] dpaa_eth: add basic XDP support Implement the XDP_DROP and XDP_PASS actions. Avoid draining and reconfiguring the buffer pool at each XDP setup/teardown by increasing the frame headroom and reserving XDP_PACKET_HEADROOM bytes from the start. Since we always reserve an entire page per buffer, this change only impacts Jumbo frame scenarios where the maximum linear frame size is reduced by 256 bytes. Multi buffer Scatter/Gather frames are now used instead in these scenarios. Allow XDP programs to access the entire buffer. The data in the received frame's headroom can be overwritten by the XDP program. Extract the relevant fields from the headroom while they are still available, before running the XDP program. Since the headroom might be resized before the frame is passed up to the stack, remove the check for a fixed headroom value when building an skb. Allow the meta data to be updated and pass the information up the stack. Scatter/Gather frames are dropped when XDP is enabled. Acked-by: Madalin Bucur Signed-off-by: Camelia Groza Reviewed-by: Maciej Fijalkowski Signed-off-by: Jakub Kicinski --- .../net/ethernet/freescale/dpaa/dpaa_eth.c | 167 ++++++++++++++++-- .../net/ethernet/freescale/dpaa/dpaa_eth.h | 2 + 2 files changed, 153 insertions(+), 16 deletions(-) diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c index 88533a2aea1d..0e9bd805ec45 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c @@ -53,6 +53,8 @@ #include #include #include +#include +#include #include #include #include "fman.h" @@ -177,7 +179,7 @@ MODULE_PARM_DESC(tx_timeout, "The Tx timeout in ms"); #define DPAA_HWA_SIZE (DPAA_PARSE_RESULTS_SIZE + DPAA_TIME_STAMP_SIZE \ + DPAA_HASH_RESULTS_SIZE) #define DPAA_RX_PRIV_DATA_DEFAULT_SIZE (DPAA_TX_PRIV_DATA_SIZE + \ - dpaa_rx_extra_headroom) + XDP_PACKET_HEADROOM - DPAA_HWA_SIZE) #ifdef CONFIG_DPAA_ERRATUM_A050385 #define DPAA_RX_PRIV_DATA_A050385_SIZE (DPAA_A050385_ALIGN - DPAA_HWA_SIZE) #define DPAA_RX_PRIV_DATA_SIZE (fman_has_errata_a050385() ? \ @@ -1733,7 +1735,6 @@ static struct sk_buff *contig_fd_to_skb(const struct dpaa_priv *priv, SKB_DATA_ALIGN(sizeof(struct skb_shared_info))); if (WARN_ONCE(!skb, "Build skb failure on Rx\n")) goto free_buffer; - WARN_ON(fd_off != priv->rx_headroom); skb_reserve(skb, fd_off); skb_put(skb, qm_fd_get_length(fd)); @@ -2349,12 +2350,62 @@ static enum qman_cb_dqrr_result rx_error_dqrr(struct qman_portal *portal, return qman_cb_dqrr_consume; } +static u32 dpaa_run_xdp(struct dpaa_priv *priv, struct qm_fd *fd, void *vaddr, + unsigned int *xdp_meta_len) +{ + ssize_t fd_off = qm_fd_get_offset(fd); + struct bpf_prog *xdp_prog; + struct xdp_buff xdp; + u32 xdp_act; + + rcu_read_lock(); + + xdp_prog = READ_ONCE(priv->xdp_prog); + if (!xdp_prog) { + rcu_read_unlock(); + return XDP_PASS; + } + + xdp.data = vaddr + fd_off; + xdp.data_meta = xdp.data; + xdp.data_hard_start = xdp.data - XDP_PACKET_HEADROOM; + xdp.data_end = xdp.data + qm_fd_get_length(fd); + xdp.frame_sz = DPAA_BP_RAW_SIZE; + + xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp); + + /* Update the length and the offset of the FD */ + qm_fd_set_contig(fd, xdp.data - vaddr, xdp.data_end - xdp.data); + + switch (xdp_act) { + case XDP_PASS: + *xdp_meta_len = xdp.data - xdp.data_meta; + break; + default: + bpf_warn_invalid_xdp_action(xdp_act); + fallthrough; + case XDP_ABORTED: + trace_xdp_exception(priv->net_dev, xdp_prog, xdp_act); + fallthrough; + case XDP_DROP: + /* Free the buffer */ + free_pages((unsigned long)vaddr, 0); + break; + } + + rcu_read_unlock(); + + return xdp_act; +} + static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal, struct qman_fq *fq, const struct qm_dqrr_entry *dq, bool sched_napi) { + bool ts_valid = false, hash_valid = false; struct skb_shared_hwtstamps *shhwtstamps; + unsigned int skb_len, xdp_meta_len = 0; struct rtnl_link_stats64 *percpu_stats; struct dpaa_percpu_priv *percpu_priv; const struct qm_fd *fd = &dq->fd; @@ -2362,12 +2413,14 @@ static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal, enum qm_fd_format fd_format; struct net_device *net_dev; u32 fd_status, hash_offset; + struct qm_sg_entry *sgt; struct dpaa_bp *dpaa_bp; struct dpaa_priv *priv; - unsigned int skb_len; struct sk_buff *skb; int *count_ptr; + u32 xdp_act; void *vaddr; + u32 hash; u64 ns; fd_status = be32_to_cpu(fd->status); @@ -2423,35 +2476,67 @@ static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal, count_ptr = this_cpu_ptr(dpaa_bp->percpu_count); (*count_ptr)--; - if (likely(fd_format == qm_fd_contig)) + /* Extract the timestamp stored in the headroom before running XDP */ + if (priv->rx_tstamp) { + if (!fman_port_get_tstamp(priv->mac_dev->port[RX], vaddr, &ns)) + ts_valid = true; + else + WARN_ONCE(1, "fman_port_get_tstamp failed!\n"); + } + + /* Extract the hash stored in the headroom before running XDP */ + if (net_dev->features & NETIF_F_RXHASH && priv->keygen_in_use && + !fman_port_get_hash_result_offset(priv->mac_dev->port[RX], + &hash_offset)) { + hash = be32_to_cpu(*(u32 *)(vaddr + hash_offset)); + hash_valid = true; + } + + if (likely(fd_format == qm_fd_contig)) { + xdp_act = dpaa_run_xdp(priv, (struct qm_fd *)fd, vaddr, + &xdp_meta_len); + if (xdp_act != XDP_PASS) { + percpu_stats->rx_packets++; + percpu_stats->rx_bytes += qm_fd_get_length(fd); + return qman_cb_dqrr_consume; + } skb = contig_fd_to_skb(priv, fd); - else + } else { + /* XDP doesn't support S/G frames. Return the fragments to the + * buffer pool and release the SGT. + */ + if (READ_ONCE(priv->xdp_prog)) { + WARN_ONCE(1, "S/G frames not supported under XDP\n"); + sgt = vaddr + qm_fd_get_offset(fd); + dpaa_release_sgt_members(sgt); + free_pages((unsigned long)vaddr, 0); + return qman_cb_dqrr_consume; + } skb = sg_fd_to_skb(priv, fd); + } if (!skb) return qman_cb_dqrr_consume; - if (priv->rx_tstamp) { + if (xdp_meta_len) + skb_metadata_set(skb, xdp_meta_len); + + /* Set the previously extracted timestamp */ + if (ts_valid) { shhwtstamps = skb_hwtstamps(skb); memset(shhwtstamps, 0, sizeof(*shhwtstamps)); - - if (!fman_port_get_tstamp(priv->mac_dev->port[RX], vaddr, &ns)) - shhwtstamps->hwtstamp = ns_to_ktime(ns); - else - dev_warn(net_dev->dev.parent, "fman_port_get_tstamp failed!\n"); + shhwtstamps->hwtstamp = ns_to_ktime(ns); } skb->protocol = eth_type_trans(skb, net_dev); - if (net_dev->features & NETIF_F_RXHASH && priv->keygen_in_use && - !fman_port_get_hash_result_offset(priv->mac_dev->port[RX], - &hash_offset)) { + /* Set the previously extracted hash */ + if (hash_valid) { enum pkt_hash_types type; /* if L4 exists, it was used in the hash generation */ type = be32_to_cpu(fd->status) & FM_FD_STAT_L4CV ? PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3; - skb_set_hash(skb, be32_to_cpu(*(u32 *)(vaddr + hash_offset)), - type); + skb_set_hash(skb, hash, type); } skb_len = skb->len; @@ -2671,6 +2756,55 @@ static int dpaa_eth_stop(struct net_device *net_dev) return err; } +static int dpaa_setup_xdp(struct net_device *net_dev, struct netdev_bpf *bpf) +{ + struct dpaa_priv *priv = netdev_priv(net_dev); + struct bpf_prog *old_prog; + int err, max_contig_data; + bool up; + + max_contig_data = priv->dpaa_bp->size - priv->rx_headroom; + + /* S/G fragments are not supported in XDP-mode */ + if (bpf->prog && + (net_dev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN > max_contig_data)) { + NL_SET_ERR_MSG_MOD(bpf->extack, "MTU too large for XDP"); + dev_warn(net_dev->dev.parent, + "The maximum MTU for XDP is %d\n", + max_contig_data - VLAN_ETH_HLEN - ETH_FCS_LEN); + return -EINVAL; + } + + up = netif_running(net_dev); + + if (up) + dpaa_eth_stop(net_dev); + + old_prog = xchg(&priv->xdp_prog, bpf->prog); + if (old_prog) + bpf_prog_put(old_prog); + + if (up) { + err = dpaa_open(net_dev); + if (err) { + NL_SET_ERR_MSG_MOD(bpf->extack, "dpaa_open() failed"); + return err; + } + } + + return 0; +} + +static int dpaa_xdp(struct net_device *net_dev, struct netdev_bpf *xdp) +{ + switch (xdp->command) { + case XDP_SETUP_PROG: + return dpaa_setup_xdp(net_dev, xdp); + default: + return -EINVAL; + } +} + static int dpaa_ts_ioctl(struct net_device *dev, struct ifreq *rq, int cmd) { struct dpaa_priv *priv = netdev_priv(dev); @@ -2737,6 +2871,7 @@ static const struct net_device_ops dpaa_ops = { .ndo_set_rx_mode = dpaa_set_rx_mode, .ndo_do_ioctl = dpaa_ioctl, .ndo_setup_tc = dpaa_setup_tc, + .ndo_bpf = dpaa_xdp, }; static int dpaa_napi_add(struct net_device *net_dev) diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h index da30e5dc55b3..94e8613d100a 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h @@ -196,6 +196,8 @@ struct dpaa_priv { bool tx_tstamp; /* Tx timestamping enabled */ bool rx_tstamp; /* Rx timestamping enabled */ + + struct bpf_prog *xdp_prog; }; /* from dpaa_ethtool.c */ From 828eadbaccc1b8090514d4b7ccc92b4876a0bd12 Mon Sep 17 00:00:00 2001 From: Camelia Groza Date: Wed, 25 Nov 2020 18:53:32 +0200 Subject: [PATCH 3/7] dpaa_eth: limit the possible MTU range when XDP is enabled Implement the ndo_change_mtu callback to prevent users from setting an MTU that would permit processing of S/G frames. The maximum MTU size is dependent on the buffer size. Acked-by: Madalin Bucur Signed-off-by: Camelia Groza Reviewed-by: Maciej Fijalkowski Signed-off-by: Jakub Kicinski --- .../net/ethernet/freescale/dpaa/dpaa_eth.c | 39 +++++++++++++++---- 1 file changed, 31 insertions(+), 8 deletions(-) diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c index 0e9bd805ec45..ac3f04c3698a 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c @@ -2756,22 +2756,44 @@ static int dpaa_eth_stop(struct net_device *net_dev) return err; } +static bool xdp_validate_mtu(struct dpaa_priv *priv, int mtu) +{ + int max_contig_data = priv->dpaa_bp->size - priv->rx_headroom; + + /* We do not support S/G fragments when XDP is enabled. + * Limit the MTU in relation to the buffer size. + */ + if (mtu + VLAN_ETH_HLEN + ETH_FCS_LEN > max_contig_data) { + dev_warn(priv->net_dev->dev.parent, + "The maximum MTU for XDP is %d\n", + max_contig_data - VLAN_ETH_HLEN - ETH_FCS_LEN); + return false; + } + + return true; +} + +static int dpaa_change_mtu(struct net_device *net_dev, int new_mtu) +{ + struct dpaa_priv *priv = netdev_priv(net_dev); + + if (priv->xdp_prog && !xdp_validate_mtu(priv, new_mtu)) + return -EINVAL; + + net_dev->mtu = new_mtu; + return 0; +} + static int dpaa_setup_xdp(struct net_device *net_dev, struct netdev_bpf *bpf) { struct dpaa_priv *priv = netdev_priv(net_dev); struct bpf_prog *old_prog; - int err, max_contig_data; + int err; bool up; - max_contig_data = priv->dpaa_bp->size - priv->rx_headroom; - /* S/G fragments are not supported in XDP-mode */ - if (bpf->prog && - (net_dev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN > max_contig_data)) { + if (bpf->prog && !xdp_validate_mtu(priv, net_dev->mtu)) { NL_SET_ERR_MSG_MOD(bpf->extack, "MTU too large for XDP"); - dev_warn(net_dev->dev.parent, - "The maximum MTU for XDP is %d\n", - max_contig_data - VLAN_ETH_HLEN - ETH_FCS_LEN); return -EINVAL; } @@ -2871,6 +2893,7 @@ static const struct net_device_ops dpaa_ops = { .ndo_set_rx_mode = dpaa_set_rx_mode, .ndo_do_ioctl = dpaa_ioctl, .ndo_setup_tc = dpaa_setup_tc, + .ndo_change_mtu = dpaa_change_mtu, .ndo_bpf = dpaa_xdp, }; From d57e57d0cd04be5ee533a0e1e6d13ba9251114b7 Mon Sep 17 00:00:00 2001 From: Camelia Groza Date: Wed, 25 Nov 2020 18:53:33 +0200 Subject: [PATCH 4/7] dpaa_eth: add XDP_TX support Use an xdp_frame structure for managing the frame. Store a backpointer to the structure at the start of the buffer before enqueueing for cleanup on TX confirmation. Reserve DPAA_TX_PRIV_DATA_SIZE bytes from the frame size shared with the XDP program for this purpose. Use the XDP API for freeing the buffer when it returns to the driver on the TX confirmation path. The frame queues are shared with the netstack. The DPAA driver is a LLTX driver so no explicit locking is required on transmission. This approach will be reused for XDP REDIRECT. Acked-by: Madalin Bucur Signed-off-by: Camelia Groza Reviewed-by: Maciej Fijalkowski Signed-off-by: Jakub Kicinski --- .../net/ethernet/freescale/dpaa/dpaa_eth.c | 129 +++++++++++++++++- .../net/ethernet/freescale/dpaa/dpaa_eth.h | 2 + 2 files changed, 126 insertions(+), 5 deletions(-) diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c index ac3f04c3698a..c2ed28d22da0 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c @@ -1130,6 +1130,25 @@ static int dpaa_fq_init(struct dpaa_fq *dpaa_fq, bool td_enable) dpaa_fq->fqid = qman_fq_fqid(fq); + if (dpaa_fq->fq_type == FQ_TYPE_RX_DEFAULT || + dpaa_fq->fq_type == FQ_TYPE_RX_PCD) { + err = xdp_rxq_info_reg(&dpaa_fq->xdp_rxq, dpaa_fq->net_dev, + dpaa_fq->fqid); + if (err) { + dev_err(dev, "xdp_rxq_info_reg() = %d\n", err); + return err; + } + + err = xdp_rxq_info_reg_mem_model(&dpaa_fq->xdp_rxq, + MEM_TYPE_PAGE_ORDER0, NULL); + if (err) { + dev_err(dev, "xdp_rxq_info_reg_mem_model() = %d\n", + err); + xdp_rxq_info_unreg(&dpaa_fq->xdp_rxq); + return err; + } + } + return 0; } @@ -1159,6 +1178,11 @@ static int dpaa_fq_free_entry(struct device *dev, struct qman_fq *fq) } } + if ((dpaa_fq->fq_type == FQ_TYPE_RX_DEFAULT || + dpaa_fq->fq_type == FQ_TYPE_RX_PCD) && + xdp_rxq_info_is_reg(&dpaa_fq->xdp_rxq)) + xdp_rxq_info_unreg(&dpaa_fq->xdp_rxq); + qman_destroy_fq(fq); list_del(&dpaa_fq->list); @@ -1625,6 +1649,9 @@ static int dpaa_eth_refill_bpools(struct dpaa_priv *priv) * * Return the skb backpointer, since for S/G frames the buffer containing it * gets freed here. + * + * No skb backpointer is set when transmitting XDP frames. Cleanup the buffer + * and return NULL in this case. */ static struct sk_buff *dpaa_cleanup_tx_fd(const struct dpaa_priv *priv, const struct qm_fd *fd, bool ts) @@ -1664,13 +1691,21 @@ static struct sk_buff *dpaa_cleanup_tx_fd(const struct dpaa_priv *priv, } } else { dma_unmap_single(priv->tx_dma_dev, addr, - priv->tx_headroom + qm_fd_get_length(fd), + qm_fd_get_offset(fd) + qm_fd_get_length(fd), dma_dir); } swbp = (struct dpaa_eth_swbp *)vaddr; skb = swbp->skb; + /* No skb backpointer is set when running XDP. An xdp_frame + * backpointer is saved instead. + */ + if (!skb) { + xdp_return_frame(swbp->xdpf); + return NULL; + } + /* DMA unmapping is required before accessing the HW provided info */ if (ts && priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) { @@ -2350,11 +2385,76 @@ static enum qman_cb_dqrr_result rx_error_dqrr(struct qman_portal *portal, return qman_cb_dqrr_consume; } +static int dpaa_xdp_xmit_frame(struct net_device *net_dev, + struct xdp_frame *xdpf) +{ + struct dpaa_priv *priv = netdev_priv(net_dev); + struct rtnl_link_stats64 *percpu_stats; + struct dpaa_percpu_priv *percpu_priv; + struct dpaa_eth_swbp *swbp; + struct netdev_queue *txq; + void *buff_start; + struct qm_fd fd; + dma_addr_t addr; + int err; + + percpu_priv = this_cpu_ptr(priv->percpu_priv); + percpu_stats = &percpu_priv->stats; + + if (xdpf->headroom < DPAA_TX_PRIV_DATA_SIZE) { + err = -EINVAL; + goto out_error; + } + + buff_start = xdpf->data - xdpf->headroom; + + /* Leave empty the skb backpointer at the start of the buffer. + * Save the XDP frame for easy cleanup on confirmation. + */ + swbp = (struct dpaa_eth_swbp *)buff_start; + swbp->skb = NULL; + swbp->xdpf = xdpf; + + qm_fd_clear_fd(&fd); + fd.bpid = FSL_DPAA_BPID_INV; + fd.cmd |= cpu_to_be32(FM_FD_CMD_FCO); + qm_fd_set_contig(&fd, xdpf->headroom, xdpf->len); + + addr = dma_map_single(priv->tx_dma_dev, buff_start, + xdpf->headroom + xdpf->len, + DMA_TO_DEVICE); + if (unlikely(dma_mapping_error(priv->tx_dma_dev, addr))) { + err = -EINVAL; + goto out_error; + } + + qm_fd_addr_set64(&fd, addr); + + /* Bump the trans_start */ + txq = netdev_get_tx_queue(net_dev, smp_processor_id()); + txq->trans_start = jiffies; + + err = dpaa_xmit(priv, percpu_stats, smp_processor_id(), &fd); + if (err) { + dma_unmap_single(priv->tx_dma_dev, addr, + qm_fd_get_offset(&fd) + qm_fd_get_length(&fd), + DMA_TO_DEVICE); + goto out_error; + } + + return 0; + +out_error: + percpu_stats->tx_errors++; + return err; +} + static u32 dpaa_run_xdp(struct dpaa_priv *priv, struct qm_fd *fd, void *vaddr, - unsigned int *xdp_meta_len) + struct dpaa_fq *dpaa_fq, unsigned int *xdp_meta_len) { ssize_t fd_off = qm_fd_get_offset(fd); struct bpf_prog *xdp_prog; + struct xdp_frame *xdpf; struct xdp_buff xdp; u32 xdp_act; @@ -2370,7 +2470,8 @@ static u32 dpaa_run_xdp(struct dpaa_priv *priv, struct qm_fd *fd, void *vaddr, xdp.data_meta = xdp.data; xdp.data_hard_start = xdp.data - XDP_PACKET_HEADROOM; xdp.data_end = xdp.data + qm_fd_get_length(fd); - xdp.frame_sz = DPAA_BP_RAW_SIZE; + xdp.frame_sz = DPAA_BP_RAW_SIZE - DPAA_TX_PRIV_DATA_SIZE; + xdp.rxq = &dpaa_fq->xdp_rxq; xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp); @@ -2380,6 +2481,22 @@ static u32 dpaa_run_xdp(struct dpaa_priv *priv, struct qm_fd *fd, void *vaddr, switch (xdp_act) { case XDP_PASS: *xdp_meta_len = xdp.data - xdp.data_meta; + break; + case XDP_TX: + /* We can access the full headroom when sending the frame + * back out + */ + xdp.data_hard_start = vaddr; + xdp.frame_sz = DPAA_BP_RAW_SIZE; + xdpf = xdp_convert_buff_to_frame(&xdp); + if (unlikely(!xdpf)) { + free_pages((unsigned long)vaddr, 0); + break; + } + + if (dpaa_xdp_xmit_frame(priv->net_dev, xdpf)) + xdp_return_frame_rx_napi(xdpf); + break; default: bpf_warn_invalid_xdp_action(xdp_act); @@ -2415,6 +2532,7 @@ static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal, u32 fd_status, hash_offset; struct qm_sg_entry *sgt; struct dpaa_bp *dpaa_bp; + struct dpaa_fq *dpaa_fq; struct dpaa_priv *priv; struct sk_buff *skb; int *count_ptr; @@ -2423,9 +2541,10 @@ static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal, u32 hash; u64 ns; + dpaa_fq = container_of(fq, struct dpaa_fq, fq_base); fd_status = be32_to_cpu(fd->status); fd_format = qm_fd_get_format(fd); - net_dev = ((struct dpaa_fq *)fq)->net_dev; + net_dev = dpaa_fq->net_dev; priv = netdev_priv(net_dev); dpaa_bp = dpaa_bpid2pool(dq->fd.bpid); if (!dpaa_bp) @@ -2494,7 +2613,7 @@ static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal, if (likely(fd_format == qm_fd_contig)) { xdp_act = dpaa_run_xdp(priv, (struct qm_fd *)fd, vaddr, - &xdp_meta_len); + dpaa_fq, &xdp_meta_len); if (xdp_act != XDP_PASS) { percpu_stats->rx_packets++; percpu_stats->rx_bytes += qm_fd_get_length(fd); diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h index 94e8613d100a..5c8d52aa86eb 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h @@ -68,6 +68,7 @@ struct dpaa_fq { u16 channel; u8 wq; enum dpaa_fq_type fq_type; + struct xdp_rxq_info xdp_rxq; }; struct dpaa_fq_cbs { @@ -150,6 +151,7 @@ struct dpaa_buffer_layout { */ struct dpaa_eth_swbp { struct sk_buff *skb; + struct xdp_frame *xdpf; }; struct dpaa_priv { From a1e031ffb422bb89df9ad9c018420d0deff7f2e3 Mon Sep 17 00:00:00 2001 From: Camelia Groza Date: Wed, 25 Nov 2020 18:53:34 +0200 Subject: [PATCH 5/7] dpaa_eth: add XDP_REDIRECT support After transmission, the frame is returned on confirmation queues for cleanup. For this, store a backpointer to the xdp_frame in the private reserved area at the start of the TX buffer. No TX batching support is implemented at this time. Acked-by: Madalin Bucur Signed-off-by: Camelia Groza Reviewed-by: Maciej Fijalkowski Signed-off-by: Jakub Kicinski --- .../net/ethernet/freescale/dpaa/dpaa_eth.c | 48 ++++++++++++++++++- .../net/ethernet/freescale/dpaa/dpaa_eth.h | 1 + 2 files changed, 48 insertions(+), 1 deletion(-) diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c index c2ed28d22da0..dd08c4617476 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c @@ -2305,8 +2305,11 @@ static int dpaa_eth_poll(struct napi_struct *napi, int budget) { struct dpaa_napi_portal *np = container_of(napi, struct dpaa_napi_portal, napi); + int cleaned; - int cleaned = qman_p_poll_dqrr(np->p, budget); + np->xdp_act = 0; + + cleaned = qman_p_poll_dqrr(np->p, budget); if (cleaned < budget) { napi_complete_done(napi, cleaned); @@ -2315,6 +2318,9 @@ static int dpaa_eth_poll(struct napi_struct *napi, int budget) qman_p_irqsource_add(np->p, QM_PIRQ_DQRI); } + if (np->xdp_act & XDP_REDIRECT) + xdp_do_flush(); + return cleaned; } @@ -2457,6 +2463,7 @@ static u32 dpaa_run_xdp(struct dpaa_priv *priv, struct qm_fd *fd, void *vaddr, struct xdp_frame *xdpf; struct xdp_buff xdp; u32 xdp_act; + int err; rcu_read_lock(); @@ -2497,6 +2504,17 @@ static u32 dpaa_run_xdp(struct dpaa_priv *priv, struct qm_fd *fd, void *vaddr, if (dpaa_xdp_xmit_frame(priv->net_dev, xdpf)) xdp_return_frame_rx_napi(xdpf); + break; + case XDP_REDIRECT: + /* Allow redirect to use the full headroom */ + xdp.data_hard_start = vaddr; + xdp.frame_sz = DPAA_BP_RAW_SIZE; + + err = xdp_do_redirect(priv->net_dev, &xdp, xdp_prog); + if (err) { + trace_xdp_exception(priv->net_dev, xdp_prog, xdp_act); + free_pages((unsigned long)vaddr, 0); + } break; default: bpf_warn_invalid_xdp_action(xdp_act); @@ -2527,6 +2545,7 @@ static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal, struct dpaa_percpu_priv *percpu_priv; const struct qm_fd *fd = &dq->fd; dma_addr_t addr = qm_fd_addr(fd); + struct dpaa_napi_portal *np; enum qm_fd_format fd_format; struct net_device *net_dev; u32 fd_status, hash_offset; @@ -2541,6 +2560,7 @@ static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal, u32 hash; u64 ns; + np = container_of(&portal, struct dpaa_napi_portal, p); dpaa_fq = container_of(fq, struct dpaa_fq, fq_base); fd_status = be32_to_cpu(fd->status); fd_format = qm_fd_get_format(fd); @@ -2614,6 +2634,7 @@ static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal, if (likely(fd_format == qm_fd_contig)) { xdp_act = dpaa_run_xdp(priv, (struct qm_fd *)fd, vaddr, dpaa_fq, &xdp_meta_len); + np->xdp_act |= xdp_act; if (xdp_act != XDP_PASS) { percpu_stats->rx_packets++; percpu_stats->rx_bytes += qm_fd_get_length(fd); @@ -2946,6 +2967,30 @@ static int dpaa_xdp(struct net_device *net_dev, struct netdev_bpf *xdp) } } +static int dpaa_xdp_xmit(struct net_device *net_dev, int n, + struct xdp_frame **frames, u32 flags) +{ + struct xdp_frame *xdpf; + int i, err, drops = 0; + + if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) + return -EINVAL; + + if (!netif_running(net_dev)) + return -ENETDOWN; + + for (i = 0; i < n; i++) { + xdpf = frames[i]; + err = dpaa_xdp_xmit_frame(net_dev, xdpf); + if (err) { + xdp_return_frame_rx_napi(xdpf); + drops++; + } + } + + return n - drops; +} + static int dpaa_ts_ioctl(struct net_device *dev, struct ifreq *rq, int cmd) { struct dpaa_priv *priv = netdev_priv(dev); @@ -3014,6 +3059,7 @@ static const struct net_device_ops dpaa_ops = { .ndo_setup_tc = dpaa_setup_tc, .ndo_change_mtu = dpaa_change_mtu, .ndo_bpf = dpaa_xdp, + .ndo_xdp_xmit = dpaa_xdp_xmit, }; static int dpaa_napi_add(struct net_device *net_dev) diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h index 5c8d52aa86eb..daf894a97050 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h @@ -127,6 +127,7 @@ struct dpaa_napi_portal { struct napi_struct napi; struct qman_portal *p; bool down; + int xdp_act; }; struct dpaa_percpu_priv { From d7af04486d9f58752bf840901908108b1ba08aba Mon Sep 17 00:00:00 2001 From: Camelia Groza Date: Wed, 25 Nov 2020 18:53:35 +0200 Subject: [PATCH 6/7] dpaa_eth: rename current skb A050385 erratum workaround Explicitly point that the current workaround addresses skbs. This change is in preparation for adding a workaround for XDP scenarios. Acked-by: Madalin Bucur Signed-off-by: Camelia Groza Reviewed-by: Maciej Fijalkowski Signed-off-by: Jakub Kicinski --- drivers/net/ethernet/freescale/dpaa/dpaa_eth.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c index dd08c4617476..c63f5fe0674e 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c @@ -2105,7 +2105,7 @@ static inline int dpaa_xmit(struct dpaa_priv *priv, } #ifdef CONFIG_DPAA_ERRATUM_A050385 -static int dpaa_a050385_wa(struct net_device *net_dev, struct sk_buff **s) +static int dpaa_a050385_wa_skb(struct net_device *net_dev, struct sk_buff **s) { struct dpaa_priv *priv = netdev_priv(net_dev); struct sk_buff *new_skb, *skb = *s; @@ -2221,7 +2221,7 @@ dpaa_start_xmit(struct sk_buff *skb, struct net_device *net_dev) #ifdef CONFIG_DPAA_ERRATUM_A050385 if (unlikely(fman_has_errata_a050385())) { - if (dpaa_a050385_wa(net_dev, &skb)) + if (dpaa_a050385_wa_skb(net_dev, &skb)) goto enomem; nonlinear = skb_is_nonlinear(skb); } From ae680bcbd06ae9189e0102f8cdb5fd938b00725f Mon Sep 17 00:00:00 2001 From: Camelia Groza Date: Wed, 25 Nov 2020 18:53:36 +0200 Subject: [PATCH 7/7] dpaa_eth: implement the A050385 erratum workaround for XDP For XDP TX, even tough we start out with correctly aligned buffers, the XDP program might change the data's alignment. For REDIRECT, we have no control over the alignment either. Create a new workaround for xdp_frame structures to verify the erratum conditions and move the data to a fresh buffer if necessary. Create a new xdp_frame for managing the new buffer and free the old one using the XDP API. Due to alignment constraints, all frames have a 256 byte headroom that is offered fully to XDP under the erratum. If the XDP program uses all of it, the data needs to be move to make room for the xdpf backpointer. Disable the metadata support since the information can be lost. Acked-by: Madalin Bucur Signed-off-by: Camelia Groza Reviewed-by: Maciej Fijalkowski Signed-off-by: Jakub Kicinski --- .../net/ethernet/freescale/dpaa/dpaa_eth.c | 84 ++++++++++++++++++- 1 file changed, 82 insertions(+), 2 deletions(-) diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c index c63f5fe0674e..947b3d2f9c7e 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c @@ -2171,6 +2171,52 @@ workaround: return 0; } + +static int dpaa_a050385_wa_xdpf(struct dpaa_priv *priv, + struct xdp_frame **init_xdpf) +{ + struct xdp_frame *new_xdpf, *xdpf = *init_xdpf; + void *new_buff; + struct page *p; + + /* Check the data alignment and make sure the headroom is large + * enough to store the xdpf backpointer. Use an aligned headroom + * value. + * + * Due to alignment constraints, we give XDP access to the full 256 + * byte frame headroom. If the XDP program uses all of it, copy the + * data to a new buffer and make room for storing the backpointer. + */ + if (PTR_IS_ALIGNED(xdpf->data, DPAA_A050385_ALIGN) && + xdpf->headroom >= priv->tx_headroom) { + xdpf->headroom = priv->tx_headroom; + return 0; + } + + p = dev_alloc_pages(0); + if (unlikely(!p)) + return -ENOMEM; + + /* Copy the data to the new buffer at a properly aligned offset */ + new_buff = page_address(p); + memcpy(new_buff + priv->tx_headroom, xdpf->data, xdpf->len); + + /* Create an XDP frame around the new buffer in a similar fashion + * to xdp_convert_buff_to_frame. + */ + new_xdpf = new_buff; + new_xdpf->data = new_buff + priv->tx_headroom; + new_xdpf->len = xdpf->len; + new_xdpf->headroom = priv->tx_headroom; + new_xdpf->frame_sz = DPAA_BP_RAW_SIZE; + new_xdpf->mem.type = MEM_TYPE_PAGE_ORDER0; + + /* Release the initial buffer */ + xdp_return_frame_rx_napi(xdpf); + + *init_xdpf = new_xdpf; + return 0; +} #endif static netdev_tx_t @@ -2407,6 +2453,15 @@ static int dpaa_xdp_xmit_frame(struct net_device *net_dev, percpu_priv = this_cpu_ptr(priv->percpu_priv); percpu_stats = &percpu_priv->stats; +#ifdef CONFIG_DPAA_ERRATUM_A050385 + if (unlikely(fman_has_errata_a050385())) { + if (dpaa_a050385_wa_xdpf(priv, &xdpf)) { + err = -ENOMEM; + goto out_error; + } + } +#endif + if (xdpf->headroom < DPAA_TX_PRIV_DATA_SIZE) { err = -EINVAL; goto out_error; @@ -2480,6 +2535,20 @@ static u32 dpaa_run_xdp(struct dpaa_priv *priv, struct qm_fd *fd, void *vaddr, xdp.frame_sz = DPAA_BP_RAW_SIZE - DPAA_TX_PRIV_DATA_SIZE; xdp.rxq = &dpaa_fq->xdp_rxq; + /* We reserve a fixed headroom of 256 bytes under the erratum and we + * offer it all to XDP programs to use. If no room is left for the + * xdpf backpointer on TX, we will need to copy the data. + * Disable metadata support since data realignments might be required + * and the information can be lost. + */ +#ifdef CONFIG_DPAA_ERRATUM_A050385 + if (unlikely(fman_has_errata_a050385())) { + xdp_set_data_meta_invalid(&xdp); + xdp.data_hard_start = vaddr; + xdp.frame_sz = DPAA_BP_RAW_SIZE; + } +#endif + xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp); /* Update the length and the offset of the FD */ @@ -2487,7 +2556,12 @@ static u32 dpaa_run_xdp(struct dpaa_priv *priv, struct qm_fd *fd, void *vaddr, switch (xdp_act) { case XDP_PASS: +#ifdef CONFIG_DPAA_ERRATUM_A050385 + *xdp_meta_len = xdp_data_meta_unsupported(&xdp) ? 0 : + xdp.data - xdp.data_meta; +#else *xdp_meta_len = xdp.data - xdp.data_meta; +#endif break; case XDP_TX: /* We can access the full headroom when sending the frame @@ -3191,10 +3265,16 @@ static u16 dpaa_get_headroom(struct dpaa_buffer_layout *bl, */ headroom = (u16)(bl[port].priv_data_size + DPAA_HWA_SIZE); - if (port == RX) + if (port == RX) { +#ifdef CONFIG_DPAA_ERRATUM_A050385 + if (unlikely(fman_has_errata_a050385())) + headroom = XDP_PACKET_HEADROOM; +#endif + return ALIGN(headroom, DPAA_FD_RX_DATA_ALIGNMENT); - else + } else { return ALIGN(headroom, DPAA_FD_DATA_ALIGNMENT); + } } static int dpaa_eth_probe(struct platform_device *pdev)