// SPDX-License-Identifier: (GPL-2.0+ OR BSD-3-Clause)
/* Copyright 2014-2016 Freescale Semiconductor Inc.
 * Copyright 2016-2020 NXP
 */
#include <linux/init.h>
#include <linux/module.h>
#include <linux/platform_device.h>
#include <linux/etherdevice.h>
#include <linux/of_net.h>
#include <linux/interrupt.h>
#include <linux/msi.h>
#include <linux/kthread.h>
#include <linux/iommu.h>
#include <linux/fsl/mc.h>
#include <linux/bpf.h>
#include <linux/bpf_trace.h>
#include <linux/fsl/ptp_qoriq.h>
#include <linux/ptp_classify.h>
#include <net/pkt_cls.h>
#include <net/sock.h>

#include "dpaa2-eth.h"

/* CREATE_TRACE_POINTS only needs to be defined once. Other dpa files
 * using trace events only need to #include <trace/events/sched.h>
 */
#define CREATE_TRACE_POINTS
#include "dpaa2-eth-trace.h"

MODULE_LICENSE("Dual BSD/GPL");
MODULE_AUTHOR("Freescale Semiconductor, Inc");
MODULE_DESCRIPTION("Freescale DPAA2 Ethernet Driver");

struct ptp_qoriq *dpaa2_ptp;
EXPORT_SYMBOL(dpaa2_ptp);

static void *dpaa2_iova_to_virt(struct iommu_domain *domain,
                                dma_addr_t iova_addr)
{
        phys_addr_t phys_addr;

        phys_addr = domain ? iommu_iova_to_phys(domain, iova_addr) : iova_addr;

        return phys_to_virt(phys_addr);
}

static void dpaa2_eth_validate_rx_csum(struct dpaa2_eth_priv *priv,
                                       u32 fd_status,
                                       struct sk_buff *skb)
{
        skb_checksum_none_assert(skb);

        /* HW checksum validation is disabled, nothing to do here */
        if (!(priv->net_dev->features & NETIF_F_RXCSUM))
                return;

        /* Read checksum validation bits */
        if (!((fd_status & DPAA2_FAS_L3CV) &&
              (fd_status & DPAA2_FAS_L4CV)))
                return;

        /* Inform the stack there's no need to compute L3/L4 csum anymore */
        skb->ip_summed = CHECKSUM_UNNECESSARY;
}

/* Free a received FD.
 * Not to be used for Tx conf FDs or on any other paths.
 */
static void dpaa2_eth_free_rx_fd(struct dpaa2_eth_priv *priv,
                                 const struct dpaa2_fd *fd,
                                 void *vaddr)
{
        struct device *dev = priv->net_dev->dev.parent;
        dma_addr_t addr = dpaa2_fd_get_addr(fd);
        u8 fd_format = dpaa2_fd_get_format(fd);
        struct dpaa2_sg_entry *sgt;
        void *sg_vaddr;
        int i;

        /* If single buffer frame, just free the data buffer */
        if (fd_format == dpaa2_fd_single)
                goto free_buf;
        else if (fd_format != dpaa2_fd_sg)
                /* We don't support any other format */
                return;

        /* For S/G frames, we first need to free all SG entries
         * except the first one, which was taken care of already
         */
        sgt = vaddr + dpaa2_fd_get_offset(fd);
        for (i = 1; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
                addr = dpaa2_sg_get_addr(&sgt[i]);
                sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
                dma_unmap_page(dev, addr, priv->rx_buf_size,
                               DMA_BIDIRECTIONAL);

                free_pages((unsigned long)sg_vaddr, 0);
                if (dpaa2_sg_is_final(&sgt[i]))
                        break;
        }

free_buf:
        free_pages((unsigned long)vaddr, 0);
}

/* Build a linear skb based on a single-buffer frame descriptor */
static struct sk_buff *dpaa2_eth_build_linear_skb(struct dpaa2_eth_channel *ch,
                                                  const struct dpaa2_fd *fd,
                                                  void *fd_vaddr)
{
        struct sk_buff *skb = NULL;
        u16 fd_offset = dpaa2_fd_get_offset(fd);
        u32 fd_length = dpaa2_fd_get_len(fd);

        ch->buf_count--;

        skb = build_skb(fd_vaddr, DPAA2_ETH_RX_BUF_RAW_SIZE);
        if (unlikely(!skb))
                return NULL;

        skb_reserve(skb, fd_offset);
        skb_put(skb, fd_length);

        return skb;
}

/* Build a non linear (fragmented) skb based on a S/G table */
static struct sk_buff *dpaa2_eth_build_frag_skb(struct dpaa2_eth_priv *priv,
                                                struct dpaa2_eth_channel *ch,
                                                struct dpaa2_sg_entry *sgt)
{
        struct sk_buff *skb = NULL;
        struct device *dev = priv->net_dev->dev.parent;
        void *sg_vaddr;
        dma_addr_t sg_addr;
        u16 sg_offset;
        u32 sg_length;
        struct page *page, *head_page;
        int page_offset;
        int i;

        for (i = 0; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
                struct dpaa2_sg_entry *sge = &sgt[i];

                /* NOTE: We only support SG entries in dpaa2_sg_single format,
                 * but this is the only format we may receive from HW anyway
                 */

                /* Get the address and length from the S/G entry */
                sg_addr = dpaa2_sg_get_addr(sge);
                sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, sg_addr);
                dma_unmap_page(dev, sg_addr, priv->rx_buf_size,
                               DMA_BIDIRECTIONAL);

                sg_length = dpaa2_sg_get_len(sge);

                if (i == 0) {
                        /* We build the skb around the first data buffer */
                        skb = build_skb(sg_vaddr, DPAA2_ETH_RX_BUF_RAW_SIZE);
                        if (unlikely(!skb)) {
                                /* Free the first SG entry now, since we already
                                 * unmapped it and obtained the virtual address
                                 */
                                free_pages((unsigned long)sg_vaddr, 0);

                                /* We still need to subtract the buffers used
                                 * by this FD from our software counter
                                 */
                                while (!dpaa2_sg_is_final(&sgt[i]) &&
                                       i < DPAA2_ETH_MAX_SG_ENTRIES)
                                        i++;
                                break;
                        }

                        sg_offset = dpaa2_sg_get_offset(sge);
                        skb_reserve(skb, sg_offset);
                        skb_put(skb, sg_length);
                } else {
                        /* Rest of the data buffers are stored as skb frags */
                        page = virt_to_page(sg_vaddr);
                        head_page = virt_to_head_page(sg_vaddr);

                        /* Offset in page (which may be compound).
                         * Data in subsequent SG entries is stored from the
                         * beginning of the buffer, so we don't need to add the
                         * sg_offset.
                         */
                        page_offset = ((unsigned long)sg_vaddr &
                                (PAGE_SIZE - 1)) +
                                (page_address(page) - page_address(head_page));

                        skb_add_rx_frag(skb, i - 1, head_page, page_offset,
                                        sg_length, priv->rx_buf_size);
                }

                if (dpaa2_sg_is_final(sge))
                        break;
        }

        WARN_ONCE(i == DPAA2_ETH_MAX_SG_ENTRIES, "Final bit not set in SGT");

        /* Count all data buffers + SG table buffer */
        ch->buf_count -= i + 2;

        return skb;
}

/* Free buffers acquired from the buffer pool or which were meant to
 * be released in the pool
 */
static void dpaa2_eth_free_bufs(struct dpaa2_eth_priv *priv, u64 *buf_array,
                                int count)
{
        struct device *dev = priv->net_dev->dev.parent;
        void *vaddr;
        int i;

        for (i = 0; i < count; i++) {
                vaddr = dpaa2_iova_to_virt(priv->iommu_domain, buf_array[i]);
                dma_unmap_page(dev, buf_array[i], priv->rx_buf_size,
                               DMA_BIDIRECTIONAL);
                free_pages((unsigned long)vaddr, 0);
        }
}

static void dpaa2_eth_recycle_buf(struct dpaa2_eth_priv *priv,
                                  struct dpaa2_eth_channel *ch,
                                  dma_addr_t addr)
{
        int retries = 0;
        int err;

        ch->recycled_bufs[ch->recycled_bufs_cnt++] = addr;
        if (ch->recycled_bufs_cnt < DPAA2_ETH_BUFS_PER_CMD)
                return;

        while ((err = dpaa2_io_service_release(ch->dpio, priv->bpid,
                                               ch->recycled_bufs,
                                               ch->recycled_bufs_cnt)) == -EBUSY) {
                if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
                        break;
                cpu_relax();
        }

        if (err) {
                dpaa2_eth_free_bufs(priv, ch->recycled_bufs, ch->recycled_bufs_cnt);
                ch->buf_count -= ch->recycled_bufs_cnt;
        }

        ch->recycled_bufs_cnt = 0;
}

static int dpaa2_eth_xdp_flush(struct dpaa2_eth_priv *priv,
                               struct dpaa2_eth_fq *fq,
                               struct dpaa2_eth_xdp_fds *xdp_fds)
{
        int total_enqueued = 0, retries = 0, enqueued;
        struct dpaa2_eth_drv_stats *percpu_extras;
        int num_fds, err, max_retries;
        struct dpaa2_fd *fds;

        percpu_extras = this_cpu_ptr(priv->percpu_extras);

        /* try to enqueue all the FDs until the max number of retries is hit */
        fds = xdp_fds->fds;
        num_fds = xdp_fds->num;
        max_retries = num_fds * DPAA2_ETH_ENQUEUE_RETRIES;
        while (total_enqueued < num_fds && retries < max_retries) {
                err = priv->enqueue(priv, fq, &fds[total_enqueued],
                                    0, num_fds - total_enqueued, &enqueued);
                if (err == -EBUSY) {
                        percpu_extras->tx_portal_busy += ++retries;
                        continue;
                }
                total_enqueued += enqueued;
        }
        xdp_fds->num = 0;

        return total_enqueued;
}

static void dpaa2_eth_xdp_tx_flush(struct dpaa2_eth_priv *priv,
                                   struct dpaa2_eth_channel *ch,
                                   struct dpaa2_eth_fq *fq)
{
        struct rtnl_link_stats64 *percpu_stats;
        struct dpaa2_fd *fds;
        int enqueued, i;

        percpu_stats = this_cpu_ptr(priv->percpu_stats);

        // enqueue the array of XDP_TX frames
        enqueued = dpaa2_eth_xdp_flush(priv, fq, &fq->xdp_tx_fds);

        /* update statistics */
        percpu_stats->tx_packets += enqueued;
        fds = fq->xdp_tx_fds.fds;
        for (i = 0; i < enqueued; i++) {
                percpu_stats->tx_bytes += dpaa2_fd_get_len(&fds[i]);
                ch->stats.xdp_tx++;
        }
        for (i = enqueued; i < fq->xdp_tx_fds.num; i++) {
                dpaa2_eth_recycle_buf(priv, ch, dpaa2_fd_get_addr(&fds[i]));
                percpu_stats->tx_errors++;
                ch->stats.xdp_tx_err++;
        }
        fq->xdp_tx_fds.num = 0;
}

static void dpaa2_eth_xdp_enqueue(struct dpaa2_eth_priv *priv,
                                  struct dpaa2_eth_channel *ch,
                                  struct dpaa2_fd *fd,
                                  void *buf_start, u16 queue_id)
{
        struct dpaa2_faead *faead;
        struct dpaa2_fd *dest_fd;
        struct dpaa2_eth_fq *fq;
        u32 ctrl, frc;

        /* Mark the egress frame hardware annotation area as valid */
        frc = dpaa2_fd_get_frc(fd);
        dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);
        dpaa2_fd_set_ctrl(fd, DPAA2_FD_CTRL_ASAL);

        /* Instruct hardware to release the FD buffer directly into
         * the buffer pool once transmission is completed, instead of
         * sending a Tx confirmation frame to us
         */
        ctrl = DPAA2_FAEAD_A4V | DPAA2_FAEAD_A2V | DPAA2_FAEAD_EBDDV;
        faead = dpaa2_get_faead(buf_start, false);
        faead->ctrl = cpu_to_le32(ctrl);
        faead->conf_fqid = 0;

        fq = &priv->fq[queue_id];
        dest_fd = &fq->xdp_tx_fds.fds[fq->xdp_tx_fds.num++];
        memcpy(dest_fd, fd, sizeof(*dest_fd));

        if (fq->xdp_tx_fds.num < DEV_MAP_BULK_SIZE)
                return;

        dpaa2_eth_xdp_tx_flush(priv, ch, fq);
}

static u32 dpaa2_eth_run_xdp(struct dpaa2_eth_priv *priv,
                             struct dpaa2_eth_channel *ch,
                             struct dpaa2_eth_fq *rx_fq,
                             struct dpaa2_fd *fd, void *vaddr)
{
        dma_addr_t addr = dpaa2_fd_get_addr(fd);
        struct bpf_prog *xdp_prog;
        struct xdp_buff xdp;
        u32 xdp_act = XDP_PASS;
        int err, offset;

        rcu_read_lock();

        xdp_prog = READ_ONCE(ch->xdp.prog);
        if (!xdp_prog)
                goto out;

        offset = dpaa2_fd_get_offset(fd) - XDP_PACKET_HEADROOM;
        xdp_init_buff(&xdp, DPAA2_ETH_RX_BUF_RAW_SIZE - offset, &ch->xdp_rxq);
        xdp_prepare_buff(&xdp, vaddr + offset, XDP_PACKET_HEADROOM,
                         dpaa2_fd_get_len(fd), false);

        xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp);

        /* xdp.data pointer may have changed */
        dpaa2_fd_set_offset(fd, xdp.data - vaddr);
        dpaa2_fd_set_len(fd, xdp.data_end - xdp.data);

        switch (xdp_act) {
        case XDP_PASS:
                break;
        case XDP_TX:
                dpaa2_eth_xdp_enqueue(priv, ch, fd, vaddr, rx_fq->flowid);
                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:
                dpaa2_eth_recycle_buf(priv, ch, addr);
                ch->stats.xdp_drop++;
                break;
        case XDP_REDIRECT:
                dma_unmap_page(priv->net_dev->dev.parent, addr,
                               priv->rx_buf_size, DMA_BIDIRECTIONAL);
                ch->buf_count--;

                /* Allow redirect use of full headroom */
                xdp.data_hard_start = vaddr;
                xdp.frame_sz = DPAA2_ETH_RX_BUF_RAW_SIZE;

                err = xdp_do_redirect(priv->net_dev, &xdp, xdp_prog);
                if (unlikely(err)) {
                        addr = dma_map_page(priv->net_dev->dev.parent,
                                            virt_to_page(vaddr), 0,
                                            priv->rx_buf_size, DMA_BIDIRECTIONAL);
                        if (unlikely(dma_mapping_error(priv->net_dev->dev.parent, addr))) {
                                free_pages((unsigned long)vaddr, 0);
                        } else {
                                ch->buf_count++;
                                dpaa2_eth_recycle_buf(priv, ch, addr);
                        }
                        ch->stats.xdp_drop++;
                } else {
                        ch->stats.xdp_redirect++;
                }
                break;
        }

        ch->xdp.res |= xdp_act;
out:
        rcu_read_unlock();
        return xdp_act;
}

static struct sk_buff *dpaa2_eth_copybreak(struct dpaa2_eth_channel *ch,
                                           const struct dpaa2_fd *fd,
                                           void *fd_vaddr)
{
        u16 fd_offset = dpaa2_fd_get_offset(fd);
        struct dpaa2_eth_priv *priv = ch->priv;
        u32 fd_length = dpaa2_fd_get_len(fd);
        struct sk_buff *skb = NULL;
        unsigned int skb_len;

        if (fd_length > priv->rx_copybreak)
                return NULL;

        skb_len = fd_length + dpaa2_eth_needed_headroom(NULL);

        skb = napi_alloc_skb(&ch->napi, skb_len);
        if (!skb)
                return NULL;

        skb_reserve(skb, dpaa2_eth_needed_headroom(NULL));
        skb_put(skb, fd_length);

        memcpy(skb->data, fd_vaddr + fd_offset, fd_length);

        dpaa2_eth_recycle_buf(priv, ch, dpaa2_fd_get_addr(fd));

        return skb;
}

/* Main Rx frame processing routine */
static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
                         struct dpaa2_eth_channel *ch,
                         const struct dpaa2_fd *fd,
                         struct dpaa2_eth_fq *fq)
{
        dma_addr_t addr = dpaa2_fd_get_addr(fd);
        u8 fd_format = dpaa2_fd_get_format(fd);
        void *vaddr;
        struct sk_buff *skb;
        struct rtnl_link_stats64 *percpu_stats;
        struct dpaa2_eth_drv_stats *percpu_extras;
        struct device *dev = priv->net_dev->dev.parent;
        struct dpaa2_fas *fas;
        void *buf_data;
        u32 status = 0;
        u32 xdp_act;

        /* Tracing point */
        trace_dpaa2_rx_fd(priv->net_dev, fd);

        vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
        dma_sync_single_for_cpu(dev, addr, priv->rx_buf_size,
                                DMA_BIDIRECTIONAL);

        fas = dpaa2_get_fas(vaddr, false);
        prefetch(fas);
        buf_data = vaddr + dpaa2_fd_get_offset(fd);
        prefetch(buf_data);

        percpu_stats = this_cpu_ptr(priv->percpu_stats);
        percpu_extras = this_cpu_ptr(priv->percpu_extras);

        if (fd_format == dpaa2_fd_single) {
                xdp_act = dpaa2_eth_run_xdp(priv, ch, fq, (struct dpaa2_fd *)fd, vaddr);
                if (xdp_act != XDP_PASS) {
                        percpu_stats->rx_packets++;
                        percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
                        return;
                }

                skb = dpaa2_eth_copybreak(ch, fd, vaddr);
                if (!skb) {
                        dma_unmap_page(dev, addr, priv->rx_buf_size,
                                       DMA_BIDIRECTIONAL);
                        skb = dpaa2_eth_build_linear_skb(ch, fd, vaddr);
                }
        } else if (fd_format == dpaa2_fd_sg) {
                WARN_ON(priv->xdp_prog);

                dma_unmap_page(dev, addr, priv->rx_buf_size,
                               DMA_BIDIRECTIONAL);
                skb = dpaa2_eth_build_frag_skb(priv, ch, buf_data);
                free_pages((unsigned long)vaddr, 0);
                percpu_extras->rx_sg_frames++;
                percpu_extras->rx_sg_bytes += dpaa2_fd_get_len(fd);
        } else {
                /* We don't support any other format */
                goto err_frame_format;
        }

        if (unlikely(!skb))
                goto err_build_skb;

        prefetch(skb->data);

        /* Get the timestamp value */
        if (priv->rx_tstamp) {
                struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
                __le64 *ts = dpaa2_get_ts(vaddr, false);
                u64 ns;

                memset(shhwtstamps, 0, sizeof(*shhwtstamps));

                ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
                shhwtstamps->hwtstamp = ns_to_ktime(ns);
        }

        /* Check if we need to validate the L4 csum */
        if (likely(dpaa2_fd_get_frc(fd) & DPAA2_FD_FRC_FASV)) {
                status = le32_to_cpu(fas->status);
                dpaa2_eth_validate_rx_csum(priv, status, skb);
        }

        skb->protocol = eth_type_trans(skb, priv->net_dev);
        skb_record_rx_queue(skb, fq->flowid);

        percpu_stats->rx_packets++;
        percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);

        list_add_tail(&skb->list, ch->rx_list);

        return;

err_build_skb:
        dpaa2_eth_free_rx_fd(priv, fd, vaddr);
err_frame_format:
        percpu_stats->rx_dropped++;
}

/* Processing of Rx frames received on the error FQ
 * We check and print the error bits and then free the frame
 */
static void dpaa2_eth_rx_err(struct dpaa2_eth_priv *priv,
                             struct dpaa2_eth_channel *ch,
                             const struct dpaa2_fd *fd,
                             struct dpaa2_eth_fq *fq __always_unused)
{
        struct device *dev = priv->net_dev->dev.parent;
        dma_addr_t addr = dpaa2_fd_get_addr(fd);
        u8 fd_format = dpaa2_fd_get_format(fd);
        struct rtnl_link_stats64 *percpu_stats;
        struct dpaa2_eth_trap_item *trap_item;
        struct dpaa2_fapr *fapr;
        struct sk_buff *skb;
        void *buf_data;
        void *vaddr;

        vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
        dma_sync_single_for_cpu(dev, addr, priv->rx_buf_size,
                                DMA_BIDIRECTIONAL);

        buf_data = vaddr + dpaa2_fd_get_offset(fd);

        if (fd_format == dpaa2_fd_single) {
                dma_unmap_page(dev, addr, priv->rx_buf_size,
                               DMA_BIDIRECTIONAL);
                skb = dpaa2_eth_build_linear_skb(ch, fd, vaddr);
        } else if (fd_format == dpaa2_fd_sg) {
                dma_unmap_page(dev, addr, priv->rx_buf_size,
                               DMA_BIDIRECTIONAL);
                skb = dpaa2_eth_build_frag_skb(priv, ch, buf_data);
                free_pages((unsigned long)vaddr, 0);
        } else {
                /* We don't support any other format */
                dpaa2_eth_free_rx_fd(priv, fd, vaddr);
                goto err_frame_format;
        }

        fapr = dpaa2_get_fapr(vaddr, false);
        trap_item = dpaa2_eth_dl_get_trap(priv, fapr);
        if (trap_item)
                devlink_trap_report(priv->devlink, skb, trap_item->trap_ctx,
                                    &priv->devlink_port, NULL);
        consume_skb(skb);

err_frame_format:
        percpu_stats = this_cpu_ptr(priv->percpu_stats);
        percpu_stats->rx_errors++;
        ch->buf_count--;
}

/* Consume all frames pull-dequeued into the store. This is the simplest way to
 * make sure we don't accidentally issue another volatile dequeue which would
 * overwrite (leak) frames already in the store.
 *
 * Observance of NAPI budget is not our concern, leaving that to the caller.
 */
static int dpaa2_eth_consume_frames(struct dpaa2_eth_channel *ch,
                                    struct dpaa2_eth_fq **src)
{
        struct dpaa2_eth_priv *priv = ch->priv;
        struct dpaa2_eth_fq *fq = NULL;
        struct dpaa2_dq *dq;
        const struct dpaa2_fd *fd;
        int cleaned = 0, retries = 0;
        int is_last;

        do {
                dq = dpaa2_io_store_next(ch->store, &is_last);
                if (unlikely(!dq)) {
                        /* If we're here, we *must* have placed a
                         * volatile dequeue comnmand, so keep reading through
                         * the store until we get some sort of valid response
                         * token (either a valid frame or an "empty dequeue")
                         */
                        if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES) {
                                netdev_err_once(priv->net_dev,
                                                "Unable to read a valid dequeue response\n");
                                return -ETIMEDOUT;
                        }
                        continue;
                }

                fd = dpaa2_dq_fd(dq);
                fq = (struct dpaa2_eth_fq *)(uintptr_t)dpaa2_dq_fqd_ctx(dq);

                fq->consume(priv, ch, fd, fq);
                cleaned++;
                retries = 0;
        } while (!is_last);

        if (!cleaned)
                return 0;

        fq->stats.frames += cleaned;
        ch->stats.frames += cleaned;

        /* A dequeue operation only pulls frames from a single queue
         * into the store. Return the frame queue as an out param.
         */
        if (src)
                *src = fq;

        return cleaned;
}

static int dpaa2_eth_ptp_parse(struct sk_buff *skb,
                               u8 *msgtype, u8 *twostep, u8 *udp,
                               u16 *correction_offset,
                               u16 *origintimestamp_offset)
{
        unsigned int ptp_class;
        struct ptp_header *hdr;
        unsigned int type;
        u8 *base;

        ptp_class = ptp_classify_raw(skb);
        if (ptp_class == PTP_CLASS_NONE)
                return -EINVAL;

        hdr = ptp_parse_header(skb, ptp_class);
        if (!hdr)
                return -EINVAL;

        *msgtype = ptp_get_msgtype(hdr, ptp_class);
        *twostep = hdr->flag_field[0] & 0x2;

        type = ptp_class & PTP_CLASS_PMASK;
        if (type == PTP_CLASS_IPV4 ||
            type == PTP_CLASS_IPV6)
                *udp = 1;
        else
                *udp = 0;

        base = skb_mac_header(skb);
        *correction_offset = (u8 *)&hdr->correction - base;
        *origintimestamp_offset = (u8 *)hdr + sizeof(struct ptp_header) - base;

        return 0;
}

/* Configure the egress frame annotation for timestamp update */
static void dpaa2_eth_enable_tx_tstamp(struct dpaa2_eth_priv *priv,
                                       struct dpaa2_fd *fd,
                                       void *buf_start,
                                       struct sk_buff *skb)
{
        struct ptp_tstamp origin_timestamp;
        struct dpni_single_step_cfg cfg;
        u8 msgtype, twostep, udp;
        struct dpaa2_faead *faead;
        struct dpaa2_fas *fas;
        struct timespec64 ts;
        u16 offset1, offset2;
        u32 ctrl, frc;
        __le64 *ns;
        u8 *data;

        /* Mark the egress frame annotation area as valid */
        frc = dpaa2_fd_get_frc(fd);
        dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);

        /* Set hardware annotation size */
        ctrl = dpaa2_fd_get_ctrl(fd);
        dpaa2_fd_set_ctrl(fd, ctrl | DPAA2_FD_CTRL_ASAL);

        /* enable UPD (update prepanded data) bit in FAEAD field of
         * hardware frame annotation area
         */
        ctrl = DPAA2_FAEAD_A2V | DPAA2_FAEAD_UPDV | DPAA2_FAEAD_UPD;
        faead = dpaa2_get_faead(buf_start, true);
        faead->ctrl = cpu_to_le32(ctrl);

        if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
                if (dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
                                        &offset1, &offset2) ||
                    msgtype != PTP_MSGTYPE_SYNC || twostep) {
                        WARN_ONCE(1, "Bad packet for one-step timestamping\n");
                        return;
                }

                /* Mark the frame annotation status as valid */
                frc = dpaa2_fd_get_frc(fd);
                dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FASV);

                /* Mark the PTP flag for one step timestamping */
                fas = dpaa2_get_fas(buf_start, true);
                fas->status = cpu_to_le32(DPAA2_FAS_PTP);

                dpaa2_ptp->caps.gettime64(&dpaa2_ptp->caps, &ts);
                ns = dpaa2_get_ts(buf_start, true);
                *ns = cpu_to_le64(timespec64_to_ns(&ts) /
                                  DPAA2_PTP_CLK_PERIOD_NS);

                /* Update current time to PTP message originTimestamp field */
                ns_to_ptp_tstamp(&origin_timestamp, le64_to_cpup(ns));
                data = skb_mac_header(skb);
                *(__be16 *)(data + offset2) = htons(origin_timestamp.sec_msb);
                *(__be32 *)(data + offset2 + 2) =
                        htonl(origin_timestamp.sec_lsb);
                *(__be32 *)(data + offset2 + 6) = htonl(origin_timestamp.nsec);

                cfg.en = 1;
                cfg.ch_update = udp;
                cfg.offset = offset1;
                cfg.peer_delay = 0;

                if (dpni_set_single_step_cfg(priv->mc_io, 0, priv->mc_token,
                                             &cfg))
                        WARN_ONCE(1, "Failed to set single step register");
        }
}

/* Create a frame descriptor based on a fragmented skb */
static int dpaa2_eth_build_sg_fd(struct dpaa2_eth_priv *priv,
                                 struct sk_buff *skb,
                                 struct dpaa2_fd *fd,
                                 void **swa_addr)
{
        struct device *dev = priv->net_dev->dev.parent;
        void *sgt_buf = NULL;
        dma_addr_t addr;
        int nr_frags = skb_shinfo(skb)->nr_frags;
        struct dpaa2_sg_entry *sgt;
        int i, err;
        int sgt_buf_size;
        struct scatterlist *scl, *crt_scl;
        int num_sg;
        int num_dma_bufs;
        struct dpaa2_eth_swa *swa;

        /* Create and map scatterlist.
         * We don't advertise NETIF_F_FRAGLIST, so skb_to_sgvec() will not have
         * to go beyond nr_frags+1.
         * Note: We don't support chained scatterlists
         */
        if (unlikely(PAGE_SIZE / sizeof(struct scatterlist) < nr_frags + 1))
                return -EINVAL;

        scl = kmalloc_array(nr_frags + 1, sizeof(struct scatterlist), GFP_ATOMIC);
        if (unlikely(!scl))
                return -ENOMEM;

        sg_init_table(scl, nr_frags + 1);
        num_sg = skb_to_sgvec(skb, scl, 0, skb->len);
        if (unlikely(num_sg < 0)) {
                err = -ENOMEM;
                goto dma_map_sg_failed;
        }
        num_dma_bufs = dma_map_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
        if (unlikely(!num_dma_bufs)) {
                err = -ENOMEM;
                goto dma_map_sg_failed;
        }

        /* Prepare the HW SGT structure */
        sgt_buf_size = priv->tx_data_offset +
                       sizeof(struct dpaa2_sg_entry) *  num_dma_bufs;
        sgt_buf = napi_alloc_frag_align(sgt_buf_size, DPAA2_ETH_TX_BUF_ALIGN);
        if (unlikely(!sgt_buf)) {
                err = -ENOMEM;
                goto sgt_buf_alloc_failed;
        }
        memset(sgt_buf, 0, sgt_buf_size);

        sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);

        /* Fill in the HW SGT structure.
         *
         * sgt_buf is zeroed out, so the following fields are implicit
         * in all sgt entries:
         *   - offset is 0
         *   - format is 'dpaa2_sg_single'
         */
        for_each_sg(scl, crt_scl, num_dma_bufs, i) {
                dpaa2_sg_set_addr(&sgt[i], sg_dma_address(crt_scl));
                dpaa2_sg_set_len(&sgt[i], sg_dma_len(crt_scl));
        }
        dpaa2_sg_set_final(&sgt[i - 1], true);

        /* Store the skb backpointer in the SGT buffer.
         * Fit the scatterlist and the number of buffers alongside the
         * skb backpointer in the software annotation area. We'll need
         * all of them on Tx Conf.
         */
        *swa_addr = (void *)sgt_buf;
        swa = (struct dpaa2_eth_swa *)sgt_buf;
        swa->type = DPAA2_ETH_SWA_SG;
        swa->sg.skb = skb;
        swa->sg.scl = scl;
        swa->sg.num_sg = num_sg;
        swa->sg.sgt_size = sgt_buf_size;

        /* Separately map the SGT buffer */
        addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
        if (unlikely(dma_mapping_error(dev, addr))) {
                err = -ENOMEM;
                goto dma_map_single_failed;
        }
        dpaa2_fd_set_offset(fd, priv->tx_data_offset);
        dpaa2_fd_set_format(fd, dpaa2_fd_sg);
        dpaa2_fd_set_addr(fd, addr);
        dpaa2_fd_set_len(fd, skb->len);
        dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);

        return 0;

dma_map_single_failed:
        skb_free_frag(sgt_buf);
sgt_buf_alloc_failed:
        dma_unmap_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
dma_map_sg_failed:
        kfree(scl);
        return err;
}

/* Create a SG frame descriptor based on a linear skb.
 *
 * This function is used on the Tx path when the skb headroom is not large
 * enough for the HW requirements, thus instead of realloc-ing the skb we
 * create a SG frame descriptor with only one entry.
 */
static int dpaa2_eth_build_sg_fd_single_buf(struct dpaa2_eth_priv *priv,
                                            struct sk_buff *skb,
                                            struct dpaa2_fd *fd,
                                            void **swa_addr)
{
        struct device *dev = priv->net_dev->dev.parent;
        struct dpaa2_eth_sgt_cache *sgt_cache;
        struct dpaa2_sg_entry *sgt;
        struct dpaa2_eth_swa *swa;
        dma_addr_t addr, sgt_addr;
        void *sgt_buf = NULL;
        int sgt_buf_size;
        int err;

        /* Prepare the HW SGT structure */
        sgt_cache = this_cpu_ptr(priv->sgt_cache);
        sgt_buf_size = priv->tx_data_offset + sizeof(struct dpaa2_sg_entry);

        if (sgt_cache->count == 0)
                sgt_buf = kzalloc(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN,
                                  GFP_ATOMIC);
        else
                sgt_buf = sgt_cache->buf[--sgt_cache->count];
        if (unlikely(!sgt_buf))
                return -ENOMEM;

        sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
        sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);

        addr = dma_map_single(dev, skb->data, skb->len, DMA_BIDIRECTIONAL);
        if (unlikely(dma_mapping_error(dev, addr))) {
                err = -ENOMEM;
                goto data_map_failed;
        }

        /* Fill in the HW SGT structure */
        dpaa2_sg_set_addr(sgt, addr);
        dpaa2_sg_set_len(sgt, skb->len);
        dpaa2_sg_set_final(sgt, true);

        /* Store the skb backpointer in the SGT buffer */
        *swa_addr = (void *)sgt_buf;
        swa = (struct dpaa2_eth_swa *)sgt_buf;
        swa->type = DPAA2_ETH_SWA_SINGLE;
        swa->single.skb = skb;
        swa->single.sgt_size = sgt_buf_size;

        /* Separately map the SGT buffer */
        sgt_addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
        if (unlikely(dma_mapping_error(dev, sgt_addr))) {
                err = -ENOMEM;
                goto sgt_map_failed;
        }

        dpaa2_fd_set_offset(fd, priv->tx_data_offset);
        dpaa2_fd_set_format(fd, dpaa2_fd_sg);
        dpaa2_fd_set_addr(fd, sgt_addr);
        dpaa2_fd_set_len(fd, skb->len);
        dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);

        return 0;

sgt_map_failed:
        dma_unmap_single(dev, addr, skb->len, DMA_BIDIRECTIONAL);
data_map_failed:
        if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
                kfree(sgt_buf);
        else
                sgt_cache->buf[sgt_cache->count++] = sgt_buf;

        return err;
}

/* Create a frame descriptor based on a linear skb */
static int dpaa2_eth_build_single_fd(struct dpaa2_eth_priv *priv,
                                     struct sk_buff *skb,
                                     struct dpaa2_fd *fd,
                                     void **swa_addr)
{
        struct device *dev = priv->net_dev->dev.parent;
        u8 *buffer_start, *aligned_start;
        struct dpaa2_eth_swa *swa;
        dma_addr_t addr;

        buffer_start = skb->data - dpaa2_eth_needed_headroom(skb);

        /* If there's enough room to align the FD address, do it.
         * It will help hardware optimize accesses.
         */
        aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
                                  DPAA2_ETH_TX_BUF_ALIGN);
        if (aligned_start >= skb->head)
                buffer_start = aligned_start;

        /* Store a backpointer to the skb at the beginning of the buffer
         * (in the private data area) such that we can release it
         * on Tx confirm
         */
        *swa_addr = (void *)buffer_start;
        swa = (struct dpaa2_eth_swa *)buffer_start;
        swa->type = DPAA2_ETH_SWA_SINGLE;
        swa->single.skb = skb;

        addr = dma_map_single(dev, buffer_start,
                              skb_tail_pointer(skb) - buffer_start,
                              DMA_BIDIRECTIONAL);
        if (unlikely(dma_mapping_error(dev, addr)))
                return -ENOMEM;

        dpaa2_fd_set_addr(fd, addr);
        dpaa2_fd_set_offset(fd, (u16)(skb->data - buffer_start));
        dpaa2_fd_set_len(fd, skb->len);
        dpaa2_fd_set_format(fd, dpaa2_fd_single);
        dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);

        return 0;
}

/* FD freeing routine on the Tx path
 *
 * DMA-unmap and free FD and possibly SGT buffer allocated on Tx. The skb
 * back-pointed to is also freed.
 * This can be called either from dpaa2_eth_tx_conf() or on the error path of
 * dpaa2_eth_tx().
 */
static void dpaa2_eth_free_tx_fd(struct dpaa2_eth_priv *priv,
                                 struct dpaa2_eth_fq *fq,
                                 const struct dpaa2_fd *fd, bool in_napi)

{
        struct device *dev = priv->net_dev->dev.parent;
        dma_addr_t fd_addr, sg_addr;
        struct sk_buff *skb = NULL;
        unsigned char *buffer_start;
        struct dpaa2_eth_swa *swa;
        u8 fd_format = dpaa2_fd_get_format(fd);
        u32 fd_len = dpaa2_fd_get_len(fd);

        struct dpaa2_eth_sgt_cache *sgt_cache;
        struct dpaa2_sg_entry *sgt;

        fd_addr = dpaa2_fd_get_addr(fd);
        buffer_start = dpaa2_iova_to_virt(priv->iommu_domain, fd_addr);
        swa = (struct dpaa2_eth_swa *)buffer_start;

        if (fd_format == dpaa2_fd_single) {
                if (swa->type == DPAA2_ETH_SWA_SINGLE) {
                        skb = swa->single.skb;
                        /* Accessing the skb buffer is safe before dma unmap,
                         * because we didn't map the actual skb shell.
                         */
                        dma_unmap_single(dev, fd_addr,
                                         skb_tail_pointer(skb) - buffer_start,
                                         DMA_BIDIRECTIONAL);
                } else {
                        WARN_ONCE(swa->type != DPAA2_ETH_SWA_XDP, "Wrong SWA type");
                        dma_unmap_single(dev, fd_addr, swa->xdp.dma_size,
                                         DMA_BIDIRECTIONAL);
                }
        } else if (fd_format == dpaa2_fd_sg) {
                if (swa->type == DPAA2_ETH_SWA_SG) {
                        skb = swa->sg.skb;

                        /* Unmap the scatterlist */
                        dma_unmap_sg(dev, swa->sg.scl, swa->sg.num_sg,
                                     DMA_BIDIRECTIONAL);
                        kfree(swa->sg.scl);

                        /* Unmap the SGT buffer */
                        dma_unmap_single(dev, fd_addr, swa->sg.sgt_size,
                                         DMA_BIDIRECTIONAL);
                } else {
                        skb = swa->single.skb;

                        /* Unmap the SGT Buffer */
                        dma_unmap_single(dev, fd_addr, swa->single.sgt_size,
                                         DMA_BIDIRECTIONAL);

                        sgt = (struct dpaa2_sg_entry *)(buffer_start +
                                                        priv->tx_data_offset);
                        sg_addr = dpaa2_sg_get_addr(sgt);
                        dma_unmap_single(dev, sg_addr, skb->len, DMA_BIDIRECTIONAL);
                }
        } else {
                netdev_dbg(priv->net_dev, "Invalid FD format\n");
                return;
        }

        if (swa->type != DPAA2_ETH_SWA_XDP && in_napi) {
                fq->dq_frames++;
                fq->dq_bytes += fd_len;
        }

        if (swa->type == DPAA2_ETH_SWA_XDP) {
                xdp_return_frame(swa->xdp.xdpf);
                return;
        }

        /* Get the timestamp value */
        if (skb->cb[0] == TX_TSTAMP) {
                struct skb_shared_hwtstamps shhwtstamps;
                __le64 *ts = dpaa2_get_ts(buffer_start, true);
                u64 ns;

                memset(&shhwtstamps, 0, sizeof(shhwtstamps));

                ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
                shhwtstamps.hwtstamp = ns_to_ktime(ns);
                skb_tstamp_tx(skb, &shhwtstamps);
        } else if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
                mutex_unlock(&priv->onestep_tstamp_lock);
        }

        /* Free SGT buffer allocated on tx */
        if (fd_format != dpaa2_fd_single) {
                sgt_cache = this_cpu_ptr(priv->sgt_cache);
                if (swa->type == DPAA2_ETH_SWA_SG) {
                        skb_free_frag(buffer_start);
                } else {
                        if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
                                kfree(buffer_start);
                        else
                                sgt_cache->buf[sgt_cache->count++] = buffer_start;
                }
        }

        /* Move on with skb release */
        napi_consume_skb(skb, in_napi);
}

static netdev_tx_t __dpaa2_eth_tx(struct sk_buff *skb,
                                  struct net_device *net_dev)
{
        struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
        struct dpaa2_fd fd;
        struct rtnl_link_stats64 *percpu_stats;
        struct dpaa2_eth_drv_stats *percpu_extras;
        struct dpaa2_eth_fq *fq;
        struct netdev_queue *nq;
        u16 queue_mapping;
        unsigned int needed_headroom;
        u32 fd_len;
        u8 prio = 0;
        int err, i;
        void *swa;

        percpu_stats = this_cpu_ptr(priv->percpu_stats);
        percpu_extras = this_cpu_ptr(priv->percpu_extras);

        needed_headroom = dpaa2_eth_needed_headroom(skb);

        /* We'll be holding a back-reference to the skb until Tx Confirmation;
         * we don't want that overwritten by a concurrent Tx with a cloned skb.
         */
        skb = skb_unshare(skb, GFP_ATOMIC);
        if (unlikely(!skb)) {
                /* skb_unshare() has already freed the skb */
                percpu_stats->tx_dropped++;
                return NETDEV_TX_OK;
        }

        /* Setup the FD fields */
        memset(&fd, 0, sizeof(fd));

        if (skb_is_nonlinear(skb)) {
                err = dpaa2_eth_build_sg_fd(priv, skb, &fd, &swa);
                percpu_extras->tx_sg_frames++;
                percpu_extras->tx_sg_bytes += skb->len;
        } else if (skb_headroom(skb) < needed_headroom) {
                err = dpaa2_eth_build_sg_fd_single_buf(priv, skb, &fd, &swa);
                percpu_extras->tx_sg_frames++;
                percpu_extras->tx_sg_bytes += skb->len;
                percpu_extras->tx_converted_sg_frames++;
                percpu_extras->tx_converted_sg_bytes += skb->len;
        } else {
                err = dpaa2_eth_build_single_fd(priv, skb, &fd, &swa);
        }

        if (unlikely(err)) {
                percpu_stats->tx_dropped++;
                goto err_build_fd;
        }

        if (skb->cb[0])
                dpaa2_eth_enable_tx_tstamp(priv, &fd, swa, skb);

        /* Tracing point */
        trace_dpaa2_tx_fd(net_dev, &fd);

        /* TxConf FQ selection relies on queue id from the stack.
         * In case of a forwarded frame from another DPNI interface, we choose
         * a queue affined to the same core that processed the Rx frame
         */
        queue_mapping = skb_get_queue_mapping(skb);

        if (net_dev->num_tc) {
                prio = netdev_txq_to_tc(net_dev, queue_mapping);
                /* Hardware interprets priority level 0 as being the highest,
                 * so we need to do a reverse mapping to the netdev tc index
                 */
                prio = net_dev->num_tc - prio - 1;
                /* We have only one FQ array entry for all Tx hardware queues
                 * with the same flow id (but different priority levels)
                 */
                queue_mapping %= dpaa2_eth_queue_count(priv);
        }
        fq = &priv->fq[queue_mapping];

        fd_len = dpaa2_fd_get_len(&fd);
        nq = netdev_get_tx_queue(net_dev, queue_mapping);
        netdev_tx_sent_queue(nq, fd_len);

        /* Everything that happens after this enqueues might race with
         * the Tx confirmation callback for this frame
         */
        for (i = 0; i < DPAA2_ETH_ENQUEUE_RETRIES; i++) {
                err = priv->enqueue(priv, fq, &fd, prio, 1, NULL);
                if (err != -EBUSY)
                        break;
        }
        percpu_extras->tx_portal_busy += i;
        if (unlikely(err < 0)) {
                percpu_stats->tx_errors++;
                /* Clean up everything, including freeing the skb */
                dpaa2_eth_free_tx_fd(priv, fq, &fd, false);
                netdev_tx_completed_queue(nq, 1, fd_len);
        } else {
                percpu_stats->tx_packets++;
                percpu_stats->tx_bytes += fd_len;
        }

        return NETDEV_TX_OK;

err_build_fd:
        dev_kfree_skb(skb);

        return NETDEV_TX_OK;
}

static void dpaa2_eth_tx_onestep_tstamp(struct work_struct *work)
{
        struct dpaa2_eth_priv *priv = container_of(work, struct dpaa2_eth_priv,
                                                   tx_onestep_tstamp);
        struct sk_buff *skb;

        while (true) {
                skb = skb_dequeue(&priv->tx_skbs);
                if (!skb)
                        return;

                /* Lock just before TX one-step timestamping packet,
                 * and release the lock in dpaa2_eth_free_tx_fd when
                 * confirm the packet has been sent on hardware, or
                 * when clean up during transmit failure.
                 */
                mutex_lock(&priv->onestep_tstamp_lock);
                __dpaa2_eth_tx(skb, priv->net_dev);
        }
}

static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
{
        struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
        u8 msgtype, twostep, udp;
        u16 offset1, offset2;

        /* Utilize skb->cb[0] for timestamping request per skb */
        skb->cb[0] = 0;

        if ((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) && dpaa2_ptp) {
                if (priv->tx_tstamp_type == HWTSTAMP_TX_ON)
                        skb->cb[0] = TX_TSTAMP;
                else if (priv->tx_tstamp_type == HWTSTAMP_TX_ONESTEP_SYNC)
                        skb->cb[0] = TX_TSTAMP_ONESTEP_SYNC;
        }

        /* TX for one-step timestamping PTP Sync packet */
        if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
                if (!dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
                                         &offset1, &offset2))
                        if (msgtype == PTP_MSGTYPE_SYNC && twostep == 0) {
                                skb_queue_tail(&priv->tx_skbs, skb);
                                queue_work(priv->dpaa2_ptp_wq,
                                           &priv->tx_onestep_tstamp);
                                return NETDEV_TX_OK;
                        }
                /* Use two-step timestamping if not one-step timestamping
                 * PTP Sync packet
                 */
                skb->cb[0] = TX_TSTAMP;
        }

        /* TX for other packets */
        return __dpaa2_eth_tx(skb, net_dev);
}

/* Tx confirmation frame processing routine */
static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
                              struct dpaa2_eth_channel *ch __always_unused,
                              const struct dpaa2_fd *fd,
                              struct dpaa2_eth_fq *fq)
{
        struct rtnl_link_stats64 *percpu_stats;
        struct dpaa2_eth_drv_stats *percpu_extras;
        u32 fd_len = dpaa2_fd_get_len(fd);
        u32 fd_errors;

        /* Tracing point */
        trace_dpaa2_tx_conf_fd(priv->net_dev, fd);

        percpu_extras = this_cpu_ptr(priv->percpu_extras);
        percpu_extras->tx_conf_frames++;
        percpu_extras->tx_conf_bytes += fd_len;

        /* Check frame errors in the FD field */
        fd_errors = dpaa2_fd_get_ctrl(fd) & DPAA2_FD_TX_ERR_MASK;
        dpaa2_eth_free_tx_fd(priv, fq, fd, true);

        if (likely(!fd_errors))
                return;

        if (net_ratelimit())
                netdev_dbg(priv->net_dev, "TX frame FD error: 0x%08x\n",
                           fd_errors);

        percpu_stats = this_cpu_ptr(priv->percpu_stats);
        /* Tx-conf logically pertains to the egress path. */
        percpu_stats->tx_errors++;
}

static int dpaa2_eth_set_rx_vlan_filtering(struct dpaa2_eth_priv *priv,
                                           bool enable)
{
        int err;

        err = dpni_enable_vlan_filter(priv->mc_io, 0, priv->mc_token, enable);

        if (err) {
                netdev_err(priv->net_dev,
                           "dpni_enable_vlan_filter failed\n");
                return err;
        }

        return 0;
}

static int dpaa2_eth_set_rx_csum(struct dpaa2_eth_priv *priv, bool enable)
{
        int err;

        err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
                               DPNI_OFF_RX_L3_CSUM, enable);
        if (err) {
                netdev_err(priv->net_dev,
                           "dpni_set_offload(RX_L3_CSUM) failed\n");
                return err;
        }

        err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
                               DPNI_OFF_RX_L4_CSUM, enable);
        if (err) {
                netdev_err(priv->net_dev,
                           "dpni_set_offload(RX_L4_CSUM) failed\n");
                return err;
        }

        return 0;
}

static int dpaa2_eth_set_tx_csum(struct dpaa2_eth_priv *priv, bool enable)
{
        int err;

        err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
                               DPNI_OFF_TX_L3_CSUM, enable);
        if (err) {
                netdev_err(priv->net_dev, "dpni_set_offload(TX_L3_CSUM) failed\n");
                return err;
        }

        err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
                               DPNI_OFF_TX_L4_CSUM, enable);
        if (err) {
                netdev_err(priv->net_dev, "dpni_set_offload(TX_L4_CSUM) failed\n");
                return err;
        }

        return 0;
}

/* Perform a single release command to add buffers
 * to the specified buffer pool
 */
static int dpaa2_eth_add_bufs(struct dpaa2_eth_priv *priv,
                              struct dpaa2_eth_channel *ch, u16 bpid)
{
        struct device *dev = priv->net_dev->dev.parent;
        u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
        struct page *page;
        dma_addr_t addr;
        int retries = 0;
        int i, err;

        for (i = 0; i < DPAA2_ETH_BUFS_PER_CMD; i++) {
                /* Allocate buffer visible to WRIOP + skb shared info +
                 * alignment padding
                 */
                /* allocate one page for each Rx buffer. WRIOP sees
                 * the entire page except for a tailroom reserved for
                 * skb shared info
                 */
                page = dev_alloc_pages(0);
                if (!page)
                        goto err_alloc;

                addr = dma_map_page(dev, page, 0, priv->rx_buf_size,
                                    DMA_BIDIRECTIONAL);
                if (unlikely(dma_mapping_error(dev, addr)))
                        goto err_map;

                buf_array[i] = addr;

                /* tracing point */
                trace_dpaa2_eth_buf_seed(priv->net_dev,
                                         page, DPAA2_ETH_RX_BUF_RAW_SIZE,
                                         addr, priv->rx_buf_size,
                                         bpid);
        }

release_bufs:
        /* In case the portal is busy, retry until successful */
        while ((err = dpaa2_io_service_release(ch->dpio, bpid,
                                               buf_array, i)) == -EBUSY) {
                if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
                        break;
                cpu_relax();
        }

        /* If release command failed, clean up and bail out;
         * not much else we can do about it
         */
        if (err) {
                dpaa2_eth_free_bufs(priv, buf_array, i);
                return 0;
        }

        return i;

err_map:
        __free_pages(page, 0);
err_alloc:
        /* If we managed to allocate at least some buffers,
         * release them to hardware
         */
        if (i)
                goto release_bufs;

        return 0;
}

static int dpaa2_eth_seed_pool(struct dpaa2_eth_priv *priv, u16 bpid)
{
        int i, j;
        int new_count;

        for (j = 0; j < priv->num_channels; j++) {
                for (i = 0; i < DPAA2_ETH_NUM_BUFS;
                     i += DPAA2_ETH_BUFS_PER_CMD) {
                        new_count = dpaa2_eth_add_bufs(priv, priv->channel[j], bpid);
                        priv->channel[j]->buf_count += new_count;

                        if (new_count < DPAA2_ETH_BUFS_PER_CMD) {
                                return -ENOMEM;
                        }
                }
        }

        return 0;
}

/*
 * Drain the specified number of buffers from the DPNI's private buffer pool.
 * @count must not exceeed DPAA2_ETH_BUFS_PER_CMD
 */
static void dpaa2_eth_drain_bufs(struct dpaa2_eth_priv *priv, int count)
{
        u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
        int retries = 0;
        int ret;

        do {
                ret = dpaa2_io_service_acquire(NULL, priv->bpid,
                                               buf_array, count);
                if (ret < 0) {
                        if (ret == -EBUSY &&
                            retries++ < DPAA2_ETH_SWP_BUSY_RETRIES)
                                continue;
                        netdev_err(priv->net_dev, "dpaa2_io_service_acquire() failed\n");
                        return;
                }
                dpaa2_eth_free_bufs(priv, buf_array, ret);
                retries = 0;
        } while (ret);
}

static void dpaa2_eth_drain_pool(struct dpaa2_eth_priv *priv)
{
        int i;

        dpaa2_eth_drain_bufs(priv, DPAA2_ETH_BUFS_PER_CMD);
        dpaa2_eth_drain_bufs(priv, 1);

        for (i = 0; i < priv->num_channels; i++)
                priv->channel[i]->buf_count = 0;
}

/* Function is called from softirq context only, so we don't need to guard
 * the access to percpu count
 */
static int dpaa2_eth_refill_pool(struct dpaa2_eth_priv *priv,
                                 struct dpaa2_eth_channel *ch,
                                 u16 bpid)
{
        int new_count;

        if (likely(ch->buf_count >= DPAA2_ETH_REFILL_THRESH))
                return 0;

        do {
                new_count = dpaa2_eth_add_bufs(priv, ch, bpid);
                if (unlikely(!new_count)) {
                        /* Out of memory; abort for now, we'll try later on */
                        break;
                }
                ch->buf_count += new_count;
        } while (ch->buf_count < DPAA2_ETH_NUM_BUFS);

        if (unlikely(ch->buf_count < DPAA2_ETH_NUM_BUFS))
                return -ENOMEM;

        return 0;
}

static void dpaa2_eth_sgt_cache_drain(struct dpaa2_eth_priv *priv)
{
        struct dpaa2_eth_sgt_cache *sgt_cache;
        u16 count;
        int k, i;

        for_each_possible_cpu(k) {
                sgt_cache = per_cpu_ptr(priv->sgt_cache, k);
                count = sgt_cache->count;

                for (i = 0; i < count; i++)
                        kfree(sgt_cache->buf[i]);
                sgt_cache->count = 0;
        }
}

static int dpaa2_eth_pull_channel(struct dpaa2_eth_channel *ch)
{
        int err;
        int dequeues = -1;

        /* Retry while portal is busy */
        do {
                err = dpaa2_io_service_pull_channel(ch->dpio, ch->ch_id,
                                                    ch->store);
                dequeues++;
                cpu_relax();
        } while (err == -EBUSY && dequeues < DPAA2_ETH_SWP_BUSY_RETRIES);

        ch->stats.dequeue_portal_busy += dequeues;
        if (unlikely(err))
                ch->stats.pull_err++;

        return err;
}

/* NAPI poll routine
 *
 * Frames are dequeued from the QMan channel associated with this NAPI context.
 * Rx, Tx confirmation and (if configured) Rx error frames all count
 * towards the NAPI budget.
 */
static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
{
        struct dpaa2_eth_channel *ch;
        struct dpaa2_eth_priv *priv;
        int rx_cleaned = 0, txconf_cleaned = 0;
        struct dpaa2_eth_fq *fq, *txc_fq = NULL;
        struct netdev_queue *nq;
        int store_cleaned, work_done;
        struct list_head rx_list;
        int retries = 0;
        u16 flowid;
        int err;

        ch = container_of(napi, struct dpaa2_eth_channel, napi);
        ch->xdp.res = 0;
        priv = ch->priv;

        INIT_LIST_HEAD(&rx_list);
        ch->rx_list = &rx_list;

        do {
                err = dpaa2_eth_pull_channel(ch);
                if (unlikely(err))
                        break;

                /* Refill pool if appropriate */
                dpaa2_eth_refill_pool(priv, ch, priv->bpid);

                store_cleaned = dpaa2_eth_consume_frames(ch, &fq);
                if (store_cleaned <= 0)
                        break;
                if (fq->type == DPAA2_RX_FQ) {
                        rx_cleaned += store_cleaned;
                        flowid = fq->flowid;
                } else {
                        txconf_cleaned += store_cleaned;
                        /* We have a single Tx conf FQ on this channel */
                        txc_fq = fq;
                }

                /* If we either consumed the whole NAPI budget with Rx frames
                 * or we reached the Tx confirmations threshold, we're done.
                 */
                if (rx_cleaned >= budget ||
                    txconf_cleaned >= DPAA2_ETH_TXCONF_PER_NAPI) {
                        work_done = budget;
                        goto out;
                }
        } while (store_cleaned);

        /* We didn't consume the entire budget, so finish napi and
         * re-enable data availability notifications
         */
        napi_complete_done(napi, rx_cleaned);
        do {
                err = dpaa2_io_service_rearm(ch->dpio, &ch->nctx);
                cpu_relax();
        } while (err == -EBUSY && retries++ < DPAA2_ETH_SWP_BUSY_RETRIES);
        WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
                  ch->nctx.desired_cpu);

        work_done = max(rx_cleaned, 1);

out:
        netif_receive_skb_list(ch->rx_list);

        if (txc_fq && txc_fq->dq_frames) {
                nq = netdev_get_tx_queue(priv->net_dev, txc_fq->flowid);
                netdev_tx_completed_queue(nq, txc_fq->dq_frames,
                                          txc_fq->dq_bytes);
                txc_fq->dq_frames = 0;
                txc_fq->dq_bytes = 0;
        }

        if (ch->xdp.res & XDP_REDIRECT)
                xdp_do_flush_map();
        else if (rx_cleaned && ch->xdp.res & XDP_TX)
                dpaa2_eth_xdp_tx_flush(priv, ch, &priv->fq[flowid]);

        return work_done;
}

static void dpaa2_eth_enable_ch_napi(struct dpaa2_eth_priv *priv)
{
        struct dpaa2_eth_channel *ch;
        int i;

        for (i = 0; i < priv->num_channels; i++) {
                ch = priv->channel[i];
                napi_enable(&ch->napi);
        }
}

static void dpaa2_eth_disable_ch_napi(struct dpaa2_eth_priv *priv)
{
        struct dpaa2_eth_channel *ch;
        int i;

        for (i = 0; i < priv->num_channels; i++) {
                ch = priv->channel[i];
                napi_disable(&ch->napi);
        }
}

void dpaa2_eth_set_rx_taildrop(struct dpaa2_eth_priv *priv,
                               bool tx_pause, bool pfc)
{
        struct dpni_taildrop td = {0};
        struct dpaa2_eth_fq *fq;
        int i, err;

        /* FQ taildrop: threshold is in bytes, per frame queue. Enabled if
         * flow control is disabled (as it might interfere with either the
         * buffer pool depletion trigger for pause frames or with the group
         * congestion trigger for PFC frames)
         */
        td.enable = !tx_pause;
        if (priv->rx_fqtd_enabled == td.enable)
                goto set_cgtd;

        td.threshold = DPAA2_ETH_FQ_TAILDROP_THRESH;
        td.units = DPNI_CONGESTION_UNIT_BYTES;

        for (i = 0; i < priv->num_fqs; i++) {
                fq = &priv->fq[i];
                if (fq->type != DPAA2_RX_FQ)
                        continue;
                err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
                                        DPNI_CP_QUEUE, DPNI_QUEUE_RX,
                                        fq->tc, fq->flowid, &td);
                if (err) {
                        netdev_err(priv->net_dev,
                                   "dpni_set_taildrop(FQ) failed\n");
                        return;
                }
        }

        priv->rx_fqtd_enabled = td.enable;

set_cgtd:
        /* Congestion group taildrop: threshold is in frames, per group
         * of FQs belonging to the same traffic class
         * Enabled if general Tx pause disabled or if PFCs are enabled
         * (congestion group threhsold for PFC generation is lower than the
         * CG taildrop threshold, so it won't interfere with it; we also
         * want frames in non-PFC enabled traffic classes to be kept in check)
         */
        td.enable = !tx_pause || pfc;
        if (priv->rx_cgtd_enabled == td.enable)
                return;

        td.threshold = DPAA2_ETH_CG_TAILDROP_THRESH(priv);
        td.units = DPNI_CONGESTION_UNIT_FRAMES;
        for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
                err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
                                        DPNI_CP_GROUP, DPNI_QUEUE_RX,
                                        i, 0, &td);
                if (err) {
                        netdev_err(priv->net_dev,
                                   "dpni_set_taildrop(CG) failed\n");
                        return;
                }
        }

        priv->rx_cgtd_enabled = td.enable;
}

static int dpaa2_eth_link_state_update(struct dpaa2_eth_priv *priv)
{
        struct dpni_link_state state = {0};
        bool tx_pause;
        int err;

        err = dpni_get_link_state(priv->mc_io, 0, priv->mc_token, &state);
        if (unlikely(err)) {
                netdev_err(priv->net_dev,
                           "dpni_get_link_state() failed\n");
                return err;
        }

        /* If Tx pause frame settings have changed, we need to update
         * Rx FQ taildrop configuration as well. We configure taildrop
         * only when pause frame generation is disabled.
         */
        tx_pause = dpaa2_eth_tx_pause_enabled(state.options);
        dpaa2_eth_set_rx_taildrop(priv, tx_pause, priv->pfc_enabled);

        /* When we manage the MAC/PHY using phylink there is no need
         * to manually update the netif_carrier.
         */
        if (dpaa2_eth_is_type_phy(priv))
                goto out;

        /* Chech link state; speed / duplex changes are not treated yet */
        if (priv->link_state.up == state.up)
                goto out;

        if (state.up) {
                netif_carrier_on(priv->net_dev);
                netif_tx_start_all_queues(priv->net_dev);
        } else {
                netif_tx_stop_all_queues(priv->net_dev);
                netif_carrier_off(priv->net_dev);
        }

        netdev_info(priv->net_dev, "Link Event: state %s\n",
                    state.up ? "up" : "down");

out:
        priv->link_state = state;

        return 0;
}

static int dpaa2_eth_open(struct net_device *net_dev)
{
        struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
        int err;

        err = dpaa2_eth_seed_pool(priv, priv->bpid);
        if (err) {
                /* Not much to do; the buffer pool, though not filled up,
                 * may still contain some buffers which would enable us
                 * to limp on.
                 */
                netdev_err(net_dev, "Buffer seeding failed for DPBP %d (bpid=%d)\n",
                           priv->dpbp_dev->obj_desc.id, priv->bpid);
        }

        if (!dpaa2_eth_is_type_phy(priv)) {
                /* We'll only start the txqs when the link is actually ready;
                 * make sure we don't race against the link up notification,
                 * which may come immediately after dpni_enable();
                 */
                netif_tx_stop_all_queues(net_dev);

                /* Also, explicitly set carrier off, otherwise
                 * netif_carrier_ok() will return true and cause 'ip link show'
                 * to report the LOWER_UP flag, even though the link
                 * notification wasn't even received.
                 */
                netif_carrier_off(net_dev);
        }
        dpaa2_eth_enable_ch_napi(priv);

        err = dpni_enable(priv->mc_io, 0, priv->mc_token);
        if (err < 0) {
                netdev_err(net_dev, "dpni_enable() failed\n");
                goto enable_err;
        }

        if (dpaa2_eth_is_type_phy(priv))
                phylink_start(priv->mac->phylink);

        return 0;

enable_err:
        dpaa2_eth_disable_ch_napi(priv);
        dpaa2_eth_drain_pool(priv);
        return err;
}

/* Total number of in-flight frames on ingress queues */
static u32 dpaa2_eth_ingress_fq_count(struct dpaa2_eth_priv *priv)
{
        struct dpaa2_eth_fq *fq;
        u32 fcnt = 0, bcnt = 0, total = 0;
        int i, err;

        for (i = 0; i < priv->num_fqs; i++) {
                fq = &priv->fq[i];
                err = dpaa2_io_query_fq_count(NULL, fq->fqid, &fcnt, &bcnt);
                if (err) {
                        netdev_warn(priv->net_dev, "query_fq_count failed");
                        break;
                }
                total += fcnt;
        }

        return total;
}

static void dpaa2_eth_wait_for_ingress_fq_empty(struct dpaa2_eth_priv *priv)
{
        int retries = 10;
        u32 pending;

        do {
                pending = dpaa2_eth_ingress_fq_count(priv);
                if (pending)
                        msleep(100);
        } while (pending && --retries);
}

#define DPNI_TX_PENDING_VER_MAJOR       7
#define DPNI_TX_PENDING_VER_MINOR       13
static void dpaa2_eth_wait_for_egress_fq_empty(struct dpaa2_eth_priv *priv)
{
        union dpni_statistics stats;
        int retries = 10;
        int err;

        if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_TX_PENDING_VER_MAJOR,
                                   DPNI_TX_PENDING_VER_MINOR) < 0)
                goto out;

        do {
                err = dpni_get_statistics(priv->mc_io, 0, priv->mc_token, 6,
                                          &stats);
                if (err)
                        goto out;
                if (stats.page_6.tx_pending_frames == 0)
                        return;
        } while (--retries);

out:
        msleep(500);
}

static int dpaa2_eth_stop(struct net_device *net_dev)
{
        struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
        int dpni_enabled = 0;
        int retries = 10;

        if (dpaa2_eth_is_type_phy(priv)) {
                phylink_stop(priv->mac->phylink);
        } else {
                netif_tx_stop_all_queues(net_dev);
                netif_carrier_off(net_dev);
        }

        /* On dpni_disable(), the MC firmware will:
         * - stop MAC Rx and wait for all Rx frames to be enqueued to software
         * - cut off WRIOP dequeues from egress FQs and wait until transmission
         * of all in flight Tx frames is finished (and corresponding Tx conf
         * frames are enqueued back to software)
         *
         * Before calling dpni_disable(), we wait for all Tx frames to arrive
         * on WRIOP. After it finishes, wait until all remaining frames on Rx
         * and Tx conf queues are consumed on NAPI poll.
         */
        dpaa2_eth_wait_for_egress_fq_empty(priv);

        do {
                dpni_disable(priv->mc_io, 0, priv->mc_token);
                dpni_is_enabled(priv->mc_io, 0, priv->mc_token, &dpni_enabled);
                if (dpni_enabled)
                        /* Allow the hardware some slack */
                        msleep(100);
        } while (dpni_enabled && --retries);
        if (!retries) {
                netdev_warn(net_dev, "Retry count exceeded disabling DPNI\n");
                /* Must go on and disable NAPI nonetheless, so we don't crash at
                 * the next "ifconfig up"
                 */
        }

        dpaa2_eth_wait_for_ingress_fq_empty(priv);
        dpaa2_eth_disable_ch_napi(priv);

        /* Empty the buffer pool */
        dpaa2_eth_drain_pool(priv);

        /* Empty the Scatter-Gather Buffer cache */
        dpaa2_eth_sgt_cache_drain(priv);

        return 0;
}

static int dpaa2_eth_set_addr(struct net_device *net_dev, void *addr)
{
        struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
        struct device *dev = net_dev->dev.parent;
        int err;

        err = eth_mac_addr(net_dev, addr);
        if (err < 0) {
                dev_err(dev, "eth_mac_addr() failed (%d)\n", err);
                return err;
        }

        err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
                                        net_dev->dev_addr);
        if (err) {
                dev_err(dev, "dpni_set_primary_mac_addr() failed (%d)\n", err);
                return err;
        }

        return 0;
}

/** Fill in counters maintained by the GPP driver. These may be different from
 * the hardware counters obtained by ethtool.
 */
static void dpaa2_eth_get_stats(struct net_device *net_dev,
                                struct rtnl_link_stats64 *stats)
{
        struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
        struct rtnl_link_stats64 *percpu_stats;
        u64 *cpustats;
        u64 *netstats = (u64 *)stats;
        int i, j;
        int num = sizeof(struct rtnl_link_stats64) / sizeof(u64);

        for_each_possible_cpu(i) {
                percpu_stats = per_cpu_ptr(priv->percpu_stats, i);
                cpustats = (u64 *)percpu_stats;
                for (j = 0; j < num; j++)
                        netstats[j] += cpustats[j];
        }
}

/* Copy mac unicast addresses from @net_dev to @priv.
 * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
 */
static void dpaa2_eth_add_uc_hw_addr(const struct net_device *net_dev,
                                     struct dpaa2_eth_priv *priv)
{
        struct netdev_hw_addr *ha;
        int err;

        netdev_for_each_uc_addr(ha, net_dev) {
                err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
                                        ha->addr);
                if (err)
                        netdev_warn(priv->net_dev,
                                    "Could not add ucast MAC %pM to the filtering table (err %d)\n",
                                    ha->addr, err);
        }
}

/* Copy mac multicast addresses from @net_dev to @priv
 * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
 */
static void dpaa2_eth_add_mc_hw_addr(const struct net_device *net_dev,
                                     struct dpaa2_eth_priv *priv)
{
        struct netdev_hw_addr *ha;
        int err;

        netdev_for_each_mc_addr(ha, net_dev) {
                err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
                                        ha->addr);