]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
net: hisilicon: new hip04 ethernet driver
authordingtianhong <dingtianhong@huawei.com>
Wed, 14 Jan 2015 06:34:14 +0000 (14:34 +0800)
committerDavid S. Miller <davem@davemloft.net>
Wed, 14 Jan 2015 06:52:45 +0000 (01:52 -0500)
Support Hisilicon hip04 ethernet driver, including 100M / 1000M controller.
The controller has no tx done interrupt, reclaim xmitted buffer in the poll.

v13: Fix the problem of alignment parameters for function and checkpatch warming.

v12: According Alex's suggestion, modify the changelog and add MODULE_DEVICE_TABLE
     for hip04 ethernet.

v11: Add ethtool support for tx coalecse getting and setting, the xmit_more
     is not supported for this patch, but I think it could work for hip04,
     will support it later after some tests for performance better.

     Here are some performance test results by ping and iperf(add tx_coalesce_frames/users),
     it looks that the performance and latency is more better by tx_coalesce_frames/usecs.

     - Before:
     $ ping 192.168.1.1 ...
     === 192.168.1.1 ping statistics ===
     24 packets transmitted, 24 received, 0% packet loss, time 22999ms
     rtt min/avg/max/mdev = 0.180/0.202/0.403/0.043 ms

     $ iperf -c 192.168.1.1 ...
     [ ID] Interval       Transfer     Bandwidth
     [  3]  0.0- 1.0 sec   115 MBytes   945 Mbits/sec

     - After:
     $ ping 192.168.1.1 ...
     === 192.168.1.1 ping statistics ===
     24 packets transmitted, 24 received, 0% packet loss, time 22999ms
     rtt min/avg/max/mdev = 0.178/0.190/0.380/0.041 ms

     $ iperf -c 192.168.1.1 ...
     [ ID] Interval       Transfer     Bandwidth
     [  3]  0.0- 1.0 sec   115 MBytes   965 Mbits/sec

v10: According David Miller and Arnd Bergmann's suggestion, add some modification
     for v9 version
     - drop the workqueue
     - batch cleanup based on tx_coalesce_frames/usecs for better throughput
     - use a reasonable default tx timeout (200us, could be shorted
       based on measurements) with a range timer
     - fix napi poll function return value
     - use a lockless queue for cleanup

Signed-off-by: Zhangfei Gao <zhangfei.gao@linaro.org>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/hisilicon/Makefile
drivers/net/ethernet/hisilicon/hip04_eth.c [new file with mode: 0644]

index 40115a7e2ed50aa8323aa86df3d7b4c46f7dc21c..6c14540a4dc5e18c6049aaa24e6e8184af99ddd7 100644 (file)
@@ -3,4 +3,4 @@
 #
 
 obj-$(CONFIG_HIX5HD2_GMAC) += hix5hd2_gmac.o
-obj-$(CONFIG_HIP04_ETH) += hip04_mdio.o
+obj-$(CONFIG_HIP04_ETH) += hip04_mdio.o hip04_eth.o
diff --git a/drivers/net/ethernet/hisilicon/hip04_eth.c b/drivers/net/ethernet/hisilicon/hip04_eth.c
new file mode 100644 (file)
index 0000000..525214e
--- /dev/null
@@ -0,0 +1,969 @@
+
+/* Copyright (c) 2014 Linaro Ltd.
+ * Copyright (c) 2014 Hisilicon Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ */
+
+#include <linux/module.h>
+#include <linux/etherdevice.h>
+#include <linux/platform_device.h>
+#include <linux/interrupt.h>
+#include <linux/ktime.h>
+#include <linux/of_address.h>
+#include <linux/phy.h>
+#include <linux/of_mdio.h>
+#include <linux/of_net.h>
+#include <linux/mfd/syscon.h>
+#include <linux/regmap.h>
+
+#define PPE_CFG_RX_ADDR                        0x100
+#define PPE_CFG_POOL_GRP               0x300
+#define PPE_CFG_RX_BUF_SIZE            0x400
+#define PPE_CFG_RX_FIFO_SIZE           0x500
+#define PPE_CURR_BUF_CNT               0xa200
+
+#define GE_DUPLEX_TYPE                 0x08
+#define GE_MAX_FRM_SIZE_REG            0x3c
+#define GE_PORT_MODE                   0x40
+#define GE_PORT_EN                     0x44
+#define GE_SHORT_RUNTS_THR_REG         0x50
+#define GE_TX_LOCAL_PAGE_REG           0x5c
+#define GE_TRANSMIT_CONTROL_REG                0x60
+#define GE_CF_CRC_STRIP_REG            0x1b0
+#define GE_MODE_CHANGE_REG             0x1b4
+#define GE_RECV_CONTROL_REG            0x1e0
+#define GE_STATION_MAC_ADDRESS         0x210
+#define PPE_CFG_CPU_ADD_ADDR           0x580
+#define PPE_CFG_MAX_FRAME_LEN_REG      0x408
+#define PPE_CFG_BUS_CTRL_REG           0x424
+#define PPE_CFG_RX_CTRL_REG            0x428
+#define PPE_CFG_RX_PKT_MODE_REG                0x438
+#define PPE_CFG_QOS_VMID_GEN           0x500
+#define PPE_CFG_RX_PKT_INT             0x538
+#define PPE_INTEN                      0x600
+#define PPE_INTSTS                     0x608
+#define PPE_RINT                       0x604
+#define PPE_CFG_STS_MODE               0x700
+#define PPE_HIS_RX_PKT_CNT             0x804
+
+/* REG_INTERRUPT */
+#define RCV_INT                                BIT(10)
+#define RCV_NOBUF                      BIT(8)
+#define RCV_DROP                       BIT(7)
+#define TX_DROP                                BIT(6)
+#define DEF_INT_ERR                    (RCV_NOBUF | RCV_DROP | TX_DROP)
+#define DEF_INT_MASK                   (RCV_INT | DEF_INT_ERR)
+
+/* TX descriptor config */
+#define TX_FREE_MEM                    BIT(0)
+#define TX_READ_ALLOC_L3               BIT(1)
+#define TX_FINISH_CACHE_INV            BIT(2)
+#define TX_CLEAR_WB                    BIT(4)
+#define TX_L3_CHECKSUM                 BIT(5)
+#define TX_LOOP_BACK                   BIT(11)
+
+/* RX error */
+#define RX_PKT_DROP                    BIT(0)
+#define RX_L2_ERR                      BIT(1)
+#define RX_PKT_ERR                     (RX_PKT_DROP | RX_L2_ERR)
+
+#define SGMII_SPEED_1000               0x08
+#define SGMII_SPEED_100                        0x07
+#define SGMII_SPEED_10                 0x06
+#define MII_SPEED_100                  0x01
+#define MII_SPEED_10                   0x00
+
+#define GE_DUPLEX_FULL                 BIT(0)
+#define GE_DUPLEX_HALF                 0x00
+#define GE_MODE_CHANGE_EN              BIT(0)
+
+#define GE_TX_AUTO_NEG                 BIT(5)
+#define GE_TX_ADD_CRC                  BIT(6)
+#define GE_TX_SHORT_PAD_THROUGH                BIT(7)
+
+#define GE_RX_STRIP_CRC                        BIT(0)
+#define GE_RX_STRIP_PAD                        BIT(3)
+#define GE_RX_PAD_EN                   BIT(4)
+
+#define GE_AUTO_NEG_CTL                        BIT(0)
+
+#define GE_RX_INT_THRESHOLD            BIT(6)
+#define GE_RX_TIMEOUT                  0x04
+
+#define GE_RX_PORT_EN                  BIT(1)
+#define GE_TX_PORT_EN                  BIT(2)
+
+#define PPE_CFG_STS_RX_PKT_CNT_RC      BIT(12)
+
+#define PPE_CFG_RX_PKT_ALIGN           BIT(18)
+#define PPE_CFG_QOS_VMID_MODE          BIT(14)
+#define PPE_CFG_QOS_VMID_GRP_SHIFT     8
+
+#define PPE_CFG_RX_FIFO_FSFU           BIT(11)
+#define PPE_CFG_RX_DEPTH_SHIFT         16
+#define PPE_CFG_RX_START_SHIFT         0
+#define PPE_CFG_RX_CTRL_ALIGN_SHIFT    11
+
+#define PPE_CFG_BUS_LOCAL_REL          BIT(14)
+#define PPE_CFG_BUS_BIG_ENDIEN         BIT(0)
+
+#define RX_DESC_NUM                    128
+#define TX_DESC_NUM                    256
+#define TX_NEXT(N)                     (((N) + 1) & (TX_DESC_NUM-1))
+#define RX_NEXT(N)                     (((N) + 1) & (RX_DESC_NUM-1))
+
+#define GMAC_PPE_RX_PKT_MAX_LEN                379
+#define GMAC_MAX_PKT_LEN               1516
+#define GMAC_MIN_PKT_LEN               31
+#define RX_BUF_SIZE                    1600
+#define RESET_TIMEOUT                  1000
+#define TX_TIMEOUT                     (6 * HZ)
+
+#define DRV_NAME                       "hip04-ether"
+#define DRV_VERSION                    "v1.0"
+
+#define HIP04_MAX_TX_COALESCE_USECS    200
+#define HIP04_MIN_TX_COALESCE_USECS    100
+#define HIP04_MAX_TX_COALESCE_FRAMES   200
+#define HIP04_MIN_TX_COALESCE_FRAMES   100
+
+struct tx_desc {
+       u32 send_addr;
+       u32 send_size;
+       u32 next_addr;
+       u32 cfg;
+       u32 wb_addr;
+} __aligned(64);
+
+struct rx_desc {
+       u16 reserved_16;
+       u16 pkt_len;
+       u32 reserve1[3];
+       u32 pkt_err;
+       u32 reserve2[4];
+};
+
+struct hip04_priv {
+       void __iomem *base;
+       int phy_mode;
+       int chan;
+       unsigned int port;
+       unsigned int speed;
+       unsigned int duplex;
+       unsigned int reg_inten;
+
+       struct napi_struct napi;
+       struct net_device *ndev;
+
+       struct tx_desc *tx_desc;
+       dma_addr_t tx_desc_dma;
+       struct sk_buff *tx_skb[TX_DESC_NUM];
+       dma_addr_t tx_phys[TX_DESC_NUM];
+       unsigned int tx_head;
+
+       int tx_coalesce_frames;
+       int tx_coalesce_usecs;
+       struct hrtimer tx_coalesce_timer;
+
+       unsigned char *rx_buf[RX_DESC_NUM];
+       dma_addr_t rx_phys[RX_DESC_NUM];
+       unsigned int rx_head;
+       unsigned int rx_buf_size;
+
+       struct device_node *phy_node;
+       struct phy_device *phy;
+       struct regmap *map;
+       struct work_struct tx_timeout_task;
+
+       /* written only by tx cleanup */
+       unsigned int tx_tail ____cacheline_aligned_in_smp;
+};
+
+static inline unsigned int tx_count(unsigned int head, unsigned int tail)
+{
+       return (head - tail) % (TX_DESC_NUM - 1);
+}
+
+static void hip04_config_port(struct net_device *ndev, u32 speed, u32 duplex)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+       u32 val;
+
+       priv->speed = speed;
+       priv->duplex = duplex;
+
+       switch (priv->phy_mode) {
+       case PHY_INTERFACE_MODE_SGMII:
+               if (speed == SPEED_1000)
+                       val = SGMII_SPEED_1000;
+               else if (speed == SPEED_100)
+                       val = SGMII_SPEED_100;
+               else
+                       val = SGMII_SPEED_10;
+               break;
+       case PHY_INTERFACE_MODE_MII:
+               if (speed == SPEED_100)
+                       val = MII_SPEED_100;
+               else
+                       val = MII_SPEED_10;
+               break;
+       default:
+               netdev_warn(ndev, "not supported mode\n");
+               val = MII_SPEED_10;
+               break;
+       }
+       writel_relaxed(val, priv->base + GE_PORT_MODE);
+
+       val = duplex ? GE_DUPLEX_FULL : GE_DUPLEX_HALF;
+       writel_relaxed(val, priv->base + GE_DUPLEX_TYPE);
+
+       val = GE_MODE_CHANGE_EN;
+       writel_relaxed(val, priv->base + GE_MODE_CHANGE_REG);
+}
+
+static void hip04_reset_ppe(struct hip04_priv *priv)
+{
+       u32 val, tmp, timeout = 0;
+
+       do {
+               regmap_read(priv->map, priv->port * 4 + PPE_CURR_BUF_CNT, &val);
+               regmap_read(priv->map, priv->port * 4 + PPE_CFG_RX_ADDR, &tmp);
+               if (timeout++ > RESET_TIMEOUT)
+                       break;
+       } while (val & 0xfff);
+}
+
+static void hip04_config_fifo(struct hip04_priv *priv)
+{
+       u32 val;
+
+       val = readl_relaxed(priv->base + PPE_CFG_STS_MODE);
+       val |= PPE_CFG_STS_RX_PKT_CNT_RC;
+       writel_relaxed(val, priv->base + PPE_CFG_STS_MODE);
+
+       val = BIT(priv->port);
+       regmap_write(priv->map, priv->port * 4 + PPE_CFG_POOL_GRP, val);
+
+       val = priv->port << PPE_CFG_QOS_VMID_GRP_SHIFT;
+       val |= PPE_CFG_QOS_VMID_MODE;
+       writel_relaxed(val, priv->base + PPE_CFG_QOS_VMID_GEN);
+
+       val = RX_BUF_SIZE;
+       regmap_write(priv->map, priv->port * 4 + PPE_CFG_RX_BUF_SIZE, val);
+
+       val = RX_DESC_NUM << PPE_CFG_RX_DEPTH_SHIFT;
+       val |= PPE_CFG_RX_FIFO_FSFU;
+       val |= priv->chan << PPE_CFG_RX_START_SHIFT;
+       regmap_write(priv->map, priv->port * 4 + PPE_CFG_RX_FIFO_SIZE, val);
+
+       val = NET_IP_ALIGN << PPE_CFG_RX_CTRL_ALIGN_SHIFT;
+       writel_relaxed(val, priv->base + PPE_CFG_RX_CTRL_REG);
+
+       val = PPE_CFG_RX_PKT_ALIGN;
+       writel_relaxed(val, priv->base + PPE_CFG_RX_PKT_MODE_REG);
+
+       val = PPE_CFG_BUS_LOCAL_REL | PPE_CFG_BUS_BIG_ENDIEN;
+       writel_relaxed(val, priv->base + PPE_CFG_BUS_CTRL_REG);
+
+       val = GMAC_PPE_RX_PKT_MAX_LEN;
+       writel_relaxed(val, priv->base + PPE_CFG_MAX_FRAME_LEN_REG);
+
+       val = GMAC_MAX_PKT_LEN;
+       writel_relaxed(val, priv->base + GE_MAX_FRM_SIZE_REG);
+
+       val = GMAC_MIN_PKT_LEN;
+       writel_relaxed(val, priv->base + GE_SHORT_RUNTS_THR_REG);
+
+       val = readl_relaxed(priv->base + GE_TRANSMIT_CONTROL_REG);
+       val |= GE_TX_AUTO_NEG | GE_TX_ADD_CRC | GE_TX_SHORT_PAD_THROUGH;
+       writel_relaxed(val, priv->base + GE_TRANSMIT_CONTROL_REG);
+
+       val = GE_RX_STRIP_CRC;
+       writel_relaxed(val, priv->base + GE_CF_CRC_STRIP_REG);
+
+       val = readl_relaxed(priv->base + GE_RECV_CONTROL_REG);
+       val |= GE_RX_STRIP_PAD | GE_RX_PAD_EN;
+       writel_relaxed(val, priv->base + GE_RECV_CONTROL_REG);
+
+       val = GE_AUTO_NEG_CTL;
+       writel_relaxed(val, priv->base + GE_TX_LOCAL_PAGE_REG);
+}
+
+static void hip04_mac_enable(struct net_device *ndev)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+       u32 val;
+
+       /* enable tx & rx */
+       val = readl_relaxed(priv->base + GE_PORT_EN);
+       val |= GE_RX_PORT_EN | GE_TX_PORT_EN;
+       writel_relaxed(val, priv->base + GE_PORT_EN);
+
+       /* clear rx int */
+       val = RCV_INT;
+       writel_relaxed(val, priv->base + PPE_RINT);
+
+       /* config recv int */
+       val = GE_RX_INT_THRESHOLD | GE_RX_TIMEOUT;
+       writel_relaxed(val, priv->base + PPE_CFG_RX_PKT_INT);
+
+       /* enable interrupt */
+       priv->reg_inten = DEF_INT_MASK;
+       writel_relaxed(priv->reg_inten, priv->base + PPE_INTEN);
+}
+
+static void hip04_mac_disable(struct net_device *ndev)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+       u32 val;
+
+       /* disable int */
+       priv->reg_inten &= ~(DEF_INT_MASK);
+       writel_relaxed(priv->reg_inten, priv->base + PPE_INTEN);
+
+       /* disable tx & rx */
+       val = readl_relaxed(priv->base + GE_PORT_EN);
+       val &= ~(GE_RX_PORT_EN | GE_TX_PORT_EN);
+       writel_relaxed(val, priv->base + GE_PORT_EN);
+}
+
+static void hip04_set_xmit_desc(struct hip04_priv *priv, dma_addr_t phys)
+{
+       writel(phys, priv->base + PPE_CFG_CPU_ADD_ADDR);
+}
+
+static void hip04_set_recv_desc(struct hip04_priv *priv, dma_addr_t phys)
+{
+       regmap_write(priv->map, priv->port * 4 + PPE_CFG_RX_ADDR, phys);
+}
+
+static u32 hip04_recv_cnt(struct hip04_priv *priv)
+{
+       return readl(priv->base + PPE_HIS_RX_PKT_CNT);
+}
+
+static void hip04_update_mac_address(struct net_device *ndev)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+
+       writel_relaxed(((ndev->dev_addr[0] << 8) | (ndev->dev_addr[1])),
+                      priv->base + GE_STATION_MAC_ADDRESS);
+       writel_relaxed(((ndev->dev_addr[2] << 24) | (ndev->dev_addr[3] << 16) |
+                       (ndev->dev_addr[4] << 8) | (ndev->dev_addr[5])),
+                      priv->base + GE_STATION_MAC_ADDRESS + 4);
+}
+
+static int hip04_set_mac_address(struct net_device *ndev, void *addr)
+{
+       eth_mac_addr(ndev, addr);
+       hip04_update_mac_address(ndev);
+       return 0;
+}
+
+static int hip04_tx_reclaim(struct net_device *ndev, bool force)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+       unsigned tx_tail = priv->tx_tail;
+       struct tx_desc *desc;
+       unsigned int bytes_compl = 0, pkts_compl = 0;
+       unsigned int count;
+
+       smp_rmb();
+       count = tx_count(ACCESS_ONCE(priv->tx_head), tx_tail);
+       if (count == 0)
+               goto out;
+
+       while (count) {
+               desc = &priv->tx_desc[tx_tail];
+               if (desc->send_addr != 0) {
+                       if (force)
+                               desc->send_addr = 0;
+                       else
+                               break;
+               }
+
+               if (priv->tx_phys[tx_tail]) {
+                       dma_unmap_single(&ndev->dev, priv->tx_phys[tx_tail],
+                                        priv->tx_skb[tx_tail]->len,
+                                        DMA_TO_DEVICE);
+                       priv->tx_phys[tx_tail] = 0;
+               }
+               pkts_compl++;
+               bytes_compl += priv->tx_skb[tx_tail]->len;
+               dev_kfree_skb(priv->tx_skb[tx_tail]);
+               priv->tx_skb[tx_tail] = NULL;
+               tx_tail = TX_NEXT(tx_tail);
+               count--;
+       }
+
+       priv->tx_tail = tx_tail;
+       smp_wmb(); /* Ensure tx_tail visible to xmit */
+
+out:
+       if (pkts_compl || bytes_compl)
+               netdev_completed_queue(ndev, pkts_compl, bytes_compl);
+
+       if (unlikely(netif_queue_stopped(ndev)) && (count < (TX_DESC_NUM - 1)))
+               netif_wake_queue(ndev);
+
+       return count;
+}
+
+static int hip04_mac_start_xmit(struct sk_buff *skb, struct net_device *ndev)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+       struct net_device_stats *stats = &ndev->stats;
+       unsigned int tx_head = priv->tx_head, count;
+       struct tx_desc *desc = &priv->tx_desc[tx_head];
+       dma_addr_t phys;
+
+       smp_rmb();
+       count = tx_count(tx_head, ACCESS_ONCE(priv->tx_tail));
+       if (count == (TX_DESC_NUM - 1)) {
+               netif_stop_queue(ndev);
+               return NETDEV_TX_BUSY;
+       }
+
+       phys = dma_map_single(&ndev->dev, skb->data, skb->len, DMA_TO_DEVICE);
+       if (dma_mapping_error(&ndev->dev, phys)) {
+               dev_kfree_skb(skb);
+               return NETDEV_TX_OK;
+       }
+
+       priv->tx_skb[tx_head] = skb;
+       priv->tx_phys[tx_head] = phys;
+       desc->send_addr = cpu_to_be32(phys);
+       desc->send_size = cpu_to_be32(skb->len);
+       desc->cfg = cpu_to_be32(TX_CLEAR_WB | TX_FINISH_CACHE_INV);
+       phys = priv->tx_desc_dma + tx_head * sizeof(struct tx_desc);
+       desc->wb_addr = cpu_to_be32(phys);
+       skb_tx_timestamp(skb);
+
+       hip04_set_xmit_desc(priv, phys);
+       priv->tx_head = TX_NEXT(tx_head);
+       count++;
+       netdev_sent_queue(ndev, skb->len);
+
+       stats->tx_bytes += skb->len;
+       stats->tx_packets++;
+
+       /* Ensure tx_head update visible to tx reclaim */
+       smp_wmb();
+
+       /* queue is getting full, better start cleaning up now */
+       if (count >= priv->tx_coalesce_frames) {
+               if (napi_schedule_prep(&priv->napi)) {
+                       /* disable rx interrupt and timer */
+                       priv->reg_inten &= ~(RCV_INT);
+                       writel_relaxed(DEF_INT_MASK & ~RCV_INT,
+                                      priv->base + PPE_INTEN);
+                       hrtimer_cancel(&priv->tx_coalesce_timer);
+                       __napi_schedule(&priv->napi);
+               }
+       } else if (!hrtimer_is_queued(&priv->tx_coalesce_timer)) {
+               /* cleanup not pending yet, start a new timer */
+               hrtimer_start_expires(&priv->tx_coalesce_timer,
+                                     HRTIMER_MODE_REL);
+       }
+
+       return NETDEV_TX_OK;
+}
+
+static int hip04_rx_poll(struct napi_struct *napi, int budget)
+{
+       struct hip04_priv *priv = container_of(napi, struct hip04_priv, napi);
+       struct net_device *ndev = priv->ndev;
+       struct net_device_stats *stats = &ndev->stats;
+       unsigned int cnt = hip04_recv_cnt(priv);
+       struct rx_desc *desc;
+       struct sk_buff *skb;
+       unsigned char *buf;
+       bool last = false;
+       dma_addr_t phys;
+       int rx = 0;
+       int tx_remaining;
+       u16 len;
+       u32 err;
+
+       while (cnt && !last) {
+               buf = priv->rx_buf[priv->rx_head];
+               skb = build_skb(buf, priv->rx_buf_size);
+               if (unlikely(!skb))
+                       net_dbg_ratelimited("build_skb failed\n");
+
+               dma_unmap_single(&ndev->dev, priv->rx_phys[priv->rx_head],
+                                RX_BUF_SIZE, DMA_FROM_DEVICE);
+               priv->rx_phys[priv->rx_head] = 0;
+
+               desc = (struct rx_desc *)skb->data;
+               len = be16_to_cpu(desc->pkt_len);
+               err = be32_to_cpu(desc->pkt_err);
+
+               if (0 == len) {
+                       dev_kfree_skb_any(skb);
+                       last = true;
+               } else if ((err & RX_PKT_ERR) || (len >= GMAC_MAX_PKT_LEN)) {
+                       dev_kfree_skb_any(skb);
+                       stats->rx_dropped++;
+                       stats->rx_errors++;
+               } else {
+                       skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
+                       skb_put(skb, len);
+                       skb->protocol = eth_type_trans(skb, ndev);
+                       napi_gro_receive(&priv->napi, skb);
+                       stats->rx_packets++;
+                       stats->rx_bytes += len;
+                       rx++;
+               }
+
+               buf = netdev_alloc_frag(priv->rx_buf_size);
+               if (!buf)
+                       goto done;
+               phys = dma_map_single(&ndev->dev, buf,
+                                     RX_BUF_SIZE, DMA_FROM_DEVICE);
+               if (dma_mapping_error(&ndev->dev, phys))
+                       goto done;
+               priv->rx_buf[priv->rx_head] = buf;
+               priv->rx_phys[priv->rx_head] = phys;
+               hip04_set_recv_desc(priv, phys);
+
+               priv->rx_head = RX_NEXT(priv->rx_head);
+               if (rx >= budget)
+                       goto done;
+
+               if (--cnt == 0)
+                       cnt = hip04_recv_cnt(priv);
+       }
+
+       if (!(priv->reg_inten & RCV_INT)) {
+               /* enable rx interrupt */
+               priv->reg_inten |= RCV_INT;
+               writel_relaxed(priv->reg_inten, priv->base + PPE_INTEN);
+       }
+       napi_complete(napi);
+done:
+       /* clean up tx descriptors and start a new timer if necessary */
+       tx_remaining = hip04_tx_reclaim(ndev, false);
+       if (rx < budget && tx_remaining)
+               hrtimer_start_expires(&priv->tx_coalesce_timer, HRTIMER_MODE_REL);
+
+       return rx;
+}
+
+static irqreturn_t hip04_mac_interrupt(int irq, void *dev_id)
+{
+       struct net_device *ndev = (struct net_device *)dev_id;
+       struct hip04_priv *priv = netdev_priv(ndev);
+       struct net_device_stats *stats = &ndev->stats;
+       u32 ists = readl_relaxed(priv->base + PPE_INTSTS);
+
+       if (!ists)
+               return IRQ_NONE;
+
+       writel_relaxed(DEF_INT_MASK, priv->base + PPE_RINT);
+
+       if (unlikely(ists & DEF_INT_ERR)) {
+               if (ists & (RCV_NOBUF | RCV_DROP))
+                       stats->rx_errors++;
+                       stats->rx_dropped++;
+                       netdev_err(ndev, "rx drop\n");
+               if (ists & TX_DROP) {
+                       stats->tx_dropped++;
+                       netdev_err(ndev, "tx drop\n");
+               }
+       }
+
+       if (ists & RCV_INT && napi_schedule_prep(&priv->napi)) {
+               /* disable rx interrupt */
+               priv->reg_inten &= ~(RCV_INT);
+               writel_relaxed(DEF_INT_MASK & ~RCV_INT, priv->base + PPE_INTEN);
+               hrtimer_cancel(&priv->tx_coalesce_timer);
+               __napi_schedule(&priv->napi);
+       }
+
+       return IRQ_HANDLED;
+}
+
+enum hrtimer_restart tx_done(struct hrtimer *hrtimer)
+{
+       struct hip04_priv *priv;
+
+       priv = container_of(hrtimer, struct hip04_priv, tx_coalesce_timer);
+
+       if (napi_schedule_prep(&priv->napi)) {
+               /* disable rx interrupt */
+               priv->reg_inten &= ~(RCV_INT);
+               writel_relaxed(DEF_INT_MASK & ~RCV_INT, priv->base + PPE_INTEN);
+               __napi_schedule(&priv->napi);
+       }
+
+       return HRTIMER_NORESTART;
+}
+
+static void hip04_adjust_link(struct net_device *ndev)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+       struct phy_device *phy = priv->phy;
+
+       if ((priv->speed != phy->speed) || (priv->duplex != phy->duplex)) {
+               hip04_config_port(ndev, phy->speed, phy->duplex);
+               phy_print_status(phy);
+       }
+}
+
+static int hip04_mac_open(struct net_device *ndev)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+       int i;
+
+       priv->rx_head = 0;
+       priv->tx_head = 0;
+       priv->tx_tail = 0;
+       hip04_reset_ppe(priv);
+
+       for (i = 0; i < RX_DESC_NUM; i++) {
+               dma_addr_t phys;
+
+               phys = dma_map_single(&ndev->dev, priv->rx_buf[i],
+                                     RX_BUF_SIZE, DMA_FROM_DEVICE);
+               if (dma_mapping_error(&ndev->dev, phys))
+                       return -EIO;
+
+               priv->rx_phys[i] = phys;
+               hip04_set_recv_desc(priv, phys);
+       }
+
+       if (priv->phy)
+               phy_start(priv->phy);
+
+       netdev_reset_queue(ndev);
+       netif_start_queue(ndev);
+       hip04_mac_enable(ndev);
+       napi_enable(&priv->napi);
+
+       return 0;
+}
+
+static int hip04_mac_stop(struct net_device *ndev)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+       int i;
+
+       napi_disable(&priv->napi);
+       netif_stop_queue(ndev);
+       hip04_mac_disable(ndev);
+       hip04_tx_reclaim(ndev, true);
+       hip04_reset_ppe(priv);
+
+       if (priv->phy)
+               phy_stop(priv->phy);
+
+       for (i = 0; i < RX_DESC_NUM; i++) {
+               if (priv->rx_phys[i]) {
+                       dma_unmap_single(&ndev->dev, priv->rx_phys[i],
+                                        RX_BUF_SIZE, DMA_FROM_DEVICE);
+                       priv->rx_phys[i] = 0;
+               }
+       }
+
+       return 0;
+}
+
+static void hip04_timeout(struct net_device *ndev)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+
+       schedule_work(&priv->tx_timeout_task);
+}
+
+static void hip04_tx_timeout_task(struct work_struct *work)
+{
+       struct hip04_priv *priv;
+
+       priv = container_of(work, struct hip04_priv, tx_timeout_task);
+       hip04_mac_stop(priv->ndev);
+       hip04_mac_open(priv->ndev);
+}
+
+static struct net_device_stats *hip04_get_stats(struct net_device *ndev)
+{
+       return &ndev->stats;
+}
+
+static int hip04_get_coalesce(struct net_device *netdev,
+                             struct ethtool_coalesce *ec)
+{
+       struct hip04_priv *priv = netdev_priv(netdev);
+
+       ec->tx_coalesce_usecs = priv->tx_coalesce_usecs;
+       ec->tx_max_coalesced_frames = priv->tx_coalesce_frames;
+
+       return 0;
+}
+
+static int hip04_set_coalesce(struct net_device *netdev,
+                             struct ethtool_coalesce *ec)
+{
+       struct hip04_priv *priv = netdev_priv(netdev);
+
+       /* Check not supported parameters  */
+       if ((ec->rx_max_coalesced_frames) || (ec->rx_coalesce_usecs_irq) ||
+           (ec->rx_max_coalesced_frames_irq) || (ec->tx_coalesce_usecs_irq) ||
+           (ec->use_adaptive_rx_coalesce) || (ec->use_adaptive_tx_coalesce) ||
+           (ec->pkt_rate_low) || (ec->rx_coalesce_usecs_low) ||
+           (ec->rx_max_coalesced_frames_low) || (ec->tx_coalesce_usecs_high) ||
+           (ec->tx_max_coalesced_frames_low) || (ec->pkt_rate_high) ||
+           (ec->tx_coalesce_usecs_low) || (ec->rx_coalesce_usecs_high) ||
+           (ec->rx_max_coalesced_frames_high) || (ec->rx_coalesce_usecs) ||
+           (ec->tx_max_coalesced_frames_irq) ||
+           (ec->stats_block_coalesce_usecs) ||
+           (ec->tx_max_coalesced_frames_high) || (ec->rate_sample_interval))
+               return -EOPNOTSUPP;
+
+       if ((ec->tx_coalesce_usecs > HIP04_MAX_TX_COALESCE_USECS ||
+            ec->tx_coalesce_usecs < HIP04_MIN_TX_COALESCE_USECS) ||
+           (ec->tx_max_coalesced_frames > HIP04_MAX_TX_COALESCE_FRAMES ||
+            ec->tx_max_coalesced_frames < HIP04_MIN_TX_COALESCE_FRAMES))
+               return -EINVAL;
+
+       priv->tx_coalesce_usecs = ec->tx_coalesce_usecs;
+       priv->tx_coalesce_frames = ec->tx_max_coalesced_frames;
+
+       return 0;
+}
+
+static void hip04_get_drvinfo(struct net_device *netdev,
+                             struct ethtool_drvinfo *drvinfo)
+{
+       strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
+       strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
+}
+
+static struct ethtool_ops hip04_ethtool_ops = {
+       .get_coalesce           = hip04_get_coalesce,
+       .set_coalesce           = hip04_set_coalesce,
+       .get_drvinfo            = hip04_get_drvinfo,
+};
+
+static struct net_device_ops hip04_netdev_ops = {
+       .ndo_open               = hip04_mac_open,
+       .ndo_stop               = hip04_mac_stop,
+       .ndo_get_stats          = hip04_get_stats,
+       .ndo_start_xmit         = hip04_mac_start_xmit,
+       .ndo_set_mac_address    = hip04_set_mac_address,
+       .ndo_tx_timeout         = hip04_timeout,
+       .ndo_validate_addr      = eth_validate_addr,
+       .ndo_change_mtu         = eth_change_mtu,
+};
+
+static int hip04_alloc_ring(struct net_device *ndev, struct device *d)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+       int i;
+
+       priv->tx_desc = dma_alloc_coherent(d,
+                                          TX_DESC_NUM * sizeof(struct tx_desc),
+                                          &priv->tx_desc_dma, GFP_KERNEL);
+       if (!priv->tx_desc)
+               return -ENOMEM;
+
+       priv->rx_buf_size = RX_BUF_SIZE +
+                           SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
+       for (i = 0; i < RX_DESC_NUM; i++) {
+               priv->rx_buf[i] = netdev_alloc_frag(priv->rx_buf_size);
+               if (!priv->rx_buf[i])
+                       return -ENOMEM;
+       }
+
+       return 0;
+}
+
+static void hip04_free_ring(struct net_device *ndev, struct device *d)
+{
+       struct hip04_priv *priv = netdev_priv(ndev);
+       int i;
+
+       for (i = 0; i < RX_DESC_NUM; i++)
+               if (priv->rx_buf[i])
+                       put_page(virt_to_head_page(priv->rx_buf[i]));
+
+       for (i = 0; i < TX_DESC_NUM; i++)
+               if (priv->tx_skb[i])
+                       dev_kfree_skb_any(priv->tx_skb[i]);
+
+       dma_free_coherent(d, TX_DESC_NUM * sizeof(struct tx_desc),
+                         priv->tx_desc, priv->tx_desc_dma);
+}
+
+static int hip04_mac_probe(struct platform_device *pdev)
+{
+       struct device *d = &pdev->dev;
+       struct device_node *node = d->of_node;
+       struct of_phandle_args arg;
+       struct net_device *ndev;
+       struct hip04_priv *priv;
+       struct resource *res;
+       unsigned int irq;
+       ktime_t txtime;
+       int ret;
+
+       ndev = alloc_etherdev(sizeof(struct hip04_priv));
+       if (!ndev)
+               return -ENOMEM;
+
+       priv = netdev_priv(ndev);
+       priv->ndev = ndev;
+       platform_set_drvdata(pdev, ndev);
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       priv->base = devm_ioremap_resource(d, res);
+       if (IS_ERR(priv->base)) {
+               ret = PTR_ERR(priv->base);
+               goto init_fail;
+       }
+
+       ret = of_parse_phandle_with_fixed_args(node, "port-handle", 2, 0, &arg);
+       if (ret < 0) {
+               dev_warn(d, "no port-handle\n");
+               goto init_fail;
+       }
+
+       priv->port = arg.args[0];
+       priv->chan = arg.args[1] * RX_DESC_NUM;
+
+       hrtimer_init(&priv->tx_coalesce_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+
+       /* BQL will try to keep the TX queue as short as possible, but it can't
+        * be faster than tx_coalesce_usecs, so we need a fast timeout here,
+        * but also long enough to gather up enough frames to ensure we don't
+        * get more interrupts than necessary.
+        * 200us is enough for 16 frames of 1500 bytes at gigabit ethernet rate
+        */
+       priv->tx_coalesce_frames = TX_DESC_NUM * 3 / 4;
+       priv->tx_coalesce_usecs = 200;
+       /* allow timer to fire after half the time at the earliest */
+       txtime = ktime_set(0, priv->tx_coalesce_usecs * NSEC_PER_USEC / 2);
+       hrtimer_set_expires_range(&priv->tx_coalesce_timer, txtime, txtime);
+       priv->tx_coalesce_timer.function = tx_done;
+
+       priv->map = syscon_node_to_regmap(arg.np);
+       if (IS_ERR(priv->map)) {
+               dev_warn(d, "no syscon hisilicon,hip04-ppe\n");
+               ret = PTR_ERR(priv->map);
+               goto init_fail;
+       }
+
+       priv->phy_mode = of_get_phy_mode(node);
+       if (priv->phy_mode < 0) {
+               dev_warn(d, "not find phy-mode\n");
+               ret = -EINVAL;
+               goto init_fail;
+       }
+
+       irq = platform_get_irq(pdev, 0);
+       if (irq <= 0) {
+               ret = -EINVAL;
+               goto init_fail;
+       }
+
+       ret = devm_request_irq(d, irq, hip04_mac_interrupt,
+                              0, pdev->name, ndev);
+       if (ret) {
+               netdev_err(ndev, "devm_request_irq failed\n");
+               goto init_fail;
+       }
+
+       priv->phy_node = of_parse_phandle(node, "phy-handle", 0);
+       if (priv->phy_node) {
+               priv->phy = of_phy_connect(ndev, priv->phy_node,
+                                          &hip04_adjust_link,
+                                          0, priv->phy_mode);
+               if (!priv->phy) {
+                       ret = -EPROBE_DEFER;
+                       goto init_fail;
+               }
+       }
+
+       INIT_WORK(&priv->tx_timeout_task, hip04_tx_timeout_task);
+
+       ether_setup(ndev);
+       ndev->netdev_ops = &hip04_netdev_ops;
+       ndev->ethtool_ops = &hip04_ethtool_ops;
+       ndev->watchdog_timeo = TX_TIMEOUT;
+       ndev->priv_flags |= IFF_UNICAST_FLT;
+       ndev->irq = irq;
+       netif_napi_add(ndev, &priv->napi, hip04_rx_poll, NAPI_POLL_WEIGHT);
+       SET_NETDEV_DEV(ndev, &pdev->dev);
+
+       hip04_reset_ppe(priv);
+       if (priv->phy_mode == PHY_INTERFACE_MODE_MII)
+               hip04_config_port(ndev, SPEED_100, DUPLEX_FULL);
+
+       hip04_config_fifo(priv);
+       random_ether_addr(ndev->dev_addr);
+       hip04_update_mac_address(ndev);
+
+       ret = hip04_alloc_ring(ndev, d);
+       if (ret) {
+               netdev_err(ndev, "alloc ring fail\n");
+               goto alloc_fail;
+       }
+
+       ret = register_netdev(ndev);
+       if (ret) {
+               free_netdev(ndev);
+               goto alloc_fail;
+       }
+
+       return 0;
+
+alloc_fail:
+       hip04_free_ring(ndev, d);
+init_fail:
+       of_node_put(priv->phy_node);
+       free_netdev(ndev);
+       return ret;
+}
+
+static int hip04_remove(struct platform_device *pdev)
+{
+       struct net_device *ndev = platform_get_drvdata(pdev);
+       struct hip04_priv *priv = netdev_priv(ndev);
+       struct device *d = &pdev->dev;
+
+       if (priv->phy)
+               phy_disconnect(priv->phy);
+
+       hip04_free_ring(ndev, d);
+       unregister_netdev(ndev);
+       free_irq(ndev->irq, ndev);
+       of_node_put(priv->phy_node);
+       cancel_work_sync(&priv->tx_timeout_task);
+       free_netdev(ndev);
+
+       return 0;
+}
+
+static const struct of_device_id hip04_mac_match[] = {
+       { .compatible = "hisilicon,hip04-mac" },
+       { }
+};
+
+MODULE_DEVICE_TABLE(of, hip04_mac_match);
+
+static struct platform_driver hip04_mac_driver = {
+       .probe  = hip04_mac_probe,
+       .remove = hip04_remove,
+       .driver = {
+               .name           = DRV_NAME,
+               .owner          = THIS_MODULE,
+               .of_match_table = hip04_mac_match,
+       },
+};
+module_platform_driver(hip04_mac_driver);
+
+MODULE_DESCRIPTION("HISILICON P04 Ethernet driver");