net: bcmgenet: add main driver file
authorFlorian Fainelli <f.fainelli@gmail.com>
Fri, 14 Feb 2014 00:08:47 +0000 (16:08 -0800)
committerDavid S. Miller <davem@davemloft.net>
Fri, 14 Feb 2014 05:27:58 +0000 (00:27 -0500)
This patch adds the BCMGENET main driver file which supports the
following:

- GENET hardware from V1 to V4
- support for reading the UniMAC MIB counters statistics
- support for the 5 transmit queues
- support for RX/TX checksum offload and SG

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/broadcom/genet/bcmgenet.c [new file with mode: 0644]

diff --git a/drivers/net/ethernet/broadcom/genet/bcmgenet.c b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
new file mode 100644 (file)
index 0000000..0ebc297
--- /dev/null
@@ -0,0 +1,2595 @@
+/*
+ * Broadcom GENET (Gigabit Ethernet) controller driver
+ *
+ * Copyright (c) 2014 Broadcom Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ */
+
+#define pr_fmt(fmt)                            "bcmgenet: " fmt
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/sched.h>
+#include <linux/types.h>
+#include <linux/fcntl.h>
+#include <linux/interrupt.h>
+#include <linux/string.h>
+#include <linux/if_ether.h>
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <linux/delay.h>
+#include <linux/platform_device.h>
+#include <linux/dma-mapping.h>
+#include <linux/pm.h>
+#include <linux/clk.h>
+#include <linux/version.h>
+#include <linux/of.h>
+#include <linux/of_address.h>
+#include <linux/of_irq.h>
+#include <linux/of_net.h>
+#include <linux/of_platform.h>
+#include <net/arp.h>
+
+#include <linux/mii.h>
+#include <linux/ethtool.h>
+#include <linux/netdevice.h>
+#include <linux/inetdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/skbuff.h>
+#include <linux/in.h>
+#include <linux/ip.h>
+#include <linux/ipv6.h>
+#include <linux/phy.h>
+
+#include <asm/unaligned.h>
+
+#include "bcmgenet.h"
+
+/* Maximum number of hardware queues, downsized if needed */
+#define GENET_MAX_MQ_CNT       4
+
+/* Default highest priority queue for multi queue support */
+#define GENET_Q0_PRIORITY      0
+
+#define GENET_DEFAULT_BD_CNT   \
+       (TOTAL_DESC - priv->hw_params->tx_queues * priv->hw_params->bds_cnt)
+
+#define RX_BUF_LENGTH          2048
+#define SKB_ALIGNMENT          32
+
+/* Tx/Rx DMA register offset, skip 256 descriptors */
+#define WORDS_PER_BD(p)                (p->hw_params->words_per_bd)
+#define DMA_DESC_SIZE          (WORDS_PER_BD(priv) * sizeof(u32))
+
+#define GENET_TDMA_REG_OFF     (priv->hw_params->tdma_offset + \
+                               TOTAL_DESC * DMA_DESC_SIZE)
+
+#define GENET_RDMA_REG_OFF     (priv->hw_params->rdma_offset + \
+                               TOTAL_DESC * DMA_DESC_SIZE)
+
+static inline void dmadesc_set_length_status(struct bcmgenet_priv *priv,
+                                               void __iomem *d, u32 value)
+{
+       __raw_writel(value, d + DMA_DESC_LENGTH_STATUS);
+}
+
+static inline u32 dmadesc_get_length_status(struct bcmgenet_priv *priv,
+                                               void __iomem *d)
+{
+       return __raw_readl(d + DMA_DESC_LENGTH_STATUS);
+}
+
+static inline void dmadesc_set_addr(struct bcmgenet_priv *priv,
+                                   void __iomem *d,
+                                   dma_addr_t addr)
+{
+       __raw_writel(lower_32_bits(addr), d + DMA_DESC_ADDRESS_LO);
+
+       /* Register writes to GISB bus can take couple hundred nanoseconds
+        * and are done for each packet, save these expensive writes unless
+        * the platform is explicitely configured for 64-bits/LPAE.
+        */
+#ifdef CONFIG_PHYS_ADDR_T_64BIT
+       if (priv->hw_params->flags & GENET_HAS_40BITS)
+               __raw_writel(upper_32_bits(addr), d + DMA_DESC_ADDRESS_HI);
+#endif
+}
+
+/* Combined address + length/status setter */
+static inline void dmadesc_set(struct bcmgenet_priv *priv,
+                               void __iomem *d, dma_addr_t addr, u32 val)
+{
+       dmadesc_set_length_status(priv, d, val);
+       dmadesc_set_addr(priv, d, addr);
+}
+
+static inline dma_addr_t dmadesc_get_addr(struct bcmgenet_priv *priv,
+                                         void __iomem *d)
+{
+       dma_addr_t addr;
+
+       addr = __raw_readl(d + DMA_DESC_ADDRESS_LO);
+
+       /* Register writes to GISB bus can take couple hundred nanoseconds
+        * and are done for each packet, save these expensive writes unless
+        * the platform is explicitely configured for 64-bits/LPAE.
+        */
+#ifdef CONFIG_PHYS_ADDR_T_64BIT
+       if (priv->hw_params->flags & GENET_HAS_40BITS)
+               addr |= (u64)__raw_readl(d + DMA_DESC_ADDRESS_HI) << 32;
+#endif
+       return addr;
+}
+
+#define GENET_VER_FMT  "%1d.%1d EPHY: 0x%04x"
+
+#define GENET_MSG_DEFAULT      (NETIF_MSG_DRV | NETIF_MSG_PROBE | \
+                               NETIF_MSG_LINK)
+
+static inline u32 bcmgenet_rbuf_ctrl_get(struct bcmgenet_priv *priv)
+{
+       if (GENET_IS_V1(priv))
+               return bcmgenet_rbuf_readl(priv, RBUF_FLUSH_CTRL_V1);
+       else
+               return bcmgenet_sys_readl(priv, SYS_RBUF_FLUSH_CTRL);
+}
+
+static inline void bcmgenet_rbuf_ctrl_set(struct bcmgenet_priv *priv, u32 val)
+{
+       if (GENET_IS_V1(priv))
+               bcmgenet_rbuf_writel(priv, val, RBUF_FLUSH_CTRL_V1);
+       else
+               bcmgenet_sys_writel(priv, val, SYS_RBUF_FLUSH_CTRL);
+}
+
+/* These macros are defined to deal with register map change
+ * between GENET1.1 and GENET2. Only those currently being used
+ * by driver are defined.
+ */
+static inline u32 bcmgenet_tbuf_ctrl_get(struct bcmgenet_priv *priv)
+{
+       if (GENET_IS_V1(priv))
+               return bcmgenet_rbuf_readl(priv, TBUF_CTRL_V1);
+       else
+               return __raw_readl(priv->base +
+                               priv->hw_params->tbuf_offset + TBUF_CTRL);
+}
+
+static inline void bcmgenet_tbuf_ctrl_set(struct bcmgenet_priv *priv, u32 val)
+{
+       if (GENET_IS_V1(priv))
+               bcmgenet_rbuf_writel(priv, val, TBUF_CTRL_V1);
+       else
+               __raw_writel(val, priv->base +
+                               priv->hw_params->tbuf_offset + TBUF_CTRL);
+}
+
+static inline u32 bcmgenet_bp_mc_get(struct bcmgenet_priv *priv)
+{
+       if (GENET_IS_V1(priv))
+               return bcmgenet_rbuf_readl(priv, TBUF_BP_MC_V1);
+       else
+               return __raw_readl(priv->base +
+                               priv->hw_params->tbuf_offset + TBUF_BP_MC);
+}
+
+static inline void bcmgenet_bp_mc_set(struct bcmgenet_priv *priv, u32 val)
+{
+       if (GENET_IS_V1(priv))
+               bcmgenet_rbuf_writel(priv, val, TBUF_BP_MC_V1);
+       else
+               __raw_writel(val, priv->base +
+                               priv->hw_params->tbuf_offset + TBUF_BP_MC);
+}
+
+/* RX/TX DMA register accessors */
+enum dma_reg {
+       DMA_RING_CFG = 0,
+       DMA_CTRL,
+       DMA_STATUS,
+       DMA_SCB_BURST_SIZE,
+       DMA_ARB_CTRL,
+       DMA_PRIORITY,
+       DMA_RING_PRIORITY,
+};
+
+static const u8 bcmgenet_dma_regs_v3plus[] = {
+       [DMA_RING_CFG]          = 0x00,
+       [DMA_CTRL]              = 0x04,
+       [DMA_STATUS]            = 0x08,
+       [DMA_SCB_BURST_SIZE]    = 0x0C,
+       [DMA_ARB_CTRL]          = 0x2C,
+       [DMA_PRIORITY]          = 0x30,
+       [DMA_RING_PRIORITY]     = 0x38,
+};
+
+static const u8 bcmgenet_dma_regs_v2[] = {
+       [DMA_RING_CFG]          = 0x00,
+       [DMA_CTRL]              = 0x04,
+       [DMA_STATUS]            = 0x08,
+       [DMA_SCB_BURST_SIZE]    = 0x0C,
+       [DMA_ARB_CTRL]          = 0x30,
+       [DMA_PRIORITY]          = 0x34,
+       [DMA_RING_PRIORITY]     = 0x3C,
+};
+
+static const u8 bcmgenet_dma_regs_v1[] = {
+       [DMA_CTRL]              = 0x00,
+       [DMA_STATUS]            = 0x04,
+       [DMA_SCB_BURST_SIZE]    = 0x0C,
+       [DMA_ARB_CTRL]          = 0x30,
+       [DMA_PRIORITY]          = 0x34,
+       [DMA_RING_PRIORITY]     = 0x3C,
+};
+
+/* Set at runtime once bcmgenet version is known */
+static const u8 *bcmgenet_dma_regs;
+
+static inline struct bcmgenet_priv *dev_to_priv(struct device *dev)
+{
+       return netdev_priv(dev_get_drvdata(dev));
+}
+
+static inline u32 bcmgenet_tdma_readl(struct bcmgenet_priv *priv,
+                                       enum dma_reg r)
+{
+       return __raw_readl(priv->base + GENET_TDMA_REG_OFF +
+                       DMA_RINGS_SIZE + bcmgenet_dma_regs[r]);
+}
+
+static inline void bcmgenet_tdma_writel(struct bcmgenet_priv *priv,
+                                       u32 val, enum dma_reg r)
+{
+       __raw_writel(val, priv->base + GENET_TDMA_REG_OFF +
+                       DMA_RINGS_SIZE + bcmgenet_dma_regs[r]);
+}
+
+static inline u32 bcmgenet_rdma_readl(struct bcmgenet_priv *priv,
+                                       enum dma_reg r)
+{
+       return __raw_readl(priv->base + GENET_RDMA_REG_OFF +
+                       DMA_RINGS_SIZE + bcmgenet_dma_regs[r]);
+}
+
+static inline void bcmgenet_rdma_writel(struct bcmgenet_priv *priv,
+                                       u32 val, enum dma_reg r)
+{
+       __raw_writel(val, priv->base + GENET_RDMA_REG_OFF +
+                       DMA_RINGS_SIZE + bcmgenet_dma_regs[r]);
+}
+
+/* RDMA/TDMA ring registers and accessors
+ * we merge the common fields and just prefix with T/D the registers
+ * having different meaning depending on the direction
+ */
+enum dma_ring_reg {
+       TDMA_READ_PTR = 0,
+       RDMA_WRITE_PTR = TDMA_READ_PTR,
+       TDMA_READ_PTR_HI,
+       RDMA_WRITE_PTR_HI = TDMA_READ_PTR_HI,
+       TDMA_CONS_INDEX,
+       RDMA_PROD_INDEX = TDMA_CONS_INDEX,
+       TDMA_PROD_INDEX,
+       RDMA_CONS_INDEX = TDMA_PROD_INDEX,
+       DMA_RING_BUF_SIZE,
+       DMA_START_ADDR,
+       DMA_START_ADDR_HI,
+       DMA_END_ADDR,
+       DMA_END_ADDR_HI,
+       DMA_MBUF_DONE_THRESH,
+       TDMA_FLOW_PERIOD,
+       RDMA_XON_XOFF_THRESH = TDMA_FLOW_PERIOD,
+       TDMA_WRITE_PTR,
+       RDMA_READ_PTR = TDMA_WRITE_PTR,
+       TDMA_WRITE_PTR_HI,
+       RDMA_READ_PTR_HI = TDMA_WRITE_PTR_HI
+};
+
+/* GENET v4 supports 40-bits pointer addressing
+ * for obvious reasons the LO and HI word parts
+ * are contiguous, but this offsets the other
+ * registers.
+ */
+static const u8 genet_dma_ring_regs_v4[] = {
+       [TDMA_READ_PTR]                 = 0x00,
+       [TDMA_READ_PTR_HI]              = 0x04,
+       [TDMA_CONS_INDEX]               = 0x08,
+       [TDMA_PROD_INDEX]               = 0x0C,
+       [DMA_RING_BUF_SIZE]             = 0x10,
+       [DMA_START_ADDR]                = 0x14,
+       [DMA_START_ADDR_HI]             = 0x18,
+       [DMA_END_ADDR]                  = 0x1C,
+       [DMA_END_ADDR_HI]               = 0x20,
+       [DMA_MBUF_DONE_THRESH]          = 0x24,
+       [TDMA_FLOW_PERIOD]              = 0x28,
+       [TDMA_WRITE_PTR]                = 0x2C,
+       [TDMA_WRITE_PTR_HI]             = 0x30,
+};
+
+static const u8 genet_dma_ring_regs_v123[] = {
+       [TDMA_READ_PTR]                 = 0x00,
+       [TDMA_CONS_INDEX]               = 0x04,
+       [TDMA_PROD_INDEX]               = 0x08,
+       [DMA_RING_BUF_SIZE]             = 0x0C,
+       [DMA_START_ADDR]                = 0x10,
+       [DMA_END_ADDR]                  = 0x14,
+       [DMA_MBUF_DONE_THRESH]          = 0x18,
+       [TDMA_FLOW_PERIOD]              = 0x1C,
+       [TDMA_WRITE_PTR]                = 0x20,
+};
+
+/* Set at runtime once GENET version is known */
+static const u8 *genet_dma_ring_regs;
+
+static inline u32 bcmgenet_tdma_ring_readl(struct bcmgenet_priv *priv,
+                                               unsigned int ring,
+                                               enum dma_ring_reg r)
+{
+       return __raw_readl(priv->base + GENET_TDMA_REG_OFF +
+                       (DMA_RING_SIZE * ring) +
+                       genet_dma_ring_regs[r]);
+}
+
+static inline void bcmgenet_tdma_ring_writel(struct bcmgenet_priv *priv,
+                                               unsigned int ring,
+                                               u32 val,
+                                               enum dma_ring_reg r)
+{
+       __raw_writel(val, priv->base + GENET_TDMA_REG_OFF +
+                       (DMA_RING_SIZE * ring) +
+                       genet_dma_ring_regs[r]);
+}
+
+static inline u32 bcmgenet_rdma_ring_readl(struct bcmgenet_priv *priv,
+                                               unsigned int ring,
+                                               enum dma_ring_reg r)
+{
+       return __raw_readl(priv->base + GENET_RDMA_REG_OFF +
+                       (DMA_RING_SIZE * ring) +
+                       genet_dma_ring_regs[r]);
+}
+
+static inline void bcmgenet_rdma_ring_writel(struct bcmgenet_priv *priv,
+                                               unsigned int ring,
+                                               u32 val,
+                                               enum dma_ring_reg r)
+{
+       __raw_writel(val, priv->base + GENET_RDMA_REG_OFF +
+                       (DMA_RING_SIZE * ring) +
+                       genet_dma_ring_regs[r]);
+}
+
+static int bcmgenet_get_settings(struct net_device *dev,
+               struct ethtool_cmd *cmd)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+
+       if (!netif_running(dev))
+               return -EINVAL;
+
+       if (!priv->phydev)
+               return -ENODEV;
+
+       return phy_ethtool_gset(priv->phydev, cmd);
+}
+
+static int bcmgenet_set_settings(struct net_device *dev,
+               struct ethtool_cmd *cmd)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+
+       if (!netif_running(dev))
+               return -EINVAL;
+
+       if (!priv->phydev)
+               return -ENODEV;
+
+       return phy_ethtool_sset(priv->phydev, cmd);
+}
+
+static int bcmgenet_set_rx_csum(struct net_device *dev,
+                               netdev_features_t wanted)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       u32 rbuf_chk_ctrl;
+       bool rx_csum_en;
+
+       rx_csum_en = !!(wanted & NETIF_F_RXCSUM);
+
+       rbuf_chk_ctrl = bcmgenet_rbuf_readl(priv, RBUF_CHK_CTRL);
+
+       /* enable rx checksumming */
+       if (rx_csum_en)
+               rbuf_chk_ctrl |= RBUF_RXCHK_EN;
+       else
+               rbuf_chk_ctrl &= ~RBUF_RXCHK_EN;
+       priv->desc_rxchk_en = rx_csum_en;
+       bcmgenet_rbuf_writel(priv, rbuf_chk_ctrl, RBUF_CHK_CTRL);
+
+       return 0;
+}
+
+static int bcmgenet_set_tx_csum(struct net_device *dev,
+                               netdev_features_t wanted)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       bool desc_64b_en;
+       u32 tbuf_ctrl, rbuf_ctrl;
+
+       tbuf_ctrl = bcmgenet_tbuf_ctrl_get(priv);
+       rbuf_ctrl = bcmgenet_rbuf_readl(priv, RBUF_CTRL);
+
+       desc_64b_en = !!(wanted & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM));
+
+       /* enable 64 bytes descriptor in both directions (RBUF and TBUF) */
+       if (desc_64b_en) {
+               tbuf_ctrl |= RBUF_64B_EN;
+               rbuf_ctrl |= RBUF_64B_EN;
+       } else {
+               tbuf_ctrl &= ~RBUF_64B_EN;
+               rbuf_ctrl &= ~RBUF_64B_EN;
+       }
+       priv->desc_64b_en = desc_64b_en;
+
+       bcmgenet_tbuf_ctrl_set(priv, tbuf_ctrl);
+       bcmgenet_rbuf_writel(priv, rbuf_ctrl, RBUF_CTRL);
+
+       return 0;
+}
+
+static int bcmgenet_set_features(struct net_device *dev,
+               netdev_features_t features)
+{
+       netdev_features_t changed = features ^ dev->features;
+       netdev_features_t wanted = dev->wanted_features;
+       int ret = 0;
+
+       if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
+               ret = bcmgenet_set_tx_csum(dev, wanted);
+       if (changed & (NETIF_F_RXCSUM))
+               ret = bcmgenet_set_rx_csum(dev, wanted);
+
+       return ret;
+}
+
+static u32 bcmgenet_get_msglevel(struct net_device *dev)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+
+       return priv->msg_enable;
+}
+
+static void bcmgenet_set_msglevel(struct net_device *dev, u32 level)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+
+       priv->msg_enable = level;
+}
+
+/* standard ethtool support functions. */
+enum bcmgenet_stat_type {
+       BCMGENET_STAT_NETDEV = -1,
+       BCMGENET_STAT_MIB_RX,
+       BCMGENET_STAT_MIB_TX,
+       BCMGENET_STAT_RUNT,
+       BCMGENET_STAT_MISC,
+};
+
+struct bcmgenet_stats {
+       char stat_string[ETH_GSTRING_LEN];
+       int stat_sizeof;
+       int stat_offset;
+       enum bcmgenet_stat_type type;
+       /* reg offset from UMAC base for misc counters */
+       u16 reg_offset;
+};
+
+#define STAT_NETDEV(m) { \
+       .stat_string = __stringify(m), \
+       .stat_sizeof = sizeof(((struct net_device_stats *)0)->m), \
+       .stat_offset = offsetof(struct net_device_stats, m), \
+       .type = BCMGENET_STAT_NETDEV, \
+}
+
+#define STAT_GENET_MIB(str, m, _type) { \
+       .stat_string = str, \
+       .stat_sizeof = sizeof(((struct bcmgenet_priv *)0)->m), \
+       .stat_offset = offsetof(struct bcmgenet_priv, m), \
+       .type = _type, \
+}
+
+#define STAT_GENET_MIB_RX(str, m) STAT_GENET_MIB(str, m, BCMGENET_STAT_MIB_RX)
+#define STAT_GENET_MIB_TX(str, m) STAT_GENET_MIB(str, m, BCMGENET_STAT_MIB_TX)
+#define STAT_GENET_RUNT(str, m) STAT_GENET_MIB(str, m, BCMGENET_STAT_RUNT)
+
+#define STAT_GENET_MISC(str, m, offset) { \
+       .stat_string = str, \
+       .stat_sizeof = sizeof(((struct bcmgenet_priv *)0)->m), \
+       .stat_offset = offsetof(struct bcmgenet_priv, m), \
+       .type = BCMGENET_STAT_MISC, \
+       .reg_offset = offset, \
+}
+
+
+/* There is a 0xC gap between the end of RX and beginning of TX stats and then
+ * between the end of TX stats and the beginning of the RX RUNT
+ */
+#define BCMGENET_STAT_OFFSET   0xc
+
+/* Hardware counters must be kept in sync because the order/offset
+ * is important here (order in structure declaration = order in hardware)
+ */
+static const struct bcmgenet_stats bcmgenet_gstrings_stats[] = {
+       /* general stats */
+       STAT_NETDEV(rx_packets),
+       STAT_NETDEV(tx_packets),
+       STAT_NETDEV(rx_bytes),
+       STAT_NETDEV(tx_bytes),
+       STAT_NETDEV(rx_errors),
+       STAT_NETDEV(tx_errors),
+       STAT_NETDEV(rx_dropped),
+       STAT_NETDEV(tx_dropped),
+       STAT_NETDEV(multicast),
+       /* UniMAC RSV counters */
+       STAT_GENET_MIB_RX("rx_64_octets", mib.rx.pkt_cnt.cnt_64),
+       STAT_GENET_MIB_RX("rx_65_127_oct", mib.rx.pkt_cnt.cnt_127),
+       STAT_GENET_MIB_RX("rx_128_255_oct", mib.rx.pkt_cnt.cnt_255),
+       STAT_GENET_MIB_RX("rx_256_511_oct", mib.rx.pkt_cnt.cnt_511),
+       STAT_GENET_MIB_RX("rx_512_1023_oct", mib.rx.pkt_cnt.cnt_1023),
+       STAT_GENET_MIB_RX("rx_1024_1518_oct", mib.rx.pkt_cnt.cnt_1518),
+       STAT_GENET_MIB_RX("rx_vlan_1519_1522_oct", mib.rx.pkt_cnt.cnt_mgv),
+       STAT_GENET_MIB_RX("rx_1522_2047_oct", mib.rx.pkt_cnt.cnt_2047),
+       STAT_GENET_MIB_RX("rx_2048_4095_oct", mib.rx.pkt_cnt.cnt_4095),
+       STAT_GENET_MIB_RX("rx_4096_9216_oct", mib.rx.pkt_cnt.cnt_9216),
+       STAT_GENET_MIB_RX("rx_pkts", mib.rx.pkt),
+       STAT_GENET_MIB_RX("rx_bytes", mib.rx.bytes),
+       STAT_GENET_MIB_RX("rx_multicast", mib.rx.mca),
+       STAT_GENET_MIB_RX("rx_broadcast", mib.rx.bca),
+       STAT_GENET_MIB_RX("rx_fcs", mib.rx.fcs),
+       STAT_GENET_MIB_RX("rx_control", mib.rx.cf),
+       STAT_GENET_MIB_RX("rx_pause", mib.rx.pf),
+       STAT_GENET_MIB_RX("rx_unknown", mib.rx.uo),
+       STAT_GENET_MIB_RX("rx_align", mib.rx.aln),
+       STAT_GENET_MIB_RX("rx_outrange", mib.rx.flr),
+       STAT_GENET_MIB_RX("rx_code", mib.rx.cde),
+       STAT_GENET_MIB_RX("rx_carrier", mib.rx.fcr),
+       STAT_GENET_MIB_RX("rx_oversize", mib.rx.ovr),
+       STAT_GENET_MIB_RX("rx_jabber", mib.rx.jbr),
+       STAT_GENET_MIB_RX("rx_mtu_err", mib.rx.mtue),
+       STAT_GENET_MIB_RX("rx_good_pkts", mib.rx.pok),
+       STAT_GENET_MIB_RX("rx_unicast", mib.rx.uc),
+       STAT_GENET_MIB_RX("rx_ppp", mib.rx.ppp),
+       STAT_GENET_MIB_RX("rx_crc", mib.rx.rcrc),
+       /* UniMAC TSV counters */
+       STAT_GENET_MIB_TX("tx_64_octets", mib.tx.pkt_cnt.cnt_64),
+       STAT_GENET_MIB_TX("tx_65_127_oct", mib.tx.pkt_cnt.cnt_127),
+       STAT_GENET_MIB_TX("tx_128_255_oct", mib.tx.pkt_cnt.cnt_255),
+       STAT_GENET_MIB_TX("tx_256_511_oct", mib.tx.pkt_cnt.cnt_511),
+       STAT_GENET_MIB_TX("tx_512_1023_oct", mib.tx.pkt_cnt.cnt_1023),
+       STAT_GENET_MIB_TX("tx_1024_1518_oct", mib.tx.pkt_cnt.cnt_1518),
+       STAT_GENET_MIB_TX("tx_vlan_1519_1522_oct", mib.tx.pkt_cnt.cnt_mgv),
+       STAT_GENET_MIB_TX("tx_1522_2047_oct", mib.tx.pkt_cnt.cnt_2047),
+       STAT_GENET_MIB_TX("tx_2048_4095_oct", mib.tx.pkt_cnt.cnt_4095),
+       STAT_GENET_MIB_TX("tx_4096_9216_oct", mib.tx.pkt_cnt.cnt_9216),
+       STAT_GENET_MIB_TX("tx_pkts", mib.tx.pkts),
+       STAT_GENET_MIB_TX("tx_multicast", mib.tx.mca),
+       STAT_GENET_MIB_TX("tx_broadcast", mib.tx.bca),
+       STAT_GENET_MIB_TX("tx_pause", mib.tx.pf),
+       STAT_GENET_MIB_TX("tx_control", mib.tx.cf),
+       STAT_GENET_MIB_TX("tx_fcs_err", mib.tx.fcs),
+       STAT_GENET_MIB_TX("tx_oversize", mib.tx.ovr),
+       STAT_GENET_MIB_TX("tx_defer", mib.tx.drf),
+       STAT_GENET_MIB_TX("tx_excess_defer", mib.tx.edf),
+       STAT_GENET_MIB_TX("tx_single_col", mib.tx.scl),
+       STAT_GENET_MIB_TX("tx_multi_col", mib.tx.mcl),
+       STAT_GENET_MIB_TX("tx_late_col", mib.tx.lcl),
+       STAT_GENET_MIB_TX("tx_excess_col", mib.tx.ecl),
+       STAT_GENET_MIB_TX("tx_frags", mib.tx.frg),
+       STAT_GENET_MIB_TX("tx_total_col", mib.tx.ncl),
+       STAT_GENET_MIB_TX("tx_jabber", mib.tx.jbr),
+       STAT_GENET_MIB_TX("tx_bytes", mib.tx.bytes),
+       STAT_GENET_MIB_TX("tx_good_pkts", mib.tx.pok),
+       STAT_GENET_MIB_TX("tx_unicast", mib.tx.uc),
+       /* UniMAC RUNT counters */
+       STAT_GENET_RUNT("rx_runt_pkts", mib.rx_runt_cnt),
+       STAT_GENET_RUNT("rx_runt_valid_fcs", mib.rx_runt_fcs),
+       STAT_GENET_RUNT("rx_runt_inval_fcs_align", mib.rx_runt_fcs_align),
+       STAT_GENET_RUNT("rx_runt_bytes", mib.rx_runt_bytes),
+       /* Misc UniMAC counters */
+       STAT_GENET_MISC("rbuf_ovflow_cnt", mib.rbuf_ovflow_cnt,
+                       UMAC_RBUF_OVFL_CNT),
+       STAT_GENET_MISC("rbuf_err_cnt", mib.rbuf_err_cnt, UMAC_RBUF_ERR_CNT),
+       STAT_GENET_MISC("mdf_err_cnt", mib.mdf_err_cnt, UMAC_MDF_ERR_CNT),
+};
+
+#define BCMGENET_STATS_LEN     ARRAY_SIZE(bcmgenet_gstrings_stats)
+
+static void bcmgenet_get_drvinfo(struct net_device *dev,
+               struct ethtool_drvinfo *info)
+{
+       strlcpy(info->driver, "bcmgenet", sizeof(info->driver));
+       strlcpy(info->version, "v2.0", sizeof(info->version));
+       info->n_stats = BCMGENET_STATS_LEN;
+
+}
+
+static int bcmgenet_get_sset_count(struct net_device *dev, int string_set)
+{
+       switch (string_set) {
+       case ETH_SS_STATS:
+               return BCMGENET_STATS_LEN;
+       default:
+               return -EOPNOTSUPP;
+       }
+}
+
+static void bcmgenet_get_strings(struct net_device *dev,
+                               u32 stringset, u8 *data)
+{
+       int i;
+
+       switch (stringset) {
+       case ETH_SS_STATS:
+               for (i = 0; i < BCMGENET_STATS_LEN; i++) {
+                       memcpy(data + i * ETH_GSTRING_LEN,
+                               bcmgenet_gstrings_stats[i].stat_string,
+                               ETH_GSTRING_LEN);
+               }
+               break;
+       }
+}
+
+static void bcmgenet_update_mib_counters(struct bcmgenet_priv *priv)
+{
+       int i, j = 0;
+
+       for (i = 0; i < BCMGENET_STATS_LEN; i++) {
+               const struct bcmgenet_stats *s;
+               u8 offset = 0;
+               u32 val = 0;
+               char *p;
+
+               s = &bcmgenet_gstrings_stats[i];
+               switch (s->type) {
+               case BCMGENET_STAT_NETDEV:
+                       continue;
+               case BCMGENET_STAT_MIB_RX:
+               case BCMGENET_STAT_MIB_TX:
+               case BCMGENET_STAT_RUNT:
+                       if (s->type != BCMGENET_STAT_MIB_RX)
+                               offset = BCMGENET_STAT_OFFSET;
+                       val = bcmgenet_umac_readl(priv, UMAC_MIB_START +
+                                                               j + offset);
+                       break;
+               case BCMGENET_STAT_MISC:
+                       val = bcmgenet_umac_readl(priv, s->reg_offset);
+                       /* clear if overflowed */
+                       if (val == ~0)
+                               bcmgenet_umac_writel(priv, 0, s->reg_offset);
+                       break;
+               }
+
+               j += s->stat_sizeof;
+               p = (char *)priv + s->stat_offset;
+               *(u32 *)p = val;
+       }
+}
+
+static void bcmgenet_get_ethtool_stats(struct net_device *dev,
+                                       struct ethtool_stats *stats,
+                                       u64 *data)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       int i;
+
+       if (netif_running(dev))
+               bcmgenet_update_mib_counters(priv);
+
+       for (i = 0; i < BCMGENET_STATS_LEN; i++) {
+               const struct bcmgenet_stats *s;
+               char *p;
+
+               s = &bcmgenet_gstrings_stats[i];
+               if (s->type == BCMGENET_STAT_NETDEV)
+                       p = (char *)&dev->stats;
+               else
+                       p = (char *)priv;
+               p += s->stat_offset;
+               data[i] = *(u32 *)p;
+       }
+}
+
+/* standard ethtool support functions. */
+static struct ethtool_ops bcmgenet_ethtool_ops = {
+       .get_strings            = bcmgenet_get_strings,
+       .get_sset_count         = bcmgenet_get_sset_count,
+       .get_ethtool_stats      = bcmgenet_get_ethtool_stats,
+       .get_settings           = bcmgenet_get_settings,
+       .set_settings           = bcmgenet_set_settings,
+       .get_drvinfo            = bcmgenet_get_drvinfo,
+       .get_link               = ethtool_op_get_link,
+       .get_msglevel           = bcmgenet_get_msglevel,
+       .set_msglevel           = bcmgenet_set_msglevel,
+};
+
+/* Power down the unimac, based on mode. */
+static void bcmgenet_power_down(struct bcmgenet_priv *priv,
+                               enum bcmgenet_power_mode mode)
+{
+       u32 reg;
+
+       switch (mode) {
+       case GENET_POWER_CABLE_SENSE:
+               if (priv->phydev)
+                       phy_detach(priv->phydev);
+               break;
+
+       case GENET_POWER_PASSIVE:
+               /* Power down LED */
+               bcmgenet_mii_reset(priv->dev);
+               if (priv->hw_params->flags & GENET_HAS_EXT) {
+                       reg = bcmgenet_ext_readl(priv, EXT_EXT_PWR_MGMT);
+                       reg |= (EXT_PWR_DOWN_PHY |
+                               EXT_PWR_DOWN_DLL | EXT_PWR_DOWN_BIAS);
+                       bcmgenet_ext_writel(priv, reg, EXT_EXT_PWR_MGMT);
+               }
+               break;
+       default:
+               break;
+       }
+}
+
+static void bcmgenet_power_up(struct bcmgenet_priv *priv,
+                               enum bcmgenet_power_mode mode)
+{
+       u32 reg;
+
+       if (!(priv->hw_params->flags & GENET_HAS_EXT))
+               return;
+
+       reg = bcmgenet_ext_readl(priv, EXT_EXT_PWR_MGMT);
+
+       switch (mode) {
+       case GENET_POWER_PASSIVE:
+               reg &= ~(EXT_PWR_DOWN_DLL | EXT_PWR_DOWN_PHY |
+                               EXT_PWR_DOWN_BIAS);
+               /* fallthrough */
+       case GENET_POWER_CABLE_SENSE:
+               /* enable APD */
+               reg |= EXT_PWR_DN_EN_LD;
+               break;
+       default:
+               break;
+       }
+
+       bcmgenet_ext_writel(priv, reg, EXT_EXT_PWR_MGMT);
+       bcmgenet_mii_reset(priv->dev);
+}
+
+/* ioctl handle special commands that are not present in ethtool. */
+static int bcmgenet_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       int val = 0;
+
+       if (!netif_running(dev))
+               return -EINVAL;
+
+       switch (cmd) {
+       case SIOCGMIIPHY:
+       case SIOCGMIIREG:
+       case SIOCSMIIREG:
+               if (!priv->phydev)
+                       val = -ENODEV;
+               else
+                       val = phy_mii_ioctl(priv->phydev, rq, cmd);
+               break;
+
+       default:
+               val = -EINVAL;
+               break;
+       }
+
+       return val;
+}
+
+static struct enet_cb *bcmgenet_get_txcb(struct bcmgenet_priv *priv,
+                                        struct bcmgenet_tx_ring *ring)
+{
+       struct enet_cb *tx_cb_ptr;
+
+       tx_cb_ptr = ring->cbs;
+       tx_cb_ptr += ring->write_ptr - ring->cb_ptr;
+       tx_cb_ptr->bd_addr = priv->tx_bds + ring->write_ptr * DMA_DESC_SIZE;
+       /* Advancing local write pointer */
+       if (ring->write_ptr == ring->end_ptr)
+               ring->write_ptr = ring->cb_ptr;
+       else
+               ring->write_ptr++;
+
+       return tx_cb_ptr;
+}
+
+/* Simple helper to free a control block's resources */
+static void bcmgenet_free_cb(struct enet_cb *cb)
+{
+       dev_kfree_skb_any(cb->skb);
+       cb->skb = NULL;
+       dma_unmap_addr_set(cb, dma_addr, 0);
+}
+
+static inline void bcmgenet_tx_ring16_int_disable(struct bcmgenet_priv *priv,
+                                                 struct bcmgenet_tx_ring *ring)
+{
+       bcmgenet_intrl2_0_writel(priv,
+                       UMAC_IRQ_TXDMA_BDONE | UMAC_IRQ_TXDMA_PDONE,
+                       INTRL2_CPU_MASK_SET);
+}
+
+static inline void bcmgenet_tx_ring16_int_enable(struct bcmgenet_priv *priv,
+                                                struct bcmgenet_tx_ring *ring)
+{
+       bcmgenet_intrl2_0_writel(priv,
+                       UMAC_IRQ_TXDMA_BDONE | UMAC_IRQ_TXDMA_PDONE,
+                       INTRL2_CPU_MASK_CLEAR);
+}
+
+static inline void bcmgenet_tx_ring_int_enable(struct bcmgenet_priv *priv,
+                                               struct bcmgenet_tx_ring *ring)
+{
+       bcmgenet_intrl2_1_writel(priv,
+                       (1 << ring->index), INTRL2_CPU_MASK_CLEAR);
+       priv->int1_mask &= ~(1 << ring->index);
+}
+
+static inline void bcmgenet_tx_ring_int_disable(struct bcmgenet_priv *priv,
+                                               struct bcmgenet_tx_ring *ring)
+{
+       bcmgenet_intrl2_1_writel(priv,
+                       (1 << ring->index), INTRL2_CPU_MASK_SET);
+       priv->int1_mask |= (1 << ring->index);
+}
+
+/* Unlocked version of the reclaim routine */
+static void __bcmgenet_tx_reclaim(struct net_device *dev,
+                               struct bcmgenet_tx_ring *ring)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       int last_tx_cn, last_c_index, num_tx_bds;
+       struct enet_cb *tx_cb_ptr;
+       unsigned int c_index;
+
+       /* Compute how many buffers are transmited since last xmit call */
+       c_index = bcmgenet_tdma_ring_readl(priv, ring->index, TDMA_CONS_INDEX);
+
+       last_c_index = ring->c_index;
+       num_tx_bds = ring->size;
+
+       c_index &= (num_tx_bds - 1);
+
+       if (c_index >= last_c_index)
+               last_tx_cn = c_index - last_c_index;
+       else
+               last_tx_cn = num_tx_bds - last_c_index + c_index;
+
+       netif_dbg(priv, tx_done, dev,
+                       "%s ring=%d index=%d last_tx_cn=%d last_index=%d\n",
+                       __func__, ring->index,
+                       c_index, last_tx_cn, last_c_index);
+
+       /* Reclaim transmitted buffers */
+       while (last_tx_cn-- > 0) {
+               tx_cb_ptr = ring->cbs + last_c_index;
+               if (tx_cb_ptr->skb) {
+                       dev->stats.tx_bytes += tx_cb_ptr->skb->len;
+                       dma_unmap_single(&dev->dev,
+                                       dma_unmap_addr(tx_cb_ptr, dma_addr),
+                                       tx_cb_ptr->skb->len,
+                                       DMA_TO_DEVICE);
+                       bcmgenet_free_cb(tx_cb_ptr);
+               } else if (dma_unmap_addr(tx_cb_ptr, dma_addr)) {
+                       dev->stats.tx_bytes +=
+                               dma_unmap_len(tx_cb_ptr, dma_len);
+                       dma_unmap_page(&dev->dev,
+                                       dma_unmap_addr(tx_cb_ptr, dma_addr),
+                                       dma_unmap_len(tx_cb_ptr, dma_len),
+                                       DMA_TO_DEVICE);
+                       dma_unmap_addr_set(tx_cb_ptr, dma_addr, 0);
+               }
+               dev->stats.tx_packets++;
+               ring->free_bds += 1;
+
+               last_c_index++;
+               last_c_index &= (num_tx_bds - 1);
+       }
+
+       if (ring->free_bds > (MAX_SKB_FRAGS + 1))
+               ring->int_disable(priv, ring);
+
+       if (__netif_subqueue_stopped(dev, ring->queue))
+               netif_wake_subqueue(dev, ring->queue);
+
+       ring->c_index = c_index;
+}
+
+static void bcmgenet_tx_reclaim(struct net_device *dev,
+               struct bcmgenet_tx_ring *ring)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&ring->lock, flags);
+       __bcmgenet_tx_reclaim(dev, ring);
+       spin_unlock_irqrestore(&ring->lock, flags);
+}
+
+static void bcmgenet_tx_reclaim_all(struct net_device *dev)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       int i;
+
+       if (netif_is_multiqueue(dev)) {
+               for (i = 0; i < priv->hw_params->tx_queues; i++)
+                       bcmgenet_tx_reclaim(dev, &priv->tx_rings[i]);
+       }
+
+       bcmgenet_tx_reclaim(dev, &priv->tx_rings[DESC_INDEX]);
+}
+
+/* Transmits a single SKB (either head of a fragment or a single SKB)
+ * caller must hold priv->lock
+ */
+static int bcmgenet_xmit_single(struct net_device *dev,
+                               struct sk_buff *skb,
+                               u16 dma_desc_flags,
+                               struct bcmgenet_tx_ring *ring)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       struct device *kdev = &priv->pdev->dev;
+       struct enet_cb *tx_cb_ptr;
+       unsigned int skb_len;
+       dma_addr_t mapping;
+       u32 length_status;
+       int ret;
+
+       tx_cb_ptr = bcmgenet_get_txcb(priv, ring);
+
+       if (unlikely(!tx_cb_ptr))
+               BUG();
+
+       tx_cb_ptr->skb = skb;
+
+       skb_len = skb_headlen(skb) < ETH_ZLEN ? ETH_ZLEN : skb_headlen(skb);
+
+       mapping = dma_map_single(kdev, skb->data, skb_len, DMA_TO_DEVICE);
+       ret = dma_mapping_error(kdev, mapping);
+       if (ret) {
+               netif_err(priv, tx_err, dev, "Tx DMA map failed\n");
+               dev_kfree_skb(skb);
+               return ret;
+       }
+
+       dma_unmap_addr_set(tx_cb_ptr, dma_addr, mapping);
+       dma_unmap_len_set(tx_cb_ptr, dma_len, skb->len);
+       length_status = (skb_len << DMA_BUFLENGTH_SHIFT) | dma_desc_flags |
+                       (priv->hw_params->qtag_mask << DMA_TX_QTAG_SHIFT) |
+                       DMA_TX_APPEND_CRC;
+
+       if (skb->ip_summed == CHECKSUM_PARTIAL)
+               length_status |= DMA_TX_DO_CSUM;
+
+       dmadesc_set(priv, tx_cb_ptr->bd_addr, mapping, length_status);
+
+       /* Decrement total BD count and advance our write pointer */
+       ring->free_bds -= 1;
+       ring->prod_index += 1;
+       ring->prod_index &= DMA_P_INDEX_MASK;
+
+       return 0;
+}
+
+/* Transmit a SKB fragement */
+static int bcmgenet_xmit_frag(struct net_device *dev,
+                               skb_frag_t *frag,
+                               u16 dma_desc_flags,
+                               struct bcmgenet_tx_ring *ring)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       struct device *kdev = &priv->pdev->dev;
+       struct enet_cb *tx_cb_ptr;
+       dma_addr_t mapping;
+       int ret;
+
+       tx_cb_ptr = bcmgenet_get_txcb(priv, ring);
+
+       if (unlikely(!tx_cb_ptr))
+               BUG();
+       tx_cb_ptr->skb = NULL;
+
+       mapping = skb_frag_dma_map(kdev, frag, 0,
+               skb_frag_size(frag), DMA_TO_DEVICE);
+       ret = dma_mapping_error(kdev, mapping);
+       if (ret) {
+               netif_err(priv, tx_err, dev, "%s: Tx DMA map failed\n",
+                               __func__);
+               return ret;
+       }
+
+       dma_unmap_addr_set(tx_cb_ptr, dma_addr, mapping);
+       dma_unmap_len_set(tx_cb_ptr, dma_len, frag->size);
+
+       dmadesc_set(priv, tx_cb_ptr->bd_addr, mapping,
+                       (frag->size << DMA_BUFLENGTH_SHIFT) | dma_desc_flags |
+                       (priv->hw_params->qtag_mask << DMA_TX_QTAG_SHIFT));
+
+
+       ring->free_bds -= 1;
+       ring->prod_index += 1;
+       ring->prod_index &= DMA_P_INDEX_MASK;
+
+       return 0;
+}
+
+/* Reallocate the SKB to put enough headroom in front of it and insert
+ * the transmit checksum offsets in the descriptors
+ */
+static int bcmgenet_put_tx_csum(struct net_device *dev, struct sk_buff *skb)
+{
+       struct status_64 *status = NULL;
+       struct sk_buff *new_skb;
+       u16 offset;
+       u8 ip_proto;
+       u16 ip_ver;
+       u32 tx_csum_info;
+
+       if (unlikely(skb_headroom(skb) < sizeof(*status))) {
+               /* If 64 byte status block enabled, must make sure skb has
+                * enough headroom for us to insert 64B status block.
+                */
+               new_skb = skb_realloc_headroom(skb, sizeof(*status));
+               dev_kfree_skb(skb);
+               if (!new_skb) {
+                       dev->stats.tx_errors++;
+                       dev->stats.tx_dropped++;
+                       return -ENOMEM;
+               }
+               skb = new_skb;
+       }
+
+       skb_push(skb, sizeof(*status));
+       status = (struct status_64 *)skb->data;
+
+       if (skb->ip_summed  == CHECKSUM_PARTIAL) {
+               ip_ver = htons(skb->protocol);
+               switch (ip_ver) {
+               case ETH_P_IP:
+                       ip_proto = ip_hdr(skb)->protocol;
+                       break;
+               case ETH_P_IPV6:
+                       ip_proto = ipv6_hdr(skb)->nexthdr;
+                       break;
+               default:
+                       return 0;
+               }
+
+               offset = skb_checksum_start_offset(skb) - sizeof(*status);
+               tx_csum_info = (offset << STATUS_TX_CSUM_START_SHIFT) |
+                               (offset + skb->csum_offset);
+
+               /* Set the length valid bit for TCP and UDP and just set
+                * the special UDP flag for IPv4, else just set to 0.
+                */
+               if (ip_proto == IPPROTO_TCP || ip_proto == IPPROTO_UDP) {
+                       tx_csum_info |= STATUS_TX_CSUM_LV;
+                       if (ip_proto == IPPROTO_UDP && ip_ver == ETH_P_IP)
+                               tx_csum_info |= STATUS_TX_CSUM_PROTO_UDP;
+               } else
+                       tx_csum_info = 0;
+
+               status->tx_csum_info = tx_csum_info;
+       }
+
+       return 0;
+}
+
+static netdev_tx_t bcmgenet_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       struct bcmgenet_tx_ring *ring = NULL;
+       unsigned long flags = 0;
+       int nr_frags, index;
+       u16 dma_desc_flags;
+       int ret;
+       int i;
+
+       index = skb_get_queue_mapping(skb);
+       /* Mapping strategy:
+        * queue_mapping = 0, unclassified, packet xmited through ring16
+        * queue_mapping = 1, goes to ring 0. (highest priority queue
+        * queue_mapping = 2, goes to ring 1.
+        * queue_mapping = 3, goes to ring 2.
+        * queue_mapping = 4, goes to ring 3.
+        */
+       if (index == 0)
+               index = DESC_INDEX;
+       else
+               index -= 1;
+
+       if ((index != DESC_INDEX) && (index > priv->hw_params->tx_queues - 1)) {
+               netdev_err(dev, "%s: queue_mapping %d is invalid\n",
+                               __func__, skb_get_queue_mapping(skb));
+               dev->stats.tx_errors++;
+               dev->stats.tx_dropped++;
+               ret = NETDEV_TX_OK;
+               goto out;
+       }
+       nr_frags = skb_shinfo(skb)->nr_frags;
+       ring = &priv->tx_rings[index];
+
+       spin_lock_irqsave(&ring->lock, flags);
+       if (ring->free_bds <= nr_frags + 1) {
+               netif_stop_subqueue(dev, ring->queue);
+               netdev_err(dev, "%s: tx ring %d full when queue %d awake\n",
+                               __func__, index, ring->queue);
+               ret = NETDEV_TX_BUSY;
+               goto out;
+       }
+
+       /* reclaim xmited skb every 8 packets. */
+       /*if (ring->free_bds < ring->size - 8)*/
+               /*__bcmgenet_tx_reclaim(dev, ring);*/
+
+       /* set the SKB transmit checksum */
+       if (priv->desc_64b_en) {
+               ret = bcmgenet_put_tx_csum(dev, skb);
+               if (ret) {
+                       ret = NETDEV_TX_OK;
+                       goto out;
+               }
+       }
+
+       dma_desc_flags = DMA_SOP;
+       if (nr_frags == 0)
+               dma_desc_flags |= DMA_EOP;
+
+       /* Transmit single SKB or head of fragment list */
+       ret = bcmgenet_xmit_single(dev, skb, dma_desc_flags, ring);
+       if (ret) {
+               ret = NETDEV_TX_OK;
+               goto out;
+       }
+
+       /* xmit fragment */
+       for (i = 0; i < nr_frags; i++) {
+               ret = bcmgenet_xmit_frag(dev,
+                               &skb_shinfo(skb)->frags[i],
+                               (i == nr_frags - 1) ? DMA_EOP : 0, ring);
+               if (ret) {
+                       ret = NETDEV_TX_OK;
+                       goto out;
+               }
+       }
+
+       /* we kept a software copy of how much we should advance the TDMA
+        * producer index, now write it down to the hardware
+        */
+       bcmgenet_tdma_ring_writel(priv, ring->index,
+                       ring->prod_index, TDMA_PROD_INDEX);
+
+       if (ring->free_bds <= (MAX_SKB_FRAGS + 1)) {
+               netif_stop_subqueue(dev, ring->queue);
+               ring->int_enable(priv, ring);
+       }
+
+out:
+       spin_unlock_irqrestore(&ring->lock, flags);
+
+       return ret;
+}
+
+
+static int bcmgenet_rx_refill(struct bcmgenet_priv *priv,
+                               struct enet_cb *cb)
+{
+       struct device *kdev = &priv->pdev->dev;
+       struct sk_buff *skb;
+       dma_addr_t mapping;
+       int ret;
+
+       skb = netdev_alloc_skb(priv->dev,
+                               priv->rx_buf_len + SKB_ALIGNMENT);
+       if (!skb)
+               return -ENOMEM;
+
+       /* a caller did not release this control block */
+       WARN_ON(cb->skb != NULL);
+       cb->skb = skb;
+       mapping = dma_map_single(kdev, skb->data,
+                       priv->rx_buf_len, DMA_FROM_DEVICE);
+       ret = dma_mapping_error(kdev, mapping);
+       if (ret) {
+               bcmgenet_free_cb(cb);
+               netif_err(priv, rx_err, priv->dev,
+                               "%s DMA map failed\n", __func__);
+               return ret;
+       }
+
+       dma_unmap_addr_set(cb, dma_addr, mapping);
+       /* assign packet, prepare descriptor, and advance pointer */
+
+       dmadesc_set_addr(priv, priv->rx_bd_assign_ptr, mapping);
+
+       /* turn on the newly assigned BD for DMA to use */
+       priv->rx_bd_assign_index++;
+       priv->rx_bd_assign_index &= (priv->num_rx_bds - 1);
+
+       priv->rx_bd_assign_ptr = priv->rx_bds +
+               (priv->rx_bd_assign_index * DMA_DESC_SIZE);
+
+       return 0;
+}
+
+/* bcmgenet_desc_rx - descriptor based rx process.
+ * this could be called from bottom half, or from NAPI polling method.
+ */
+static unsigned int bcmgenet_desc_rx(struct bcmgenet_priv *priv,
+                                    unsigned int budget)
+{
+       struct net_device *dev = priv->dev;
+       struct enet_cb *cb;
+       struct sk_buff *skb;
+       u32 dma_length_status;
+       unsigned long dma_flag;
+       int len, err;
+       unsigned int rxpktprocessed = 0, rxpkttoprocess;
+       unsigned int p_index;
+       unsigned int chksum_ok = 0;
+
+       p_index = bcmgenet_rdma_ring_readl(priv,
+                       DESC_INDEX, RDMA_PROD_INDEX);
+       p_index &= DMA_P_INDEX_MASK;
+
+       if (p_index < priv->rx_c_index)
+               rxpkttoprocess = (DMA_C_INDEX_MASK + 1) -
+                       priv->rx_c_index + p_index;
+       else
+               rxpkttoprocess = p_index - priv->rx_c_index;
+
+       netif_dbg(priv, rx_status, dev,
+               "RDMA: rxpkttoprocess=%d\n", rxpkttoprocess);
+
+       while ((rxpktprocessed < rxpkttoprocess) &&
+                       (rxpktprocessed < budget)) {
+
+               /* Unmap the packet contents such that we can use the
+                * RSV from the 64 bytes descriptor when enabled and save
+                * a 32-bits register read
+                */
+               cb = &priv->rx_cbs[priv->rx_read_ptr];
+               skb = cb->skb;
+               dma_unmap_single(&dev->dev, dma_unmap_addr(cb, dma_addr),
+                               priv->rx_buf_len, DMA_FROM_DEVICE);
+
+               if (!priv->desc_64b_en) {
+                       dma_length_status = dmadesc_get_length_status(priv,
+                                                       priv->rx_bds +
+                                                       (priv->rx_read_ptr *
+                                                        DMA_DESC_SIZE));
+               } else {
+                       struct status_64 *status;
+                       status = (struct status_64 *)skb->data;
+                       dma_length_status = status->length_status;
+               }
+
+               /* DMA flags and length are still valid no matter how
+                * we got the Receive Status Vector (64B RSB or register)
+                */
+               dma_flag = dma_length_status & 0xffff;
+               len = dma_length_status >> DMA_BUFLENGTH_SHIFT;
+
+               netif_dbg(priv, rx_status, dev,
+                       "%s: p_ind=%d c_ind=%d read_ptr=%d len_stat=0x%08x\n",
+                       __func__, p_index, priv->rx_c_index, priv->rx_read_ptr,
+                       dma_length_status);
+
+               rxpktprocessed++;
+
+               priv->rx_read_ptr++;
+               priv->rx_read_ptr &= (priv->num_rx_bds - 1);
+
+               /* out of memory, just drop packets at the hardware level */
+               if (unlikely(!skb)) {
+                       dev->stats.rx_dropped++;
+                       dev->stats.rx_errors++;
+                       goto refill;
+               }
+
+               if (unlikely(!(dma_flag & DMA_EOP) || !(dma_flag & DMA_SOP))) {
+                       netif_err(priv, rx_status, dev,
+                                       "Droping fragmented packet!\n");
+                       dev->stats.rx_dropped++;
+                       dev->stats.rx_errors++;
+                       dev_kfree_skb_any(cb->skb);
+                       cb->skb = NULL;
+                       goto refill;
+               }
+               /* report errors */
+               if (unlikely(dma_flag & (DMA_RX_CRC_ERROR |
+                                               DMA_RX_OV |
+                                               DMA_RX_NO |
+                                               DMA_RX_LG |
+                                               DMA_RX_RXER))) {
+                       netif_err(priv, rx_status, dev, "dma_flag=0x%x\n",
+                                               (unsigned int)dma_flag);
+                       if (dma_flag & DMA_RX_CRC_ERROR)
+                               dev->stats.rx_crc_errors++;
+                       if (dma_flag & DMA_RX_OV)
+                               dev->stats.rx_over_errors++;
+                       if (dma_flag & DMA_RX_NO)
+                               dev->stats.rx_frame_errors++;
+                       if (dma_flag & DMA_RX_LG)
+                               dev->stats.rx_length_errors++;
+                       dev->stats.rx_dropped++;
+                       dev->stats.rx_errors++;
+
+                       /* discard the packet and advance consumer index.*/
+                       dev_kfree_skb_any(cb->skb);
+                       cb->skb = NULL;
+                       goto refill;
+               } /* error packet */
+
+               chksum_ok = (dma_flag & priv->dma_rx_chk_bit) &&
+                               priv->desc_rxchk_en;
+
+               skb_put(skb, len);
+               if (priv->desc_64b_en) {
+                       skb_pull(skb, 64);
+                       len -= 64;
+               }
+
+               if (likely(chksum_ok))
+                       skb->ip_summed = CHECKSUM_UNNECESSARY;
+
+               /* remove hardware 2bytes added for IP alignment */
+               skb_pull(skb, 2);
+               len -= 2;
+
+               if (priv->crc_fwd_en) {
+                       skb_trim(skb, len - ETH_FCS_LEN);
+                       len -= ETH_FCS_LEN;
+               }
+
+               /*Finish setting up the received SKB and send it to the kernel*/
+               skb->protocol = eth_type_trans(skb, priv->dev);
+               dev->stats.rx_packets++;
+               dev->stats.rx_bytes += len;
+               if (dma_flag & DMA_RX_MULT)
+                       dev->stats.multicast++;
+
+               /* Notify kernel */
+               napi_gro_receive(&priv->napi, skb);
+               cb->skb = NULL;
+               netif_dbg(priv, rx_status, dev, "pushed up to kernel\n");
+
+               /* refill RX path on the current control block */
+refill:
+               err = bcmgenet_rx_refill(priv, cb);
+               if (err)
+                       netif_err(priv, rx_err, dev, "Rx refill failed\n");
+       }
+
+       return rxpktprocessed;
+}
+
+/* Assign skb to RX DMA descriptor. */
+static int bcmgenet_alloc_rx_buffers(struct bcmgenet_priv *priv)
+{
+       struct enet_cb *cb;
+       int ret = 0;
+       int i;
+
+       netif_dbg(priv, hw, priv->dev, "%s:\n", __func__);
+
+       /* loop here for each buffer needing assign */
+       for (i = 0; i < priv->num_rx_bds; i++) {
+               cb = &priv->rx_cbs[priv->rx_bd_assign_index];
+               if (cb->skb)
+                       continue;
+
+               /* set the DMA descriptor length once and for all
+                * it will only change if we support dynamically sizing
+                * priv->rx_buf_len, but we do not
+                */
+               dmadesc_set_length_status(priv, priv->rx_bd_assign_ptr,
+                               priv->rx_buf_len << DMA_BUFLENGTH_SHIFT);
+
+               ret = bcmgenet_rx_refill(priv, cb);
+               if (ret)
+                       break;
+
+       }
+
+       return ret;
+}
+
+static void bcmgenet_free_rx_buffers(struct bcmgenet_priv *priv)
+{
+       struct enet_cb *cb;
+       int i;
+
+       for (i = 0; i < priv->num_rx_bds; i++) {
+               cb = &priv->rx_cbs[i];
+
+               if (dma_unmap_addr(cb, dma_addr)) {
+                       dma_unmap_single(&priv->dev->dev,
+                                       dma_unmap_addr(cb, dma_addr),
+                                       priv->rx_buf_len, DMA_FROM_DEVICE);
+                       dma_unmap_addr_set(cb, dma_addr, 0);
+               }
+
+               if (cb->skb)
+                       bcmgenet_free_cb(cb);
+       }
+}
+
+static int reset_umac(struct bcmgenet_priv *priv)
+{
+       struct device *kdev = &priv->pdev->dev;
+       unsigned int timeout = 0;
+       u32 reg;
+
+       /* 7358a0/7552a0: bad default in RBUF_FLUSH_CTRL.umac_sw_rst */
+       bcmgenet_rbuf_ctrl_set(priv, 0);
+       udelay(10);
+
+       /* disable MAC while updating its registers */
+       bcmgenet_umac_writel(priv, 0, UMAC_CMD);
+
+       /* issue soft reset, wait for it to complete */
+       bcmgenet_umac_writel(priv, CMD_SW_RESET, UMAC_CMD);
+       while (timeout++ < 1000) {
+               reg = bcmgenet_umac_readl(priv, UMAC_CMD);
+               if (!(reg & CMD_SW_RESET))
+                       return 0;
+
+               udelay(1);
+       }
+
+       if (timeout == 1000) {
+               dev_err(kdev,
+                       "timeout waiting for MAC to come out of resetn\n");
+               return -ETIMEDOUT;
+       }
+
+       return 0;
+}
+
+static int init_umac(struct bcmgenet_priv *priv)
+{
+       struct device *kdev = &priv->pdev->dev;
+       int ret;
+       u32 reg, cpu_mask_clear;
+
+       dev_dbg(&priv->pdev->dev, "bcmgenet: init_umac\n");
+
+       ret = reset_umac(priv);
+       if (ret)
+               return ret;
+
+       bcmgenet_umac_writel(priv, 0, UMAC_CMD);
+       /* clear tx/rx counter */
+       bcmgenet_umac_writel(priv,
+               MIB_RESET_RX | MIB_RESET_TX | MIB_RESET_RUNT, UMAC_MIB_CTRL);
+       bcmgenet_umac_writel(priv, 0, UMAC_MIB_CTRL);
+
+       bcmgenet_umac_writel(priv, ENET_MAX_MTU_SIZE, UMAC_MAX_FRAME_LEN);
+
+       /* init rx registers, enable ip header optimization */
+       reg = bcmgenet_rbuf_readl(priv, RBUF_CTRL);
+       reg |= RBUF_ALIGN_2B;
+       bcmgenet_rbuf_writel(priv, reg, RBUF_CTRL);
+
+       if (!GENET_IS_V1(priv) && !GENET_IS_V2(priv))
+               bcmgenet_rbuf_writel(priv, 1, RBUF_TBUF_SIZE_CTRL);
+
+       /* Mask all interrupts.*/
+       bcmgenet_intrl2_0_writel(priv, 0xFFFFFFFF, INTRL2_CPU_MASK_SET);
+       bcmgenet_intrl2_0_writel(priv, 0xFFFFFFFF, INTRL2_CPU_CLEAR);
+       bcmgenet_intrl2_0_writel(priv, 0, INTRL2_CPU_MASK_CLEAR);
+
+       cpu_mask_clear = UMAC_IRQ_RXDMA_BDONE;
+
+       dev_dbg(kdev, "%s:Enabling RXDMA_BDONE interrupt\n", __func__);
+
+       /* Monitor cable plug/unpluged event for internal PHY */
+       if (phy_is_internal(priv->phydev))
+               cpu_mask_clear |= (UMAC_IRQ_LINK_DOWN | UMAC_IRQ_LINK_UP);
+       else if (priv->ext_phy)
+               cpu_mask_clear |= (UMAC_IRQ_LINK_DOWN | UMAC_IRQ_LINK_UP);
+       else if (priv->phy_interface == PHY_INTERFACE_MODE_MOCA) {
+               reg = bcmgenet_bp_mc_get(priv);
+               reg |= BIT(priv->hw_params->bp_in_en_shift);
+
+               /* bp_mask: back pressure mask */
+               if (netif_is_multiqueue(priv->dev))
+                       reg |= priv->hw_params->bp_in_mask;
+               else
+                       reg &= ~priv->hw_params->bp_in_mask;
+               bcmgenet_bp_mc_set(priv, reg);
+       }
+
+       /* Enable MDIO interrupts on GENET v3+ */
+       if (priv->hw_params->flags & GENET_HAS_MDIO_INTR)
+               cpu_mask_clear |= UMAC_IRQ_MDIO_DONE | UMAC_IRQ_MDIO_ERROR;
+
+       bcmgenet_intrl2_0_writel(priv, cpu_mask_clear,
+               INTRL2_CPU_MASK_CLEAR);
+
+       /* Enable rx/tx engine.*/
+       dev_dbg(kdev, "done init umac\n");
+
+       return 0;
+}
+
+/* Initialize all house-keeping variables for a TX ring, along
+ * with corresponding hardware registers
+ */
+static void bcmgenet_init_tx_ring(struct bcmgenet_priv *priv,
+                                 unsigned int index, unsigned int size,
+                                 unsigned int write_ptr, unsigned int end_ptr)
+{
+       struct bcmgenet_tx_ring *ring = &priv->tx_rings[index];
+       u32 words_per_bd = WORDS_PER_BD(priv);
+       u32 flow_period_val = 0;
+       unsigned int first_bd;
+
+       spin_lock_init(&ring->lock);
+       ring->index = index;
+       if (index == DESC_INDEX) {
+               ring->queue = 0;
+               ring->int_enable = bcmgenet_tx_ring16_int_enable;
+               ring->int_disable = bcmgenet_tx_ring16_int_disable;
+       } else {
+               ring->queue = index + 1;
+               ring->int_enable = bcmgenet_tx_ring_int_enable;
+               ring->int_disable = bcmgenet_tx_ring_int_disable;
+       }
+       ring->cbs = priv->tx_cbs + write_ptr;
+       ring->size = size;
+       ring->c_index = 0;
+       ring->free_bds = size;
+       ring->write_ptr = write_ptr;
+       ring->cb_ptr = write_ptr;
+       ring->end_ptr = end_ptr - 1;
+       ring->prod_index = 0;
+
+       /* Set flow period for ring != 16 */
+       if (index != DESC_INDEX)
+               flow_period_val = ENET_MAX_MTU_SIZE << 16;
+
+       bcmgenet_tdma_ring_writel(priv, index, 0, TDMA_PROD_INDEX);
+       bcmgenet_tdma_ring_writel(priv, index, 0, TDMA_CONS_INDEX);
+       bcmgenet_tdma_ring_writel(priv, index, 1, DMA_MBUF_DONE_THRESH);
+       /* Disable rate control for now */
+       bcmgenet_tdma_ring_writel(priv, index, flow_period_val,
+                       TDMA_FLOW_PERIOD);
+       /* Unclassified traffic goes to ring 16 */
+       bcmgenet_tdma_ring_writel(priv, index,
+                       ((size << DMA_RING_SIZE_SHIFT) | RX_BUF_LENGTH),
+                       DMA_RING_BUF_SIZE);
+
+       first_bd = write_ptr;
+
+       /* Set start and end address, read and write pointers */
+       bcmgenet_tdma_ring_writel(priv, index, first_bd * words_per_bd,
+                       DMA_START_ADDR);
+       bcmgenet_tdma_ring_writel(priv, index, first_bd * words_per_bd,
+                       TDMA_READ_PTR);
+       bcmgenet_tdma_ring_writel(priv, index, first_bd,
+                       TDMA_WRITE_PTR);
+       bcmgenet_tdma_ring_writel(priv, index, end_ptr * words_per_bd - 1,
+                       DMA_END_ADDR);
+}
+
+/* Initialize a RDMA ring */
+static int bcmgenet_init_rx_ring(struct bcmgenet_priv *priv,
+                                 unsigned int index, unsigned int size)
+{
+       u32 words_per_bd = WORDS_PER_BD(priv);
+       int ret;
+
+       priv->num_rx_bds = TOTAL_DESC;
+       priv->rx_bds = priv->base + priv->hw_params->rdma_offset;
+       priv->rx_bd_assign_ptr = priv->rx_bds;
+       priv->rx_bd_assign_index = 0;
+       priv->rx_c_index = 0;
+       priv->rx_read_ptr = 0;
+       priv->rx_cbs = kzalloc(priv->num_rx_bds * sizeof(struct enet_cb),
+                               GFP_KERNEL);
+       if (!priv->rx_cbs)
+               return -ENOMEM;
+
+       ret = bcmgenet_alloc_rx_buffers(priv);
+       if (ret) {
+               kfree(priv->rx_cbs);
+               return ret;
+       }
+
+       bcmgenet_rdma_ring_writel(priv, index, 0, RDMA_WRITE_PTR);
+       bcmgenet_rdma_ring_writel(priv, index, 0, RDMA_PROD_INDEX);
+       bcmgenet_rdma_ring_writel(priv, index, 0, RDMA_CONS_INDEX);
+       bcmgenet_rdma_ring_writel(priv, index,
+               ((size << DMA_RING_SIZE_SHIFT) | RX_BUF_LENGTH),
+               DMA_RING_BUF_SIZE);
+       bcmgenet_rdma_ring_writel(priv, index, 0, DMA_START_ADDR);
+       bcmgenet_rdma_ring_writel(priv, index,
+               words_per_bd * size - 1, DMA_END_ADDR);
+       bcmgenet_rdma_ring_writel(priv, index,
+                       (DMA_FC_THRESH_LO << DMA_XOFF_THRESHOLD_SHIFT) |
+                       DMA_FC_THRESH_HI, RDMA_XON_XOFF_THRESH);
+       bcmgenet_rdma_ring_writel(priv, index, 0, RDMA_READ_PTR);
+
+       return ret;
+}
+
+/* init multi xmit queues, only available for GENET2+
+ * the queue is partitioned as follows:
+ *
+ * queue 0 - 3 is priority based, each one has 32 descriptors,
+ * with queue 0 being the highest priority queue.
+ *
+ * queue 16 is the default tx queue with GENET_DEFAULT_BD_CNT
+ * descriptors: 256 - (number of tx queues * bds per queues) = 128
+ * descriptors.
+ *
+ * The transmit control block pool is then partitioned as following:
+ * - tx_cbs[0...127] are for queue 16
+ * - tx_ring_cbs[0] points to tx_cbs[128..159]
+ * - tx_ring_cbs[1] points to tx_cbs[160..191]
+ * - tx_ring_cbs[2] points to tx_cbs[192..223]
+ * - tx_ring_cbs[3] points to tx_cbs[224..255]
+ */
+static void bcmgenet_init_multiq(struct net_device *dev)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       unsigned int i, dma_enable;
+       u32 reg, dma_ctrl, ring_cfg = 0, dma_priority = 0;
+
+       if (!netif_is_multiqueue(dev)) {
+               netdev_warn(dev, "called with non multi queue aware HW\n");
+               return;
+       }
+
+       dma_ctrl = bcmgenet_tdma_readl(priv, DMA_CTRL);
+       dma_enable = dma_ctrl & DMA_EN;
+       dma_ctrl &= ~DMA_EN;
+       bcmgenet_tdma_writel(priv, dma_ctrl, DMA_CTRL);
+
+       /* Enable strict priority arbiter mode */
+       bcmgenet_tdma_writel(priv, DMA_ARBITER_SP, DMA_ARB_CTRL);
+
+       for (i = 0; i < priv->hw_params->tx_queues; i++) {
+               /* first 64 tx_cbs are reserved for default tx queue
+                * (ring 16)
+                */
+               bcmgenet_init_tx_ring(priv, i, priv->hw_params->bds_cnt,
+                                       i * priv->hw_params->bds_cnt,
+                                       (i + 1) * priv->hw_params->bds_cnt);
+
+               /* Configure ring as decriptor ring and setup priority */
+               ring_cfg |= 1 << i;
+               dma_priority |= ((GENET_Q0_PRIORITY + i) <<
+                               (GENET_MAX_MQ_CNT + 1) * i);
+               dma_ctrl |= 1 << (i + DMA_RING_BUF_EN_SHIFT);
+       }
+
+       /* Enable rings */
+       reg = bcmgenet_tdma_readl(priv, DMA_RING_CFG);
+       reg |= ring_cfg;
+       bcmgenet_tdma_writel(priv, reg, DMA_RING_CFG);
+
+       /* Use configured rings priority and set ring #16 priority */
+       reg = bcmgenet_tdma_readl(priv, DMA_RING_PRIORITY);
+       reg |= ((GENET_Q0_PRIORITY + priv->hw_params->tx_queues) << 20);
+       reg |= dma_priority;
+       bcmgenet_tdma_writel(priv, reg, DMA_PRIORITY);
+
+       /* Configure ring as descriptor ring and re-enable DMA if enabled */
+       reg = bcmgenet_tdma_readl(priv, DMA_CTRL);
+       reg |= dma_ctrl;
+       if (dma_enable)
+               reg |= DMA_EN;
+       bcmgenet_tdma_writel(priv, reg, DMA_CTRL);
+}
+
+static void bcmgenet_fini_dma(struct bcmgenet_priv *priv)
+{
+       int i;
+
+       /* disable DMA */
+       bcmgenet_rdma_writel(priv, 0, DMA_CTRL);
+       bcmgenet_tdma_writel(priv, 0, DMA_CTRL);
+
+       for (i = 0; i < priv->num_tx_bds; i++) {
+               if (priv->tx_cbs[i].skb != NULL) {
+                       dev_kfree_skb(priv->tx_cbs[i].skb);
+                       priv->tx_cbs[i].skb = NULL;
+               }
+       }
+
+       bcmgenet_free_rx_buffers(priv);
+       kfree(priv->rx_cbs);
+       kfree(priv->tx_cbs);
+}
+
+/* init_edma: Initialize DMA control register */
+static int bcmgenet_init_dma(struct bcmgenet_priv *priv)
+{
+       int ret;
+
+       netif_dbg(priv, hw, priv->dev, "bcmgenet: init_edma\n");
+
+       /* by default, enable ring 16 (descriptor based) */
+       ret = bcmgenet_init_rx_ring(priv, DESC_INDEX, TOTAL_DESC);
+       if (ret) {
+               netdev_err(priv->dev, "failed to initialize RX ring\n");
+               return ret;
+       }
+
+       /* init rDma */
+       bcmgenet_rdma_writel(priv, DMA_MAX_BURST_LENGTH, DMA_SCB_BURST_SIZE);
+
+       /* Init tDma */
+       bcmgenet_tdma_writel(priv, DMA_MAX_BURST_LENGTH, DMA_SCB_BURST_SIZE);
+
+       /* Initialize commont TX ring structures */
+       priv->tx_bds = priv->base + priv->hw_params->tdma_offset;
+       priv->num_tx_bds = TOTAL_DESC;
+       priv->tx_cbs = kzalloc(priv->num_tx_bds * sizeof(struct enet_cb),
+                               GFP_KERNEL);
+       if (!priv->tx_cbs) {
+               bcmgenet_fini_dma(priv);
+               return -ENOMEM;
+       }
+
+       /* initialize multi xmit queue */
+       bcmgenet_init_multiq(priv->dev);
+
+       /* initialize special ring 16 */
+       bcmgenet_init_tx_ring(priv, DESC_INDEX, GENET_DEFAULT_BD_CNT,
+                       priv->hw_params->tx_queues * priv->hw_params->bds_cnt,
+                       TOTAL_DESC);
+
+       return 0;
+}
+
+/* NAPI polling method*/
+static int bcmgenet_poll(struct napi_struct *napi, int budget)
+{
+       struct bcmgenet_priv *priv = container_of(napi,
+                       struct bcmgenet_priv, napi);
+       unsigned int work_done;
+
+       /* tx reclaim */
+       bcmgenet_tx_reclaim(priv->dev, &priv->tx_rings[DESC_INDEX]);
+
+       work_done = bcmgenet_desc_rx(priv, budget);
+
+       /* Advancing our consumer index*/
+       priv->rx_c_index += work_done;
+       priv->rx_c_index &= DMA_C_INDEX_MASK;
+       bcmgenet_rdma_ring_writel(priv, DESC_INDEX,
+                               priv->rx_c_index, RDMA_CONS_INDEX);
+       if (work_done < budget) {
+               napi_complete(napi);
+               bcmgenet_intrl2_0_writel(priv,
+                       UMAC_IRQ_RXDMA_BDONE, INTRL2_CPU_MASK_CLEAR);
+       }
+
+       return work_done;
+}
+
+/* Interrupt bottom half */
+static void bcmgenet_irq_task(struct work_struct *work)
+{
+       struct bcmgenet_priv *priv = container_of(
+                       work, struct bcmgenet_priv, bcmgenet_irq_work);
+
+       netif_dbg(priv, intr, priv->dev, "%s\n", __func__);
+
+       /* Link UP/DOWN event */
+       if ((priv->hw_params->flags & GENET_HAS_MDIO_INTR) &&
+               (priv->irq0_stat & (UMAC_IRQ_LINK_UP|UMAC_IRQ_LINK_DOWN))) {
+               if (priv->phydev)
+                       phy_mac_interrupt(priv->phydev,
+                               (priv->irq0_stat & UMAC_IRQ_LINK_UP));
+               priv->irq0_stat &= ~(UMAC_IRQ_LINK_UP|UMAC_IRQ_LINK_DOWN);
+       }
+}
+
+/* bcmgenet_isr1: interrupt handler for ring buffer. */
+static irqreturn_t bcmgenet_isr1(int irq, void *dev_id)
+{
+       struct bcmgenet_priv *priv = dev_id;
+       unsigned int index;
+
+       /* Save irq status for bottom-half processing. */
+       priv->irq1_stat =
+               bcmgenet_intrl2_1_readl(priv, INTRL2_CPU_STAT) &
+               ~priv->int1_mask;
+       /* clear inerrupts*/
+       bcmgenet_intrl2_1_writel(priv, priv->irq1_stat, INTRL2_CPU_CLEAR);
+
+       netif_dbg(priv, intr, priv->dev,
+               "%s: IRQ=0x%x\n", __func__, priv->irq1_stat);
+       /* Check the MBDONE interrupts.
+        * packet is done, reclaim descriptors
+        */
+       if (priv->irq1_stat & 0x0000ffff) {
+               index = 0;
+               for (index = 0; index < 16; index++) {
+                       if (priv->irq1_stat & (1 << index))
+                               bcmgenet_tx_reclaim(priv->dev,
+                                               &priv->tx_rings[index]);
+               }
+       }
+       return IRQ_HANDLED;
+}
+
+/* bcmgenet_isr0: Handle various interrupts. */
+static irqreturn_t bcmgenet_isr0(int irq, void *dev_id)
+{
+       struct bcmgenet_priv *priv = dev_id;
+
+       /* Save irq status for bottom-half processing. */
+       priv->irq0_stat =
+               bcmgenet_intrl2_0_readl(priv, INTRL2_CPU_STAT) &
+               ~bcmgenet_intrl2_0_readl(priv, INTRL2_CPU_MASK_STATUS);
+       /* clear inerrupts*/
+       bcmgenet_intrl2_0_writel(priv, priv->irq0_stat, INTRL2_CPU_CLEAR);
+
+       netif_dbg(priv, intr, priv->dev,
+               "IRQ=0x%x\n", priv->irq0_stat);
+
+       if (priv->irq0_stat & (UMAC_IRQ_RXDMA_BDONE | UMAC_IRQ_RXDMA_PDONE)) {
+               /* We use NAPI(software interrupt throttling, if
+                * Rx Descriptor throttling is not used.
+                * Disable interrupt, will be enabled in the poll method.
+                */
+               if (likely(napi_schedule_prep(&priv->napi))) {
+                       bcmgenet_intrl2_0_writel(priv,
+                               UMAC_IRQ_RXDMA_BDONE, INTRL2_CPU_MASK_SET);
+                       __napi_schedule(&priv->napi);
+               }
+       }
+       if (priv->irq0_stat &
+                       (UMAC_IRQ_TXDMA_BDONE | UMAC_IRQ_TXDMA_PDONE)) {
+               /* Tx reclaim */
+               bcmgenet_tx_reclaim(priv->dev, &priv->tx_rings[DESC_INDEX]);
+       }
+       if (priv->irq0_stat & (UMAC_IRQ_PHY_DET_R |
+                               UMAC_IRQ_PHY_DET_F |
+                               UMAC_IRQ_LINK_UP |
+                               UMAC_IRQ_LINK_DOWN |
+                               UMAC_IRQ_HFB_SM |
+                               UMAC_IRQ_HFB_MM |
+                               UMAC_IRQ_MPD_R)) {
+               /* all other interested interrupts handled in bottom half */
+               schedule_work(&priv->bcmgenet_irq_work);
+       }
+
+       if ((priv->hw_params->flags & GENET_HAS_MDIO_INTR) &&
+               priv->irq0_stat & (UMAC_IRQ_MDIO_DONE | UMAC_IRQ_MDIO_ERROR)) {
+               priv->irq0_stat &= ~(UMAC_IRQ_MDIO_DONE | UMAC_IRQ_MDIO_ERROR);
+               wake_up(&priv->wq);
+       }
+
+       return IRQ_HANDLED;
+}
+
+static void bcmgenet_umac_reset(struct bcmgenet_priv *priv)
+{
+       u32 reg;
+
+       reg = bcmgenet_rbuf_ctrl_get(priv);
+       reg |= BIT(1);
+       bcmgenet_rbuf_ctrl_set(priv, reg);
+       udelay(10);
+
+       reg &= ~BIT(1);
+       bcmgenet_rbuf_ctrl_set(priv, reg);
+       udelay(10);
+}
+
+static void bcmgenet_set_hw_addr(struct bcmgenet_priv *priv,
+                                 unsigned char *addr)
+{
+       bcmgenet_umac_writel(priv, (addr[0] << 24) | (addr[1] << 16) |
+                       (addr[2] << 8) | addr[3], UMAC_MAC0);
+       bcmgenet_umac_writel(priv, (addr[4] << 8) | addr[5], UMAC_MAC1);
+}
+
+static int bcmgenet_wol_resume(struct bcmgenet_priv *priv)
+{
+       int ret;
+
+       /* From WOL-enabled suspend, switch to regular clock */
+       clk_disable(priv->clk_wol);
+       /* init umac registers to synchronize s/w with h/w */
+       ret = init_umac(priv);
+       if (ret)
+               return ret;
+
+       if (priv->phydev)
+               phy_init_hw(priv->phydev);
+       /* Speed settings must be restored */
+       bcmgenet_mii_config(priv->dev);
+
+       return 0;
+}
+
+/* Returns a reusable dma control register value */
+static u32 bcmgenet_dma_disable(struct bcmgenet_priv *priv)
+{
+       u32 reg;
+       u32 dma_ctrl;
+
+       /* disable DMA */
+       dma_ctrl = 1 << (DESC_INDEX + DMA_RING_BUF_EN_SHIFT) | DMA_EN;
+       reg = bcmgenet_tdma_readl(priv, DMA_CTRL);
+       reg &= ~dma_ctrl;
+       bcmgenet_tdma_writel(priv, reg, DMA_CTRL);
+
+       reg = bcmgenet_rdma_readl(priv, DMA_CTRL);
+       reg &= ~dma_ctrl;
+       bcmgenet_rdma_writel(priv, reg, DMA_CTRL);
+
+       bcmgenet_umac_writel(priv, 1, UMAC_TX_FLUSH);
+       udelay(10);
+       bcmgenet_umac_writel(priv, 0, UMAC_TX_FLUSH);
+
+       return dma_ctrl;
+}
+
+static void bcmgenet_enable_dma(struct bcmgenet_priv *priv, u32 dma_ctrl)
+{
+       u32 reg;
+
+       reg = bcmgenet_rdma_readl(priv, DMA_CTRL);
+       reg |= dma_ctrl;
+       bcmgenet_rdma_writel(priv, reg, DMA_CTRL);
+
+       reg = bcmgenet_tdma_readl(priv, DMA_CTRL);
+       reg |= dma_ctrl;
+       bcmgenet_tdma_writel(priv, reg, DMA_CTRL);
+}
+
+static int bcmgenet_open(struct net_device *dev)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       unsigned long dma_ctrl;
+       u32 reg;
+       int ret;
+
+       netif_dbg(priv, ifup, dev, "bcmgenet_open\n");
+
+       /* Turn on the clock */
+       if (!IS_ERR(priv->clk))
+               clk_prepare_enable(priv->clk);
+
+       /* take MAC out of reset */
+       bcmgenet_umac_reset(priv);
+
+       ret = init_umac(priv);
+       if (ret)
+               goto err_clk_disable;
+
+       /* disable ethernet MAC while updating its registers */
+       reg = bcmgenet_umac_readl(priv, UMAC_CMD);
+       reg &= ~(CMD_TX_EN | CMD_RX_EN);
+       bcmgenet_umac_writel(priv, reg, UMAC_CMD);
+
+       bcmgenet_set_hw_addr(priv, dev->dev_addr);
+
+       if (priv->wol_enabled) {
+               ret = bcmgenet_wol_resume(priv);
+               if (ret)
+                       return ret;
+       }
+
+       if (phy_is_internal(priv->phydev)) {
+               reg = bcmgenet_ext_readl(priv, EXT_EXT_PWR_MGMT);
+               reg |= EXT_ENERGY_DET_MASK;
+               bcmgenet_ext_writel(priv, reg, EXT_EXT_PWR_MGMT);
+       }
+
+       /* Disable RX/TX DMA and flush TX queues */
+       dma_ctrl = bcmgenet_dma_disable(priv);
+
+       /* Reinitialize TDMA and RDMA and SW housekeeping */
+       ret = bcmgenet_init_dma(priv);
+       if (ret) {
+               netdev_err(dev, "failed to initialize DMA\n");
+               goto err_fini_dma;
+       }
+
+       /* Always enable ring 16 - descriptor ring */
+       bcmgenet_enable_dma(priv, dma_ctrl);
+
+       ret = request_irq(priv->irq0, bcmgenet_isr0, IRQF_SHARED,
+                       dev->name, priv);
+       if (ret < 0) {
+               netdev_err(dev, "can't request IRQ %d\n", priv->irq0);
+               goto err_fini_dma;
+       }
+
+       ret = request_irq(priv->irq1, bcmgenet_isr1, IRQF_SHARED,
+                               dev->name, priv);
+       if (ret < 0) {
+               netdev_err(dev, "can't request IRQ %d\n", priv->irq1);
+               goto err_irq0;
+       }
+
+       /* Start the network engine */
+       napi_enable(&priv->napi);
+
+       reg = bcmgenet_umac_readl(priv, UMAC_CMD);
+       reg |= (CMD_TX_EN | CMD_RX_EN);
+       bcmgenet_umac_writel(priv, reg, UMAC_CMD);
+
+       /* Make sure we reflect the value of CRC_CMD_FWD */
+       priv->crc_fwd_en = !!(reg & CMD_CRC_FWD);
+
+       device_set_wakeup_capable(&dev->dev, 1);
+
+       if (phy_is_internal(priv->phydev))
+               bcmgenet_power_up(priv, GENET_POWER_PASSIVE);
+
+       netif_tx_start_all_queues(dev);
+
+       if (priv->phydev)
+               phy_start(priv->phydev);
+
+       return 0;
+
+err_irq0:
+       free_irq(priv->irq0, dev);
+err_fini_dma:
+       bcmgenet_fini_dma(priv);
+err_clk_disable:
+       if (!IS_ERR(priv->clk))
+               clk_disable_unprepare(priv->clk);
+       return ret;
+}
+
+static int bcmgenet_dma_teardown(struct bcmgenet_priv *priv)
+{
+       int ret = 0;
+       int timeout = 0;
+       u32 reg;
+
+       /* Disable TDMA to stop add more frames in TX DMA */
+       reg = bcmgenet_tdma_readl(priv, DMA_CTRL);
+       reg &= ~DMA_EN;
+       bcmgenet_tdma_writel(priv, reg, DMA_CTRL);
+
+       /* Check TDMA status register to confirm TDMA is disabled */
+       while (timeout++ < DMA_TIMEOUT_VAL) {
+               reg = bcmgenet_tdma_readl(priv, DMA_STATUS);
+               if (reg & DMA_DISABLED)
+                       break;
+
+               udelay(1);
+       }
+
+       if (timeout == DMA_TIMEOUT_VAL) {
+               netdev_warn(priv->dev,
+                       "Timed out while disabling TX DMA\n");
+               ret = -ETIMEDOUT;
+       }
+
+       /* Wait 10ms for packet drain in both tx and rx dma */
+       usleep_range(10000, 20000);
+
+       /* Disable RDMA */
+       reg = bcmgenet_rdma_readl(priv, DMA_CTRL);
+       reg &= ~DMA_EN;
+       bcmgenet_rdma_writel(priv, reg, DMA_CTRL);
+
+       timeout = 0;
+       /* Check RDMA status register to confirm RDMA is disabled */
+       while (timeout++ < DMA_TIMEOUT_VAL) {
+               reg = bcmgenet_rdma_readl(priv, DMA_STATUS);
+               if (reg & DMA_DISABLED)
+                       break;
+
+               udelay(1);
+       }
+
+       if (timeout == DMA_TIMEOUT_VAL) {
+               netdev_warn(priv->dev,
+                       "Timed out while disabling RX DMA\n");
+                       ret = -ETIMEDOUT;
+       }
+
+       return ret;
+}
+
+static int bcmgenet_close(struct net_device *dev)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       int ret;
+       u32 reg;
+
+       netif_dbg(priv, ifdown, dev, "bcmgenet_close\n");
+
+       if (priv->phydev)
+               phy_stop(priv->phydev);
+
+       /* Disable MAC receive */
+       reg = bcmgenet_umac_readl(priv, UMAC_CMD);
+       reg &= ~CMD_RX_EN;
+       bcmgenet_umac_writel(priv, reg, UMAC_CMD);
+
+       netif_tx_stop_all_queues(dev);
+
+       ret = bcmgenet_dma_teardown(priv);
+       if (ret)
+               return ret;
+
+       /* Disable MAC transmit. TX DMA disabled have to done before this */
+       reg = bcmgenet_umac_readl(priv, UMAC_CMD);
+       reg &= ~CMD_TX_EN;
+       bcmgenet_umac_writel(priv, reg, UMAC_CMD);
+
+       napi_disable(&priv->napi);
+
+       /* tx reclaim */
+       bcmgenet_tx_reclaim_all(dev);
+       bcmgenet_fini_dma(priv);
+
+       free_irq(priv->irq0, priv);
+       free_irq(priv->irq1, priv);
+
+       /* Wait for pending work items to complete - we are stopping
+        * the clock now. Since interrupts are disabled, no new work
+        * will be scheduled.
+        */
+       cancel_work_sync(&priv->bcmgenet_irq_work);
+
+       if (phy_is_internal(priv->phydev))
+               bcmgenet_power_down(priv, GENET_POWER_PASSIVE);
+
+       if (priv->wol_enabled)
+               clk_enable(priv->clk_wol);
+
+       if (!IS_ERR(priv->clk))
+               clk_disable_unprepare(priv->clk);
+
+       return 0;
+}
+
+static void bcmgenet_timeout(struct net_device *dev)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+
+       netif_dbg(priv, tx_err, dev, "bcmgenet_timeout\n");
+
+       dev->trans_start = jiffies;
+
+       dev->stats.tx_errors++;
+
+       netif_tx_wake_all_queues(dev);
+}
+
+#define MAX_MC_COUNT   16
+
+static inline void bcmgenet_set_mdf_addr(struct bcmgenet_priv *priv,
+                                        unsigned char *addr,
+                                        int *i,
+                                        int *mc)
+{
+       u32 reg;
+
+       bcmgenet_umac_writel(priv,
+                       addr[0] << 8 | addr[1], UMAC_MDF_ADDR + (*i * 4));
+       bcmgenet_umac_writel(priv,
+                       addr[2] << 24 | addr[3] << 16 |
+                       addr[4] << 8 | addr[5],
+                       UMAC_MDF_ADDR + ((*i + 1) * 4));
+       reg = bcmgenet_umac_readl(priv, UMAC_MDF_CTRL);
+       reg |= (1 << (MAX_MC_COUNT - *mc));
+       bcmgenet_umac_writel(priv, reg, UMAC_MDF_CTRL);
+       *i += 2;
+       (*mc)++;
+}
+
+static void bcmgenet_set_rx_mode(struct net_device *dev)
+{
+       struct bcmgenet_priv *priv = netdev_priv(dev);
+       struct netdev_hw_addr *ha;
+       int i, mc;
+       u32 reg;
+
+       netif_dbg(priv, hw, dev, "%s: %08X\n", __func__, dev->flags);
+
+       /* Promiscous mode */
+       reg = bcmgenet_umac_readl(priv, UMAC_CMD);
+       if (dev->flags & IFF_PROMISC) {
+               reg |= CMD_PROMISC;
+               bcmgenet_umac_writel(priv, reg, UMAC_CMD);
+               bcmgenet_umac_writel(priv, 0, UMAC_MDF_CTRL);
+               return;
+       } else {
+               reg &= ~CMD_PROMISC;
+               bcmgenet_umac_writel(priv, reg, UMAC_CMD);
+       }
+
+       /* UniMac doesn't support ALLMULTI */
+       if (dev->flags & IFF_ALLMULTI) {
+               netdev_warn(dev, "ALLMULTI is not supported\n");
+               return;
+       }
+
+       /* update MDF filter */
+       i = 0;
+       mc = 0;
+       /* Broadcast */
+       bcmgenet_set_mdf_addr(priv, dev->broadcast, &i, &mc);
+       /* my own address.*/
+       bcmgenet_set_mdf_addr(priv, dev->dev_addr, &i, &mc);
+       /* Unicast list*/
+       if (netdev_uc_count(dev) > (MAX_MC_COUNT - mc))
+               return;
+
+       if (!netdev_uc_empty(dev))
+               netdev_for_each_uc_addr(ha, dev)
+                       bcmgenet_set_mdf_addr(priv, ha->addr, &i, &mc);
+       /* Multicast */
+       if (netdev_mc_empty(dev) || netdev_mc_count(dev) >= (MAX_MC_COUNT - mc))
+               return;
+
+       netdev_for_each_mc_addr(ha, dev)
+               bcmgenet_set_mdf_addr(priv, ha->addr, &i, &mc);
+}
+
+/* Set the hardware MAC address. */
+static int bcmgenet_set_mac_addr(struct net_device *dev, void *p)
+{
+       struct sockaddr *addr = p;
+
+       /* Setting the MAC address at the hardware level is not possible
+        * without disabling the UniMAC RX/TX enable bits.
+        */
+       if (netif_running(dev))
+               return -EBUSY;
+
+       ether_addr_copy(dev->dev_addr, addr->sa_data);
+
+       return 0;
+}
+
+static u16 bcmgenet_select_queue(struct net_device *dev,
+               struct sk_buff *skb, void *accel_priv)
+{
+       return netif_is_multiqueue(dev) ? skb->queue_mapping : 0;
+}
+
+static const struct net_device_ops bcmgenet_netdev_ops = {
+       .ndo_open               = bcmgenet_open,
+       .ndo_stop               = bcmgenet_close,
+       .ndo_start_xmit         = bcmgenet_xmit,
+       .ndo_select_queue       = bcmgenet_select_queue,
+       .ndo_tx_timeout         = bcmgenet_timeout,
+       .ndo_set_rx_mode        = bcmgenet_set_rx_mode,
+       .ndo_set_mac_address    = bcmgenet_set_mac_addr,
+       .ndo_do_ioctl           = bcmgenet_ioctl,
+       .ndo_set_features       = bcmgenet_set_features,
+};
+
+/* Array of GENET hardware parameters/characteristics */
+static struct bcmgenet_hw_params bcmgenet_hw_params[] = {
+       [GENET_V1] = {
+               .tx_queues = 0,
+               .rx_queues = 0,
+               .bds_cnt = 0,
+               .bp_in_en_shift = 16,
+               .bp_in_mask = 0xffff,
+               .hfb_filter_cnt = 16,
+               .qtag_mask = 0x1F,
+               .hfb_offset = 0x1000,
+               .rdma_offset = 0x2000,
+               .tdma_offset = 0x3000,
+               .words_per_bd = 2,
+       },
+       [GENET_V2] = {
+               .tx_queues = 4,
+               .rx_queues = 4,
+               .bds_cnt = 32,
+               .bp_in_en_shift = 16,
+               .bp_in_mask = 0xffff,
+               .hfb_filter_cnt = 16,
+               .qtag_mask = 0x1F,
+               .tbuf_offset = 0x0600,
+               .hfb_offset = 0x1000,
+               .hfb_reg_offset = 0x2000,
+               .rdma_offset = 0x3000,
+               .tdma_offset = 0x4000,
+               .words_per_bd = 2,
+               .flags = GENET_HAS_EXT,
+       },
+       [GENET_V3] = {
+               .tx_queues = 4,
+               .rx_queues = 4,
+               .bds_cnt = 32,
+               .bp_in_en_shift = 17,
+               .bp_in_mask = 0x1ffff,
+               .hfb_filter_cnt = 48,
+               .qtag_mask = 0x3F,
+               .tbuf_offset = 0x0600,
+               .hfb_offset = 0x8000,
+               .hfb_reg_offset = 0xfc00,
+               .rdma_offset = 0x10000,
+               .tdma_offset = 0x11000,
+               .words_per_bd = 2,
+               .flags = GENET_HAS_EXT | GENET_HAS_MDIO_INTR,
+       },
+       [GENET_V4] = {
+               .tx_queues = 4,
+               .rx_queues = 4,
+               .bds_cnt = 32,
+               .bp_in_en_shift = 17,
+               .bp_in_mask = 0x1ffff,
+               .hfb_filter_cnt = 48,
+               .qtag_mask = 0x3F,
+               .tbuf_offset = 0x0600,
+               .hfb_offset = 0x8000,
+               .hfb_reg_offset = 0xfc00,
+               .rdma_offset = 0x2000,
+               .tdma_offset = 0x4000,
+               .words_per_bd = 3,
+               .flags = GENET_HAS_40BITS | GENET_HAS_EXT | GENET_HAS_MDIO_INTR,
+       },
+};
+
+/* Infer hardware parameters from the detected GENET version */
+static void bcmgenet_set_hw_params(struct bcmgenet_priv *priv)
+{
+       struct bcmgenet_hw_params *params;
+       u32 reg;
+       u8 major;
+
+       if (GENET_IS_V4(priv)) {
+               bcmgenet_dma_regs = bcmgenet_dma_regs_v3plus;
+               genet_dma_ring_regs = genet_dma_ring_regs_v4;
+               priv->dma_rx_chk_bit = DMA_RX_CHK_V3PLUS;
+               priv->version = GENET_V4;
+       } else if (GENET_IS_V3(priv)) {
+               bcmgenet_dma_regs = bcmgenet_dma_regs_v3plus;
+               genet_dma_ring_regs = genet_dma_ring_regs_v123;
+               priv->dma_rx_chk_bit = DMA_RX_CHK_V3PLUS;
+               priv->version = GENET_V3;
+       } else if (GENET_IS_V2(priv)) {
+               bcmgenet_dma_regs = bcmgenet_dma_regs_v2;
+               genet_dma_ring_regs = genet_dma_ring_regs_v123;
+               priv->dma_rx_chk_bit = DMA_RX_CHK_V12;
+               priv->version = GENET_V2;
+       } else if (GENET_IS_V1(priv)) {
+               bcmgenet_dma_regs = bcmgenet_dma_regs_v1;
+               genet_dma_ring_regs = genet_dma_ring_regs_v123;
+               priv->dma_rx_chk_bit = DMA_RX_CHK_V12;
+               priv->version = GENET_V1;
+       }
+
+       /* enum genet_version starts at 1 */
+       priv->hw_params = &bcmgenet_hw_params[priv->version];
+       params = priv->hw_params;
+
+       /* Read GENET HW version */
+       reg = bcmgenet_sys_readl(priv, SYS_REV_CTRL);
+       major = (reg >> 24 & 0x0f);
+       if (major == 5)
+               major = 4;
+       else if (major == 0)
+               major = 1;
+       if (major != priv->version) {
+               dev_err(&priv->pdev->dev,
+                       "GENET version mismatch, got: %d, configured for: %d\n",
+                       major, priv->version);
+       }
+
+       /* Print the GENET core version */
+       dev_info(&priv->pdev->dev, "GENET " GENET_VER_FMT,
+               major, (reg >> 16) & 0x0f, reg & 0xffff);
+
+#ifdef CONFIG_PHYS_ADDR_T_64BIT
+       if (!(params->flags & GENET_HAS_40BITS))
+               pr_warn("GENET does not support 40-bits PA\n");
+#endif
+
+       pr_debug("Configuration for version: %d\n"
+               "TXq: %1d, RXq: %1d, BDs: %1d\n"
+               "BP << en: %2d, BP msk: 0x%05x\n"
+               "HFB count: %2d, QTAQ msk: 0x%05x\n"
+               "TBUF: 0x%04x, HFB: 0x%04x, HFBreg: 0x%04x\n"
+               "RDMA: 0x%05x, TDMA: 0x%05x\n"
+               "Words/BD: %d\n",
+               priv->version,
+               params->tx_queues, params->rx_queues, params->bds_cnt,
+               params->bp_in_en_shift, params->bp_in_mask,
+               params->hfb_filter_cnt, params->qtag_mask,
+               params->tbuf_offset, params->hfb_offset,
+               params->hfb_reg_offset,
+               params->rdma_offset, params->tdma_offset,
+               params->words_per_bd);
+}
+
+static const struct of_device_id bcmgenet_match[] = {
+       { .compatible = "brcm,genet-v1", .data = (void *)GENET_V1 },
+       { .compatible = "brcm,genet-v2", .data = (void *)GENET_V2 },
+       { .compatible = "brcm,genet-v3", .data = (void *)GENET_V3 },
+       { .compatible = "brcm,genet-v4", .data = (void *)GENET_V4 },
+       { },
+};
+
+static int bcmgenet_probe(struct platform_device *pdev)
+{
+       struct device_node *dn = pdev->dev.of_node;
+       const struct of_device_id *of_id;
+       struct bcmgenet_priv *priv;
+       struct net_device *dev;
+       const void *macaddr;
+       struct resource *r;
+       int err = -EIO;
+
+       /* Up to GENET_MAX_MQ_CNT + 1 TX queues and a single RX queue */
+       dev = alloc_etherdev_mqs(sizeof(*priv), GENET_MAX_MQ_CNT + 1, 1);
+       if (!dev) {
+               dev_err(&pdev->dev, "can't allocate net device\n");
+               return -ENOMEM;
+       }
+
+       of_id = of_match_node(bcmgenet_match, dn);
+       if (!of_id)
+               return -EINVAL;
+
+       priv = netdev_priv(dev);
+       priv->irq0 = platform_get_irq(pdev, 0);
+       priv->irq1 = platform_get_irq(pdev, 1);
+       if (!priv->irq0 || !priv->irq1) {
+               dev_err(&pdev->dev, "can't find IRQs\n");
+               err = -EINVAL;
+               goto err;
+       }
+
+       macaddr = of_get_mac_address(dn);
+       if (!macaddr) {
+               dev_err(&pdev->dev, "can't find MAC address\n");
+               err = -EINVAL;
+               goto err;
+       }
+
+       r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       priv->base = devm_request_and_ioremap(&pdev->dev, r);
+       if (!priv->base) {
+               dev_err(&pdev->dev, "can't ioremap\n");
+               err = -EINVAL;
+               goto err;
+       }
+
+       SET_NETDEV_DEV(dev, &pdev->dev);
+       dev_set_drvdata(&pdev->dev, dev);
+       ether_addr_copy(dev->dev_addr, macaddr);
+       dev->watchdog_timeo = 2 * HZ;
+       SET_ETHTOOL_OPS(dev, &bcmgenet_ethtool_ops);
+       dev->netdev_ops = &bcmgenet_netdev_ops;
+       netif_napi_add(dev, &priv->napi, bcmgenet_poll, 64);
+
+       priv->msg_enable = netif_msg_init(-1, GENET_MSG_DEFAULT);
+
+       /* Set hardware features */
+       dev->hw_features |= NETIF_F_SG | NETIF_F_IP_CSUM |
+               NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
+
+       /* Set the needed headroom to account for any possible
+        * features enabling/disabling at runtime
+        */
+       dev->needed_headroom += 64;
+
+       netdev_boot_setup_check(dev);
+
+       priv->dev = dev;
+       priv->pdev = pdev;
+       priv->version = (enum bcmgenet_version)of_id->data;
+
+       bcmgenet_set_hw_params(priv);
+
+       spin_lock_init(&priv->lock);
+       /* Mii wait queue */
+       init_waitqueue_head(&priv->wq);
+       /* Always use RX_BUF_LENGTH (2KB) buffer for all chips */
+       priv->rx_buf_len = RX_BUF_LENGTH;
+       INIT_WORK(&priv->bcmgenet_irq_work, bcmgenet_irq_task);
+
+       priv->clk = devm_clk_get(&priv->pdev->dev, "enet");
+       if (IS_ERR(priv->clk))
+               dev_warn(&priv->pdev->dev, "failed to get enet clock\n");
+
+       priv->clk_wol = devm_clk_get(&priv->pdev->dev, "enet-wol");
+       if (IS_ERR(priv->clk_wol))
+               dev_warn(&priv->pdev->dev, "failed to get enet-wol clock\n");
+
+       if (!IS_ERR(priv->clk))
+               clk_prepare_enable(priv->clk);
+
+       err = reset_umac(priv);
+       if (err)
+               goto err_clk_disable;
+
+       err = bcmgenet_mii_init(dev);
+       if (err)
+               goto err_clk_disable;
+
+       /* setup number of real queues  + 1 (GENET_V1 has 0 hardware queues
+        * just the ring 16 descriptor based TX
+        */
+       netif_set_real_num_tx_queues(priv->dev, priv->hw_params->tx_queues + 1);
+       netif_set_real_num_rx_queues(priv->dev, priv->hw_params->rx_queues + 1);
+
+       err = register_netdev(dev);
+       if (err)
+               goto err_clk_disable;
+
+       /* Turn off the main clock, WOL clock is handled separately */
+       if (!IS_ERR(priv->clk))
+               clk_disable_unprepare(priv->clk);
+
+       return err;
+
+err_clk_disable:
+       if (!IS_ERR(priv->clk))
+               clk_disable_unprepare(priv->clk);
+err:
+       free_netdev(dev);
+       return err;
+}
+
+static int bcmgenet_remove(struct platform_device *pdev)
+{
+       struct bcmgenet_priv *priv = dev_to_priv(&pdev->dev);
+
+       dev_set_drvdata(&pdev->dev, NULL);
+       unregister_netdev(priv->dev);
+       bcmgenet_mii_exit(priv->dev);
+       free_netdev(priv->dev);
+
+       return 0;
+}
+
+
+static struct platform_driver bcmgenet_driver = {
+       .probe  = bcmgenet_probe,
+       .remove = bcmgenet_remove,
+       .driver = {
+               .name   = "bcmgenet",
+               .owner  = THIS_MODULE,
+               .of_match_table = bcmgenet_match,
+       },
+};
+module_platform_driver(bcmgenet_driver);
+
+MODULE_AUTHOR("Broadcom Corporation");
+MODULE_DESCRIPTION("Broadcom GENET Ethernet controller driver");
+MODULE_ALIAS("platform:bcmgenet");
+MODULE_LICENSE("GPL");