net: netcp: Add Keystone NetCP core ethernet driver
authorKaricheri, Muralidharan <m-karicheri2@ti.com>
Fri, 16 Jan 2015 00:12:50 +0000 (19:12 -0500)
committerDavid S. Miller <davem@davemloft.net>
Mon, 19 Jan 2015 20:07:39 +0000 (15:07 -0500)
The network coprocessor (NetCP) is a hardware accelerator available in
Keystone SoCs that processes Ethernet packets. NetCP consists of following
hardware components

 1 Gigabit Ethernet (GbE) subsystem with a Ethernet switch sub-module to
   send and receive packets.
 2 Packet Accelerator (PA) module to perform packet classification
   operations such as header matching, and packet modification operations
   such as checksum generation.
 3 Security Accelerator(SA) capable of performing IPSec operations on
   ingress/egress packets.
 4 An optional 10 Gigabit Ethernet Subsystem (XGbE) which includes a
   3-port Ethernet switch sub-module capable of 10Gb/s and 1Gb/s rates
   per Ethernet port.
 5 Packet DMA and Queue Management Subsystem (QMSS) to enqueue and dequeue
   packets and DMA the packets between memory and NetCP hardware components
   described above.

NetCP core driver make use of the Keystone Navigator driver API to allocate
DMA channel for the Ethenet device and to handle packet queue/de-queue,
Please refer API's in include/linux/soc/ti/knav_dma.h and
drivers/soc/ti/knav_qmss.h for details.

NetCP driver consists of NetCP core driver and at a minimum Gigabit
Ethernet (GBE) module (1) driver to implement the Network device function.
Other modules (2,3) can be optionally added to achieve supported hardware
acceleration function. The initial version of the driver include NetCP
core driver and GBE driver modules.

Please refer Documentation/devicetree/bindings/net/keystone-netcp.txt
for design of the driver.

Cc: David Miller <davem@davemloft.net>
Cc: Rob Herring <robh+dt@kernel.org>
Cc: Grant Likely <grant.likely@linaro.org>
Cc: Santosh Shilimkar <santosh.shilimkar@kernel.org>
Cc: Pawel Moll <pawel.moll@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Ian Campbell <ijc+devicetree@hellion.org.uk>
Cc: Kumar Gala <galak@codeaurora.org>
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Wingman Kwok <w-kwok2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
MAINTAINERS
drivers/net/ethernet/ti/Kconfig
drivers/net/ethernet/ti/Makefile
drivers/net/ethernet/ti/netcp.h [new file with mode: 0644]
drivers/net/ethernet/ti/netcp_core.c [new file with mode: 0644]

index 9b91d9f..e1ff4ce 100644 (file)
@@ -9609,6 +9609,13 @@ F:       drivers/power/lp8788-charger.c
 F:     drivers/regulator/lp8788-*.c
 F:     include/linux/mfd/lp8788*.h
 
+TI NETCP ETHERNET DRIVER
+M:     Wingman Kwok <w-kwok2@ti.com>
+M:     Murali Karicheri <m-karicheri2@ti.com>
+L:     netdev@vger.kernel.org
+S:     Maintained
+F:     drivers/net/ethernet/ti/netcp*
+
 TI TWL4030 SERIES SOC CODEC DRIVER
 M:     Peter Ujfalusi <peter.ujfalusi@ti.com>
 L:     alsa-devel@alsa-project.org (moderated for non-subscribers)
index 605dd90..e11bcfa 100644 (file)
@@ -73,12 +73,23 @@ config TI_CPSW
 config TI_CPTS
        boolean "TI Common Platform Time Sync (CPTS) Support"
        depends on TI_CPSW
+       depends on TI_CPSW || TI_KEYSTONE_NET
        select PTP_1588_CLOCK
        ---help---
          This driver supports the Common Platform Time Sync unit of
          the CPSW Ethernet Switch. The unit can time stamp PTP UDP/IPv4
          and Layer 2 packets, and the driver offers a PTP Hardware Clock.
 
+config TI_KEYSTONE_NETCP
+       tristate "TI Keystone NETCP Ethernet subsystem Support"
+       depends on OF
+       depends on KEYSTONE_NAVIGATOR_DMA && KEYSTONE_NAVIGATOR_QMSS
+       ---help---
+         This driver supports TI's Keystone NETCP Ethernet subsystem.
+
+         To compile this driver as a module, choose M here: the module
+         will be called keystone_netcp.
+
 config TLAN
        tristate "TI ThunderLAN support"
        depends on (PCI || EISA)
index 9cfaab8..4e8a8e4 100644 (file)
@@ -10,3 +10,6 @@ obj-$(CONFIG_TI_DAVINCI_CPDMA) += davinci_cpdma.o
 obj-$(CONFIG_TI_CPSW_PHY_SEL) += cpsw-phy-sel.o
 obj-$(CONFIG_TI_CPSW) += ti_cpsw.o
 ti_cpsw-y := cpsw_ale.o cpsw.o cpts.o
+
+obj-$(CONFIG_TI_KEYSTONE_NETCP) += keystone_netcp.o
+keystone_netcp-y := netcp_core.o
diff --git a/drivers/net/ethernet/ti/netcp.h b/drivers/net/ethernet/ti/netcp.h
new file mode 100644 (file)
index 0000000..906e9bc
--- /dev/null
@@ -0,0 +1,229 @@
+/*
+ * NetCP driver local header
+ *
+ * Copyright (C) 2014 Texas Instruments Incorporated
+ * Authors:    Sandeep Nair <sandeep_n@ti.com>
+ *             Sandeep Paulraj <s-paulraj@ti.com>
+ *             Cyril Chemparathy <cyril@ti.com>
+ *             Santosh Shilimkar <santosh.shilimkar@ti.com>
+ *             Wingman Kwok <w-kwok2@ti.com>
+ *             Murali Karicheri <m-karicheri2@ti.com>
+ *
+ * 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 version 2.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+#ifndef __NETCP_H__
+#define __NETCP_H__
+
+#include <linux/netdevice.h>
+#include <linux/soc/ti/knav_dma.h>
+
+/* Maximum Ethernet frame size supported by Keystone switch */
+#define NETCP_MAX_FRAME_SIZE           9504
+
+#define SGMII_LINK_MAC_MAC_AUTONEG     0
+#define SGMII_LINK_MAC_PHY             1
+#define SGMII_LINK_MAC_MAC_FORCED      2
+#define SGMII_LINK_MAC_FIBER           3
+#define SGMII_LINK_MAC_PHY_NO_MDIO     4
+#define XGMII_LINK_MAC_PHY             10
+#define XGMII_LINK_MAC_MAC_FORCED      11
+
+struct netcp_device;
+
+struct netcp_tx_pipe {
+       struct netcp_device     *netcp_device;
+       void                    *dma_queue;
+       unsigned int            dma_queue_id;
+       u8                      dma_psflags;
+       void                    *dma_channel;
+       const char              *dma_chan_name;
+};
+
+#define ADDR_NEW                       BIT(0)
+#define ADDR_VALID                     BIT(1)
+
+enum netcp_addr_type {
+       ADDR_ANY,
+       ADDR_DEV,
+       ADDR_UCAST,
+       ADDR_MCAST,
+       ADDR_BCAST
+};
+
+struct netcp_addr {
+       struct netcp_intf       *netcp;
+       unsigned char           addr[ETH_ALEN];
+       enum netcp_addr_type    type;
+       unsigned int            flags;
+       struct list_head        node;
+};
+
+struct netcp_intf {
+       struct device           *dev;
+       struct device           *ndev_dev;
+       struct net_device       *ndev;
+       bool                    big_endian;
+       unsigned int            tx_compl_qid;
+       void                    *tx_pool;
+       struct list_head        txhook_list_head;
+       unsigned int            tx_pause_threshold;
+       void                    *tx_compl_q;
+
+       unsigned int            tx_resume_threshold;
+       void                    *rx_queue;
+       void                    *rx_pool;
+       struct list_head        rxhook_list_head;
+       unsigned int            rx_queue_id;
+       void                    *rx_fdq[KNAV_DMA_FDQ_PER_CHAN];
+       u32                     rx_buffer_sizes[KNAV_DMA_FDQ_PER_CHAN];
+       struct napi_struct      rx_napi;
+       struct napi_struct      tx_napi;
+
+       void                    *rx_channel;
+       const char              *dma_chan_name;
+       u32                     rx_pool_size;
+       u32                     rx_pool_region_id;
+       u32                     tx_pool_size;
+       u32                     tx_pool_region_id;
+       struct list_head        module_head;
+       struct list_head        interface_list;
+       struct list_head        addr_list;
+       bool                    netdev_registered;
+       bool                    primary_module_attached;
+
+       /* Lock used for protecting Rx/Tx hook list management */
+       spinlock_t              lock;
+       struct netcp_device     *netcp_device;
+       struct device_node      *node_interface;
+
+       /* DMA configuration data */
+       u32                     msg_enable;
+       u32                     rx_queue_depths[KNAV_DMA_FDQ_PER_CHAN];
+};
+
+#define        NETCP_PSDATA_LEN                KNAV_DMA_NUM_PS_WORDS
+struct netcp_packet {
+       struct sk_buff          *skb;
+       u32                     *epib;
+       u32                     *psdata;
+       unsigned int            psdata_len;
+       struct netcp_intf       *netcp;
+       struct netcp_tx_pipe    *tx_pipe;
+       bool                    rxtstamp_complete;
+       void                    *ts_context;
+
+       int     (*txtstamp_complete)(void *ctx, struct netcp_packet *pkt);
+};
+
+static inline u32 *netcp_push_psdata(struct netcp_packet *p_info,
+                                    unsigned int bytes)
+{
+       u32 *buf;
+       unsigned int words;
+
+       if ((bytes & 0x03) != 0)
+               return NULL;
+       words = bytes >> 2;
+
+       if ((p_info->psdata_len + words) > NETCP_PSDATA_LEN)
+               return NULL;
+
+       p_info->psdata_len += words;
+       buf = &p_info->psdata[NETCP_PSDATA_LEN - p_info->psdata_len];
+       return buf;
+}
+
+static inline int netcp_align_psdata(struct netcp_packet *p_info,
+                                    unsigned int byte_align)
+{
+       int padding;
+
+       switch (byte_align) {
+       case 0:
+               padding = -EINVAL;
+               break;
+       case 1:
+       case 2:
+       case 4:
+               padding = 0;
+               break;
+       case 8:
+               padding = (p_info->psdata_len << 2) % 8;
+               break;
+       case 16:
+               padding = (p_info->psdata_len << 2) % 16;
+               break;
+       default:
+               padding = (p_info->psdata_len << 2) % byte_align;
+               break;
+       }
+       return padding;
+}
+
+struct netcp_module {
+       const char              *name;
+       struct module           *owner;
+       bool                    primary;
+
+       /* probe/remove: called once per NETCP instance */
+       int     (*probe)(struct netcp_device *netcp_device,
+                        struct device *device, struct device_node *node,
+                        void **inst_priv);
+       int     (*remove)(struct netcp_device *netcp_device, void *inst_priv);
+
+       /* attach/release: called once per network interface */
+       int     (*attach)(void *inst_priv, struct net_device *ndev,
+                         struct device_node *node, void **intf_priv);
+       int     (*release)(void *intf_priv);
+       int     (*open)(void *intf_priv, struct net_device *ndev);
+       int     (*close)(void *intf_priv, struct net_device *ndev);
+       int     (*add_addr)(void *intf_priv, struct netcp_addr *naddr);
+       int     (*del_addr)(void *intf_priv, struct netcp_addr *naddr);
+       int     (*add_vid)(void *intf_priv, int vid);
+       int     (*del_vid)(void *intf_priv, int vid);
+       int     (*ioctl)(void *intf_priv, struct ifreq *req, int cmd);
+
+       /* used internally */
+       struct list_head        module_list;
+       struct list_head        interface_list;
+};
+
+int netcp_register_module(struct netcp_module *module);
+void netcp_unregister_module(struct netcp_module *module);
+void *netcp_module_get_intf_data(struct netcp_module *module,
+                                struct netcp_intf *intf);
+
+int netcp_txpipe_init(struct netcp_tx_pipe *tx_pipe,
+                     struct netcp_device *netcp_device,
+                     const char *dma_chan_name, unsigned int dma_queue_id);
+int netcp_txpipe_open(struct netcp_tx_pipe *tx_pipe);
+int netcp_txpipe_close(struct netcp_tx_pipe *tx_pipe);
+
+typedef int netcp_hook_rtn(int order, void *data, struct netcp_packet *packet);
+int netcp_register_txhook(struct netcp_intf *netcp_priv, int order,
+                         netcp_hook_rtn *hook_rtn, void *hook_data);
+int netcp_unregister_txhook(struct netcp_intf *netcp_priv, int order,
+                           netcp_hook_rtn *hook_rtn, void *hook_data);
+int netcp_register_rxhook(struct netcp_intf *netcp_priv, int order,
+                         netcp_hook_rtn *hook_rtn, void *hook_data);
+int netcp_unregister_rxhook(struct netcp_intf *netcp_priv, int order,
+                           netcp_hook_rtn *hook_rtn, void *hook_data);
+void *netcp_device_find_module(struct netcp_device *netcp_device,
+                              const char *name);
+
+/* SGMII functions */
+int netcp_sgmii_reset(void __iomem *sgmii_ofs, int port);
+int netcp_sgmii_get_port_link(void __iomem *sgmii_ofs, int port);
+int netcp_sgmii_config(void __iomem *sgmii_ofs, int port, u32 interface);
+
+/* XGBE SERDES init functions */
+int netcp_xgbe_serdes_init(void __iomem *serdes_regs, void __iomem *xgbe_regs);
+
+#endif /* __NETCP_H__ */
diff --git a/drivers/net/ethernet/ti/netcp_core.c b/drivers/net/ethernet/ti/netcp_core.c
new file mode 100644 (file)
index 0000000..ba3002e
--- /dev/null
@@ -0,0 +1,2141 @@
+/*
+ * Keystone NetCP Core driver
+ *
+ * Copyright (C) 2014 Texas Instruments Incorporated
+ * Authors:    Sandeep Nair <sandeep_n@ti.com>
+ *             Sandeep Paulraj <s-paulraj@ti.com>
+ *             Cyril Chemparathy <cyril@ti.com>
+ *             Santosh Shilimkar <santosh.shilimkar@ti.com>
+ *             Murali Karicheri <m-karicheri2@ti.com>
+ *             Wingman Kwok <w-kwok2@ti.com>
+ *
+ * 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 version 2.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/of_net.h>
+#include <linux/of_address.h>
+#include <linux/if_vlan.h>
+#include <linux/pm_runtime.h>
+#include <linux/platform_device.h>
+#include <linux/soc/ti/knav_qmss.h>
+#include <linux/soc/ti/knav_dma.h>
+
+#include "netcp.h"
+
+#define NETCP_SOP_OFFSET       (NET_IP_ALIGN + NET_SKB_PAD)
+#define NETCP_NAPI_WEIGHT      64
+#define NETCP_TX_TIMEOUT       (5 * HZ)
+#define NETCP_MIN_PACKET_SIZE  ETH_ZLEN
+#define NETCP_MAX_MCAST_ADDR   16
+
+#define NETCP_EFUSE_REG_INDEX  0
+
+#define NETCP_MOD_PROBE_SKIPPED        1
+#define NETCP_MOD_PROBE_FAILED 2
+
+#define NETCP_DEBUG (NETIF_MSG_HW      | NETIF_MSG_WOL         |       \
+                   NETIF_MSG_DRV       | NETIF_MSG_LINK        |       \
+                   NETIF_MSG_IFUP      | NETIF_MSG_INTR        |       \
+                   NETIF_MSG_PROBE     | NETIF_MSG_TIMER       |       \
+                   NETIF_MSG_IFDOWN    | NETIF_MSG_RX_ERR      |       \
+                   NETIF_MSG_TX_ERR    | NETIF_MSG_TX_DONE     |       \
+                   NETIF_MSG_PKTDATA   | NETIF_MSG_TX_QUEUED   |       \
+                   NETIF_MSG_RX_STATUS)
+
+#define knav_queue_get_id(q)   knav_queue_device_control(q, \
+                               KNAV_QUEUE_GET_ID, (unsigned long)NULL)
+
+#define knav_queue_enable_notify(q) knav_queue_device_control(q,       \
+                                       KNAV_QUEUE_ENABLE_NOTIFY,       \
+                                       (unsigned long)NULL)
+
+#define knav_queue_disable_notify(q) knav_queue_device_control(q,      \
+                                       KNAV_QUEUE_DISABLE_NOTIFY,      \
+                                       (unsigned long)NULL)
+
+#define knav_queue_get_count(q)        knav_queue_device_control(q, \
+                               KNAV_QUEUE_GET_COUNT, (unsigned long)NULL)
+
+#define for_each_netcp_module(module)                  \
+       list_for_each_entry(module, &netcp_modules, module_list)
+
+#define for_each_netcp_device_module(netcp_device, inst_modpriv) \
+       list_for_each_entry(inst_modpriv, \
+               &((netcp_device)->modpriv_head), inst_list)
+
+#define for_each_module(netcp, intf_modpriv)                   \
+       list_for_each_entry(intf_modpriv, &netcp->module_head, intf_list)
+
+/* Module management structures */
+struct netcp_device {
+       struct list_head        device_list;
+       struct list_head        interface_head;
+       struct list_head        modpriv_head;
+       struct device           *device;
+};
+
+struct netcp_inst_modpriv {
+       struct netcp_device     *netcp_device;
+       struct netcp_module     *netcp_module;
+       struct list_head        inst_list;
+       void                    *module_priv;
+};
+
+struct netcp_intf_modpriv {
+       struct netcp_intf       *netcp_priv;
+       struct netcp_module     *netcp_module;
+       struct list_head        intf_list;
+       void                    *module_priv;
+};
+
+static LIST_HEAD(netcp_devices);
+static LIST_HEAD(netcp_modules);
+static DEFINE_MUTEX(netcp_modules_lock);
+
+static int netcp_debug_level = -1;
+module_param(netcp_debug_level, int, 0);
+MODULE_PARM_DESC(netcp_debug_level, "Netcp debug level (NETIF_MSG bits) (0=none,...,16=all)");
+
+/* Helper functions - Get/Set */
+static void get_pkt_info(u32 *buff, u32 *buff_len, u32 *ndesc,
+                        struct knav_dma_desc *desc)
+{
+       *buff_len = desc->buff_len;
+       *buff = desc->buff;
+       *ndesc = desc->next_desc;
+}
+
+static void get_pad_info(u32 *pad0, u32 *pad1, struct knav_dma_desc *desc)
+{
+       *pad0 = desc->pad[0];
+       *pad1 = desc->pad[1];
+}
+
+static void get_org_pkt_info(u32 *buff, u32 *buff_len,
+                            struct knav_dma_desc *desc)
+{
+       *buff = desc->orig_buff;
+       *buff_len = desc->orig_len;
+}
+
+static void get_words(u32 *words, int num_words, u32 *desc)
+{
+       int i;
+
+       for (i = 0; i < num_words; i++)
+               words[i] = desc[i];
+}
+
+static void set_pkt_info(u32 buff, u32 buff_len, u32 ndesc,
+                        struct knav_dma_desc *desc)
+{
+       desc->buff_len = buff_len;
+       desc->buff = buff;
+       desc->next_desc = ndesc;
+}
+
+static void set_desc_info(u32 desc_info, u32 pkt_info,
+                         struct knav_dma_desc *desc)
+{
+       desc->desc_info = desc_info;
+       desc->packet_info = pkt_info;
+}
+
+static void set_pad_info(u32 pad0, u32 pad1, struct knav_dma_desc *desc)
+{
+       desc->pad[0] = pad0;
+       desc->pad[1] = pad1;
+}
+
+static void set_org_pkt_info(u32 buff, u32 buff_len,
+                            struct knav_dma_desc *desc)
+{
+       desc->orig_buff = buff;
+       desc->orig_len = buff_len;
+}
+
+static void set_words(u32 *words, int num_words, u32 *desc)
+{
+       int i;
+
+       for (i = 0; i < num_words; i++)
+               desc[i] = words[i];
+}
+
+/* Read the e-fuse value as 32 bit values to be endian independent */
+static int emac_arch_get_mac_addr(char *x, void __iomem *efuse_mac)
+{
+       unsigned int addr0, addr1;
+
+       addr1 = readl(efuse_mac + 4);
+       addr0 = readl(efuse_mac);
+
+       x[0] = (addr1 & 0x0000ff00) >> 8;
+       x[1] = addr1 & 0x000000ff;
+       x[2] = (addr0 & 0xff000000) >> 24;
+       x[3] = (addr0 & 0x00ff0000) >> 16;
+       x[4] = (addr0 & 0x0000ff00) >> 8;
+       x[5] = addr0 & 0x000000ff;
+
+       return 0;
+}
+
+static const char *netcp_node_name(struct device_node *node)
+{
+       const char *name;
+
+       if (of_property_read_string(node, "label", &name) < 0)
+               name = node->name;
+       if (!name)
+               name = "unknown";
+       return name;
+}
+
+/* Module management routines */
+static int netcp_register_interface(struct netcp_intf *netcp)
+{
+       int ret;
+
+       ret = register_netdev(netcp->ndev);
+       if (!ret)
+               netcp->netdev_registered = true;
+       return ret;
+}
+
+static int netcp_module_probe(struct netcp_device *netcp_device,
+                             struct netcp_module *module)
+{
+       struct device *dev = netcp_device->device;
+       struct device_node *devices, *interface, *node = dev->of_node;
+       struct device_node *child;
+       struct netcp_inst_modpriv *inst_modpriv;
+       struct netcp_intf *netcp_intf;
+       struct netcp_module *tmp;
+       bool primary_module_registered = false;
+       int ret;
+
+       /* Find this module in the sub-tree for this device */
+       devices = of_get_child_by_name(node, "netcp-devices");
+       if (!devices) {
+               dev_err(dev, "could not find netcp-devices node\n");
+               return NETCP_MOD_PROBE_SKIPPED;
+       }
+
+       for_each_available_child_of_node(devices, child) {
+               const char *name = netcp_node_name(child);
+
+               if (!strcasecmp(module->name, name))
+                       break;
+       }
+
+       of_node_put(devices);
+       /* If module not used for this device, skip it */
+       if (!child) {
+               dev_warn(dev, "module(%s) not used for device\n", module->name);
+               return NETCP_MOD_PROBE_SKIPPED;
+       }
+
+       inst_modpriv = devm_kzalloc(dev, sizeof(*inst_modpriv), GFP_KERNEL);
+       if (!inst_modpriv) {
+               of_node_put(child);
+               return -ENOMEM;
+       }
+
+       inst_modpriv->netcp_device = netcp_device;
+       inst_modpriv->netcp_module = module;
+       list_add_tail(&inst_modpriv->inst_list, &netcp_device->modpriv_head);
+
+       ret = module->probe(netcp_device, dev, child,
+                           &inst_modpriv->module_priv);
+       of_node_put(child);
+       if (ret) {
+               dev_err(dev, "Probe of module(%s) failed with %d\n",
+                       module->name, ret);
+               list_del(&inst_modpriv->inst_list);
+               devm_kfree(dev, inst_modpriv);
+               return NETCP_MOD_PROBE_FAILED;
+       }
+
+       /* Attach modules only if the primary module is probed */
+       for_each_netcp_module(tmp) {
+               if (tmp->primary)
+                       primary_module_registered = true;
+       }
+
+       if (!primary_module_registered)
+               return 0;
+
+       /* Attach module to interfaces */
+       list_for_each_entry(netcp_intf, &netcp_device->interface_head,
+                           interface_list) {
+               struct netcp_intf_modpriv *intf_modpriv;
+
+               /* If interface not registered then register now */
+               if (!netcp_intf->netdev_registered)
+                       ret = netcp_register_interface(netcp_intf);
+
+               if (ret)
+                       return -ENODEV;
+
+               intf_modpriv = devm_kzalloc(dev, sizeof(*intf_modpriv),
+                                           GFP_KERNEL);
+               if (!intf_modpriv)
+                       return -ENOMEM;
+
+               interface = of_parse_phandle(netcp_intf->node_interface,
+                                            module->name, 0);
+
+               intf_modpriv->netcp_priv = netcp_intf;
+               intf_modpriv->netcp_module = module;
+               list_add_tail(&intf_modpriv->intf_list,
+                             &netcp_intf->module_head);
+
+               ret = module->attach(inst_modpriv->module_priv,
+                                    netcp_intf->ndev, interface,
+                                    &intf_modpriv->module_priv);
+               of_node_put(interface);
+               if (ret) {
+                       dev_dbg(dev, "Attach of module %s declined with %d\n",
+                               module->name, ret);
+                       list_del(&intf_modpriv->intf_list);
+                       devm_kfree(dev, intf_modpriv);
+                       continue;
+               }
+       }
+       return 0;
+}
+
+int netcp_register_module(struct netcp_module *module)
+{
+       struct netcp_device *netcp_device;
+       struct netcp_module *tmp;
+       int ret;
+
+       if (!module->name) {
+               WARN(1, "error registering netcp module: no name\n");
+               return -EINVAL;
+       }
+
+       if (!module->probe) {
+               WARN(1, "error registering netcp module: no probe\n");
+               return -EINVAL;
+       }
+
+       mutex_lock(&netcp_modules_lock);
+
+       for_each_netcp_module(tmp) {
+               if (!strcasecmp(tmp->name, module->name)) {
+                       mutex_unlock(&netcp_modules_lock);
+                       return -EEXIST;
+               }
+       }
+       list_add_tail(&module->module_list, &netcp_modules);
+
+       list_for_each_entry(netcp_device, &netcp_devices, device_list) {
+               ret = netcp_module_probe(netcp_device, module);
+               if (ret < 0)
+                       goto fail;
+       }
+
+       mutex_unlock(&netcp_modules_lock);
+       return 0;
+
+fail:
+       mutex_unlock(&netcp_modules_lock);
+       netcp_unregister_module(module);
+       return ret;
+}
+
+static void netcp_release_module(struct netcp_device *netcp_device,
+                                struct netcp_module *module)
+{
+       struct netcp_inst_modpriv *inst_modpriv, *inst_tmp;
+       struct netcp_intf *netcp_intf, *netcp_tmp;
+       struct device *dev = netcp_device->device;
+
+       /* Release the module from each interface */
+       list_for_each_entry_safe(netcp_intf, netcp_tmp,
+                                &netcp_device->interface_head,
+                                interface_list) {
+               struct netcp_intf_modpriv *intf_modpriv, *intf_tmp;
+
+               list_for_each_entry_safe(intf_modpriv, intf_tmp,
+                                        &netcp_intf->module_head,
+                                        intf_list) {
+                       if (intf_modpriv->netcp_module == module) {
+                               module->release(intf_modpriv->module_priv);
+                               list_del(&intf_modpriv->intf_list);
+                               devm_kfree(dev, intf_modpriv);
+                               break;
+                       }
+               }
+       }
+
+       /* Remove the module from each instance */
+       list_for_each_entry_safe(inst_modpriv, inst_tmp,
+                                &netcp_device->modpriv_head, inst_list) {
+               if (inst_modpriv->netcp_module == module) {
+                       module->remove(netcp_device,
+                                      inst_modpriv->module_priv);
+                       list_del(&inst_modpriv->inst_list);
+                       devm_kfree(dev, inst_modpriv);
+                       break;
+               }
+       }
+}
+
+void netcp_unregister_module(struct netcp_module *module)
+{
+       struct netcp_device *netcp_device;
+       struct netcp_module *module_tmp;
+
+       mutex_lock(&netcp_modules_lock);
+
+       list_for_each_entry(netcp_device, &netcp_devices, device_list) {
+               netcp_release_module(netcp_device, module);
+       }
+
+       /* Remove the module from the module list */
+       for_each_netcp_module(module_tmp) {
+               if (module == module_tmp) {
+                       list_del(&module->module_list);
+                       break;
+               }
+       }
+
+       mutex_unlock(&netcp_modules_lock);
+}
+
+void *netcp_module_get_intf_data(struct netcp_module *module,
+                                struct netcp_intf *intf)
+{
+       struct netcp_intf_modpriv *intf_modpriv;
+
+       list_for_each_entry(intf_modpriv, &intf->module_head, intf_list)
+               if (intf_modpriv->netcp_module == module)
+                       return intf_modpriv->module_priv;
+       return NULL;
+}
+
+/* Module TX and RX Hook management */
+struct netcp_hook_list {
+       struct list_head         list;
+       netcp_hook_rtn          *hook_rtn;
+       void                    *hook_data;
+       int                      order;
+};
+
+int netcp_register_txhook(struct netcp_intf *netcp_priv, int order,
+                         netcp_hook_rtn *hook_rtn, void *hook_data)
+{
+       struct netcp_hook_list *entry;
+       struct netcp_hook_list *next;
+       unsigned long flags;
+
+       entry = devm_kzalloc(netcp_priv->dev, sizeof(*entry), GFP_KERNEL);
+       if (!entry)
+               return -ENOMEM;
+
+       entry->hook_rtn  = hook_rtn;
+       entry->hook_data = hook_data;
+       entry->order     = order;
+
+       spin_lock_irqsave(&netcp_priv->lock, flags);
+       list_for_each_entry(next, &netcp_priv->txhook_list_head, list) {
+               if (next->order > order)
+                       break;
+       }
+       __list_add(&entry->list, next->list.prev, &next->list);
+       spin_unlock_irqrestore(&netcp_priv->lock, flags);
+
+       return 0;
+}
+
+int netcp_unregister_txhook(struct netcp_intf *netcp_priv, int order,
+                           netcp_hook_rtn *hook_rtn, void *hook_data)
+{
+       struct netcp_hook_list *next, *n;
+       unsigned long flags;
+
+       spin_lock_irqsave(&netcp_priv->lock, flags);
+       list_for_each_entry_safe(next, n, &netcp_priv->txhook_list_head, list) {
+               if ((next->order     == order) &&
+                   (next->hook_rtn  == hook_rtn) &&
+                   (next->hook_data == hook_data)) {
+                       list_del(&next->list);
+                       spin_unlock_irqrestore(&netcp_priv->lock, flags);
+                       devm_kfree(netcp_priv->dev, next);
+                       return 0;
+               }
+       }
+       spin_unlock_irqrestore(&netcp_priv->lock, flags);
+       return -ENOENT;
+}
+
+int netcp_register_rxhook(struct netcp_intf *netcp_priv, int order,
+                         netcp_hook_rtn *hook_rtn, void *hook_data)
+{
+       struct netcp_hook_list *entry;
+       struct netcp_hook_list *next;
+       unsigned long flags;
+
+       entry = devm_kzalloc(netcp_priv->dev, sizeof(*entry), GFP_KERNEL);
+       if (!entry)
+               return -ENOMEM;
+
+       entry->hook_rtn  = hook_rtn;
+       entry->hook_data = hook_data;
+       entry->order     = order;
+
+       spin_lock_irqsave(&netcp_priv->lock, flags);
+       list_for_each_entry(next, &netcp_priv->rxhook_list_head, list) {
+               if (next->order > order)
+                       break;
+       }
+       __list_add(&entry->list, next->list.prev, &next->list);
+       spin_unlock_irqrestore(&netcp_priv->lock, flags);
+
+       return 0;
+}
+
+int netcp_unregister_rxhook(struct netcp_intf *netcp_priv, int order,
+                           netcp_hook_rtn *hook_rtn, void *hook_data)
+{
+       struct netcp_hook_list *next, *n;
+       unsigned long flags;
+
+       spin_lock_irqsave(&netcp_priv->lock, flags);
+       list_for_each_entry_safe(next, n, &netcp_priv->rxhook_list_head, list) {
+               if ((next->order     == order) &&
+                   (next->hook_rtn  == hook_rtn) &&
+                   (next->hook_data == hook_data)) {
+                       list_del(&next->list);
+                       spin_unlock_irqrestore(&netcp_priv->lock, flags);
+                       devm_kfree(netcp_priv->dev, next);
+                       return 0;
+               }
+       }
+       spin_unlock_irqrestore(&netcp_priv->lock, flags);
+
+       return -ENOENT;
+}
+
+static void netcp_frag_free(bool is_frag, void *ptr)
+{
+       if (is_frag)
+               put_page(virt_to_head_page(ptr));
+       else
+               kfree(ptr);
+}
+
+static void netcp_free_rx_desc_chain(struct netcp_intf *netcp,
+                                    struct knav_dma_desc *desc)
+{
+       struct knav_dma_desc *ndesc;
+       dma_addr_t dma_desc, dma_buf;
+       unsigned int buf_len, dma_sz = sizeof(*ndesc);
+       void *buf_ptr;
+       u32 tmp;
+
+       get_words(&dma_desc, 1, &desc->next_desc);
+
+       while (dma_desc) {
+               ndesc = knav_pool_desc_unmap(netcp->rx_pool, dma_desc, dma_sz);
+               if (unlikely(!ndesc)) {
+                       dev_err(netcp->ndev_dev, "failed to unmap Rx desc\n");
+                       break;
+               }
+               get_pkt_info(&dma_buf, &tmp, &dma_desc, ndesc);
+               get_pad_info((u32 *)&buf_ptr, &tmp, ndesc);
+               dma_unmap_page(netcp->dev, dma_buf, PAGE_SIZE, DMA_FROM_DEVICE);
+               __free_page(buf_ptr);
+               knav_pool_desc_put(netcp->rx_pool, desc);
+       }
+
+       get_pad_info((u32 *)&buf_ptr, &buf_len, desc);
+       if (buf_ptr)
+               netcp_frag_free(buf_len <= PAGE_SIZE, buf_ptr);
+       knav_pool_desc_put(netcp->rx_pool, desc);
+}
+
+static void netcp_empty_rx_queue(struct netcp_intf *netcp)
+{
+       struct knav_dma_desc *desc;
+       unsigned int dma_sz;
+       dma_addr_t dma;
+
+       for (; ;) {
+               dma = knav_queue_pop(netcp->rx_queue, &dma_sz);
+               if (!dma)
+                       break;
+
+               desc = knav_pool_desc_unmap(netcp->rx_pool, dma, dma_sz);
+               if (unlikely(!desc)) {
+                       dev_err(netcp->ndev_dev, "%s: failed to unmap Rx desc\n",
+                               __func__);
+                       netcp->ndev->stats.rx_errors++;
+                       continue;
+               }
+               netcp_free_rx_desc_chain(netcp, desc);
+               netcp->ndev->stats.rx_dropped++;
+       }
+}
+
+static int netcp_process_one_rx_packet(struct netcp_intf *netcp)
+{
+       unsigned int dma_sz, buf_len, org_buf_len;
+       struct knav_dma_desc *desc, *ndesc;
+       unsigned int pkt_sz = 0, accum_sz;
+       struct netcp_hook_list *rx_hook;
+       dma_addr_t dma_desc, dma_buff;
+       struct netcp_packet p_info;
+       struct sk_buff *skb;
+       void *org_buf_ptr;
+       u32 tmp;
+
+       dma_desc = knav_queue_pop(netcp->rx_queue, &dma_sz);
+       if (!dma_desc)
+               return -1;
+
+       desc = knav_pool_desc_unmap(netcp->rx_pool, dma_desc, dma_sz);
+       if (unlikely(!desc)) {
+               dev_err(netcp->ndev_dev, "failed to unmap Rx desc\n");
+               return 0;
+       }
+
+       get_pkt_info(&dma_buff, &buf_len, &dma_desc, desc);
+       get_pad_info((u32 *)&org_buf_ptr, &org_buf_len, desc);
+
+       if (unlikely(!org_buf_ptr)) {
+               dev_err(netcp->ndev_dev, "NULL bufptr in desc\n");
+               goto free_desc;
+       }
+
+       pkt_sz &= KNAV_DMA_DESC_PKT_LEN_MASK;
+       accum_sz = buf_len;
+       dma_unmap_single(netcp->dev, dma_buff, buf_len, DMA_FROM_DEVICE);
+
+       /* Build a new sk_buff for the primary buffer */
+       skb = build_skb(org_buf_ptr, org_buf_len);
+       if (unlikely(!skb)) {
+               dev_err(netcp->ndev_dev, "build_skb() failed\n");
+               goto free_desc;
+       }
+
+       /* update data, tail and len */
+       skb_reserve(skb, NETCP_SOP_OFFSET);
+       __skb_put(skb, buf_len);
+
+       /* Fill in the page fragment list */
+       while (dma_desc) {
+               struct page *page;
+
+               ndesc = knav_pool_desc_unmap(netcp->rx_pool, dma_desc, dma_sz);
+               if (unlikely(!ndesc)) {
+                       dev_err(netcp->ndev_dev, "failed to unmap Rx desc\n");
+                       goto free_desc;
+               }
+
+               get_pkt_info(&dma_buff, &buf_len, &dma_desc, ndesc);
+               get_pad_info((u32 *)&page, &tmp, ndesc);
+
+               if (likely(dma_buff && buf_len && page)) {
+                       dma_unmap_page(netcp->dev, dma_buff, PAGE_SIZE,
+                                      DMA_FROM_DEVICE);
+               } else {
+                       dev_err(netcp->ndev_dev, "Bad Rx desc dma_buff(%p), len(%d), page(%p)\n",
+                               (void *)dma_buff, buf_len, page);
+                       goto free_desc;
+               }
+
+               skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
+                               offset_in_page(dma_buff), buf_len, PAGE_SIZE);
+               accum_sz += buf_len;
+
+               /* Free the descriptor */
+               knav_pool_desc_put(netcp->rx_pool, ndesc);
+       }
+
+       /* Free the primary descriptor */
+       knav_pool_desc_put(netcp->rx_pool, desc);
+
+       /* check for packet len and warn */
+       if (unlikely(pkt_sz != accum_sz))
+               dev_dbg(netcp->ndev_dev, "mismatch in packet size(%d) & sum of fragments(%d)\n",
+                       pkt_sz, accum_sz);
+
+       /* Remove ethernet FCS from the packet */
+       __pskb_trim(skb, skb->len - ETH_FCS_LEN);
+
+       /* Call each of the RX hooks */
+       p_info.skb = skb;
+       p_info.rxtstamp_complete = false;
+       list_for_each_entry(rx_hook, &netcp->rxhook_list_head, list) {
+               int ret;
+
+               ret = rx_hook->hook_rtn(rx_hook->order, rx_hook->hook_data,
+                                       &p_info);
+               if (unlikely(ret)) {
+                       dev_err(netcp->ndev_dev, "RX hook %d failed: %d\n",
+                               rx_hook->order, ret);
+                       netcp->ndev->stats.rx_errors++;
+                       dev_kfree_skb(skb);
+                       return 0;
+               }
+       }
+
+       netcp->ndev->last_rx = jiffies;
+       netcp->ndev->stats.rx_packets++;
+       netcp->ndev->stats.rx_bytes += skb->len;
+
+       /* push skb up the stack */
+       skb->protocol = eth_type_trans(skb, netcp->ndev);
+       netif_receive_skb(skb);
+       return 0;
+
+free_desc:
+       netcp_free_rx_desc_chain(netcp, desc);
+       netcp->ndev->stats.rx_errors++;
+       return 0;
+}
+
+static int netcp_process_rx_packets(struct netcp_intf *netcp,
+                                   unsigned int budget)
+{
+       int i;
+
+       for (i = 0; (i < budget) && !netcp_process_one_rx_packet(netcp); i++)
+               ;
+       return i;
+}
+
+/* Release descriptors and attached buffers from Rx FDQ */
+static void netcp_free_rx_buf(struct netcp_intf *netcp, int fdq)
+{
+       struct knav_dma_desc *desc;
+       unsigned int buf_len, dma_sz;
+       dma_addr_t dma;
+       void *buf_ptr;
+       u32 tmp;
+
+       /* Allocate descriptor */
+       while ((dma = knav_queue_pop(netcp->rx_fdq[fdq], &dma_sz))) {
+               desc = knav_pool_desc_unmap(netcp->rx_pool, dma, dma_sz);
+               if (unlikely(!desc)) {
+                       dev_err(netcp->ndev_dev, "failed to unmap Rx desc\n");
+                       continue;
+               }
+
+               get_org_pkt_info(&dma, &buf_len, desc);
+               get_pad_info((u32 *)&buf_ptr, &tmp, desc);
+
+               if (unlikely(!dma)) {
+                       dev_err(netcp->ndev_dev, "NULL orig_buff in desc\n");
+                       knav_pool_desc_put(netcp->rx_pool, desc);
+                       continue;
+               }
+
+               if (unlikely(!buf_ptr)) {
+                       dev_err(netcp->ndev_dev, "NULL bufptr in desc\n");
+                       knav_pool_desc_put(netcp->rx_pool, desc);
+                       continue;
+               }
+
+               if (fdq == 0) {
+                       dma_unmap_single(netcp->dev, dma, buf_len,
+                                        DMA_FROM_DEVICE);
+                       netcp_frag_free((buf_len <= PAGE_SIZE), buf_ptr);
+               } else {
+                       dma_unmap_page(netcp->dev, dma, buf_len,
+                                      DMA_FROM_DEVICE);
+                       __free_page(buf_ptr);
+               }
+
+               knav_pool_desc_put(netcp->rx_pool, desc);
+       }
+}
+
+static void netcp_rxpool_free(struct netcp_intf *netcp)
+{
+       int i;
+
+       for (i = 0; i < KNAV_DMA_FDQ_PER_CHAN &&
+            !IS_ERR_OR_NULL(netcp->rx_fdq[i]); i++)
+               netcp_free_rx_buf(netcp, i);
+
+       if (knav_pool_count(netcp->rx_pool) != netcp->rx_pool_size)
+               dev_err(netcp->ndev_dev, "Lost Rx (%d) descriptors\n",
+                       netcp->rx_pool_size - knav_pool_count(netcp->rx_pool));
+
+       knav_pool_destroy(netcp->rx_pool);
+       netcp->rx_pool = NULL;
+}
+
+static void netcp_allocate_rx_buf(struct netcp_intf *netcp, int fdq)
+{
+       struct knav_dma_desc *hwdesc;
+       unsigned int buf_len, dma_sz;
+       u32 desc_info, pkt_info;
+       struct page *page;
+       dma_addr_t dma;
+       void *bufptr;
+       u32 pad[2];
+
+       /* Allocate descriptor */
+       hwdesc = knav_pool_desc_get(netcp->rx_pool);
+       if (IS_ERR_OR_NULL(hwdesc)) {
+               dev_dbg(netcp->ndev_dev, "out of rx pool desc\n");
+               return;
+       }
+
+       if (likely(fdq == 0)) {
+               unsigned int primary_buf_len;
+               /* Allocate a primary receive queue entry */
+               buf_len = netcp->rx_buffer_sizes[0] + NETCP_SOP_OFFSET;
+               primary_buf_len = SKB_DATA_ALIGN(buf_len) +
+                               SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
+
+               if (primary_buf_len <= PAGE_SIZE) {
+                       bufptr = netdev_alloc_frag(primary_buf_len);
+                       pad[1] = primary_buf_len;
+               } else {
+                       bufptr = kmalloc(primary_buf_len, GFP_ATOMIC |
+                                        GFP_DMA32 | __GFP_COLD);
+                       pad[1] = 0;
+               }
+
+               if (unlikely(!bufptr)) {
+                       dev_warn_ratelimited(netcp->ndev_dev, "Primary RX buffer alloc failed\n");
+                       goto fail;
+               }
+               dma = dma_map_single(netcp->dev, bufptr, buf_len,
+                                    DMA_TO_DEVICE);
+               pad[0] = (u32)bufptr;
+
+       } else {
+               /* Allocate a secondary receive queue entry */
+               page = alloc_page(GFP_ATOMIC | GFP_DMA32 | __GFP_COLD);
+               if (unlikely(!page)) {
+                       dev_warn_ratelimited(netcp->ndev_dev, "Secondary page alloc failed\n");
+                       goto fail;
+               }
+               buf_len = PAGE_SIZE;
+               dma = dma_map_page(netcp->dev, page, 0, buf_len, DMA_TO_DEVICE);
+               pad[0] = (u32)page;
+               pad[1] = 0;
+       }
+
+       desc_info =  KNAV_DMA_DESC_PS_INFO_IN_DESC;
+       desc_info |= buf_len & KNAV_DMA_DESC_PKT_LEN_MASK;
+       pkt_info =  KNAV_DMA_DESC_HAS_EPIB;
+       pkt_info |= KNAV_DMA_NUM_PS_WORDS << KNAV_DMA_DESC_PSLEN_SHIFT;
+       pkt_info |= (netcp->rx_queue_id & KNAV_DMA_DESC_RETQ_MASK) <<
+                   KNAV_DMA_DESC_RETQ_SHIFT;
+       set_org_pkt_info(dma, buf_len, hwdesc);
+       set_pad_info(pad[0], pad[1], hwdesc);
+       set_desc_info(desc_info, pkt_info, hwdesc);
+
+       /* Push to FDQs */
+       knav_pool_desc_map(netcp->rx_pool, hwdesc, sizeof(*hwdesc), &dma,
+                          &dma_sz);
+       knav_queue_push(netcp->rx_fdq[fdq], dma, sizeof(*hwdesc), 0);
+       return;
+
+fail:
+       knav_pool_desc_put(netcp->rx_pool, hwdesc);
+}
+
+/* Refill Rx FDQ with descriptors & attached buffers */
+static void netcp_rxpool_refill(struct netcp_intf *netcp)
+{
+       u32 fdq_deficit[KNAV_DMA_FDQ_PER_CHAN] = {0};
+       int i;
+
+       /* Calculate the FDQ deficit and refill */
+       for (i = 0; i < KNAV_DMA_FDQ_PER_CHAN && netcp->rx_fdq[i]; i++) {
+               fdq_deficit[i] = netcp->rx_queue_depths[i] -
+                                knav_queue_get_count(netcp->rx_fdq[i]);
+
+               while (fdq_deficit[i]--)
+                       netcp_allocate_rx_buf(netcp, i);
+       } /* end for fdqs */
+}
+
+/* NAPI poll */
+static int netcp_rx_poll(struct napi_struct *napi, int budget)
+{
+       struct netcp_intf *netcp = container_of(napi, struct netcp_intf,
+                                               rx_napi);
+       unsigned int packets;
+
+       packets = netcp_process_rx_packets(netcp, budget);
+
+       if (packets < budget) {
+               napi_complete(&netcp->rx_napi);
+               knav_queue_enable_notify(netcp->rx_queue);
+       }
+
+       netcp_rxpool_refill(netcp);
+       return packets;
+}
+
+static void netcp_rx_notify(void *arg)
+{
+       struct netcp_intf *netcp = arg;
+
+       knav_queue_disable_notify(netcp->rx_queue);
+       napi_schedule(&netcp->rx_napi);
+}
+
+static void netcp_free_tx_desc_chain(struct netcp_intf *netcp,
+                                    struct knav_dma_desc *desc,
+                                    unsigned int desc_sz)
+{
+       struct knav_dma_desc *ndesc = desc;
+       dma_addr_t dma_desc, dma_buf;
+       unsigned int buf_len;
+
+       while (ndesc) {
+               get_pkt_info(&dma_buf, &buf_len, &dma_desc, ndesc);
+
+               if (dma_buf && buf_len)
+                       dma_unmap_single(netcp->dev, dma_buf, buf_len,
+                                        DMA_TO_DEVICE);
+               else
+                       dev_warn(netcp->ndev_dev, "bad Tx desc buf(%p), len(%d)\n",
+                                (void *)dma_buf, buf_len);
+
+               knav_pool_desc_put(netcp->tx_pool, ndesc);
+               ndesc = NULL;
+               if (dma_desc) {
+                       ndesc = knav_pool_desc_unmap(netcp->tx_pool, dma_desc,
+                                                    desc_sz);
+                       if (!ndesc)
+                               dev_err(netcp->ndev_dev, "failed to unmap Tx desc\n");
+               }
+       }
+}
+
+static int netcp_process_tx_compl_packets(struct netcp_intf *netcp,
+                                         unsigned int budget)
+{
+       struct knav_dma_desc *desc;
+       struct sk_buff *skb;
+       unsigned int dma_sz;
+       dma_addr_t dma;
+       int pkts = 0;
+       u32 tmp;
+
+       while (budget--) {
+               dma = knav_queue_pop(netcp->tx_compl_q, &dma_sz);
+               if (!dma)
+                       break;
+               desc = knav_pool_desc_unmap(netcp->tx_pool, dma, dma_sz);
+               if (unlikely(!desc)) {
+                       dev_err(netcp->ndev_dev, "failed to unmap Tx desc\n");
+                       netcp->ndev->stats.tx_errors++;
+                       continue;
+               }
+
+               get_pad_info((u32 *)&skb, &tmp, desc);
+               netcp_free_tx_desc_chain(netcp, desc, dma_sz);
+               if (!skb) {
+                       dev_err(netcp->ndev_dev, "No skb in Tx desc\n");
+                       netcp->ndev->stats.tx_errors++;
+                       continue;
+               }
+
+               if (netif_subqueue_stopped(netcp->ndev, skb) &&
+                   netif_running(netcp->ndev) &&
+                   (knav_pool_count(netcp->tx_pool) >
+                   netcp->tx_resume_threshold)) {
+                       u16 subqueue = skb_get_queue_mapping(skb);
+
+                       netif_wake_subqueue(netcp->ndev, subqueue);
+               }
+
+               netcp->ndev->stats.tx_packets++;
+               netcp->ndev->stats.tx_bytes += skb->len;
+               dev_kfree_skb(skb);
+               pkts++;
+       }
+       return pkts;
+}
+
+static int netcp_tx_poll(struct napi_struct *napi, int budget)
+{
+       int packets;
+       struct netcp_intf *netcp = container_of(napi, struct netcp_intf,
+                                               tx_napi);
+
+       packets = netcp_process_tx_compl_packets(netcp, budget);
+       if (packets < budget) {
+               napi_complete(&netcp->tx_napi);
+               knav_queue_enable_notify(netcp->tx_compl_q);
+       }
+
+       return packets;
+}
+
+static void netcp_tx_notify(void *arg)
+{
+       struct netcp_intf *netcp = arg;
+
+       knav_queue_disable_notify(netcp->tx_compl_q);
+       napi_schedule(&netcp->tx_napi);
+}
+
+static struct knav_dma_desc*
+netcp_tx_map_skb(struct sk_buff *skb, struct netcp_intf *netcp)
+{
+       struct knav_dma_desc *desc, *ndesc, *pdesc;
+       unsigned int pkt_len = skb_headlen(skb);
+       struct device *dev = netcp->dev;
+       dma_addr_t dma_addr;
+       unsigned int dma_sz;
+       int i;
+
+       /* Map the linear buffer */
+       dma_addr = dma_map_single(dev, skb->data, pkt_len, DMA_TO_DEVICE);
+       if (unlikely(!dma_addr)) {
+               dev_err(netcp->ndev_dev, "Failed to map skb buffer\n");
+               return NULL;
+       }
+
+       desc = knav_pool_desc_get(netcp->tx_pool);
+       if (unlikely(IS_ERR_OR_NULL(desc))) {
+               dev_err(netcp->ndev_dev, "out of TX desc\n");
+               dma_unmap_single(dev, dma_addr, pkt_len, DMA_TO_DEVICE);
+               return NULL;
+       }
+
+       set_pkt_info(dma_addr, pkt_len, 0, desc);
+       if (skb_is_nonlinear(skb)) {
+               prefetchw(skb_shinfo(skb));
+       } else {
+               desc->next_desc = 0;
+               goto upd_pkt_len;
+       }
+
+       pdesc = desc;
+
+       /* Handle the case where skb is fragmented in pages */
+       for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+               skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+               struct page *page = skb_frag_page(frag);
+               u32 page_offset = frag->page_offset;
+               u32 buf_len = skb_frag_size(frag);
+               dma_addr_t desc_dma;
+               u32 pkt_info;
+
+               dma_addr = dma_map_page(dev, page, page_offset, buf_len,
+                                       DMA_TO_DEVICE);
+               if (unlikely(!dma_addr)) {
+                       dev_err(netcp->ndev_dev, "Failed to map skb page\n");
+                       goto free_descs;
+               }
+
+               ndesc = knav_pool_desc_get(netcp->tx_pool);
+               if (unlikely(IS_ERR_OR_NULL(ndesc))) {
+                       dev_err(netcp->ndev_dev, "out of TX desc for frags\n");
+                       dma_unmap_page(dev, dma_addr, buf_len, DMA_TO_DEVICE);
+                       goto free_descs;
+               }
+
+               desc_dma = knav_pool_desc_virt_to_dma(netcp->tx_pool,
+                                                     (void *)ndesc);
+               pkt_info =
+                       (netcp->tx_compl_qid & KNAV_DMA_DESC_RETQ_MASK) <<
+                               KNAV_DMA_DESC_RETQ_SHIFT;
+               set_pkt_info(dma_addr, buf_len, 0, ndesc);
+               set_words(&desc_dma, 1, &pdesc->next_desc);
+               pkt_len += buf_len;
+               if (pdesc != desc)
+                       knav_pool_desc_map(netcp->tx_pool, pdesc,
+                                          sizeof(*pdesc), &desc_dma, &dma_sz);
+               pdesc = ndesc;
+       }
+       if (pdesc != desc)
+               knav_pool_desc_map(netcp->tx_pool, pdesc, sizeof(*pdesc),
+                                  &dma_addr, &dma_sz);
+
+       /* frag list based linkage is not supported for now. */
+       if (skb_shinfo(skb)->frag_list) {
+               dev_err_ratelimited(netcp->ndev_dev, "NETIF_F_FRAGLIST not supported\n");
+               goto free_descs;
+       }
+
+upd_pkt_len:
+       WARN_ON(pkt_len != skb->len);
+
+       pkt_len &= KNAV_DMA_DESC_PKT_LEN_MASK;
+       set_words(&pkt_len, 1, &desc->desc_info);
+       return desc;
+
+free_descs:
+       netcp_free_tx_desc_chain(netcp, desc, sizeof(*desc));
+       return NULL;
+}
+
+static int netcp_tx_submit_skb(struct netcp_intf *netcp,
+                              struct sk_buff *skb,
+                              struct knav_dma_desc *desc)
+{
+       struct netcp_tx_pipe *tx_pipe = NULL;
+       struct netcp_hook_list *tx_hook;
+       struct netcp_packet p_info;
+       u32 packet_info = 0;
+       unsigned int dma_sz;
+       dma_addr_t dma;
+       int ret = 0;
+
+       p_info.netcp = netcp;
+       p_info.skb = skb;
+       p_info.tx_pipe = NULL;
+       p_info.psdata_len = 0;
+       p_info.ts_context = NULL;
+       p_info.txtstamp_complete = NULL;
+       p_info.epib = desc->epib;
+       p_info.psdata = desc->psdata;
+       memset(p_info.epib, 0, KNAV_DMA_NUM_EPIB_WORDS * sizeof(u32));
+
+       /* Find out where to inject the packet for transmission */
+       list_for_each_entry(tx_hook, &netcp->txhook_list_head, list) {
+               ret = tx_hook->hook_rtn(tx_hook->order, tx_hook->hook_data,
+                                       &p_info);
+               if (unlikely(ret != 0)) {
+                       dev_err(netcp->ndev_dev, "TX hook %d rejected the packet with reason(%d)\n",
+                               tx_hook->order, ret);
+                       ret = (ret < 0) ? ret : NETDEV_TX_OK;
+                       goto out;
+               }
+       }
+
+       /* Make sure some TX hook claimed the packet */
+       tx_pipe = p_info.tx_pipe;
+       if (!tx_pipe) {
+               dev_err(netcp->ndev_dev, "No TX hook claimed the packet!\n");
+               ret = -ENXIO;
+               goto out;
+       }
+
+       /* update descriptor */
+       if (p_info.psdata_len) {
+               u32 *psdata = p_info.psdata;
+
+               memmove(p_info.psdata, p_info.psdata + p_info.psdata_len,
+                       p_info.psdata_len);
+               set_words(psdata, p_info.psdata_len, psdata);
+               packet_info |=
+                       (p_info.psdata_len & KNAV_DMA_DESC_PSLEN_MASK) <<
+                       KNAV_DMA_DESC_PSLEN_SHIFT;
+       }
+
+       packet_info |= KNAV_DMA_DESC_HAS_EPIB |
+               ((netcp->tx_compl_qid & KNAV_DMA_DESC_RETQ_MASK) <<
+               KNAV_DMA_DESC_RETQ_SHIFT) |
+               ((tx_pipe->dma_psflags & KNAV_DMA_DESC_PSFLAG_MASK) <<
+               KNAV_DMA_DESC_PSFLAG_SHIFT);
+
+       set_words(&packet_info, 1, &desc->packet_info);
+       set_words((u32 *)&skb, 1, &desc->pad[0]);
+
+       /* submit packet descriptor */
+       ret = knav_pool_desc_map(netcp->tx_pool, desc, sizeof(*desc), &dma,
+                                &dma_sz);
+       if (unlikely(ret)) {
+               dev_err(netcp->ndev_dev, "%s() failed to map desc\n", __func__);
+               ret = -ENOMEM;
+               goto out;
+       }
+       skb_tx_timestamp(skb);
+       knav_queue_push(tx_pipe->dma_queue, dma, dma_sz, 0);
+
+out:
+       return ret;
+}
+
+/* Submit the packet */
+static int netcp_ndo_start_xmit(struct sk_buff *skb, struct net_device *ndev)
+{
+       struct netcp_intf *netcp = netdev_priv(ndev);
+       int subqueue = skb_get_queue_mapping(skb);
+       struct knav_dma_desc *desc;
+       int desc_count, ret = 0;
+
+       if (unlikely(skb->len <= 0)) {
+               dev_kfree_skb(skb);
+               return NETDEV_TX_OK;
+       }
+
+       if (unlikely(skb->len < NETCP_MIN_PACKET_SIZE)) {
+               ret = skb_padto(skb, NETCP_MIN_PACKET_SIZE);
+               if (ret < 0) {
+                       /* If we get here, the skb has already been dropped */
+                       dev_warn(netcp->ndev_dev, "padding failed (%d), packet dropped\n",
+                                ret);
+                       ndev->stats.tx_dropped++;
+                       return ret;
+               }
+               skb->len = NETCP_MIN_PACKET_SIZE;
+       }
+
+       desc = netcp_tx_map_skb(skb, netcp);
+       if (unlikely(!desc)) {
+               netif_stop_subqueue(ndev, subqueue);
+               ret = -ENOBUFS;
+               goto drop;
+       }
+
+       ret = netcp_tx_submit_skb(netcp, skb, desc);
+       if (ret)
+               goto drop;
+
+       ndev->trans_start = jiffies;
+
+       /* Check Tx pool count & stop subqueue if needed */
+       desc_count = knav_pool_count(netcp->tx_pool);
+       if (desc_count < netcp->tx_pause_threshold) {
+               dev_dbg(netcp->ndev_dev, "pausing tx, count(%d)\n", desc_count);
+               netif_stop_subqueue(ndev, subqueue);
+       }
+       return NETDEV_TX_OK;
+
+drop:
+       ndev->stats.tx_dropped++;
+       if (desc)
+               netcp_free_tx_desc_chain(netcp, desc, sizeof(*desc));
+       dev_kfree_skb(skb);
+       return ret;
+}
+
+int netcp_txpipe_close(struct netcp_tx_pipe *tx_pipe)
+{
+       if (tx_pipe->dma_channel) {
+               knav_dma_close_channel(tx_pipe->dma_channel);
+               tx_pipe->dma_channel = NULL;
+       }
+       return 0;
+}
+
+int netcp_txpipe_open(struct netcp_tx_pipe *tx_pipe)
+{
+       struct device *dev = tx_pipe->netcp_device->device;
+       struct knav_dma_cfg config;
+       int ret = 0;
+       u8 name[16];
+
+       memset(&config, 0, sizeof(config));
+       config.direction = DMA_MEM_TO_DEV;
+       config.u.tx.filt_einfo = false;
+       config.u.tx.filt_pswords = false;
+       config.u.tx.priority = DMA_PRIO_MED_L;
+
+       tx_pipe->dma_channel = knav_dma_open_channel(dev,
+                               tx_pipe->dma_chan_name, &config);
+       if (IS_ERR_OR_NULL(tx_pipe->dma_channel)) {
+               dev_err(dev, "failed opening tx chan(%s)\n",
+                       tx_pipe->dma_chan_name);
+               goto err;
+       }
+
+       snprintf(name, sizeof(name), "tx-pipe-%s", dev_name(dev));
+       tx_pipe->dma_queue = knav_queue_open(name, tx_pipe->dma_queue_id,
+                                            KNAV_QUEUE_SHARED);
+       if (IS_ERR(tx_pipe->dma_queue)) {
+               dev_err(dev, "Could not open DMA queue for channel \"%s\": %d\n",
+                       name, ret);
+               ret = PTR_ERR(tx_pipe->dma_queue);
+               goto err;
+       }
+
+       dev_dbg(dev, "opened tx pipe %s\n", name);
+       return 0;
+
+err:
+       if (!IS_ERR_OR_NULL(tx_pipe->dma_channel))
+               knav_dma_close_channel(tx_pipe->dma_channel);
+       tx_pipe->dma_channel = NULL;
+       return ret;
+}
+
+int netcp_txpipe_init(struct netcp_tx_pipe *tx_pipe,
+                     struct netcp_device *netcp_device,
+                     const char *dma_chan_name, unsigned int dma_queue_id)
+{
+       memset(tx_pipe, 0, sizeof(*tx_pipe));
+       tx_pipe->netcp_device = netcp_device;
+       tx_pipe->dma_chan_name = dma_chan_name;
+       tx_pipe->dma_queue_id = dma_queue_id;
+       return 0;
+}
+
+static struct netcp_addr *netcp_addr_find(struct netcp_intf *netcp,
+                                         const u8 *addr,
+                                         enum netcp_addr_type type)
+{
+       struct netcp_addr *naddr;
+
+       list_for_each_entry(naddr, &netcp->addr_list, node) {
+               if (naddr->type != type)
+                       continue;
+               if (addr && memcmp(addr, naddr->addr, ETH_ALEN))
+                       continue;
+               return naddr;
+       }
+
+       return NULL;
+}
+
+static struct netcp_addr *netcp_addr_add(struct netcp_intf *netcp,
+                                        const u8 *addr,
+                                        enum netcp_addr_type type)
+{
+       struct netcp_addr *naddr;
+
+       naddr = devm_kmalloc(netcp->dev, sizeof(*naddr), GFP_ATOMIC);
+       if (!naddr)
+               return NULL;
+
+       naddr->type = type;
+       naddr->flags = 0;
+       naddr->netcp = netcp;
+       if (addr)
+               ether_addr_copy(naddr->addr, addr);
+       else
+               memset(naddr->addr, 0, ETH_ALEN);
+       list_add_tail(&naddr->node, &netcp->addr_list);
+
+       return naddr;
+}
+
+static void netcp_addr_del(struct netcp_intf *netcp, struct netcp_addr *naddr)
+{
+       list_del(&naddr->node);
+       devm_kfree(netcp->dev, naddr);
+}
+
+static void netcp_addr_clear_mark(struct netcp_intf *netcp)
+{
+       struct netcp_addr *naddr;
+
+       list_for_each_entry(naddr, &netcp->addr_list, node)
+               naddr->flags = 0;
+}
+
+static void netcp_addr_add_mark(struct netcp_intf *netcp, const u8 *addr,
+                               enum netcp_addr_type type)
+{
+       struct netcp_addr *naddr;
+
+       naddr = netcp_addr_find(netcp, addr, type);
+       if (naddr) {
+               naddr->flags |= ADDR_VALID;
+               return;
+       }
+
+       naddr = netcp_addr_add(netcp, addr, type);
+       if (!WARN_ON(!naddr))
+               naddr->flags |= ADDR_NEW;
+}
+
+static void netcp_addr_sweep_del(struct netcp_intf *netcp)
+{
+       struct netcp_addr *naddr, *tmp;
+       struct netcp_intf_modpriv *priv;
+       struct netcp_module *module;
+       int error;
+
+       list_for_each_entry_safe(naddr, tmp, &netcp->addr_list, node) {
+               if (naddr->flags & (ADDR_VALID | ADDR_NEW))
+                       continue;
+               dev_dbg(netcp->ndev_dev, "deleting address %pM, type %x\n",
+                       naddr->addr, naddr->type);
+               mutex_lock(&netcp_modules_lock);
+               for_each_module(netcp, priv) {
+                       module = priv->netcp_module;
+                       if (!module->del_addr)
+                               continue;
+                       error = module->del_addr(priv->module_priv,
+                                                naddr);
+                       WARN_ON(error);
+               }
+               mutex_unlock(&netcp_modules_lock);
+               netcp_addr_del(netcp, naddr);
+       }
+}
+
+static void netcp_addr_sweep_add(struct netcp_intf *netcp)
+{
+       struct netcp_addr *naddr, *tmp;
+       struct netcp_intf_modpriv *priv;
+       struct netcp_module *module;
+       int error;
+
+       list_for_each_entry_safe(naddr, tmp, &netcp->addr_list, node) {
+               if (!(naddr->flags & ADDR_NEW))
+                       continue;
+               dev_dbg(netcp->ndev_dev, "adding address %pM, type %x\n",
+                       naddr->addr, naddr->type);
+               mutex_lock(&netcp_modules_lock);
+               for_each_module(netcp, priv) {
+                       module = priv->netcp_module;
+                       if (!module->add_addr)
+                               continue;
+                       error = module->add_addr(priv->module_priv, naddr);
+                       WARN_ON(error);
+               }
+               mutex_unlock(&netcp_modules_lock);
+       }
+}
+
+static void netcp_set_rx_mode(struct net_device *ndev)
+{
+       struct netcp_intf *netcp = netdev_priv(ndev);
+       struct netdev_hw_addr *ndev_addr;
+       bool promisc;
+
+       promisc = (ndev->flags & IFF_PROMISC ||
+                  ndev->flags & IFF_ALLMULTI ||
+                  netdev_mc_count(ndev) > NETCP_MAX_MCAST_ADDR);
+
+       /* first clear all marks */
+       netcp_addr_clear_mark(netcp);
+
+       /* next add new entries, mark existing ones */
+       netcp_addr_add_mark(netcp, ndev->broadcast, ADDR_BCAST);
+       for_each_dev_addr(ndev, ndev_addr)
+               netcp_addr_add_mark(netcp, ndev_addr->addr, ADDR_DEV);
+       netdev_for_each_uc_addr(ndev_addr, ndev)
+               netcp_addr_add_mark(netcp, ndev_addr->addr, ADDR_UCAST);
+       netdev_for_each_mc_addr(ndev_addr, ndev)
+               netcp_addr_add_mark(netcp, ndev_addr->addr, ADDR_MCAST);
+
+       if (promisc)
+               netcp_addr_add_mark(netcp, NULL, ADDR_ANY);
+
+       /* finally sweep and callout into modules */
+       netcp_addr_sweep_del(netcp);
+       netcp_addr_sweep_add(netcp);
+}
+
+static void netcp_free_navigator_resources(struct netcp_intf *netcp)
+{
+       int i;
+
+       if (netcp->rx_channel) {
+               knav_dma_close_channel(netcp->rx_channel);
+               netcp->rx_channel = NULL;
+       }
+
+       if (!IS_ERR_OR_NULL(netcp->rx_pool))
+               netcp_rxpool_free(netcp);
+
+       if (!IS_ERR_OR_NULL(netcp->rx_queue)) {
+               knav_queue_close(netcp->rx_queue);
+               netcp->rx_queue = NULL;
+       }
+
+       for (i = 0; i < KNAV_DMA_FDQ_PER_CHAN &&
+            !IS_ERR_OR_NULL(netcp->rx_fdq[i]) ; ++i) {
+               knav_queue_close(netcp->rx_fdq[i]);
+               netcp->rx_fdq[i] = NULL;
+       }
+
+       if (!IS_ERR_OR_NULL(netcp->tx_compl_q)) {
+               knav_queue_close(netcp->tx_compl_q);
+               netcp->tx_compl_q = NULL;
+       }
+
+       if (!IS_ERR_OR_NULL(netcp->tx_pool)) {
+               knav_pool_destroy(netcp->tx_pool);
+               netcp->tx_pool = NULL;
+       }
+}
+
+static int netcp_setup_navigator_resources(struct net_device *ndev)
+{
+       struct netcp_intf *netcp = netdev_priv(ndev);
+       struct knav_queue_notify_config notify_cfg;
+       struct knav_dma_cfg config;
+       u32 last_fdq = 0;
+       u8 name[16];
+       int ret;
+       int i;
+
+       /* Create Rx/Tx descriptor pools */
+       snprintf(name, sizeof(name), "rx-pool-%s", ndev->name);
+       netcp->rx_pool = knav_pool_create(name, netcp->rx_pool_size,
+                                               netcp->rx_pool_region_id);
+       if (IS_ERR_OR_NULL(netcp->rx_pool)) {
+               dev_err(netcp->ndev_dev, "Couldn't create rx pool\n");
+               ret = PTR_ERR(netcp->rx_pool);
+               goto fail;
+       }
+
+       snprintf(name, sizeof(name), "tx-pool-%s", ndev->name);
+       netcp->tx_pool = knav_pool_create(name, netcp->tx_pool_size,
+                                               netcp->tx_pool_region_id);
+       if (IS_ERR_OR_NULL(netcp->tx_pool)) {
+               dev_err(netcp->ndev_dev, "Couldn't create tx pool\n");
+               ret = PTR_ERR(netcp->tx_pool);
+               goto fail;
+       }
+
+       /* open Tx completion queue */
+       snprintf(name, sizeof(name), "tx-compl-%s", ndev->name);
+       netcp->tx_compl_q = knav_queue_open(name, netcp->tx_compl_qid, 0);
+       if (IS_ERR_OR_NULL(netcp->tx_compl_q)) {
+               ret = PTR_ERR(netcp->tx_compl_q);
+               goto fail;
+       }
+       netcp->tx_compl_qid = knav_queue_get_id(netcp->tx_compl_q);
+
+       /* Set notification for Tx completion */
+       notify_cfg.fn = netcp_tx_notify;
+       notify_cfg.fn_arg = netcp;
+       ret = knav_queue_device_control(netcp->tx_compl_q,
+                                       KNAV_QUEUE_SET_NOTIFIER,
+                                       (unsigned long)&notify_cfg);
+       if (ret)
+               goto fail;
+
+       knav_queue_disable_notify(netcp->tx_compl_q);
+
+       /* open Rx completion queue */
+       snprintf(name, sizeof(name), "rx-compl-%s", ndev->name);
+       netcp->rx_queue = knav_queue_open(name, netcp->rx_queue_id, 0);
+       if (IS_ERR_OR_NULL(netcp->rx_queue)) {
+               ret = PTR_ERR(netcp->rx_queue);
+               goto fail;
+       }
+       netcp->rx_queue_id = knav_queue_get_id(netcp->rx_queue);
+
+       /* Set notification for Rx completion */
+       notify_cfg.fn = netcp_rx_notify;
+       notify_cfg.fn_arg = netcp;
+       ret = knav_queue_device_control(netcp->rx_queue,
+                                       KNAV_QUEUE_SET_NOTIFIER,
+                                       (unsigned long)&notify_cfg);
+       if (ret)
+               goto fail;
+
+       knav_queue_disable_notify(netcp->rx_queue);
+
+       /* open Rx FDQs */
+       for (i = 0; i < KNAV_DMA_FDQ_PER_CHAN &&
+            netcp->rx_queue_depths[i] && netcp->rx_buffer_sizes[i]; ++i) {
+               snprintf(name, sizeof(name), "rx-fdq-%s-%d", ndev->name, i);
+               netcp->rx_fdq[i] = knav_queue_open(name, KNAV_QUEUE_GP, 0);
+               if (IS_ERR_OR_NULL(netcp->rx_fdq[i])) {
+                       ret = PTR_ERR(netcp->rx_fdq[i]);
+                       goto fail;
+               }
+       }
+
+       memset(&config, 0, sizeof(config));
+       config.direction                = DMA_DEV_TO_MEM;
+       config.u.rx.einfo_present       = true;
+       config.u.rx.psinfo_present      = true;
+       config.u.rx.err_mode            = DMA_DROP;
+       config.u.rx.desc_type           = DMA_DESC_HOST;
+       config.u.rx.psinfo_at_sop       = false;
+       config.u.rx.sop_offset          = NETCP_SOP_OFFSET;
+       config.u.rx.dst_q               = netcp->rx_queue_id;
+       config.u.rx.thresh              = DMA_THRESH_NONE;
+
+       for (i = 0; i < KNAV_DMA_FDQ_PER_CHAN; ++i) {
+               if (netcp->rx_fdq[i])
+                       last_fdq = knav_queue_get_id(netcp->rx_fdq[i]);
+               config.u.rx.fdq[i] = last_fdq;
+       }
+
+       netcp->rx_channel = knav_dma_open_channel(netcp->netcp_device->device,
+                                       netcp->dma_chan_name, &config);
+       if (IS_ERR_OR_NULL(netcp->rx_channel)) {
+               dev_err(netcp->ndev_dev, "failed opening rx chan(%s\n",
+                       netcp->dma_chan_name);
+               goto fail;
+       }
+
+       dev_dbg(netcp->ndev_dev, "opened RX channel: %p\n", netcp->rx_channel);
+       return 0;
+
+fail:
+       netcp_free_navigator_resources(netcp);
+       return ret;
+}
+
+/* Open the device */
+static int netcp_ndo_open(struct net_device *ndev)
+{
+       struct netcp_intf *netcp = netdev_priv(ndev);
+       struct netcp_intf_modpriv *intf_modpriv;
+       struct netcp_module *module;
+       int ret;
+
+       netif_carrier_off(ndev);
+       ret = netcp_setup_navigator_resources(ndev);
+       if (ret) {
+               dev_err(netcp->ndev_dev, "Failed to setup navigator resources\n");
+               goto fail;
+       }
+
+       mutex_lock(&netcp_modules_lock);
+       for_each_module(netcp, intf_modpriv) {
+               module = intf_modpriv->netcp_module;
+               if (module->open) {
+                       ret = module->open(intf_modpriv->module_priv, ndev);
+                       if (ret != 0) {
+                               dev_err(netcp->ndev_dev, "module open failed\n");
+                               goto fail_open;
+                       }
+               }
+       }
+       mutex_unlock(&netcp_modules_lock);
+
+       netcp_rxpool_refill(netcp);
+       napi_enable(&netcp->rx_napi);
+       napi_enable(&netcp->tx_napi);
+       knav_queue_enable_notify(netcp->tx_compl_q);
+       knav_queue_enable_notify(netcp->rx_queue);
+       netif_tx_wake_all_queues(ndev);
+       dev_dbg(netcp->ndev_dev, "netcp device %s opened\n", ndev->name);
+       return 0;
+
+fail_open:
+       for_each_module(netcp, intf_modpriv) {
+               module = intf_modpriv->netcp_module;
+               if (module->close)
+                       module->close(intf_modpriv->module_priv, ndev);
+       }
+       mutex_unlock(&netcp_modules_lock);
+
+fail:
+       netcp_free_navigator_resources(netcp);
+       return ret;
+}
+
+/* Close the device */
+static int netcp_ndo_stop(struct net_device *ndev)
+{
+       struct netcp_intf *netcp = netdev_priv(ndev);
+       struct netcp_intf_modpriv *intf_modpriv;
+       struct netcp_module *module;
+       int err = 0;
+
+       netif_tx_stop_all_queues(ndev);
+       netif_carrier_off(ndev);
+       netcp_addr_clear_mark(netcp);
+       netcp_addr_sweep_del(netcp);
+       knav_queue_disable_notify(netcp->rx_queue);
+       knav_queue_disable_notify(netcp->tx_compl_q);
+       napi_disable(&netcp->rx_napi);
+       napi_disable(&netcp->tx_napi);
+
+       mutex_lock(&netcp_modules_lock);
+       for_each_module(netcp, intf_modpriv) {
+               module = intf_modpriv->netcp_module;
+               if (module->close) {
+                       err = module->close(intf_modpriv->module_priv, ndev);
+                       if (err != 0)
+                               dev_err(netcp->ndev_dev, "Close failed\n");
+               }
+       }
+       mutex_unlock(&netcp_modules_lock);
+
+       /* Recycle Rx descriptors from completion queue */
+       netcp_empty_rx_queue(netcp);
+
+       /* Recycle Tx descriptors from completion queue */
+       netcp_process_tx_compl_packets(netcp, netcp->tx_pool_size);
+
+       if (knav_pool_count(netcp->tx_pool) != netcp->tx_pool_size)
+               dev_err(netcp->ndev_dev, "Lost (%d) Tx descs\n",
+                       netcp->tx_pool_size - knav_pool_count(netcp->tx_pool));
+
+       netcp_free_navigator_resources(netcp);
+       dev_dbg(netcp->ndev_dev, "netcp device %s stopped\n", ndev->name);
+       return 0;
+}
+
+static int netcp_ndo_ioctl(struct net_device *ndev,
+                          struct ifreq *req, int cmd)
+{
+       struct netcp_intf *netcp = netdev_priv(ndev);
+       struct netcp_intf_modpriv *intf_modpriv;
+       struct netcp_module *module;
+       int ret = -1, err = -EOPNOTSUPP;
+
+       if (!netif_running(ndev))
+               return -EINVAL;
+
+       mutex_lock(&netcp_modules_lock);
+       for_each_module(netcp, intf_modpriv) {
+               module = intf_modpriv->netcp_module;
+               if (!module->ioctl)
+                       continue;
+
+               err = module->ioctl(intf_modpriv->module_priv, req, cmd);
+               if ((err < 0) && (err != -EOPNOTSUPP)) {
+                       ret = err;
+                       goto out;
+               }
+               if (err == 0)
+                       ret = err;
+       }
+
+out:
+       mutex_unlock(&netcp_modules_lock);
+       return (ret == 0) ? 0 : err;
+}
+
+static int netcp_ndo_change_mtu(struct net_device *ndev, int new_mtu)
+{
+       struct netcp_intf *netcp = netdev_priv(ndev);
+
+       /* MTU < 68 is an error for IPv4 traffic */
+       if ((new_mtu < 68) ||
+           (new_mtu > (NETCP_MAX_FRAME_SIZE - ETH_HLEN - ETH_FCS_LEN))) {
+               dev_err(netcp->ndev_dev, "Invalid mtu size = %d\n", new_mtu);
+               return -EINVAL;
+       }
+
+       ndev->mtu = new_mtu;
+       return 0;
+}
+
+static void netcp_ndo_tx_timeout(struct net_device *ndev)
+{
+       struct netcp_intf *netcp = netdev_priv(ndev);
+       unsigned int descs = knav_pool_count(netcp->tx_pool);
+
+       dev_err(netcp->ndev_dev, "transmit timed out tx descs(%d)\n", descs);
+       netcp_process_tx_compl_packets(netcp, netcp->tx_pool_size);
+       ndev->trans_start = jiffies;
+       netif_tx_wake_all_queues(ndev);
+}
+
+static int netcp_rx_add_vid(struct net_device *ndev, __be16 proto, u16 vid)
+{
+       struct netcp_intf *netcp = netdev_priv(ndev);
+       struct netcp_intf_modpriv *intf_modpriv;
+       struct netcp_module *module;
+       int err = 0;
+
+       dev_dbg(netcp->ndev_dev, "adding rx vlan id: %d\n", vid);
+
+       mutex_lock(&netcp_modules_lock);
+       for_each_module(netcp, intf_modpriv) {
+               module = intf_modpriv->netcp_module;
+               if ((module->add_vid) && (vid != 0)) {
+                       err = module->add_vid(intf_modpriv->module_priv, vid);
+                       if (err != 0) {
+                               dev_err(netcp->ndev_dev, "Could not add vlan id = %d\n",
+                                       vid);
+                               break;
+                       }
+               }
+       }
+       mutex_unlock(&netcp_modules_lock);
+       return err;
+}
+
+static int netcp_rx_kill_vid(struct net_device *ndev, __be16 proto, u16 vid)
+{
+       struct netcp_intf *netcp = netdev_priv(ndev);
+       struct netcp_intf_modpriv *intf_modpriv;
+       struct netcp_module *module;
+       int err = 0;
+
+       dev_dbg(netcp->ndev_dev, "removing rx vlan id: %d\n", vid);
+
+       mutex_lock(&netcp_modules_lock);
+       for_each_module(netcp, intf_modpriv) {
+               module = intf_modpriv->netcp_module;
+               if (module->del_vid) {
+                       err = module->del_vid(intf_modpriv->module_priv, vid);
+                       if (err != 0) {
+                               dev_err(netcp->ndev_dev, "Could not delete vlan id = %d\n",
+                                       vid);
+                               break;
+                       }
+               }
+       }
+       mutex_unlock(&netcp_modules_lock);
+       return err;
+}
+
+static u16 netcp_select_queue(struct net_device *dev, struct sk_buff *skb,
+                             void *accel_priv,
+                             select_queue_fallback_t fallback)
+{
+       return 0;
+}
+
+static int netcp_setup_tc(struct net_device *dev, u8 num_tc)
+{
+       int i;
+
+       /* setup tc must be called under rtnl lock */
+       ASSERT_RTNL();
+
+       /* Sanity-check the number of traffic classes requested */
+       if ((dev->real_num_tx_queues <= 1) ||
+           (dev->real_num_tx_queues < num_tc))
+               return -EINVAL;
+
+       /* Configure traffic class to queue mappings */
+       if (num_tc) {
+               netdev_set_num_tc(dev, num_tc);
+               for (i = 0; i < num_tc; i++)
+                       netdev_set_tc_queue(dev, i, 1, i);
+       } else {
+               netdev_reset_tc(dev);
+       }
+
+       return 0;
+}
+
+static const struct net_device_ops netcp_netdev_ops = {
+       .ndo_open               = netcp_ndo_open,
+       .ndo_stop               = netcp_ndo_stop,
+       .ndo_start_xmit         = netcp_ndo_start_xmit,
+       .ndo_set_rx_mode        = netcp_set_rx_mode,
+       .ndo_do_ioctl           = netcp_ndo_ioctl,
+       .ndo_change_mtu         = netcp_ndo_change_mtu,
+       .ndo_set_mac_address    = eth_mac_addr,
+       .ndo_validate_addr      = eth_validate_addr,
+       .ndo_vlan_rx_add_vid    = netcp_rx_add_vid,
+       .ndo_vlan_rx_kill_vid   = netcp_rx_kill_vid,
+       .ndo_tx_timeout         = netcp_ndo_tx_timeout,
+       .ndo_select_queue       = netcp_select_queue,
+       .ndo_setup_tc           = netcp_setup_tc,
+};
+
+static int netcp_create_interface(struct netcp_device *netcp_device,
+                                 struct device_node *node_interface)
+{
+       struct device *dev = netcp_device->device;
+       struct device_node *node = dev->of_node;
+       struct netcp_intf *netcp;
+       struct net_device *ndev;
+       resource_size_t size;
+       struct resource res;
+       void __iomem *efuse = NULL;
+       u32 efuse_mac = 0;
+       const void *mac_addr;
+       u8 efuse_mac_addr[6];
+       u32 temp[2];
+       int ret = 0;
+
+       ndev = alloc_etherdev_mqs(sizeof(*netcp), 1, 1);
+       if (!ndev) {
+               dev_err(dev, "Error allocating netdev\n");
+               return -ENOMEM;
+       }
+
+       ndev->features |= NETIF_F_SG;
+       ndev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
+       ndev->hw_features = ndev->features;
+       ndev->vlan_features |=  NETIF_F_SG;
+
+       netcp = netdev_priv(ndev);
+       spin_lock_init(&netcp->lock);
+       INIT_LIST_HEAD(&netcp->module_head);
+       INIT_LIST_HEAD(&netcp->txhook_list_head);
+       INIT_LIST_HEAD(&netcp->rxhook_list_head);
+       INIT_LIST_HEAD(&netcp->addr_list);
+       netcp->netcp_device = netcp_device;
+       netcp->dev = netcp_device->device;
+       netcp->ndev = ndev;
+       netcp->ndev_dev  = &ndev->dev;
+       netcp->msg_enable = netif_msg_init(netcp_debug_level, NETCP_DEBUG);
+       netcp->tx_pause_threshold = MAX_SKB_FRAGS;
+       netcp->tx_resume_threshold = netcp->tx_pause_threshold;
+       netcp->node_interface = node_interface;
+
+       ret = of_property_read_u32(node_interface, "efuse-mac", &efuse_mac);
+       if (efuse_mac) {
+               if (of_address_to_resource(node, NETCP_EFUSE_REG_INDEX, &res)) {
+                       dev_err(dev, "could not find efuse-mac reg resource\n");
+                       ret = -ENODEV;
+                       goto quit;
+               }
+               size = resource_size(&res);
+
+               if (!devm_request_mem_region(dev, res.start, size,
+                                            dev_name(dev))) {
+                       dev_err(dev, "could not reserve resource\n");
+                       ret = -ENOMEM;
+                       goto quit;
+               }
+
+               efuse = devm_ioremap_nocache(dev, res.start, size);
+               if (!efuse) {
+                       dev_err(dev, "could not map resource\n");
+                       devm_release_mem_region(dev, res.start, size);
+                       ret = -ENOMEM;
+                       goto quit;
+               }
+
+               emac_arch_get_mac_addr(efuse_mac_addr, efuse);
+               if (is_valid_ether_addr(efuse_mac_addr))
+                       ether_addr_copy(ndev->dev_addr, efuse_mac_addr);
+               else
+                       random_ether_addr(ndev->dev_addr);
+
+               devm_iounmap(dev, efuse);
+               devm_release_mem_region(dev, res.start, size);
+       } else {
+               mac_addr = of_get_mac_address(node_interface);
+               if (mac_addr)
+                       ether_addr_copy(ndev->dev_addr, mac_addr);
+               else
+                       random_ether_addr(ndev->dev_addr);
+       }
+
+       ret = of_property_read_string(node_interface, "rx-channel",
+                                     &netcp->dma_chan_name);
+       if (ret < 0) {
+               dev_err(dev, "missing \"rx-channel\" parameter\n");
+               ret = -ENODEV;
+               goto quit;
+       }
+
+       ret = of_property_read_u32(node_interface, "rx-queue",
+                                  &netcp->rx_queue_id);
+       if (ret < 0) {
+               dev_warn(dev, "missing \"rx-queue\" parameter\n");
+               netcp->rx_queue_id = KNAV_QUEUE_QPEND;
+       }
+
+       ret = of_property_read_u32_array(node_interface, "rx-queue-depth",
+                                        netcp->rx_queue_depths,
+                                        KNAV_DMA_FDQ_PER_CHAN);
+       if (ret < 0) {
+               dev_err(dev, "missing \"rx-queue-depth\" parameter\n");
+               netcp->rx_queue_depths[0] = 128;
+       }
+
+       ret = of_property_read_u32_array(node_interface, "rx-buffer-size",
+                                        netcp->rx_buffer_sizes,
+                                        KNAV_DMA_FDQ_PER_CHAN);
+       if (ret) {
+               dev_err(dev, "missing \"rx-buffer-size\" parameter\n");
+               netcp->rx_buffer_sizes[0] = 1536;
+       }
+
+       ret = of_property_read_u32_array(node_interface, "rx-pool", temp, 2);
+       if (ret < 0) {
+               dev_err(dev, "missing \"rx-pool\" parameter\n");
+               ret = -ENODEV;
+               goto quit;
+       }
+       netcp->rx_pool_size = temp[0];
+       netcp->rx_pool_region_id = temp[1];
+
+       ret = of_property_read_u32_array(node_interface, "tx-pool", temp, 2);
+       if (ret < 0) {
+               dev_err(dev, "missing \"tx-pool\" parameter\n");
+               ret = -ENODEV;
+               goto quit;
+       }
+       netcp->tx_pool_size = temp[0];
+       netcp->tx_pool_region_id = temp[1];
+
+       if (netcp->tx_pool_size < MAX_SKB_FRAGS) {
+               dev_err(dev, "tx-pool size too small, must be atleast(%ld)\n",
+                       MAX_SKB_FRAGS);
+               ret = -ENODEV;
+               goto quit;
+       }
+
+       ret = of_property_read_u32(node_interface, "tx-completion-queue",
+                                  &netcp->tx_compl_qid);
+       if (ret < 0) {
+               dev_warn(dev, "missing \"tx-completion-queue\" parameter\n");
+               netcp->tx_compl_qid = KNAV_QUEUE_QPEND;
+       }
+
+       /* NAPI register */
+       netif_napi_add(ndev, &netcp->rx_napi, netcp_rx_poll, NETCP_NAPI_WEIGHT);
+       netif_napi_add(ndev, &netcp->tx_napi, netcp_tx_poll, NETCP_NAPI_WEIGHT);
+
+       /* Register the network device */
+       ndev->dev_id            = 0;
+       ndev->watchdog_timeo    = NETCP_TX_TIMEOUT;
+       ndev->netdev_ops        = &netcp_netdev_ops;
+       SET_NETDEV_DEV(ndev, dev);
+
+       list_add_tail(&netcp->interface_list, &netcp_device->interface_head);
+       return 0;
+
+quit:
+       free_netdev(ndev);
+       return ret;
+}
+
+static void netcp_delete_interface(struct netcp_device *netcp_device,
+                                  struct net_device *ndev)
+{
+       struct netcp_intf_modpriv *intf_modpriv, *tmp;
+       struct netcp_intf *netcp = netdev_priv(ndev);
+       struct netcp_module *module;
+
+       dev_dbg(netcp_device->device, "Removing interface \"%s\"\n",
+               ndev->name);
+
+       /* Notify each of the modules that the interface is going away */
+       list_for_each_entry_safe(intf_modpriv, tmp, &netcp->module_head,
+                                intf_list) {
+               module = intf_modpriv->netcp_module;
+               dev_dbg(netcp_device->device, "Releasing module \"%s\"\n",
+                       module->name);
+               if (module->release)
+                       module->release(intf_modpriv->module_priv);
+               list_del(&intf_modpriv->intf_list);
+               kfree(intf_modpriv);
+       }
+       WARN(!list_empty(&netcp->module_head), "%s interface module list is not empty!\n",
+            ndev->name);
+
+       list_del(&netcp->interface_list);
+
+       of_node_put(netcp->node_interface);
+       unregister_netdev(ndev);
+       netif_napi_del(&netcp->rx_napi);
+       free_netdev(ndev);
+}
+
+static int netcp_probe(struct platform_device *pdev)
+{
+       struct device_node *node = pdev->dev.of_node;
+       struct netcp_intf *netcp_intf, *netcp_tmp;
+       struct device_node *child, *interfaces;
+       struct netcp_device *netcp_device;
+       struct device *dev = &pdev->dev;
+       struct netcp_module *module;
+       int ret;
+
+       if (!node) {
+               dev_err(dev, "could not find device info\n");
+               return -ENODEV;
+       }
+
+       /* Allocate a new NETCP device instance */
+       netcp_device = devm_kzalloc(dev, sizeof(*netcp_device), GFP_KERNEL);
+       if (!netcp_device)
+               return -ENOMEM;
+
+       pm_runtime_enable(&pdev->dev);
+       ret = pm_runtime_get_sync(&pdev->dev);
+       if (ret < 0) {
+               dev_err(dev, "Failed to enable NETCP power-domain\n");
+               pm_runtime_disable(&pdev->dev);
+               return ret;
+       }
+
+       /* Initialize the NETCP device instance */
+       INIT_LIST_HEAD(&netcp_device->interface_head);
+       INIT_LIST_HEAD(&netcp_device->modpriv_head);
+       netcp_device->device = dev;
+       platform_set_drvdata(pdev, netcp_device);
+
+       /* create interfaces */
+       interfaces = of_get_child_by_name(node, "netcp-interfaces");
+       if (!interfaces) {
+               dev_err(dev, "could not find netcp-interfaces node\n");
+               ret = -ENODEV;
+               goto probe_quit;
+       }
+
+       for_each_available_child_of_node(interfaces, child) {
+               ret = netcp_create_interface(netcp_device, child);
+               if (ret) {
+                       dev_err(dev, "could not create interface(%s)\n",
+                               child->name);
+                       goto probe_quit_interface;
+               }
+       }
+
+       /* Add the device instance to the list */
+       list_add_tail(&netcp_device->device_list, &netcp_devices);
+
+       /* Probe & attach any modules already registered */
+       mutex_lock(&netcp_modules_lock);
+       for_each_netcp_module(module) {
+               ret = netcp_module_probe(netcp_device, module);
+               if (ret < 0)
+                       dev_err(dev, "module(%s) probe failed\n", module->name);
+       }
+       mutex_unlock(&netcp_modules_lock);
+       return 0;
+
+probe_quit_interface:
+       list_for_each_entry_safe(netcp_intf, netcp_tmp,
+                                &netcp_device->interface_head,
+                                interface_list) {
+               netcp_delete_interface(netcp_device, netcp_intf->ndev);
+       }
+
+probe_quit:
+       pm_runtime_put_sync(&pdev->dev);
+       pm_runtime_disable(&pdev->dev);
+       platform_set_drvdata(pdev, NULL);
+       return ret;
+}
+
+static int netcp_remove(struct platform_device *pdev)
+{
+       struct netcp_device *netcp_device = platform_get_drvdata(pdev);
+       struct netcp_inst_modpriv *inst_modpriv, *tmp;
+       struct netcp_module *module;
+
+       list_for_each_entry_safe(inst_modpriv, tmp, &netcp_device->modpriv_head,
+                                inst_list) {
+               module = inst_modpriv->netcp_module;
+               dev_dbg(&pdev->dev, "Removing module \"%s\"\n", module->name);
+               module->remove(netcp_device, inst_modpriv->module_priv);
+               list_del(&inst_modpriv->inst_list);
+               kfree(inst_modpriv);
+       }
+       WARN(!list_empty(&netcp_device->interface_head), "%s interface list not empty!\n",
+            pdev->name);
+
+       devm_kfree(&pdev->dev, netcp_device);
+       pm_runtime_put_sync(&pdev->dev);
+       pm_runtime_disable(&pdev->dev);
+       platform_set_drvdata(pdev, NULL);
+       return 0;
+}
+
+static struct of_device_id of_match[] = {
+       { .compatible = "ti,netcp-1.0", },
+       {},
+};
+MODULE_DEVICE_TABLE(of, of_match);
+
+static struct platform_driver netcp_driver = {
+       .driver = {
+               .name           = "netcp-1.0",
+               .owner          = THIS_MODULE,
+               .of_match_table = of_match,
+       },
+       .probe = netcp_probe,
+       .remove = netcp_remove,
+};
+module_platform_driver(netcp_driver);
+
+MODULE_LICENSE("GPL v2");
+MODULE_DESCRIPTION("TI NETCP driver for Keystone SOCs");
+MODULE_AUTHOR("Sandeep Nair <sandeep_n@ti.com");