Revert "staging: octeon: delete driver"
authorChris Packham <chris.packham@alliedtelesis.co.nz>
Wed, 5 Feb 2020 00:11:12 +0000 (13:11 +1300)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 12 Feb 2020 21:50:32 +0000 (13:50 -0800)
This reverts commit 710d7fbe21ee2ceab121f1f84a20edf68f9f9742.
Re-instate the code so subsequent commits can clean it up and get it
building properly.

Signed-off-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
Link: https://lore.kernel.org/r/20200205001116.14096-3-chris.packham@alliedtelesis.co.nz
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
21 files changed:
drivers/staging/Kconfig
drivers/staging/Makefile
drivers/staging/octeon/Kconfig [new file with mode: 0644]
drivers/staging/octeon/Makefile [new file with mode: 0644]
drivers/staging/octeon/TODO [new file with mode: 0644]
drivers/staging/octeon/ethernet-defines.h [new file with mode: 0644]
drivers/staging/octeon/ethernet-mdio.c [new file with mode: 0644]
drivers/staging/octeon/ethernet-mdio.h [new file with mode: 0644]
drivers/staging/octeon/ethernet-mem.c [new file with mode: 0644]
drivers/staging/octeon/ethernet-mem.h [new file with mode: 0644]
drivers/staging/octeon/ethernet-rgmii.c [new file with mode: 0644]
drivers/staging/octeon/ethernet-rx.c [new file with mode: 0644]
drivers/staging/octeon/ethernet-rx.h [new file with mode: 0644]
drivers/staging/octeon/ethernet-sgmii.c [new file with mode: 0644]
drivers/staging/octeon/ethernet-spi.c [new file with mode: 0644]
drivers/staging/octeon/ethernet-tx.c [new file with mode: 0644]
drivers/staging/octeon/ethernet-tx.h [new file with mode: 0644]
drivers/staging/octeon/ethernet-util.h [new file with mode: 0644]
drivers/staging/octeon/ethernet.c [new file with mode: 0644]
drivers/staging/octeon/octeon-ethernet.h [new file with mode: 0644]
drivers/staging/octeon/octeon-stubs.h [new file with mode: 0644]

index 9ec982d16e1cf90e225cf1a15dac4506cc8a754e..f46c1f7d0d39f4e629918f8db625c977fa5be1c7 100644 (file)
@@ -42,6 +42,8 @@ source "drivers/staging/rtl8188eu/Kconfig"
 
 source "drivers/staging/rts5208/Kconfig"
 
+source "drivers/staging/octeon/Kconfig"
+
 source "drivers/staging/octeon-usb/Kconfig"
 
 source "drivers/staging/vt6655/Kconfig"
index be3f0a814bf2ceb0e3365414c7f36e2d4d9eb0ad..bb26d753066c6d0ddd7e3df97fad484767699574 100644 (file)
@@ -12,6 +12,7 @@ obj-$(CONFIG_R8712U)          += rtl8712/
 obj-$(CONFIG_R8188EU)          += rtl8188eu/
 obj-$(CONFIG_RTS5208)          += rts5208/
 obj-$(CONFIG_NETLOGIC_XLR_NET) += netlogic/
+obj-$(CONFIG_OCTEON_ETHERNET)  += octeon/
 obj-$(CONFIG_OCTEON_USB)       += octeon-usb/
 obj-$(CONFIG_VT6655)           += vt6655/
 obj-$(CONFIG_VT6656)           += vt6656/
diff --git a/drivers/staging/octeon/Kconfig b/drivers/staging/octeon/Kconfig
new file mode 100644 (file)
index 0000000..e7f4ddc
--- /dev/null
@@ -0,0 +1,16 @@
+# SPDX-License-Identifier: GPL-2.0
+config OCTEON_ETHERNET
+       tristate "Cavium Networks Octeon Ethernet support"
+       depends on CAVIUM_OCTEON_SOC || COMPILE_TEST
+       depends on NETDEVICES
+       depends on BROKEN
+       select PHYLIB
+       select MDIO_OCTEON
+       help
+         This driver supports the builtin ethernet ports on Cavium
+         Networks' products in the Octeon family. This driver supports the
+         CN3XXX and CN5XXX Octeon processors.
+
+         To compile this driver as a module, choose M here.  The module
+         will be called octeon-ethernet.
+
diff --git a/drivers/staging/octeon/Makefile b/drivers/staging/octeon/Makefile
new file mode 100644 (file)
index 0000000..3887cf5
--- /dev/null
@@ -0,0 +1,19 @@
+# SPDX-License-Identifier: GPL-2.0
+#
+# Copyright (C) 2005-2009 Cavium Networks
+#
+
+#
+# Makefile for Cavium OCTEON on-board ethernet driver
+#
+
+obj-${CONFIG_OCTEON_ETHERNET} :=  octeon-ethernet.o
+
+octeon-ethernet-y := ethernet.o
+octeon-ethernet-y += ethernet-mdio.o
+octeon-ethernet-y += ethernet-mem.o
+octeon-ethernet-y += ethernet-rgmii.o
+octeon-ethernet-y += ethernet-rx.o
+octeon-ethernet-y += ethernet-sgmii.o
+octeon-ethernet-y += ethernet-spi.o
+octeon-ethernet-y += ethernet-tx.o
diff --git a/drivers/staging/octeon/TODO b/drivers/staging/octeon/TODO
new file mode 100644 (file)
index 0000000..67a0a1f
--- /dev/null
@@ -0,0 +1,9 @@
+This driver is functional and supports Ethernet on OCTEON+/OCTEON2/OCTEON3
+chips at least up to CN7030.
+
+TODO:
+       - general code review and clean up
+       - make driver self-contained instead of being split between staging and
+         arch/mips/cavium-octeon.
+
+Contact: Aaro Koskinen <aaro.koskinen@iki.fi>
diff --git a/drivers/staging/octeon/ethernet-defines.h b/drivers/staging/octeon/ethernet-defines.h
new file mode 100644 (file)
index 0000000..ef9e767
--- /dev/null
@@ -0,0 +1,40 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+/*
+ * A few defines are used to control the operation of this driver:
+ *  USE_ASYNC_IOBDMA
+ *      Use asynchronous IO access to hardware. This uses Octeon's asynchronous
+ *      IOBDMAs to issue IO accesses without stalling. Set this to zero
+ *      to disable this. Note that IOBDMAs require CVMSEG.
+ *  REUSE_SKBUFFS_WITHOUT_FREE
+ *      Allows the TX path to free an skbuff into the FPA hardware pool. This
+ *      can significantly improve performance for forwarding and bridging, but
+ *      may be somewhat dangerous. Checks are made, but if any buffer is reused
+ *      without the proper Linux cleanup, the networking stack may have very
+ *      bizarre bugs.
+ */
+#ifndef __ETHERNET_DEFINES_H__
+#define __ETHERNET_DEFINES_H__
+
+#ifdef CONFIG_NETFILTER
+#define REUSE_SKBUFFS_WITHOUT_FREE  0
+#else
+#define REUSE_SKBUFFS_WITHOUT_FREE  1
+#endif
+
+#define USE_ASYNC_IOBDMA            (CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE > 0)
+
+/* Maximum number of SKBs to try to free per xmit packet. */
+#define MAX_OUT_QUEUE_DEPTH 1000
+
+#define FAU_TOTAL_TX_TO_CLEAN (CVMX_FAU_REG_END - sizeof(u32))
+#define FAU_NUM_PACKET_BUFFERS_TO_FREE (FAU_TOTAL_TX_TO_CLEAN - sizeof(u32))
+
+#define TOTAL_NUMBER_OF_PORTS       (CVMX_PIP_NUM_INPUT_PORTS + 1)
+
+#endif /* __ETHERNET_DEFINES_H__ */
diff --git a/drivers/staging/octeon/ethernet-mdio.c b/drivers/staging/octeon/ethernet-mdio.c
new file mode 100644 (file)
index 0000000..c798672
--- /dev/null
@@ -0,0 +1,178 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+#include <linux/kernel.h>
+#include <linux/ethtool.h>
+#include <linux/phy.h>
+#include <linux/ratelimit.h>
+#include <linux/of_mdio.h>
+#include <generated/utsrelease.h>
+#include <net/dst.h>
+
+#include "octeon-ethernet.h"
+#include "ethernet-defines.h"
+#include "ethernet-mdio.h"
+#include "ethernet-util.h"
+
+static void cvm_oct_get_drvinfo(struct net_device *dev,
+                               struct ethtool_drvinfo *info)
+{
+       strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
+       strlcpy(info->version, UTS_RELEASE, sizeof(info->version));
+       strlcpy(info->bus_info, "Builtin", sizeof(info->bus_info));
+}
+
+static int cvm_oct_nway_reset(struct net_device *dev)
+{
+       if (!capable(CAP_NET_ADMIN))
+               return -EPERM;
+
+       if (dev->phydev)
+               return phy_start_aneg(dev->phydev);
+
+       return -EINVAL;
+}
+
+const struct ethtool_ops cvm_oct_ethtool_ops = {
+       .get_drvinfo = cvm_oct_get_drvinfo,
+       .nway_reset = cvm_oct_nway_reset,
+       .get_link = ethtool_op_get_link,
+       .get_link_ksettings = phy_ethtool_get_link_ksettings,
+       .set_link_ksettings = phy_ethtool_set_link_ksettings,
+};
+
+/**
+ * cvm_oct_ioctl - IOCTL support for PHY control
+ * @dev:    Device to change
+ * @rq:     the request
+ * @cmd:    the command
+ *
+ * Returns Zero on success
+ */
+int cvm_oct_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
+{
+       if (!netif_running(dev))
+               return -EINVAL;
+
+       if (!dev->phydev)
+               return -EINVAL;
+
+       return phy_mii_ioctl(dev->phydev, rq, cmd);
+}
+
+void cvm_oct_note_carrier(struct octeon_ethernet *priv,
+                         union cvmx_helper_link_info li)
+{
+       if (li.s.link_up) {
+               pr_notice_ratelimited("%s: %u Mbps %s duplex, port %d, queue %d\n",
+                                     netdev_name(priv->netdev), li.s.speed,
+                                     (li.s.full_duplex) ? "Full" : "Half",
+                                     priv->port, priv->queue);
+       } else {
+               pr_notice_ratelimited("%s: Link down\n",
+                                     netdev_name(priv->netdev));
+       }
+}
+
+void cvm_oct_adjust_link(struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       union cvmx_helper_link_info link_info;
+
+       link_info.u64           = 0;
+       link_info.s.link_up     = dev->phydev->link ? 1 : 0;
+       link_info.s.full_duplex = dev->phydev->duplex ? 1 : 0;
+       link_info.s.speed       = dev->phydev->speed;
+       priv->link_info         = link_info.u64;
+
+       /*
+        * The polling task need to know about link status changes.
+        */
+       if (priv->poll)
+               priv->poll(dev);
+
+       if (priv->last_link != dev->phydev->link) {
+               priv->last_link = dev->phydev->link;
+               cvmx_helper_link_set(priv->port, link_info);
+               cvm_oct_note_carrier(priv, link_info);
+       }
+}
+
+int cvm_oct_common_stop(struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       int interface = INTERFACE(priv->port);
+       union cvmx_helper_link_info link_info;
+       union cvmx_gmxx_prtx_cfg gmx_cfg;
+       int index = INDEX(priv->port);
+
+       gmx_cfg.u64 = cvmx_read_csr(CVMX_GMXX_PRTX_CFG(index, interface));
+       gmx_cfg.s.en = 0;
+       cvmx_write_csr(CVMX_GMXX_PRTX_CFG(index, interface), gmx_cfg.u64);
+
+       priv->poll = NULL;
+
+       if (dev->phydev)
+               phy_disconnect(dev->phydev);
+
+       if (priv->last_link) {
+               link_info.u64 = 0;
+               priv->last_link = 0;
+
+               cvmx_helper_link_set(priv->port, link_info);
+               cvm_oct_note_carrier(priv, link_info);
+       }
+       return 0;
+}
+
+/**
+ * cvm_oct_phy_setup_device - setup the PHY
+ *
+ * @dev:    Device to setup
+ *
+ * Returns Zero on success, negative on failure
+ */
+int cvm_oct_phy_setup_device(struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       struct device_node *phy_node;
+       struct phy_device *phydev = NULL;
+
+       if (!priv->of_node)
+               goto no_phy;
+
+       phy_node = of_parse_phandle(priv->of_node, "phy-handle", 0);
+       if (!phy_node && of_phy_is_fixed_link(priv->of_node)) {
+               int rc;
+
+               rc = of_phy_register_fixed_link(priv->of_node);
+               if (rc)
+                       return rc;
+
+               phy_node = of_node_get(priv->of_node);
+       }
+       if (!phy_node)
+               goto no_phy;
+
+       phydev = of_phy_connect(dev, phy_node, cvm_oct_adjust_link, 0,
+                               priv->phy_mode);
+       of_node_put(phy_node);
+
+       if (!phydev)
+               return -ENODEV;
+
+       priv->last_link = 0;
+       phy_start(phydev);
+
+       return 0;
+no_phy:
+       /* If there is no phy, assume a direct MAC connection and that
+        * the link is up.
+        */
+       netif_carrier_on(dev);
+       return 0;
+}
diff --git a/drivers/staging/octeon/ethernet-mdio.h b/drivers/staging/octeon/ethernet-mdio.h
new file mode 100644 (file)
index 0000000..e3771d4
--- /dev/null
@@ -0,0 +1,28 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/ip.h>
+#include <linux/string.h>
+#include <linux/ethtool.h>
+#include <linux/seq_file.h>
+#include <linux/proc_fs.h>
+#include <net/dst.h>
+#ifdef CONFIG_XFRM
+#include <linux/xfrm.h>
+#include <net/xfrm.h>
+#endif /* CONFIG_XFRM */
+
+extern const struct ethtool_ops cvm_oct_ethtool_ops;
+
+void octeon_mdiobus_force_mod_depencency(void);
+
+int cvm_oct_ioctl(struct net_device *dev, struct ifreq *rq, int cmd);
+int cvm_oct_phy_setup_device(struct net_device *dev);
diff --git a/drivers/staging/octeon/ethernet-mem.c b/drivers/staging/octeon/ethernet-mem.c
new file mode 100644 (file)
index 0000000..5325949
--- /dev/null
@@ -0,0 +1,154 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2010 Cavium Networks
+ */
+
+#include <linux/kernel.h>
+#include <linux/netdevice.h>
+#include <linux/slab.h>
+
+#include "octeon-ethernet.h"
+#include "ethernet-mem.h"
+#include "ethernet-defines.h"
+
+/**
+ * cvm_oct_fill_hw_skbuff - fill the supplied hardware pool with skbuffs
+ * @pool:     Pool to allocate an skbuff for
+ * @size:     Size of the buffer needed for the pool
+ * @elements: Number of buffers to allocate
+ *
+ * Returns the actual number of buffers allocated.
+ */
+static int cvm_oct_fill_hw_skbuff(int pool, int size, int elements)
+{
+       int freed = elements;
+
+       while (freed) {
+               struct sk_buff *skb = dev_alloc_skb(size + 256);
+
+               if (unlikely(!skb))
+                       break;
+               skb_reserve(skb, 256 - (((unsigned long)skb->data) & 0x7f));
+               *(struct sk_buff **)(skb->data - sizeof(void *)) = skb;
+               cvmx_fpa_free(skb->data, pool, size / 128);
+               freed--;
+       }
+       return elements - freed;
+}
+
+/**
+ * cvm_oct_free_hw_skbuff- free hardware pool skbuffs
+ * @pool:     Pool to allocate an skbuff for
+ * @size:     Size of the buffer needed for the pool
+ * @elements: Number of buffers to allocate
+ */
+static void cvm_oct_free_hw_skbuff(int pool, int size, int elements)
+{
+       char *memory;
+
+       do {
+               memory = cvmx_fpa_alloc(pool);
+               if (memory) {
+                       struct sk_buff *skb =
+                           *(struct sk_buff **)(memory - sizeof(void *));
+                       elements--;
+                       dev_kfree_skb(skb);
+               }
+       } while (memory);
+
+       if (elements < 0)
+               pr_warn("Freeing of pool %u had too many skbuffs (%d)\n",
+                       pool, elements);
+       else if (elements > 0)
+               pr_warn("Freeing of pool %u is missing %d skbuffs\n",
+                       pool, elements);
+}
+
+/**
+ * cvm_oct_fill_hw_memory - fill a hardware pool with memory.
+ * @pool:     Pool to populate
+ * @size:     Size of each buffer in the pool
+ * @elements: Number of buffers to allocate
+ *
+ * Returns the actual number of buffers allocated.
+ */
+static int cvm_oct_fill_hw_memory(int pool, int size, int elements)
+{
+       char *memory;
+       char *fpa;
+       int freed = elements;
+
+       while (freed) {
+               /*
+                * FPA memory must be 128 byte aligned.  Since we are
+                * aligning we need to save the original pointer so we
+                * can feed it to kfree when the memory is returned to
+                * the kernel.
+                *
+                * We allocate an extra 256 bytes to allow for
+                * alignment and space for the original pointer saved
+                * just before the block.
+                */
+               memory = kmalloc(size + 256, GFP_ATOMIC);
+               if (unlikely(!memory)) {
+                       pr_warn("Unable to allocate %u bytes for FPA pool %d\n",
+                               elements * size, pool);
+                       break;
+               }
+               fpa = (char *)(((unsigned long)memory + 256) & ~0x7fUL);
+               *((char **)fpa - 1) = memory;
+               cvmx_fpa_free(fpa, pool, 0);
+               freed--;
+       }
+       return elements - freed;
+}
+
+/**
+ * cvm_oct_free_hw_memory - Free memory allocated by cvm_oct_fill_hw_memory
+ * @pool:     FPA pool to free
+ * @size:     Size of each buffer in the pool
+ * @elements: Number of buffers that should be in the pool
+ */
+static void cvm_oct_free_hw_memory(int pool, int size, int elements)
+{
+       char *memory;
+       char *fpa;
+
+       do {
+               fpa = cvmx_fpa_alloc(pool);
+               if (fpa) {
+                       elements--;
+                       fpa = (char *)phys_to_virt(cvmx_ptr_to_phys(fpa));
+                       memory = *((char **)fpa - 1);
+                       kfree(memory);
+               }
+       } while (fpa);
+
+       if (elements < 0)
+               pr_warn("Freeing of pool %u had too many buffers (%d)\n",
+                       pool, elements);
+       else if (elements > 0)
+               pr_warn("Warning: Freeing of pool %u is missing %d buffers\n",
+                       pool, elements);
+}
+
+int cvm_oct_mem_fill_fpa(int pool, int size, int elements)
+{
+       int freed;
+
+       if (pool == CVMX_FPA_PACKET_POOL)
+               freed = cvm_oct_fill_hw_skbuff(pool, size, elements);
+       else
+               freed = cvm_oct_fill_hw_memory(pool, size, elements);
+       return freed;
+}
+
+void cvm_oct_mem_empty_fpa(int pool, int size, int elements)
+{
+       if (pool == CVMX_FPA_PACKET_POOL)
+               cvm_oct_free_hw_skbuff(pool, size, elements);
+       else
+               cvm_oct_free_hw_memory(pool, size, elements);
+}
diff --git a/drivers/staging/octeon/ethernet-mem.h b/drivers/staging/octeon/ethernet-mem.h
new file mode 100644 (file)
index 0000000..692dcdb
--- /dev/null
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+int cvm_oct_mem_fill_fpa(int pool, int size, int elements);
+void cvm_oct_mem_empty_fpa(int pool, int size, int elements);
diff --git a/drivers/staging/octeon/ethernet-rgmii.c b/drivers/staging/octeon/ethernet-rgmii.c
new file mode 100644 (file)
index 0000000..0c4fac3
--- /dev/null
@@ -0,0 +1,158 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+#include <linux/kernel.h>
+#include <linux/netdevice.h>
+#include <linux/interrupt.h>
+#include <linux/phy.h>
+#include <linux/ratelimit.h>
+#include <net/dst.h>
+
+#include "octeon-ethernet.h"
+#include "ethernet-defines.h"
+#include "ethernet-util.h"
+#include "ethernet-mdio.h"
+
+static DEFINE_SPINLOCK(global_register_lock);
+
+static void cvm_oct_set_hw_preamble(struct octeon_ethernet *priv, bool enable)
+{
+       union cvmx_gmxx_rxx_frm_ctl gmxx_rxx_frm_ctl;
+       union cvmx_ipd_sub_port_fcs ipd_sub_port_fcs;
+       union cvmx_gmxx_rxx_int_reg gmxx_rxx_int_reg;
+       int interface = INTERFACE(priv->port);
+       int index = INDEX(priv->port);
+
+       /* Set preamble checking. */
+       gmxx_rxx_frm_ctl.u64 = cvmx_read_csr(CVMX_GMXX_RXX_FRM_CTL(index,
+                                                                  interface));
+       gmxx_rxx_frm_ctl.s.pre_chk = enable;
+       cvmx_write_csr(CVMX_GMXX_RXX_FRM_CTL(index, interface),
+                      gmxx_rxx_frm_ctl.u64);
+
+       /* Set FCS stripping. */
+       ipd_sub_port_fcs.u64 = cvmx_read_csr(CVMX_IPD_SUB_PORT_FCS);
+       if (enable)
+               ipd_sub_port_fcs.s.port_bit |= 1ull << priv->port;
+       else
+               ipd_sub_port_fcs.s.port_bit &=
+                                       0xffffffffull ^ (1ull << priv->port);
+       cvmx_write_csr(CVMX_IPD_SUB_PORT_FCS, ipd_sub_port_fcs.u64);
+
+       /* Clear any error bits. */
+       gmxx_rxx_int_reg.u64 = cvmx_read_csr(CVMX_GMXX_RXX_INT_REG(index,
+                                                                  interface));
+       cvmx_write_csr(CVMX_GMXX_RXX_INT_REG(index, interface),
+                      gmxx_rxx_int_reg.u64);
+}
+
+static void cvm_oct_check_preamble_errors(struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       union cvmx_helper_link_info link_info;
+       unsigned long flags;
+
+       link_info.u64 = priv->link_info;
+
+       /*
+        * Take the global register lock since we are going to
+        * touch registers that affect more than one port.
+        */
+       spin_lock_irqsave(&global_register_lock, flags);
+
+       if (link_info.s.speed == 10 && priv->last_speed == 10) {
+               /*
+                * Read the GMXX_RXX_INT_REG[PCTERR] bit and see if we are
+                * getting preamble errors.
+                */
+               int interface = INTERFACE(priv->port);
+               int index = INDEX(priv->port);
+               union cvmx_gmxx_rxx_int_reg gmxx_rxx_int_reg;
+
+               gmxx_rxx_int_reg.u64 = cvmx_read_csr(CVMX_GMXX_RXX_INT_REG
+                                                       (index, interface));
+               if (gmxx_rxx_int_reg.s.pcterr) {
+                       /*
+                        * We are getting preamble errors at 10Mbps. Most
+                        * likely the PHY is giving us packets with misaligned
+                        * preambles. In order to get these packets we need to
+                        * disable preamble checking and do it in software.
+                        */
+                       cvm_oct_set_hw_preamble(priv, false);
+                       printk_ratelimited("%s: Using 10Mbps with software preamble removal\n",
+                                          dev->name);
+               }
+       } else {
+               /*
+                * Since the 10Mbps preamble workaround is allowed we need to
+                * enable preamble checking, FCS stripping, and clear error
+                * bits on every speed change. If errors occur during 10Mbps
+                * operation the above code will change this stuff
+                */
+               if (priv->last_speed != link_info.s.speed)
+                       cvm_oct_set_hw_preamble(priv, true);
+               priv->last_speed = link_info.s.speed;
+       }
+       spin_unlock_irqrestore(&global_register_lock, flags);
+}
+
+static void cvm_oct_rgmii_poll(struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       union cvmx_helper_link_info link_info;
+       bool status_change;
+
+       link_info = cvmx_helper_link_get(priv->port);
+       if (priv->link_info != link_info.u64 &&
+           cvmx_helper_link_set(priv->port, link_info))
+               link_info.u64 = priv->link_info;
+       status_change = priv->link_info != link_info.u64;
+       priv->link_info = link_info.u64;
+
+       cvm_oct_check_preamble_errors(dev);
+
+       if (likely(!status_change))
+               return;
+
+       /* Tell core. */
+       if (link_info.s.link_up) {
+               if (!netif_carrier_ok(dev))
+                       netif_carrier_on(dev);
+       } else if (netif_carrier_ok(dev)) {
+               netif_carrier_off(dev);
+       }
+       cvm_oct_note_carrier(priv, link_info);
+}
+
+int cvm_oct_rgmii_open(struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       int ret;
+
+       ret = cvm_oct_common_open(dev, cvm_oct_rgmii_poll);
+       if (ret)
+               return ret;
+
+       if (dev->phydev) {
+               /*
+                * In phydev mode, we need still periodic polling for the
+                * preamble error checking, and we also need to call this
+                * function on every link state change.
+                *
+                * Only true RGMII ports need to be polled. In GMII mode, port
+                * 0 is really a RGMII port.
+                */
+               if ((priv->imode == CVMX_HELPER_INTERFACE_MODE_GMII &&
+                    priv->port  == 0) ||
+                   (priv->imode == CVMX_HELPER_INTERFACE_MODE_RGMII)) {
+                       priv->poll = cvm_oct_check_preamble_errors;
+                       cvm_oct_check_preamble_errors(dev);
+               }
+       }
+
+       return 0;
+}
diff --git a/drivers/staging/octeon/ethernet-rx.c b/drivers/staging/octeon/ethernet-rx.c
new file mode 100644 (file)
index 0000000..2c16230
--- /dev/null
@@ -0,0 +1,538 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2010 Cavium Networks
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/cache.h>
+#include <linux/cpumask.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/ip.h>
+#include <linux/string.h>
+#include <linux/prefetch.h>
+#include <linux/ratelimit.h>
+#include <linux/smp.h>
+#include <linux/interrupt.h>
+#include <net/dst.h>
+#ifdef CONFIG_XFRM
+#include <linux/xfrm.h>
+#include <net/xfrm.h>
+#endif /* CONFIG_XFRM */
+
+#include "octeon-ethernet.h"
+#include "ethernet-defines.h"
+#include "ethernet-mem.h"
+#include "ethernet-rx.h"
+#include "ethernet-util.h"
+
+static atomic_t oct_rx_ready = ATOMIC_INIT(0);
+
+static struct oct_rx_group {
+       int irq;
+       int group;
+       struct napi_struct napi;
+} oct_rx_group[16];
+
+/**
+ * cvm_oct_do_interrupt - interrupt handler.
+ * @irq: Interrupt number.
+ * @napi_id: Cookie to identify the NAPI instance.
+ *
+ * The interrupt occurs whenever the POW has packets in our group.
+ *
+ */
+static irqreturn_t cvm_oct_do_interrupt(int irq, void *napi_id)
+{
+       /* Disable the IRQ and start napi_poll. */
+       disable_irq_nosync(irq);
+       napi_schedule(napi_id);
+
+       return IRQ_HANDLED;
+}
+
+/**
+ * cvm_oct_check_rcv_error - process receive errors
+ * @work: Work queue entry pointing to the packet.
+ *
+ * Returns Non-zero if the packet can be dropped, zero otherwise.
+ */
+static inline int cvm_oct_check_rcv_error(struct cvmx_wqe *work)
+{
+       int port;
+
+       if (octeon_has_feature(OCTEON_FEATURE_PKND))
+               port = work->word0.pip.cn68xx.pknd;
+       else
+               port = work->word1.cn38xx.ipprt;
+
+       if ((work->word2.snoip.err_code == 10) && (work->word1.len <= 64)) {
+               /*
+                * Ignore length errors on min size packets. Some
+                * equipment incorrectly pads packets to 64+4FCS
+                * instead of 60+4FCS.  Note these packets still get
+                * counted as frame errors.
+                */
+       } else if (work->word2.snoip.err_code == 5 ||
+                  work->word2.snoip.err_code == 7) {
+               /*
+                * We received a packet with either an alignment error
+                * or a FCS error. This may be signalling that we are
+                * running 10Mbps with GMXX_RXX_FRM_CTL[PRE_CHK]
+                * off. If this is the case we need to parse the
+                * packet to determine if we can remove a non spec
+                * preamble and generate a correct packet.
+                */
+               int interface = cvmx_helper_get_interface_num(port);
+               int index = cvmx_helper_get_interface_index_num(port);
+               union cvmx_gmxx_rxx_frm_ctl gmxx_rxx_frm_ctl;
+
+               gmxx_rxx_frm_ctl.u64 =
+                   cvmx_read_csr(CVMX_GMXX_RXX_FRM_CTL(index, interface));
+               if (gmxx_rxx_frm_ctl.s.pre_chk == 0) {
+                       u8 *ptr =
+                           cvmx_phys_to_ptr(work->packet_ptr.s.addr);
+                       int i = 0;
+
+                       while (i < work->word1.len - 1) {
+                               if (*ptr != 0x55)
+                                       break;
+                               ptr++;
+                               i++;
+                       }
+
+                       if (*ptr == 0xd5) {
+                               /* Port received 0xd5 preamble */
+                               work->packet_ptr.s.addr += i + 1;
+                               work->word1.len -= i + 5;
+                       } else if ((*ptr & 0xf) == 0xd) {
+                               /* Port received 0xd preamble */
+                               work->packet_ptr.s.addr += i;
+                               work->word1.len -= i + 4;
+                               for (i = 0; i < work->word1.len; i++) {
+                                       *ptr =
+                                           ((*ptr & 0xf0) >> 4) |
+                                           ((*(ptr + 1) & 0xf) << 4);
+                                       ptr++;
+                               }
+                       } else {
+                               printk_ratelimited("Port %d unknown preamble, packet dropped\n",
+                                                  port);
+                               cvm_oct_free_work(work);
+                               return 1;
+                       }
+               }
+       } else {
+               printk_ratelimited("Port %d receive error code %d, packet dropped\n",
+                                  port, work->word2.snoip.err_code);
+               cvm_oct_free_work(work);
+               return 1;
+       }
+
+       return 0;
+}
+
+static void copy_segments_to_skb(struct cvmx_wqe *work, struct sk_buff *skb)
+{
+       int segments = work->word2.s.bufs;
+       union cvmx_buf_ptr segment_ptr = work->packet_ptr;
+       int len = work->word1.len;
+       int segment_size;
+
+       while (segments--) {
+               union cvmx_buf_ptr next_ptr;
+
+               next_ptr = *(union cvmx_buf_ptr *)
+                       cvmx_phys_to_ptr(segment_ptr.s.addr - 8);
+
+               /*
+                * Octeon Errata PKI-100: The segment size is wrong.
+                *
+                * Until it is fixed, calculate the segment size based on
+                * the packet pool buffer size.
+                * When it is fixed, the following line should be replaced
+                * with this one:
+                * int segment_size = segment_ptr.s.size;
+                */
+               segment_size =
+                       CVMX_FPA_PACKET_POOL_SIZE -
+                       (segment_ptr.s.addr -
+                        (((segment_ptr.s.addr >> 7) -
+                          segment_ptr.s.back) << 7));
+
+               /* Don't copy more than what is left in the packet */
+               if (segment_size > len)
+                       segment_size = len;
+
+               /* Copy the data into the packet */
+               skb_put_data(skb, cvmx_phys_to_ptr(segment_ptr.s.addr),
+                            segment_size);
+               len -= segment_size;
+               segment_ptr = next_ptr;
+       }
+}
+
+static int cvm_oct_poll(struct oct_rx_group *rx_group, int budget)
+{
+       const int       coreid = cvmx_get_core_num();
+       u64     old_group_mask;
+       u64     old_scratch;
+       int             rx_count = 0;
+       int             did_work_request = 0;
+       int             packet_not_copied;
+
+       /* Prefetch cvm_oct_device since we know we need it soon */
+       prefetch(cvm_oct_device);
+
+       if (USE_ASYNC_IOBDMA) {
+               /* Save scratch in case userspace is using it */
+               CVMX_SYNCIOBDMA;
+               old_scratch = cvmx_scratch_read64(CVMX_SCR_SCRATCH);
+       }
+
+       /* Only allow work for our group (and preserve priorities) */
+       if (OCTEON_IS_MODEL(OCTEON_CN68XX)) {
+               old_group_mask = cvmx_read_csr(CVMX_SSO_PPX_GRP_MSK(coreid));
+               cvmx_write_csr(CVMX_SSO_PPX_GRP_MSK(coreid),
+                              BIT(rx_group->group));
+               cvmx_read_csr(CVMX_SSO_PPX_GRP_MSK(coreid)); /* Flush */
+       } else {
+               old_group_mask = cvmx_read_csr(CVMX_POW_PP_GRP_MSKX(coreid));
+               cvmx_write_csr(CVMX_POW_PP_GRP_MSKX(coreid),
+                              (old_group_mask & ~0xFFFFull) |
+                              BIT(rx_group->group));
+       }
+
+       if (USE_ASYNC_IOBDMA) {
+               cvmx_pow_work_request_async(CVMX_SCR_SCRATCH, CVMX_POW_NO_WAIT);
+               did_work_request = 1;
+       }
+
+       while (rx_count < budget) {
+               struct sk_buff *skb = NULL;
+               struct sk_buff **pskb = NULL;
+               int skb_in_hw;
+               struct cvmx_wqe *work;
+               int port;
+
+               if (USE_ASYNC_IOBDMA && did_work_request)
+                       work = cvmx_pow_work_response_async(CVMX_SCR_SCRATCH);
+               else
+                       work = cvmx_pow_work_request_sync(CVMX_POW_NO_WAIT);
+
+               prefetch(work);
+               did_work_request = 0;
+               if (!work) {
+                       if (OCTEON_IS_MODEL(OCTEON_CN68XX)) {
+                               cvmx_write_csr(CVMX_SSO_WQ_IQ_DIS,
+                                              BIT(rx_group->group));
+                               cvmx_write_csr(CVMX_SSO_WQ_INT,
+                                              BIT(rx_group->group));
+                       } else {
+                               union cvmx_pow_wq_int wq_int;
+
+                               wq_int.u64 = 0;
+                               wq_int.s.iq_dis = BIT(rx_group->group);
+                               wq_int.s.wq_int = BIT(rx_group->group);
+                               cvmx_write_csr(CVMX_POW_WQ_INT, wq_int.u64);
+                       }
+                       break;
+               }
+               pskb = (struct sk_buff **)
+                       (cvm_oct_get_buffer_ptr(work->packet_ptr) -
+                       sizeof(void *));
+               prefetch(pskb);
+
+               if (USE_ASYNC_IOBDMA && rx_count < (budget - 1)) {
+                       cvmx_pow_work_request_async_nocheck(CVMX_SCR_SCRATCH,
+                                                           CVMX_POW_NO_WAIT);
+                       did_work_request = 1;
+               }
+               rx_count++;
+
+               skb_in_hw = work->word2.s.bufs == 1;
+               if (likely(skb_in_hw)) {
+                       skb = *pskb;
+                       prefetch(&skb->head);
+                       prefetch(&skb->len);
+               }
+
+               if (octeon_has_feature(OCTEON_FEATURE_PKND))
+                       port = work->word0.pip.cn68xx.pknd;
+               else
+                       port = work->word1.cn38xx.ipprt;
+
+               prefetch(cvm_oct_device[port]);
+
+               /* Immediately throw away all packets with receive errors */
+               if (unlikely(work->word2.snoip.rcv_error)) {
+                       if (cvm_oct_check_rcv_error(work))
+                               continue;
+               }
+
+               /*
+                * We can only use the zero copy path if skbuffs are
+                * in the FPA pool and the packet fits in a single
+                * buffer.
+                */
+               if (likely(skb_in_hw)) {
+                       skb->data = skb->head + work->packet_ptr.s.addr -
+                               cvmx_ptr_to_phys(skb->head);
+                       prefetch(skb->data);
+                       skb->len = work->word1.len;
+                       skb_set_tail_pointer(skb, skb->len);
+                       packet_not_copied = 1;
+               } else {
+                       /*
+                        * We have to copy the packet. First allocate
+                        * an skbuff for it.
+                        */
+                       skb = dev_alloc_skb(work->word1.len);
+                       if (!skb) {
+                               cvm_oct_free_work(work);
+                               continue;
+                       }
+
+                       /*
+                        * Check if we've received a packet that was
+                        * entirely stored in the work entry.
+                        */
+                       if (unlikely(work->word2.s.bufs == 0)) {
+                               u8 *ptr = work->packet_data;
+
+                               if (likely(!work->word2.s.not_IP)) {
+                                       /*
+                                        * The beginning of the packet
+                                        * moves for IP packets.
+                                        */
+                                       if (work->word2.s.is_v6)
+                                               ptr += 2;
+                                       else
+                                               ptr += 6;
+                               }
+                               skb_put_data(skb, ptr, work->word1.len);
+                               /* No packet buffers to free */
+                       } else {
+                               copy_segments_to_skb(work, skb);
+                       }
+                       packet_not_copied = 0;
+               }
+               if (likely((port < TOTAL_NUMBER_OF_PORTS) &&
+                          cvm_oct_device[port])) {
+                       struct net_device *dev = cvm_oct_device[port];
+
+                       /*
+                        * Only accept packets for devices that are
+                        * currently up.
+                        */
+                       if (likely(dev->flags & IFF_UP)) {
+                               skb->protocol = eth_type_trans(skb, dev);
+                               skb->dev = dev;
+
+                               if (unlikely(work->word2.s.not_IP ||
+                                            work->word2.s.IP_exc ||
+                                            work->word2.s.L4_error ||
+                                            !work->word2.s.tcp_or_udp))
+                                       skb->ip_summed = CHECKSUM_NONE;
+                               else
+                                       skb->ip_summed = CHECKSUM_UNNECESSARY;
+
+                               /* Increment RX stats for virtual ports */
+                               if (port >= CVMX_PIP_NUM_INPUT_PORTS) {
+                                       dev->stats.rx_packets++;
+                                       dev->stats.rx_bytes += skb->len;
+                               }
+                               netif_receive_skb(skb);
+                       } else {
+                               /*
+                                * Drop any packet received for a device that
+                                * isn't up.
+                                */
+                               dev->stats.rx_dropped++;
+                               dev_kfree_skb_irq(skb);
+                       }
+               } else {
+                       /*
+                        * Drop any packet received for a device that
+                        * doesn't exist.
+                        */
+                       printk_ratelimited("Port %d not controlled by Linux, packet dropped\n",
+                                          port);
+                       dev_kfree_skb_irq(skb);
+               }
+               /*
+                * Check to see if the skbuff and work share the same
+                * packet buffer.
+                */
+               if (likely(packet_not_copied)) {
+                       /*
+                        * This buffer needs to be replaced, increment
+                        * the number of buffers we need to free by
+                        * one.
+                        */
+                       cvmx_fau_atomic_add32(FAU_NUM_PACKET_BUFFERS_TO_FREE,
+                                             1);
+
+                       cvmx_fpa_free(work, CVMX_FPA_WQE_POOL, 1);
+               } else {
+                       cvm_oct_free_work(work);
+               }
+       }
+       /* Restore the original POW group mask */
+       if (OCTEON_IS_MODEL(OCTEON_CN68XX)) {
+               cvmx_write_csr(CVMX_SSO_PPX_GRP_MSK(coreid), old_group_mask);
+               cvmx_read_csr(CVMX_SSO_PPX_GRP_MSK(coreid)); /* Flush */
+       } else {
+               cvmx_write_csr(CVMX_POW_PP_GRP_MSKX(coreid), old_group_mask);
+       }
+
+       if (USE_ASYNC_IOBDMA) {
+               /* Restore the scratch area */
+               cvmx_scratch_write64(CVMX_SCR_SCRATCH, old_scratch);
+       }
+       cvm_oct_rx_refill_pool(0);
+
+       return rx_count;
+}
+
+/**
+ * cvm_oct_napi_poll - the NAPI poll function.
+ * @napi: The NAPI instance.
+ * @budget: Maximum number of packets to receive.
+ *
+ * Returns the number of packets processed.
+ */
+static int cvm_oct_napi_poll(struct napi_struct *napi, int budget)
+{
+       struct oct_rx_group *rx_group = container_of(napi, struct oct_rx_group,
+                                                    napi);
+       int rx_count;
+
+       rx_count = cvm_oct_poll(rx_group, budget);
+
+       if (rx_count < budget) {
+               /* No more work */
+               napi_complete_done(napi, rx_count);
+               enable_irq(rx_group->irq);
+       }
+       return rx_count;
+}
+
+#ifdef CONFIG_NET_POLL_CONTROLLER
+/**
+ * cvm_oct_poll_controller - poll for receive packets
+ * device.
+ *
+ * @dev:    Device to poll. Unused
+ */
+void cvm_oct_poll_controller(struct net_device *dev)
+{
+       int i;
+
+       if (!atomic_read(&oct_rx_ready))
+               return;
+
+       for (i = 0; i < ARRAY_SIZE(oct_rx_group); i++) {
+               if (!(pow_receive_groups & BIT(i)))
+                       continue;
+
+               cvm_oct_poll(&oct_rx_group[i], 16);
+       }
+}
+#endif
+
+void cvm_oct_rx_initialize(void)
+{
+       int i;
+       struct net_device *dev_for_napi = NULL;
+
+       for (i = 0; i < TOTAL_NUMBER_OF_PORTS; i++) {
+               if (cvm_oct_device[i]) {
+                       dev_for_napi = cvm_oct_device[i];
+                       break;
+               }
+       }
+
+       if (!dev_for_napi)
+               panic("No net_devices were allocated.");
+
+       for (i = 0; i < ARRAY_SIZE(oct_rx_group); i++) {
+               int ret;
+
+               if (!(pow_receive_groups & BIT(i)))
+                       continue;
+
+               netif_napi_add(dev_for_napi, &oct_rx_group[i].napi,
+                              cvm_oct_napi_poll, rx_napi_weight);
+               napi_enable(&oct_rx_group[i].napi);
+
+               oct_rx_group[i].irq = OCTEON_IRQ_WORKQ0 + i;
+               oct_rx_group[i].group = i;
+
+               /* Register an IRQ handler to receive POW interrupts */
+               ret = request_irq(oct_rx_group[i].irq, cvm_oct_do_interrupt, 0,
+                                 "Ethernet", &oct_rx_group[i].napi);
+               if (ret)
+                       panic("Could not acquire Ethernet IRQ %d\n",
+                             oct_rx_group[i].irq);
+
+               disable_irq_nosync(oct_rx_group[i].irq);
+
+               /* Enable POW interrupt when our port has at least one packet */
+               if (OCTEON_IS_MODEL(OCTEON_CN68XX)) {
+                       union cvmx_sso_wq_int_thrx int_thr;
+                       union cvmx_pow_wq_int_pc int_pc;
+
+                       int_thr.u64 = 0;
+                       int_thr.s.tc_en = 1;
+                       int_thr.s.tc_thr = 1;
+                       cvmx_write_csr(CVMX_SSO_WQ_INT_THRX(i), int_thr.u64);
+
+                       int_pc.u64 = 0;
+                       int_pc.s.pc_thr = 5;
+                       cvmx_write_csr(CVMX_SSO_WQ_INT_PC, int_pc.u64);
+               } else {
+                       union cvmx_pow_wq_int_thrx int_thr;
+                       union cvmx_pow_wq_int_pc int_pc;
+
+                       int_thr.u64 = 0;
+                       int_thr.s.tc_en = 1;
+                       int_thr.s.tc_thr = 1;
+                       cvmx_write_csr(CVMX_POW_WQ_INT_THRX(i), int_thr.u64);
+
+                       int_pc.u64 = 0;
+                       int_pc.s.pc_thr = 5;
+                       cvmx_write_csr(CVMX_POW_WQ_INT_PC, int_pc.u64);
+               }
+
+               /* Schedule NAPI now. This will indirectly enable the
+                * interrupt.
+                */
+               napi_schedule(&oct_rx_group[i].napi);
+       }
+       atomic_inc(&oct_rx_ready);
+}
+
+void cvm_oct_rx_shutdown(void)
+{
+       int i;
+
+       for (i = 0; i < ARRAY_SIZE(oct_rx_group); i++) {
+               if (!(pow_receive_groups & BIT(i)))
+                       continue;
+
+               /* Disable POW interrupt */
+               if (OCTEON_IS_MODEL(OCTEON_CN68XX))
+                       cvmx_write_csr(CVMX_SSO_WQ_INT_THRX(i), 0);
+               else
+                       cvmx_write_csr(CVMX_POW_WQ_INT_THRX(i), 0);
+
+               /* Free the interrupt handler */
+               free_irq(oct_rx_group[i].irq, cvm_oct_device);
+
+               netif_napi_del(&oct_rx_group[i].napi);
+       }
+}
diff --git a/drivers/staging/octeon/ethernet-rx.h b/drivers/staging/octeon/ethernet-rx.h
new file mode 100644 (file)
index 0000000..ff6482f
--- /dev/null
@@ -0,0 +1,31 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+void cvm_oct_poll_controller(struct net_device *dev);
+void cvm_oct_rx_initialize(void);
+void cvm_oct_rx_shutdown(void);
+
+static inline void cvm_oct_rx_refill_pool(int fill_threshold)
+{
+       int number_to_free;
+       int num_freed;
+       /* Refill the packet buffer pool */
+       number_to_free =
+               cvmx_fau_fetch_and_add32(FAU_NUM_PACKET_BUFFERS_TO_FREE, 0);
+
+       if (number_to_free > fill_threshold) {
+               cvmx_fau_atomic_add32(FAU_NUM_PACKET_BUFFERS_TO_FREE,
+                                     -number_to_free);
+               num_freed = cvm_oct_mem_fill_fpa(CVMX_FPA_PACKET_POOL,
+                                                CVMX_FPA_PACKET_POOL_SIZE,
+                                                number_to_free);
+               if (num_freed != number_to_free) {
+                       cvmx_fau_atomic_add32(FAU_NUM_PACKET_BUFFERS_TO_FREE,
+                                             number_to_free - num_freed);
+               }
+       }
+}
diff --git a/drivers/staging/octeon/ethernet-sgmii.c b/drivers/staging/octeon/ethernet-sgmii.c
new file mode 100644 (file)
index 0000000..d7fbd91
--- /dev/null
@@ -0,0 +1,30 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+#include <linux/phy.h>
+#include <linux/kernel.h>
+#include <linux/netdevice.h>
+#include <linux/ratelimit.h>
+#include <net/dst.h>
+
+#include "octeon-ethernet.h"
+#include "ethernet-defines.h"
+#include "ethernet-util.h"
+#include "ethernet-mdio.h"
+
+int cvm_oct_sgmii_open(struct net_device *dev)
+{
+       return cvm_oct_common_open(dev, cvm_oct_link_poll);
+}
+
+int cvm_oct_sgmii_init(struct net_device *dev)
+{
+       cvm_oct_common_init(dev);
+
+       /* FIXME: Need autoneg logic */
+       return 0;
+}
diff --git a/drivers/staging/octeon/ethernet-spi.c b/drivers/staging/octeon/ethernet-spi.c
new file mode 100644 (file)
index 0000000..c582403
--- /dev/null
@@ -0,0 +1,226 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+#include <linux/kernel.h>
+#include <linux/netdevice.h>
+#include <linux/interrupt.h>
+#include <net/dst.h>
+
+#include "octeon-ethernet.h"
+#include "ethernet-defines.h"
+#include "ethernet-util.h"
+
+static int number_spi_ports;
+static int need_retrain[2] = { 0, 0 };
+
+static void cvm_oct_spxx_int_pr(union cvmx_spxx_int_reg spx_int_reg, int index)
+{
+       if (spx_int_reg.s.spf)
+               pr_err("SPI%d: SRX Spi4 interface down\n", index);
+       if (spx_int_reg.s.calerr)
+               pr_err("SPI%d: SRX Spi4 Calendar table parity error\n", index);
+       if (spx_int_reg.s.syncerr)
+               pr_err("SPI%d: SRX Consecutive Spi4 DIP4 errors have exceeded SPX_ERR_CTL[ERRCNT]\n",
+                      index);
+       if (spx_int_reg.s.diperr)
+               pr_err("SPI%d: SRX Spi4 DIP4 error\n", index);
+       if (spx_int_reg.s.tpaovr)
+               pr_err("SPI%d: SRX Selected port has hit TPA overflow\n",
+                      index);
+       if (spx_int_reg.s.rsverr)
+               pr_err("SPI%d: SRX Spi4 reserved control word detected\n",
+                      index);
+       if (spx_int_reg.s.drwnng)
+               pr_err("SPI%d: SRX Spi4 receive FIFO drowning/overflow\n",
+                      index);
+       if (spx_int_reg.s.clserr)
+               pr_err("SPI%d: SRX Spi4 packet closed on non-16B alignment without EOP\n",
+                      index);
+       if (spx_int_reg.s.spiovr)
+               pr_err("SPI%d: SRX Spi4 async FIFO overflow\n", index);
+       if (spx_int_reg.s.abnorm)
+               pr_err("SPI%d: SRX Abnormal packet termination (ERR bit)\n",
+                      index);
+       if (spx_int_reg.s.prtnxa)
+               pr_err("SPI%d: SRX Port out of range\n", index);
+}
+
+static void cvm_oct_stxx_int_pr(union cvmx_stxx_int_reg stx_int_reg, int index)
+{
+       if (stx_int_reg.s.syncerr)
+               pr_err("SPI%d: STX Interface encountered a fatal error\n",
+                      index);
+       if (stx_int_reg.s.frmerr)
+               pr_err("SPI%d: STX FRMCNT has exceeded STX_DIP_CNT[MAXFRM]\n",
+                      index);
+       if (stx_int_reg.s.unxfrm)
+               pr_err("SPI%d: STX Unexpected framing sequence\n", index);
+       if (stx_int_reg.s.nosync)
+               pr_err("SPI%d: STX ERRCNT has exceeded STX_DIP_CNT[MAXDIP]\n",
+                      index);
+       if (stx_int_reg.s.diperr)
+               pr_err("SPI%d: STX DIP2 error on the Spi4 Status channel\n",
+                      index);
+       if (stx_int_reg.s.datovr)
+               pr_err("SPI%d: STX Spi4 FIFO overflow error\n", index);
+       if (stx_int_reg.s.ovrbst)
+               pr_err("SPI%d: STX Transmit packet burst too big\n", index);
+       if (stx_int_reg.s.calpar1)
+               pr_err("SPI%d: STX Calendar Table Parity Error Bank%d\n",
+                      index, 1);
+       if (stx_int_reg.s.calpar0)
+               pr_err("SPI%d: STX Calendar Table Parity Error Bank%d\n",
+                      index, 0);
+}
+
+static irqreturn_t cvm_oct_spi_spx_int(int index)
+{
+       union cvmx_spxx_int_reg spx_int_reg;
+       union cvmx_stxx_int_reg stx_int_reg;
+
+       spx_int_reg.u64 = cvmx_read_csr(CVMX_SPXX_INT_REG(index));
+       cvmx_write_csr(CVMX_SPXX_INT_REG(index), spx_int_reg.u64);
+       if (!need_retrain[index]) {
+               spx_int_reg.u64 &= cvmx_read_csr(CVMX_SPXX_INT_MSK(index));
+               cvm_oct_spxx_int_pr(spx_int_reg, index);
+       }
+
+       stx_int_reg.u64 = cvmx_read_csr(CVMX_STXX_INT_REG(index));
+       cvmx_write_csr(CVMX_STXX_INT_REG(index), stx_int_reg.u64);
+       if (!need_retrain[index]) {
+               stx_int_reg.u64 &= cvmx_read_csr(CVMX_STXX_INT_MSK(index));
+               cvm_oct_stxx_int_pr(stx_int_reg, index);
+       }
+
+       cvmx_write_csr(CVMX_SPXX_INT_MSK(index), 0);
+       cvmx_write_csr(CVMX_STXX_INT_MSK(index), 0);
+       need_retrain[index] = 1;
+
+       return IRQ_HANDLED;
+}
+
+static irqreturn_t cvm_oct_spi_rml_interrupt(int cpl, void *dev_id)
+{
+       irqreturn_t return_status = IRQ_NONE;
+       union cvmx_npi_rsl_int_blocks rsl_int_blocks;
+
+       /* Check and see if this interrupt was caused by the GMX block */
+       rsl_int_blocks.u64 = cvmx_read_csr(CVMX_NPI_RSL_INT_BLOCKS);
+       if (rsl_int_blocks.s.spx1) /* 19 - SPX1_INT_REG & STX1_INT_REG */
+               return_status = cvm_oct_spi_spx_int(1);
+
+       if (rsl_int_blocks.s.spx0) /* 18 - SPX0_INT_REG & STX0_INT_REG */
+               return_status = cvm_oct_spi_spx_int(0);
+
+       return return_status;
+}
+
+static void cvm_oct_spi_enable_error_reporting(int interface)
+{
+       union cvmx_spxx_int_msk spxx_int_msk;
+       union cvmx_stxx_int_msk stxx_int_msk;
+
+       spxx_int_msk.u64 = cvmx_read_csr(CVMX_SPXX_INT_MSK(interface));
+       spxx_int_msk.s.calerr = 1;
+       spxx_int_msk.s.syncerr = 1;
+       spxx_int_msk.s.diperr = 1;
+       spxx_int_msk.s.tpaovr = 1;
+       spxx_int_msk.s.rsverr = 1;
+       spxx_int_msk.s.drwnng = 1;
+       spxx_int_msk.s.clserr = 1;
+       spxx_int_msk.s.spiovr = 1;
+       spxx_int_msk.s.abnorm = 1;
+       spxx_int_msk.s.prtnxa = 1;
+       cvmx_write_csr(CVMX_SPXX_INT_MSK(interface), spxx_int_msk.u64);
+
+       stxx_int_msk.u64 = cvmx_read_csr(CVMX_STXX_INT_MSK(interface));
+       stxx_int_msk.s.frmerr = 1;
+       stxx_int_msk.s.unxfrm = 1;
+       stxx_int_msk.s.nosync = 1;
+       stxx_int_msk.s.diperr = 1;
+       stxx_int_msk.s.datovr = 1;
+       stxx_int_msk.s.ovrbst = 1;
+       stxx_int_msk.s.calpar1 = 1;
+       stxx_int_msk.s.calpar0 = 1;
+       cvmx_write_csr(CVMX_STXX_INT_MSK(interface), stxx_int_msk.u64);
+}
+
+static void cvm_oct_spi_poll(struct net_device *dev)
+{
+       static int spi4000_port;
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       int interface;
+
+       for (interface = 0; interface < 2; interface++) {
+               if ((priv->port == interface * 16) && need_retrain[interface]) {
+                       if (cvmx_spi_restart_interface
+                           (interface, CVMX_SPI_MODE_DUPLEX, 10) == 0) {
+                               need_retrain[interface] = 0;
+                               cvm_oct_spi_enable_error_reporting(interface);
+                       }
+               }
+
+               /*
+                * The SPI4000 TWSI interface is very slow. In order
+                * not to bring the system to a crawl, we only poll a
+                * single port every second. This means negotiation
+                * speed changes take up to 10 seconds, but at least
+                * we don't waste absurd amounts of time waiting for
+                * TWSI.
+                */
+               if (priv->port == spi4000_port) {
+                       /*
+                        * This function does nothing if it is called on an
+                        * interface without a SPI4000.
+                        */
+                       cvmx_spi4000_check_speed(interface, priv->port);
+                       /*
+                        * Normal ordering increments. By decrementing
+                        * we only match once per iteration.
+                        */
+                       spi4000_port--;
+                       if (spi4000_port < 0)
+                               spi4000_port = 10;
+               }
+       }
+}
+
+int cvm_oct_spi_init(struct net_device *dev)
+{
+       int r;
+       struct octeon_ethernet *priv = netdev_priv(dev);
+
+       if (number_spi_ports == 0) {
+               r = request_irq(OCTEON_IRQ_RML, cvm_oct_spi_rml_interrupt,
+                               IRQF_SHARED, "SPI", &number_spi_ports);
+               if (r)
+                       return r;
+       }
+       number_spi_ports++;
+
+       if ((priv->port == 0) || (priv->port == 16)) {
+               cvm_oct_spi_enable_error_reporting(INTERFACE(priv->port));
+               priv->poll = cvm_oct_spi_poll;
+       }
+       cvm_oct_common_init(dev);
+       return 0;
+}
+
+void cvm_oct_spi_uninit(struct net_device *dev)
+{
+       int interface;
+
+       cvm_oct_common_uninit(dev);
+       number_spi_ports--;
+       if (number_spi_ports == 0) {
+               for (interface = 0; interface < 2; interface++) {
+                       cvmx_write_csr(CVMX_SPXX_INT_MSK(interface), 0);
+                       cvmx_write_csr(CVMX_STXX_INT_MSK(interface), 0);
+               }
+               free_irq(OCTEON_IRQ_RML, &number_spi_ports);
+       }
+}
diff --git a/drivers/staging/octeon/ethernet-tx.c b/drivers/staging/octeon/ethernet-tx.c
new file mode 100644 (file)
index 0000000..b334cf8
--- /dev/null
@@ -0,0 +1,717 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2010 Cavium Networks
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/ip.h>
+#include <linux/ratelimit.h>
+#include <linux/string.h>
+#include <linux/interrupt.h>
+#include <net/dst.h>
+#ifdef CONFIG_XFRM
+#include <linux/xfrm.h>
+#include <net/xfrm.h>
+#endif /* CONFIG_XFRM */
+
+#include <linux/atomic.h>
+#include <net/sch_generic.h>
+
+#include "octeon-ethernet.h"
+#include "ethernet-defines.h"
+#include "ethernet-tx.h"
+#include "ethernet-util.h"
+
+#define CVM_OCT_SKB_CB(skb)    ((u64 *)((skb)->cb))
+
+/*
+ * You can define GET_SKBUFF_QOS() to override how the skbuff output
+ * function determines which output queue is used. The default
+ * implementation always uses the base queue for the port. If, for
+ * example, you wanted to use the skb->priority field, define
+ * GET_SKBUFF_QOS as: #define GET_SKBUFF_QOS(skb) ((skb)->priority)
+ */
+#ifndef GET_SKBUFF_QOS
+#define GET_SKBUFF_QOS(skb) 0
+#endif
+
+static void cvm_oct_tx_do_cleanup(unsigned long arg);
+static DECLARE_TASKLET(cvm_oct_tx_cleanup_tasklet, cvm_oct_tx_do_cleanup, 0);
+
+/* Maximum number of SKBs to try to free per xmit packet. */
+#define MAX_SKB_TO_FREE (MAX_OUT_QUEUE_DEPTH * 2)
+
+static inline int cvm_oct_adjust_skb_to_free(int skb_to_free, int fau)
+{
+       int undo;
+
+       undo = skb_to_free > 0 ? MAX_SKB_TO_FREE : skb_to_free +
+                                                  MAX_SKB_TO_FREE;
+       if (undo > 0)
+               cvmx_fau_atomic_add32(fau, -undo);
+       skb_to_free = -skb_to_free > MAX_SKB_TO_FREE ? MAX_SKB_TO_FREE :
+                                                      -skb_to_free;
+       return skb_to_free;
+}
+
+static void cvm_oct_kick_tx_poll_watchdog(void)
+{
+       union cvmx_ciu_timx ciu_timx;
+
+       ciu_timx.u64 = 0;
+       ciu_timx.s.one_shot = 1;
+       ciu_timx.s.len = cvm_oct_tx_poll_interval;
+       cvmx_write_csr(CVMX_CIU_TIMX(1), ciu_timx.u64);
+}
+
+static void cvm_oct_free_tx_skbs(struct net_device *dev)
+{
+       int skb_to_free;
+       int qos, queues_per_port;
+       int total_freed = 0;
+       int total_remaining = 0;
+       unsigned long flags;
+       struct octeon_ethernet *priv = netdev_priv(dev);
+
+       queues_per_port = cvmx_pko_get_num_queues(priv->port);
+       /* Drain any pending packets in the free list */
+       for (qos = 0; qos < queues_per_port; qos++) {
+               if (skb_queue_len(&priv->tx_free_list[qos]) == 0)
+                       continue;
+               skb_to_free = cvmx_fau_fetch_and_add32(priv->fau + qos * 4,
+                                                      MAX_SKB_TO_FREE);
+               skb_to_free = cvm_oct_adjust_skb_to_free(skb_to_free,
+                                                        priv->fau + qos * 4);
+               total_freed += skb_to_free;
+               if (skb_to_free > 0) {
+                       struct sk_buff *to_free_list = NULL;
+
+                       spin_lock_irqsave(&priv->tx_free_list[qos].lock, flags);
+                       while (skb_to_free > 0) {
+                               struct sk_buff *t;
+
+                               t = __skb_dequeue(&priv->tx_free_list[qos]);
+                               t->next = to_free_list;
+                               to_free_list = t;
+                               skb_to_free--;
+                       }
+                       spin_unlock_irqrestore(&priv->tx_free_list[qos].lock,
+                                              flags);
+                       /* Do the actual freeing outside of the lock. */
+                       while (to_free_list) {
+                               struct sk_buff *t = to_free_list;
+
+                               to_free_list = to_free_list->next;
+                               dev_kfree_skb_any(t);
+                       }
+               }
+               total_remaining += skb_queue_len(&priv->tx_free_list[qos]);
+       }
+       if (total_remaining < MAX_OUT_QUEUE_DEPTH && netif_queue_stopped(dev))
+               netif_wake_queue(dev);
+       if (total_remaining)
+               cvm_oct_kick_tx_poll_watchdog();
+}
+
+/**
+ * cvm_oct_xmit - transmit a packet
+ * @skb:    Packet to send
+ * @dev:    Device info structure
+ *
+ * Returns Always returns NETDEV_TX_OK
+ */
+int cvm_oct_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       union cvmx_pko_command_word0 pko_command;
+       union cvmx_buf_ptr hw_buffer;
+       u64 old_scratch;
+       u64 old_scratch2;
+       int qos;
+       int i;
+       enum {QUEUE_CORE, QUEUE_HW, QUEUE_DROP} queue_type;
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       struct sk_buff *to_free_list;
+       int skb_to_free;
+       int buffers_to_free;
+       u32 total_to_clean;
+       unsigned long flags;
+#if REUSE_SKBUFFS_WITHOUT_FREE
+       unsigned char *fpa_head;
+#endif
+
+       /*
+        * Prefetch the private data structure.  It is larger than the
+        * one cache line.
+        */
+       prefetch(priv);
+
+       /*
+        * The check on CVMX_PKO_QUEUES_PER_PORT_* is designed to
+        * completely remove "qos" in the event neither interface
+        * supports multiple queues per port.
+        */
+       if ((CVMX_PKO_QUEUES_PER_PORT_INTERFACE0 > 1) ||
+           (CVMX_PKO_QUEUES_PER_PORT_INTERFACE1 > 1)) {
+               qos = GET_SKBUFF_QOS(skb);
+               if (qos <= 0)
+                       qos = 0;
+               else if (qos >= cvmx_pko_get_num_queues(priv->port))
+                       qos = 0;
+       } else {
+               qos = 0;
+       }
+
+       if (USE_ASYNC_IOBDMA) {
+               /* Save scratch in case userspace is using it */
+               CVMX_SYNCIOBDMA;
+               old_scratch = cvmx_scratch_read64(CVMX_SCR_SCRATCH);
+               old_scratch2 = cvmx_scratch_read64(CVMX_SCR_SCRATCH + 8);
+
+               /*
+                * Fetch and increment the number of packets to be
+                * freed.
+                */
+               cvmx_fau_async_fetch_and_add32(CVMX_SCR_SCRATCH + 8,
+                                              FAU_NUM_PACKET_BUFFERS_TO_FREE,
+                                              0);
+               cvmx_fau_async_fetch_and_add32(CVMX_SCR_SCRATCH,
+                                              priv->fau + qos * 4,
+                                              MAX_SKB_TO_FREE);
+       }
+
+       /*
+        * We have space for 6 segment pointers, If there will be more
+        * than that, we must linearize.
+        */
+       if (unlikely(skb_shinfo(skb)->nr_frags > 5)) {
+               if (unlikely(__skb_linearize(skb))) {
+                       queue_type = QUEUE_DROP;
+                       if (USE_ASYNC_IOBDMA) {
+                               /*
+                                * Get the number of skbuffs in use
+                                * by the hardware
+                                */
+                               CVMX_SYNCIOBDMA;
+                               skb_to_free =
+                                       cvmx_scratch_read64(CVMX_SCR_SCRATCH);
+                       } else {
+                               /*
+                                * Get the number of skbuffs in use
+                                * by the hardware
+                                */
+                               skb_to_free =
+                                    cvmx_fau_fetch_and_add32(priv->fau +
+                                                             qos * 4,
+                                                             MAX_SKB_TO_FREE);
+                       }
+                       skb_to_free = cvm_oct_adjust_skb_to_free(skb_to_free,
+                                                                priv->fau +
+                                                                qos * 4);
+                       spin_lock_irqsave(&priv->tx_free_list[qos].lock, flags);
+                       goto skip_xmit;
+               }
+       }
+
+       /*
+        * The CN3XXX series of parts has an errata (GMX-401) which
+        * causes the GMX block to hang if a collision occurs towards
+        * the end of a <68 byte packet. As a workaround for this, we
+        * pad packets to be 68 bytes whenever we are in half duplex
+        * mode. We don't handle the case of having a small packet but
+        * no room to add the padding.  The kernel should always give
+        * us at least a cache line
+        */
+       if ((skb->len < 64) && OCTEON_IS_MODEL(OCTEON_CN3XXX)) {
+               union cvmx_gmxx_prtx_cfg gmx_prt_cfg;
+               int interface = INTERFACE(priv->port);
+               int index = INDEX(priv->port);
+
+               if (interface < 2) {
+                       /* We only need to pad packet in half duplex mode */
+                       gmx_prt_cfg.u64 =
+                           cvmx_read_csr(CVMX_GMXX_PRTX_CFG(index, interface));
+                       if (gmx_prt_cfg.s.duplex == 0) {
+                               int add_bytes = 64 - skb->len;
+
+                               if ((skb_tail_pointer(skb) + add_bytes) <=
+                                   skb_end_pointer(skb))
+                                       __skb_put_zero(skb, add_bytes);
+                       }
+               }
+       }
+
+       /* Build the PKO command */
+       pko_command.u64 = 0;
+#ifdef __LITTLE_ENDIAN
+       pko_command.s.le = 1;
+#endif
+       pko_command.s.n2 = 1;   /* Don't pollute L2 with the outgoing packet */
+       pko_command.s.segs = 1;
+       pko_command.s.total_bytes = skb->len;
+       pko_command.s.size0 = CVMX_FAU_OP_SIZE_32;
+       pko_command.s.subone0 = 1;
+
+       pko_command.s.dontfree = 1;
+
+       /* Build the PKO buffer pointer */
+       hw_buffer.u64 = 0;
+       if (skb_shinfo(skb)->nr_frags == 0) {
+               hw_buffer.s.addr = XKPHYS_TO_PHYS((uintptr_t)skb->data);
+               hw_buffer.s.pool = 0;
+               hw_buffer.s.size = skb->len;
+       } else {
+               hw_buffer.s.addr = XKPHYS_TO_PHYS((uintptr_t)skb->data);
+               hw_buffer.s.pool = 0;
+               hw_buffer.s.size = skb_headlen(skb);
+               CVM_OCT_SKB_CB(skb)[0] = hw_buffer.u64;
+               for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+                       skb_frag_t *fs = skb_shinfo(skb)->frags + i;
+
+                       hw_buffer.s.addr =
+                               XKPHYS_TO_PHYS((uintptr_t)skb_frag_address(fs));
+                       hw_buffer.s.size = skb_frag_size(fs);
+                       CVM_OCT_SKB_CB(skb)[i + 1] = hw_buffer.u64;
+               }
+               hw_buffer.s.addr =
+                       XKPHYS_TO_PHYS((uintptr_t)CVM_OCT_SKB_CB(skb));
+               hw_buffer.s.size = skb_shinfo(skb)->nr_frags + 1;
+               pko_command.s.segs = skb_shinfo(skb)->nr_frags + 1;
+               pko_command.s.gather = 1;
+               goto dont_put_skbuff_in_hw;
+       }
+
+       /*
+        * See if we can put this skb in the FPA pool. Any strange
+        * behavior from the Linux networking stack will most likely
+        * be caused by a bug in the following code. If some field is
+        * in use by the network stack and gets carried over when a
+        * buffer is reused, bad things may happen.  If in doubt and
+        * you dont need the absolute best performance, disable the
+        * define REUSE_SKBUFFS_WITHOUT_FREE. The reuse of buffers has
+        * shown a 25% increase in performance under some loads.
+        */
+#if REUSE_SKBUFFS_WITHOUT_FREE
+       fpa_head = skb->head + 256 - ((unsigned long)skb->head & 0x7f);
+       if (unlikely(skb->data < fpa_head)) {
+               /* TX buffer beginning can't meet FPA alignment constraints */
+               goto dont_put_skbuff_in_hw;
+       }
+       if (unlikely
+           ((skb_end_pointer(skb) - fpa_head) < CVMX_FPA_PACKET_POOL_SIZE)) {
+               /* TX buffer isn't large enough for the FPA */
+               goto dont_put_skbuff_in_hw;
+       }
+       if (unlikely(skb_shared(skb))) {
+               /* TX buffer sharing data with someone else */
+               goto dont_put_skbuff_in_hw;
+       }
+       if (unlikely(skb_cloned(skb))) {
+               /* TX buffer has been cloned */
+               goto dont_put_skbuff_in_hw;
+       }
+       if (unlikely(skb_header_cloned(skb))) {
+               /* TX buffer header has been cloned */
+               goto dont_put_skbuff_in_hw;
+       }
+       if (unlikely(skb->destructor)) {
+               /* TX buffer has a destructor */
+               goto dont_put_skbuff_in_hw;
+       }
+       if (unlikely(skb_shinfo(skb)->nr_frags)) {
+               /* TX buffer has fragments */
+               goto dont_put_skbuff_in_hw;
+       }
+       if (unlikely
+           (skb->truesize !=
+            sizeof(*skb) + skb_end_offset(skb))) {
+               /* TX buffer truesize has been changed */
+               goto dont_put_skbuff_in_hw;
+       }
+
+       /*
+        * We can use this buffer in the FPA.  We don't need the FAU
+        * update anymore
+        */
+       pko_command.s.dontfree = 0;
+
+       hw_buffer.s.back = ((unsigned long)skb->data >> 7) -
+                          ((unsigned long)fpa_head >> 7);
+
+       *(struct sk_buff **)(fpa_head - sizeof(void *)) = skb;
+
+       /*
+        * The skbuff will be reused without ever being freed. We must
+        * cleanup a bunch of core things.
+        */
+       dst_release(skb_dst(skb));
+       skb_dst_set(skb, NULL);
+       skb_ext_reset(skb);
+       nf_reset_ct(skb);
+
+#ifdef CONFIG_NET_SCHED
+       skb->tc_index = 0;
+       skb_reset_tc(skb);
+#endif /* CONFIG_NET_SCHED */
+#endif /* REUSE_SKBUFFS_WITHOUT_FREE */
+
+dont_put_skbuff_in_hw:
+
+       /* Check if we can use the hardware checksumming */
+       if ((skb->protocol == htons(ETH_P_IP)) &&
+           (ip_hdr(skb)->version == 4) &&
+           (ip_hdr(skb)->ihl == 5) &&
+           ((ip_hdr(skb)->frag_off == 0) ||
+            (ip_hdr(skb)->frag_off == htons(1 << 14))) &&
+           ((ip_hdr(skb)->protocol == IPPROTO_TCP) ||
+            (ip_hdr(skb)->protocol == IPPROTO_UDP))) {
+               /* Use hardware checksum calc */
+               pko_command.s.ipoffp1 = skb_network_offset(skb) + 1;
+       }
+
+       if (USE_ASYNC_IOBDMA) {
+               /* Get the number of skbuffs in use by the hardware */
+               CVMX_SYNCIOBDMA;
+               skb_to_free = cvmx_scratch_read64(CVMX_SCR_SCRATCH);
+               buffers_to_free = cvmx_scratch_read64(CVMX_SCR_SCRATCH + 8);
+       } else {
+               /* Get the number of skbuffs in use by the hardware */
+               skb_to_free = cvmx_fau_fetch_and_add32(priv->fau + qos * 4,
+                                                      MAX_SKB_TO_FREE);
+               buffers_to_free =
+                   cvmx_fau_fetch_and_add32(FAU_NUM_PACKET_BUFFERS_TO_FREE, 0);
+       }
+
+       skb_to_free = cvm_oct_adjust_skb_to_free(skb_to_free,
+                                                priv->fau + qos * 4);
+
+       /*
+        * If we're sending faster than the receive can free them then
+        * don't do the HW free.
+        */
+       if ((buffers_to_free < -100) && !pko_command.s.dontfree)
+               pko_command.s.dontfree = 1;
+
+       if (pko_command.s.dontfree) {
+               queue_type = QUEUE_CORE;
+               pko_command.s.reg0 = priv->fau + qos * 4;
+       } else {
+               queue_type = QUEUE_HW;
+       }
+       if (USE_ASYNC_IOBDMA)
+               cvmx_fau_async_fetch_and_add32(CVMX_SCR_SCRATCH,
+                                              FAU_TOTAL_TX_TO_CLEAN, 1);
+
+       spin_lock_irqsave(&priv->tx_free_list[qos].lock, flags);
+
+       /* Drop this packet if we have too many already queued to the HW */
+       if (unlikely(skb_queue_len(&priv->tx_free_list[qos]) >=
+                    MAX_OUT_QUEUE_DEPTH)) {
+               if (dev->tx_queue_len != 0) {
+                       /* Drop the lock when notifying the core.  */
+                       spin_unlock_irqrestore(&priv->tx_free_list[qos].lock,
+                                              flags);
+                       netif_stop_queue(dev);
+                       spin_lock_irqsave(&priv->tx_free_list[qos].lock,
+                                         flags);
+               } else {
+                       /* If not using normal queueing.  */
+                       queue_type = QUEUE_DROP;
+                       goto skip_xmit;
+               }
+       }
+
+       cvmx_pko_send_packet_prepare(priv->port, priv->queue + qos,
+                                    CVMX_PKO_LOCK_NONE);
+
+       /* Send the packet to the output queue */
+       if (unlikely(cvmx_pko_send_packet_finish(priv->port,
+                                                priv->queue + qos,
+                                                pko_command, hw_buffer,
+                                                CVMX_PKO_LOCK_NONE))) {
+               printk_ratelimited("%s: Failed to send the packet\n",
+                                  dev->name);
+               queue_type = QUEUE_DROP;
+       }
+skip_xmit:
+       to_free_list = NULL;
+
+       switch (queue_type) {
+       case QUEUE_DROP:
+               skb->next = to_free_list;
+               to_free_list = skb;
+               dev->stats.tx_dropped++;
+               break;
+       case QUEUE_HW:
+               cvmx_fau_atomic_add32(FAU_NUM_PACKET_BUFFERS_TO_FREE, -1);
+               break;
+       case QUEUE_CORE:
+               __skb_queue_tail(&priv->tx_free_list[qos], skb);
+               break;
+       default:
+               BUG();
+       }
+
+       while (skb_to_free > 0) {
+               struct sk_buff *t = __skb_dequeue(&priv->tx_free_list[qos]);
+
+               t->next = to_free_list;
+               to_free_list = t;
+               skb_to_free--;
+       }
+
+       spin_unlock_irqrestore(&priv->tx_free_list[qos].lock, flags);
+
+       /* Do the actual freeing outside of the lock. */
+       while (to_free_list) {
+               struct sk_buff *t = to_free_list;
+
+               to_free_list = to_free_list->next;
+               dev_kfree_skb_any(t);
+       }
+
+       if (USE_ASYNC_IOBDMA) {
+               CVMX_SYNCIOBDMA;
+               total_to_clean = cvmx_scratch_read64(CVMX_SCR_SCRATCH);
+               /* Restore the scratch area */
+               cvmx_scratch_write64(CVMX_SCR_SCRATCH, old_scratch);
+               cvmx_scratch_write64(CVMX_SCR_SCRATCH + 8, old_scratch2);
+       } else {
+               total_to_clean =
+                       cvmx_fau_fetch_and_add32(FAU_TOTAL_TX_TO_CLEAN, 1);
+       }
+
+       if (total_to_clean & 0x3ff) {
+               /*
+                * Schedule the cleanup tasklet every 1024 packets for
+                * the pathological case of high traffic on one port
+                * delaying clean up of packets on a different port
+                * that is blocked waiting for the cleanup.
+                */
+               tasklet_schedule(&cvm_oct_tx_cleanup_tasklet);
+       }
+
+       cvm_oct_kick_tx_poll_watchdog();
+
+       return NETDEV_TX_OK;
+}
+
+/**
+ * cvm_oct_xmit_pow - transmit a packet to the POW
+ * @skb:    Packet to send
+ * @dev:    Device info structure
+
+ * Returns Always returns zero
+ */
+int cvm_oct_xmit_pow(struct sk_buff *skb, struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       void *packet_buffer;
+       void *copy_location;
+
+       /* Get a work queue entry */
+       struct cvmx_wqe *work = cvmx_fpa_alloc(CVMX_FPA_WQE_POOL);
+
+       if (unlikely(!work)) {
+               printk_ratelimited("%s: Failed to allocate a work queue entry\n",
+                                  dev->name);
+               dev->stats.tx_dropped++;
+               dev_kfree_skb_any(skb);
+               return 0;
+       }
+
+       /* Get a packet buffer */
+       packet_buffer = cvmx_fpa_alloc(CVMX_FPA_PACKET_POOL);
+       if (unlikely(!packet_buffer)) {
+               printk_ratelimited("%s: Failed to allocate a packet buffer\n",
+                                  dev->name);
+               cvmx_fpa_free(work, CVMX_FPA_WQE_POOL, 1);
+               dev->stats.tx_dropped++;
+               dev_kfree_skb_any(skb);
+               return 0;
+       }
+
+       /*
+        * Calculate where we need to copy the data to. We need to
+        * leave 8 bytes for a next pointer (unused). We also need to
+        * include any configure skip. Then we need to align the IP
+        * packet src and dest into the same 64bit word. The below
+        * calculation may add a little extra, but that doesn't
+        * hurt.
+        */
+       copy_location = packet_buffer + sizeof(u64);
+       copy_location += ((CVMX_HELPER_FIRST_MBUFF_SKIP + 7) & 0xfff8) + 6;
+
+       /*
+        * We have to copy the packet since whoever processes this
+        * packet will free it to a hardware pool. We can't use the
+        * trick of counting outstanding packets like in
+        * cvm_oct_xmit.
+        */
+       memcpy(copy_location, skb->data, skb->len);
+
+       /*
+        * Fill in some of the work queue fields. We may need to add
+        * more if the software at the other end needs them.
+        */
+       if (!OCTEON_IS_MODEL(OCTEON_CN68XX))
+               work->word0.pip.cn38xx.hw_chksum = skb->csum;
+       work->word1.len = skb->len;
+       cvmx_wqe_set_port(work, priv->port);
+       cvmx_wqe_set_qos(work, priv->port & 0x7);
+       cvmx_wqe_set_grp(work, pow_send_group);
+       work->word1.tag_type = CVMX_HELPER_INPUT_TAG_TYPE;
+       work->word1.tag = pow_send_group;       /* FIXME */
+       /* Default to zero. Sets of zero later are commented out */
+       work->word2.u64 = 0;
+       work->word2.s.bufs = 1;
+       work->packet_ptr.u64 = 0;
+       work->packet_ptr.s.addr = cvmx_ptr_to_phys(copy_location);
+       work->packet_ptr.s.pool = CVMX_FPA_PACKET_POOL;
+       work->packet_ptr.s.size = CVMX_FPA_PACKET_POOL_SIZE;
+       work->packet_ptr.s.back = (copy_location - packet_buffer) >> 7;
+
+       if (skb->protocol == htons(ETH_P_IP)) {
+               work->word2.s.ip_offset = 14;
+#if 0
+               work->word2.s.vlan_valid = 0;   /* FIXME */
+               work->word2.s.vlan_cfi = 0;     /* FIXME */
+               work->word2.s.vlan_id = 0;      /* FIXME */
+               work->word2.s.dec_ipcomp = 0;   /* FIXME */
+#endif
+               work->word2.s.tcp_or_udp =
+                   (ip_hdr(skb)->protocol == IPPROTO_TCP) ||
+                   (ip_hdr(skb)->protocol == IPPROTO_UDP);
+#if 0
+               /* FIXME */
+               work->word2.s.dec_ipsec = 0;
+               /* We only support IPv4 right now */
+               work->word2.s.is_v6 = 0;
+               /* Hardware would set to zero */
+               work->word2.s.software = 0;
+               /* No error, packet is internal */
+               work->word2.s.L4_error = 0;
+#endif
+               work->word2.s.is_frag = !((ip_hdr(skb)->frag_off == 0) ||
+                                         (ip_hdr(skb)->frag_off ==
+                                             cpu_to_be16(1 << 14)));
+#if 0
+               /* Assume Linux is sending a good packet */
+               work->word2.s.IP_exc = 0;
+#endif
+               work->word2.s.is_bcast = (skb->pkt_type == PACKET_BROADCAST);
+               work->word2.s.is_mcast = (skb->pkt_type == PACKET_MULTICAST);
+#if 0
+               /* This is an IP packet */
+               work->word2.s.not_IP = 0;
+               /* No error, packet is internal */
+               work->word2.s.rcv_error = 0;
+               /* No error, packet is internal */
+               work->word2.s.err_code = 0;
+#endif
+
+               /*
+                * When copying the data, include 4 bytes of the
+                * ethernet header to align the same way hardware
+                * does.
+                */
+               memcpy(work->packet_data, skb->data + 10,
+                      sizeof(work->packet_data));
+       } else {
+#if 0
+               work->word2.snoip.vlan_valid = 0;       /* FIXME */
+               work->word2.snoip.vlan_cfi = 0; /* FIXME */
+               work->word2.snoip.vlan_id = 0;  /* FIXME */
+               work->word2.snoip.software = 0; /* Hardware would set to zero */
+#endif
+               work->word2.snoip.is_rarp = skb->protocol == htons(ETH_P_RARP);
+               work->word2.snoip.is_arp = skb->protocol == htons(ETH_P_ARP);
+               work->word2.snoip.is_bcast =
+                   (skb->pkt_type == PACKET_BROADCAST);
+               work->word2.snoip.is_mcast =
+                   (skb->pkt_type == PACKET_MULTICAST);
+               work->word2.snoip.not_IP = 1;   /* IP was done up above */
+#if 0
+               /* No error, packet is internal */
+               work->word2.snoip.rcv_error = 0;
+               /* No error, packet is internal */
+               work->word2.snoip.err_code = 0;
+#endif
+               memcpy(work->packet_data, skb->data, sizeof(work->packet_data));
+       }
+
+       /* Submit the packet to the POW */
+       cvmx_pow_work_submit(work, work->word1.tag, work->word1.tag_type,
+                            cvmx_wqe_get_qos(work), cvmx_wqe_get_grp(work));
+       dev->stats.tx_packets++;
+       dev->stats.tx_bytes += skb->len;
+       dev_consume_skb_any(skb);
+       return 0;
+}
+
+/**
+ * cvm_oct_tx_shutdown_dev - free all skb that are currently queued for TX.
+ * @dev:    Device being shutdown
+ *
+ */
+void cvm_oct_tx_shutdown_dev(struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       unsigned long flags;
+       int qos;
+
+       for (qos = 0; qos < 16; qos++) {
+               spin_lock_irqsave(&priv->tx_free_list[qos].lock, flags);
+               while (skb_queue_len(&priv->tx_free_list[qos]))
+                       dev_kfree_skb_any(__skb_dequeue
+                                         (&priv->tx_free_list[qos]));
+               spin_unlock_irqrestore(&priv->tx_free_list[qos].lock, flags);
+       }
+}
+
+static void cvm_oct_tx_do_cleanup(unsigned long arg)
+{
+       int port;
+
+       for (port = 0; port < TOTAL_NUMBER_OF_PORTS; port++) {
+               if (cvm_oct_device[port]) {
+                       struct net_device *dev = cvm_oct_device[port];
+
+                       cvm_oct_free_tx_skbs(dev);
+               }
+       }
+}
+
+static irqreturn_t cvm_oct_tx_cleanup_watchdog(int cpl, void *dev_id)
+{
+       /* Disable the interrupt.  */
+       cvmx_write_csr(CVMX_CIU_TIMX(1), 0);
+       /* Do the work in the tasklet.  */
+       tasklet_schedule(&cvm_oct_tx_cleanup_tasklet);
+       return IRQ_HANDLED;
+}
+
+void cvm_oct_tx_initialize(void)
+{
+       int i;
+
+       /* Disable the interrupt.  */
+       cvmx_write_csr(CVMX_CIU_TIMX(1), 0);
+       /* Register an IRQ handler to receive CIU_TIMX(1) interrupts */
+       i = request_irq(OCTEON_IRQ_TIMER1,
+                       cvm_oct_tx_cleanup_watchdog, 0,
+                       "Ethernet", cvm_oct_device);
+
+       if (i)
+               panic("Could not acquire Ethernet IRQ %d\n", OCTEON_IRQ_TIMER1);
+}
+
+void cvm_oct_tx_shutdown(void)
+{
+       /* Free the interrupt handler */
+       free_irq(OCTEON_IRQ_TIMER1, cvm_oct_device);
+}
diff --git a/drivers/staging/octeon/ethernet-tx.h b/drivers/staging/octeon/ethernet-tx.h
new file mode 100644 (file)
index 0000000..78936e9
--- /dev/null
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+int cvm_oct_xmit(struct sk_buff *skb, struct net_device *dev);
+int cvm_oct_xmit_pow(struct sk_buff *skb, struct net_device *dev);
+int cvm_oct_transmit_qos(struct net_device *dev, void *work_queue_entry,
+                        int do_free, int qos);
+void cvm_oct_tx_initialize(void);
+void cvm_oct_tx_shutdown(void);
+void cvm_oct_tx_shutdown_dev(struct net_device *dev);
diff --git a/drivers/staging/octeon/ethernet-util.h b/drivers/staging/octeon/ethernet-util.h
new file mode 100644 (file)
index 0000000..2af83a1
--- /dev/null
@@ -0,0 +1,47 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+/**
+ * cvm_oct_get_buffer_ptr - convert packet data address to pointer
+ * @packet_ptr: Packet data hardware address
+ *
+ * Returns Packet buffer pointer
+ */
+static inline void *cvm_oct_get_buffer_ptr(union cvmx_buf_ptr packet_ptr)
+{
+       return cvmx_phys_to_ptr(((packet_ptr.s.addr >> 7) - packet_ptr.s.back)
+                               << 7);
+}
+
+/**
+ * INTERFACE - convert IPD port to logical interface
+ * @ipd_port: Port to check
+ *
+ * Returns Logical interface
+ */
+static inline int INTERFACE(int ipd_port)
+{
+       int interface;
+
+       if (ipd_port == CVMX_PIP_NUM_INPUT_PORTS)
+               return 10;
+       interface = cvmx_helper_get_interface_num(ipd_port);
+       if (interface >= 0)
+               return interface;
+       panic("Illegal ipd_port %d passed to %s\n", ipd_port, __func__);
+}
+
+/**
+ * INDEX - convert IPD/PKO port number to the port's interface index
+ * @ipd_port: Port to check
+ *
+ * Returns Index into interface port list
+ */
+static inline int INDEX(int ipd_port)
+{
+       return cvmx_helper_get_interface_index_num(ipd_port);
+}
diff --git a/drivers/staging/octeon/ethernet.c b/drivers/staging/octeon/ethernet.c
new file mode 100644 (file)
index 0000000..f42c381
--- /dev/null
@@ -0,0 +1,992 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2007 Cavium Networks
+ */
+
+#include <linux/platform_device.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/phy.h>
+#include <linux/slab.h>
+#include <linux/interrupt.h>
+#include <linux/of_net.h>
+#include <linux/if_ether.h>
+#include <linux/if_vlan.h>
+
+#include <net/dst.h>
+
+#include "octeon-ethernet.h"
+#include "ethernet-defines.h"
+#include "ethernet-mem.h"
+#include "ethernet-rx.h"
+#include "ethernet-tx.h"
+#include "ethernet-mdio.h"
+#include "ethernet-util.h"
+
+#define OCTEON_MAX_MTU 65392
+
+static int num_packet_buffers = 1024;
+module_param(num_packet_buffers, int, 0444);
+MODULE_PARM_DESC(num_packet_buffers, "\n"
+       "\tNumber of packet buffers to allocate and store in the\n"
+       "\tFPA. By default, 1024 packet buffers are used.\n");
+
+static int pow_receive_group = 15;
+module_param(pow_receive_group, int, 0444);
+MODULE_PARM_DESC(pow_receive_group, "\n"
+       "\tPOW group to receive packets from. All ethernet hardware\n"
+       "\twill be configured to send incoming packets to this POW\n"
+       "\tgroup. Also any other software can submit packets to this\n"
+       "\tgroup for the kernel to process.");
+
+static int receive_group_order;
+module_param(receive_group_order, int, 0444);
+MODULE_PARM_DESC(receive_group_order, "\n"
+       "\tOrder (0..4) of receive groups to take into use. Ethernet hardware\n"
+       "\twill be configured to send incoming packets to multiple POW\n"
+       "\tgroups. pow_receive_group parameter is ignored when multiple\n"
+       "\tgroups are taken into use and groups are allocated starting\n"
+       "\tfrom 0. By default, a single group is used.\n");
+
+int pow_send_group = -1;
+module_param(pow_send_group, int, 0644);
+MODULE_PARM_DESC(pow_send_group, "\n"
+       "\tPOW group to send packets to other software on. This\n"
+       "\tcontrols the creation of the virtual device pow0.\n"
+       "\talways_use_pow also depends on this value.");
+
+int always_use_pow;
+module_param(always_use_pow, int, 0444);
+MODULE_PARM_DESC(always_use_pow, "\n"
+       "\tWhen set, always send to the pow group. This will cause\n"
+       "\tpackets sent to real ethernet devices to be sent to the\n"
+       "\tPOW group instead of the hardware. Unless some other\n"
+       "\tapplication changes the config, packets will still be\n"
+       "\treceived from the low level hardware. Use this option\n"
+       "\tto allow a CVMX app to intercept all packets from the\n"
+       "\tlinux kernel. You must specify pow_send_group along with\n"
+       "\tthis option.");
+
+char pow_send_list[128] = "";
+module_param_string(pow_send_list, pow_send_list, sizeof(pow_send_list), 0444);
+MODULE_PARM_DESC(pow_send_list, "\n"
+       "\tComma separated list of ethernet devices that should use the\n"
+       "\tPOW for transmit instead of the actual ethernet hardware. This\n"
+       "\tis a per port version of always_use_pow. always_use_pow takes\n"
+       "\tprecedence over this list. For example, setting this to\n"
+       "\t\"eth2,spi3,spi7\" would cause these three devices to transmit\n"
+       "\tusing the pow_send_group.");
+
+int rx_napi_weight = 32;
+module_param(rx_napi_weight, int, 0444);
+MODULE_PARM_DESC(rx_napi_weight, "The NAPI WEIGHT parameter.");
+
+/* Mask indicating which receive groups are in use. */
+int pow_receive_groups;
+
+/*
+ * cvm_oct_poll_queue_stopping - flag to indicate polling should stop.
+ *
+ * Set to one right before cvm_oct_poll_queue is destroyed.
+ */
+atomic_t cvm_oct_poll_queue_stopping = ATOMIC_INIT(0);
+
+/*
+ * Array of every ethernet device owned by this driver indexed by
+ * the ipd input port number.
+ */
+struct net_device *cvm_oct_device[TOTAL_NUMBER_OF_PORTS];
+
+u64 cvm_oct_tx_poll_interval;
+
+static void cvm_oct_rx_refill_worker(struct work_struct *work);
+static DECLARE_DELAYED_WORK(cvm_oct_rx_refill_work, cvm_oct_rx_refill_worker);
+
+static void cvm_oct_rx_refill_worker(struct work_struct *work)
+{
+       /*
+        * FPA 0 may have been drained, try to refill it if we need
+        * more than num_packet_buffers / 2, otherwise normal receive
+        * processing will refill it.  If it were drained, no packets
+        * could be received so cvm_oct_napi_poll would never be
+        * invoked to do the refill.
+        */
+       cvm_oct_rx_refill_pool(num_packet_buffers / 2);
+
+       if (!atomic_read(&cvm_oct_poll_queue_stopping))
+               schedule_delayed_work(&cvm_oct_rx_refill_work, HZ);
+}
+
+static void cvm_oct_periodic_worker(struct work_struct *work)
+{
+       struct octeon_ethernet *priv = container_of(work,
+                                                   struct octeon_ethernet,
+                                                   port_periodic_work.work);
+
+       if (priv->poll)
+               priv->poll(cvm_oct_device[priv->port]);
+
+       cvm_oct_device[priv->port]->netdev_ops->ndo_get_stats
+                                               (cvm_oct_device[priv->port]);
+
+       if (!atomic_read(&cvm_oct_poll_queue_stopping))
+               schedule_delayed_work(&priv->port_periodic_work, HZ);
+}
+
+static void cvm_oct_configure_common_hw(void)
+{
+       /* Setup the FPA */
+       cvmx_fpa_enable();
+       cvm_oct_mem_fill_fpa(CVMX_FPA_PACKET_POOL, CVMX_FPA_PACKET_POOL_SIZE,
+                            num_packet_buffers);
+       cvm_oct_mem_fill_fpa(CVMX_FPA_WQE_POOL, CVMX_FPA_WQE_POOL_SIZE,
+                            num_packet_buffers);
+       if (CVMX_FPA_OUTPUT_BUFFER_POOL != CVMX_FPA_PACKET_POOL)
+               cvm_oct_mem_fill_fpa(CVMX_FPA_OUTPUT_BUFFER_POOL,
+                                    CVMX_FPA_OUTPUT_BUFFER_POOL_SIZE, 1024);
+
+#ifdef __LITTLE_ENDIAN
+       {
+               union cvmx_ipd_ctl_status ipd_ctl_status;
+
+               ipd_ctl_status.u64 = cvmx_read_csr(CVMX_IPD_CTL_STATUS);
+               ipd_ctl_status.s.pkt_lend = 1;
+               ipd_ctl_status.s.wqe_lend = 1;
+               cvmx_write_csr(CVMX_IPD_CTL_STATUS, ipd_ctl_status.u64);
+       }
+#endif
+
+       cvmx_helper_setup_red(num_packet_buffers / 4, num_packet_buffers / 8);
+}
+
+/**
+ * cvm_oct_free_work- Free a work queue entry
+ *
+ * @work_queue_entry: Work queue entry to free
+ *
+ * Returns Zero on success, Negative on failure.
+ */
+int cvm_oct_free_work(void *work_queue_entry)
+{
+       struct cvmx_wqe *work = work_queue_entry;
+
+       int segments = work->word2.s.bufs;
+       union cvmx_buf_ptr segment_ptr = work->packet_ptr;
+
+       while (segments--) {
+               union cvmx_buf_ptr next_ptr = *(union cvmx_buf_ptr *)
+                       cvmx_phys_to_ptr(segment_ptr.s.addr - 8);
+               if (unlikely(!segment_ptr.s.i))
+                       cvmx_fpa_free(cvm_oct_get_buffer_ptr(segment_ptr),
+                                     segment_ptr.s.pool,
+                                     CVMX_FPA_PACKET_POOL_SIZE / 128);
+               segment_ptr = next_ptr;
+       }
+       cvmx_fpa_free(work, CVMX_FPA_WQE_POOL, 1);
+
+       return 0;
+}
+EXPORT_SYMBOL(cvm_oct_free_work);
+
+/**
+ * cvm_oct_common_get_stats - get the low level ethernet statistics
+ * @dev:    Device to get the statistics from
+ *
+ * Returns Pointer to the statistics
+ */
+static struct net_device_stats *cvm_oct_common_get_stats(struct net_device *dev)
+{
+       cvmx_pip_port_status_t rx_status;
+       cvmx_pko_port_status_t tx_status;
+       struct octeon_ethernet *priv = netdev_priv(dev);
+
+       if (priv->port < CVMX_PIP_NUM_INPUT_PORTS) {
+               if (octeon_is_simulation()) {
+                       /* The simulator doesn't support statistics */
+                       memset(&rx_status, 0, sizeof(rx_status));
+                       memset(&tx_status, 0, sizeof(tx_status));
+               } else {
+                       cvmx_pip_get_port_status(priv->port, 1, &rx_status);
+                       cvmx_pko_get_port_status(priv->port, 1, &tx_status);
+               }
+
+               dev->stats.rx_packets += rx_status.inb_packets;
+               dev->stats.tx_packets += tx_status.packets;
+               dev->stats.rx_bytes += rx_status.inb_octets;
+               dev->stats.tx_bytes += tx_status.octets;
+               dev->stats.multicast += rx_status.multicast_packets;
+               dev->stats.rx_crc_errors += rx_status.inb_errors;
+               dev->stats.rx_frame_errors += rx_status.fcs_align_err_packets;
+               dev->stats.rx_dropped += rx_status.dropped_packets;
+       }
+
+       return &dev->stats;
+}
+
+/**
+ * cvm_oct_common_change_mtu - change the link MTU
+ * @dev:     Device to change
+ * @new_mtu: The new MTU
+ *
+ * Returns Zero on success
+ */
+static int cvm_oct_common_change_mtu(struct net_device *dev, int new_mtu)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       int interface = INTERFACE(priv->port);
+#if IS_ENABLED(CONFIG_VLAN_8021Q)
+       int vlan_bytes = VLAN_HLEN;
+#else
+       int vlan_bytes = 0;
+#endif
+       int mtu_overhead = ETH_HLEN + ETH_FCS_LEN + vlan_bytes;
+
+       dev->mtu = new_mtu;
+
+       if ((interface < 2) &&
+           (cvmx_helper_interface_get_mode(interface) !=
+               CVMX_HELPER_INTERFACE_MODE_SPI)) {
+               int index = INDEX(priv->port);
+               /* Add ethernet header and FCS, and VLAN if configured. */
+               int max_packet = new_mtu + mtu_overhead;
+
+               if (OCTEON_IS_MODEL(OCTEON_CN3XXX) ||
+                   OCTEON_IS_MODEL(OCTEON_CN58XX)) {
+                       /* Signal errors on packets larger than the MTU */
+                       cvmx_write_csr(CVMX_GMXX_RXX_FRM_MAX(index, interface),
+                                      max_packet);
+               } else {
+                       /*
+                        * Set the hardware to truncate packets larger
+                        * than the MTU and smaller the 64 bytes.
+                        */
+                       union cvmx_pip_frm_len_chkx frm_len_chk;
+
+                       frm_len_chk.u64 = 0;
+                       frm_len_chk.s.minlen = VLAN_ETH_ZLEN;
+                       frm_len_chk.s.maxlen = max_packet;
+                       cvmx_write_csr(CVMX_PIP_FRM_LEN_CHKX(interface),
+                                      frm_len_chk.u64);
+               }
+               /*
+                * Set the hardware to truncate packets larger than
+                * the MTU. The jabber register must be set to a
+                * multiple of 8 bytes, so round up.
+                */
+               cvmx_write_csr(CVMX_GMXX_RXX_JABBER(index, interface),
+                              (max_packet + 7) & ~7u);
+       }
+       return 0;
+}
+
+/**
+ * cvm_oct_common_set_multicast_list - set the multicast list
+ * @dev:    Device to work on
+ */
+static void cvm_oct_common_set_multicast_list(struct net_device *dev)
+{
+       union cvmx_gmxx_prtx_cfg gmx_cfg;
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       int interface = INTERFACE(priv->port);
+
+       if ((interface < 2) &&
+           (cvmx_helper_interface_get_mode(interface) !=
+               CVMX_HELPER_INTERFACE_MODE_SPI)) {
+               union cvmx_gmxx_rxx_adr_ctl control;
+               int index = INDEX(priv->port);
+
+               control.u64 = 0;
+               control.s.bcst = 1;     /* Allow broadcast MAC addresses */
+
+               if (!netdev_mc_empty(dev) || (dev->flags & IFF_ALLMULTI) ||
+                   (dev->flags & IFF_PROMISC))
+                       /* Force accept multicast packets */
+                       control.s.mcst = 2;
+               else
+                       /* Force reject multicast packets */
+                       control.s.mcst = 1;
+
+               if (dev->flags & IFF_PROMISC)
+                       /*
+                        * Reject matches if promisc. Since CAM is
+                        * shut off, should accept everything.
+                        */
+                       control.s.cam_mode = 0;
+               else
+                       /* Filter packets based on the CAM */
+                       control.s.cam_mode = 1;
+
+               gmx_cfg.u64 =
+                   cvmx_read_csr(CVMX_GMXX_PRTX_CFG(index, interface));
+               cvmx_write_csr(CVMX_GMXX_PRTX_CFG(index, interface),
+                              gmx_cfg.u64 & ~1ull);
+
+               cvmx_write_csr(CVMX_GMXX_RXX_ADR_CTL(index, interface),
+                              control.u64);
+               if (dev->flags & IFF_PROMISC)
+                       cvmx_write_csr(CVMX_GMXX_RXX_ADR_CAM_EN
+                                      (index, interface), 0);
+               else
+                       cvmx_write_csr(CVMX_GMXX_RXX_ADR_CAM_EN
+                                      (index, interface), 1);
+
+               cvmx_write_csr(CVMX_GMXX_PRTX_CFG(index, interface),
+                              gmx_cfg.u64);
+       }
+}
+
+static int cvm_oct_set_mac_filter(struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       union cvmx_gmxx_prtx_cfg gmx_cfg;
+       int interface = INTERFACE(priv->port);
+
+       if ((interface < 2) &&
+           (cvmx_helper_interface_get_mode(interface) !=
+               CVMX_HELPER_INTERFACE_MODE_SPI)) {
+               int i;
+               u8 *ptr = dev->dev_addr;
+               u64 mac = 0;
+               int index = INDEX(priv->port);
+
+               for (i = 0; i < 6; i++)
+                       mac = (mac << 8) | (u64)ptr[i];
+
+               gmx_cfg.u64 =
+                   cvmx_read_csr(CVMX_GMXX_PRTX_CFG(index, interface));
+               cvmx_write_csr(CVMX_GMXX_PRTX_CFG(index, interface),
+                              gmx_cfg.u64 & ~1ull);
+
+               cvmx_write_csr(CVMX_GMXX_SMACX(index, interface), mac);
+               cvmx_write_csr(CVMX_GMXX_RXX_ADR_CAM0(index, interface),
+                              ptr[0]);
+               cvmx_write_csr(CVMX_GMXX_RXX_ADR_CAM1(index, interface),
+                              ptr[1]);
+               cvmx_write_csr(CVMX_GMXX_RXX_ADR_CAM2(index, interface),
+                              ptr[2]);
+               cvmx_write_csr(CVMX_GMXX_RXX_ADR_CAM3(index, interface),
+                              ptr[3]);
+               cvmx_write_csr(CVMX_GMXX_RXX_ADR_CAM4(index, interface),
+                              ptr[4]);
+               cvmx_write_csr(CVMX_GMXX_RXX_ADR_CAM5(index, interface),
+                              ptr[5]);
+               cvm_oct_common_set_multicast_list(dev);
+               cvmx_write_csr(CVMX_GMXX_PRTX_CFG(index, interface),
+                              gmx_cfg.u64);
+       }
+       return 0;
+}
+
+/**
+ * cvm_oct_common_set_mac_address - set the hardware MAC address for a device
+ * @dev:    The device in question.
+ * @addr:   Socket address.
+ *
+ * Returns Zero on success
+ */
+static int cvm_oct_common_set_mac_address(struct net_device *dev, void *addr)
+{
+       int r = eth_mac_addr(dev, addr);
+
+       if (r)
+               return r;
+       return cvm_oct_set_mac_filter(dev);
+}
+
+/**
+ * cvm_oct_common_init - per network device initialization
+ * @dev:    Device to initialize
+ *
+ * Returns Zero on success
+ */
+int cvm_oct_common_init(struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       const u8 *mac = NULL;
+
+       if (priv->of_node)
+               mac = of_get_mac_address(priv->of_node);
+
+       if (!IS_ERR_OR_NULL(mac))
+               ether_addr_copy(dev->dev_addr, mac);
+       else
+               eth_hw_addr_random(dev);
+
+       /*
+        * Force the interface to use the POW send if always_use_pow
+        * was specified or it is in the pow send list.
+        */
+       if ((pow_send_group != -1) &&
+           (always_use_pow || strstr(pow_send_list, dev->name)))
+               priv->queue = -1;
+
+       if (priv->queue != -1)
+               dev->features |= NETIF_F_SG | NETIF_F_IP_CSUM;
+
+       /* We do our own locking, Linux doesn't need to */
+       dev->features |= NETIF_F_LLTX;
+       dev->ethtool_ops = &cvm_oct_ethtool_ops;
+
+       cvm_oct_set_mac_filter(dev);
+       dev_set_mtu(dev, dev->mtu);
+
+       /*
+        * Zero out stats for port so we won't mistakenly show
+        * counters from the bootloader.
+        */
+       memset(dev->netdev_ops->ndo_get_stats(dev), 0,
+              sizeof(struct net_device_stats));
+
+       if (dev->netdev_ops->ndo_stop)
+               dev->netdev_ops->ndo_stop(dev);
+
+       return 0;
+}
+
+void cvm_oct_common_uninit(struct net_device *dev)
+{
+       if (dev->phydev)
+               phy_disconnect(dev->phydev);
+}
+
+int cvm_oct_common_open(struct net_device *dev,
+                       void (*link_poll)(struct net_device *))
+{
+       union cvmx_gmxx_prtx_cfg gmx_cfg;
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       int interface = INTERFACE(priv->port);
+       int index = INDEX(priv->port);
+       union cvmx_helper_link_info link_info;
+       int rv;
+
+       rv = cvm_oct_phy_setup_device(dev);
+       if (rv)
+               return rv;
+
+       gmx_cfg.u64 = cvmx_read_csr(CVMX_GMXX_PRTX_CFG(index, interface));
+       gmx_cfg.s.en = 1;
+       if (octeon_has_feature(OCTEON_FEATURE_PKND))
+               gmx_cfg.s.pknd = priv->port;
+       cvmx_write_csr(CVMX_GMXX_PRTX_CFG(index, interface), gmx_cfg.u64);
+
+       if (octeon_is_simulation())
+               return 0;
+
+       if (dev->phydev) {
+               int r = phy_read_status(dev->phydev);
+
+               if (r == 0 && dev->phydev->link == 0)
+                       netif_carrier_off(dev);
+               cvm_oct_adjust_link(dev);
+       } else {
+               link_info = cvmx_helper_link_get(priv->port);
+               if (!link_info.s.link_up)
+                       netif_carrier_off(dev);
+               priv->poll = link_poll;
+               link_poll(dev);
+       }
+
+       return 0;
+}
+
+void cvm_oct_link_poll(struct net_device *dev)
+{
+       struct octeon_ethernet *priv = netdev_priv(dev);
+       union cvmx_helper_link_info link_info;
+
+       link_info = cvmx_helper_link_get(priv->port);
+       if (link_info.u64 == priv->link_info)
+               return;
+
+       if (cvmx_helper_link_set(priv->port, link_info))
+               link_info.u64 = priv->link_info;
+       else
+               priv->link_info = link_info.u64;
+
+       if (link_info.s.link_up) {
+               if (!netif_carrier_ok(dev))
+                       netif_carrier_on(dev);
+       } else if (netif_carrier_ok(dev)) {
+               netif_carrier_off(dev);
+       }
+       cvm_oct_note_carrier(priv, link_info);
+}
+
+static int cvm_oct_xaui_open(struct net_device *dev)
+{
+       return cvm_oct_common_open(dev, cvm_oct_link_poll);
+}
+
+static const struct net_device_ops cvm_oct_npi_netdev_ops = {
+       .ndo_init               = cvm_oct_common_init,
+       .ndo_uninit             = cvm_oct_common_uninit,
+       .ndo_start_xmit         = cvm_oct_xmit,
+       .ndo_set_rx_mode        = cvm_oct_common_set_multicast_list,
+       .ndo_set_mac_address    = cvm_oct_common_set_mac_address,
+       .ndo_do_ioctl           = cvm_oct_ioctl,
+       .ndo_change_mtu         = cvm_oct_common_change_mtu,
+       .ndo_get_stats          = cvm_oct_common_get_stats,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = cvm_oct_poll_controller,
+#endif
+};
+
+static const struct net_device_ops cvm_oct_xaui_netdev_ops = {
+       .ndo_init               = cvm_oct_common_init,
+       .ndo_uninit             = cvm_oct_common_uninit,
+       .ndo_open               = cvm_oct_xaui_open,
+       .ndo_stop               = cvm_oct_common_stop,
+       .ndo_start_xmit         = cvm_oct_xmit,
+       .ndo_set_rx_mode        = cvm_oct_common_set_multicast_list,
+       .ndo_set_mac_address    = cvm_oct_common_set_mac_address,
+       .ndo_do_ioctl           = cvm_oct_ioctl,
+       .ndo_change_mtu         = cvm_oct_common_change_mtu,
+       .ndo_get_stats          = cvm_oct_common_get_stats,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = cvm_oct_poll_controller,
+#endif
+};
+
+static const struct net_device_ops cvm_oct_sgmii_netdev_ops = {
+       .ndo_init               = cvm_oct_sgmii_init,
+       .ndo_uninit             = cvm_oct_common_uninit,
+       .ndo_open               = cvm_oct_sgmii_open,
+       .ndo_stop               = cvm_oct_common_stop,
+       .ndo_start_xmit         = cvm_oct_xmit,
+       .ndo_set_rx_mode        = cvm_oct_common_set_multicast_list,
+       .ndo_set_mac_address    = cvm_oct_common_set_mac_address,
+       .ndo_do_ioctl           = cvm_oct_ioctl,
+       .ndo_change_mtu         = cvm_oct_common_change_mtu,
+       .ndo_get_stats          = cvm_oct_common_get_stats,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = cvm_oct_poll_controller,
+#endif
+};
+
+static const struct net_device_ops cvm_oct_spi_netdev_ops = {
+       .ndo_init               = cvm_oct_spi_init,
+       .ndo_uninit             = cvm_oct_spi_uninit,
+       .ndo_start_xmit         = cvm_oct_xmit,
+       .ndo_set_rx_mode        = cvm_oct_common_set_multicast_list,
+       .ndo_set_mac_address    = cvm_oct_common_set_mac_address,
+       .ndo_do_ioctl           = cvm_oct_ioctl,
+       .ndo_change_mtu         = cvm_oct_common_change_mtu,
+       .ndo_get_stats          = cvm_oct_common_get_stats,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = cvm_oct_poll_controller,
+#endif
+};
+
+static const struct net_device_ops cvm_oct_rgmii_netdev_ops = {
+       .ndo_init               = cvm_oct_common_init,
+       .ndo_uninit             = cvm_oct_common_uninit,
+       .ndo_open               = cvm_oct_rgmii_open,
+       .ndo_stop               = cvm_oct_common_stop,
+       .ndo_start_xmit         = cvm_oct_xmit,
+       .ndo_set_rx_mode        = cvm_oct_common_set_multicast_list,
+       .ndo_set_mac_address    = cvm_oct_common_set_mac_address,
+       .ndo_do_ioctl           = cvm_oct_ioctl,
+       .ndo_change_mtu         = cvm_oct_common_change_mtu,
+       .ndo_get_stats          = cvm_oct_common_get_stats,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = cvm_oct_poll_controller,
+#endif
+};
+
+static const struct net_device_ops cvm_oct_pow_netdev_ops = {
+       .ndo_init               = cvm_oct_common_init,
+       .ndo_start_xmit         = cvm_oct_xmit_pow,
+       .ndo_set_rx_mode        = cvm_oct_common_set_multicast_list,
+       .ndo_set_mac_address    = cvm_oct_common_set_mac_address,
+       .ndo_do_ioctl           = cvm_oct_ioctl,
+       .ndo_change_mtu         = cvm_oct_common_change_mtu,
+       .ndo_get_stats          = cvm_oct_common_get_stats,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = cvm_oct_poll_controller,
+#endif
+};
+
+static struct device_node *cvm_oct_of_get_child
+                               (const struct device_node *parent, int reg_val)
+{
+       struct device_node *node = NULL;
+       int size;
+       const __be32 *addr;
+
+       for (;;) {
+               node = of_get_next_child(parent, node);
+               if (!node)
+                       break;
+               addr = of_get_property(node, "reg", &size);
+               if (addr && (be32_to_cpu(*addr) == reg_val))
+                       break;
+       }
+       return node;
+}
+
+static struct device_node *cvm_oct_node_for_port(struct device_node *pip,
+                                                int interface, int port)
+{
+       struct device_node *ni, *np;
+
+       ni = cvm_oct_of_get_child(pip, interface);
+       if (!ni)
+               return NULL;
+
+       np = cvm_oct_of_get_child(ni, port);
+       of_node_put(ni);
+
+       return np;
+}
+
+static void cvm_set_rgmii_delay(struct octeon_ethernet *priv, int iface,
+                               int port)
+{
+       struct device_node *np = priv->of_node;
+       u32 delay_value;
+       bool rx_delay;
+       bool tx_delay;
+
+       /* By default, both RX/TX delay is enabled in
+        * __cvmx_helper_rgmii_enable().
+        */
+       rx_delay = true;
+       tx_delay = true;
+
+       if (!of_property_read_u32(np, "rx-delay", &delay_value)) {
+               cvmx_write_csr(CVMX_ASXX_RX_CLK_SETX(port, iface), delay_value);
+               rx_delay = delay_value > 0;
+       }
+       if (!of_property_read_u32(np, "tx-delay", &delay_value)) {
+               cvmx_write_csr(CVMX_ASXX_TX_CLK_SETX(port, iface), delay_value);
+               tx_delay = delay_value > 0;
+       }
+
+       if (!rx_delay && !tx_delay)
+               priv->phy_mode = PHY_INTERFACE_MODE_RGMII_ID;
+       else if (!rx_delay)
+               priv->phy_mode = PHY_INTERFACE_MODE_RGMII_RXID;
+       else if (!tx_delay)
+               priv->phy_mode = PHY_INTERFACE_MODE_RGMII_TXID;
+       else
+               priv->phy_mode = PHY_INTERFACE_MODE_RGMII;
+}
+
+static int cvm_oct_probe(struct platform_device *pdev)
+{
+       int num_interfaces;
+       int interface;
+       int fau = FAU_NUM_PACKET_BUFFERS_TO_FREE;
+       int qos;
+       struct device_node *pip;
+       int mtu_overhead = ETH_HLEN + ETH_FCS_LEN;
+
+#if IS_ENABLED(CONFIG_VLAN_8021Q)
+       mtu_overhead += VLAN_HLEN;
+#endif
+
+       octeon_mdiobus_force_mod_depencency();
+
+       pip = pdev->dev.of_node;
+       if (!pip) {
+               pr_err("Error: No 'pip' in /aliases\n");
+               return -EINVAL;
+       }
+
+       cvm_oct_configure_common_hw();
+
+       cvmx_helper_initialize_packet_io_global();
+
+       if (receive_group_order) {
+               if (receive_group_order > 4)
+                       receive_group_order = 4;
+               pow_receive_groups = (1 << (1 << receive_group_order)) - 1;
+       } else {
+               pow_receive_groups = BIT(pow_receive_group);
+       }
+
+       /* Change the input group for all ports before input is enabled */
+       num_interfaces = cvmx_helper_get_number_of_interfaces();
+       for (interface = 0; interface < num_interfaces; interface++) {
+               int num_ports = cvmx_helper_ports_on_interface(interface);
+               int port;
+
+               for (port = cvmx_helper_get_ipd_port(interface, 0);
+                    port < cvmx_helper_get_ipd_port(interface, num_ports);
+                    port++) {
+                       union cvmx_pip_prt_tagx pip_prt_tagx;
+
+                       pip_prt_tagx.u64 =
+                           cvmx_read_csr(CVMX_PIP_PRT_TAGX(port));
+
+                       if (receive_group_order) {
+                               int tag_mask;
+
+                               /* We support only 16 groups at the moment, so
+                                * always disable the two additional "hidden"
+                                * tag_mask bits on CN68XX.
+                                */
+                               if (OCTEON_IS_MODEL(OCTEON_CN68XX))
+                                       pip_prt_tagx.u64 |= 0x3ull << 44;
+
+                               tag_mask = ~((1 << receive_group_order) - 1);
+                               pip_prt_tagx.s.grptagbase       = 0;
+                               pip_prt_tagx.s.grptagmask       = tag_mask;
+                               pip_prt_tagx.s.grptag           = 1;
+                               pip_prt_tagx.s.tag_mode         = 0;
+                               pip_prt_tagx.s.inc_prt_flag     = 1;
+                               pip_prt_tagx.s.ip6_dprt_flag    = 1;
+                               pip_prt_tagx.s.ip4_dprt_flag    = 1;
+                               pip_prt_tagx.s.ip6_sprt_flag    = 1;
+                               pip_prt_tagx.s.ip4_sprt_flag    = 1;
+                               pip_prt_tagx.s.ip6_dst_flag     = 1;
+                               pip_prt_tagx.s.ip4_dst_flag     = 1;
+                               pip_prt_tagx.s.ip6_src_flag     = 1;
+                               pip_prt_tagx.s.ip4_src_flag     = 1;
+                               pip_prt_tagx.s.grp              = 0;
+                       } else {
+                               pip_prt_tagx.s.grptag   = 0;
+                               pip_prt_tagx.s.grp      = pow_receive_group;
+                       }
+
+                       cvmx_write_csr(CVMX_PIP_PRT_TAGX(port),
+                                      pip_prt_tagx.u64);
+               }
+       }
+
+       cvmx_helper_ipd_and_packet_input_enable();
+
+       memset(cvm_oct_device, 0, sizeof(cvm_oct_device));
+
+       /*
+        * Initialize the FAU used for counting packet buffers that
+        * need to be freed.
+        */
+       cvmx_fau_atomic_write32(FAU_NUM_PACKET_BUFFERS_TO_FREE, 0);
+
+       /* Initialize the FAU used for counting tx SKBs that need to be freed */
+       cvmx_fau_atomic_write32(FAU_TOTAL_TX_TO_CLEAN, 0);
+
+       if ((pow_send_group != -1)) {
+               struct net_device *dev;
+
+               dev = alloc_etherdev(sizeof(struct octeon_ethernet));
+               if (dev) {
+                       /* Initialize the device private structure. */
+                       struct octeon_ethernet *priv = netdev_priv(dev);
+
+                       SET_NETDEV_DEV(dev, &pdev->dev);
+                       dev->netdev_ops = &cvm_oct_pow_netdev_ops;
+                       priv->imode = CVMX_HELPER_INTERFACE_MODE_DISABLED;
+                       priv->port = CVMX_PIP_NUM_INPUT_PORTS;
+                       priv->queue = -1;
+                       strscpy(dev->name, "pow%d", sizeof(dev->name));
+                       for (qos = 0; qos < 16; qos++)
+                               skb_queue_head_init(&priv->tx_free_list[qos]);
+                       dev->min_mtu = VLAN_ETH_ZLEN - mtu_overhead;
+                       dev->max_mtu = OCTEON_MAX_MTU - mtu_overhead;
+
+                       if (register_netdev(dev) < 0) {
+                               pr_err("Failed to register ethernet device for POW\n");
+                               free_netdev(dev);
+                       } else {
+                               cvm_oct_device[CVMX_PIP_NUM_INPUT_PORTS] = dev;
+                               pr_info("%s: POW send group %d, receive group %d\n",
+                                       dev->name, pow_send_group,
+                                       pow_receive_group);
+                       }
+               } else {
+                       pr_err("Failed to allocate ethernet device for POW\n");
+               }
+       }
+
+       num_interfaces = cvmx_helper_get_number_of_interfaces();
+       for (interface = 0; interface < num_interfaces; interface++) {
+               cvmx_helper_interface_mode_t imode =
+                   cvmx_helper_interface_get_mode(interface);
+               int num_ports = cvmx_helper_ports_on_interface(interface);
+               int port;
+               int port_index;
+
+               for (port_index = 0,
+                    port = cvmx_helper_get_ipd_port(interface, 0);
+                    port < cvmx_helper_get_ipd_port(interface, num_ports);
+                    port_index++, port++) {
+                       struct octeon_ethernet *priv;
+                       struct net_device *dev =
+                           alloc_etherdev(sizeof(struct octeon_ethernet));
+                       if (!dev) {
+                               pr_err("Failed to allocate ethernet device for port %d\n",
+                                      port);
+                               continue;
+                       }
+
+                       /* Initialize the device private structure. */
+                       SET_NETDEV_DEV(dev, &pdev->dev);
+                       priv = netdev_priv(dev);
+                       priv->netdev = dev;
+                       priv->of_node = cvm_oct_node_for_port(pip, interface,
+                                                             port_index);
+
+                       INIT_DELAYED_WORK(&priv->port_periodic_work,
+                                         cvm_oct_periodic_worker);
+                       priv->imode = imode;
+                       priv->port = port;
+                       priv->queue = cvmx_pko_get_base_queue(priv->port);
+                       priv->fau = fau - cvmx_pko_get_num_queues(port) * 4;
+                       priv->phy_mode = PHY_INTERFACE_MODE_NA;
+                       for (qos = 0; qos < 16; qos++)
+                               skb_queue_head_init(&priv->tx_free_list[qos]);
+                       for (qos = 0; qos < cvmx_pko_get_num_queues(port);
+                            qos++)
+                               cvmx_fau_atomic_write32(priv->fau + qos * 4, 0);
+                       dev->min_mtu = VLAN_ETH_ZLEN - mtu_overhead;
+                       dev->max_mtu = OCTEON_MAX_MTU - mtu_overhead;
+
+                       switch (priv->imode) {
+                       /* These types don't support ports to IPD/PKO */
+                       case CVMX_HELPER_INTERFACE_MODE_DISABLED:
+                       case CVMX_HELPER_INTERFACE_MODE_PCIE:
+                       case CVMX_HELPER_INTERFACE_MODE_PICMG:
+                               break;
+
+                       case CVMX_HELPER_INTERFACE_MODE_NPI:
+                               dev->netdev_ops = &cvm_oct_npi_netdev_ops;
+                               strscpy(dev->name, "npi%d", sizeof(dev->name));
+                               break;
+
+                       case CVMX_HELPER_INTERFACE_MODE_XAUI:
+                               dev->netdev_ops = &cvm_oct_xaui_netdev_ops;
+                               strscpy(dev->name, "xaui%d", sizeof(dev->name));
+                               break;
+
+                       case CVMX_HELPER_INTERFACE_MODE_LOOP:
+                               dev->netdev_ops = &cvm_oct_npi_netdev_ops;
+                               strscpy(dev->name, "loop%d", sizeof(dev->name));
+                               break;
+
+                       case CVMX_HELPER_INTERFACE_MODE_SGMII:
+                               priv->phy_mode = PHY_INTERFACE_MODE_SGMII;
+                               dev->netdev_ops = &cvm_oct_sgmii_netdev_ops;
+                               strscpy(dev->name, "eth%d", sizeof(dev->name));
+                               break;
+
+                       case CVMX_HELPER_INTERFACE_MODE_SPI:
+                               dev->netdev_ops = &cvm_oct_spi_netdev_ops;
+                               strscpy(dev->name, "spi%d", sizeof(dev->name));
+                               break;
+
+                       case CVMX_HELPER_INTERFACE_MODE_GMII:
+                               priv->phy_mode = PHY_INTERFACE_MODE_GMII;
+                               dev->netdev_ops = &cvm_oct_rgmii_netdev_ops;
+                               strscpy(dev->name, "eth%d", sizeof(dev->name));
+                               break;
+
+                       case CVMX_HELPER_INTERFACE_MODE_RGMII:
+                               dev->netdev_ops = &cvm_oct_rgmii_netdev_ops;
+                               strscpy(dev->name, "eth%d", sizeof(dev->name));
+                               cvm_set_rgmii_delay(priv, interface,
+                                                   port_index);
+                               break;
+                       }
+
+                       if (!dev->netdev_ops) {
+                               free_netdev(dev);
+                       } else if (register_netdev(dev) < 0) {
+                               pr_err("Failed to register ethernet device for interface %d, port %d\n",
+                                      interface, priv->port);
+                               free_netdev(dev);
+                       } else {
+                               cvm_oct_device[priv->port] = dev;
+                               fau -=
+                                   cvmx_pko_get_num_queues(priv->port) *
+                                   sizeof(u32);
+                               schedule_delayed_work(&priv->port_periodic_work,
+                                                     HZ);
+                       }
+               }
+       }
+
+       cvm_oct_tx_initialize();
+       cvm_oct_rx_initialize();
+
+       /*
+        * 150 uS: about 10 1500-byte packets at 1GE.
+        */
+       cvm_oct_tx_poll_interval = 150 * (octeon_get_clock_rate() / 1000000);
+
+       schedule_delayed_work(&cvm_oct_rx_refill_work, HZ);
+
+       return 0;
+}
+
+static int cvm_oct_remove(struct platform_device *pdev)
+{
+       int port;
+
+       cvmx_ipd_disable();
+
+       atomic_inc_return(&cvm_oct_poll_queue_stopping);
+       cancel_delayed_work_sync(&cvm_oct_rx_refill_work);
+
+       cvm_oct_rx_shutdown();
+       cvm_oct_tx_shutdown();
+
+       cvmx_pko_disable();
+
+       /* Free the ethernet devices */
+       for (port = 0; port < TOTAL_NUMBER_OF_PORTS; port++) {
+               if (cvm_oct_device[port]) {
+                       struct net_device *dev = cvm_oct_device[port];
+                       struct octeon_ethernet *priv = netdev_priv(dev);
+
+                       cancel_delayed_work_sync(&priv->port_periodic_work);
+
+                       cvm_oct_tx_shutdown_dev(dev);
+                       unregister_netdev(dev);
+                       free_netdev(dev);
+                       cvm_oct_device[port] = NULL;
+               }
+       }
+
+       cvmx_pko_shutdown();
+
+       cvmx_ipd_free_ptr();
+
+       /* Free the HW pools */
+       cvm_oct_mem_empty_fpa(CVMX_FPA_PACKET_POOL, CVMX_FPA_PACKET_POOL_SIZE,
+                             num_packet_buffers);
+       cvm_oct_mem_empty_fpa(CVMX_FPA_WQE_POOL, CVMX_FPA_WQE_POOL_SIZE,
+                             num_packet_buffers);
+       if (CVMX_FPA_OUTPUT_BUFFER_POOL != CVMX_FPA_PACKET_POOL)
+               cvm_oct_mem_empty_fpa(CVMX_FPA_OUTPUT_BUFFER_POOL,
+                                     CVMX_FPA_OUTPUT_BUFFER_POOL_SIZE, 128);
+       return 0;
+}
+
+static const struct of_device_id cvm_oct_match[] = {
+       {
+               .compatible = "cavium,octeon-3860-pip",
+       },
+       {},
+};
+MODULE_DEVICE_TABLE(of, cvm_oct_match);
+
+static struct platform_driver cvm_oct_driver = {
+       .probe          = cvm_oct_probe,
+       .remove         = cvm_oct_remove,
+       .driver         = {
+               .name   = KBUILD_MODNAME,
+               .of_match_table = cvm_oct_match,
+       },
+};
+
+module_platform_driver(cvm_oct_driver);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Cavium Networks <support@caviumnetworks.com>");
+MODULE_DESCRIPTION("Cavium Networks Octeon ethernet driver.");
diff --git a/drivers/staging/octeon/octeon-ethernet.h b/drivers/staging/octeon/octeon-ethernet.h
new file mode 100644 (file)
index 0000000..a614070
--- /dev/null
@@ -0,0 +1,107 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * This file is based on code from OCTEON SDK by Cavium Networks.
+ *
+ * Copyright (c) 2003-2010 Cavium Networks
+ */
+
+/*
+ * External interface for the Cavium Octeon ethernet driver.
+ */
+#ifndef OCTEON_ETHERNET_H
+#define OCTEON_ETHERNET_H
+
+#include <linux/of.h>
+#include <linux/phy.h>
+
+#ifdef CONFIG_CAVIUM_OCTEON_SOC
+
+#include <asm/octeon/octeon.h>
+
+#include <asm/octeon/cvmx-asxx-defs.h>
+#include <asm/octeon/cvmx-config.h>
+#include <asm/octeon/cvmx-fau.h>
+#include <asm/octeon/cvmx-gmxx-defs.h>
+#include <asm/octeon/cvmx-helper.h>
+#include <asm/octeon/cvmx-helper-util.h>
+#include <asm/octeon/cvmx-ipd.h>
+#include <asm/octeon/cvmx-ipd-defs.h>
+#include <asm/octeon/cvmx-npi-defs.h>
+#include <asm/octeon/cvmx-pip.h>
+#include <asm/octeon/cvmx-pko.h>
+#include <asm/octeon/cvmx-pow.h>
+#include <asm/octeon/cvmx-scratch.h>
+#include <asm/octeon/cvmx-spi.h>
+#include <asm/octeon/cvmx-spxx-defs.h>
+#include <asm/octeon/cvmx-stxx-defs.h>
+#include <asm/octeon/cvmx-wqe.h>
+
+#else
+
+#include "octeon-stubs.h"
+
+#endif
+
+/**
+ * This is the definition of the Ethernet driver's private
+ * driver state stored in netdev_priv(dev).
+ */
+struct octeon_ethernet {
+       /* PKO hardware output port */
+       int port;
+       /* PKO hardware queue for the port */
+       int queue;
+       /* Hardware fetch and add to count outstanding tx buffers */
+       int fau;
+       /* My netdev. */
+       struct net_device *netdev;
+       /*
+        * Type of port. This is one of the enums in
+        * cvmx_helper_interface_mode_t
+        */
+       int imode;
+       /* PHY mode */
+       phy_interface_t phy_mode;
+       /* List of outstanding tx buffers per queue */
+       struct sk_buff_head tx_free_list[16];
+       unsigned int last_speed;
+       unsigned int last_link;
+       /* Last negotiated link state */
+       u64 link_info;
+       /* Called periodically to check link status */
+       void (*poll)(struct net_device *dev);
+       struct delayed_work     port_periodic_work;
+       struct device_node      *of_node;
+};
+
+int cvm_oct_free_work(void *work_queue_entry);
+
+int cvm_oct_rgmii_open(struct net_device *dev);
+
+int cvm_oct_sgmii_init(struct net_device *dev);
+int cvm_oct_sgmii_open(struct net_device *dev);
+
+int cvm_oct_spi_init(struct net_device *dev);
+void cvm_oct_spi_uninit(struct net_device *dev);
+
+int cvm_oct_common_init(struct net_device *dev);
+void cvm_oct_common_uninit(struct net_device *dev);
+void cvm_oct_adjust_link(struct net_device *dev);
+int cvm_oct_common_stop(struct net_device *dev);
+int cvm_oct_common_open(struct net_device *dev,
+                       void (*link_poll)(struct net_device *));
+void cvm_oct_note_carrier(struct octeon_ethernet *priv,
+                         union cvmx_helper_link_info li);
+void cvm_oct_link_poll(struct net_device *dev);
+
+extern int always_use_pow;
+extern int pow_send_group;
+extern int pow_receive_groups;
+extern char pow_send_list[];
+extern struct net_device *cvm_oct_device[];
+extern atomic_t cvm_oct_poll_queue_stopping;
+extern u64 cvm_oct_tx_poll_interval;
+
+extern int rx_napi_weight;
+
+#endif
diff --git a/drivers/staging/octeon/octeon-stubs.h b/drivers/staging/octeon/octeon-stubs.h
new file mode 100644 (file)
index 0000000..79213c0
--- /dev/null
@@ -0,0 +1,1433 @@
+#define CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE       512
+
+#ifndef XKPHYS_TO_PHYS
+# define XKPHYS_TO_PHYS(p)                     (p)
+#endif
+
+#define OCTEON_IRQ_WORKQ0 0
+#define OCTEON_IRQ_RML 0
+#define OCTEON_IRQ_TIMER1 0
+#define OCTEON_IS_MODEL(x) 0
+#define octeon_has_feature(x)  0
+#define octeon_get_clock_rate()        0
+
+#define CVMX_SYNCIOBDMA                do { } while(0)
+
+#define CVMX_HELPER_INPUT_TAG_TYPE     0
+#define CVMX_HELPER_FIRST_MBUFF_SKIP   7
+#define CVMX_FAU_REG_END               (2048)
+#define CVMX_FPA_OUTPUT_BUFFER_POOL        (2)
+#define CVMX_FPA_OUTPUT_BUFFER_POOL_SIZE    16
+#define CVMX_FPA_PACKET_POOL               (0)
+#define CVMX_FPA_PACKET_POOL_SIZE          16
+#define CVMX_FPA_WQE_POOL                  (1)
+#define CVMX_FPA_WQE_POOL_SIZE             16
+#define CVMX_GMXX_RXX_ADR_CAM_EN(a, b) ((a)+(b))
+#define CVMX_GMXX_RXX_ADR_CTL(a, b)    ((a)+(b))
+#define CVMX_GMXX_PRTX_CFG(a, b)       ((a)+(b))
+#define CVMX_GMXX_RXX_FRM_MAX(a, b)    ((a)+(b))
+#define CVMX_GMXX_RXX_JABBER(a, b)     ((a)+(b))
+#define CVMX_IPD_CTL_STATUS            0
+#define CVMX_PIP_FRM_LEN_CHKX(a)       (a)
+#define CVMX_PIP_NUM_INPUT_PORTS       1
+#define CVMX_SCR_SCRATCH               0
+#define CVMX_PKO_QUEUES_PER_PORT_INTERFACE0    2
+#define CVMX_PKO_QUEUES_PER_PORT_INTERFACE1    2
+#define CVMX_IPD_SUB_PORT_FCS          0
+#define CVMX_SSO_WQ_IQ_DIS             0
+#define CVMX_SSO_WQ_INT                        0
+#define CVMX_POW_WQ_INT                        0
+#define CVMX_SSO_WQ_INT_PC             0
+#define CVMX_NPI_RSL_INT_BLOCKS                0
+#define CVMX_POW_WQ_INT_PC             0
+
+union cvmx_pip_wqe_word2 {
+       uint64_t u64;
+       struct {
+               uint64_t bufs:8;
+               uint64_t ip_offset:8;
+               uint64_t vlan_valid:1;
+               uint64_t vlan_stacked:1;
+               uint64_t unassigned:1;
+               uint64_t vlan_cfi:1;
+               uint64_t vlan_id:12;
+               uint64_t pr:4;
+               uint64_t unassigned2:8;
+               uint64_t dec_ipcomp:1;
+               uint64_t tcp_or_udp:1;
+               uint64_t dec_ipsec:1;
+               uint64_t is_v6:1;
+               uint64_t software:1;
+               uint64_t L4_error:1;
+               uint64_t is_frag:1;
+               uint64_t IP_exc:1;
+               uint64_t is_bcast:1;
+               uint64_t is_mcast:1;
+               uint64_t not_IP:1;
+               uint64_t rcv_error:1;
+               uint64_t err_code:8;
+       } s;
+       struct {
+               uint64_t bufs:8;
+               uint64_t ip_offset:8;
+               uint64_t vlan_valid:1;
+               uint64_t vlan_stacked:1;
+               uint64_t unassigned:1;
+               uint64_t vlan_cfi:1;
+               uint64_t vlan_id:12;
+               uint64_t port:12;
+               uint64_t dec_ipcomp:1;
+               uint64_t tcp_or_udp:1;
+               uint64_t dec_ipsec:1;
+               uint64_t is_v6:1;
+               uint64_t software:1;
+               uint64_t L4_error:1;
+               uint64_t is_frag:1;
+               uint64_t IP_exc:1;
+               uint64_t is_bcast:1;
+               uint64_t is_mcast:1;
+               uint64_t not_IP:1;
+               uint64_t rcv_error:1;
+               uint64_t err_code:8;
+       } s_cn68xx;
+
+       struct {
+               uint64_t unused1:16;
+               uint64_t vlan:16;
+               uint64_t unused2:32;
+       } svlan;
+       struct {
+               uint64_t bufs:8;
+               uint64_t unused:8;
+               uint64_t vlan_valid:1;
+               uint64_t vlan_stacked:1;
+               uint64_t unassigned:1;
+               uint64_t vlan_cfi:1;
+               uint64_t vlan_id:12;
+               uint64_t pr:4;
+               uint64_t unassigned2:12;
+               uint64_t software:1;
+               uint64_t unassigned3:1;
+               uint64_t is_rarp:1;
+               uint64_t is_arp:1;
+               uint64_t is_bcast:1;
+               uint64_t is_mcast:1;
+               uint64_t not_IP:1;
+               uint64_t rcv_error:1;
+               uint64_t err_code:8;
+       } snoip;
+
+};
+
+union cvmx_pip_wqe_word0 {
+       struct {
+               uint64_t next_ptr:40;
+               uint8_t unused;
+               __wsum hw_chksum;
+       } cn38xx;
+       struct {
+               uint64_t pknd:6;        /* 0..5 */
+               uint64_t unused2:2;     /* 6..7 */
+               uint64_t bpid:6;        /* 8..13 */
+               uint64_t unused1:18;    /* 14..31 */
+               uint64_t l2ptr:8;       /* 32..39 */
+               uint64_t l3ptr:8;       /* 40..47 */
+               uint64_t unused0:8;     /* 48..55 */
+               uint64_t l4ptr:8;       /* 56..63 */
+       } cn68xx;
+};
+
+union cvmx_wqe_word0 {
+       uint64_t u64;
+       union cvmx_pip_wqe_word0 pip;
+};
+
+union cvmx_wqe_word1 {
+       uint64_t u64;
+       struct {
+               uint64_t tag:32;
+               uint64_t tag_type:2;
+               uint64_t varies:14;
+               uint64_t len:16;
+       };
+       struct {
+               uint64_t tag:32;
+               uint64_t tag_type:2;
+               uint64_t zero_2:3;
+               uint64_t grp:6;
+               uint64_t zero_1:1;
+               uint64_t qos:3;
+               uint64_t zero_0:1;
+               uint64_t len:16;
+       } cn68xx;
+       struct {
+               uint64_t tag:32;
+               uint64_t tag_type:2;
+               uint64_t zero_2:1;
+               uint64_t grp:4;
+               uint64_t qos:3;
+               uint64_t ipprt:6;
+               uint64_t len:16;
+       } cn38xx;
+};
+
+union cvmx_buf_ptr {
+       void *ptr;
+       uint64_t u64;
+       struct {
+               uint64_t i:1;
+               uint64_t back:4;
+               uint64_t pool:3;
+               uint64_t size:16;
+               uint64_t addr:40;
+       } s;
+};
+
+struct cvmx_wqe {
+       union cvmx_wqe_word0 word0;
+       union cvmx_wqe_word1 word1;
+       union cvmx_pip_wqe_word2 word2;
+       union cvmx_buf_ptr packet_ptr;
+       uint8_t packet_data[96];
+};
+
+union cvmx_helper_link_info {
+       uint64_t u64;
+       struct {
+               uint64_t reserved_20_63:44;
+               uint64_t link_up:1;         /**< Is the physical link up? */
+               uint64_t full_duplex:1;     /**< 1 if the link is full duplex */
+               uint64_t speed:18;          /**< Speed of the link in Mbps */
+       } s;
+};
+
+enum cvmx_fau_reg_32 {
+       CVMX_FAU_REG_32_START   = 0,
+};
+
+enum cvmx_fau_op_size {
+       CVMX_FAU_OP_SIZE_8 = 0,
+       CVMX_FAU_OP_SIZE_16 = 1,
+       CVMX_FAU_OP_SIZE_32 = 2,
+       CVMX_FAU_OP_SIZE_64 = 3
+};
+
+typedef enum {
+       CVMX_SPI_MODE_UNKNOWN = 0,
+       CVMX_SPI_MODE_TX_HALFPLEX = 1,
+       CVMX_SPI_MODE_RX_HALFPLEX = 2,
+       CVMX_SPI_MODE_DUPLEX = 3
+} cvmx_spi_mode_t;
+
+typedef enum {
+       CVMX_HELPER_INTERFACE_MODE_DISABLED,
+       CVMX_HELPER_INTERFACE_MODE_RGMII,
+       CVMX_HELPER_INTERFACE_MODE_GMII,
+       CVMX_HELPER_INTERFACE_MODE_SPI,
+       CVMX_HELPER_INTERFACE_MODE_PCIE,
+       CVMX_HELPER_INTERFACE_MODE_XAUI,
+       CVMX_HELPER_INTERFACE_MODE_SGMII,
+       CVMX_HELPER_INTERFACE_MODE_PICMG,
+       CVMX_HELPER_INTERFACE_MODE_NPI,
+       CVMX_HELPER_INTERFACE_MODE_LOOP,
+} cvmx_helper_interface_mode_t;
+
+typedef enum {
+       CVMX_POW_WAIT = 1,
+       CVMX_POW_NO_WAIT = 0,
+} cvmx_pow_wait_t;
+
+typedef enum {
+       CVMX_PKO_LOCK_NONE = 0,
+       CVMX_PKO_LOCK_ATOMIC_TAG = 1,
+       CVMX_PKO_LOCK_CMD_QUEUE = 2,
+} cvmx_pko_lock_t;
+
+typedef enum {
+       CVMX_PKO_SUCCESS,
+       CVMX_PKO_INVALID_PORT,
+       CVMX_PKO_INVALID_QUEUE,
+       CVMX_PKO_INVALID_PRIORITY,
+       CVMX_PKO_NO_MEMORY,
+       CVMX_PKO_PORT_ALREADY_SETUP,
+       CVMX_PKO_CMD_QUEUE_INIT_ERROR
+} cvmx_pko_status_t;
+
+enum cvmx_pow_tag_type {
+       CVMX_POW_TAG_TYPE_ORDERED   = 0L,
+       CVMX_POW_TAG_TYPE_ATOMIC    = 1L,
+       CVMX_POW_TAG_TYPE_NULL      = 2L,
+       CVMX_POW_TAG_TYPE_NULL_NULL = 3L
+};
+
+union cvmx_ipd_ctl_status {
+       uint64_t u64;
+       struct cvmx_ipd_ctl_status_s {
+               uint64_t reserved_18_63:46;
+               uint64_t use_sop:1;
+               uint64_t rst_done:1;
+               uint64_t clken:1;
+               uint64_t no_wptr:1;
+               uint64_t pq_apkt:1;
+               uint64_t pq_nabuf:1;
+               uint64_t ipd_full:1;
+               uint64_t pkt_off:1;
+               uint64_t len_m8:1;
+               uint64_t reset:1;
+               uint64_t addpkt:1;
+               uint64_t naddbuf:1;
+               uint64_t pkt_lend:1;
+               uint64_t wqe_lend:1;
+               uint64_t pbp_en:1;
+               uint64_t opc_mode:2;
+               uint64_t ipd_en:1;
+       } s;
+       struct cvmx_ipd_ctl_status_cn30xx {
+               uint64_t reserved_10_63:54;
+               uint64_t len_m8:1;
+               uint64_t reset:1;
+               uint64_t addpkt:1;
+               uint64_t naddbuf:1;
+               uint64_t pkt_lend:1;
+               uint64_t wqe_lend:1;
+               uint64_t pbp_en:1;
+               uint64_t opc_mode:2;
+               uint64_t ipd_en:1;
+       } cn30xx;
+       struct cvmx_ipd_ctl_status_cn38xxp2 {
+               uint64_t reserved_9_63:55;
+               uint64_t reset:1;
+               uint64_t addpkt:1;
+               uint64_t naddbuf:1;
+               uint64_t pkt_lend:1;
+               uint64_t wqe_lend:1;
+               uint64_t pbp_en:1;
+               uint64_t opc_mode:2;
+               uint64_t ipd_en:1;
+       } cn38xxp2;
+       struct cvmx_ipd_ctl_status_cn50xx {
+               uint64_t reserved_15_63:49;
+               uint64_t no_wptr:1;
+               uint64_t pq_apkt:1;
+               uint64_t pq_nabuf:1;
+               uint64_t ipd_full:1;
+               uint64_t pkt_off:1;
+               uint64_t len_m8:1;
+               uint64_t reset:1;
+               uint64_t addpkt:1;
+               uint64_t naddbuf:1;
+               uint64_t pkt_lend:1;
+               uint64_t wqe_lend:1;
+               uint64_t pbp_en:1;
+               uint64_t opc_mode:2;
+               uint64_t ipd_en:1;
+       } cn50xx;
+       struct cvmx_ipd_ctl_status_cn58xx {
+               uint64_t reserved_12_63:52;
+               uint64_t ipd_full:1;
+               uint64_t pkt_off:1;
+               uint64_t len_m8:1;
+               uint64_t reset:1;
+               uint64_t addpkt:1;
+               uint64_t naddbuf:1;
+               uint64_t pkt_lend:1;
+               uint64_t wqe_lend:1;
+               uint64_t pbp_en:1;
+               uint64_t opc_mode:2;
+               uint64_t ipd_en:1;
+       } cn58xx;
+       struct cvmx_ipd_ctl_status_cn63xxp1 {
+               uint64_t reserved_16_63:48;
+               uint64_t clken:1;
+               uint64_t no_wptr:1;
+               uint64_t pq_apkt:1;
+               uint64_t pq_nabuf:1;
+               uint64_t ipd_full:1;
+               uint64_t pkt_off:1;
+               uint64_t len_m8:1;
+               uint64_t reset:1;
+               uint64_t addpkt:1;
+               uint64_t naddbuf:1;
+               uint64_t pkt_lend:1;
+               uint64_t wqe_lend:1;
+               uint64_t pbp_en:1;
+               uint64_t opc_mode:2;
+               uint64_t ipd_en:1;
+       } cn63xxp1;
+};
+
+union cvmx_ipd_sub_port_fcs {
+       uint64_t u64;
+       struct cvmx_ipd_sub_port_fcs_s {
+               uint64_t port_bit:32;
+               uint64_t reserved_32_35:4;
+               uint64_t port_bit2:4;
+               uint64_t reserved_40_63:24;
+       } s;
+       struct cvmx_ipd_sub_port_fcs_cn30xx {
+               uint64_t port_bit:3;
+               uint64_t reserved_3_63:61;
+       } cn30xx;
+       struct cvmx_ipd_sub_port_fcs_cn38xx {
+               uint64_t port_bit:32;
+               uint64_t reserved_32_63:32;
+       } cn38xx;
+};
+
+union cvmx_ipd_sub_port_qos_cnt {
+       uint64_t u64;
+       struct cvmx_ipd_sub_port_qos_cnt_s {
+               uint64_t cnt:32;
+               uint64_t port_qos:9;
+               uint64_t reserved_41_63:23;
+       } s;
+};
+typedef struct {
+       uint32_t dropped_octets;
+       uint32_t dropped_packets;
+       uint32_t pci_raw_packets;
+       uint32_t octets;
+       uint32_t packets;
+       uint32_t multicast_packets;
+       uint32_t broadcast_packets;
+       uint32_t len_64_packets;
+       uint32_t len_65_127_packets;
+       uint32_t len_128_255_packets;
+       uint32_t len_256_511_packets;
+       uint32_t len_512_1023_packets;
+       uint32_t len_1024_1518_packets;
+       uint32_t len_1519_max_packets;
+       uint32_t fcs_align_err_packets;
+       uint32_t runt_packets;
+       uint32_t runt_crc_packets;
+       uint32_t oversize_packets;
+       uint32_t oversize_crc_packets;
+       uint32_t inb_packets;
+       uint64_t inb_octets;
+       uint16_t inb_errors;
+} cvmx_pip_port_status_t;
+
+typedef struct {
+       uint32_t packets;
+       uint64_t octets;
+       uint64_t doorbell;
+} cvmx_pko_port_status_t;
+
+union cvmx_pip_frm_len_chkx {
+       uint64_t u64;
+       struct cvmx_pip_frm_len_chkx_s {
+               uint64_t reserved_32_63:32;
+               uint64_t maxlen:16;
+               uint64_t minlen:16;
+       } s;
+};
+
+union cvmx_gmxx_rxx_frm_ctl {
+       uint64_t u64;
+       struct cvmx_gmxx_rxx_frm_ctl_s {
+               uint64_t pre_chk:1;
+               uint64_t pre_strp:1;
+               uint64_t ctl_drp:1;
+               uint64_t ctl_bck:1;
+               uint64_t ctl_mcst:1;
+               uint64_t ctl_smac:1;
+               uint64_t pre_free:1;
+               uint64_t vlan_len:1;
+               uint64_t pad_len:1;
+               uint64_t pre_align:1;
+               uint64_t null_dis:1;
+               uint64_t reserved_11_11:1;
+               uint64_t ptp_mode:1;
+               uint64_t reserved_13_63:51;
+       } s;
+       struct cvmx_gmxx_rxx_frm_ctl_cn30xx {
+               uint64_t pre_chk:1;
+               uint64_t pre_strp:1;
+               uint64_t ctl_drp:1;
+               uint64_t ctl_bck:1;
+               uint64_t ctl_mcst:1;
+               uint64_t ctl_smac:1;
+               uint64_t pre_free:1;
+               uint64_t vlan_len:1;
+               uint64_t pad_len:1;
+               uint64_t reserved_9_63:55;
+       } cn30xx;
+       struct cvmx_gmxx_rxx_frm_ctl_cn31xx {
+               uint64_t pre_chk:1;
+               uint64_t pre_strp:1;
+               uint64_t ctl_drp:1;
+               uint64_t ctl_bck:1;
+               uint64_t ctl_mcst:1;
+               uint64_t ctl_smac:1;
+               uint64_t pre_free:1;
+               uint64_t vlan_len:1;
+               uint64_t reserved_8_63:56;
+       } cn31xx;
+       struct cvmx_gmxx_rxx_frm_ctl_cn50xx {
+               uint64_t pre_chk:1;
+               uint64_t pre_strp:1;
+               uint64_t ctl_drp:1;
+               uint64_t ctl_bck:1;
+               uint64_t ctl_mcst:1;
+               uint64_t ctl_smac:1;
+               uint64_t pre_free:1;
+               uint64_t reserved_7_8:2;
+               uint64_t pre_align:1;
+               uint64_t null_dis:1;
+               uint64_t reserved_11_63:53;
+       } cn50xx;
+       struct cvmx_gmxx_rxx_frm_ctl_cn56xxp1 {
+               uint64_t pre_chk:1;
+               uint64_t pre_strp:1;
+               uint64_t ctl_drp:1;
+               uint64_t ctl_bck:1;
+               uint64_t ctl_mcst:1;
+               uint64_t ctl_smac:1;
+               uint64_t pre_free:1;
+               uint64_t reserved_7_8:2;
+               uint64_t pre_align:1;
+               uint64_t reserved_10_63:54;
+       } cn56xxp1;
+       struct cvmx_gmxx_rxx_frm_ctl_cn58xx {
+               uint64_t pre_chk:1;
+               uint64_t pre_strp:1;
+               uint64_t ctl_drp:1;
+               uint64_t ctl_bck:1;
+               uint64_t ctl_mcst:1;
+               uint64_t ctl_smac:1;
+               uint64_t pre_free:1;
+               uint64_t vlan_len:1;
+               uint64_t pad_len:1;
+               uint64_t pre_align:1;
+               uint64_t null_dis:1;
+               uint64_t reserved_11_63:53;
+       } cn58xx;
+       struct cvmx_gmxx_rxx_frm_ctl_cn61xx {
+               uint64_t pre_chk:1;
+               uint64_t pre_strp:1;
+               uint64_t ctl_drp:1;
+               uint64_t ctl_bck:1;
+               uint64_t ctl_mcst:1;
+               uint64_t ctl_smac:1;
+               uint64_t pre_free:1;
+               uint64_t reserved_7_8:2;
+               uint64_t pre_align:1;
+               uint64_t null_dis:1;
+               uint64_t reserved_11_11:1;
+               uint64_t ptp_mode:1;
+               uint64_t reserved_13_63:51;
+       } cn61xx;
+};
+
+union cvmx_gmxx_rxx_int_reg {
+       uint64_t u64;
+       struct cvmx_gmxx_rxx_int_reg_s {
+               uint64_t minerr:1;
+               uint64_t carext:1;
+               uint64_t maxerr:1;
+               uint64_t jabber:1;
+               uint64_t fcserr:1;
+               uint64_t alnerr:1;
+               uint64_t lenerr:1;
+               uint64_t rcverr:1;
+               uint64_t skperr:1;
+               uint64_t niberr:1;
+               uint64_t ovrerr:1;
+               uint64_t pcterr:1;
+               uint64_t rsverr:1;
+               uint64_t falerr:1;
+               uint64_t coldet:1;
+               uint64_t ifgerr:1;
+               uint64_t phy_link:1;
+               uint64_t phy_spd:1;
+               uint64_t phy_dupx:1;
+               uint64_t pause_drp:1;
+               uint64_t loc_fault:1;
+               uint64_t rem_fault:1;
+               uint64_t bad_seq:1;
+               uint64_t bad_term:1;
+               uint64_t unsop:1;
+               uint64_t uneop:1;
+               uint64_t undat:1;
+               uint64_t hg2fld:1;
+               uint64_t hg2cc:1;
+               uint64_t reserved_29_63:35;
+       } s;
+       struct cvmx_gmxx_rxx_int_reg_cn30xx {
+               uint64_t minerr:1;
+               uint64_t carext:1;
+               uint64_t maxerr:1;
+               uint64_t jabber:1;
+               uint64_t fcserr:1;
+               uint64_t alnerr:1;
+               uint64_t lenerr:1;
+               uint64_t rcverr:1;
+               uint64_t skperr:1;
+               uint64_t niberr:1;
+               uint64_t ovrerr:1;
+               uint64_t pcterr:1;
+               uint64_t rsverr:1;
+               uint64_t falerr:1;
+               uint64_t coldet:1;
+               uint64_t ifgerr:1;
+               uint64_t phy_link:1;
+               uint64_t phy_spd:1;
+               uint64_t phy_dupx:1;
+               uint64_t reserved_19_63:45;
+       } cn30xx;
+       struct cvmx_gmxx_rxx_int_reg_cn50xx {
+               uint64_t reserved_0_0:1;
+               uint64_t carext:1;
+               uint64_t reserved_2_2:1;
+               uint64_t jabber:1;
+               uint64_t fcserr:1;
+               uint64_t alnerr:1;
+               uint64_t reserved_6_6:1;
+               uint64_t rcverr:1;
+               uint64_t skperr:1;
+               uint64_t niberr:1;
+               uint64_t ovrerr:1;
+               uint64_t pcterr:1;
+               uint64_t rsverr:1;
+               uint64_t falerr:1;
+               uint64_t coldet:1;
+               uint64_t ifgerr:1;
+               uint64_t phy_link:1;
+               uint64_t phy_spd:1;
+               uint64_t phy_dupx:1;
+               uint64_t pause_drp:1;
+               uint64_t reserved_20_63:44;
+       } cn50xx;
+       struct cvmx_gmxx_rxx_int_reg_cn52xx {
+               uint64_t reserved_0_0:1;
+               uint64_t carext:1;
+               uint64_t reserved_2_2:1;
+               uint64_t jabber:1;
+               uint64_t fcserr:1;
+               uint64_t reserved_5_6:2;
+               uint64_t rcverr:1;
+               uint64_t skperr:1;
+               uint64_t reserved_9_9:1;
+               uint64_t ovrerr:1;
+               uint64_t pcterr:1;
+               uint64_t rsverr:1;
+               uint64_t falerr:1;
+               uint64_t coldet:1;
+               uint64_t ifgerr:1;
+               uint64_t reserved_16_18:3;
+               uint64_t pause_drp:1;
+               uint64_t loc_fault:1;
+               uint64_t rem_fault:1;
+               uint64_t bad_seq:1;
+               uint64_t bad_term:1;
+               uint64_t unsop:1;
+               uint64_t uneop:1;
+               uint64_t undat:1;
+               uint64_t hg2fld:1;
+               uint64_t hg2cc:1;
+               uint64_t reserved_29_63:35;
+       } cn52xx;
+       struct cvmx_gmxx_rxx_int_reg_cn56xxp1 {
+               uint64_t reserved_0_0:1;
+               uint64_t carext:1;
+               uint64_t reserved_2_2:1;
+               uint64_t jabber:1;
+               uint64_t fcserr:1;
+               uint64_t reserved_5_6:2;
+               uint64_t rcverr:1;
+               uint64_t skperr:1;
+               uint64_t reserved_9_9:1;
+               uint64_t ovrerr:1;
+               uint64_t pcterr:1;
+               uint64_t rsverr:1;
+               uint64_t falerr:1;
+               uint64_t coldet:1;
+               uint64_t ifgerr:1;
+               uint64_t reserved_16_18:3;
+               uint64_t pause_drp:1;
+               uint64_t loc_fault:1;
+               uint64_t rem_fault:1;
+               uint64_t bad_seq:1;
+               uint64_t bad_term:1;
+               uint64_t unsop:1;
+               uint64_t uneop:1;
+               uint64_t undat:1;
+               uint64_t reserved_27_63:37;
+       } cn56xxp1;
+       struct cvmx_gmxx_rxx_int_reg_cn58xx {
+               uint64_t minerr:1;
+               uint64_t carext:1;
+               uint64_t maxerr:1;
+               uint64_t jabber:1;
+               uint64_t fcserr:1;
+               uint64_t alnerr:1;
+               uint64_t lenerr:1;
+               uint64_t rcverr:1;
+               uint64_t skperr:1;
+               uint64_t niberr:1;
+               uint64_t ovrerr:1;
+               uint64_t pcterr:1;
+               uint64_t rsverr:1;
+               uint64_t falerr:1;
+               uint64_t coldet:1;
+               uint64_t ifgerr:1;
+               uint64_t phy_link:1;
+               uint64_t phy_spd:1;
+               uint64_t phy_dupx:1;
+               uint64_t pause_drp:1;
+               uint64_t reserved_20_63:44;
+       } cn58xx;
+       struct cvmx_gmxx_rxx_int_reg_cn61xx {
+               uint64_t minerr:1;
+               uint64_t carext:1;
+               uint64_t reserved_2_2:1;
+               uint64_t jabber:1;
+               uint64_t fcserr:1;
+               uint64_t reserved_5_6:2;
+               uint64_t rcverr:1;
+               uint64_t skperr:1;
+               uint64_t reserved_9_9:1;
+               uint64_t ovrerr:1;
+               uint64_t pcterr:1;
+               uint64_t rsverr:1;
+               uint64_t falerr:1;
+               uint64_t coldet:1;
+               uint64_t ifgerr:1;
+               uint64_t reserved_16_18:3;
+               uint64_t pause_drp:1;
+               uint64_t loc_fault:1;
+               uint64_t rem_fault:1;
+               uint64_t bad_seq:1;
+               uint64_t bad_term:1;
+               uint64_t unsop:1;
+               uint64_t uneop:1;
+               uint64_t undat:1;
+               uint64_t hg2fld:1;
+               uint64_t hg2cc:1;
+               uint64_t reserved_29_63:35;
+       } cn61xx;
+};
+
+union cvmx_gmxx_prtx_cfg {
+       uint64_t u64;
+       struct cvmx_gmxx_prtx_cfg_s {
+               uint64_t reserved_22_63:42;
+               uint64_t pknd:6;
+               uint64_t reserved_14_15:2;
+               uint64_t tx_idle:1;
+               uint64_t rx_idle:1;
+               uint64_t reserved_9_11:3;
+               uint64_t speed_msb:1;
+               uint64_t reserved_4_7:4;
+               uint64_t slottime:1;
+               uint64_t duplex:1;
+               uint64_t speed:1;
+               uint64_t en:1;
+       } s;
+       struct cvmx_gmxx_prtx_cfg_cn30xx {
+               uint64_t reserved_4_63:60;
+               uint64_t slottime:1;
+               uint64_t duplex:1;
+               uint64_t speed:1;
+               uint64_t en:1;
+       } cn30xx;
+       struct cvmx_gmxx_prtx_cfg_cn52xx {
+               uint64_t reserved_14_63:50;
+               uint64_t tx_idle:1;
+               uint64_t rx_idle:1;
+               uint64_t reserved_9_11:3;
+               uint64_t speed_msb:1;
+               uint64_t reserved_4_7:4;
+               uint64_t slottime:1;
+               uint64_t duplex:1;
+               uint64_t speed:1;
+               uint64_t en:1;
+       } cn52xx;
+};
+
+union cvmx_gmxx_rxx_adr_ctl {
+       uint64_t u64;
+       struct cvmx_gmxx_rxx_adr_ctl_s {
+               uint64_t reserved_4_63:60;
+               uint64_t cam_mode:1;
+               uint64_t mcst:2;
+               uint64_t bcst:1;
+       } s;
+};
+
+union cvmx_pip_prt_tagx {
+       uint64_t u64;
+       struct cvmx_pip_prt_tagx_s {
+               uint64_t reserved_54_63:10;
+               uint64_t portadd_en:1;
+               uint64_t inc_hwchk:1;
+               uint64_t reserved_50_51:2;
+               uint64_t grptagbase_msb:2;
+               uint64_t reserved_46_47:2;
+               uint64_t grptagmask_msb:2;
+               uint64_t reserved_42_43:2;
+               uint64_t grp_msb:2;
+               uint64_t grptagbase:4;
+               uint64_t grptagmask:4;
+               uint64_t grptag:1;
+               uint64_t grptag_mskip:1;
+               uint64_t tag_mode:2;
+               uint64_t inc_vs:2;
+               uint64_t inc_vlan:1;
+               uint64_t inc_prt_flag:1;
+               uint64_t ip6_dprt_flag:1;
+               uint64_t ip4_dprt_flag:1;
+               uint64_t ip6_sprt_flag:1;
+               uint64_t ip4_sprt_flag:1;
+               uint64_t ip6_nxth_flag:1;
+               uint64_t ip4_pctl_flag:1;
+               uint64_t ip6_dst_flag:1;
+               uint64_t ip4_dst_flag:1;
+               uint64_t ip6_src_flag:1;
+               uint64_t ip4_src_flag:1;
+               uint64_t tcp6_tag_type:2;
+               uint64_t tcp4_tag_type:2;
+               uint64_t ip6_tag_type:2;
+               uint64_t ip4_tag_type:2;
+               uint64_t non_tag_type:2;
+               uint64_t grp:4;
+       } s;
+       struct cvmx_pip_prt_tagx_cn30xx {
+               uint64_t reserved_40_63:24;
+               uint64_t grptagbase:4;
+               uint64_t grptagmask:4;
+               uint64_t grptag:1;
+               uint64_t reserved_30_30:1;
+               uint64_t tag_mode:2;
+               uint64_t inc_vs:2;
+               uint64_t inc_vlan:1;
+               uint64_t inc_prt_flag:1;
+               uint64_t ip6_dprt_flag:1;
+               uint64_t ip4_dprt_flag:1;
+               uint64_t ip6_sprt_flag:1;
+               uint64_t ip4_sprt_flag:1;
+               uint64_t ip6_nxth_flag:1;
+               uint64_t ip4_pctl_flag:1;
+               uint64_t ip6_dst_flag:1;
+               uint64_t ip4_dst_flag:1;
+               uint64_t ip6_src_flag:1;
+               uint64_t ip4_src_flag:1;
+               uint64_t tcp6_tag_type:2;
+               uint64_t tcp4_tag_type:2;
+               uint64_t ip6_tag_type:2;
+               uint64_t ip4_tag_type:2;
+               uint64_t non_tag_type:2;
+               uint64_t grp:4;
+       } cn30xx;
+       struct cvmx_pip_prt_tagx_cn50xx {
+               uint64_t reserved_40_63:24;
+               uint64_t grptagbase:4;
+               uint64_t grptagmask:4;
+               uint64_t grptag:1;
+               uint64_t grptag_mskip:1;
+               uint64_t tag_mode:2;
+               uint64_t inc_vs:2;
+               uint64_t inc_vlan:1;
+               uint64_t inc_prt_flag:1;
+               uint64_t ip6_dprt_flag:1;
+               uint64_t ip4_dprt_flag:1;
+               uint64_t ip6_sprt_flag:1;
+               uint64_t ip4_sprt_flag:1;
+               uint64_t ip6_nxth_flag:1;
+               uint64_t ip4_pctl_flag:1;
+               uint64_t ip6_dst_flag:1;
+               uint64_t ip4_dst_flag:1;
+               uint64_t ip6_src_flag:1;
+               uint64_t ip4_src_flag:1;
+               uint64_t tcp6_tag_type:2;
+               uint64_t tcp4_tag_type:2;
+               uint64_t ip6_tag_type:2;
+               uint64_t ip4_tag_type:2;
+               uint64_t non_tag_type:2;
+               uint64_t grp:4;
+       } cn50xx;
+};
+
+union cvmx_spxx_int_reg {
+       uint64_t u64;
+       struct cvmx_spxx_int_reg_s {
+               uint64_t reserved_32_63:32;
+               uint64_t spf:1;
+               uint64_t reserved_12_30:19;
+               uint64_t calerr:1;
+               uint64_t syncerr:1;
+               uint64_t diperr:1;
+               uint64_t tpaovr:1;
+               uint64_t rsverr:1;
+               uint64_t drwnng:1;
+               uint64_t clserr:1;
+               uint64_t spiovr:1;
+               uint64_t reserved_2_3:2;
+               uint64_t abnorm:1;
+               uint64_t prtnxa:1;
+       } s;
+};
+
+union cvmx_spxx_int_msk {
+       uint64_t u64;
+       struct cvmx_spxx_int_msk_s {
+               uint64_t reserved_12_63:52;
+               uint64_t calerr:1;
+               uint64_t syncerr:1;
+               uint64_t diperr:1;
+               uint64_t tpaovr:1;
+               uint64_t rsverr:1;
+               uint64_t drwnng:1;
+               uint64_t clserr:1;
+               uint64_t spiovr:1;
+               uint64_t reserved_2_3:2;
+               uint64_t abnorm:1;
+               uint64_t prtnxa:1;
+       } s;
+};
+
+union cvmx_pow_wq_int {
+       uint64_t u64;
+       struct cvmx_pow_wq_int_s {
+               uint64_t wq_int:16;
+               uint64_t iq_dis:16;
+               uint64_t reserved_32_63:32;
+       } s;
+};
+
+union cvmx_sso_wq_int_thrx {
+       uint64_t u64;
+       struct {
+               uint64_t iq_thr:12;
+               uint64_t reserved_12_13:2;
+               uint64_t ds_thr:12;
+               uint64_t reserved_26_27:2;
+               uint64_t tc_thr:4;
+               uint64_t tc_en:1;
+               uint64_t reserved_33_63:31;
+       } s;
+};
+
+union cvmx_stxx_int_reg {
+       uint64_t u64;
+       struct cvmx_stxx_int_reg_s {
+               uint64_t reserved_9_63:55;
+               uint64_t syncerr:1;
+               uint64_t frmerr:1;
+               uint64_t unxfrm:1;
+               uint64_t nosync:1;
+               uint64_t diperr:1;
+               uint64_t datovr:1;
+               uint64_t ovrbst:1;
+               uint64_t calpar1:1;
+               uint64_t calpar0:1;
+       } s;
+};
+
+union cvmx_stxx_int_msk {
+       uint64_t u64;
+       struct cvmx_stxx_int_msk_s {
+               uint64_t reserved_8_63:56;
+               uint64_t frmerr:1;
+               uint64_t unxfrm:1;
+               uint64_t nosync:1;
+               uint64_t diperr:1;
+               uint64_t datovr:1;
+               uint64_t ovrbst:1;
+               uint64_t calpar1:1;
+               uint64_t calpar0:1;
+       } s;
+};
+
+union cvmx_pow_wq_int_pc {
+       uint64_t u64;
+       struct cvmx_pow_wq_int_pc_s {
+               uint64_t reserved_0_7:8;
+               uint64_t pc_thr:20;
+               uint64_t reserved_28_31:4;
+               uint64_t pc:28;
+               uint64_t reserved_60_63:4;
+       } s;
+};
+
+union cvmx_pow_wq_int_thrx {
+       uint64_t u64;
+       struct cvmx_pow_wq_int_thrx_s {
+               uint64_t reserved_29_63:35;
+               uint64_t tc_en:1;
+               uint64_t tc_thr:4;
+               uint64_t reserved_23_23:1;
+               uint64_t ds_thr:11;
+               uint64_t reserved_11_11:1;
+               uint64_t iq_thr:11;
+       } s;
+       struct cvmx_pow_wq_int_thrx_cn30xx {
+               uint64_t reserved_29_63:35;
+               uint64_t tc_en:1;
+               uint64_t tc_thr:4;
+               uint64_t reserved_18_23:6;
+               uint64_t ds_thr:6;
+               uint64_t reserved_6_11:6;
+               uint64_t iq_thr:6;
+       } cn30xx;
+       struct cvmx_pow_wq_int_thrx_cn31xx {
+               uint64_t reserved_29_63:35;
+               uint64_t tc_en:1;
+               uint64_t tc_thr:4;
+               uint64_t reserved_20_23:4;
+               uint64_t ds_thr:8;
+               uint64_t reserved_8_11:4;
+               uint64_t iq_thr:8;
+       } cn31xx;
+       struct cvmx_pow_wq_int_thrx_cn52xx {
+               uint64_t reserved_29_63:35;
+               uint64_t tc_en:1;
+               uint64_t tc_thr:4;
+               uint64_t reserved_21_23:3;
+               uint64_t ds_thr:9;
+               uint64_t reserved_9_11:3;
+               uint64_t iq_thr:9;
+       } cn52xx;
+       struct cvmx_pow_wq_int_thrx_cn63xx {
+               uint64_t reserved_29_63:35;
+               uint64_t tc_en:1;
+               uint64_t tc_thr:4;
+               uint64_t reserved_22_23:2;
+               uint64_t ds_thr:10;
+               uint64_t reserved_10_11:2;
+               uint64_t iq_thr:10;
+       } cn63xx;
+};
+
+union cvmx_npi_rsl_int_blocks {
+       uint64_t u64;
+       struct cvmx_npi_rsl_int_blocks_s {
+               uint64_t reserved_32_63:32;
+               uint64_t rint_31:1;
+               uint64_t iob:1;
+               uint64_t reserved_28_29:2;
+               uint64_t rint_27:1;
+               uint64_t rint_26:1;
+               uint64_t rint_25:1;
+               uint64_t rint_24:1;
+               uint64_t asx1:1;
+               uint64_t asx0:1;
+               uint64_t rint_21:1;
+               uint64_t pip:1;
+               uint64_t spx1:1;
+               uint64_t spx0:1;
+               uint64_t lmc:1;
+               uint64_t l2c:1;
+               uint64_t rint_15:1;
+               uint64_t reserved_13_14:2;
+               uint64_t pow:1;
+               uint64_t tim:1;
+               uint64_t pko:1;
+               uint64_t ipd:1;
+               uint64_t rint_8:1;
+               uint64_t zip:1;
+               uint64_t dfa:1;
+               uint64_t fpa:1;
+               uint64_t key:1;
+               uint64_t npi:1;
+               uint64_t gmx1:1;
+               uint64_t gmx0:1;
+               uint64_t mio:1;
+       } s;
+       struct cvmx_npi_rsl_int_blocks_cn30xx {
+               uint64_t reserved_32_63:32;
+               uint64_t rint_31:1;
+               uint64_t iob:1;
+               uint64_t rint_29:1;
+               uint64_t rint_28:1;
+               uint64_t rint_27:1;
+               uint64_t rint_26:1;
+               uint64_t rint_25:1;
+               uint64_t rint_24:1;
+               uint64_t asx1:1;
+               uint64_t asx0:1;
+               uint64_t rint_21:1;
+               uint64_t pip:1;
+               uint64_t spx1:1;
+               uint64_t spx0:1;
+               uint64_t lmc:1;
+               uint64_t l2c:1;
+               uint64_t rint_15:1;
+               uint64_t rint_14:1;
+               uint64_t usb:1;
+               uint64_t pow:1;
+               uint64_t tim:1;
+               uint64_t pko:1;
+               uint64_t ipd:1;
+               uint64_t rint_8:1;
+               uint64_t zip:1;
+               uint64_t dfa:1;
+               uint64_t fpa:1;
+               uint64_t key:1;
+               uint64_t npi:1;
+               uint64_t gmx1:1;
+               uint64_t gmx0:1;
+               uint64_t mio:1;
+       } cn30xx;
+       struct cvmx_npi_rsl_int_blocks_cn38xx {
+               uint64_t reserved_32_63:32;
+               uint64_t rint_31:1;
+               uint64_t iob:1;
+               uint64_t rint_29:1;
+               uint64_t rint_28:1;
+               uint64_t rint_27:1;
+               uint64_t rint_26:1;
+               uint64_t rint_25:1;
+               uint64_t rint_24:1;
+               uint64_t asx1:1;
+               uint64_t asx0:1;
+               uint64_t rint_21:1;
+               uint64_t pip:1;
+               uint64_t spx1:1;
+               uint64_t spx0:1;
+               uint64_t lmc:1;
+               uint64_t l2c:1;
+               uint64_t rint_15:1;
+               uint64_t rint_14:1;
+               uint64_t rint_13:1;
+               uint64_t pow:1;
+               uint64_t tim:1;
+               uint64_t pko:1;
+               uint64_t ipd:1;
+               uint64_t rint_8:1;
+               uint64_t zip:1;
+               uint64_t dfa:1;
+               uint64_t fpa:1;
+               uint64_t key:1;
+               uint64_t npi:1;
+               uint64_t gmx1:1;
+               uint64_t gmx0:1;
+               uint64_t mio:1;
+       } cn38xx;
+       struct cvmx_npi_rsl_int_blocks_cn50xx {
+               uint64_t reserved_31_63:33;
+               uint64_t iob:1;
+               uint64_t lmc1:1;
+               uint64_t agl:1;
+               uint64_t reserved_24_27:4;
+               uint64_t asx1:1;
+               uint64_t asx0:1;
+               uint64_t reserved_21_21:1;
+               uint64_t pip:1;
+               uint64_t spx1:1;
+               uint64_t spx0:1;
+               uint64_t lmc:1;
+               uint64_t l2c:1;
+               uint64_t reserved_15_15:1;
+               uint64_t rad:1;
+               uint64_t usb:1;
+               uint64_t pow:1;
+               uint64_t tim:1;
+               uint64_t pko:1;
+               uint64_t ipd:1;
+               uint64_t reserved_8_8:1;
+               uint64_t zip:1;
+               uint64_t dfa:1;
+               uint64_t fpa:1;
+               uint64_t key:1;
+               uint64_t npi:1;
+               uint64_t gmx1:1;
+               uint64_t gmx0:1;
+               uint64_t mio:1;
+       } cn50xx;
+};
+
+union cvmx_pko_command_word0 {
+       uint64_t u64;
+       struct {
+               uint64_t total_bytes:16;
+               uint64_t segs:6;
+               uint64_t dontfree:1;
+               uint64_t ignore_i:1;
+               uint64_t ipoffp1:7;
+               uint64_t gather:1;
+               uint64_t rsp:1;
+               uint64_t wqp:1;
+               uint64_t n2:1;
+               uint64_t le:1;
+               uint64_t reg0:11;
+               uint64_t subone0:1;
+               uint64_t reg1:11;
+               uint64_t subone1:1;
+               uint64_t size0:2;
+               uint64_t size1:2;
+       } s;
+};
+
+union cvmx_ciu_timx {
+       uint64_t u64;
+       struct cvmx_ciu_timx_s {
+               uint64_t reserved_37_63:27;
+               uint64_t one_shot:1;
+               uint64_t len:36;
+       } s;
+};
+
+union cvmx_gmxx_rxx_rx_inbnd {
+       uint64_t u64;
+       struct cvmx_gmxx_rxx_rx_inbnd_s {
+               uint64_t status:1;
+               uint64_t speed:2;
+               uint64_t duplex:1;
+               uint64_t reserved_4_63:60;
+       } s;
+};
+
+static inline int32_t cvmx_fau_fetch_and_add32(enum cvmx_fau_reg_32 reg,
+                                              int32_t value)
+{
+       return value;
+}
+
+static inline void cvmx_fau_atomic_add32(enum cvmx_fau_reg_32 reg,
+                                        int32_t value)
+{ }
+
+static inline void cvmx_fau_atomic_write32(enum cvmx_fau_reg_32 reg,
+                                          int32_t value)
+{ }
+
+static inline uint64_t cvmx_scratch_read64(uint64_t address)
+{
+       return 0;
+}
+
+static inline void cvmx_scratch_write64(uint64_t address, uint64_t value)
+{ }
+
+static inline int cvmx_wqe_get_grp(struct cvmx_wqe *work)
+{
+       return 0;
+}
+
+static inline void *cvmx_phys_to_ptr(uint64_t physical_address)
+{
+       return (void *)(uintptr_t)(physical_address);
+}
+
+static inline uint64_t cvmx_ptr_to_phys(void *ptr)
+{
+       return (unsigned long)ptr;
+}
+
+static inline int cvmx_helper_get_interface_num(int ipd_port)
+{
+       return ipd_port;
+}
+
+static inline int cvmx_helper_get_interface_index_num(int ipd_port)
+{
+       return ipd_port;
+}
+
+static inline void cvmx_fpa_enable(void)
+{ }
+
+static inline uint64_t cvmx_read_csr(uint64_t csr_addr)
+{
+       return 0;
+}
+
+static inline void cvmx_write_csr(uint64_t csr_addr, uint64_t val)
+{ }
+
+static inline int cvmx_helper_setup_red(int pass_thresh, int drop_thresh)
+{
+       return 0;
+}
+
+static inline void *cvmx_fpa_alloc(uint64_t pool)
+{
+       return NULL;
+}
+
+static inline void cvmx_fpa_free(void *ptr, uint64_t pool,
+                                uint64_t num_cache_lines)
+{ }
+
+static inline int octeon_is_simulation(void)
+{
+       return 1;
+}
+
+static inline void cvmx_pip_get_port_status(uint64_t port_num, uint64_t clear,
+                                           cvmx_pip_port_status_t *status)
+{ }
+
+static inline void cvmx_pko_get_port_status(uint64_t port_num, uint64_t clear,
+                                           cvmx_pko_port_status_t *status)
+{ }
+
+static inline cvmx_helper_interface_mode_t cvmx_helper_interface_get_mode(int
+                                                                  interface)
+{
+       return 0;
+}
+
+static inline union cvmx_helper_link_info cvmx_helper_link_get(int ipd_port)
+{
+       union cvmx_helper_link_info ret = { .u64 = 0 };
+
+       return ret;
+}
+
+static inline int cvmx_helper_link_set(int ipd_port,
+                                      union cvmx_helper_link_info link_info)
+{
+       return 0;
+}
+
+static inline int cvmx_helper_initialize_packet_io_global(void)
+{
+       return 0;
+}
+
+static inline int cvmx_helper_get_number_of_interfaces(void)
+{
+       return 2;
+}
+
+static inline int cvmx_helper_ports_on_interface(int interface)
+{
+       return 1;
+}
+
+static inline int cvmx_helper_get_ipd_port(int interface, int port)
+{
+       return 0;
+}
+
+static inline int cvmx_helper_ipd_and_packet_input_enable(void)
+{
+       return 0;
+}
+
+static inline void cvmx_ipd_disable(void)
+{ }
+
+static inline void cvmx_ipd_free_ptr(void)
+{ }
+
+static inline void cvmx_pko_disable(void)
+{ }
+
+static inline void cvmx_pko_shutdown(void)
+{ }
+
+static inline int cvmx_pko_get_base_queue_per_core(int port, int core)
+{
+       return port;
+}
+
+static inline int cvmx_pko_get_base_queue(int port)
+{
+       return port;
+}
+
+static inline int cvmx_pko_get_num_queues(int port)
+{
+       return port;
+}
+
+static inline unsigned int cvmx_get_core_num(void)
+{
+       return 0;
+}
+
+static inline void cvmx_pow_work_request_async_nocheck(int scr_addr,
+                                                      cvmx_pow_wait_t wait)
+{ }
+
+static inline void cvmx_pow_work_request_async(int scr_addr,
+                                                      cvmx_pow_wait_t wait)
+{ }
+
+static inline struct cvmx_wqe *cvmx_pow_work_response_async(int scr_addr)
+{
+       struct cvmx_wqe *wqe = (void *)(unsigned long)scr_addr;
+
+       return wqe;
+}
+
+static inline struct cvmx_wqe *cvmx_pow_work_request_sync(cvmx_pow_wait_t wait)
+{
+       return (void *)(unsigned long)wait;
+}
+
+static inline int cvmx_spi_restart_interface(int interface,
+                                       cvmx_spi_mode_t mode, int timeout)
+{
+       return 0;
+}
+
+static inline void cvmx_fau_async_fetch_and_add32(uint64_t scraddr,
+                                                 enum cvmx_fau_reg_32 reg,
+                                                 int32_t value)
+{ }
+
+static inline union cvmx_gmxx_rxx_rx_inbnd cvmx_spi4000_check_speed(
+       int interface,
+       int port)
+{
+       union cvmx_gmxx_rxx_rx_inbnd r;
+
+       r.u64 = 0;
+       return r;
+}
+
+static inline void cvmx_pko_send_packet_prepare(uint64_t port, uint64_t queue,
+                                               cvmx_pko_lock_t use_locking)
+{ }
+
+static inline cvmx_pko_status_t cvmx_pko_send_packet_finish(uint64_t port,
+               uint64_t queue, union cvmx_pko_command_word0 pko_command,
+               union cvmx_buf_ptr packet, cvmx_pko_lock_t use_locking)
+{
+       return 0;
+}
+
+static inline void cvmx_wqe_set_port(struct cvmx_wqe *work, int port)
+{ }
+
+static inline void cvmx_wqe_set_qos(struct cvmx_wqe *work, int qos)
+{ }
+
+static inline int cvmx_wqe_get_qos(struct cvmx_wqe *work)
+{
+       return 0;
+}
+
+static inline void cvmx_wqe_set_grp(struct cvmx_wqe *work, int grp)
+{ }
+
+static inline void cvmx_pow_work_submit(struct cvmx_wqe *wqp, uint32_t tag,
+                                       enum cvmx_pow_tag_type tag_type,
+                                       uint64_t qos, uint64_t grp)
+{ }
+
+#define CVMX_ASXX_RX_CLK_SETX(a, b)    ((a)+(b))
+#define CVMX_ASXX_TX_CLK_SETX(a, b)    ((a)+(b))
+#define CVMX_CIU_TIMX(a)               (a)
+#define CVMX_GMXX_RXX_ADR_CAM0(a, b)   ((a)+(b))
+#define CVMX_GMXX_RXX_ADR_CAM1(a, b)   ((a)+(b))
+#define CVMX_GMXX_RXX_ADR_CAM2(a, b)   ((a)+(b))
+#define CVMX_GMXX_RXX_ADR_CAM3(a, b)   ((a)+(b))
+#define CVMX_GMXX_RXX_ADR_CAM4(a, b)   ((a)+(b))
+#define CVMX_GMXX_RXX_ADR_CAM5(a, b)   ((a)+(b))
+#define CVMX_GMXX_RXX_FRM_CTL(a, b)    ((a)+(b))
+#define CVMX_GMXX_RXX_INT_REG(a, b)    ((a)+(b))
+#define CVMX_GMXX_SMACX(a, b)          ((a)+(b))
+#define CVMX_PIP_PRT_TAGX(a)           (a)
+#define CVMX_POW_PP_GRP_MSKX(a)                (a)
+#define CVMX_POW_WQ_INT_THRX(a)                (a)
+#define CVMX_SPXX_INT_MSK(a)           (a)
+#define CVMX_SPXX_INT_REG(a)           (a)
+#define CVMX_SSO_PPX_GRP_MSK(a)                (a)
+#define CVMX_SSO_WQ_INT_THRX(a)                (a)
+#define CVMX_STXX_INT_MSK(a)           (a)
+#define CVMX_STXX_INT_REG(a)           (a)