[PATCH] Add Myri-10G Ethernet driver
authorBrice Goglin <brice@myri.com>
Tue, 23 May 2006 10:10:15 +0000 (06:10 -0400)
committerJeff Garzik <jeff@garzik.org>
Wed, 24 May 2006 04:27:31 +0000 (00:27 -0400)
Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Andrew J. Gallatin <gallatin@myri.com>
 drivers/net/Kconfig                            |   17
 drivers/net/Makefile                           |    1
 drivers/net/myri10ge/Makefile                  |    5
 drivers/net/myri10ge/myri10ge.c                | 2851 +++++++++++++++
 drivers/net/myri10ge/myri10ge_mcp.h            |  205 +
 drivers/net/myri10ge/myri10ge_mcp_gen_header.h |   58
 include/linux/pci_ids.h                        |    1
 7 files changed, 3138 insertions(+)
Signed-off-by: Jeff Garzik <jeff@garzik.org>
drivers/net/Kconfig
drivers/net/Makefile
drivers/net/myri10ge/Makefile [new file with mode: 0644]
drivers/net/myri10ge/myri10ge.c [new file with mode: 0644]
drivers/net/myri10ge/myri10ge_mcp.h [new file with mode: 0644]
drivers/net/myri10ge/myri10ge_mcp_gen_header.h [new file with mode: 0644]
include/linux/pci_ids.h

index 68bc073b8b31f9e65cc300482cecb8db2f27a16a..f499a3bc629f3acf472d9e8f284d51e54926b663 100644 (file)
@@ -2327,6 +2327,23 @@ config S2IO_NAPI
 
          If in doubt, say N.
 
+config MYRI10GE
+       tristate "Myricom Myri-10G Ethernet support"
+       depends on PCI
+       select FW_LOADER
+       select CRC32
+       ---help---
+         This driver supports Myricom Myri-10G Dual Protocol interface in
+         Ethernet mode. If the eeprom on your board is not recent enough,
+         you will need a newer firmware image.
+         You may get this image or more information, at:
+
+         <http://www.myri.com/Myri-10G/>
+
+         To compile this driver as a module, choose M here and read
+         <file:Documentation/networking/net-modules.txt>.  The module
+         will be called myri10ge.
+
 endmenu
 
 source "drivers/net/tokenring/Kconfig"
index b01cc9a3cb10db0b012843b5a06f38f099aaef13..1eced3287507e94462b820408a3a2a8b0c90567f 100644 (file)
@@ -192,6 +192,7 @@ obj-$(CONFIG_R8169) += r8169.o
 obj-$(CONFIG_AMD8111_ETH) += amd8111e.o
 obj-$(CONFIG_IBMVETH) += ibmveth.o
 obj-$(CONFIG_S2IO) += s2io.o
+obj-$(CONFIG_MYRI10GE) += myri10ge/
 obj-$(CONFIG_SMC91X) += smc91x.o
 obj-$(CONFIG_SMC911X) += smc911x.o
 obj-$(CONFIG_DM9000) += dm9000.o
diff --git a/drivers/net/myri10ge/Makefile b/drivers/net/myri10ge/Makefile
new file mode 100644 (file)
index 0000000..5df8916
--- /dev/null
@@ -0,0 +1,5 @@
+#
+# Makefile for the Myricom Myri-10G ethernet driver
+#
+
+obj-$(CONFIG_MYRI10GE) += myri10ge.o
diff --git a/drivers/net/myri10ge/myri10ge.c b/drivers/net/myri10ge/myri10ge.c
new file mode 100644 (file)
index 0000000..87933cb
--- /dev/null
@@ -0,0 +1,2851 @@
+/*************************************************************************
+ * myri10ge.c: Myricom Myri-10G Ethernet driver.
+ *
+ * Copyright (C) 2005, 2006 Myricom, Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. Neither the name of Myricom, Inc. nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *
+ * If the eeprom on your board is not recent enough, you will need to get a
+ * newer firmware image at:
+ *   http://www.myri.com/scs/download-Myri10GE.html
+ *
+ * Contact Information:
+ *   <help@myri.com>
+ *   Myricom, Inc., 325N Santa Anita Avenue, Arcadia, CA 91006
+ *************************************************************************/
+
+#include <linux/tcp.h>
+#include <linux/netdevice.h>
+#include <linux/skbuff.h>
+#include <linux/string.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/etherdevice.h>
+#include <linux/if_ether.h>
+#include <linux/if_vlan.h>
+#include <linux/ip.h>
+#include <linux/inet.h>
+#include <linux/in.h>
+#include <linux/ethtool.h>
+#include <linux/firmware.h>
+#include <linux/delay.h>
+#include <linux/version.h>
+#include <linux/timer.h>
+#include <linux/vmalloc.h>
+#include <linux/crc32.h>
+#include <linux/moduleparam.h>
+#include <linux/io.h>
+#include <net/checksum.h>
+#include <asm/byteorder.h>
+#include <asm/io.h>
+#include <asm/pci.h>
+#include <asm/processor.h>
+#ifdef CONFIG_MTRR
+#include <asm/mtrr.h>
+#endif
+
+#include "myri10ge_mcp.h"
+#include "myri10ge_mcp_gen_header.h"
+
+#define MYRI10GE_VERSION_STR "0.9.0"
+
+MODULE_DESCRIPTION("Myricom 10G driver (10GbE)");
+MODULE_AUTHOR("Maintainer: help@myri.com");
+MODULE_VERSION(MYRI10GE_VERSION_STR);
+MODULE_LICENSE("Dual BSD/GPL");
+
+#define MYRI10GE_MAX_ETHER_MTU 9014
+
+#define MYRI10GE_ETH_STOPPED 0
+#define MYRI10GE_ETH_STOPPING 1
+#define MYRI10GE_ETH_STARTING 2
+#define MYRI10GE_ETH_RUNNING 3
+#define MYRI10GE_ETH_OPEN_FAILED 4
+
+#define MYRI10GE_EEPROM_STRINGS_SIZE 256
+#define MYRI10GE_MAX_SEND_DESC_TSO ((65536 / 2048) * 2)
+
+#define MYRI10GE_NO_CONFIRM_DATA 0xffffffff
+#define MYRI10GE_NO_RESPONSE_RESULT 0xffffffff
+
+struct myri10ge_rx_buffer_state {
+       struct sk_buff *skb;
+        DECLARE_PCI_UNMAP_ADDR(bus)
+        DECLARE_PCI_UNMAP_LEN(len)
+};
+
+struct myri10ge_tx_buffer_state {
+       struct sk_buff *skb;
+       int last;
+        DECLARE_PCI_UNMAP_ADDR(bus)
+        DECLARE_PCI_UNMAP_LEN(len)
+};
+
+struct myri10ge_cmd {
+       u32 data0;
+       u32 data1;
+       u32 data2;
+};
+
+struct myri10ge_rx_buf {
+       struct mcp_kreq_ether_recv __iomem *lanai;      /* lanai ptr for recv ring */
+       u8 __iomem *wc_fifo;    /* w/c rx dma addr fifo address */
+       struct mcp_kreq_ether_recv *shadow;     /* host shadow of recv ring */
+       struct myri10ge_rx_buffer_state *info;
+       int cnt;
+       int alloc_fail;
+       int mask;               /* number of rx slots -1 */
+};
+
+struct myri10ge_tx_buf {
+       struct mcp_kreq_ether_send __iomem *lanai;      /* lanai ptr for sendq */
+       u8 __iomem *wc_fifo;    /* w/c send fifo address */
+       struct mcp_kreq_ether_send *req_list;   /* host shadow of sendq */
+       char *req_bytes;
+       struct myri10ge_tx_buffer_state *info;
+       int mask;               /* number of transmit slots -1  */
+       int boundary;           /* boundary transmits cannot cross */
+       int req ____cacheline_aligned;  /* transmit slots submitted     */
+       int pkt_start;          /* packets started */
+       int done ____cacheline_aligned; /* transmit slots completed     */
+       int pkt_done;           /* packets completed */
+};
+
+struct myri10ge_rx_done {
+       struct mcp_slot *entry;
+       dma_addr_t bus;
+       int cnt;
+       int idx;
+};
+
+struct myri10ge_priv {
+       int running;            /* running?             */
+       int csum_flag;          /* rx_csums?            */
+       struct myri10ge_tx_buf tx;      /* transmit ring        */
+       struct myri10ge_rx_buf rx_small;
+       struct myri10ge_rx_buf rx_big;
+       struct myri10ge_rx_done rx_done;
+       int small_bytes;
+       struct net_device *dev;
+       struct net_device_stats stats;
+       u8 __iomem *sram;
+       int sram_size;
+       unsigned long board_span;
+       unsigned long iomem_base;
+       u32 __iomem *irq_claim;
+       u32 __iomem *irq_deassert;
+       char *mac_addr_string;
+       struct mcp_cmd_response *cmd;
+       dma_addr_t cmd_bus;
+       struct mcp_irq_data *fw_stats;
+       dma_addr_t fw_stats_bus;
+       struct pci_dev *pdev;
+       int msi_enabled;
+       unsigned int link_state;
+       unsigned int rdma_tags_available;
+       int intr_coal_delay;
+       u32 __iomem *intr_coal_delay_ptr;
+       int mtrr;
+       int wake_queue;
+       int stop_queue;
+       int down_cnt;
+       wait_queue_head_t down_wq;
+       struct work_struct watchdog_work;
+       struct timer_list watchdog_timer;
+       int watchdog_tx_done;
+       int watchdog_resets;
+       int tx_linearized;
+       int pause;
+       char *fw_name;
+       char eeprom_strings[MYRI10GE_EEPROM_STRINGS_SIZE];
+       char fw_version[128];
+       u8 mac_addr[6];         /* eeprom mac address */
+       unsigned long serial_number;
+       int vendor_specific_offset;
+       u32 devctl;
+       u16 msi_flags;
+       u32 pm_state[16];
+       u32 read_dma;
+       u32 write_dma;
+       u32 read_write_dma;
+};
+
+static char *myri10ge_fw_unaligned = "myri10ge_ethp_z8e.dat";
+static char *myri10ge_fw_aligned = "myri10ge_eth_z8e.dat";
+
+static char *myri10ge_fw_name = NULL;
+module_param(myri10ge_fw_name, charp, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(myri10ge_fw_name, "Firmware image name\n");
+
+static int myri10ge_ecrc_enable = 1;
+module_param(myri10ge_ecrc_enable, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_ecrc_enable, "Enable Extended CRC on PCI-E\n");
+
+static int myri10ge_max_intr_slots = 1024;
+module_param(myri10ge_max_intr_slots, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_max_intr_slots, "Interrupt queue slots\n");
+
+static int myri10ge_small_bytes = -1;  /* -1 == auto */
+module_param(myri10ge_small_bytes, int, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(myri10ge_small_bytes, "Threshold of small packets\n");
+
+static int myri10ge_msi = 1;   /* enable msi by default */
+module_param(myri10ge_msi, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_msi, "Enable Message Signalled Interrupts\n");
+
+static int myri10ge_intr_coal_delay = 25;
+module_param(myri10ge_intr_coal_delay, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_intr_coal_delay, "Interrupt coalescing delay\n");
+
+static int myri10ge_flow_control = 1;
+module_param(myri10ge_flow_control, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_flow_control, "Pause parameter\n");
+
+static int myri10ge_deassert_wait = 1;
+module_param(myri10ge_deassert_wait, int, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(myri10ge_deassert_wait,
+                "Wait when deasserting legacy interrupts\n");
+
+static int myri10ge_force_firmware = 0;
+module_param(myri10ge_force_firmware, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_force_firmware,
+                "Force firmware to assume aligned completions\n");
+
+static int myri10ge_skb_cross_4k = 0;
+module_param(myri10ge_skb_cross_4k, int, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(myri10ge_skb_cross_4k,
+                "Can a small skb cross a 4KB boundary?\n");
+
+static int myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN;
+module_param(myri10ge_initial_mtu, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_initial_mtu, "Initial MTU\n");
+
+static int myri10ge_napi_weight = 64;
+module_param(myri10ge_napi_weight, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_napi_weight, "Set NAPI weight\n");
+
+static int myri10ge_watchdog_timeout = 1;
+module_param(myri10ge_watchdog_timeout, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_watchdog_timeout, "Set watchdog timeout\n");
+
+static int myri10ge_max_irq_loops = 1048576;
+module_param(myri10ge_max_irq_loops, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_max_irq_loops,
+                "Set stuck legacy IRQ detection threshold\n");
+
+#define MYRI10GE_FW_OFFSET 1024*1024
+#define MYRI10GE_HIGHPART_TO_U32(X) \
+(sizeof (X) == 8) ? ((u32)((u64)(X) >> 32)) : (0)
+#define MYRI10GE_LOWPART_TO_U32(X) ((u32)(X))
+
+#define myri10ge_pio_copy(to,from,size) __iowrite64_copy(to,from,size/8)
+
+static int
+myri10ge_send_cmd(struct myri10ge_priv *mgp, u32 cmd,
+                 struct myri10ge_cmd *data, int atomic)
+{
+       struct mcp_cmd *buf;
+       char buf_bytes[sizeof(*buf) + 8];
+       struct mcp_cmd_response *response = mgp->cmd;
+       char __iomem *cmd_addr = mgp->sram + MXGEFW_CMD_OFFSET;
+       u32 dma_low, dma_high, result, value;
+       int sleep_total = 0;
+
+       /* ensure buf is aligned to 8 bytes */
+       buf = (struct mcp_cmd *)ALIGN((unsigned long)buf_bytes, 8);
+
+       buf->data0 = htonl(data->data0);
+       buf->data1 = htonl(data->data1);
+       buf->data2 = htonl(data->data2);
+       buf->cmd = htonl(cmd);
+       dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
+       dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
+
+       buf->response_addr.low = htonl(dma_low);
+       buf->response_addr.high = htonl(dma_high);
+       response->result = MYRI10GE_NO_RESPONSE_RESULT;
+       mb();
+       myri10ge_pio_copy(cmd_addr, buf, sizeof(*buf));
+
+       /* wait up to 15ms. Longest command is the DMA benchmark,
+        * which is capped at 5ms, but runs from a timeout handler
+        * that runs every 7.8ms. So a 15ms timeout leaves us with
+        * a 2.2ms margin
+        */
+       if (atomic) {
+               /* if atomic is set, do not sleep,
+                * and try to get the completion quickly
+                * (1ms will be enough for those commands) */
+               for (sleep_total = 0;
+                    sleep_total < 1000
+                    && response->result == MYRI10GE_NO_RESPONSE_RESULT;
+                    sleep_total += 10)
+                       udelay(10);
+       } else {
+               /* use msleep for most command */
+               for (sleep_total = 0;
+                    sleep_total < 15
+                    && response->result == MYRI10GE_NO_RESPONSE_RESULT;
+                    sleep_total++)
+                       msleep(1);
+       }
+
+       result = ntohl(response->result);
+       value = ntohl(response->data);
+       if (result != MYRI10GE_NO_RESPONSE_RESULT) {
+               if (result == 0) {
+                       data->data0 = value;
+                       return 0;
+               } else {
+                       dev_err(&mgp->pdev->dev,
+                               "command %d failed, result = %d\n",
+                               cmd, result);
+                       return -ENXIO;
+               }
+       }
+
+       dev_err(&mgp->pdev->dev, "command %d timed out, result = %d\n",
+               cmd, result);
+       return -EAGAIN;
+}
+
+/*
+ * The eeprom strings on the lanaiX have the format
+ * SN=x\0
+ * MAC=x:x:x:x:x:x\0
+ * PT:ddd mmm xx xx:xx:xx xx\0
+ * PV:ddd mmm xx xx:xx:xx xx\0
+ */
+static int myri10ge_read_mac_addr(struct myri10ge_priv *mgp)
+{
+       char *ptr, *limit;
+       int i;
+
+       ptr = mgp->eeprom_strings;
+       limit = mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE;
+
+       while (*ptr != '\0' && ptr < limit) {
+               if (memcmp(ptr, "MAC=", 4) == 0) {
+                       ptr += 4;
+                       mgp->mac_addr_string = ptr;
+                       for (i = 0; i < 6; i++) {
+                               if ((ptr + 2) > limit)
+                                       goto abort;
+                               mgp->mac_addr[i] =
+                                   simple_strtoul(ptr, &ptr, 16);
+                               ptr += 1;
+                       }
+               }
+               if (memcmp((const void *)ptr, "SN=", 3) == 0) {
+                       ptr += 3;
+                       mgp->serial_number = simple_strtoul(ptr, &ptr, 10);
+               }
+               while (ptr < limit && *ptr++) ;
+       }
+
+       return 0;
+
+abort:
+       dev_err(&mgp->pdev->dev, "failed to parse eeprom_strings\n");
+       return -ENXIO;
+}
+
+/*
+ * Enable or disable periodic RDMAs from the host to make certain
+ * chipsets resend dropped PCIe messages
+ */
+
+static void myri10ge_dummy_rdma(struct myri10ge_priv *mgp, int enable)
+{
+       char __iomem *submit;
+       u32 buf[16];
+       u32 dma_low, dma_high;
+       int i;
+
+       /* clear confirmation addr */
+       mgp->cmd->data = 0;
+       mb();
+
+       /* send a rdma command to the PCIe engine, and wait for the
+        * response in the confirmation address.  The firmware should
+        * write a -1 there to indicate it is alive and well
+        */
+       dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
+       dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
+
+       buf[0] = htonl(dma_high);       /* confirm addr MSW */
+       buf[1] = htonl(dma_low);        /* confirm addr LSW */
+       buf[2] = htonl(MYRI10GE_NO_CONFIRM_DATA);       /* confirm data */
+       buf[3] = htonl(dma_high);       /* dummy addr MSW */
+       buf[4] = htonl(dma_low);        /* dummy addr LSW */
+       buf[5] = htonl(enable); /* enable? */
+
+       submit = mgp->sram + 0xfc01c0;
+
+       myri10ge_pio_copy(submit, &buf, sizeof(buf));
+       for (i = 0; mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20; i++)
+               msleep(1);
+       if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA)
+               dev_err(&mgp->pdev->dev, "dummy rdma %s failed\n",
+                       (enable ? "enable" : "disable"));
+}
+
+static int
+myri10ge_validate_firmware(struct myri10ge_priv *mgp,
+                          struct mcp_gen_header *hdr)
+{
+       struct device *dev = &mgp->pdev->dev;
+       int major, minor;
+
+       /* check firmware type */
+       if (ntohl(hdr->mcp_type) != MCP_TYPE_ETH) {
+               dev_err(dev, "Bad firmware type: 0x%x\n", ntohl(hdr->mcp_type));
+               return -EINVAL;
+       }
+
+       /* save firmware version for ethtool */
+       strncpy(mgp->fw_version, hdr->version, sizeof(mgp->fw_version));
+
+       sscanf(mgp->fw_version, "%d.%d", &major, &minor);
+
+       if (!(major == MXGEFW_VERSION_MAJOR && minor == MXGEFW_VERSION_MINOR)) {
+               dev_err(dev, "Found firmware version %s\n", mgp->fw_version);
+               dev_err(dev, "Driver needs %d.%d\n", MXGEFW_VERSION_MAJOR,
+                       MXGEFW_VERSION_MINOR);
+               return -EINVAL;
+       }
+       return 0;
+}
+
+static int myri10ge_load_hotplug_firmware(struct myri10ge_priv *mgp, u32 * size)
+{
+       unsigned crc, reread_crc;
+       const struct firmware *fw;
+       struct device *dev = &mgp->pdev->dev;
+       struct mcp_gen_header *hdr;
+       size_t hdr_offset;
+       int status;
+
+       if ((status = request_firmware(&fw, mgp->fw_name, dev)) < 0) {
+               dev_err(dev, "Unable to load %s firmware image via hotplug\n",
+                       mgp->fw_name);
+               status = -EINVAL;
+               goto abort_with_nothing;
+       }
+
+       /* check size */
+
+       if (fw->size >= mgp->sram_size - MYRI10GE_FW_OFFSET ||
+           fw->size < MCP_HEADER_PTR_OFFSET + 4) {
+               dev_err(dev, "Firmware size invalid:%d\n", (int)fw->size);
+               status = -EINVAL;
+               goto abort_with_fw;
+       }
+
+       /* check id */
+       hdr_offset = ntohl(*(u32 *) (fw->data + MCP_HEADER_PTR_OFFSET));
+       if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > fw->size) {
+               dev_err(dev, "Bad firmware file\n");
+               status = -EINVAL;
+               goto abort_with_fw;
+       }
+       hdr = (void *)(fw->data + hdr_offset);
+
+       status = myri10ge_validate_firmware(mgp, hdr);
+       if (status != 0)
+               goto abort_with_fw;
+
+       crc = crc32(~0, fw->data, fw->size);
+       memcpy_toio(mgp->sram + MYRI10GE_FW_OFFSET, fw->data, fw->size);
+       /* corruption checking is good for parity recovery and buggy chipset */
+       memcpy_fromio(fw->data, mgp->sram + MYRI10GE_FW_OFFSET, fw->size);
+       reread_crc = crc32(~0, fw->data, fw->size);
+       if (crc != reread_crc) {
+               dev_err(dev, "CRC failed(fw-len=%u), got 0x%x (expect 0x%x)\n",
+                       (unsigned)fw->size, reread_crc, crc);
+               status = -EIO;
+               goto abort_with_fw;
+       }
+       *size = (u32) fw->size;
+
+abort_with_fw:
+       release_firmware(fw);
+
+abort_with_nothing:
+       return status;
+}
+
+static int myri10ge_adopt_running_firmware(struct myri10ge_priv *mgp)
+{
+       struct mcp_gen_header *hdr;
+       struct device *dev = &mgp->pdev->dev;
+       const size_t bytes = sizeof(struct mcp_gen_header);
+       size_t hdr_offset;
+       int status;
+
+       /* find running firmware header */
+       hdr_offset = ntohl(__raw_readl(mgp->sram + MCP_HEADER_PTR_OFFSET));
+
+       if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > mgp->sram_size) {
+               dev_err(dev, "Running firmware has bad header offset (%d)\n",
+                       (int)hdr_offset);
+               return -EIO;
+       }
+
+       /* copy header of running firmware from SRAM to host memory to
+        * validate firmware */
+       hdr = kmalloc(bytes, GFP_KERNEL);
+       if (hdr == NULL) {
+               dev_err(dev, "could not malloc firmware hdr\n");
+               return -ENOMEM;
+       }
+       memcpy_fromio(hdr, mgp->sram + hdr_offset, bytes);
+       status = myri10ge_validate_firmware(mgp, hdr);
+       kfree(hdr);
+       return status;
+}
+
+static int myri10ge_load_firmware(struct myri10ge_priv *mgp)
+{
+       char __iomem *submit;
+       u32 buf[16];
+       u32 dma_low, dma_high, size;
+       int status, i;
+
+       status = myri10ge_load_hotplug_firmware(mgp, &size);
+       if (status) {
+               dev_warn(&mgp->pdev->dev, "hotplug firmware loading failed\n");
+
+               /* Do not attempt to adopt firmware if there
+                * was a bad crc */
+               if (status == -EIO)
+                       return status;
+
+               status = myri10ge_adopt_running_firmware(mgp);
+               if (status != 0) {
+                       dev_err(&mgp->pdev->dev,
+                               "failed to adopt running firmware\n");
+                       return status;
+               }
+               dev_info(&mgp->pdev->dev,
+                        "Successfully adopted running firmware\n");
+               if (mgp->tx.boundary == 4096) {
+                       dev_warn(&mgp->pdev->dev,
+                                "Using firmware currently running on NIC"
+                                ".  For optimal\n");
+                       dev_warn(&mgp->pdev->dev,
+                                "performance consider loading optimized "
+                                "firmware\n");
+                       dev_warn(&mgp->pdev->dev, "via hotplug\n");
+               }
+
+               mgp->fw_name = "adopted";
+               mgp->tx.boundary = 2048;
+               return status;
+       }
+
+       /* clear confirmation addr */
+       mgp->cmd->data = 0;
+       mb();
+
+       /* send a reload command to the bootstrap MCP, and wait for the
+        *  response in the confirmation address.  The firmware should
+        * write a -1 there to indicate it is alive and well
+        */
+       dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
+       dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
+
+       buf[0] = htonl(dma_high);       /* confirm addr MSW */
+       buf[1] = htonl(dma_low);        /* confirm addr LSW */
+       buf[2] = htonl(MYRI10GE_NO_CONFIRM_DATA);       /* confirm data */
+
+       /* FIX: All newest firmware should un-protect the bottom of
+        * the sram before handoff. However, the very first interfaces
+        * do not. Therefore the handoff copy must skip the first 8 bytes
+        */
+       buf[3] = htonl(MYRI10GE_FW_OFFSET + 8); /* where the code starts */
+       buf[4] = htonl(size - 8);       /* length of code */
+       buf[5] = htonl(8);      /* where to copy to */
+       buf[6] = htonl(0);      /* where to jump to */
+
+       submit = mgp->sram + 0xfc0000;
+
+       myri10ge_pio_copy(submit, &buf, sizeof(buf));
+       mb();
+       msleep(1);
+       mb();
+       i = 0;
+       while (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20) {
+               msleep(1);
+               i++;
+       }
+       if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA) {
+               dev_err(&mgp->pdev->dev, "handoff failed\n");
+               return -ENXIO;
+       }
+       dev_info(&mgp->pdev->dev, "handoff confirmed\n");
+       myri10ge_dummy_rdma(mgp, mgp->tx.boundary != 4096);
+
+       return 0;
+}
+
+static int myri10ge_update_mac_address(struct myri10ge_priv *mgp, u8 * addr)
+{
+       struct myri10ge_cmd cmd;
+       int status;
+
+       cmd.data0 = ((addr[0] << 24) | (addr[1] << 16)
+                    | (addr[2] << 8) | addr[3]);
+
+       cmd.data1 = ((addr[4] << 8) | (addr[5]));
+
+       status = myri10ge_send_cmd(mgp, MXGEFW_SET_MAC_ADDRESS, &cmd, 0);
+       return status;
+}
+
+static int myri10ge_change_pause(struct myri10ge_priv *mgp, int pause)
+{
+       struct myri10ge_cmd cmd;
+       int status, ctl;
+
+       ctl = pause ? MXGEFW_ENABLE_FLOW_CONTROL : MXGEFW_DISABLE_FLOW_CONTROL;
+       status = myri10ge_send_cmd(mgp, ctl, &cmd, 0);
+
+       if (status) {
+               printk(KERN_ERR
+                      "myri10ge: %s: Failed to set flow control mode\n",
+                      mgp->dev->name);
+               return status;
+       }
+       mgp->pause = pause;
+       return 0;
+}
+
+static void
+myri10ge_change_promisc(struct myri10ge_priv *mgp, int promisc, int atomic)
+{
+       struct myri10ge_cmd cmd;
+       int status, ctl;
+
+       ctl = promisc ? MXGEFW_ENABLE_PROMISC : MXGEFW_DISABLE_PROMISC;
+       status = myri10ge_send_cmd(mgp, ctl, &cmd, atomic);
+       if (status)
+               printk(KERN_ERR "myri10ge: %s: Failed to set promisc mode\n",
+                      mgp->dev->name);
+}
+
+static int myri10ge_reset(struct myri10ge_priv *mgp)
+{
+       struct myri10ge_cmd cmd;
+       int status;
+       size_t bytes;
+       u32 len;
+
+       /* try to send a reset command to the card to see if it
+        * is alive */
+       memset(&cmd, 0, sizeof(cmd));
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_RESET, &cmd, 0);
+       if (status != 0) {
+               dev_err(&mgp->pdev->dev, "failed reset\n");
+               return -ENXIO;
+       }
+
+       /* Now exchange information about interrupts  */
+
+       bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
+       memset(mgp->rx_done.entry, 0, bytes);
+       cmd.data0 = (u32) bytes;
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_SIZE, &cmd, 0);
+       cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
+       cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
+       status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_DMA, &cmd, 0);
+
+       status |=
+           myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_ACK_OFFSET, &cmd, 0);
+       mgp->irq_claim = (__iomem u32 *) (mgp->sram + cmd.data0);
+       if (!mgp->msi_enabled) {
+               status |= myri10ge_send_cmd
+                   (mgp, MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET, &cmd, 0);
+               mgp->irq_deassert = (__iomem u32 *) (mgp->sram + cmd.data0);
+
+       }
+       status |= myri10ge_send_cmd
+           (mgp, MXGEFW_CMD_GET_INTR_COAL_DELAY_OFFSET, &cmd, 0);
+       mgp->intr_coal_delay_ptr = (__iomem u32 *) (mgp->sram + cmd.data0);
+       if (status != 0) {
+               dev_err(&mgp->pdev->dev, "failed set interrupt parameters\n");
+               return status;
+       }
+       __raw_writel(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
+
+       /* Run a small DMA test.
+        * The magic multipliers to the length tell the firmware
+        * to do DMA read, write, or read+write tests.  The
+        * results are returned in cmd.data0.  The upper 16
+        * bits or the return is the number of transfers completed.
+        * The lower 16 bits is the time in 0.5us ticks that the
+        * transfers took to complete.
+        */
+
+       len = mgp->tx.boundary;
+
+       cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
+       cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
+       cmd.data2 = len * 0x10000;
+       status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0);
+       if (status == 0)
+               mgp->read_dma = ((cmd.data0 >> 16) * len * 2) /
+                   (cmd.data0 & 0xffff);
+       else
+               dev_warn(&mgp->pdev->dev, "DMA read benchmark failed: %d\n",
+                        status);
+       cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
+       cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
+       cmd.data2 = len * 0x1;
+       status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0);
+       if (status == 0)
+               mgp->write_dma = ((cmd.data0 >> 16) * len * 2) /
+                   (cmd.data0 & 0xffff);
+       else
+               dev_warn(&mgp->pdev->dev, "DMA write benchmark failed: %d\n",
+                        status);
+
+       cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
+       cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
+       cmd.data2 = len * 0x10001;
+       status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0);
+       if (status == 0)
+               mgp->read_write_dma = ((cmd.data0 >> 16) * len * 2 * 2) /
+                   (cmd.data0 & 0xffff);
+       else
+               dev_warn(&mgp->pdev->dev,
+                        "DMA read/write benchmark failed: %d\n", status);
+
+       memset(mgp->rx_done.entry, 0, bytes);
+
+       /* reset mcp/driver shared state back to 0 */
+       mgp->tx.req = 0;
+       mgp->tx.done = 0;
+       mgp->tx.pkt_start = 0;
+       mgp->tx.pkt_done = 0;
+       mgp->rx_big.cnt = 0;
+       mgp->rx_small.cnt = 0;
+       mgp->rx_done.idx = 0;
+       mgp->rx_done.cnt = 0;
+       status = myri10ge_update_mac_address(mgp, mgp->dev->dev_addr);
+       myri10ge_change_promisc(mgp, 0, 0);
+       myri10ge_change_pause(mgp, mgp->pause);
+       return status;
+}
+
+static inline void
+myri10ge_submit_8rx(struct mcp_kreq_ether_recv __iomem * dst,
+                   struct mcp_kreq_ether_recv *src)
+{
+       u32 low;
+
+       low = src->addr_low;
+       src->addr_low = DMA_32BIT_MASK;
+       myri10ge_pio_copy(dst, src, 8 * sizeof(*src));
+       mb();
+       src->addr_low = low;
+       __raw_writel(low, &dst->addr_low);
+       mb();
+}
+
+/*
+ * Set of routunes to get a new receive buffer.  Any buffer which
+ * crosses a 4KB boundary must start on a 4KB boundary due to PCIe
+ * wdma restrictions. We also try to align any smaller allocation to
+ * at least a 16 byte boundary for efficiency.  We assume the linux
+ * memory allocator works by powers of 2, and will not return memory
+ * smaller than 2KB which crosses a 4KB boundary.  If it does, we fall
+ * back to allocating 2x as much space as required.
+ *
+ * We intend to replace large (>4KB) skb allocations by using
+ * pages directly and building a fraglist in the near future.
+ */
+
+static inline struct sk_buff *myri10ge_alloc_big(int bytes)
+{
+       struct sk_buff *skb;
+       unsigned long data, roundup;
+
+       skb = dev_alloc_skb(bytes + 4096 + MXGEFW_PAD);
+       if (skb == NULL)
+               return NULL;
+
+       /* Correct skb->truesize so that socket buffer
+        * accounting is not confused the rounding we must
+        * do to satisfy alignment constraints.
+        */
+       skb->truesize -= 4096;
+
+       data = (unsigned long)(skb->data);
+       roundup = (-data) & (4095);
+       skb_reserve(skb, roundup);
+       return skb;
+}
+
+/* Allocate 2x as much space as required and use whichever portion
+ * does not cross a 4KB boundary */
+static inline struct sk_buff *myri10ge_alloc_small_safe(unsigned int bytes)
+{
+       struct sk_buff *skb;
+       unsigned long data, boundary;
+
+       skb = dev_alloc_skb(2 * (bytes + MXGEFW_PAD) - 1);
+       if (unlikely(skb == NULL))
+               return NULL;
+
+       /* Correct skb->truesize so that socket buffer
+        * accounting is not confused the rounding we must
+        * do to satisfy alignment constraints.
+        */
+       skb->truesize -= bytes + MXGEFW_PAD;
+
+       data = (unsigned long)(skb->data);
+       boundary = (data + 4095UL) & ~4095UL;
+       if ((boundary - data) >= (bytes + MXGEFW_PAD))
+               return skb;
+
+       skb_reserve(skb, boundary - data);
+       return skb;
+}
+
+/* Allocate just enough space, and verify that the allocated
+ * space does not cross a 4KB boundary */
+static inline struct sk_buff *myri10ge_alloc_small(int bytes)
+{
+       struct sk_buff *skb;
+       unsigned long roundup, data, end;
+
+       skb = dev_alloc_skb(bytes + 16 + MXGEFW_PAD);
+       if (unlikely(skb == NULL))
+               return NULL;
+
+       /* Round allocated buffer to 16 byte boundary */
+       data = (unsigned long)(skb->data);
+       roundup = (-data) & 15UL;
+       skb_reserve(skb, roundup);
+       /* Verify that the data buffer does not cross a page boundary */
+       data = (unsigned long)(skb->data);
+       end = data + bytes + MXGEFW_PAD - 1;
+       if (unlikely(((end >> 12) != (data >> 12)) && (data & 4095UL))) {
+               printk(KERN_NOTICE
+                      "myri10ge_alloc_small: small skb crossed 4KB boundary\n");
+               myri10ge_skb_cross_4k = 1;
+               dev_kfree_skb_any(skb);
+               skb = myri10ge_alloc_small_safe(bytes);
+       }
+       return skb;
+}
+
+static inline int
+myri10ge_getbuf(struct myri10ge_rx_buf *rx, struct pci_dev *pdev, int bytes,
+               int idx)
+{
+       struct sk_buff *skb;
+       dma_addr_t bus;
+       int len, retval = 0;
+
+       bytes += VLAN_HLEN;     /* account for 802.1q vlan tag */
+
+       if ((bytes + MXGEFW_PAD) > (4096 - 16) /* linux overhead */ )
+               skb = myri10ge_alloc_big(bytes);
+       else if (myri10ge_skb_cross_4k)
+               skb = myri10ge_alloc_small_safe(bytes);
+       else
+               skb = myri10ge_alloc_small(bytes);
+
+       if (unlikely(skb == NULL)) {
+               rx->alloc_fail++;
+               retval = -ENOBUFS;
+               goto done;
+       }
+
+       /* set len so that it only covers the area we
+        * need mapped for DMA */
+       len = bytes + MXGEFW_PAD;
+
+       bus = pci_map_single(pdev, skb->data, len, PCI_DMA_FROMDEVICE);
+       rx->info[idx].skb = skb;
+       pci_unmap_addr_set(&rx->info[idx], bus, bus);
+       pci_unmap_len_set(&rx->info[idx], len, len);
+       rx->shadow[idx].addr_low = htonl(MYRI10GE_LOWPART_TO_U32(bus));
+       rx->shadow[idx].addr_high = htonl(MYRI10GE_HIGHPART_TO_U32(bus));
+
+done:
+       /* copy 8 descriptors (64-bytes) to the mcp at a time */
+       if ((idx & 7) == 7) {
+               if (rx->wc_fifo == NULL)
+                       myri10ge_submit_8rx(&rx->lanai[idx - 7],
+                                           &rx->shadow[idx - 7]);
+               else {
+                       mb();
+                       myri10ge_pio_copy(rx->wc_fifo,
+                                         &rx->shadow[idx - 7], 64);
+               }
+       }
+       return retval;
+}
+
+static inline void myri10ge_vlan_ip_csum(struct sk_buff *skb, u16 hw_csum)
+{
+       struct vlan_hdr *vh = (struct vlan_hdr *)(skb->data);
+
+       if ((skb->protocol == ntohs(ETH_P_8021Q)) &&
+           (vh->h_vlan_encapsulated_proto == htons(ETH_P_IP) ||
+            vh->h_vlan_encapsulated_proto == htons(ETH_P_IPV6))) {
+               skb->csum = hw_csum;
+               skb->ip_summed = CHECKSUM_HW;
+       }
+}
+
+static inline unsigned long
+myri10ge_rx_done(struct myri10ge_priv *mgp, struct myri10ge_rx_buf *rx,
+                int bytes, int len, int csum)
+{
+       dma_addr_t bus;
+       struct sk_buff *skb;
+       int idx, unmap_len;
+
+       idx = rx->cnt & rx->mask;
+       rx->cnt++;
+
+       /* save a pointer to the received skb */
+       skb = rx->info[idx].skb;
+       bus = pci_unmap_addr(&rx->info[idx], bus);
+       unmap_len = pci_unmap_len(&rx->info[idx], len);
+
+       /* try to replace the received skb */
+       if (myri10ge_getbuf(rx, mgp->pdev, bytes, idx)) {
+               /* drop the frame -- the old skbuf is re-cycled */
+               mgp->stats.rx_dropped += 1;
+               return 0;
+       }
+
+       /* unmap the recvd skb */
+       pci_unmap_single(mgp->pdev, bus, unmap_len, PCI_DMA_FROMDEVICE);
+
+       /* mcp implicitly skips 1st bytes so that packet is properly
+        * aligned */
+       skb_reserve(skb, MXGEFW_PAD);
+
+       /* set the length of the frame */
+       skb_put(skb, len);
+
+       skb->protocol = eth_type_trans(skb, mgp->dev);
+       skb->dev = mgp->dev;
+       if (mgp->csum_flag) {
+               if ((skb->protocol == ntohs(ETH_P_IP)) ||
+                   (skb->protocol == ntohs(ETH_P_IPV6))) {
+                       skb->csum = ntohs((u16) csum);
+                       skb->ip_summed = CHECKSUM_HW;
+               } else
+                       myri10ge_vlan_ip_csum(skb, ntohs((u16) csum));
+       }
+
+       netif_receive_skb(skb);
+       mgp->dev->last_rx = jiffies;
+       return 1;
+}
+
+static inline void myri10ge_tx_done(struct myri10ge_priv *mgp, int mcp_index)
+{
+       struct pci_dev *pdev = mgp->pdev;
+       struct myri10ge_tx_buf *tx = &mgp->tx;
+       struct sk_buff *skb;
+       int idx, len;
+       int limit = 0;
+
+       while (tx->pkt_done != mcp_index) {
+               idx = tx->done & tx->mask;
+               skb = tx->info[idx].skb;
+
+               /* Mark as free */
+               tx->info[idx].skb = NULL;
+               if (tx->info[idx].last) {
+                       tx->pkt_done++;
+                       tx->info[idx].last = 0;
+               }
+               tx->done++;
+               len = pci_unmap_len(&tx->info[idx], len);
+               pci_unmap_len_set(&tx->info[idx], len, 0);
+               if (skb) {
+                       mgp->stats.tx_bytes += skb->len;
+                       mgp->stats.tx_packets++;
+                       dev_kfree_skb_irq(skb);
+                       if (len)
+                               pci_unmap_single(pdev,
+                                                pci_unmap_addr(&tx->info[idx],
+                                                               bus), len,
+                                                PCI_DMA_TODEVICE);
+               } else {
+                       if (len)
+                               pci_unmap_page(pdev,
+                                              pci_unmap_addr(&tx->info[idx],
+                                                             bus), len,
+                                              PCI_DMA_TODEVICE);
+               }
+
+               /* limit potential for livelock by only handling
+                * 2 full tx rings per call */
+               if (unlikely(++limit > 2 * tx->mask))
+                       break;
+       }
+       /* start the queue if we've stopped it */
+       if (netif_queue_stopped(mgp->dev)
+           && tx->req - tx->done < (tx->mask >> 1)) {
+               mgp->wake_queue++;
+               netif_wake_queue(mgp->dev);
+       }
+}
+
+static inline void myri10ge_clean_rx_done(struct myri10ge_priv *mgp, int *limit)
+{
+       struct myri10ge_rx_done *rx_done = &mgp->rx_done;
+       unsigned long rx_bytes = 0;
+       unsigned long rx_packets = 0;
+       unsigned long rx_ok;
+
+       int idx = rx_done->idx;
+       int cnt = rx_done->cnt;
+       u16 length;
+       u16 checksum;
+
+       while (rx_done->entry[idx].length != 0 && *limit != 0) {
+               length = ntohs(rx_done->entry[idx].length);
+               rx_done->entry[idx].length = 0;
+               checksum = ntohs(rx_done->entry[idx].checksum);
+               if (length <= mgp->small_bytes)
+                       rx_ok = myri10ge_rx_done(mgp, &mgp->rx_small,
+                                                mgp->small_bytes,
+                                                length, checksum);
+               else
+                       rx_ok = myri10ge_rx_done(mgp, &mgp->rx_big,
+                                                mgp->dev->mtu + ETH_HLEN,
+                                                length, checksum);
+               rx_packets += rx_ok;
+               rx_bytes += rx_ok * (unsigned long)length;
+               cnt++;
+               idx = cnt & (myri10ge_max_intr_slots - 1);
+
+               /* limit potential for livelock by only handling a
+                * limited number of frames. */
+               (*limit)--;
+       }
+       rx_done->idx = idx;
+       rx_done->cnt = cnt;
+       mgp->stats.rx_packets += rx_packets;
+       mgp->stats.rx_bytes += rx_bytes;
+}
+
+static inline void myri10ge_check_statblock(struct myri10ge_priv *mgp)
+{
+       struct mcp_irq_data *stats = mgp->fw_stats;
+
+       if (unlikely(stats->stats_updated)) {
+               if (mgp->link_state != stats->link_up) {
+                       mgp->link_state = stats->link_up;
+                       if (mgp->link_state) {
+                               printk(KERN_INFO "myri10ge: %s: link up\n",
+                                      mgp->dev->name);
+                               netif_carrier_on(mgp->dev);
+                       } else {
+                               printk(KERN_INFO "myri10ge: %s: link down\n",
+                                      mgp->dev->name);
+                               netif_carrier_off(mgp->dev);
+                       }
+               }
+               if (mgp->rdma_tags_available !=
+                   ntohl(mgp->fw_stats->rdma_tags_available)) {
+                       mgp->rdma_tags_available =
+                           ntohl(mgp->fw_stats->rdma_tags_available);
+                       printk(KERN_WARNING "myri10ge: %s: RDMA timed out! "
+                              "%d tags left\n", mgp->dev->name,
+                              mgp->rdma_tags_available);
+               }
+               mgp->down_cnt += stats->link_down;
+               if (stats->link_down)
+                       wake_up(&mgp->down_wq);
+       }
+}
+
+static int myri10ge_poll(struct net_device *netdev, int *budget)
+{
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+       struct myri10ge_rx_done *rx_done = &mgp->rx_done;
+       int limit, orig_limit, work_done;
+
+       /* process as many rx events as NAPI will allow */
+       limit = min(*budget, netdev->quota);
+       orig_limit = limit;
+       myri10ge_clean_rx_done(mgp, &limit);
+       work_done = orig_limit - limit;
+       *budget -= work_done;
+       netdev->quota -= work_done;
+
+       if (rx_done->entry[rx_done->idx].length == 0 || !netif_running(netdev)) {
+               netif_rx_complete(netdev);
+               __raw_writel(htonl(3), mgp->irq_claim);
+               return 0;
+       }
+       return 1;
+}
+
+static irqreturn_t myri10ge_intr(int irq, void *arg, struct pt_regs *regs)
+{
+       struct myri10ge_priv *mgp = arg;
+       struct mcp_irq_data *stats = mgp->fw_stats;
+       struct myri10ge_tx_buf *tx = &mgp->tx;
+       u32 send_done_count;
+       int i;
+
+       /* make sure it is our IRQ, and that the DMA has finished */
+       if (unlikely(!stats->valid))
+               return (IRQ_NONE);
+
+       /* low bit indicates receives are present, so schedule
+        * napi poll handler */
+       if (stats->valid & 1)
+               netif_rx_schedule(mgp->dev);
+
+       if (!mgp->msi_enabled) {
+               __raw_writel(0, mgp->irq_deassert);
+               if (!myri10ge_deassert_wait)
+                       stats->valid = 0;
+               mb();
+       } else
+               stats->valid = 0;
+
+       /* Wait for IRQ line to go low, if using INTx */
+       i = 0;
+       while (1) {
+               i++;
+               /* check for transmit completes and receives */
+               send_done_count = ntohl(stats->send_done_count);
+               if (send_done_count != tx->pkt_done)
+                       myri10ge_tx_done(mgp, (int)send_done_count);
+               if (unlikely(i > myri10ge_max_irq_loops)) {
+                       printk(KERN_WARNING "myri10ge: %s: irq stuck?\n",
+                              mgp->dev->name);
+                       stats->valid = 0;
+                       schedule_work(&mgp->watchdog_work);
+               }
+               if (likely(stats->valid == 0))
+                       break;
+               cpu_relax();
+               barrier();
+       }
+
+       myri10ge_check_statblock(mgp);
+
+       __raw_writel(htonl(3), mgp->irq_claim + 1);
+       return (IRQ_HANDLED);
+}
+
+static int
+myri10ge_get_settings(struct net_device *netdev, struct ethtool_cmd *cmd)
+{
+       cmd->autoneg = AUTONEG_DISABLE;
+       cmd->speed = SPEED_10000;
+       cmd->duplex = DUPLEX_FULL;
+       return 0;
+}
+
+static void
+myri10ge_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *info)
+{
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+
+       strlcpy(info->driver, "myri10ge", sizeof(info->driver));
+       strlcpy(info->version, MYRI10GE_VERSION_STR, sizeof(info->version));
+       strlcpy(info->fw_version, mgp->fw_version, sizeof(info->fw_version));
+       strlcpy(info->bus_info, pci_name(mgp->pdev), sizeof(info->bus_info));
+}
+
+static int
+myri10ge_get_coalesce(struct net_device *netdev, struct ethtool_coalesce *coal)
+{
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+       coal->rx_coalesce_usecs = mgp->intr_coal_delay;
+       return 0;
+}
+
+static int
+myri10ge_set_coalesce(struct net_device *netdev, struct ethtool_coalesce *coal)
+{
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+
+       mgp->intr_coal_delay = coal->rx_coalesce_usecs;
+       __raw_writel(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
+       return 0;
+}
+
+static void
+myri10ge_get_pauseparam(struct net_device *netdev,
+                       struct ethtool_pauseparam *pause)
+{
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+
+       pause->autoneg = 0;
+       pause->rx_pause = mgp->pause;
+       pause->tx_pause = mgp->pause;
+}
+
+static int
+myri10ge_set_pauseparam(struct net_device *netdev,
+                       struct ethtool_pauseparam *pause)
+{
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+
+       if (pause->tx_pause != mgp->pause)
+               return myri10ge_change_pause(mgp, pause->tx_pause);
+       if (pause->rx_pause != mgp->pause)
+               return myri10ge_change_pause(mgp, pause->tx_pause);
+       if (pause->autoneg != 0)
+               return -EINVAL;
+       return 0;
+}
+
+static void
+myri10ge_get_ringparam(struct net_device *netdev,
+                      struct ethtool_ringparam *ring)
+{
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+
+       ring->rx_mini_max_pending = mgp->rx_small.mask + 1;
+       ring->rx_max_pending = mgp->rx_big.mask + 1;
+       ring->rx_jumbo_max_pending = 0;
+       ring->tx_max_pending = mgp->rx_small.mask + 1;
+       ring->rx_mini_pending = ring->rx_mini_max_pending;
+       ring->rx_pending = ring->rx_max_pending;
+       ring->rx_jumbo_pending = ring->rx_jumbo_max_pending;
+       ring->tx_pending = ring->tx_max_pending;
+}
+
+static u32 myri10ge_get_rx_csum(struct net_device *netdev)
+{
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+       if (mgp->csum_flag)
+               return 1;
+       else
+               return 0;
+}
+
+static int myri10ge_set_rx_csum(struct net_device *netdev, u32 csum_enabled)
+{
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+       if (csum_enabled)
+               mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
+       else
+               mgp->csum_flag = 0;
+       return 0;
+}
+
+static const char myri10ge_gstrings_stats[][ETH_GSTRING_LEN] = {
+       "rx_packets", "tx_packets", "rx_bytes", "tx_bytes", "rx_errors",
+       "tx_errors", "rx_dropped", "tx_dropped", "multicast", "collisions",
+       "rx_length_errors", "rx_over_errors", "rx_crc_errors",
+       "rx_frame_errors", "rx_fifo_errors", "rx_missed_errors",
+       "tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors",
+       "tx_heartbeat_errors", "tx_window_errors",
+       /* device-specific stats */
+       "read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs",
+       "serial_number", "tx_pkt_start", "tx_pkt_done",
+       "tx_req", "tx_done", "rx_small_cnt", "rx_big_cnt",
+       "wake_queue", "stop_queue", "watchdog_resets", "tx_linearized",
+       "link_up", "dropped_link_overflow", "dropped_link_error_or_filtered",
+       "dropped_runt", "dropped_overrun", "dropped_no_small_buffer",
+       "dropped_no_big_buffer"
+};
+
+#define MYRI10GE_NET_STATS_LEN      21
+#define MYRI10GE_STATS_LEN  sizeof(myri10ge_gstrings_stats) / ETH_GSTRING_LEN
+
+static void
+myri10ge_get_strings(struct net_device *netdev, u32 stringset, u8 * data)
+{
+       switch (stringset) {
+       case ETH_SS_STATS:
+               memcpy(data, *myri10ge_gstrings_stats,
+                      sizeof(myri10ge_gstrings_stats));
+               break;
+       }
+}
+
+static int myri10ge_get_stats_count(struct net_device *netdev)
+{
+       return MYRI10GE_STATS_LEN;
+}
+
+static void
+myri10ge_get_ethtool_stats(struct net_device *netdev,
+                          struct ethtool_stats *stats, u64 * data)
+{
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+       int i;
+
+       for (i = 0; i < MYRI10GE_NET_STATS_LEN; i++)
+               data[i] = ((unsigned long *)&mgp->stats)[i];
+
+       data[i++] = (unsigned int)mgp->read_dma;
+       data[i++] = (unsigned int)mgp->write_dma;
+       data[i++] = (unsigned int)mgp->read_write_dma;
+       data[i++] = (unsigned int)mgp->serial_number;
+       data[i++] = (unsigned int)mgp->tx.pkt_start;
+       data[i++] = (unsigned int)mgp->tx.pkt_done;
+       data[i++] = (unsigned int)mgp->tx.req;
+       data[i++] = (unsigned int)mgp->tx.done;
+       data[i++] = (unsigned int)mgp->rx_small.cnt;
+       data[i++] = (unsigned int)mgp->rx_big.cnt;
+       data[i++] = (unsigned int)mgp->wake_queue;
+       data[i++] = (unsigned int)mgp->stop_queue;
+       data[i++] = (unsigned int)mgp->watchdog_resets;
+       data[i++] = (unsigned int)mgp->tx_linearized;
+       data[i++] = (unsigned int)ntohl(mgp->fw_stats->link_up);
+       data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_link_overflow);
+       data[i++] =
+           (unsigned int)ntohl(mgp->fw_stats->dropped_link_error_or_filtered);
+       data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_runt);
+       data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_overrun);
+       data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_no_small_buffer);
+       data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_no_big_buffer);
+}
+
+static struct ethtool_ops myri10ge_ethtool_ops = {
+       .get_settings = myri10ge_get_settings,
+       .get_drvinfo = myri10ge_get_drvinfo,
+       .get_coalesce = myri10ge_get_coalesce,
+       .set_coalesce = myri10ge_set_coalesce,
+       .get_pauseparam = myri10ge_get_pauseparam,
+       .set_pauseparam = myri10ge_set_pauseparam,
+       .get_ringparam = myri10ge_get_ringparam,
+       .get_rx_csum = myri10ge_get_rx_csum,
+       .set_rx_csum = myri10ge_set_rx_csum,
+       .get_tx_csum = ethtool_op_get_tx_csum,
+       .set_tx_csum = ethtool_op_set_tx_csum,
+       .get_sg = ethtool_op_get_sg,
+       .set_sg = ethtool_op_set_sg,
+#ifdef NETIF_F_TSO
+       .get_tso = ethtool_op_get_tso,
+       .set_tso = ethtool_op_set_tso,
+#endif
+       .get_strings = myri10ge_get_strings,
+       .get_stats_count = myri10ge_get_stats_count,
+       .get_ethtool_stats = myri10ge_get_ethtool_stats
+};
+
+static int myri10ge_allocate_rings(struct net_device *dev)
+{
+       struct myri10ge_priv *mgp;
+       struct myri10ge_cmd cmd;
+       int tx_ring_size, rx_ring_size;
+       int tx_ring_entries, rx_ring_entries;
+       int i, status;
+       size_t bytes;
+
+       mgp = netdev_priv(dev);
+
+       /* get ring sizes */
+
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_RING_SIZE, &cmd, 0);
+       tx_ring_size = cmd.data0;
+       status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0);
+       rx_ring_size = cmd.data0;
+
+       tx_ring_entries = tx_ring_size / sizeof(struct mcp_kreq_ether_send);
+       rx_ring_entries = rx_ring_size / sizeof(struct mcp_dma_addr);
+       mgp->tx.mask = tx_ring_entries - 1;
+       mgp->rx_small.mask = mgp->rx_big.mask = rx_ring_entries - 1;
+
+       /* allocate the host shadow rings */
+
+       bytes = 8 + (MYRI10GE_MAX_SEND_DESC_TSO + 4)
+           * sizeof(*mgp->tx.req_list);
+       mgp->tx.req_bytes = kzalloc(bytes, GFP_KERNEL);
+       if (mgp->tx.req_bytes == NULL)
+               goto abort_with_nothing;
+
+       /* ensure req_list entries are aligned to 8 bytes */
+       mgp->tx.req_list = (struct mcp_kreq_ether_send *)
+           ALIGN((unsigned long)mgp->tx.req_bytes, 8);
+
+       bytes = rx_ring_entries * sizeof(*mgp->rx_small.shadow);
+       mgp->rx_small.shadow = kzalloc(bytes, GFP_KERNEL);
+       if (mgp->rx_small.shadow == NULL)
+               goto abort_with_tx_req_bytes;
+
+       bytes = rx_ring_entries * sizeof(*mgp->rx_big.shadow);
+       mgp->rx_big.shadow = kzalloc(bytes, GFP_KERNEL);
+       if (mgp->rx_big.shadow == NULL)
+               goto abort_with_rx_small_shadow;
+
+       /* allocate the host info rings */
+
+       bytes = tx_ring_entries * sizeof(*mgp->tx.info);
+       mgp->tx.info = kzalloc(bytes, GFP_KERNEL);
+       if (mgp->tx.info == NULL)
+               goto abort_with_rx_big_shadow;
+
+       bytes = rx_ring_entries * sizeof(*mgp->rx_small.info);
+       mgp->rx_small.info = kzalloc(bytes, GFP_KERNEL);
+       if (mgp->rx_small.info == NULL)
+               goto abort_with_tx_info;
+
+       bytes = rx_ring_entries * sizeof(*mgp->rx_big.info);
+       mgp->rx_big.info = kzalloc(bytes, GFP_KERNEL);
+       if (mgp->rx_big.info == NULL)
+               goto abort_with_rx_small_info;
+
+       /* Fill the receive rings */
+
+       for (i = 0; i <= mgp->rx_small.mask; i++) {
+               status = myri10ge_getbuf(&mgp->rx_small, mgp->pdev,
+                                        mgp->small_bytes, i);
+               if (status) {
+                       printk(KERN_ERR
+                              "myri10ge: %s: alloced only %d small bufs\n",
+                              dev->name, i);
+                       goto abort_with_rx_small_ring;
+               }
+       }
+
+       for (i = 0; i <= mgp->rx_big.mask; i++) {
+               status =
+                   myri10ge_getbuf(&mgp->rx_big, mgp->pdev,
+                                   dev->mtu + ETH_HLEN, i);
+               if (status) {
+                       printk(KERN_ERR
+                              "myri10ge: %s: alloced only %d big bufs\n",
+                              dev->name, i);
+                       goto abort_with_rx_big_ring;
+               }
+       }
+
+       return 0;
+
+abort_with_rx_big_ring:
+       for (i = 0; i <= mgp->rx_big.mask; i++) {
+               if (mgp->rx_big.info[i].skb != NULL)
+                       dev_kfree_skb_any(mgp->rx_big.info[i].skb);
+               if (pci_unmap_len(&mgp->rx_big.info[i], len))
+                       pci_unmap_single(mgp->pdev,
+                                        pci_unmap_addr(&mgp->rx_big.info[i],
+                                                       bus),
+                                        pci_unmap_len(&mgp->rx_big.info[i],
+                                                      len),
+                                        PCI_DMA_FROMDEVICE);
+       }
+
+abort_with_rx_small_ring:
+       for (i = 0; i <= mgp->rx_small.mask; i++) {
+               if (mgp->rx_small.info[i].skb != NULL)
+                       dev_kfree_skb_any(mgp->rx_small.info[i].skb);
+               if (pci_unmap_len(&mgp->rx_small.info[i], len))
+                       pci_unmap_single(mgp->pdev,
+                                        pci_unmap_addr(&mgp->rx_small.info[i],
+                                                       bus),
+                                        pci_unmap_len(&mgp->rx_small.info[i],
+                                                      len),
+                                        PCI_DMA_FROMDEVICE);
+       }
+       kfree(mgp->rx_big.info);
+
+abort_with_rx_small_info:
+       kfree(mgp->rx_small.info);
+
+abort_with_tx_info:
+       kfree(mgp->tx.info);
+
+abort_with_rx_big_shadow:
+       kfree(mgp->rx_big.shadow);
+
+abort_with_rx_small_shadow:
+       kfree(mgp->rx_small.shadow);
+
+abort_with_tx_req_bytes:
+       kfree(mgp->tx.req_bytes);
+       mgp->tx.req_bytes = NULL;
+       mgp->tx.req_list = NULL;
+
+abort_with_nothing:
+       return status;
+}
+
+static void myri10ge_free_rings(struct net_device *dev)
+{
+       struct myri10ge_priv *mgp;
+       struct sk_buff *skb;
+       struct myri10ge_tx_buf *tx;
+       int i, len, idx;
+
+       mgp = netdev_priv(dev);
+
+       for (i = 0; i <= mgp->rx_big.mask; i++) {
+               if (mgp->rx_big.info[i].skb != NULL)
+                       dev_kfree_skb_any(mgp->rx_big.info[i].skb);
+               if (pci_unmap_len(&mgp->rx_big.info[i], len))
+                       pci_unmap_single(mgp->pdev,
+                                        pci_unmap_addr(&mgp->rx_big.info[i],
+                                                       bus),
+                                        pci_unmap_len(&mgp->rx_big.info[i],
+                                                      len),
+                                        PCI_DMA_FROMDEVICE);
+       }
+
+       for (i = 0; i <= mgp->rx_small.mask; i++) {
+               if (mgp->rx_small.info[i].skb != NULL)
+                       dev_kfree_skb_any(mgp->rx_small.info[i].skb);
+               if (pci_unmap_len(&mgp->rx_small.info[i], len))
+                       pci_unmap_single(mgp->pdev,
+                                        pci_unmap_addr(&mgp->rx_small.info[i],
+                                                       bus),
+                                        pci_unmap_len(&mgp->rx_small.info[i],
+                                                      len),
+                                        PCI_DMA_FROMDEVICE);
+       }
+
+       tx = &mgp->tx;
+       while (tx->done != tx->req) {
+               idx = tx->done & tx->mask;
+               skb = tx->info[idx].skb;
+
+               /* Mark as free */
+               tx->info[idx].skb = NULL;
+               tx->done++;
+               len = pci_unmap_len(&tx->info[idx], len);
+               pci_unmap_len_set(&tx->info[idx], len, 0);
+               if (skb) {
+                       mgp->stats.tx_dropped++;
+                       dev_kfree_skb_any(skb);
+                       if (len)
+                               pci_unmap_single(mgp->pdev,
+                                                pci_unmap_addr(&tx->info[idx],
+                                                               bus), len,
+                                                PCI_DMA_TODEVICE);
+               } else {
+                       if (len)
+                               pci_unmap_page(mgp->pdev,
+                                              pci_unmap_addr(&tx->info[idx],
+                                                             bus), len,
+                                              PCI_DMA_TODEVICE);
+               }
+       }
+       kfree(mgp->rx_big.info);
+
+       kfree(mgp->rx_small.info);
+
+       kfree(mgp->tx.info);
+
+       kfree(mgp->rx_big.shadow);
+
+       kfree(mgp->rx_small.shadow);
+
+       kfree(mgp->tx.req_bytes);
+       mgp->tx.req_bytes = NULL;
+       mgp->tx.req_list = NULL;
+}
+
+static int myri10ge_open(struct net_device *dev)
+{
+       struct myri10ge_priv *mgp;
+       struct myri10ge_cmd cmd;
+       int status, big_pow2;
+
+       mgp = netdev_priv(dev);
+
+       if (mgp->running != MYRI10GE_ETH_STOPPED)
+               return -EBUSY;
+
+       mgp->running = MYRI10GE_ETH_STARTING;
+       status = myri10ge_reset(mgp);
+       if (status != 0) {
+               printk(KERN_ERR "myri10ge: %s: failed reset\n", dev->name);
+               mgp->running = MYRI10GE_ETH_STOPPED;
+               return -ENXIO;
+       }
+
+       /* decide what small buffer size to use.  For good TCP rx
+        * performance, it is important to not receive 1514 byte
+        * frames into jumbo buffers, as it confuses the socket buffer
+        * accounting code, leading to drops and erratic performance.
+        */
+
+       if (dev->mtu <= ETH_DATA_LEN)
+               mgp->small_bytes = 128; /* enough for a TCP header */
+       else
+               mgp->small_bytes = ETH_FRAME_LEN;       /* enough for an ETH_DATA_LEN frame */
+
+       /* Override the small buffer size? */
+       if (myri10ge_small_bytes > 0)
+               mgp->small_bytes = myri10ge_small_bytes;
+
+       /* If the user sets an obscenely small MTU, adjust the small
+        * bytes down to nearly nothing */
+       if (mgp->small_bytes >= (dev->mtu + ETH_HLEN))
+               mgp->small_bytes = 64;
+
+       /* get the lanai pointers to the send and receive rings */
+
+       status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET, &cmd, 0);
+       mgp->tx.lanai =
+           (struct mcp_kreq_ether_send __iomem *)(mgp->sram + cmd.data0);
+
+       status |=
+           myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SMALL_RX_OFFSET, &cmd, 0);
+       mgp->rx_small.lanai =
+           (struct mcp_kreq_ether_recv __iomem *)(mgp->sram + cmd.data0);
+
+       status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_BIG_RX_OFFSET, &cmd, 0);
+       mgp->rx_big.lanai =
+           (struct mcp_kreq_ether_recv __iomem *)(mgp->sram + cmd.data0);
+
+       if (status != 0) {
+               printk(KERN_ERR
+                      "myri10ge: %s: failed to get ring sizes or locations\n",
+                      dev->name);
+               mgp->running = MYRI10GE_ETH_STOPPED;
+               return -ENXIO;
+       }
+
+       if (mgp->mtrr >= 0) {
+               mgp->tx.wc_fifo = (u8 __iomem *) mgp->sram + 0x200000;
+               mgp->rx_small.wc_fifo = (u8 __iomem *) mgp->sram + 0x300000;
+               mgp->rx_big.wc_fifo = (u8 __iomem *) mgp->sram + 0x340000;
+       } else {
+               mgp->tx.wc_fifo = NULL;
+               mgp->rx_small.wc_fifo = NULL;
+               mgp->rx_big.wc_fifo = NULL;
+       }
+
+       status = myri10ge_allocate_rings(dev);
+       if (status != 0)
+               goto abort_with_nothing;
+
+       /* Firmware needs the big buff size as a power of 2.  Lie and
+        * tell him the buffer is larger, because we only use 1
+        * buffer/pkt, and the mtu will prevent overruns.
+        */
+       big_pow2 = dev->mtu + ETH_HLEN + MXGEFW_PAD;
+       while ((big_pow2 & (big_pow2 - 1)) != 0)
+               big_pow2++;
+
+       /* now give firmware buffers sizes, and MTU */
+       cmd.data0 = dev->mtu + ETH_HLEN + VLAN_HLEN;
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_MTU, &cmd, 0);
+       cmd.data0 = mgp->small_bytes;
+       status |=
+           myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_SMALL_BUFFER_SIZE, &cmd, 0);
+       cmd.data0 = big_pow2;
+       status |=
+           myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_BIG_BUFFER_SIZE, &cmd, 0);
+       if (status) {
+               printk(KERN_ERR "myri10ge: %s: Couldn't set buffer sizes\n",
+                      dev->name);
+               goto abort_with_rings;
+       }
+
+       cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->fw_stats_bus);
+       cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->fw_stats_bus);
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA, &cmd, 0);
+       if (status) {
+               printk(KERN_ERR "myri10ge: %s: Couldn't set stats DMA\n",
+                      dev->name);
+               goto abort_with_rings;
+       }
+
+       mgp->link_state = -1;
+       mgp->rdma_tags_available = 15;
+
+       netif_poll_enable(mgp->dev);    /* must happen prior to any irq */
+
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_UP, &cmd, 0);
+       if (status) {
+               printk(KERN_ERR "myri10ge: %s: Couldn't bring up link\n",
+                      dev->name);
+               goto abort_with_rings;
+       }
+
+       mgp->wake_queue = 0;
+       mgp->stop_queue = 0;
+       mgp->running = MYRI10GE_ETH_RUNNING;
+       mgp->watchdog_timer.expires = jiffies + myri10ge_watchdog_timeout * HZ;
+       add_timer(&mgp->watchdog_timer);
+       netif_wake_queue(dev);
+       return 0;
+
+abort_with_rings:
+       myri10ge_free_rings(dev);
+
+abort_with_nothing:
+       mgp->running = MYRI10GE_ETH_STOPPED;
+       return -ENOMEM;
+}
+
+static int myri10ge_close(struct net_device *dev)
+{
+       struct myri10ge_priv *mgp;
+       struct myri10ge_cmd cmd;
+       int status, old_down_cnt;
+
+       mgp = netdev_priv(dev);
+
+       if (mgp->running != MYRI10GE_ETH_RUNNING)
+               return 0;
+
+       if (mgp->tx.req_bytes == NULL)
+               return 0;
+
+       del_timer_sync(&mgp->watchdog_timer);
+       mgp->running = MYRI10GE_ETH_STOPPING;
+       netif_poll_disable(mgp->dev);
+       netif_carrier_off(dev);
+       netif_stop_queue(dev);
+       old_down_cnt = mgp->down_cnt;
+       mb();
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_DOWN, &cmd, 0);
+       if (status)
+               printk(KERN_ERR "myri10ge: %s: Couldn't bring down link\n",
+                      dev->name);
+
+       wait_event_timeout(mgp->down_wq, old_down_cnt != mgp->down_cnt, HZ);
+       if (old_down_cnt == mgp->down_cnt)
+               printk(KERN_ERR "myri10ge: %s never got down irq\n", dev->name);
+
+       netif_tx_disable(dev);
+
+       myri10ge_free_rings(dev);
+
+       mgp->running = MYRI10GE_ETH_STOPPED;
+       return 0;
+}
+
+/* copy an array of struct mcp_kreq_ether_send's to the mcp.  Copy
+ * backwards one at a time and handle ring wraps */
+
+static inline void
+myri10ge_submit_req_backwards(struct myri10ge_tx_buf *tx,
+                             struct mcp_kreq_ether_send *src, int cnt)
+{
+       int idx, starting_slot;
+       starting_slot = tx->req;
+       while (cnt > 1) {
+               cnt--;
+               idx = (starting_slot + cnt) & tx->mask;
+               myri10ge_pio_copy(&tx->lanai[idx], &src[cnt], sizeof(*src));
+               mb();
+       }
+}
+
+/*
+ * copy an array of struct mcp_kreq_ether_send's to the mcp.  Copy
+ * at most 32 bytes at a time, so as to avoid involving the software
+ * pio handler in the nic.   We re-write the first segment's flags
+ * to mark them valid only after writing the entire chain.
+ */
+
+static inline void
+myri10ge_submit_req(struct myri10ge_tx_buf *tx, struct mcp_kreq_ether_send *src,
+                   int cnt)
+{
+       int idx, i;
+       struct mcp_kreq_ether_send __iomem *dstp, *dst;
+       struct mcp_kreq_ether_send *srcp;
+       u8 last_flags;
+
+       idx = tx->req & tx->mask;
+
+       last_flags = src->flags;
+       src->flags = 0;
+       mb();
+       dst = dstp = &tx->lanai[idx];
+       srcp = src;
+
+       if ((idx + cnt) < tx->mask) {
+               for (i = 0; i < (cnt - 1); i += 2) {
+                       myri10ge_pio_copy(dstp, srcp, 2 * sizeof(*src));
+                       mb();   /* force write every 32 bytes */
+                       srcp += 2;
+                       dstp += 2;
+               }
+       } else {
+               /* submit all but the first request, and ensure
+                * that it is submitted below */
+               myri10ge_submit_req_backwards(tx, src, cnt);
+               i = 0;
+       }
+       if (i < cnt) {
+               /* submit the first request */
+               myri10ge_pio_copy(dstp, srcp, sizeof(*src));
+               mb();           /* barrier before setting valid flag */
+       }
+
+       /* re-write the last 32-bits with the valid flags */
+       src->flags = last_flags;
+       __raw_writel(*((u32 *) src + 3), (u32 __iomem *) dst + 3);
+       tx->req += cnt;
+       mb();
+}
+
+static inline void
+myri10ge_submit_req_wc(struct myri10ge_tx_buf *tx,
+                      struct mcp_kreq_ether_send *src, int cnt)
+{
+       tx->req += cnt;
+       mb();
+       while (cnt >= 4) {
+               myri10ge_pio_copy(tx->wc_fifo, src, 64);
+               mb();
+               src += 4;
+               cnt -= 4;
+       }
+       if (cnt > 0) {
+               /* pad it to 64 bytes.  The src is 64 bytes bigger than it
+                * needs to be so that we don't overrun it */
+               myri10ge_pio_copy(tx->wc_fifo + (cnt << 18), src, 64);
+               mb();
+       }
+}
+
+/*
+ * Transmit a packet.  We need to split the packet so that a single
+ * segment does not cross myri10ge->tx.boundary, so this makes segment
+ * counting tricky.  So rather than try to count segments up front, we
+ * just give up if there are too few segments to hold a reasonably
+ * fragmented packet currently available.  If we run
+ * out of segments while preparing a packet for DMA, we just linearize
+ * it and try again.
+ */
+
+static int myri10ge_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct myri10ge_priv *mgp = netdev_priv(dev);
+       struct mcp_kreq_ether_send *req;
+       struct myri10ge_tx_buf *tx = &mgp->tx;
+       struct skb_frag_struct *frag;
+       dma_addr_t bus;
+       u32 low, high_swapped;
+       unsigned int len;
+       int idx, last_idx, avail, frag_cnt, frag_idx, count, mss, max_segments;
+       u16 pseudo_hdr_offset, cksum_offset;
+       int cum_len, seglen, boundary, rdma_count;
+       u8 flags, odd_flag;
+
+again:
+       req = tx->req_list;
+       avail = tx->mask - 1 - (tx->req - tx->done);
+
+       mss = 0;
+       max_segments = MXGEFW_MAX_SEND_DESC;
+
+#ifdef NETIF_F_TSO
+       if (skb->len > (dev->mtu + ETH_HLEN)) {
+               mss = skb_shinfo(skb)->tso_size;
+               if (mss != 0)
+                       max_segments = MYRI10GE_MAX_SEND_DESC_TSO;
+       }
+#endif                         /*NETIF_F_TSO */
+
+       if ((unlikely(avail < max_segments))) {
+               /* we are out of transmit resources */
+               mgp->stop_queue++;
+               netif_stop_queue(dev);
+               return 1;
+       }
+
+       /* Setup checksum offloading, if needed */
+       cksum_offset = 0;
+       pseudo_hdr_offset = 0;
+       odd_flag = 0;
+       flags = (MXGEFW_FLAGS_NO_TSO | MXGEFW_FLAGS_FIRST);
+       if (likely(skb->ip_summed == CHECKSUM_HW)) {
+               cksum_offset = (skb->h.raw - skb->data);
+               pseudo_hdr_offset = (skb->h.raw + skb->csum) - skb->data;
+               /* If the headers are excessively large, then we must
+                * fall back to a software checksum */
+               if (unlikely(cksum_offset > 255 || pseudo_hdr_offset > 127)) {
+                       if (skb_checksum_help(skb, 0))
+                               goto drop;
+                       cksum_offset = 0;
+                       pseudo_hdr_offset = 0;
+               } else {
+                       pseudo_hdr_offset = htons(pseudo_hdr_offset);
+                       odd_flag = MXGEFW_FLAGS_ALIGN_ODD;
+                       flags |= MXGEFW_FLAGS_CKSUM;
+               }
+       }
+
+       cum_len = 0;
+
+#ifdef NETIF_F_TSO
+       if (mss) {              /* TSO */
+               /* this removes any CKSUM flag from before */
+               flags = (MXGEFW_FLAGS_TSO_HDR | MXGEFW_FLAGS_FIRST);
+
+               /* negative cum_len signifies to the
+                * send loop that we are still in the
+                * header portion of the TSO packet.
+                * TSO header must be at most 134 bytes long */
+               cum_len = -((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
+
+               /* for TSO, pseudo_hdr_offset holds mss.
+                * The firmware figures out where to put
+                * the checksum by parsing the header. */
+               pseudo_hdr_offset = htons(mss);
+       } else
+#endif                         /*NETIF_F_TSO */
+               /* Mark small packets, and pad out tiny packets */
+       if (skb->len <= MXGEFW_SEND_SMALL_SIZE) {
+               flags |= MXGEFW_FLAGS_SMALL;
+
+               /* pad frames to at least ETH_ZLEN bytes */
+               if (unlikely(skb->len < ETH_ZLEN)) {
+                       skb = skb_padto(skb, ETH_ZLEN);
+                       if (skb == NULL) {
+                               /* The packet is gone, so we must
+                                * return 0 */
+                               mgp->stats.tx_dropped += 1;
+                               return 0;
+                       }
+                       /* adjust the len to account for the zero pad
+                        * so that the nic can know how long it is */
+                       skb->len = ETH_ZLEN;
+               }
+       }
+
+       /* map the skb for DMA */
+       len = skb->len - skb->data_len;
+       idx = tx->req & tx->mask;
+       tx->info[idx].skb = skb;
+       bus = pci_map_single(mgp->pdev, skb->data, len, PCI_DMA_TODEVICE);
+       pci_unmap_addr_set(&tx->info[idx], bus, bus);
+       pci_unmap_len_set(&tx->info[idx], len, len);
+
+       frag_cnt = skb_shinfo(skb)->nr_frags;
+       frag_idx = 0;
+       count = 0;
+       rdma_count = 0;
+
+       /* "rdma_count" is the number of RDMAs belonging to the
+        * current packet BEFORE the current send request. For
+        * non-TSO packets, this is equal to "count".
+        * For TSO packets, rdma_count needs to be reset
+        * to 0 after a segment cut.
+        *
+        * The rdma_count field of the send request is
+        * the number of RDMAs of the packet starting at
+        * that request. For TSO send requests with one ore more cuts
+        * in the middle, this is the number of RDMAs starting
+        * after the last cut in the request. All previous
+        * segments before the last cut implicitly have 1 RDMA.
+        *
+        * Since the number of RDMAs is not known beforehand,
+        * it must be filled-in retroactively - after each
+        * segmentation cut or at the end of the entire packet.
+        */
+
+       while (1) {
+               /* Break the SKB or Fragment up into pieces which
+                * do not cross mgp->tx.boundary */
+               low = MYRI10GE_LOWPART_TO_U32(bus);
+               high_swapped = htonl(MYRI10GE_HIGHPART_TO_U32(bus));
+               while (len) {
+                       u8 flags_next;
+                       int cum_len_next;
+
+                       if (unlikely(count == max_segments))
+                               goto abort_linearize;
+
+                       boundary = (low + tx->boundary) & ~(tx->boundary - 1);
+                       seglen = boundary - low;
+                       if (seglen > len)
+                               seglen = len;
+                       flags_next = flags & ~MXGEFW_FLAGS_FIRST;
+                       cum_len_next = cum_len + seglen;
+#ifdef NETIF_F_TSO
+                       if (mss) {      /* TSO */
+                               (req - rdma_count)->rdma_count = rdma_count + 1;
+
+                               if (likely(cum_len >= 0)) {     /* payload */
+                                       int next_is_first, chop;
+
+                                       chop = (cum_len_next > mss);
+                                       cum_len_next = cum_len_next % mss;
+                                       next_is_first = (cum_len_next == 0);
+                                       flags |= chop * MXGEFW_FLAGS_TSO_CHOP;
+                                       flags_next |= next_is_first *
+                                           MXGEFW_FLAGS_FIRST;
+                                       rdma_count |= -(chop | next_is_first);
+                                       rdma_count += chop & !next_is_first;
+                               } else if (likely(cum_len_next >= 0)) { /* header ends */
+                                       int small;
+
+                                       rdma_count = -1;
+                                       cum_len_next = 0;
+                                       seglen = -cum_len;
+                                       small = (mss <= MXGEFW_SEND_SMALL_SIZE);
+                                       flags_next = MXGEFW_FLAGS_TSO_PLD |
+                                           MXGEFW_FLAGS_FIRST |
+                                           (small * MXGEFW_FLAGS_SMALL);
+                               }
+                       }
+#endif                         /* NETIF_F_TSO */
+                       req->addr_high = high_swapped;
+                       req->addr_low = htonl(low);
+                       req->pseudo_hdr_offset = pseudo_hdr_offset;
+                       req->pad = 0;   /* complete solid 16-byte block; does this matter? */
+                       req->rdma_count = 1;
+                       req->length = htons(seglen);
+                       req->cksum_offset = cksum_offset;
+                       req->flags = flags | ((cum_len & 1) * odd_flag);
+
+                       low += seglen;
+                       len -= seglen;
+                       cum_len = cum_len_next;
+                       flags = flags_next;
+                       req++;
+                       count++;
+                       rdma_count++;
+                       if (unlikely(cksum_offset > seglen))
+                               cksum_offset -= seglen;
+                       else
+                               cksum_offset = 0;
+               }
+               if (frag_idx == frag_cnt)
+                       break;
+
+               /* map next fragment for DMA */
+               idx = (count + tx->req) & tx->mask;
+               frag = &skb_shinfo(skb)->frags[frag_idx];
+               frag_idx++;
+               len = frag->size;
+               bus = pci_map_page(mgp->pdev, frag->page, frag->page_offset,
+                                  len, PCI_DMA_TODEVICE);
+               pci_unmap_addr_set(&tx->info[idx], bus, bus);
+               pci_unmap_len_set(&tx->info[idx], len, len);
+       }
+
+       (req - rdma_count)->rdma_count = rdma_count;
+#ifdef NETIF_F_TSO
+       if (mss)
+               do {
+                       req--;
+                       req->flags |= MXGEFW_FLAGS_TSO_LAST;
+               } while (!(req->flags & (MXGEFW_FLAGS_TSO_CHOP |
+                                        MXGEFW_FLAGS_FIRST)));
+#endif
+       idx = ((count - 1) + tx->req) & tx->mask;
+       tx->info[idx].last = 1;
+       if (tx->wc_fifo == NULL)
+               myri10ge_submit_req(tx, tx->req_list, count);
+       else
+               myri10ge_submit_req_wc(tx, tx->req_list, count);
+       tx->pkt_start++;
+       if ((avail - count) < MXGEFW_MAX_SEND_DESC) {
+               mgp->stop_queue++;
+               netif_stop_queue(dev);
+       }
+       dev->trans_start = jiffies;
+       return 0;
+
+abort_linearize:
+       /* Free any DMA resources we've alloced and clear out the skb
+        * slot so as to not trip up assertions, and to avoid a
+        * double-free if linearizing fails */
+
+       last_idx = (idx + 1) & tx->mask;
+       idx = tx->req & tx->mask;
+       tx->info[idx].skb = NULL;
+       do {
+               len = pci_unmap_len(&tx->info[idx], len);
+               if (len) {
+                       if (tx->info[idx].skb != NULL)
+                               pci_unmap_single(mgp->pdev,
+                                                pci_unmap_addr(&tx->info[idx],
+                                                               bus), len,
+                                                PCI_DMA_TODEVICE);
+                       else
+                               pci_unmap_page(mgp->pdev,
+                                              pci_unmap_addr(&tx->info[idx],
+                                                             bus), len,
+                                              PCI_DMA_TODEVICE);
+                       pci_unmap_len_set(&tx->info[idx], len, 0);
+                       tx->info[idx].skb = NULL;
+               }
+               idx = (idx + 1) & tx->mask;
+       } while (idx != last_idx);
+       if (skb_shinfo(skb)->tso_size) {
+               printk(KERN_ERR
+                      "myri10ge: %s: TSO but wanted to linearize?!?!?\n",
+                      mgp->dev->name);
+               goto drop;
+       }
+
+       if (skb_linearize(skb, GFP_ATOMIC))
+               goto drop;
+
+       mgp->tx_linearized++;
+       goto again;
+
+drop:
+       dev_kfree_skb_any(skb);
+       mgp->stats.tx_dropped += 1;
+       return 0;
+
+}
+
+static struct net_device_stats *myri10ge_get_stats(struct net_device *dev)
+{
+       struct myri10ge_priv *mgp = netdev_priv(dev);
+       return &mgp->stats;
+}
+
+static void myri10ge_set_multicast_list(struct net_device *dev)
+{
+       /* can be called from atomic contexts,
+        * pass 1 to force atomicity in myri10ge_send_cmd() */
+       myri10ge_change_promisc(netdev_priv(dev), dev->flags & IFF_PROMISC, 1);
+}
+
+static int myri10ge_set_mac_address(struct net_device *dev, void *addr)
+{
+       struct sockaddr *sa = addr;
+       struct myri10ge_priv *mgp = netdev_priv(dev);
+       int status;
+
+       if (!is_valid_ether_addr(sa->sa_data))
+               return -EADDRNOTAVAIL;
+
+       status = myri10ge_update_mac_address(mgp, sa->sa_data);
+       if (status != 0) {
+               printk(KERN_ERR
+                      "myri10ge: %s: changing mac address failed with %d\n",
+                      dev->name, status);
+               return status;
+       }
+
+       /* change the dev structure */
+       memcpy(dev->dev_addr, sa->sa_data, 6);
+       return 0;
+}
+
+static int myri10ge_change_mtu(struct net_device *dev, int new_mtu)
+{
+       struct myri10ge_priv *mgp = netdev_priv(dev);
+       int error = 0;
+
+       if ((new_mtu < 68) || (ETH_HLEN + new_mtu > MYRI10GE_MAX_ETHER_MTU)) {
+               printk(KERN_ERR "myri10ge: %s: new mtu (%d) is not valid\n",
+                      dev->name, new_mtu);
+               return -EINVAL;
+       }
+       printk(KERN_INFO "%s: changing mtu from %d to %d\n",
+              dev->name, dev->mtu, new_mtu);
+       if (mgp->running) {
+               /* if we change the mtu on an active device, we must
+                * reset the device so the firmware sees the change */
+               myri10ge_close(dev);
+               dev->mtu = new_mtu;
+               myri10ge_open(dev);
+       } else
+               dev->mtu = new_mtu;
+
+       return error;
+}
+
+/*
+ * Enable ECRC to align PCI-E Completion packets on an 8-byte boundary.
+ * Only do it if the bridge is a root port since we don't want to disturb
+ * any other device, except if forced with myri10ge_ecrc_enable > 1.
+ */
+
+#define PCI_DEVICE_ID_NVIDIA_NFORCE_CK804_PCIE 0x005d
+
+static void myri10ge_enable_ecrc(struct myri10ge_priv *mgp)
+{
+       struct pci_dev *bridge = mgp->pdev->bus->self;
+       struct device *dev = &mgp->pdev->dev;
+       unsigned cap;
+       unsigned err_cap;
+       u16 val;
+       u8 ext_type;
+       int ret;
+
+       if (!myri10ge_ecrc_enable || !bridge)
+               return;
+
+       /* check that the bridge is a root port */
+       cap = pci_find_capability(bridge, PCI_CAP_ID_EXP);
+       pci_read_config_word(bridge, cap + PCI_CAP_FLAGS, &val);
+       ext_type = (val & PCI_EXP_FLAGS_TYPE) >> 4;
+       if (ext_type != PCI_EXP_TYPE_ROOT_PORT) {
+               if (myri10ge_ecrc_enable > 1) {
+                       struct pci_dev *old_bridge = bridge;
+
+                       /* Walk the hierarchy up to the root port
+                        * where ECRC has to be enabled */
+                       do {
+                               bridge = bridge->bus->self;
+                               if (!bridge) {
+                                       dev_err(dev,
+                                               "Failed to find root port"
+                                               " to force ECRC\n");
+                                       return;
+                               }
+                               cap =
+                                   pci_find_capability(bridge, PCI_CAP_ID_EXP);
+                               pci_read_config_word(bridge,
+                                                    cap + PCI_CAP_FLAGS, &val);
+                               ext_type = (val & PCI_EXP_FLAGS_TYPE) >> 4;
+                       } while (ext_type != PCI_EXP_TYPE_ROOT_PORT);
+
+                       dev_info(dev,
+                                "Forcing ECRC on non-root port %s"
+                                " (enabling on root port %s)\n",
+                                pci_name(old_bridge), pci_name(bridge));
+               } else {
+                       dev_err(dev,
+                               "Not enabling ECRC on non-root port %s\n",
+                               pci_name(bridge));
+                       return;
+               }
+       }
+
+       cap = pci_find_ext_capability(bridge, PCI_EXT_CAP_ID_ERR);
+       /* nvidia ext cap is not always linked in ext cap chain */
+       if (!cap
+           && bridge->vendor == PCI_VENDOR_ID_NVIDIA
+           && bridge->device == PCI_DEVICE_ID_NVIDIA_NFORCE_CK804_PCIE)
+               cap = 0x160;
+
+       if (!cap)
+               return;
+
+       ret = pci_read_config_dword(bridge, cap + PCI_ERR_CAP, &err_cap);
+       if (ret) {
+               dev_err(dev, "failed reading ext-conf-space of %s\n",
+                       pci_name(bridge));
+               dev_err(dev, "\t pci=nommconf in use? "
+                       "or buggy/incomplete/absent ACPI MCFG attr?\n");
+               return;
+       }
+       if (!(err_cap & PCI_ERR_CAP_ECRC_GENC))
+               return;
+
+       err_cap |= PCI_ERR_CAP_ECRC_GENE;
+       pci_write_config_dword(bridge, cap + PCI_ERR_CAP, err_cap);
+       dev_info(dev, "Enabled ECRC on upstream bridge %s\n", pci_name(bridge));
+       mgp->tx.boundary = 4096;
+       mgp->fw_name = myri10ge_fw_aligned;
+}
+
+/*
+ * The Lanai Z8E PCI-E interface achieves higher Read-DMA throughput
+ * when the PCI-E Completion packets are aligned on an 8-byte
+ * boundary.  Some PCI-E chip sets always align Completion packets; on
+ * the ones that do not, the alignment can be enforced by enabling
+ * ECRC generation (if supported).
+ *
+ * When PCI-E Completion packets are not aligned, it is actually more
+ * efficient to limit Read-DMA transactions to 2KB, rather than 4KB.
+ *
+ * If the driver can neither enable ECRC nor verify that it has
+ * already been enabled, then it must use a firmware image which works
+ * around unaligned completion packets (myri10ge_ethp_z8e.dat), and it
+ * should also ensure that it never gives the device a Read-DMA which is
+ * larger than 2KB by setting the tx.boundary to 2KB.  If ECRC is
+ * enabled, then the driver should use the aligned (myri10ge_eth_z8e.dat)
+ * firmware image, and set tx.boundary to 4KB.
+ */
+
+#define PCI_DEVICE_ID_SERVERWORKS_HT2000_PCIE  0x0132
+
+static void myri10ge_select_firmware(struct myri10ge_priv *mgp)
+{
+       struct pci_dev *bridge = mgp->pdev->bus->self;
+
+       mgp->tx.boundary = 2048;
+       mgp->fw_name = myri10ge_fw_unaligned;
+
+       if (myri10ge_force_firmware == 0) {
+               myri10ge_enable_ecrc(mgp);
+
+               /* Check to see if the upstream bridge is known to
+                * provide aligned completions */
+               if (bridge
+                   /* ServerWorks HT2000/HT1000 */
+                   && bridge->vendor == PCI_VENDOR_ID_SERVERWORKS
+                   && bridge->device ==
+                   PCI_DEVICE_ID_SERVERWORKS_HT2000_PCIE) {
+                       dev_info(&mgp->pdev->dev,
+                                "Assuming aligned completions (0x%x:0x%x)\n",
+                                bridge->vendor, bridge->device);
+                       mgp->tx.boundary = 4096;
+                       mgp->fw_name = myri10ge_fw_aligned;
+               }
+       } else {
+               if (myri10ge_force_firmware == 1) {
+                       dev_info(&mgp->pdev->dev,
+                                "Assuming aligned completions (forced)\n");
+                       mgp->tx.boundary = 4096;
+                       mgp->fw_name = myri10ge_fw_aligned;
+               } else {
+                       dev_info(&mgp->pdev->dev,
+                                "Assuming unaligned completions (forced)\n");
+                       mgp->tx.boundary = 2048;
+                       mgp->fw_name = myri10ge_fw_unaligned;
+               }
+       }
+       if (myri10ge_fw_name != NULL) {
+               dev_info(&mgp->pdev->dev, "overriding firmware to %s\n",
+                        myri10ge_fw_name);
+               mgp->fw_name = myri10ge_fw_name;
+       }
+}
+
+static void myri10ge_save_state(struct myri10ge_priv *mgp)
+{
+       struct pci_dev *pdev = mgp->pdev;
+       int cap;
+
+       pci_save_state(pdev);
+       /* now save PCIe and MSI state that Linux will not
+        * save for us */
+       cap = pci_find_capability(pdev, PCI_CAP_ID_EXP);
+       pci_read_config_dword(pdev, cap + PCI_EXP_DEVCTL, &mgp->devctl);
+       cap = pci_find_capability(pdev, PCI_CAP_ID_MSI);
+       pci_read_config_word(pdev, cap + PCI_MSI_FLAGS, &mgp->msi_flags);
+}
+
+static void myri10ge_restore_state(struct myri10ge_priv *mgp)
+{
+       struct pci_dev *pdev = mgp->pdev;
+       int cap;
+
+       /* restore PCIe and MSI state that linux will not */
+       cap = pci_find_capability(pdev, PCI_CAP_ID_EXP);
+       pci_write_config_dword(pdev, cap + PCI_CAP_ID_EXP, mgp->devctl);
+       cap = pci_find_capability(pdev, PCI_CAP_ID_MSI);
+       pci_write_config_word(pdev, cap + PCI_MSI_FLAGS, mgp->msi_flags);
+
+       pci_restore_state(pdev);
+}
+
+#ifdef CONFIG_PM
+
+static int myri10ge_suspend(struct pci_dev *pdev, pm_message_t state)
+{
+       struct myri10ge_priv *mgp;
+       struct net_device *netdev;
+
+       mgp = pci_get_drvdata(pdev);
+       if (mgp == NULL)
+               return -EINVAL;
+       netdev = mgp->dev;
+
+       netif_device_detach(netdev);
+       if (netif_running(netdev)) {
+               printk(KERN_INFO "myri10ge: closing %s\n", netdev->name);
+               rtnl_lock();
+               myri10ge_close(netdev);
+               rtnl_unlock();
+       }
+       myri10ge_dummy_rdma(mgp, 0);
+       free_irq(pdev->irq, mgp);
+       myri10ge_save_state(mgp);
+       pci_disable_device(pdev);
+       pci_set_power_state(pdev, pci_choose_state(pdev, state));
+       return 0;
+}
+
+static int myri10ge_resume(struct pci_dev *pdev)
+{
+       struct myri10ge_priv *mgp;
+       struct net_device *netdev;
+       int status;
+       u16 vendor;
+
+       mgp = pci_get_drvdata(pdev);
+       if (mgp == NULL)
+               return -EINVAL;
+       netdev = mgp->dev;
+       pci_set_power_state(pdev, 0);   /* zeros conf space as a side effect */
+       msleep(5);              /* give card time to respond */
+       pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor);
+       if (vendor == 0xffff) {
+               printk(KERN_ERR "myri10ge: %s: device disappeared!\n",
+                      mgp->dev->name);
+               return -EIO;
+       }
+       myri10ge_restore_state(mgp);
+       pci_enable_device(pdev);
+       pci_set_master(pdev);
+
+       status = request_irq(pdev->irq, myri10ge_intr, SA_SHIRQ,
+                            netdev->name, mgp);
+       if (status != 0) {
+               dev_err(&pdev->dev, "failed to allocate IRQ\n");
+               goto abort_with_msi;
+       }
+
+       myri10ge_reset(mgp);
+       myri10ge_dummy_rdma(mgp, mgp->tx.boundary != 4096);
+
+       /* Save configuration space to be restored if the
+        * nic resets due to a parity error */
+       myri10ge_save_state(mgp);
+
+       if (netif_running(netdev)) {
+               rtnl_lock();
+               myri10ge_open(netdev);
+               rtnl_unlock();
+       }
+       netif_device_attach(netdev);
+
+       return 0;
+
+abort_with_msi:
+       return -EIO;
+
+}
+
+#endif                         /* CONFIG_PM */
+
+static u32 myri10ge_read_reboot(struct myri10ge_priv *mgp)
+{
+       struct pci_dev *pdev = mgp->pdev;
+       int vs = mgp->vendor_specific_offset;
+       u32 reboot;
+
+       /*enter read32 mode */
+       pci_write_config_byte(pdev, vs + 0x10, 0x3);
+
+       /*read REBOOT_STATUS (0xfffffff0) */
+       pci_write_config_dword(pdev, vs + 0x18, 0xfffffff0);
+       pci_read_config_dword(pdev, vs + 0x14, &reboot);
+       return reboot;
+}
+
+/*
+ * This watchdog is used to check whether the board has suffered
+ * from a parity error and needs to be recovered.
+ */
+static void myri10ge_watchdog(void *arg)
+{
+       struct myri10ge_priv *mgp = arg;
+       u32 reboot;
+       int status;
+       u16 cmd, vendor;
+
+       mgp->watchdog_resets++;
+       pci_read_config_word(mgp->pdev, PCI_COMMAND, &cmd);
+       if ((cmd & PCI_COMMAND_MASTER) == 0) {
+               /* Bus master DMA disabled?  Check to see
+                * if the card rebooted due to a parity error
+                * For now, just report it */
+               reboot = myri10ge_read_reboot(mgp);
+               printk(KERN_ERR
+                      "myri10ge: %s: NIC rebooted (0x%x), resetting\n",
+                      mgp->dev->name, reboot);
+               /*
+                * A rebooted nic will come back with config space as
+                * it was after power was applied to PCIe bus.
+                * Attempt to restore config space which was saved
+                * when the driver was loaded, or the last time the
+                * nic was resumed from power saving mode.
+                */
+               myri10ge_restore_state(mgp);
+       } else {
+               /* if we get back -1's from our slot, perhaps somebody
+                * powered off our card.  Don't try to reset it in
+                * this case */
+               if (cmd == 0xffff) {
+                       pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor);
+                       if (vendor == 0xffff) {
+                               printk(KERN_ERR
+                                      "myri10ge: %s: device disappeared!\n",
+                                      mgp->dev->name);
+                               return;
+                       }
+               }
+               /* Perhaps it is a software error.  Try to reset */
+
+               printk(KERN_ERR "myri10ge: %s: device timeout, resetting\n",
+                      mgp->dev->name);
+               printk(KERN_INFO "myri10ge: %s: %d %d %d %d %d\n",
+                      mgp->dev->name, mgp->tx.req, mgp->tx.done,
+                      mgp->tx.pkt_start, mgp->tx.pkt_done,
+                      (int)ntohl(mgp->fw_stats->send_done_count));
+               msleep(2000);
+               printk(KERN_INFO "myri10ge: %s: %d %d %d %d %d\n",
+                      mgp->dev->name, mgp->tx.req, mgp->tx.done,
+                      mgp->tx.pkt_start, mgp->tx.pkt_done,
+                      (int)ntohl(mgp->fw_stats->send_done_count));
+       }
+       rtnl_lock();
+       myri10ge_close(mgp->dev);
+       status = myri10ge_load_firmware(mgp);
+       if (status != 0)
+               printk(KERN_ERR "myri10ge: %s: failed to load firmware\n",
+                      mgp->dev->name);
+       else
+               myri10ge_open(mgp->dev);
+       rtnl_unlock();
+}
+
+/*
+ * We use our own timer routine rather than relying upon
+ * netdev->tx_timeout because we have a very large hardware transmit
+ * queue.  Due to the large queue, the netdev->tx_timeout function
+ * cannot detect a NIC with a parity error in a timely fashion if the
+ * NIC is lightly loaded.
+ */
+static void myri10ge_watchdog_timer(unsigned long arg)
+{
+       struct myri10ge_priv *mgp;
+
+       mgp = (struct myri10ge_priv *)arg;
+       if (mgp->tx.req != mgp->tx.done &&
+           mgp->tx.done == mgp->watchdog_tx_done)
+               /* nic seems like it might be stuck.. */
+               schedule_work(&mgp->watchdog_work);
+       else
+               /* rearm timer */
+               mod_timer(&mgp->watchdog_timer,
+                         jiffies + myri10ge_watchdog_timeout * HZ);
+
+       mgp->watchdog_tx_done = mgp->tx.done;
+}
+
+static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
+{
+       struct net_device *netdev;
+       struct myri10ge_priv *mgp;
+       struct device *dev = &pdev->dev;
+       size_t bytes;
+       int i;
+       int status = -ENXIO;
+       int cap;
+       int dac_enabled;
+       u16 val;
+
+       netdev = alloc_etherdev(sizeof(*mgp));
+       if (netdev == NULL) {
+               dev_err(dev, "Could not allocate ethernet device\n");
+               return -ENOMEM;
+       }
+
+       mgp = netdev_priv(netdev);
+       memset(mgp, 0, sizeof(*mgp));
+       mgp->dev = netdev;
+       mgp->pdev = pdev;
+       mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
+       mgp->pause = myri10ge_flow_control;
+       mgp->intr_coal_delay = myri10ge_intr_coal_delay;
+       init_waitqueue_head(&mgp->down_wq);
+
+       if (pci_enable_device(pdev)) {
+               dev_err(&pdev->dev, "pci_enable_device call failed\n");
+               status = -ENODEV;
+               goto abort_with_netdev;
+       }
+       myri10ge_select_firmware(mgp);
+
+       /* Find the vendor-specific cap so we can check
+        * the reboot register later on */
+       mgp->vendor_specific_offset
+           = pci_find_capability(pdev, PCI_CAP_ID_VNDR);
+
+       /* Set our max read request to 4KB */
+       cap = pci_find_capability(pdev, PCI_CAP_ID_EXP);
+       if (cap < 64) {
+               dev_err(&pdev->dev, "Bad PCI_CAP_ID_EXP location %d\n", cap);
+               goto abort_with_netdev;
+       }
+       status = pci_read_config_word(pdev, cap + PCI_EXP_DEVCTL, &val);
+       if (status != 0) {
+               dev_err(&pdev->dev, "Error %d reading PCI_EXP_DEVCTL\n",
+                       status);
+               goto abort_with_netdev;
+       }
+       val = (val & ~PCI_EXP_DEVCTL_READRQ) | (5 << 12);
+       status = pci_write_config_word(pdev, cap + PCI_EXP_DEVCTL, val);
+       if (status != 0) {
+               dev_err(&pdev->dev, "Error %d writing PCI_EXP_DEVCTL\n",
+                       status);
+               goto abort_with_netdev;
+       }
+
+       pci_set_master(pdev);
+       dac_enabled = 1;
+       status = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
+       if (status != 0) {
+               dac_enabled = 0;
+               dev_err(&pdev->dev,
+                       "64-bit pci address mask was refused, trying 32-bit");
+               status = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
+       }
+       if (status != 0) {
+               dev_err(&pdev->dev, "Error %d setting DMA mask\n", status);
+               goto abort_with_netdev;
+       }
+       mgp->cmd = pci_alloc_consistent(pdev, sizeof(*mgp->cmd), &mgp->cmd_bus);
+       if (mgp->cmd == NULL)
+               goto abort_with_netdev;
+
+       mgp->fw_stats = pci_alloc_consistent(pdev, sizeof(*mgp->fw_stats),
+                                            &mgp->fw_stats_bus);
+       if (mgp->fw_stats == NULL)
+               goto abort_with_cmd;
+
+       mgp->board_span = pci_resource_len(pdev, 0);
+       mgp->iomem_base = pci_resource_start(pdev, 0);
+       mgp->mtrr = -1;
+#ifdef CONFIG_MTRR
+       mgp->mtrr = mtrr_add(mgp->iomem_base, mgp->board_span,
+                            MTRR_TYPE_WRCOMB, 1);
+#endif
+       /* Hack.  need to get rid of these magic numbers */
+       mgp->sram_size =
+           2 * 1024 * 1024 - (2 * (48 * 1024) + (32 * 1024)) - 0x100;
+       if (mgp->sram_size > mgp->board_span) {
+               dev_err(&pdev->dev, "board span %ld bytes too small\n",
+                       mgp->board_span);
+               goto abort_with_wc;
+       }
+       mgp->sram = ioremap(mgp->iomem_base, mgp->board_span);
+       if (mgp->sram == NULL) {
+               dev_err(&pdev->dev, "ioremap failed for %ld bytes at 0x%lx\n",
+                       mgp->board_span, mgp->iomem_base);
+               status = -ENXIO;
+               goto abort_with_wc;
+       }
+       memcpy_fromio(mgp->eeprom_strings,
+                     mgp->sram + mgp->sram_size - MYRI10GE_EEPROM_STRINGS_SIZE,
+                     MYRI10GE_EEPROM_STRINGS_SIZE);
+       memset(mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE - 2, 0, 2);
+       status = myri10ge_read_mac_addr(mgp);
+       if (status)
+               goto abort_with_ioremap;
+
+       for (i = 0; i < ETH_ALEN; i++)
+               netdev->dev_addr[i] = mgp->mac_addr[i];
+
+       /* allocate rx done ring */
+       bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
+       mgp->rx_done.entry =
+           pci_alloc_consistent(pdev, bytes, &mgp->rx_done.bus);
+       if (mgp->rx_done.entry == NULL)
+               goto abort_with_ioremap;
+       memset(mgp->rx_done.entry, 0, bytes);
+
+       status = myri10ge_load_firmware(mgp);
+       if (status != 0) {
+               dev_err(&pdev->dev, "failed to load firmware\n");
+               goto abort_with_rx_done;
+       }
+
+       status = myri10ge_reset(mgp);
+       if (status != 0) {
+               dev_err(&pdev->dev, "failed reset\n");
+               goto abort_with_firmware;
+       }
+
+       if (myri10ge_msi) {
+               status = pci_enable_msi(pdev);
+               if (status != 0)
+                       dev_err(&pdev->dev,
+                               "Error %d setting up MSI; falling back to xPIC\n",
+                               status);
+               else
+                       mgp->msi_enabled = 1;
+       }
+
+       status = request_irq(pdev->irq, myri10ge_intr, SA_SHIRQ,
+                            netdev->name, mgp);
+       if (status != 0) {
+               dev_err(&pdev->dev, "failed to allocate IRQ\n");
+               goto abort_with_firmware;
+       }
+
+       pci_set_drvdata(pdev, mgp);
+       if ((myri10ge_initial_mtu + ETH_HLEN) > MYRI10GE_MAX_ETHER_MTU)
+               myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN;
+       if ((myri10ge_initial_mtu + ETH_HLEN) < 68)
+               myri10ge_initial_mtu = 68;
+       netdev->mtu = myri10ge_initial_mtu;
+       netdev->open = myri10ge_open;
+       netdev->stop = myri10ge_close;
+       netdev->hard_start_xmit = myri10ge_xmit;
+       netdev->get_stats = myri10ge_get_stats;
+       netdev->base_addr = mgp->iomem_base;
+       netdev->irq = pdev->irq;
+       netdev->change_mtu = myri10ge_change_mtu;
+       netdev->set_multicast_list = myri10ge_set_multicast_list;
+       netdev->set_mac_address = myri10ge_set_mac_address;
+       netdev->features = NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_TSO;
+       if (dac_enabled)
+               netdev->features |= NETIF_F_HIGHDMA;
+       netdev->poll = myri10ge_poll;
+       netdev->weight = myri10ge_napi_weight;
+
+       /* Save configuration space to be restored if the
+        * nic resets due to a parity error */
+       myri10ge_save_state(mgp);
+       /* Restore state immediately since pci_save_msi_state disables MSI */
+       myri10ge_restore_state(mgp);
+
+       /* Setup the watchdog timer */
+       setup_timer(&mgp->watchdog_timer, myri10ge_watchdog_timer,
+                   (unsigned long)mgp);
+
+       SET_ETHTOOL_OPS(netdev, &myri10ge_ethtool_ops);
+       INIT_WORK(&mgp->watchdog_work, myri10ge_watchdog, mgp);
+       status = register_netdev(netdev);
+       if (status != 0) {
+               dev_err(&pdev->dev, "register_netdev failed: %d\n", status);
+               goto abort_with_irq;
+       }
+
+       printk(KERN_INFO "myri10ge: %s: %s IRQ %d, tx bndry %d, fw %s, WC %s\n",
+              netdev->name, (mgp->msi_enabled ? "MSI" : "xPIC"),
+              pdev->irq, mgp->tx.boundary, mgp->fw_name,
+              (mgp->mtrr >= 0 ? "Enabled" : "Disabled"));
+
+       return 0;
+
+abort_with_irq:
+       free_irq(pdev->irq, mgp);
+       if (mgp->msi_enabled)
+               pci_disable_msi(pdev);
+
+abort_with_firmware:
+       myri10ge_dummy_rdma(mgp, 0);
+
+abort_with_rx_done:
+       bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
+       pci_free_consistent(pdev, bytes, mgp->rx_done.entry, mgp->rx_done.bus);
+
+abort_with_ioremap:
+       iounmap(mgp->sram);
+
+abort_with_wc:
+#ifdef CONFIG_MTRR
+       if (mgp->mtrr >= 0)
+               mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
+#endif
+       pci_free_consistent(pdev, sizeof(*mgp->fw_stats),
+                           mgp->fw_stats, mgp->fw_stats_bus);
+
+abort_with_cmd:
+       pci_free_consistent(pdev, sizeof(*mgp->cmd), mgp->cmd, mgp->cmd_bus);
+
+abort_with_netdev:
+
+       free_netdev(netdev);
+       return status;
+}
+
+/*
+ * myri10ge_remove
+ *
+ * Does what is necessary to shutdown one Myrinet device. Called
+ *   once for each Myrinet card by the kernel when a module is
+ *   unloaded.
+ */
+static void myri10ge_remove(struct pci_dev *pdev)
+{
+       struct myri10ge_priv *mgp;
+       struct net_device *netdev;
+       size_t bytes;
+
+       mgp = pci_get_drvdata(pdev);
+       if (mgp == NULL)
+               return;
+
+       flush_scheduled_work();
+       netdev = mgp->dev;
+       unregister_netdev(netdev);
+       free_irq(pdev->irq, mgp);
+       if (mgp->msi_enabled)
+               pci_disable_msi(pdev);
+
+       myri10ge_dummy_rdma(mgp, 0);
+
+       bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
+       pci_free_consistent(pdev, bytes, mgp->rx_done.entry, mgp->rx_done.bus);
+
+       iounmap(mgp->sram);
+
+#ifdef CONFIG_MTRR
+       if (mgp->mtrr >= 0)
+               mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
+#endif
+       pci_free_consistent(pdev, sizeof(*mgp->fw_stats),
+                           mgp->fw_stats, mgp->fw_stats_bus);
+
+       pci_free_consistent(pdev, sizeof(*mgp->cmd), mgp->cmd, mgp->cmd_bus);
+
+       free_netdev(netdev);
+       pci_set_drvdata(pdev, NULL);
+}
+
+#define PCI_DEVICE_ID_MYIRCOM_MYRI10GE_Z8E     0x0008
+
+static struct pci_device_id myri10ge_pci_tbl[] = {
+       {PCI_DEVICE(PCI_VENDOR_ID_MYRICOM, PCI_DEVICE_ID_MYIRCOM_MYRI10GE_Z8E)},
+       {0},
+};
+
+static struct pci_driver myri10ge_driver = {
+       .name = "myri10ge",
+       .probe = myri10ge_probe,
+       .remove = myri10ge_remove,
+       .id_table = myri10ge_pci_tbl,
+#ifdef CONFIG_PM
+       .suspend = myri10ge_suspend,
+       .resume = myri10ge_resume,
+#endif
+};
+
+static __init int myri10ge_init_module(void)
+{
+       printk(KERN_INFO "%s: Version %s\n", myri10ge_driver.name,
+              MYRI10GE_VERSION_STR);
+       return pci_register_driver(&myri10ge_driver);
+}
+
+module_init(myri10ge_init_module);
+
+static __exit void myri10ge_cleanup_module(void)
+{
+       pci_unregister_driver(&myri10ge_driver);
+}
+
+module_exit(myri10ge_cleanup_module);
diff --git a/drivers/net/myri10ge/myri10ge_mcp.h b/drivers/net/myri10ge/myri10ge_mcp.h
new file mode 100644 (file)
index 0000000..ac23e18
--- /dev/null
@@ -0,0 +1,205 @@
+#ifndef __MYRI10GE_MCP_H__
+#define __MYRI10GE_MCP_H__
+
+#define MXGEFW_VERSION_MAJOR   1
+#define MXGEFW_VERSION_MINOR   4
+
+/* 8 Bytes */
+struct mcp_dma_addr {
+       u32 high;
+       u32 low;
+};
+
+/* 4 Bytes */
+struct mcp_slot {
+       u16 checksum;
+       u16 length;
+};
+
+/* 64 Bytes */
+struct mcp_cmd {
+       u32 cmd;
+       u32 data0;              /* will be low portion if data > 32 bits */
+       /* 8 */
+       u32 data1;              /* will be high portion if data > 32 bits */
+       u32 data2;              /* currently unused.. */
+       /* 16 */
+       struct mcp_dma_addr response_addr;
+       /* 24 */
+       u8 pad[40];
+};
+
+/* 8 Bytes */
+struct mcp_cmd_response {
+       u32 data;
+       u32 result;
+};
+
+/* 
+ * flags used in mcp_kreq_ether_send_t:
+ * 
+ * The SMALL flag is only needed in the first segment. It is raised
+ * for packets that are total less or equal 512 bytes.
+ * 
+ * The CKSUM flag must be set in all segments.
+ * 
+ * The PADDED flags is set if the packet needs to be padded, and it
+ * must be set for all segments.
+ * 
+ * The  MXGEFW_FLAGS_ALIGN_ODD must be set if the cumulative
+ * length of all previous segments was odd.
+ */
+
+#define MXGEFW_FLAGS_SMALL      0x1
+#define MXGEFW_FLAGS_TSO_HDR    0x1
+#define MXGEFW_FLAGS_FIRST      0x2
+#define MXGEFW_FLAGS_ALIGN_ODD  0x4
+#define MXGEFW_FLAGS_CKSUM      0x8
+#define MXGEFW_FLAGS_TSO_LAST   0x8
+#define MXGEFW_FLAGS_NO_TSO     0x10
+#define MXGEFW_FLAGS_TSO_CHOP   0x10
+#define MXGEFW_FLAGS_TSO_PLD    0x20
+
+#define MXGEFW_SEND_SMALL_SIZE  1520
+#define MXGEFW_MAX_MTU          9400
+
+union mcp_pso_or_cumlen {
+       u16 pseudo_hdr_offset;
+       u16 cum_len;
+};
+
+#define        MXGEFW_MAX_SEND_DESC 12
+#define MXGEFW_PAD         2
+
+/* 16 Bytes */
+struct mcp_kreq_ether_send {
+       u32 addr_high;
+       u32 addr_low;
+       u16 pseudo_hdr_offset;
+       u16 length;
+       u8 pad;
+       u8 rdma_count;
+       u8 cksum_offset;        /* where to start computing cksum */
+       u8 flags;               /* as defined above */
+};
+
+/* 8 Bytes */
+struct mcp_kreq_ether_recv {
+       u32 addr_high;
+       u32 addr_low;
+};
+
+/* Commands */
+
+#define MXGEFW_CMD_OFFSET 0xf80000
+
+enum myri10ge_mcp_cmd_type {
+       MXGEFW_CMD_NONE = 0,
+       /* Reset the mcp, it is left in a safe state, waiting
+        * for the driver to set all its parameters */
+       MXGEFW_CMD_RESET,
+
+       /* get the version number of the current firmware..
+        * (may be available in the eeprom strings..? */
+       MXGEFW_GET_MCP_VERSION,
+
+       /* Parameters which must be set by the driver before it can
+        * issue MXGEFW_CMD_ETHERNET_UP. They persist until the next
+        * MXGEFW_CMD_RESET is issued */
+
+       MXGEFW_CMD_SET_INTRQ_DMA,
+       MXGEFW_CMD_SET_BIG_BUFFER_SIZE, /* in bytes, power of 2 */
+       MXGEFW_CMD_SET_SMALL_BUFFER_SIZE,       /* in bytes */
+
+       /* Parameters which refer to lanai SRAM addresses where the 
+        * driver must issue PIO writes for various things */
+
+       MXGEFW_CMD_GET_SEND_OFFSET,
+       MXGEFW_CMD_GET_SMALL_RX_OFFSET,
+       MXGEFW_CMD_GET_BIG_RX_OFFSET,
+       MXGEFW_CMD_GET_IRQ_ACK_OFFSET,
+       MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET,
+
+       /* Parameters which refer to rings stored on the MCP,
+        * and whose size is controlled by the mcp */
+
+       MXGEFW_CMD_GET_SEND_RING_SIZE,  /* in bytes */
+       MXGEFW_CMD_GET_RX_RING_SIZE,    /* in bytes */
+
+       /* Parameters which refer to rings stored in the host,
+        * and whose size is controlled by the host.  Note that
+        * all must be physically contiguous and must contain 
+        * a power of 2 number of entries.  */
+
+       MXGEFW_CMD_SET_INTRQ_SIZE,      /* in bytes */
+
+       /* command to bring ethernet interface up.  Above parameters
+        * (plus mtu & mac address) must have been exchanged prior
+        * to issuing this command  */
+       MXGEFW_CMD_ETHERNET_UP,
+
+       /* command to bring ethernet interface down.  No further sends
+        * or receives may be processed until an MXGEFW_CMD_ETHERNET_UP
+        * is issued, and all interrupt queues must be flushed prior
+        * to ack'ing this command */
+
+       MXGEFW_CMD_ETHERNET_DOWN,
+
+       /* commands the driver may issue live, without resetting
+        * the nic.  Note that increasing the mtu "live" should
+        * only be done if the driver has already supplied buffers
+        * sufficiently large to handle the new mtu.  Decreasing
+        * the mtu live is safe */
+
+       MXGEFW_CMD_SET_MTU,
+       MXGEFW_CMD_GET_INTR_COAL_DELAY_OFFSET,  /* in microseconds */
+       MXGEFW_CMD_SET_STATS_INTERVAL,  /* in microseconds */
+       MXGEFW_CMD_SET_STATS_DMA,
+
+       MXGEFW_ENABLE_PROMISC,
+       MXGEFW_DISABLE_PROMISC,
+       MXGEFW_SET_MAC_ADDRESS,
+
+       MXGEFW_ENABLE_FLOW_CONTROL,
+       MXGEFW_DISABLE_FLOW_CONTROL,
+
+       /* do a DMA test
+        * data0,data1 = DMA address
+        * data2       = RDMA length (MSH), WDMA length (LSH)
+        * command return data = repetitions (MSH), 0.5-ms ticks (LSH)
+        */
+       MXGEFW_DMA_TEST
+};
+
+enum myri10ge_mcp_cmd_status {
+       MXGEFW_CMD_OK = 0,
+       MXGEFW_CMD_UNKNOWN,
+       MXGEFW_CMD_ERROR_RANGE,
+       MXGEFW_CMD_ERROR_BUSY,
+       MXGEFW_CMD_ERROR_EMPTY,
+       MXGEFW_CMD_ERROR_CLOSED,
+       MXGEFW_CMD_ERROR_HASH_ERROR,
+       MXGEFW_CMD_ERROR_BAD_PORT,
+       MXGEFW_CMD_ERROR_RESOURCES
+};
+
+/* 40 Bytes */
+struct mcp_irq_data {
+       u32 send_done_count;
+
+       u32 link_up;
+       u32 dropped_link_overflow;
+       u32 dropped_link_error_or_filtered;
+       u32 dropped_runt;
+       u32 dropped_overrun;
+       u32 dropped_no_small_buffer;
+       u32 dropped_no_big_buffer;
+       u32 rdma_tags_available;
+
+       u8 tx_stopped;
+       u8 link_down;
+       u8 stats_updated;
+       u8 valid;
+};
+
+#endif                         /* __MYRI10GE_MCP_H__ */
diff --git a/drivers/net/myri10ge/myri10ge_mcp_gen_header.h b/drivers/net/myri10ge/myri10ge_mcp_gen_header.h
new file mode 100644 (file)
index 0000000..1d3a35f
--- /dev/null
@@ -0,0 +1,58 @@
+#ifndef __MYRI10GE_MCP_GEN_HEADER_H__
+#define __MYRI10GE_MCP_GEN_HEADER_H__
+
+/* this file define a standard header used as a first entry point to
+ * exchange information between firmware/driver and driver.  The
+ * header structure can be anywhere in the mcp. It will usually be in
+ * the .data section, because some fields needs to be initialized at
+ * compile time.
+ * The 32bit word at offset MX_HEADER_PTR_OFFSET in the mcp must
+ * contains the location of the header. 
+ * 
+ * Typically a MCP will start with the following:
+ * .text
+ * .space 52    ! to help catch MEMORY_INT errors
+ * bt start     ! jump to real code
+ * nop
+ * .long _gen_mcp_header
+ * 
+ * The source will have a definition like:
+ * 
+ * mcp_gen_header_t gen_mcp_header = {
+ * .header_length = sizeof(mcp_gen_header_t),
+ * .mcp_type = MCP_TYPE_XXX,
+ * .version = "something $Id: mcp_gen_header.h,v 1.2 2006/05/13 10:04:35 bgoglin Exp $",
+ * .mcp_globals = (unsigned)&Globals
+ * };
+ */
+
+#define MCP_HEADER_PTR_OFFSET  0x3c
+
+#define MCP_TYPE_MX 0x4d582020 /* "MX  " */
+#define MCP_TYPE_PCIE 0x70636965       /* "PCIE" pcie-only MCP */
+#define MCP_TYPE_ETH 0x45544820        /* "ETH " */
+#define MCP_TYPE_MCP0 0x4d435030       /* "MCP0" */
+
+struct mcp_gen_header {
+       /* the first 4 fields are filled at compile time */
+       unsigned header_length;
+       unsigned mcp_type;
+       char version[128];
+       unsigned mcp_globals;   /* pointer to mcp-type specific structure */
+
+       /* filled by the MCP at run-time */
+       unsigned sram_size;
+       unsigned string_specs;  /* either the original STRING_SPECS or a superset */
+       unsigned string_specs_len;
+
+       /* Fields above this comment are guaranteed to be present.
+        * 
+        * Fields below this comment are extensions added in later versions
+        * of this struct, drivers should compare the header_length against
+        * offsetof(field) to check wether a given MCP implements them.
+        * 
+        * Never remove any field.  Keep everything naturally align.
+        */
+};
+
+#endif                         /* __MYRI10GE_MCP_GEN_HEADER_H__ */
index 563eeadbea46441a8b8196b35c6fa1f080e6ff79..151549a69ee0cd47acd50333c7aa98399c478737 100644 (file)
 
 #define PCI_VENDOR_ID_SAMSUNG          0x144d
 
+#define PCI_VENDOR_ID_MYRICOM          0x14c1
 
 #define PCI_VENDOR_ID_TITAN            0x14D2
 #define PCI_DEVICE_ID_TITAN_010L       0x8001