Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6.git] / drivers / net / myri10ge / myri10ge.c
index eddcee3..5319db9 100644 (file)
@@ -1,7 +1,7 @@
 /*************************************************************************
  * myri10ge.c: Myricom Myri-10G Ethernet driver.
  *
- * Copyright (C) 2005 - 2007 Myricom, Inc.
+ * Copyright (C) 2005 - 2009 Myricom, Inc.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
 #include <linux/if_ether.h>
 #include <linux/if_vlan.h>
 #include <linux/inet_lro.h>
+#include <linux/dca.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>
@@ -75,7 +75,7 @@
 #include "myri10ge_mcp.h"
 #include "myri10ge_mcp_gen_header.h"
 
-#define MYRI10GE_VERSION_STR "1.3.2-1.287"
+#define MYRI10GE_VERSION_STR "1.5.1-1.451"
 
 MODULE_DESCRIPTION("Myricom 10G driver (10GbE)");
 MODULE_AUTHOR("Maintainer: help@myri.com");
@@ -102,6 +102,8 @@ MODULE_LICENSE("Dual BSD/GPL");
 #define MYRI10GE_ALLOC_SIZE ((1 << MYRI10GE_ALLOC_ORDER) * PAGE_SIZE)
 #define MYRI10GE_MAX_FRAGS_PER_FRAME (MYRI10GE_MAX_ETHER_MTU/MYRI10GE_ALLOC_SIZE + 1)
 
+#define MYRI10GE_MAX_SLICES 32
+
 struct myri10ge_rx_buffer_state {
        struct page *page;
        int page_offset;
@@ -124,7 +126,6 @@ struct myri10ge_cmd {
 
 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;
        struct page *page;
@@ -139,7 +140,8 @@ struct myri10ge_rx_buf {
 
 struct myri10ge_tx_buf {
        struct mcp_kreq_ether_send __iomem *lanai;      /* lanai ptr for sendq */
-       u8 __iomem *wc_fifo;    /* w/c send fifo address */
+       __be32 __iomem *send_go;        /* "go" doorbell ptr */
+       __be32 __iomem *send_stop;      /* "stop" doorbell ptr */
        struct mcp_kreq_ether_send *req_list;   /* host shadow of sendq */
        char *req_bytes;
        struct myri10ge_tx_buffer_state *info;
@@ -151,6 +153,7 @@ struct myri10ge_tx_buf {
        int done ____cacheline_aligned; /* transmit slots completed     */
        int pkt_done;           /* packets completed */
        int wake_queue;
+       int queue_active;
 };
 
 struct myri10ge_rx_done {
@@ -185,6 +188,12 @@ struct myri10ge_slice_state {
        dma_addr_t fw_stats_bus;
        int watchdog_tx_done;
        int watchdog_tx_req;
+       int watchdog_rx_done;
+#ifdef CONFIG_MYRI10GE_DCA
+       int cached_dca_tag;
+       int cpu;
+       __be32 __iomem *dca_tag;
+#endif
        char irq_desc[32];
 };
 
@@ -198,7 +207,6 @@ struct myri10ge_priv {
        int big_bytes;
        int max_intr_slots;
        struct net_device *dev;
-       struct net_device_stats stats;
        spinlock_t stats_lock;
        u8 __iomem *sram;
        int sram_size;
@@ -212,6 +220,9 @@ struct myri10ge_priv {
        int msi_enabled;
        int msix_enabled;
        struct msix_entry *msix_vectors;
+#ifdef CONFIG_MYRI10GE_DCA
+       int dca_enabled;
+#endif
        u32 link_state;
        unsigned int rdma_tags_available;
        int intr_coal_delay;
@@ -244,6 +255,8 @@ struct myri10ge_priv {
        u32 read_write_dma;
        u32 link_changes;
        u32 msg_enable;
+       unsigned int board_number;
+       int rebooted;
 };
 
 static char *myri10ge_fw_unaligned = "myri10ge_ethp_z8e.dat";
@@ -255,6 +268,13 @@ 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");
 
+#define MYRI10GE_MAX_BOARDS 8
+static char *myri10ge_fw_names[MYRI10GE_MAX_BOARDS] =
+    {[0 ... (MYRI10GE_MAX_BOARDS - 1)] = NULL };
+module_param_array_named(myri10ge_fw_names, myri10ge_fw_names, charp, NULL,
+                        0444);
+MODULE_PARM_DESC(myri10ge_fw_name, "Firmware image names per board");
+
 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");
@@ -308,10 +328,6 @@ static int myri10ge_debug = -1;    /* defaults above */
 module_param(myri10ge_debug, int, 0);
 MODULE_PARM_DESC(myri10ge_debug, "Debug level (0=none,...,16=all)");
 
-static int myri10ge_lro = 1;
-module_param(myri10ge_lro, int, S_IRUGO);
-MODULE_PARM_DESC(myri10ge_lro, "Enable large receive offload");
-
 static int myri10ge_lro_max_pkts = MYRI10GE_LRO_MAX_PKTS;
 module_param(myri10ge_lro_max_pkts, int, S_IRUGO);
 MODULE_PARM_DESC(myri10ge_lro_max_pkts,
@@ -323,10 +339,6 @@ MODULE_PARM_DESC(myri10ge_fill_thresh, "Number of empty rx slots allowed");
 
 static int myri10ge_reset_recover = 1;
 
-static int myri10ge_wcfifo = 0;
-module_param(myri10ge_wcfifo, int, S_IRUGO);
-MODULE_PARM_DESC(myri10ge_wcfifo, "Enable WC Fifo when WC is enabled");
-
 static int myri10ge_max_slices = 1;
 module_param(myri10ge_max_slices, int, S_IRUGO);
 MODULE_PARM_DESC(myri10ge_max_slices, "Max tx/rx queues");
@@ -335,6 +347,10 @@ static int myri10ge_rss_hash = MXGEFW_RSS_HASH_TYPE_SRC_PORT;
 module_param(myri10ge_rss_hash, int, S_IRUGO);
 MODULE_PARM_DESC(myri10ge_rss_hash, "Type of RSS hashing to do");
 
+static int myri10ge_dca = 1;
+module_param(myri10ge_dca, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_dca, "Enable DCA if possible");
+
 #define MYRI10GE_FW_OFFSET 1024*1024
 #define MYRI10GE_HIGHPART_TO_U32(X) \
 (sizeof (X) == 8) ? ((u32)((u64)(X) >> 32)) : (0)
@@ -343,13 +359,16 @@ MODULE_PARM_DESC(myri10ge_rss_hash, "Type of RSS hashing to do");
 #define myri10ge_pio_copy(to,from,size) __iowrite64_copy(to,from,size/8)
 
 static void myri10ge_set_multicast_list(struct net_device *dev);
-static int myri10ge_sw_tso(struct sk_buff *skb, struct net_device *dev);
+static netdev_tx_t myri10ge_sw_tso(struct sk_buff *skb,
+                                        struct net_device *dev);
 
 static inline void put_be32(__be32 val, __be32 __iomem * p)
 {
        __raw_writel((__force __u32) val, (__force void __iomem *)p);
 }
 
+static struct net_device_stats *myri10ge_get_stats(struct net_device *dev);
+
 static int
 myri10ge_send_cmd(struct myri10ge_priv *mgp, u32 cmd,
                  struct myri10ge_cmd *data, int atomic)
@@ -412,6 +431,12 @@ myri10ge_send_cmd(struct myri10ge_priv *mgp, u32 cmd,
                        return -ENOSYS;
                } else if (result == MXGEFW_CMD_ERROR_UNALIGNED) {
                        return -E2BIG;
+               } else if (result == MXGEFW_CMD_ERROR_RANGE &&
+                          cmd == MXGEFW_CMD_ENABLE_RSS_QUEUES &&
+                          (data->
+                           data1 & MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES) !=
+                          0) {
+                       return -ERANGE;
                } else {
                        dev_err(&mgp->pdev->dev,
                                "command %d failed, result = %d\n",
@@ -543,6 +568,7 @@ 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;
+       unsigned char *fw_readback;
        struct mcp_gen_header *hdr;
        size_t hdr_offset;
        int status;
@@ -585,9 +611,15 @@ static int myri10ge_load_hotplug_firmware(struct myri10ge_priv *mgp, u32 * size)
                mb();
                readb(mgp->sram);
        }
+       fw_readback = vmalloc(fw->size);
+       if (!fw_readback) {
+               status = -ENOMEM;
+               goto abort_with_fw;
+       }
        /* 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);
+       memcpy_fromio(fw_readback, mgp->sram + MYRI10GE_FW_OFFSET, fw->size);
+       reread_crc = crc32(~0, fw_readback, fw->size);
+       vfree(fw_readback);
        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);
@@ -878,6 +910,9 @@ static int myri10ge_reset(struct myri10ge_priv *mgp)
        struct myri10ge_slice_state *ss;
        int i, status;
        size_t bytes;
+#ifdef CONFIG_MYRI10GE_DCA
+       unsigned long dca_tag_off;
+#endif
 
        /* try to send a reset command to the card to see if it
         * is alive */
@@ -931,9 +966,24 @@ static int myri10ge_reset(struct myri10ge_priv *mgp)
                 */
 
                cmd.data0 = mgp->num_slices;
-               cmd.data1 = 1;  /* use MSI-X */
+               cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE;
+               if (mgp->dev->real_num_tx_queues > 1)
+                       cmd.data1 |= MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES;
                status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES,
                                           &cmd, 0);
+
+               /* Firmware older than 1.4.32 only supports multiple
+                * RX queues, so if we get an error, first retry using a
+                * single TX queue before giving up */
+               if (status != 0 && mgp->dev->real_num_tx_queues > 1) {
+                       mgp->dev->real_num_tx_queues = 1;
+                       cmd.data0 = mgp->num_slices;
+                       cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE;
+                       status = myri10ge_send_cmd(mgp,
+                                                  MXGEFW_CMD_ENABLE_RSS_QUEUES,
+                                                  &cmd, 0);
+               }
+
                if (status != 0) {
                        dev_err(&mgp->pdev->dev,
                                "failed to set number of slices\n");
@@ -970,6 +1020,20 @@ static int myri10ge_reset(struct myri10ge_priv *mgp)
        }
        put_be32(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
 
+#ifdef CONFIG_MYRI10GE_DCA
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_DCA_OFFSET, &cmd, 0);
+       dca_tag_off = cmd.data0;
+       for (i = 0; i < mgp->num_slices; i++) {
+               ss = &mgp->ss[i];
+               if (status == 0) {
+                       ss->dca_tag = (__iomem __be32 *)
+                           (mgp->sram + dca_tag_off + 4 * i);
+               } else {
+                       ss->dca_tag = NULL;
+               }
+       }
+#endif                         /* CONFIG_MYRI10GE_DCA */
+
        /* reset mcp/driver shared state back to 0 */
 
        mgp->link_changes = 0;
@@ -995,6 +1059,78 @@ static int myri10ge_reset(struct myri10ge_priv *mgp)
        return status;
 }
 
+#ifdef CONFIG_MYRI10GE_DCA
+static void
+myri10ge_write_dca(struct myri10ge_slice_state *ss, int cpu, int tag)
+{
+       ss->cpu = cpu;
+       ss->cached_dca_tag = tag;
+       put_be32(htonl(tag), ss->dca_tag);
+}
+
+static inline void myri10ge_update_dca(struct myri10ge_slice_state *ss)
+{
+       int cpu = get_cpu();
+       int tag;
+
+       if (cpu != ss->cpu) {
+               tag = dca_get_tag(cpu);
+               if (ss->cached_dca_tag != tag)
+                       myri10ge_write_dca(ss, cpu, tag);
+       }
+       put_cpu();
+}
+
+static void myri10ge_setup_dca(struct myri10ge_priv *mgp)
+{
+       int err, i;
+       struct pci_dev *pdev = mgp->pdev;
+
+       if (mgp->ss[0].dca_tag == NULL || mgp->dca_enabled)
+               return;
+       if (!myri10ge_dca) {
+               dev_err(&pdev->dev, "dca disabled by administrator\n");
+               return;
+       }
+       err = dca_add_requester(&pdev->dev);
+       if (err) {
+               if (err != -ENODEV)
+                       dev_err(&pdev->dev,
+                               "dca_add_requester() failed, err=%d\n", err);
+               return;
+       }
+       mgp->dca_enabled = 1;
+       for (i = 0; i < mgp->num_slices; i++)
+               myri10ge_write_dca(&mgp->ss[i], -1, 0);
+}
+
+static void myri10ge_teardown_dca(struct myri10ge_priv *mgp)
+{
+       struct pci_dev *pdev = mgp->pdev;
+       int err;
+
+       if (!mgp->dca_enabled)
+               return;
+       mgp->dca_enabled = 0;
+       err = dca_remove_requester(&pdev->dev);
+}
+
+static int myri10ge_notify_dca_device(struct device *dev, void *data)
+{
+       struct myri10ge_priv *mgp;
+       unsigned long event;
+
+       mgp = dev_get_drvdata(dev);
+       event = *(unsigned long *)data;
+
+       if (event == DCA_PROVIDER_ADD)
+               myri10ge_setup_dca(mgp);
+       else if (event == DCA_PROVIDER_REMOVE)
+               myri10ge_teardown_dca(mgp);
+       return 0;
+}
+#endif                         /* CONFIG_MYRI10GE_DCA */
+
 static inline void
 myri10ge_submit_8rx(struct mcp_kreq_ether_recv __iomem * dst,
                    struct mcp_kreq_ether_recv *src)
@@ -1002,7 +1138,7 @@ myri10ge_submit_8rx(struct mcp_kreq_ether_recv __iomem * dst,
        __be32 low;
 
        low = src->addr_low;
-       src->addr_low = htonl(DMA_32BIT_MASK);
+       src->addr_low = htonl(DMA_BIT_MASK(32));
        myri10ge_pio_copy(dst, src, 4 * sizeof(*src));
        mb();
        myri10ge_pio_copy(dst + 4, src + 4, 4 * sizeof(*src));
@@ -1110,14 +1246,8 @@ myri10ge_alloc_rx_pages(struct myri10ge_priv *mgp, struct myri10ge_rx_buf *rx,
 
                /* copy 8 descriptors to the firmware 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);
-                       }
+                       myri10ge_submit_8rx(&rx->lanai[idx - 7],
+                                           &rx->shadow[idx - 7]);
                }
        }
 }
@@ -1168,7 +1298,7 @@ myri10ge_rx_done(struct myri10ge_slice_state *ss, struct myri10ge_rx_buf *rx,
                remainder -= MYRI10GE_ALLOC_SIZE;
        }
 
-       if (mgp->csum_flag && myri10ge_lro) {
+       if (dev->features & NETIF_F_LRO) {
                rx_frags[0].page_offset += MXGEFW_PAD;
                rx_frags[0].size -= MXGEFW_PAD;
                len -= MXGEFW_PAD;
@@ -1187,7 +1317,7 @@ myri10ge_rx_done(struct myri10ge_slice_state *ss, struct myri10ge_rx_buf *rx,
 
        skb = netdev_alloc_skb(dev, MYRI10GE_HLEN + 16);
        if (unlikely(skb == NULL)) {
-               mgp->stats.rx_dropped++;
+               ss->stats.rx_dropped++;
                do {
                        i--;
                        put_page(rx_frags[i].page);
@@ -1202,6 +1332,7 @@ myri10ge_rx_done(struct myri10ge_slice_state *ss, struct myri10ge_rx_buf *rx,
                skb_shinfo(skb)->nr_frags = 0;
        }
        skb->protocol = eth_type_trans(skb, dev);
+       skb_record_rx_queue(skb, ss - &mgp->ss[0]);
 
        if (mgp->csum_flag) {
                if ((skb->protocol == htons(ETH_P_IP)) ||
@@ -1212,7 +1343,6 @@ myri10ge_rx_done(struct myri10ge_slice_state *ss, struct myri10ge_rx_buf *rx,
                        myri10ge_vlan_ip_csum(skb, csum);
        }
        netif_receive_skb(skb);
-       dev->last_rx = jiffies;
        return 1;
 }
 
@@ -1221,6 +1351,7 @@ myri10ge_tx_done(struct myri10ge_slice_state *ss, int mcp_index)
 {
        struct pci_dev *pdev = ss->mgp->pdev;
        struct myri10ge_tx_buf *tx = &ss->tx;
+       struct netdev_queue *dev_queue;
        struct sk_buff *skb;
        int idx, len;
 
@@ -1254,11 +1385,33 @@ myri10ge_tx_done(struct myri10ge_slice_state *ss, int mcp_index)
                                               PCI_DMA_TODEVICE);
                }
        }
+
+       dev_queue = netdev_get_tx_queue(ss->dev, ss - ss->mgp->ss);
+       /*
+        * Make a minimal effort to prevent the NIC from polling an
+        * idle tx queue.  If we can't get the lock we leave the queue
+        * active. In this case, either a thread was about to start
+        * using the queue anyway, or we lost a race and the NIC will
+        * waste some of its resources polling an inactive queue for a
+        * while.
+        */
+
+       if ((ss->mgp->dev->real_num_tx_queues > 1) &&
+           __netif_tx_trylock(dev_queue)) {
+               if (tx->req == tx->done) {
+                       tx->queue_active = 0;
+                       put_be32(htonl(1), tx->send_stop);
+                       mb();
+                       mmiowb();
+               }
+               __netif_tx_unlock(dev_queue);
+       }
+
        /* start the queue if we've stopped it */
-       if (netif_queue_stopped(ss->dev)
+       if (netif_tx_queue_stopped(dev_queue)
            && tx->req - tx->done < (tx->mask >> 1)) {
                tx->wake_queue++;
-               netif_wake_queue(ss->dev);
+               netif_tx_wake_queue(dev_queue);
        }
 }
 
@@ -1267,6 +1420,7 @@ myri10ge_clean_rx_done(struct myri10ge_slice_state *ss, int budget)
 {
        struct myri10ge_rx_done *rx_done = &ss->rx_done;
        struct myri10ge_priv *mgp = ss->mgp;
+       struct net_device *netdev = mgp->dev;
        unsigned long rx_bytes = 0;
        unsigned long rx_packets = 0;
        unsigned long rx_ok;
@@ -1300,7 +1454,7 @@ myri10ge_clean_rx_done(struct myri10ge_slice_state *ss, int budget)
        ss->stats.rx_packets += rx_packets;
        ss->stats.rx_bytes += rx_bytes;
 
-       if (myri10ge_lro)
+       if (netdev->features & NETIF_F_LRO)
                lro_flush_all(&rx_done->lro_mgr);
 
        /* restock receive rings if needed */
@@ -1359,14 +1513,18 @@ static int myri10ge_poll(struct napi_struct *napi, int budget)
 {
        struct myri10ge_slice_state *ss =
            container_of(napi, struct myri10ge_slice_state, napi);
-       struct net_device *netdev = ss->mgp->dev;
        int work_done;
 
+#ifdef CONFIG_MYRI10GE_DCA
+       if (ss->mgp->dca_enabled)
+               myri10ge_update_dca(ss);
+#endif
+
        /* process as many rx events as NAPI will allow */
        work_done = myri10ge_clean_rx_done(ss, budget);
 
        if (work_done < budget) {
-               netif_rx_complete(netdev, napi);
+               napi_complete(napi);
                put_be32(htonl(3), ss->irq_claim);
        }
        return work_done;
@@ -1381,10 +1539,10 @@ static irqreturn_t myri10ge_intr(int irq, void *arg)
        u32 send_done_count;
        int i;
 
-       /* an interrupt on a non-zero slice is implicitly valid
-        * since MSI-X irqs are not shared */
-       if (ss != mgp->ss) {
-               netif_rx_schedule(ss->dev, &ss->napi);
+       /* an interrupt on a non-zero receive-only slice is implicitly
+        * valid  since MSI-X irqs are not shared */
+       if ((mgp->dev->real_num_tx_queues == 1) && (ss != mgp->ss)) {
+               napi_schedule(&ss->napi);
                return (IRQ_HANDLED);
        }
 
@@ -1395,7 +1553,7 @@ static irqreturn_t myri10ge_intr(int irq, void *arg)
        /* low bit indicates receives are present, so schedule
         * napi poll handler */
        if (stats->valid & 1)
-               netif_rx_schedule(ss->dev, &ss->napi);
+               napi_schedule(&ss->napi);
 
        if (!mgp->msi_enabled && !mgp->msix_enabled) {
                put_be32(0, mgp->irq_deassert);
@@ -1425,7 +1583,9 @@ static irqreturn_t myri10ge_intr(int irq, void *arg)
                barrier();
        }
 
-       myri10ge_check_statblock(mgp);
+       /* Only slice 0 updates stats */
+       if (ss == mgp->ss)
+               myri10ge_check_statblock(mgp);
 
        put_be32(htonl(3), ss->irq_claim + 1);
        return (IRQ_HANDLED);
@@ -1463,10 +1623,21 @@ myri10ge_get_settings(struct net_device *netdev, struct ethtool_cmd *cmd)
                        return 0;
                }
        }
-       if (*ptr == 'R' || *ptr == 'Q') {
-               /* We've found either an XFP or quad ribbon fiber */
+       if (*ptr == '2')
+               ptr++;
+       if (*ptr == 'R' || *ptr == 'Q' || *ptr == 'S') {
+               /* We've found either an XFP, quad ribbon fiber, or SFP+ */
                cmd->port = PORT_FIBRE;
+               cmd->supported |= SUPPORTED_FIBRE;
+               cmd->advertising |= ADVERTISED_FIBRE;
+       } else {
+               cmd->port = PORT_OTHER;
        }
+       if (*ptr == 'R' || *ptr == 'S')
+               cmd->transceiver = XCVR_EXTERNAL;
+       else
+               cmd->transceiver = XCVR_INTERNAL;
+
        return 0;
 }
 
@@ -1535,7 +1706,7 @@ myri10ge_get_ringparam(struct net_device *netdev,
        ring->rx_mini_max_pending = mgp->ss[0].rx_small.mask + 1;
        ring->rx_max_pending = mgp->ss[0].rx_big.mask + 1;
        ring->rx_jumbo_max_pending = 0;
-       ring->tx_max_pending = mgp->ss[0].rx_small.mask + 1;
+       ring->tx_max_pending = mgp->ss[0].tx.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;
@@ -1555,12 +1726,17 @@ static u32 myri10ge_get_rx_csum(struct net_device *netdev)
 static int myri10ge_set_rx_csum(struct net_device *netdev, u32 csum_enabled)
 {
        struct myri10ge_priv *mgp = netdev_priv(netdev);
+       int err = 0;
 
        if (csum_enabled)
                mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
-       else
+       else {
+               u32 flags = ethtool_op_get_flags(netdev);
+               err = ethtool_op_set_flags(netdev, (flags & ~ETH_FLAG_LRO));
                mgp->csum_flag = 0;
-       return 0;
+
+       }
+       return err;
 }
 
 static int myri10ge_set_tso(struct net_device *netdev, u32 tso_enabled)
@@ -1586,6 +1762,9 @@ static const char myri10ge_gstrings_main_stats[][ETH_GSTRING_LEN] = {
        "tx_boundary", "WC", "irq", "MSI", "MSIX",
        "read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs",
        "serial_number", "watchdog_resets",
+#ifdef CONFIG_MYRI10GE_DCA
+       "dca_capable_firmware", "dca_device_present",
+#endif
        "link_changes", "link_up", "dropped_link_overflow",
        "dropped_link_error_or_filtered",
        "dropped_pause", "dropped_bad_phy", "dropped_bad_crc32",
@@ -1649,8 +1828,10 @@ myri10ge_get_ethtool_stats(struct net_device *netdev,
        int slice;
        int i;
 
+       /* force stats update */
+       (void)myri10ge_get_stats(netdev);
        for (i = 0; i < MYRI10GE_NET_STATS_LEN; i++)
-               data[i] = ((unsigned long *)&mgp->stats)[i];
+               data[i] = ((unsigned long *)&netdev->stats)[i];
 
        data[i++] = (unsigned int)mgp->tx_boundary;
        data[i++] = (unsigned int)mgp->wc_enabled;
@@ -1662,6 +1843,10 @@ myri10ge_get_ethtool_stats(struct net_device *netdev,
        data[i++] = (unsigned int)mgp->read_write_dma;
        data[i++] = (unsigned int)mgp->serial_number;
        data[i++] = (unsigned int)mgp->watchdog_resets;
+#ifdef CONFIG_MYRI10GE_DCA
+       data[i++] = (unsigned int)(mgp->ss[0].dca_tag != NULL);
+       data[i++] = (unsigned int)(mgp->dca_enabled);
+#endif
        data[i++] = (unsigned int)mgp->link_changes;
 
        /* firmware stats are useful only in the first slice */
@@ -1734,7 +1919,9 @@ static const struct ethtool_ops myri10ge_ethtool_ops = {
        .get_sset_count = myri10ge_get_sset_count,
        .get_ethtool_stats = myri10ge_get_ethtool_stats,
        .set_msglevel = myri10ge_set_msglevel,
-       .get_msglevel = myri10ge_get_msglevel
+       .get_msglevel = myri10ge_get_msglevel,
+       .get_flags = ethtool_op_get_flags,
+       .set_flags = ethtool_op_set_flags
 };
 
 static int myri10ge_allocate_rings(struct myri10ge_slice_state *ss)
@@ -1776,6 +1963,7 @@ static int myri10ge_allocate_rings(struct myri10ge_slice_state *ss)
        /* ensure req_list entries are aligned to 8 bytes */
        ss->tx.req_list = (struct mcp_kreq_ether_send *)
            ALIGN((unsigned long)ss->tx.req_bytes, 8);
+       ss->tx.queue_active = 0;
 
        bytes = rx_ring_entries * sizeof(*ss->rx_small.shadow);
        ss->rx_small.shadow = kzalloc(bytes, GFP_KERNEL);
@@ -2070,6 +2258,8 @@ myri10ge_get_frag_header(struct skb_frag_struct *frag, void **mac_hdr,
        *ip_hdr = iph;
        if (iph->protocol != IPPROTO_TCP)
                return -1;
+       if (iph->frag_off & htons(IP_MF | IP_OFFSET))
+               return -1;
        *hdr_flags |= LRO_TCP;
        *tcpudp_hdr = (u8 *) (*ip_hdr) + (iph->ihl << 2);
 
@@ -2093,11 +2283,14 @@ static int myri10ge_get_txrx(struct myri10ge_priv *mgp, int slice)
        int status;
 
        ss = &mgp->ss[slice];
-       cmd.data0 = 0;          /* single slice for now */
-       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET, &cmd, 0);
-       ss->tx.lanai = (struct mcp_kreq_ether_send __iomem *)
-           (mgp->sram + cmd.data0);
-
+       status = 0;
+       if (slice == 0 || (mgp->dev->real_num_tx_queues > 1)) {
+               cmd.data0 = slice;
+               status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET,
+                                          &cmd, 0);
+               ss->tx.lanai = (struct mcp_kreq_ether_send __iomem *)
+                   (mgp->sram + cmd.data0);
+       }
        cmd.data0 = slice;
        status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SMALL_RX_OFFSET,
                                    &cmd, 0);
@@ -2109,18 +2302,10 @@ static int myri10ge_get_txrx(struct myri10ge_priv *mgp, int slice)
        ss->rx_big.lanai = (struct mcp_kreq_ether_recv __iomem *)
            (mgp->sram + cmd.data0);
 
-       if (myri10ge_wcfifo && mgp->wc_enabled) {
-               ss->tx.wc_fifo = (u8 __iomem *)
-                   mgp->sram + MXGEFW_ETH_SEND_4 + 64 * slice;
-               ss->rx_small.wc_fifo = (u8 __iomem *)
-                   mgp->sram + MXGEFW_ETH_RECV_SMALL + 64 * slice;
-               ss->rx_big.wc_fifo = (u8 __iomem *)
-                   mgp->sram + MXGEFW_ETH_RECV_BIG + 64 * slice;
-       } else {
-               ss->tx.wc_fifo = NULL;
-               ss->rx_small.wc_fifo = NULL;
-               ss->rx_big.wc_fifo = NULL;
-       }
+       ss->tx.send_go = (__iomem __be32 *)
+           (mgp->sram + MXGEFW_ETH_SEND_GO + 64 * slice);
+       ss->tx.send_stop = (__iomem __be32 *)
+           (mgp->sram + MXGEFW_ETH_SEND_STOP + 64 * slice);
        return status;
 
 }
@@ -2134,7 +2319,7 @@ static int myri10ge_set_stats(struct myri10ge_priv *mgp, int slice)
        ss = &mgp->ss[slice];
        cmd.data0 = MYRI10GE_LOWPART_TO_U32(ss->fw_stats_bus);
        cmd.data1 = MYRI10GE_HIGHPART_TO_U32(ss->fw_stats_bus);
-       cmd.data2 = sizeof(struct mcp_irq_data);
+       cmd.data2 = sizeof(struct mcp_irq_data) | (slice << 16);
        status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA_V2, &cmd, 0);
        if (status == -ENOSYS) {
                dma_addr_t bus = ss->fw_stats_bus;
@@ -2175,7 +2360,9 @@ static int myri10ge_open(struct net_device *dev)
 
        if (mgp->num_slices > 1) {
                cmd.data0 = mgp->num_slices;
-               cmd.data1 = 1;  /* use MSI-X */
+               cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE;
+               if (mgp->dev->real_num_tx_queues > 1)
+                       cmd.data1 |= MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES;
                status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES,
                                           &cmd, 0);
                if (status != 0) {
@@ -2196,6 +2383,7 @@ static int myri10ge_open(struct net_device *dev)
                        printk(KERN_ERR
                               "myri10ge: %s: failed to setup rss tables\n",
                               dev->name);
+                       goto abort_with_nothing;
                }
 
                /* just enable an identity mapping */
@@ -2266,7 +2454,11 @@ static int myri10ge_open(struct net_device *dev)
                status = myri10ge_allocate_rings(ss);
                if (status != 0)
                        goto abort_with_rings;
-               if (slice == 0)
+
+               /* only firmware which supports multiple TX queues
+                * supports setting up the tx stats on non-zero
+                * slices */
+               if (slice == 0 || mgp->dev->real_num_tx_queues > 1)
                        status = myri10ge_set_stats(mgp, slice);
                if (status) {
                        printk(KERN_ERR
@@ -2284,6 +2476,7 @@ static int myri10ge_open(struct net_device *dev)
                lro_mgr->lro_arr = ss->rx_done.lro_desc;
                lro_mgr->get_frag_header = myri10ge_get_frag_header;
                lro_mgr->max_aggr = myri10ge_lro_max_pkts;
+               lro_mgr->frag_align_pad = 2;
                if (lro_mgr->max_aggr > MAX_SKB_FRAGS)
                        lro_mgr->max_aggr = MAX_SKB_FRAGS;
 
@@ -2332,10 +2525,15 @@ static int myri10ge_open(struct net_device *dev)
        mgp->running = MYRI10GE_ETH_RUNNING;
        mgp->watchdog_timer.expires = jiffies + myri10ge_watchdog_timeout * HZ;
        add_timer(&mgp->watchdog_timer);
-       netif_wake_queue(dev);
+       netif_tx_wake_all_queues(dev);
+
        return 0;
 
 abort_with_rings:
+       while (slice) {
+               slice--;
+               napi_disable(&mgp->ss[slice].napi);
+       }
        for (i = 0; i < mgp->num_slices; i++)
                myri10ge_free_rings(&mgp->ss[i]);
 
@@ -2365,18 +2563,24 @@ static int myri10ge_close(struct net_device *dev)
                napi_disable(&mgp->ss[i].napi);
        }
        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_stop_all_queues(dev);
+       if (mgp->rebooted == 0) {
+               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_irq(mgp);
        for (i = 0; i < mgp->num_slices; i++)
@@ -2453,27 +2657,6 @@ myri10ge_submit_req(struct myri10ge_tx_buf *tx, struct mcp_kreq_ether_send *src,
        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 + MXGEFW_ETH_SEND_OFFSET(cnt),
-                                 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
@@ -2484,25 +2667,29 @@ myri10ge_submit_req_wc(struct myri10ge_tx_buf *tx,
  * it and try again.
  */
 
-static int myri10ge_xmit(struct sk_buff *skb, struct net_device *dev)
+static netdev_tx_t myri10ge_xmit(struct sk_buff *skb,
+                                      struct net_device *dev)
 {
        struct myri10ge_priv *mgp = netdev_priv(dev);
        struct myri10ge_slice_state *ss;
        struct mcp_kreq_ether_send *req;
        struct myri10ge_tx_buf *tx;
        struct skb_frag_struct *frag;
+       struct netdev_queue *netdev_queue;
        dma_addr_t bus;
        u32 low;
        __be32 high_swapped;
        unsigned int len;
        int idx, last_idx, avail, frag_cnt, frag_idx, count, mss, max_segments;
-       u16 pseudo_hdr_offset, cksum_offset;
+       u16 pseudo_hdr_offset, cksum_offset, queue;
        int cum_len, seglen, boundary, rdma_count;
        u8 flags, odd_flag;
 
-       /* always transmit through slot 0 */
-       ss = mgp->ss;
+       queue = skb_get_queue_mapping(skb);
+       ss = &mgp->ss[queue];
+       netdev_queue = netdev_get_tx_queue(mgp->dev, queue);
        tx = &ss->tx;
+
 again:
        req = tx->req_list;
        avail = tx->mask - 1 - (tx->req - tx->done);
@@ -2518,8 +2705,8 @@ again:
        if ((unlikely(avail < max_segments))) {
                /* we are out of transmit resources */
                tx->stop_queue++;
-               netif_stop_queue(dev);
-               return 1;
+               netif_tx_stop_queue(netdev_queue);
+               return NETDEV_TX_BUSY;
        }
 
        /* Setup checksum offloading, if needed */
@@ -2580,7 +2767,7 @@ again:
                                /* The packet is gone, so we must
                                 * return 0 */
                                ss->stats.tx_dropped += 1;
-                               return 0;
+                               return NETDEV_TX_OK;
                        }
                        /* adjust the len to account for the zero pad
                         * so that the nic can know how long it is */
@@ -2710,17 +2897,21 @@ again:
                                         MXGEFW_FLAGS_FIRST)));
        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);
+       myri10ge_submit_req(tx, tx->req_list, count);
+       /* if using multiple tx queues, make sure NIC polls the
+        * current slice */
+       if ((mgp->dev->real_num_tx_queues > 1) && tx->queue_active == 0) {
+               tx->queue_active = 1;
+               put_be32(htonl(1), tx->send_go);
+               mb();
+               mmiowb();
+       }
        tx->pkt_start++;
        if ((avail - count) < MXGEFW_MAX_SEND_DESC) {
                tx->stop_queue++;
-               netif_stop_queue(dev);
+               netif_tx_stop_queue(netdev_queue);
        }
-       dev->trans_start = jiffies;
-       return 0;
+       return NETDEV_TX_OK;
 
 abort_linearize:
        /* Free any DMA resources we've alloced and clear out the skb
@@ -2764,15 +2955,17 @@ abort_linearize:
 drop:
        dev_kfree_skb_any(skb);
        ss->stats.tx_dropped += 1;
-       return 0;
+       return NETDEV_TX_OK;
 
 }
 
-static int myri10ge_sw_tso(struct sk_buff *skb, struct net_device *dev)
+static netdev_tx_t myri10ge_sw_tso(struct sk_buff *skb,
+                                        struct net_device *dev)
 {
        struct sk_buff *segs, *curr;
        struct myri10ge_priv *mgp = netdev_priv(dev);
-       int status;
+       struct myri10ge_slice_state *ss;
+       netdev_tx_t status;
 
        segs = skb_gso_segment(skb, dev->features & ~NETIF_F_TSO6);
        if (IS_ERR(segs))
@@ -2795,21 +2988,23 @@ static int myri10ge_sw_tso(struct sk_buff *skb, struct net_device *dev)
                }
        }
        dev_kfree_skb_any(skb);
-       return 0;
+       return NETDEV_TX_OK;
 
 drop:
+       ss = &mgp->ss[skb_get_queue_mapping(skb)];
        dev_kfree_skb_any(skb);
-       mgp->stats.tx_dropped += 1;
-       return 0;
+       ss->stats.tx_dropped += 1;
+       return NETDEV_TX_OK;
 }
 
 static struct net_device_stats *myri10ge_get_stats(struct net_device *dev)
 {
        struct myri10ge_priv *mgp = netdev_priv(dev);
        struct myri10ge_slice_netstats *slice_stats;
-       struct net_device_stats *stats = &mgp->stats;
+       struct net_device_stats *stats = &dev->stats;
        int i;
 
+       spin_lock(&mgp->stats_lock);
        memset(stats, 0, sizeof(*stats));
        for (i = 0; i < mgp->num_slices; i++) {
                slice_stats = &mgp->ss[i].stats;
@@ -2820,6 +3015,7 @@ static struct net_device_stats *myri10ge_get_stats(struct net_device *dev)
                stats->rx_dropped += slice_stats->rx_dropped;
                stats->tx_dropped += slice_stats->tx_dropped;
        }
+       spin_unlock(&mgp->stats_lock);
        return stats;
 }
 
@@ -2830,7 +3026,6 @@ static void myri10ge_set_multicast_list(struct net_device *dev)
        struct dev_mc_list *mc_list;
        __be32 data[2] = { 0, 0 };
        int err;
-       DECLARE_MAC_BUF(mac);
 
        /* can be called from atomic contexts,
         * pass 1 to force atomicity in myri10ge_send_cmd() */
@@ -2877,8 +3072,7 @@ static void myri10ge_set_multicast_list(struct net_device *dev)
                        printk(KERN_ERR "myri10ge: %s: Failed "
                               "MXGEFW_JOIN_MULTICAST_GROUP, error status:"
                               "%d\t", dev->name, err);
-                       printk(KERN_ERR "MAC %s\n",
-                              print_mac(mac, mc_list->dmi_addr));
+                       printk(KERN_ERR "MAC %pM\n", mc_list->dmi_addr);
                        goto abort;
                }
        }
@@ -3096,6 +3290,8 @@ abort:
 
 static void myri10ge_select_firmware(struct myri10ge_priv *mgp)
 {
+       int overridden = 0;
+
        if (myri10ge_force_firmware == 0) {
                int link_width, exp_cap;
                u16 lnk;
@@ -3129,10 +3325,18 @@ static void myri10ge_select_firmware(struct myri10ge_priv *mgp)
                }
        }
        if (myri10ge_fw_name != NULL) {
-               dev_info(&mgp->pdev->dev, "overriding firmware to %s\n",
-                        myri10ge_fw_name);
+               overridden = 1;
                mgp->fw_name = myri10ge_fw_name;
        }
+       if (mgp->board_number < MYRI10GE_MAX_BOARDS &&
+           myri10ge_fw_names[mgp->board_number] != NULL &&
+           strlen(myri10ge_fw_names[mgp->board_number])) {
+               mgp->fw_name = myri10ge_fw_names[mgp->board_number];
+               overridden = 1;
+       }
+       if (overridden)
+               dev_info(&mgp->pdev->dev, "overriding firmware to %s\n",
+                        mgp->fw_name);
 }
 
 #ifdef CONFIG_PM
@@ -3243,12 +3447,13 @@ static void myri10ge_watchdog(struct work_struct *work)
            container_of(work, struct myri10ge_priv, watchdog_work);
        struct myri10ge_tx_buf *tx;
        u32 reboot;
-       int status;
+       int status, rebooted;
        int i;
        u16 cmd, vendor;
 
        mgp->watchdog_resets++;
        pci_read_config_word(mgp->pdev, PCI_COMMAND, &cmd);
+       rebooted = 0;
        if ((cmd & PCI_COMMAND_MASTER) == 0) {
                /* Bus master DMA disabled?  Check to see
                 * if the card rebooted due to a parity error
@@ -3260,9 +3465,12 @@ static void myri10ge_watchdog(struct work_struct *work)
                       myri10ge_reset_recover ? " " : " not");
                if (myri10ge_reset_recover == 0)
                        return;
-
+               rtnl_lock();
+               mgp->rebooted = 1;
+               rebooted = 1;
+               myri10ge_close(mgp->dev);
                myri10ge_reset_recover--;
-
+               mgp->rebooted = 0;
                /*
                 * A rebooted nic will come back with config space as
                 * it was after power was applied to PCIe bus.
@@ -3295,22 +3503,25 @@ static void myri10ge_watchdog(struct work_struct *work)
                for (i = 0; i < mgp->num_slices; i++) {
                        tx = &mgp->ss[i].tx;
                        printk(KERN_INFO
-                              "myri10ge: %s: (%d): %d %d %d %d %d\n",
-                              mgp->dev->name, i, tx->req, tx->done,
-                              tx->pkt_start, tx->pkt_done,
+                              "myri10ge: %s: (%d): %d %d %d %d %d %d\n",
+                              mgp->dev->name, i, tx->queue_active, tx->req,
+                              tx->done, tx->pkt_start, tx->pkt_done,
                               (int)ntohl(mgp->ss[i].fw_stats->
                                          send_done_count));
                        msleep(2000);
                        printk(KERN_INFO
-                              "myri10ge: %s: (%d): %d %d %d %d %d\n",
-                              mgp->dev->name, i, tx->req, tx->done,
-                              tx->pkt_start, tx->pkt_done,
+                              "myri10ge: %s: (%d): %d %d %d %d %d %d\n",
+                              mgp->dev->name, i, tx->queue_active, tx->req,
+                              tx->done, tx->pkt_start, tx->pkt_done,
                               (int)ntohl(mgp->ss[i].fw_stats->
                                          send_done_count));
                }
        }
-       rtnl_lock();
-       myri10ge_close(mgp->dev);
+
+       if (!rebooted) {
+               rtnl_lock();
+               myri10ge_close(mgp->dev);
+       }
        status = myri10ge_load_firmware(mgp, 1);
        if (status != 0)
                printk(KERN_ERR "myri10ge: %s: failed to load firmware\n",
@@ -3331,12 +3542,14 @@ static void myri10ge_watchdog_timer(unsigned long arg)
 {
        struct myri10ge_priv *mgp;
        struct myri10ge_slice_state *ss;
-       int i, reset_needed;
+       int i, reset_needed, busy_slice_cnt;
        u32 rx_pause_cnt;
+       u16 cmd;
 
        mgp = (struct myri10ge_priv *)arg;
 
        rx_pause_cnt = ntohl(mgp->ss[0].fw_stats->dropped_pause);
+       busy_slice_cnt = 0;
        for (i = 0, reset_needed = 0;
             i < mgp->num_slices && reset_needed == 0; ++i) {
 
@@ -3363,15 +3576,33 @@ static void myri10ge_watchdog_timer(unsigned long arg)
                        /* nic seems like it might be stuck.. */
                        if (rx_pause_cnt != mgp->watchdog_pause) {
                                if (net_ratelimit())
-                                       printk(KERN_WARNING "myri10ge %s:"
+                                       printk(KERN_WARNING
+                                              "myri10ge %s slice %d:"
                                               "TX paused, check link partner\n",
-                                              mgp->dev->name);
+                                              mgp->dev->name, i);
                        } else {
+                               printk(KERN_WARNING
+                                      "myri10ge %s slice %d stuck:",
+                                      mgp->dev->name, i);
                                reset_needed = 1;
                        }
                }
+               if (ss->watchdog_tx_done != ss->tx.done ||
+                   ss->watchdog_rx_done != ss->rx_done.cnt) {
+                       busy_slice_cnt++;
+               }
                ss->watchdog_tx_done = ss->tx.done;
                ss->watchdog_tx_req = ss->tx.req;
+               ss->watchdog_rx_done = ss->rx_done.cnt;
+       }
+       /* if we've sent or received no traffic, poll the NIC to
+        * ensure it is still there.  Otherwise, we risk not noticing
+        * an error in a timely fashion */
+       if (busy_slice_cnt == 0) {
+               pci_read_config_word(mgp->pdev, PCI_COMMAND, &cmd);
+               if ((cmd & PCI_COMMAND_MASTER) == 0) {
+                       reset_needed = 1;
+               }
        }
        mgp->watchdog_pause = rx_pause_cnt;
 
@@ -3476,7 +3707,11 @@ static void myri10ge_probe_slices(struct myri10ge_priv *mgp)
 
        /* try to load the slice aware rss firmware */
        old_fw = mgp->fw_name;
-       if (old_fw == myri10ge_fw_aligned)
+       if (myri10ge_fw_name != NULL) {
+               dev_info(&mgp->pdev->dev, "overriding rss firmware to %s\n",
+                        myri10ge_fw_name);
+               mgp->fw_name = myri10ge_fw_name;
+       } else if (old_fw == myri10ge_fw_aligned)
                mgp->fw_name = myri10ge_fw_rss_aligned;
        else
                mgp->fw_name = myri10ge_fw_rss_unaligned;
@@ -3568,6 +3803,17 @@ abort_with_fw:
        myri10ge_load_firmware(mgp, 0);
 }
 
+static const struct net_device_ops myri10ge_netdev_ops = {
+       .ndo_open               = myri10ge_open,
+       .ndo_stop               = myri10ge_close,
+       .ndo_start_xmit         = myri10ge_xmit,
+       .ndo_get_stats          = myri10ge_get_stats,
+       .ndo_validate_addr      = eth_validate_addr,
+       .ndo_change_mtu         = myri10ge_change_mtu,
+       .ndo_set_multicast_list = myri10ge_set_multicast_list,
+       .ndo_set_mac_address    = myri10ge_set_mac_address,
+};
+
 static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 {
        struct net_device *netdev;
@@ -3576,8 +3822,10 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
        int i;
        int status = -ENXIO;
        int dac_enabled;
+       unsigned hdr_offset, ss_offset;
+       static int board_number;
 
-       netdev = alloc_etherdev(sizeof(*mgp));
+       netdev = alloc_etherdev_mq(sizeof(*mgp), MYRI10GE_MAX_SLICES);
        if (netdev == NULL) {
                dev_err(dev, "Could not allocate ethernet device\n");
                return -ENOMEM;
@@ -3592,6 +3840,7 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
        mgp->pause = myri10ge_flow_control;
        mgp->intr_coal_delay = myri10ge_intr_coal_delay;
        mgp->msg_enable = netif_msg_init(myri10ge_debug, MYRI10GE_MSG_DEFAULT);
+       mgp->board_number = board_number;
        init_waitqueue_head(&mgp->down_wq);
 
        if (pci_enable_device(pdev)) {
@@ -3610,27 +3859,28 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
        if (status != 0) {
                dev_err(&pdev->dev, "Error %d writing PCI_EXP_DEVCTL\n",
                        status);
-               goto abort_with_netdev;
+               goto abort_with_enabled;
        }
 
        pci_set_master(pdev);
        dac_enabled = 1;
-       status = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
+       status = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
        if (status != 0) {
                dac_enabled = 0;
                dev_err(&pdev->dev,
                        "64-bit pci address mask was refused, "
                        "trying 32-bit\n");
-               status = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
+               status = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
        }
        if (status != 0) {
                dev_err(&pdev->dev, "Error %d setting DMA mask\n", status);
-               goto abort_with_netdev;
+               goto abort_with_enabled;
        }
+       (void)pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
        mgp->cmd = dma_alloc_coherent(&pdev->dev, sizeof(*mgp->cmd),
                                      &mgp->cmd_bus, GFP_KERNEL);
        if (mgp->cmd == NULL)
-               goto abort_with_netdev;
+               goto abort_with_enabled;
 
        mgp->board_span = pci_resource_len(pdev, 0);
        mgp->iomem_base = pci_resource_start(pdev, 0);
@@ -3642,24 +3892,26 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
        if (mgp->mtrr >= 0)
                mgp->wc_enabled = 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);
+       mgp->sram = ioremap_wc(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;
+               goto abort_with_mtrr;
+       }
+       hdr_offset =
+           ntohl(__raw_readl(mgp->sram + MCP_HEADER_PTR_OFFSET)) & 0xffffc;
+       ss_offset = hdr_offset + offsetof(struct mcp_gen_header, string_specs);
+       mgp->sram_size = ntohl(__raw_readl(mgp->sram + ss_offset));
+       if (mgp->sram_size > mgp->board_span ||
+           mgp->sram_size <= MYRI10GE_FW_OFFSET) {
+               dev_err(&pdev->dev,
+                       "invalid sram_size %dB or board span %ldB\n",
+                       mgp->sram_size, mgp->board_span);
+               goto abort_with_ioremap;
        }
        memcpy_fromio(mgp->eeprom_strings,
-                     mgp->sram + mgp->sram_size - MYRI10GE_EEPROM_STRINGS_SIZE,
-                     MYRI10GE_EEPROM_STRINGS_SIZE);
+                     mgp->sram + mgp->sram_size, MYRI10GE_EEPROM_STRINGS_SIZE);
        memset(mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE - 2, 0, 2);
        status = myri10ge_read_mac_addr(mgp);
        if (status)
@@ -3681,30 +3933,35 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
                dev_err(&pdev->dev, "failed to alloc slice state\n");
                goto abort_with_firmware;
        }
-
+       netdev->real_num_tx_queues = mgp->num_slices;
        status = myri10ge_reset(mgp);
        if (status != 0) {
                dev_err(&pdev->dev, "failed reset\n");
                goto abort_with_slices;
        }
-
+#ifdef CONFIG_MYRI10GE_DCA
+       myri10ge_setup_dca(mgp);
+#endif
        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->netdev_ops = &myri10ge_netdev_ops;
        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->change_mtu = myri10ge_change_mtu;
-       netdev->set_multicast_list = myri10ge_set_multicast_list;
-       netdev->set_mac_address = myri10ge_set_mac_address;
        netdev->features = mgp->features;
+
        if (dac_enabled)
                netdev->features |= NETIF_F_HIGHDMA;
+       netdev->features |= NETIF_F_LRO;
+
+       netdev->vlan_features |= mgp->features;
+       if (mgp->fw_ver_tiny < 37)
+               netdev->vlan_features &= ~NETIF_F_TSO6;
+       if (mgp->fw_ver_tiny < 32)
+               netdev->vlan_features &= ~NETIF_F_TSO;
 
        /* make sure we can get an irq, and that MSI can be
         * setup (if available).  Also ensure netdev->irq
@@ -3723,6 +3980,7 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
        setup_timer(&mgp->watchdog_timer, myri10ge_watchdog_timer,
                    (unsigned long)mgp);
 
+       spin_lock_init(&mgp->stats_lock);
        SET_ETHTOOL_OPS(netdev, &myri10ge_ethtool_ops);
        INIT_WORK(&mgp->watchdog_work, myri10ge_watchdog);
        status = register_netdev(netdev);
@@ -3740,6 +3998,7 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
                         netdev->irq, mgp->tx_boundary, mgp->fw_name,
                         (mgp->wc_enabled ? "Enabled" : "Disabled"));
 
+       board_number++;
        return 0;
 
 abort_with_state:
@@ -3752,9 +4011,13 @@ abort_with_firmware:
        myri10ge_dummy_rdma(mgp, 0);
 
 abort_with_ioremap:
+       if (mgp->mac_addr_string != NULL)
+               dev_err(&pdev->dev,
+                       "myri10ge_probe() failed: MAC=%s, SN=%ld\n",
+                       mgp->mac_addr_string, mgp->serial_number);
        iounmap(mgp->sram);
 
-abort_with_wc:
+abort_with_mtrr:
 #ifdef CONFIG_MTRR
        if (mgp->mtrr >= 0)
                mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
@@ -3762,8 +4025,10 @@ abort_with_wc:
        dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
                          mgp->cmd, mgp->cmd_bus);
 
-abort_with_netdev:
+abort_with_enabled:
+       pci_disable_device(pdev);
 
+abort_with_netdev:
        free_netdev(netdev);
        return status;
 }
@@ -3788,6 +4053,9 @@ static void myri10ge_remove(struct pci_dev *pdev)
        netdev = mgp->dev;
        unregister_netdev(netdev);
 
+#ifdef CONFIG_MYRI10GE_DCA
+       myri10ge_teardown_dca(mgp);
+#endif
        myri10ge_dummy_rdma(mgp, 0);
 
        /* avoid a memory leak */
@@ -3806,6 +4074,7 @@ static void myri10ge_remove(struct pci_dev *pdev)
                          mgp->cmd, mgp->cmd_bus);
 
        free_netdev(netdev);
+       pci_disable_device(pdev);
        pci_set_drvdata(pdev, NULL);
 }
 
@@ -3819,6 +4088,8 @@ static struct pci_device_id myri10ge_pci_tbl[] = {
        {0},
 };
 
+MODULE_DEVICE_TABLE(pci, myri10ge_pci_tbl);
+
 static struct pci_driver myri10ge_driver = {
        .name = "myri10ge",
        .probe = myri10ge_probe,
@@ -3830,18 +4101,42 @@ static struct pci_driver myri10ge_driver = {
 #endif
 };
 
+#ifdef CONFIG_MYRI10GE_DCA
+static int
+myri10ge_notify_dca(struct notifier_block *nb, unsigned long event, void *p)
+{
+       int err = driver_for_each_device(&myri10ge_driver.driver,
+                                        NULL, &event,
+                                        myri10ge_notify_dca_device);
+
+       if (err)
+               return NOTIFY_BAD;
+       return NOTIFY_DONE;
+}
+
+static struct notifier_block myri10ge_dca_notifier = {
+       .notifier_call = myri10ge_notify_dca,
+       .next = NULL,
+       .priority = 0,
+};
+#endif                         /* CONFIG_MYRI10GE_DCA */
+
 static __init int myri10ge_init_module(void)
 {
        printk(KERN_INFO "%s: Version %s\n", myri10ge_driver.name,
               MYRI10GE_VERSION_STR);
 
-       if (myri10ge_rss_hash > MXGEFW_RSS_HASH_TYPE_SRC_PORT ||
-           myri10ge_rss_hash < MXGEFW_RSS_HASH_TYPE_IPV4) {
+       if (myri10ge_rss_hash > MXGEFW_RSS_HASH_TYPE_MAX) {
                printk(KERN_ERR
                       "%s: Illegal rssh hash type %d, defaulting to source port\n",
                       myri10ge_driver.name, myri10ge_rss_hash);
                myri10ge_rss_hash = MXGEFW_RSS_HASH_TYPE_SRC_PORT;
        }
+#ifdef CONFIG_MYRI10GE_DCA
+       dca_register_notify(&myri10ge_dca_notifier);
+#endif
+       if (myri10ge_max_slices > MYRI10GE_MAX_SLICES)
+               myri10ge_max_slices = MYRI10GE_MAX_SLICES;
 
        return pci_register_driver(&myri10ge_driver);
 }
@@ -3850,6 +4145,9 @@ module_init(myri10ge_init_module);
 
 static __exit void myri10ge_cleanup_module(void)
 {
+#ifdef CONFIG_MYRI10GE_DCA
+       dca_unregister_notify(&myri10ge_dca_notifier);
+#endif
        pci_unregister_driver(&myri10ge_driver);
 }