myri10ge: add multislices support
[pandora-kernel.git] / drivers / net / myri10ge / myri10ge.c
index cead81e..eddcee3 100644 (file)
@@ -144,11 +144,13 @@ struct myri10ge_tx_buf {
        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 stop_queue;
+       int linearized;
        int done ____cacheline_aligned; /* transmit slots completed     */
        int pkt_done;           /* packets completed */
+       int wake_queue;
 };
 
 struct myri10ge_rx_done {
@@ -160,51 +162,72 @@ struct myri10ge_rx_done {
        struct net_lro_desc lro_desc[MYRI10GE_MAX_LRO_DESCRIPTORS];
 };
 
-struct myri10ge_priv {
-       int running;            /* running?             */
-       int csum_flag;          /* rx_csums?            */
+struct myri10ge_slice_netstats {
+       unsigned long rx_packets;
+       unsigned long tx_packets;
+       unsigned long rx_bytes;
+       unsigned long tx_bytes;
+       unsigned long rx_dropped;
+       unsigned long tx_dropped;
+};
+
+struct myri10ge_slice_state {
        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;
+       struct net_device *dev;
+       struct napi_struct napi;
+       struct myri10ge_priv *mgp;
+       struct myri10ge_slice_netstats stats;
+       __be32 __iomem *irq_claim;
+       struct mcp_irq_data *fw_stats;
+       dma_addr_t fw_stats_bus;
+       int watchdog_tx_done;
+       int watchdog_tx_req;
+       char irq_desc[32];
+};
+
+struct myri10ge_priv {
+       struct myri10ge_slice_state *ss;
+       int tx_boundary;        /* boundary transmits cannot cross */
+       int num_slices;
+       int running;            /* running?             */
+       int csum_flag;          /* rx_csums?            */
        int small_bytes;
        int big_bytes;
+       int max_intr_slots;
        struct net_device *dev;
-       struct napi_struct napi;
        struct net_device_stats stats;
+       spinlock_t stats_lock;
        u8 __iomem *sram;
        int sram_size;
        unsigned long board_span;
        unsigned long iomem_base;
-       __be32 __iomem *irq_claim;
        __be32 __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;
+       int msix_enabled;
+       struct msix_entry *msix_vectors;
        u32 link_state;
        unsigned int rdma_tags_available;
        int intr_coal_delay;
        __be32 __iomem *intr_coal_delay_ptr;
        int mtrr;
        int wc_enabled;
-       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_tx_req;
-       int watchdog_pause;
        int watchdog_resets;
-       int tx_linearized;
+       int watchdog_pause;
        int pause;
        char *fw_name;
        char eeprom_strings[MYRI10GE_EEPROM_STRINGS_SIZE];
+       char *product_code_string;
        char fw_version[128];
        int fw_ver_major;
        int fw_ver_minor;
@@ -225,61 +248,59 @@ struct myri10ge_priv {
 
 static char *myri10ge_fw_unaligned = "myri10ge_ethp_z8e.dat";
 static char *myri10ge_fw_aligned = "myri10ge_eth_z8e.dat";
+static char *myri10ge_fw_rss_unaligned = "myri10ge_rss_ethp_z8e.dat";
+static char *myri10ge_fw_rss_aligned = "myri10ge_rss_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");
+MODULE_PARM_DESC(myri10ge_fw_name, "Firmware image name");
 
 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");
+MODULE_PARM_DESC(myri10ge_ecrc_enable, "Enable Extended CRC on PCI-E");
 
 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");
+MODULE_PARM_DESC(myri10ge_small_bytes, "Threshold of small packets");
 
 static int myri10ge_msi = 1;   /* enable msi by default */
 module_param(myri10ge_msi, int, S_IRUGO | S_IWUSR);
-MODULE_PARM_DESC(myri10ge_msi, "Enable Message Signalled Interrupts\n");
+MODULE_PARM_DESC(myri10ge_msi, "Enable Message Signalled Interrupts");
 
 static int myri10ge_intr_coal_delay = 75;
 module_param(myri10ge_intr_coal_delay, int, S_IRUGO);
-MODULE_PARM_DESC(myri10ge_intr_coal_delay, "Interrupt coalescing delay\n");
+MODULE_PARM_DESC(myri10ge_intr_coal_delay, "Interrupt coalescing delay");
 
 static int myri10ge_flow_control = 1;
 module_param(myri10ge_flow_control, int, S_IRUGO);
-MODULE_PARM_DESC(myri10ge_flow_control, "Pause parameter\n");
+MODULE_PARM_DESC(myri10ge_flow_control, "Pause parameter");
 
 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");
+                "Wait when deasserting legacy interrupts");
 
 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");
+                "Force firmware to assume aligned completions");
 
 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");
+MODULE_PARM_DESC(myri10ge_initial_mtu, "Initial MTU");
 
 static int myri10ge_napi_weight = 64;
 module_param(myri10ge_napi_weight, int, S_IRUGO);
-MODULE_PARM_DESC(myri10ge_napi_weight, "Set NAPI weight\n");
+MODULE_PARM_DESC(myri10ge_napi_weight, "Set NAPI weight");
 
 static int myri10ge_watchdog_timeout = 1;
 module_param(myri10ge_watchdog_timeout, int, S_IRUGO);
-MODULE_PARM_DESC(myri10ge_watchdog_timeout, "Set watchdog timeout\n");
+MODULE_PARM_DESC(myri10ge_watchdog_timeout, "Set watchdog timeout");
 
 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");
+                "Set stuck legacy IRQ detection threshold");
 
 #define MYRI10GE_MSG_DEFAULT NETIF_MSG_LINK
 
@@ -289,21 +310,30 @@ 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\n");
+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, "Number of LRO packets to be aggregated\n");
+MODULE_PARM_DESC(myri10ge_lro_max_pkts,
+                "Number of LRO packets to be aggregated");
 
 static int myri10ge_fill_thresh = 256;
 module_param(myri10ge_fill_thresh, int, S_IRUGO | S_IWUSR);
-MODULE_PARM_DESC(myri10ge_fill_thresh, "Number of empty rx slots allowed\n");
+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\n");
+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");
+
+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");
 
 #define MYRI10GE_FW_OFFSET 1024*1024
 #define MYRI10GE_HIGHPART_TO_U32(X) \
@@ -359,8 +389,10 @@ myri10ge_send_cmd(struct myri10ge_priv *mgp, u32 cmd,
                for (sleep_total = 0;
                     sleep_total < 1000
                     && response->result == htonl(MYRI10GE_NO_RESPONSE_RESULT);
-                    sleep_total += 10)
+                    sleep_total += 10) {
                        udelay(10);
+                       mb();
+               }
        } else {
                /* use msleep for most command */
                for (sleep_total = 0;
@@ -420,6 +452,10 @@ static int myri10ge_read_mac_addr(struct myri10ge_priv *mgp)
                                ptr += 1;
                        }
                }
+               if (memcmp(ptr, "PC=", 3) == 0) {
+                       ptr += 3;
+                       mgp->product_code_string = ptr;
+               }
                if (memcmp((const void *)ptr, "SN=", 3) == 0) {
                        ptr += 3;
                        mgp->serial_number = simple_strtoul(ptr, &ptr, 10);
@@ -442,7 +478,7 @@ abort:
 static void myri10ge_dummy_rdma(struct myri10ge_priv *mgp, int enable)
 {
        char __iomem *submit;
-       __be32 buf[16];
+       __be32 buf[16] __attribute__ ((__aligned__(8)));
        u32 dma_low, dma_high;
        int i;
 
@@ -609,17 +645,44 @@ static int myri10ge_adopt_running_firmware(struct myri10ge_priv *mgp)
        return status;
 }
 
-static int myri10ge_load_firmware(struct myri10ge_priv *mgp)
+static int myri10ge_get_firmware_capabilities(struct myri10ge_priv *mgp)
+{
+       struct myri10ge_cmd cmd;
+       int status;
+
+       /* probe for IPv6 TSO support */
+       mgp->features = NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_TSO;
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_TSO6_HDR_SIZE,
+                                  &cmd, 0);
+       if (status == 0) {
+               mgp->max_tso6 = cmd.data0;
+               mgp->features |= NETIF_F_TSO6;
+       }
+
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0);
+       if (status != 0) {
+               dev_err(&mgp->pdev->dev,
+                       "failed MXGEFW_CMD_GET_RX_RING_SIZE\n");
+               return -ENXIO;
+       }
+
+       mgp->max_intr_slots = 2 * (cmd.data0 / sizeof(struct mcp_dma_addr));
+
+       return 0;
+}
+
+static int myri10ge_load_firmware(struct myri10ge_priv *mgp, int adopt)
 {
        char __iomem *submit;
-       __be32 buf[16];
+       __be32 buf[16] __attribute__ ((__aligned__(8)));
        u32 dma_low, dma_high, size;
        int status, i;
-       struct myri10ge_cmd cmd;
 
        size = 0;
        status = myri10ge_load_hotplug_firmware(mgp, &size);
        if (status) {
+               if (!adopt)
+                       return status;
                dev_warn(&mgp->pdev->dev, "hotplug firmware loading failed\n");
 
                /* Do not attempt to adopt firmware if there
@@ -635,7 +698,7 @@ static int myri10ge_load_firmware(struct myri10ge_priv *mgp)
                }
                dev_info(&mgp->pdev->dev,
                         "Successfully adopted running firmware\n");
-               if (mgp->tx.boundary == 4096) {
+               if (mgp->tx_boundary == 4096) {
                        dev_warn(&mgp->pdev->dev,
                                 "Using firmware currently running on NIC"
                                 ".  For optimal\n");
@@ -646,7 +709,9 @@ static int myri10ge_load_firmware(struct myri10ge_priv *mgp)
                }
 
                mgp->fw_name = "adopted";
-               mgp->tx.boundary = 2048;
+               mgp->tx_boundary = 2048;
+               myri10ge_dummy_rdma(mgp, 1);
+               status = myri10ge_get_firmware_capabilities(mgp);
                return status;
        }
 
@@ -681,26 +746,18 @@ static int myri10ge_load_firmware(struct myri10ge_priv *mgp)
        msleep(1);
        mb();
        i = 0;
-       while (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20) {
-               msleep(1);
+       while (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 9) {
+               msleep(1 << i);
                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, 1);
+       status = myri10ge_get_firmware_capabilities(mgp);
 
-       /* probe for IPv6 TSO support */
-       mgp->features = NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_TSO;
-       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_TSO6_HDR_SIZE,
-                                  &cmd, 0);
-       if (status == 0) {
-               mgp->max_tso6 = cmd.data0;
-               mgp->features |= NETIF_F_TSO6;
-       }
-       return 0;
+       return status;
 }
 
 static int myri10ge_update_mac_address(struct myri10ge_priv *mgp, u8 * addr)
@@ -772,7 +829,7 @@ static int myri10ge_dma_test(struct myri10ge_priv *mgp, int test_type)
         * transfers took to complete.
         */
 
-       len = mgp->tx.boundary;
+       len = mgp->tx_boundary;
 
        cmd.data0 = MYRI10GE_LOWPART_TO_U32(dmatest_bus);
        cmd.data1 = MYRI10GE_HIGHPART_TO_U32(dmatest_bus);
@@ -818,7 +875,8 @@ abort:
 static int myri10ge_reset(struct myri10ge_priv *mgp)
 {
        struct myri10ge_cmd cmd;
-       int status;
+       struct myri10ge_slice_state *ss;
+       int i, status;
        size_t bytes;
 
        /* try to send a reset command to the card to see if it
@@ -831,20 +889,74 @@ static int myri10ge_reset(struct myri10ge_priv *mgp)
        }
 
        (void)myri10ge_dma_test(mgp, MXGEFW_DMA_TEST);
+       /*
+        * Use non-ndis mcp_slot (eg, 4 bytes total,
+        * no toeplitz hash value returned.  Older firmware will
+        * not understand this command, but will use the correct
+        * sized mcp_slot, so we ignore error returns
+        */
+       cmd.data0 = MXGEFW_RSS_MCP_SLOT_TYPE_MIN;
+       (void)myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_MCP_SLOT_TYPE, &cmd, 0);
 
        /* Now exchange information about interrupts  */
 
-       bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
-       memset(mgp->rx_done.entry, 0, bytes);
+       bytes = mgp->max_intr_slots * sizeof(*mgp->ss[0].rx_done.entry);
        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);
+
+       /*
+        * Even though we already know how many slices are supported
+        * via myri10ge_probe_slices() MXGEFW_CMD_GET_MAX_RSS_QUEUES
+        * has magic side effects, and must be called after a reset.
+        * It must be called prior to calling any RSS related cmds,
+        * including assigning an interrupt queue for anything but
+        * slice 0.  It must also be called *after*
+        * MXGEFW_CMD_SET_INTRQ_SIZE, since the intrq size is used by
+        * the firmware to compute offsets.
+        */
+
+       if (mgp->num_slices > 1) {
+
+               /* ask the maximum number of slices it supports */
+               status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_RSS_QUEUES,
+                                          &cmd, 0);
+               if (status != 0) {
+                       dev_err(&mgp->pdev->dev,
+                               "failed to get number of slices\n");
+               }
+
+               /*
+                * MXGEFW_CMD_ENABLE_RSS_QUEUES must be called prior
+                * to setting up the interrupt queue DMA
+                */
+
+               cmd.data0 = mgp->num_slices;
+               cmd.data1 = 1;  /* use MSI-X */
+               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");
+
+                       return status;
+               }
+       }
+       for (i = 0; i < mgp->num_slices; i++) {
+               ss = &mgp->ss[i];
+               cmd.data0 = MYRI10GE_LOWPART_TO_U32(ss->rx_done.bus);
+               cmd.data1 = MYRI10GE_HIGHPART_TO_U32(ss->rx_done.bus);
+               cmd.data2 = i;
+               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 __be32 *) (mgp->sram + cmd.data0);
+       for (i = 0; i < mgp->num_slices; i++) {
+               ss = &mgp->ss[i];
+               ss->irq_claim =
+                   (__iomem __be32 *) (mgp->sram + cmd.data0 + 8 * i);
+       }
        status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET,
                                    &cmd, 0);
        mgp->irq_deassert = (__iomem __be32 *) (mgp->sram + cmd.data0);
@@ -858,18 +970,25 @@ static int myri10ge_reset(struct myri10ge_priv *mgp)
        }
        put_be32(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
 
-       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;
+
        mgp->link_changes = 0;
+       for (i = 0; i < mgp->num_slices; i++) {
+               ss = &mgp->ss[i];
+
+               memset(ss->rx_done.entry, 0, bytes);
+               ss->tx.req = 0;
+               ss->tx.done = 0;
+               ss->tx.pkt_start = 0;
+               ss->tx.pkt_done = 0;
+               ss->rx_big.cnt = 0;
+               ss->rx_small.cnt = 0;
+               ss->rx_done.idx = 0;
+               ss->rx_done.cnt = 0;
+               ss->tx.wake_queue = 0;
+               ss->tx.stop_queue = 0;
+       }
+
        status = myri10ge_update_mac_address(mgp, mgp->dev->dev_addr);
        myri10ge_change_pause(mgp, mgp->pause);
        myri10ge_set_multicast_list(mgp->dev);
@@ -1020,9 +1139,10 @@ myri10ge_unmap_rx_page(struct pci_dev *pdev,
                                 * page into an skb */
 
 static inline int
-myri10ge_rx_done(struct myri10ge_priv *mgp, struct myri10ge_rx_buf *rx,
+myri10ge_rx_done(struct myri10ge_slice_state *ss, struct myri10ge_rx_buf *rx,
                 int bytes, int len, __wsum csum)
 {
+       struct myri10ge_priv *mgp = ss->mgp;
        struct sk_buff *skb;
        struct skb_frag_struct rx_frags[MYRI10GE_MAX_FRAGS_PER_FRAME];
        int i, idx, hlen, remainder;
@@ -1052,11 +1172,11 @@ myri10ge_rx_done(struct myri10ge_priv *mgp, struct myri10ge_rx_buf *rx,
                rx_frags[0].page_offset += MXGEFW_PAD;
                rx_frags[0].size -= MXGEFW_PAD;
                len -= MXGEFW_PAD;
-               lro_receive_frags(&mgp->rx_done.lro_mgr, rx_frags,
+               lro_receive_frags(&ss->rx_done.lro_mgr, rx_frags,
+                                 /* opaque, will come back in get_frag_header */
                                  len, len,
-                                /* opaque, will come back in get_frag_header */
-                                 (void *)(__force unsigned long)csum,
-                                 csum);
+                                 (void *)(__force unsigned long)csum, csum);
+
                return 1;
        }
 
@@ -1096,10 +1216,11 @@ myri10ge_rx_done(struct myri10ge_priv *mgp, struct myri10ge_rx_buf *rx,
        return 1;
 }
 
-static inline void myri10ge_tx_done(struct myri10ge_priv *mgp, int mcp_index)
+static inline void
+myri10ge_tx_done(struct myri10ge_slice_state *ss, int mcp_index)
 {
-       struct pci_dev *pdev = mgp->pdev;
-       struct myri10ge_tx_buf *tx = &mgp->tx;
+       struct pci_dev *pdev = ss->mgp->pdev;
+       struct myri10ge_tx_buf *tx = &ss->tx;
        struct sk_buff *skb;
        int idx, len;
 
@@ -1117,8 +1238,8 @@ static inline void myri10ge_tx_done(struct myri10ge_priv *mgp, int mcp_index)
                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++;
+                       ss->stats.tx_bytes += skb->len;
+                       ss->stats.tx_packets++;
                        dev_kfree_skb_irq(skb);
                        if (len)
                                pci_unmap_single(pdev,
@@ -1134,16 +1255,18 @@ static inline void myri10ge_tx_done(struct myri10ge_priv *mgp, int mcp_index)
                }
        }
        /* start the queue if we've stopped it */
-       if (netif_queue_stopped(mgp->dev)
+       if (netif_queue_stopped(ss->dev)
            && tx->req - tx->done < (tx->mask >> 1)) {
-               mgp->wake_queue++;
-               netif_wake_queue(mgp->dev);
+               tx->wake_queue++;
+               netif_wake_queue(ss->dev);
        }
 }
 
-static inline int myri10ge_clean_rx_done(struct myri10ge_priv *mgp, int budget)
+static inline int
+myri10ge_clean_rx_done(struct myri10ge_slice_state *ss, int budget)
 {
-       struct myri10ge_rx_done *rx_done = &mgp->rx_done;
+       struct myri10ge_rx_done *rx_done = &ss->rx_done;
+       struct myri10ge_priv *mgp = ss->mgp;
        unsigned long rx_bytes = 0;
        unsigned long rx_packets = 0;
        unsigned long rx_ok;
@@ -1159,40 +1282,40 @@ static inline int myri10ge_clean_rx_done(struct myri10ge_priv *mgp, int budget)
                rx_done->entry[idx].length = 0;
                checksum = csum_unfold(rx_done->entry[idx].checksum);
                if (length <= mgp->small_bytes)
-                       rx_ok = myri10ge_rx_done(mgp, &mgp->rx_small,
+                       rx_ok = myri10ge_rx_done(ss, &ss->rx_small,
                                                 mgp->small_bytes,
                                                 length, checksum);
                else
-                       rx_ok = myri10ge_rx_done(mgp, &mgp->rx_big,
+                       rx_ok = myri10ge_rx_done(ss, &ss->rx_big,
                                                 mgp->big_bytes,
                                                 length, checksum);
                rx_packets += rx_ok;
                rx_bytes += rx_ok * (unsigned long)length;
                cnt++;
-               idx = cnt & (myri10ge_max_intr_slots - 1);
+               idx = cnt & (mgp->max_intr_slots - 1);
                work_done++;
        }
        rx_done->idx = idx;
        rx_done->cnt = cnt;
-       mgp->stats.rx_packets += rx_packets;
-       mgp->stats.rx_bytes += rx_bytes;
+       ss->stats.rx_packets += rx_packets;
+       ss->stats.rx_bytes += rx_bytes;
 
        if (myri10ge_lro)
                lro_flush_all(&rx_done->lro_mgr);
 
        /* restock receive rings if needed */
-       if (mgp->rx_small.fill_cnt - mgp->rx_small.cnt < myri10ge_fill_thresh)
-               myri10ge_alloc_rx_pages(mgp, &mgp->rx_small,
+       if (ss->rx_small.fill_cnt - ss->rx_small.cnt < myri10ge_fill_thresh)
+               myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
                                        mgp->small_bytes + MXGEFW_PAD, 0);
-       if (mgp->rx_big.fill_cnt - mgp->rx_big.cnt < myri10ge_fill_thresh)
-               myri10ge_alloc_rx_pages(mgp, &mgp->rx_big, mgp->big_bytes, 0);
+       if (ss->rx_big.fill_cnt - ss->rx_big.cnt < myri10ge_fill_thresh)
+               myri10ge_alloc_rx_pages(mgp, &ss->rx_big, mgp->big_bytes, 0);
 
        return work_done;
 }
 
 static inline void myri10ge_check_statblock(struct myri10ge_priv *mgp)
 {
-       struct mcp_irq_data *stats = mgp->fw_stats;
+       struct mcp_irq_data *stats = mgp->ss[0].fw_stats;
 
        if (unlikely(stats->stats_updated)) {
                unsigned link_up = ntohl(stats->link_up);
@@ -1219,9 +1342,9 @@ static inline void myri10ge_check_statblock(struct myri10ge_priv *mgp)
                        }
                }
                if (mgp->rdma_tags_available !=
-                   ntohl(mgp->fw_stats->rdma_tags_available)) {
+                   ntohl(stats->rdma_tags_available)) {
                        mgp->rdma_tags_available =
-                           ntohl(mgp->fw_stats->rdma_tags_available);
+                           ntohl(stats->rdma_tags_available);
                        printk(KERN_WARNING "myri10ge: %s: RDMA timed out! "
                               "%d tags left\n", mgp->dev->name,
                               mgp->rdma_tags_available);
@@ -1234,29 +1357,37 @@ static inline void myri10ge_check_statblock(struct myri10ge_priv *mgp)
 
 static int myri10ge_poll(struct napi_struct *napi, int budget)
 {
-       struct myri10ge_priv *mgp =
-           container_of(napi, struct myri10ge_priv, napi);
-       struct net_device *netdev = mgp->dev;
+       struct myri10ge_slice_state *ss =
+           container_of(napi, struct myri10ge_slice_state, napi);
+       struct net_device *netdev = ss->mgp->dev;
        int work_done;
 
        /* process as many rx events as NAPI will allow */
-       work_done = myri10ge_clean_rx_done(mgp, budget);
+       work_done = myri10ge_clean_rx_done(ss, budget);
 
        if (work_done < budget) {
                netif_rx_complete(netdev, napi);
-               put_be32(htonl(3), mgp->irq_claim);
+               put_be32(htonl(3), ss->irq_claim);
        }
        return work_done;
 }
 
 static irqreturn_t myri10ge_intr(int irq, void *arg)
 {
-       struct myri10ge_priv *mgp = arg;
-       struct mcp_irq_data *stats = mgp->fw_stats;
-       struct myri10ge_tx_buf *tx = &mgp->tx;
+       struct myri10ge_slice_state *ss = arg;
+       struct myri10ge_priv *mgp = ss->mgp;
+       struct mcp_irq_data *stats = ss->fw_stats;
+       struct myri10ge_tx_buf *tx = &ss->tx;
        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);
+               return (IRQ_HANDLED);
+       }
+
        /* make sure it is our IRQ, and that the DMA has finished */
        if (unlikely(!stats->valid))
                return (IRQ_NONE);
@@ -1264,9 +1395,9 @@ 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(mgp->dev, &mgp->napi);
+               netif_rx_schedule(ss->dev, &ss->napi);
 
-       if (!mgp->msi_enabled) {
+       if (!mgp->msi_enabled && !mgp->msix_enabled) {
                put_be32(0, mgp->irq_deassert);
                if (!myri10ge_deassert_wait)
                        stats->valid = 0;
@@ -1281,7 +1412,7 @@ static irqreturn_t myri10ge_intr(int irq, void *arg)
                /* 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);
+                       myri10ge_tx_done(ss, (int)send_done_count);
                if (unlikely(i > myri10ge_max_irq_loops)) {
                        printk(KERN_WARNING "myri10ge: %s: irq stuck?\n",
                               mgp->dev->name);
@@ -1296,16 +1427,46 @@ static irqreturn_t myri10ge_intr(int irq, void *arg)
 
        myri10ge_check_statblock(mgp);
 
-       put_be32(htonl(3), mgp->irq_claim + 1);
+       put_be32(htonl(3), ss->irq_claim + 1);
        return (IRQ_HANDLED);
 }
 
 static int
 myri10ge_get_settings(struct net_device *netdev, struct ethtool_cmd *cmd)
 {
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+       char *ptr;
+       int i;
+
        cmd->autoneg = AUTONEG_DISABLE;
        cmd->speed = SPEED_10000;
        cmd->duplex = DUPLEX_FULL;
+
+       /*
+        * parse the product code to deterimine the interface type
+        * (CX4, XFP, Quad Ribbon Fiber) by looking at the character
+        * after the 3rd dash in the driver's cached copy of the
+        * EEPROM's product code string.
+        */
+       ptr = mgp->product_code_string;
+       if (ptr == NULL) {
+               printk(KERN_ERR "myri10ge: %s: Missing product code\n",
+                      netdev->name);
+               return 0;
+       }
+       for (i = 0; i < 3; i++, ptr++) {
+               ptr = strchr(ptr, '-');
+               if (ptr == NULL) {
+                       printk(KERN_ERR "myri10ge: %s: Invalid product "
+                              "code %s\n", netdev->name,
+                              mgp->product_code_string);
+                       return 0;
+               }
+       }
+       if (*ptr == 'R' || *ptr == 'Q') {
+               /* We've found either an XFP or quad ribbon fiber */
+               cmd->port = PORT_FIBRE;
+       }
        return 0;
 }
 
@@ -1324,6 +1485,7 @@ 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;
 }
@@ -1370,10 +1532,10 @@ myri10ge_get_ringparam(struct net_device *netdev,
 {
        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_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->rx_small.mask + 1;
+       ring->tx_max_pending = mgp->ss[0].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;
@@ -1383,6 +1545,7 @@ myri10ge_get_ringparam(struct net_device *netdev,
 static u32 myri10ge_get_rx_csum(struct net_device *netdev)
 {
        struct myri10ge_priv *mgp = netdev_priv(netdev);
+
        if (mgp->csum_flag)
                return 1;
        else
@@ -1392,6 +1555,7 @@ 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);
+
        if (csum_enabled)
                mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
        else
@@ -1411,7 +1575,7 @@ static int myri10ge_set_tso(struct net_device *netdev, u32 tso_enabled)
        return 0;
 }
 
-static const char myri10ge_gstrings_stats[][ETH_GSTRING_LEN] = {
+static const char myri10ge_gstrings_main_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",
@@ -1419,39 +1583,58 @@ static const char myri10ge_gstrings_stats[][ETH_GSTRING_LEN] = {
        "tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors",
        "tx_heartbeat_errors", "tx_window_errors",
        /* device-specific stats */
-       "tx_boundary", "WC", "irq", "MSI",
+       "tx_boundary", "WC", "irq", "MSI", "MSIX",
        "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",
+       "serial_number", "watchdog_resets",
        "link_changes", "link_up", "dropped_link_overflow",
        "dropped_link_error_or_filtered",
        "dropped_pause", "dropped_bad_phy", "dropped_bad_crc32",
        "dropped_unicast_filtered", "dropped_multicast_filtered",
        "dropped_runt", "dropped_overrun", "dropped_no_small_buffer",
-       "dropped_no_big_buffer", "LRO aggregated", "LRO flushed",
+       "dropped_no_big_buffer"
+};
+
+static const char myri10ge_gstrings_slice_stats[][ETH_GSTRING_LEN] = {
+       "----------- slice ---------",
+       "tx_pkt_start", "tx_pkt_done", "tx_req", "tx_done",
+       "rx_small_cnt", "rx_big_cnt",
+       "wake_queue", "stop_queue", "tx_linearized", "LRO aggregated",
+           "LRO flushed",
        "LRO avg aggr", "LRO no_desc"
 };
 
 #define MYRI10GE_NET_STATS_LEN      21
-#define MYRI10GE_STATS_LEN     ARRAY_SIZE(myri10ge_gstrings_stats)
+#define MYRI10GE_MAIN_STATS_LEN  ARRAY_SIZE(myri10ge_gstrings_main_stats)
+#define MYRI10GE_SLICE_STATS_LEN  ARRAY_SIZE(myri10ge_gstrings_slice_stats)
 
 static void
 myri10ge_get_strings(struct net_device *netdev, u32 stringset, u8 * data)
 {
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+       int i;
+
        switch (stringset) {
        case ETH_SS_STATS:
-               memcpy(data, *myri10ge_gstrings_stats,
-                      sizeof(myri10ge_gstrings_stats));
+               memcpy(data, *myri10ge_gstrings_main_stats,
+                      sizeof(myri10ge_gstrings_main_stats));
+               data += sizeof(myri10ge_gstrings_main_stats);
+               for (i = 0; i < mgp->num_slices; i++) {
+                       memcpy(data, *myri10ge_gstrings_slice_stats,
+                              sizeof(myri10ge_gstrings_slice_stats));
+                       data += sizeof(myri10ge_gstrings_slice_stats);
+               }
                break;
        }
 }
 
 static int myri10ge_get_sset_count(struct net_device *netdev, int sset)
 {
+       struct myri10ge_priv *mgp = netdev_priv(netdev);
+
        switch (sset) {
        case ETH_SS_STATS:
-               return MYRI10GE_STATS_LEN;
+               return MYRI10GE_MAIN_STATS_LEN +
+                   mgp->num_slices * MYRI10GE_SLICE_STATS_LEN;
        default:
                return -EOPNOTSUPP;
        }
@@ -1462,53 +1645,63 @@ myri10ge_get_ethtool_stats(struct net_device *netdev,
                           struct ethtool_stats *stats, u64 * data)
 {
        struct myri10ge_priv *mgp = netdev_priv(netdev);
+       struct myri10ge_slice_state *ss;
+       int slice;
        int i;
 
        for (i = 0; i < MYRI10GE_NET_STATS_LEN; i++)
                data[i] = ((unsigned long *)&mgp->stats)[i];
 
-       data[i++] = (unsigned int)mgp->tx.boundary;
+       data[i++] = (unsigned int)mgp->tx_boundary;
        data[i++] = (unsigned int)mgp->wc_enabled;
        data[i++] = (unsigned int)mgp->pdev->irq;
        data[i++] = (unsigned int)mgp->msi_enabled;
+       data[i++] = (unsigned int)mgp->msix_enabled;
        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)mgp->link_changes;
-       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_pause);
-       data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_bad_phy);
-       data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_bad_crc32);
+
+       /* firmware stats are useful only in the first slice */
+       ss = &mgp->ss[0];
+       data[i++] = (unsigned int)ntohl(ss->fw_stats->link_up);
+       data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_link_overflow);
        data[i++] =
-           (unsigned int)ntohl(mgp->fw_stats->dropped_unicast_filtered);
+           (unsigned int)ntohl(ss->fw_stats->dropped_link_error_or_filtered);
+       data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_pause);
+       data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_bad_phy);
+       data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_bad_crc32);
+       data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_unicast_filtered);
        data[i++] =
-           (unsigned int)ntohl(mgp->fw_stats->dropped_multicast_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);
-       data[i++] = mgp->rx_done.lro_mgr.stats.aggregated;
-       data[i++] = mgp->rx_done.lro_mgr.stats.flushed;
-       if (mgp->rx_done.lro_mgr.stats.flushed)
-               data[i++] = mgp->rx_done.lro_mgr.stats.aggregated /
-                   mgp->rx_done.lro_mgr.stats.flushed;
-       else
-               data[i++] = 0;
-       data[i++] = mgp->rx_done.lro_mgr.stats.no_desc;
+           (unsigned int)ntohl(ss->fw_stats->dropped_multicast_filtered);
+       data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_runt);
+       data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_overrun);
+       data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_no_small_buffer);
+       data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_no_big_buffer);
+
+       for (slice = 0; slice < mgp->num_slices; slice++) {
+               ss = &mgp->ss[slice];
+               data[i++] = slice;
+               data[i++] = (unsigned int)ss->tx.pkt_start;
+               data[i++] = (unsigned int)ss->tx.pkt_done;
+               data[i++] = (unsigned int)ss->tx.req;
+               data[i++] = (unsigned int)ss->tx.done;
+               data[i++] = (unsigned int)ss->rx_small.cnt;
+               data[i++] = (unsigned int)ss->rx_big.cnt;
+               data[i++] = (unsigned int)ss->tx.wake_queue;
+               data[i++] = (unsigned int)ss->tx.stop_queue;
+               data[i++] = (unsigned int)ss->tx.linearized;
+               data[i++] = ss->rx_done.lro_mgr.stats.aggregated;
+               data[i++] = ss->rx_done.lro_mgr.stats.flushed;
+               if (ss->rx_done.lro_mgr.stats.flushed)
+                       data[i++] = ss->rx_done.lro_mgr.stats.aggregated /
+                           ss->rx_done.lro_mgr.stats.flushed;
+               else
+                       data[i++] = 0;
+               data[i++] = ss->rx_done.lro_mgr.stats.no_desc;
+       }
 }
 
 static void myri10ge_set_msglevel(struct net_device *netdev, u32 value)
@@ -1544,21 +1737,22 @@ static const struct ethtool_ops myri10ge_ethtool_ops = {
        .get_msglevel = myri10ge_get_msglevel
 };
 
-static int myri10ge_allocate_rings(struct net_device *dev)
+static int myri10ge_allocate_rings(struct myri10ge_slice_state *ss)
 {
-       struct myri10ge_priv *mgp;
+       struct myri10ge_priv *mgp = ss->mgp;
        struct myri10ge_cmd cmd;
+       struct net_device *dev = mgp->dev;
        int tx_ring_size, rx_ring_size;
        int tx_ring_entries, rx_ring_entries;
-       int i, status;
+       int i, slice, status;
        size_t bytes;
 
-       mgp = netdev_priv(dev);
-
        /* get ring sizes */
-
+       slice = ss - mgp->ss;
+       cmd.data0 = slice;
        status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_RING_SIZE, &cmd, 0);
        tx_ring_size = cmd.data0;
+       cmd.data0 = slice;
        status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0);
        if (status != 0)
                return status;
@@ -1566,144 +1760,148 @@ static int myri10ge_allocate_rings(struct net_device *dev)
 
        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;
+       ss->tx.mask = tx_ring_entries - 1;
+       ss->rx_small.mask = ss->rx_big.mask = rx_ring_entries - 1;
 
        status = -ENOMEM;
 
        /* 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)
+           * sizeof(*ss->tx.req_list);
+       ss->tx.req_bytes = kzalloc(bytes, GFP_KERNEL);
+       if (ss->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);
+       ss->tx.req_list = (struct mcp_kreq_ether_send *)
+           ALIGN((unsigned long)ss->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)
+       bytes = rx_ring_entries * sizeof(*ss->rx_small.shadow);
+       ss->rx_small.shadow = kzalloc(bytes, GFP_KERNEL);
+       if (ss->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)
+       bytes = rx_ring_entries * sizeof(*ss->rx_big.shadow);
+       ss->rx_big.shadow = kzalloc(bytes, GFP_KERNEL);
+       if (ss->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)
+       bytes = tx_ring_entries * sizeof(*ss->tx.info);
+       ss->tx.info = kzalloc(bytes, GFP_KERNEL);
+       if (ss->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)
+       bytes = rx_ring_entries * sizeof(*ss->rx_small.info);
+       ss->rx_small.info = kzalloc(bytes, GFP_KERNEL);
+       if (ss->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)
+       bytes = rx_ring_entries * sizeof(*ss->rx_big.info);
+       ss->rx_big.info = kzalloc(bytes, GFP_KERNEL);
+       if (ss->rx_big.info == NULL)
                goto abort_with_rx_small_info;
 
        /* Fill the receive rings */
-       mgp->rx_big.cnt = 0;
-       mgp->rx_small.cnt = 0;
-       mgp->rx_big.fill_cnt = 0;
-       mgp->rx_small.fill_cnt = 0;
-       mgp->rx_small.page_offset = MYRI10GE_ALLOC_SIZE;
-       mgp->rx_big.page_offset = MYRI10GE_ALLOC_SIZE;
-       mgp->rx_small.watchdog_needed = 0;
-       mgp->rx_big.watchdog_needed = 0;
-       myri10ge_alloc_rx_pages(mgp, &mgp->rx_small,
+       ss->rx_big.cnt = 0;
+       ss->rx_small.cnt = 0;
+       ss->rx_big.fill_cnt = 0;
+       ss->rx_small.fill_cnt = 0;
+       ss->rx_small.page_offset = MYRI10GE_ALLOC_SIZE;
+       ss->rx_big.page_offset = MYRI10GE_ALLOC_SIZE;
+       ss->rx_small.watchdog_needed = 0;
+       ss->rx_big.watchdog_needed = 0;
+       myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
                                mgp->small_bytes + MXGEFW_PAD, 0);
 
-       if (mgp->rx_small.fill_cnt < mgp->rx_small.mask + 1) {
-               printk(KERN_ERR "myri10ge: %s: alloced only %d small bufs\n",
-                      dev->name, mgp->rx_small.fill_cnt);
+       if (ss->rx_small.fill_cnt < ss->rx_small.mask + 1) {
+               printk(KERN_ERR
+                      "myri10ge: %s:slice-%d: alloced only %d small bufs\n",
+                      dev->name, slice, ss->rx_small.fill_cnt);
                goto abort_with_rx_small_ring;
        }
 
-       myri10ge_alloc_rx_pages(mgp, &mgp->rx_big, mgp->big_bytes, 0);
-       if (mgp->rx_big.fill_cnt < mgp->rx_big.mask + 1) {
-               printk(KERN_ERR "myri10ge: %s: alloced only %d big bufs\n",
-                      dev->name, mgp->rx_big.fill_cnt);
+       myri10ge_alloc_rx_pages(mgp, &ss->rx_big, mgp->big_bytes, 0);
+       if (ss->rx_big.fill_cnt < ss->rx_big.mask + 1) {
+               printk(KERN_ERR
+                      "myri10ge: %s:slice-%d: alloced only %d big bufs\n",
+                      dev->name, slice, ss->rx_big.fill_cnt);
                goto abort_with_rx_big_ring;
        }
 
        return 0;
 
 abort_with_rx_big_ring:
-       for (i = mgp->rx_big.cnt; i < mgp->rx_big.fill_cnt; i++) {
-               int idx = i & mgp->rx_big.mask;
-               myri10ge_unmap_rx_page(mgp->pdev, &mgp->rx_big.info[idx],
+       for (i = ss->rx_big.cnt; i < ss->rx_big.fill_cnt; i++) {
+               int idx = i & ss->rx_big.mask;
+               myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_big.info[idx],
                                       mgp->big_bytes);
-               put_page(mgp->rx_big.info[idx].page);
+               put_page(ss->rx_big.info[idx].page);
        }
 
 abort_with_rx_small_ring:
-       for (i = mgp->rx_small.cnt; i < mgp->rx_small.fill_cnt; i++) {
-               int idx = i & mgp->rx_small.mask;
-               myri10ge_unmap_rx_page(mgp->pdev, &mgp->rx_small.info[idx],
+       for (i = ss->rx_small.cnt; i < ss->rx_small.fill_cnt; i++) {
+               int idx = i & ss->rx_small.mask;
+               myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_small.info[idx],
                                       mgp->small_bytes + MXGEFW_PAD);
-               put_page(mgp->rx_small.info[idx].page);
+               put_page(ss->rx_small.info[idx].page);
        }
 
-       kfree(mgp->rx_big.info);
+       kfree(ss->rx_big.info);
 
 abort_with_rx_small_info:
-       kfree(mgp->rx_small.info);
+       kfree(ss->rx_small.info);
 
 abort_with_tx_info:
-       kfree(mgp->tx.info);
+       kfree(ss->tx.info);
 
 abort_with_rx_big_shadow:
-       kfree(mgp->rx_big.shadow);
+       kfree(ss->rx_big.shadow);
 
 abort_with_rx_small_shadow:
-       kfree(mgp->rx_small.shadow);
+       kfree(ss->rx_small.shadow);
 
 abort_with_tx_req_bytes:
-       kfree(mgp->tx.req_bytes);
-       mgp->tx.req_bytes = NULL;
-       mgp->tx.req_list = NULL;
+       kfree(ss->tx.req_bytes);
+       ss->tx.req_bytes = NULL;
+       ss->tx.req_list = NULL;
 
 abort_with_nothing:
        return status;
 }
 
-static void myri10ge_free_rings(struct net_device *dev)
+static void myri10ge_free_rings(struct myri10ge_slice_state *ss)
 {
-       struct myri10ge_priv *mgp;
+       struct myri10ge_priv *mgp = ss->mgp;
        struct sk_buff *skb;
        struct myri10ge_tx_buf *tx;
        int i, len, idx;
 
-       mgp = netdev_priv(dev);
+       /* If not allocated, skip it */
+       if (ss->tx.req_list == NULL)
+               return;
 
-       for (i = mgp->rx_big.cnt; i < mgp->rx_big.fill_cnt; i++) {
-               idx = i & mgp->rx_big.mask;
-               if (i == mgp->rx_big.fill_cnt - 1)
-                       mgp->rx_big.info[idx].page_offset = MYRI10GE_ALLOC_SIZE;
-               myri10ge_unmap_rx_page(mgp->pdev, &mgp->rx_big.info[idx],
+       for (i = ss->rx_big.cnt; i < ss->rx_big.fill_cnt; i++) {
+               idx = i & ss->rx_big.mask;
+               if (i == ss->rx_big.fill_cnt - 1)
+                       ss->rx_big.info[idx].page_offset = MYRI10GE_ALLOC_SIZE;
+               myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_big.info[idx],
                                       mgp->big_bytes);
-               put_page(mgp->rx_big.info[idx].page);
+               put_page(ss->rx_big.info[idx].page);
        }
 
-       for (i = mgp->rx_small.cnt; i < mgp->rx_small.fill_cnt; i++) {
-               idx = i & mgp->rx_small.mask;
-               if (i == mgp->rx_small.fill_cnt - 1)
-                       mgp->rx_small.info[idx].page_offset =
+       for (i = ss->rx_small.cnt; i < ss->rx_small.fill_cnt; i++) {
+               idx = i & ss->rx_small.mask;
+               if (i == ss->rx_small.fill_cnt - 1)
+                       ss->rx_small.info[idx].page_offset =
                            MYRI10GE_ALLOC_SIZE;
-               myri10ge_unmap_rx_page(mgp->pdev, &mgp->rx_small.info[idx],
+               myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_small.info[idx],
                                       mgp->small_bytes + MXGEFW_PAD);
-               put_page(mgp->rx_small.info[idx].page);
+               put_page(ss->rx_small.info[idx].page);
        }
-       tx = &mgp->tx;
+       tx = &ss->tx;
        while (tx->done != tx->req) {
                idx = tx->done & tx->mask;
                skb = tx->info[idx].skb;
@@ -1714,7 +1912,7 @@ static void myri10ge_free_rings(struct net_device *dev)
                len = pci_unmap_len(&tx->info[idx], len);
                pci_unmap_len_set(&tx->info[idx], len, 0);
                if (skb) {
-                       mgp->stats.tx_dropped++;
+                       ss->stats.tx_dropped++;
                        dev_kfree_skb_any(skb);
                        if (len)
                                pci_unmap_single(mgp->pdev,
@@ -1729,43 +1927,85 @@ static void myri10ge_free_rings(struct net_device *dev)
                                               PCI_DMA_TODEVICE);
                }
        }
-       kfree(mgp->rx_big.info);
+       kfree(ss->rx_big.info);
 
-       kfree(mgp->rx_small.info);
+       kfree(ss->rx_small.info);
 
-       kfree(mgp->tx.info);
+       kfree(ss->tx.info);
 
-       kfree(mgp->rx_big.shadow);
+       kfree(ss->rx_big.shadow);
 
-       kfree(mgp->rx_small.shadow);
+       kfree(ss->rx_small.shadow);
 
-       kfree(mgp->tx.req_bytes);
-       mgp->tx.req_bytes = NULL;
-       mgp->tx.req_list = NULL;
+       kfree(ss->tx.req_bytes);
+       ss->tx.req_bytes = NULL;
+       ss->tx.req_list = NULL;
 }
 
 static int myri10ge_request_irq(struct myri10ge_priv *mgp)
 {
        struct pci_dev *pdev = mgp->pdev;
+       struct myri10ge_slice_state *ss;
+       struct net_device *netdev = mgp->dev;
+       int i;
        int status;
 
+       mgp->msi_enabled = 0;
+       mgp->msix_enabled = 0;
+       status = 0;
        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;
-       } else {
-               mgp->msi_enabled = 0;
+               if (mgp->num_slices > 1) {
+                       status =
+                           pci_enable_msix(pdev, mgp->msix_vectors,
+                                           mgp->num_slices);
+                       if (status == 0) {
+                               mgp->msix_enabled = 1;
+                       } else {
+                               dev_err(&pdev->dev,
+                                       "Error %d setting up MSI-X\n", status);
+                               return status;
+                       }
+               }
+               if (mgp->msix_enabled == 0) {
+                       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, IRQF_SHARED,
-                            mgp->dev->name, mgp);
-       if (status != 0) {
-               dev_err(&pdev->dev, "failed to allocate IRQ\n");
-               if (mgp->msi_enabled)
-                       pci_disable_msi(pdev);
+       if (mgp->msix_enabled) {
+               for (i = 0; i < mgp->num_slices; i++) {
+                       ss = &mgp->ss[i];
+                       snprintf(ss->irq_desc, sizeof(ss->irq_desc),
+                                "%s:slice-%d", netdev->name, i);
+                       status = request_irq(mgp->msix_vectors[i].vector,
+                                            myri10ge_intr, 0, ss->irq_desc,
+                                            ss);
+                       if (status != 0) {
+                               dev_err(&pdev->dev,
+                                       "slice %d failed to allocate IRQ\n", i);
+                               i--;
+                               while (i >= 0) {
+                                       free_irq(mgp->msix_vectors[i].vector,
+                                                &mgp->ss[i]);
+                                       i--;
+                               }
+                               pci_disable_msix(pdev);
+                               return status;
+                       }
+               }
+       } else {
+               status = request_irq(pdev->irq, myri10ge_intr, IRQF_SHARED,
+                                    mgp->dev->name, &mgp->ss[0]);
+               if (status != 0) {
+                       dev_err(&pdev->dev, "failed to allocate IRQ\n");
+                       if (mgp->msi_enabled)
+                               pci_disable_msi(pdev);
+               }
        }
        return status;
 }
@@ -1773,10 +2013,18 @@ static int myri10ge_request_irq(struct myri10ge_priv *mgp)
 static void myri10ge_free_irq(struct myri10ge_priv *mgp)
 {
        struct pci_dev *pdev = mgp->pdev;
+       int i;
 
-       free_irq(pdev->irq, mgp);
+       if (mgp->msix_enabled) {
+               for (i = 0; i < mgp->num_slices; i++)
+                       free_irq(mgp->msix_vectors[i].vector, &mgp->ss[i]);
+       } else {
+               free_irq(pdev->irq, &mgp->ss[0]);
+       }
        if (mgp->msi_enabled)
                pci_disable_msi(pdev);
+       if (mgp->msix_enabled)
+               pci_disable_msix(pdev);
 }
 
 static int
@@ -1838,14 +2086,82 @@ myri10ge_get_frag_header(struct skb_frag_struct *frag, void **mac_hdr,
        return 0;
 }
 
+static int myri10ge_get_txrx(struct myri10ge_priv *mgp, int slice)
+{
+       struct myri10ge_cmd cmd;
+       struct myri10ge_slice_state *ss;
+       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);
+
+       cmd.data0 = slice;
+       status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SMALL_RX_OFFSET,
+                                   &cmd, 0);
+       ss->rx_small.lanai = (struct mcp_kreq_ether_recv __iomem *)
+           (mgp->sram + cmd.data0);
+
+       cmd.data0 = slice;
+       status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_BIG_RX_OFFSET, &cmd, 0);
+       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;
+       }
+       return status;
+
+}
+
+static int myri10ge_set_stats(struct myri10ge_priv *mgp, int slice)
+{
+       struct myri10ge_cmd cmd;
+       struct myri10ge_slice_state *ss;
+       int status;
+
+       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);
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA_V2, &cmd, 0);
+       if (status == -ENOSYS) {
+               dma_addr_t bus = ss->fw_stats_bus;
+               if (slice != 0)
+                       return -EINVAL;
+               bus += offsetof(struct mcp_irq_data, send_done_count);
+               cmd.data0 = MYRI10GE_LOWPART_TO_U32(bus);
+               cmd.data1 = MYRI10GE_HIGHPART_TO_U32(bus);
+               status = myri10ge_send_cmd(mgp,
+                                          MXGEFW_CMD_SET_STATS_DMA_OBSOLETE,
+                                          &cmd, 0);
+               /* Firmware cannot support multicast without STATS_DMA_V2 */
+               mgp->fw_multicast_support = 0;
+       } else {
+               mgp->fw_multicast_support = 1;
+       }
+       return 0;
+}
+
 static int myri10ge_open(struct net_device *dev)
 {
-       struct myri10ge_priv *mgp;
+       struct myri10ge_slice_state *ss;
+       struct myri10ge_priv *mgp = netdev_priv(dev);
        struct myri10ge_cmd cmd;
+       int i, status, big_pow2, slice;
+       u8 *itable;
        struct net_lro_mgr *lro_mgr;
-       int status, big_pow2;
-
-       mgp = netdev_priv(dev);
 
        if (mgp->running != MYRI10GE_ETH_STOPPED)
                return -EBUSY;
@@ -1857,6 +2173,48 @@ static int myri10ge_open(struct net_device *dev)
                goto abort_with_nothing;
        }
 
+       if (mgp->num_slices > 1) {
+               cmd.data0 = mgp->num_slices;
+               cmd.data1 = 1;  /* use MSI-X */
+               status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES,
+                                          &cmd, 0);
+               if (status != 0) {
+                       printk(KERN_ERR
+                              "myri10ge: %s: failed to set number of slices\n",
+                              dev->name);
+                       goto abort_with_nothing;
+               }
+               /* setup the indirection table */
+               cmd.data0 = mgp->num_slices;
+               status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_TABLE_SIZE,
+                                          &cmd, 0);
+
+               status |= myri10ge_send_cmd(mgp,
+                                           MXGEFW_CMD_GET_RSS_TABLE_OFFSET,
+                                           &cmd, 0);
+               if (status != 0) {
+                       printk(KERN_ERR
+                              "myri10ge: %s: failed to setup rss tables\n",
+                              dev->name);
+               }
+
+               /* just enable an identity mapping */
+               itable = mgp->sram + cmd.data0;
+               for (i = 0; i < mgp->num_slices; i++)
+                       __raw_writeb(i, &itable[i]);
+
+               cmd.data0 = 1;
+               cmd.data1 = myri10ge_rss_hash;
+               status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_ENABLE,
+                                          &cmd, 0);
+               if (status != 0) {
+                       printk(KERN_ERR
+                              "myri10ge: %s: failed to enable slices\n",
+                              dev->name);
+                       goto abort_with_nothing;
+               }
+       }
+
        status = myri10ge_request_irq(mgp);
        if (status != 0)
                goto abort_with_nothing;
@@ -1880,41 +2238,6 @@ static int myri10ge_open(struct net_device *dev)
        if (myri10ge_small_bytes > 0)
                mgp->small_bytes = myri10ge_small_bytes;
 
-       /* 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;
-               goto abort_with_irq;
-       }
-
-       if (myri10ge_wcfifo && mgp->wc_enabled) {
-               mgp->tx.wc_fifo = (u8 __iomem *) mgp->sram + MXGEFW_ETH_SEND_4;
-               mgp->rx_small.wc_fifo =
-                   (u8 __iomem *) mgp->sram + MXGEFW_ETH_RECV_SMALL;
-               mgp->rx_big.wc_fifo =
-                   (u8 __iomem *) mgp->sram + MXGEFW_ETH_RECV_BIG;
-       } else {
-               mgp->tx.wc_fifo = NULL;
-               mgp->rx_small.wc_fifo = NULL;
-               mgp->rx_big.wc_fifo = NULL;
-       }
-
        /* 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.
@@ -1929,9 +2252,44 @@ static int myri10ge_open(struct net_device *dev)
                mgp->big_bytes = big_pow2;
        }
 
-       status = myri10ge_allocate_rings(dev);
-       if (status != 0)
-               goto abort_with_irq;
+       /* setup the per-slice data structures */
+       for (slice = 0; slice < mgp->num_slices; slice++) {
+               ss = &mgp->ss[slice];
+
+               status = myri10ge_get_txrx(mgp, slice);
+               if (status != 0) {
+                       printk(KERN_ERR
+                              "myri10ge: %s: failed to get ring sizes or locations\n",
+                              dev->name);
+                       goto abort_with_rings;
+               }
+               status = myri10ge_allocate_rings(ss);
+               if (status != 0)
+                       goto abort_with_rings;
+               if (slice == 0)
+                       status = myri10ge_set_stats(mgp, slice);
+               if (status) {
+                       printk(KERN_ERR
+                              "myri10ge: %s: Couldn't set stats DMA\n",
+                              dev->name);
+                       goto abort_with_rings;
+               }
+
+               lro_mgr = &ss->rx_done.lro_mgr;
+               lro_mgr->dev = dev;
+               lro_mgr->features = LRO_F_NAPI;
+               lro_mgr->ip_summed = CHECKSUM_COMPLETE;
+               lro_mgr->ip_summed_aggr = CHECKSUM_UNNECESSARY;
+               lro_mgr->max_desc = MYRI10GE_MAX_LRO_DESCRIPTORS;
+               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;
+               if (lro_mgr->max_aggr > MAX_SKB_FRAGS)
+                       lro_mgr->max_aggr = MAX_SKB_FRAGS;
+
+               /* must happen prior to any irq */
+               napi_enable(&(ss)->napi);
+       }
 
        /* now give firmware buffers sizes, and MTU */
        cmd.data0 = dev->mtu + ETH_HLEN + VLAN_HLEN;
@@ -1948,25 +2306,15 @@ static int myri10ge_open(struct net_device *dev)
                goto abort_with_rings;
        }
 
-       cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->fw_stats_bus);
-       cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->fw_stats_bus);
-       cmd.data2 = sizeof(struct mcp_irq_data);
-       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA_V2, &cmd, 0);
-       if (status == -ENOSYS) {
-               dma_addr_t bus = mgp->fw_stats_bus;
-               bus += offsetof(struct mcp_irq_data, send_done_count);
-               cmd.data0 = MYRI10GE_LOWPART_TO_U32(bus);
-               cmd.data1 = MYRI10GE_HIGHPART_TO_U32(bus);
-               status = myri10ge_send_cmd(mgp,
-                                          MXGEFW_CMD_SET_STATS_DMA_OBSOLETE,
-                                          &cmd, 0);
-               /* Firmware cannot support multicast without STATS_DMA_V2 */
-               mgp->fw_multicast_support = 0;
-       } else {
-               mgp->fw_multicast_support = 1;
-       }
-       if (status) {
-               printk(KERN_ERR "myri10ge: %s: Couldn't set stats DMA\n",
+       /*
+        * Set Linux style TSO mode; this is needed only on newer
+        *  firmware versions.  Older versions default to Linux
+        *  style TSO
+        */
+       cmd.data0 = 0;
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_TSO_MODE, &cmd, 0);
+       if (status && status != -ENOSYS) {
+               printk(KERN_ERR "myri10ge: %s: Couldn't set TSO mode\n",
                       dev->name);
                goto abort_with_rings;
        }
@@ -1974,21 +2322,6 @@ static int myri10ge_open(struct net_device *dev)
        mgp->link_state = ~0U;
        mgp->rdma_tags_available = 15;
 
-       lro_mgr = &mgp->rx_done.lro_mgr;
-       lro_mgr->dev = dev;
-       lro_mgr->features = LRO_F_NAPI;
-       lro_mgr->ip_summed = CHECKSUM_COMPLETE;
-       lro_mgr->ip_summed_aggr = CHECKSUM_UNNECESSARY;
-       lro_mgr->max_desc = MYRI10GE_MAX_LRO_DESCRIPTORS;
-       lro_mgr->lro_arr = mgp->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;
-
-       napi_enable(&mgp->napi);        /* 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",
@@ -1996,8 +2329,6 @@ static int myri10ge_open(struct net_device *dev)
                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);
@@ -2005,9 +2336,9 @@ static int myri10ge_open(struct net_device *dev)
        return 0;
 
 abort_with_rings:
-       myri10ge_free_rings(dev);
+       for (i = 0; i < mgp->num_slices; i++)
+               myri10ge_free_rings(&mgp->ss[i]);
 
-abort_with_irq:
        myri10ge_free_irq(mgp);
 
 abort_with_nothing:
@@ -2017,21 +2348,22 @@ abort_with_nothing:
 
 static int myri10ge_close(struct net_device *dev)
 {
-       struct myri10ge_priv *mgp;
+       struct myri10ge_priv *mgp = netdev_priv(dev);
        struct myri10ge_cmd cmd;
        int status, old_down_cnt;
-
-       mgp = netdev_priv(dev);
+       int i;
 
        if (mgp->running != MYRI10GE_ETH_RUNNING)
                return 0;
 
-       if (mgp->tx.req_bytes == NULL)
+       if (mgp->ss[0].tx.req_bytes == NULL)
                return 0;
 
        del_timer_sync(&mgp->watchdog_timer);
        mgp->running = MYRI10GE_ETH_STOPPING;
-       napi_disable(&mgp->napi);
+       for (i = 0; i < mgp->num_slices; i++) {
+               napi_disable(&mgp->ss[i].napi);
+       }
        netif_carrier_off(dev);
        netif_stop_queue(dev);
        old_down_cnt = mgp->down_cnt;
@@ -2047,7 +2379,8 @@ static int myri10ge_close(struct net_device *dev)
 
        netif_tx_disable(dev);
        myri10ge_free_irq(mgp);
-       myri10ge_free_rings(dev);
+       for (i = 0; i < mgp->num_slices; i++)
+               myri10ge_free_rings(&mgp->ss[i]);
 
        mgp->running = MYRI10GE_ETH_STOPPED;
        return 0;
@@ -2143,7 +2476,7 @@ myri10ge_submit_req_wc(struct myri10ge_tx_buf *tx,
 
 /*
  * Transmit a packet.  We need to split the packet so that a single
- * segment does not cross myri10ge->tx.boundary, so this makes segment
+ * 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
@@ -2154,8 +2487,9 @@ myri10ge_submit_req_wc(struct myri10ge_tx_buf *tx,
 static int 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 = &mgp->tx;
+       struct myri10ge_tx_buf *tx;
        struct skb_frag_struct *frag;
        dma_addr_t bus;
        u32 low;
@@ -2166,6 +2500,9 @@ static int myri10ge_xmit(struct sk_buff *skb, struct net_device *dev)
        int cum_len, seglen, boundary, rdma_count;
        u8 flags, odd_flag;
 
+       /* always transmit through slot 0 */
+       ss = mgp->ss;
+       tx = &ss->tx;
 again:
        req = tx->req_list;
        avail = tx->mask - 1 - (tx->req - tx->done);
@@ -2180,7 +2517,7 @@ again:
 
        if ((unlikely(avail < max_segments))) {
                /* we are out of transmit resources */
-               mgp->stop_queue++;
+               tx->stop_queue++;
                netif_stop_queue(dev);
                return 1;
        }
@@ -2242,7 +2579,7 @@ again:
                        if (skb_padto(skb, ETH_ZLEN)) {
                                /* The packet is gone, so we must
                                 * return 0 */
-                               mgp->stats.tx_dropped += 1;
+                               ss->stats.tx_dropped += 1;
                                return 0;
                        }
                        /* adjust the len to account for the zero pad
@@ -2284,7 +2621,7 @@ again:
 
        while (1) {
                /* Break the SKB or Fragment up into pieces which
-                * do not cross mgp->tx.boundary */
+                * do not cross mgp->tx_boundary */
                low = MYRI10GE_LOWPART_TO_U32(bus);
                high_swapped = htonl(MYRI10GE_HIGHPART_TO_U32(bus));
                while (len) {
@@ -2294,7 +2631,8 @@ again:
                        if (unlikely(count == max_segments))
                                goto abort_linearize;
 
-                       boundary = (low + tx->boundary) & ~(tx->boundary - 1);
+                       boundary =
+                           (low + mgp->tx_boundary) & ~(mgp->tx_boundary - 1);
                        seglen = boundary - low;
                        if (seglen > len)
                                seglen = len;
@@ -2378,7 +2716,7 @@ again:
                myri10ge_submit_req_wc(tx, tx->req_list, count);
        tx->pkt_start++;
        if ((avail - count) < MXGEFW_MAX_SEND_DESC) {
-               mgp->stop_queue++;
+               tx->stop_queue++;
                netif_stop_queue(dev);
        }
        dev->trans_start = jiffies;
@@ -2420,12 +2758,12 @@ abort_linearize:
        if (skb_linearize(skb))
                goto drop;
 
-       mgp->tx_linearized++;
+       tx->linearized++;
        goto again;
 
 drop:
        dev_kfree_skb_any(skb);
-       mgp->stats.tx_dropped += 1;
+       ss->stats.tx_dropped += 1;
        return 0;
 
 }
@@ -2433,11 +2771,11 @@ drop:
 static int myri10ge_sw_tso(struct sk_buff *skb, struct net_device *dev)
 {
        struct sk_buff *segs, *curr;
-       struct myri10ge_priv *mgp = dev->priv;
+       struct myri10ge_priv *mgp = netdev_priv(dev);
        int status;
 
        segs = skb_gso_segment(skb, dev->features & ~NETIF_F_TSO6);
-       if (unlikely(IS_ERR(segs)))
+       if (IS_ERR(segs))
                goto drop;
 
        while (segs) {
@@ -2468,19 +2806,32 @@ drop:
 static struct net_device_stats *myri10ge_get_stats(struct net_device *dev)
 {
        struct myri10ge_priv *mgp = netdev_priv(dev);
-       return &mgp->stats;
+       struct myri10ge_slice_netstats *slice_stats;
+       struct net_device_stats *stats = &mgp->stats;
+       int i;
+
+       memset(stats, 0, sizeof(*stats));
+       for (i = 0; i < mgp->num_slices; i++) {
+               slice_stats = &mgp->ss[i].stats;
+               stats->rx_packets += slice_stats->rx_packets;
+               stats->tx_packets += slice_stats->tx_packets;
+               stats->rx_bytes += slice_stats->rx_bytes;
+               stats->tx_bytes += slice_stats->tx_bytes;
+               stats->rx_dropped += slice_stats->rx_dropped;
+               stats->tx_dropped += slice_stats->tx_dropped;
+       }
+       return stats;
 }
 
 static void myri10ge_set_multicast_list(struct net_device *dev)
 {
+       struct myri10ge_priv *mgp = netdev_priv(dev);
        struct myri10ge_cmd cmd;
-       struct myri10ge_priv *mgp;
        struct dev_mc_list *mc_list;
        __be32 data[2] = { 0, 0 };
        int err;
        DECLARE_MAC_BUF(mac);
 
-       mgp = netdev_priv(dev);
        /* can be called from atomic contexts,
         * pass 1 to force atomicity in myri10ge_send_cmd() */
        myri10ge_change_promisc(mgp, dev->flags & IFF_PROMISC, 1);
@@ -2616,13 +2967,14 @@ static void myri10ge_enable_ecrc(struct myri10ge_priv *mgp)
        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;
+                       struct pci_dev *prev_bridge, *old_bridge = bridge;
 
                        /* Walk the hierarchy up to the root port
                         * where ECRC has to be enabled */
                        do {
+                               prev_bridge = bridge;
                                bridge = bridge->bus->self;
-                               if (!bridge) {
+                               if (!bridge || prev_bridge == bridge) {
                                        dev_err(dev,
                                                "Failed to find root port"
                                                " to force ECRC\n");
@@ -2679,11 +3031,11 @@ static void myri10ge_enable_ecrc(struct myri10ge_priv *mgp)
  *
  * 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
+ * around unaligned completion packets (myri10ge_rss_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.
+ * larger than 2KB by setting the tx_boundary to 2KB.  If ECRC is
+ * enabled, then the driver should use the aligned (myri10ge_rss_eth_z8e.dat)
+ * firmware image, and set tx_boundary to 4KB.
  */
 
 static void myri10ge_firmware_probe(struct myri10ge_priv *mgp)
@@ -2692,7 +3044,7 @@ static void myri10ge_firmware_probe(struct myri10ge_priv *mgp)
        struct device *dev = &pdev->dev;
        int status;
 
-       mgp->tx.boundary = 4096;
+       mgp->tx_boundary = 4096;
        /*
         * Verify the max read request size was set to 4KB
         * before trying the test with 4KB.
@@ -2704,14 +3056,14 @@ static void myri10ge_firmware_probe(struct myri10ge_priv *mgp)
        }
        if (status != 4096) {
                dev_warn(dev, "Max Read Request size != 4096 (%d)\n", status);
-               mgp->tx.boundary = 2048;
+               mgp->tx_boundary = 2048;
        }
        /*
         * load the optimized firmware (which assumes aligned PCIe
         * completions) in order to see if it works on this host.
         */
        mgp->fw_name = myri10ge_fw_aligned;
-       status = myri10ge_load_firmware(mgp);
+       status = myri10ge_load_firmware(mgp, 1);
        if (status != 0) {
                goto abort;
        }
@@ -2737,7 +3089,7 @@ static void myri10ge_firmware_probe(struct myri10ge_priv *mgp)
                         "Please install up to date fw\n");
 abort:
        /* fall back to using the unaligned firmware */
-       mgp->tx.boundary = 2048;
+       mgp->tx_boundary = 2048;
        mgp->fw_name = myri10ge_fw_unaligned;
 
 }
@@ -2758,7 +3110,7 @@ static void myri10ge_select_firmware(struct myri10ge_priv *mgp)
                if (link_width < 8) {
                        dev_info(&mgp->pdev->dev, "PCIE x%d Link\n",
                                 link_width);
-                       mgp->tx.boundary = 4096;
+                       mgp->tx_boundary = 4096;
                        mgp->fw_name = myri10ge_fw_aligned;
                } else {
                        myri10ge_firmware_probe(mgp);
@@ -2767,12 +3119,12 @@ static void myri10ge_select_firmware(struct myri10ge_priv *mgp)
                if (myri10ge_force_firmware == 1) {
                        dev_info(&mgp->pdev->dev,
                                 "Assuming aligned completions (forced)\n");
-                       mgp->tx.boundary = 4096;
+                       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->tx_boundary = 2048;
                        mgp->fw_name = myri10ge_fw_unaligned;
                }
        }
@@ -2889,8 +3241,10 @@ static void myri10ge_watchdog(struct work_struct *work)
 {
        struct myri10ge_priv *mgp =
            container_of(work, struct myri10ge_priv, watchdog_work);
+       struct myri10ge_tx_buf *tx;
        u32 reboot;
        int status;
+       int i;
        u16 cmd, vendor;
 
        mgp->watchdog_resets++;
@@ -2938,19 +3292,26 @@ static void myri10ge_watchdog(struct work_struct *work)
 
                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));
+               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,
+                              (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,
+                              (int)ntohl(mgp->ss[i].fw_stats->
+                                         send_done_count));
+               }
        }
        rtnl_lock();
        myri10ge_close(mgp->dev);
-       status = myri10ge_load_firmware(mgp);
+       status = myri10ge_load_firmware(mgp, 1);
        if (status != 0)
                printk(KERN_ERR "myri10ge: %s: failed to load firmware\n",
                       mgp->dev->name);
@@ -2969,45 +3330,242 @@ static void myri10ge_watchdog(struct work_struct *work)
 static void myri10ge_watchdog_timer(unsigned long arg)
 {
        struct myri10ge_priv *mgp;
+       struct myri10ge_slice_state *ss;
+       int i, reset_needed;
        u32 rx_pause_cnt;
 
        mgp = (struct myri10ge_priv *)arg;
 
-       if (mgp->rx_small.watchdog_needed) {
-               myri10ge_alloc_rx_pages(mgp, &mgp->rx_small,
-                                       mgp->small_bytes + MXGEFW_PAD, 1);
-               if (mgp->rx_small.fill_cnt - mgp->rx_small.cnt >=
-                   myri10ge_fill_thresh)
-                       mgp->rx_small.watchdog_needed = 0;
-       }
-       if (mgp->rx_big.watchdog_needed) {
-               myri10ge_alloc_rx_pages(mgp, &mgp->rx_big, mgp->big_bytes, 1);
-               if (mgp->rx_big.fill_cnt - mgp->rx_big.cnt >=
-                   myri10ge_fill_thresh)
-                       mgp->rx_big.watchdog_needed = 0;
-       }
-       rx_pause_cnt = ntohl(mgp->fw_stats->dropped_pause);
-
-       if (mgp->tx.req != mgp->tx.done &&
-           mgp->tx.done == mgp->watchdog_tx_done &&
-           mgp->watchdog_tx_req != mgp->watchdog_tx_done) {
-               /* nic seems like it might be stuck.. */
-               if (rx_pause_cnt != mgp->watchdog_pause) {
-                       if (net_ratelimit())
-                               printk(KERN_WARNING "myri10ge %s:"
-                                      "TX paused, check link partner\n",
-                                      mgp->dev->name);
-               } else {
-                       schedule_work(&mgp->watchdog_work);
-                       return;
+       rx_pause_cnt = ntohl(mgp->ss[0].fw_stats->dropped_pause);
+       for (i = 0, reset_needed = 0;
+            i < mgp->num_slices && reset_needed == 0; ++i) {
+
+               ss = &mgp->ss[i];
+               if (ss->rx_small.watchdog_needed) {
+                       myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
+                                               mgp->small_bytes + MXGEFW_PAD,
+                                               1);
+                       if (ss->rx_small.fill_cnt - ss->rx_small.cnt >=
+                           myri10ge_fill_thresh)
+                               ss->rx_small.watchdog_needed = 0;
+               }
+               if (ss->rx_big.watchdog_needed) {
+                       myri10ge_alloc_rx_pages(mgp, &ss->rx_big,
+                                               mgp->big_bytes, 1);
+                       if (ss->rx_big.fill_cnt - ss->rx_big.cnt >=
+                           myri10ge_fill_thresh)
+                               ss->rx_big.watchdog_needed = 0;
                }
+
+               if (ss->tx.req != ss->tx.done &&
+                   ss->tx.done == ss->watchdog_tx_done &&
+                   ss->watchdog_tx_req != ss->watchdog_tx_done) {
+                       /* nic seems like it might be stuck.. */
+                       if (rx_pause_cnt != mgp->watchdog_pause) {
+                               if (net_ratelimit())
+                                       printk(KERN_WARNING "myri10ge %s:"
+                                              "TX paused, check link partner\n",
+                                              mgp->dev->name);
+                       } else {
+                               reset_needed = 1;
+                       }
+               }
+               ss->watchdog_tx_done = ss->tx.done;
+               ss->watchdog_tx_req = ss->tx.req;
        }
-       /* rearm timer */
-       mod_timer(&mgp->watchdog_timer,
-                 jiffies + myri10ge_watchdog_timeout * HZ);
-       mgp->watchdog_tx_done = mgp->tx.done;
-       mgp->watchdog_tx_req = mgp->tx.req;
        mgp->watchdog_pause = rx_pause_cnt;
+
+       if (reset_needed) {
+               schedule_work(&mgp->watchdog_work);
+       } else {
+               /* rearm timer */
+               mod_timer(&mgp->watchdog_timer,
+                         jiffies + myri10ge_watchdog_timeout * HZ);
+       }
+}
+
+static void myri10ge_free_slices(struct myri10ge_priv *mgp)
+{
+       struct myri10ge_slice_state *ss;
+       struct pci_dev *pdev = mgp->pdev;
+       size_t bytes;
+       int i;
+
+       if (mgp->ss == NULL)
+               return;
+
+       for (i = 0; i < mgp->num_slices; i++) {
+               ss = &mgp->ss[i];
+               if (ss->rx_done.entry != NULL) {
+                       bytes = mgp->max_intr_slots *
+                           sizeof(*ss->rx_done.entry);
+                       dma_free_coherent(&pdev->dev, bytes,
+                                         ss->rx_done.entry, ss->rx_done.bus);
+                       ss->rx_done.entry = NULL;
+               }
+               if (ss->fw_stats != NULL) {
+                       bytes = sizeof(*ss->fw_stats);
+                       dma_free_coherent(&pdev->dev, bytes,
+                                         ss->fw_stats, ss->fw_stats_bus);
+                       ss->fw_stats = NULL;
+               }
+       }
+       kfree(mgp->ss);
+       mgp->ss = NULL;
+}
+
+static int myri10ge_alloc_slices(struct myri10ge_priv *mgp)
+{
+       struct myri10ge_slice_state *ss;
+       struct pci_dev *pdev = mgp->pdev;
+       size_t bytes;
+       int i;
+
+       bytes = sizeof(*mgp->ss) * mgp->num_slices;
+       mgp->ss = kzalloc(bytes, GFP_KERNEL);
+       if (mgp->ss == NULL) {
+               return -ENOMEM;
+       }
+
+       for (i = 0; i < mgp->num_slices; i++) {
+               ss = &mgp->ss[i];
+               bytes = mgp->max_intr_slots * sizeof(*ss->rx_done.entry);
+               ss->rx_done.entry = dma_alloc_coherent(&pdev->dev, bytes,
+                                                      &ss->rx_done.bus,
+                                                      GFP_KERNEL);
+               if (ss->rx_done.entry == NULL)
+                       goto abort;
+               memset(ss->rx_done.entry, 0, bytes);
+               bytes = sizeof(*ss->fw_stats);
+               ss->fw_stats = dma_alloc_coherent(&pdev->dev, bytes,
+                                                 &ss->fw_stats_bus,
+                                                 GFP_KERNEL);
+               if (ss->fw_stats == NULL)
+                       goto abort;
+               ss->mgp = mgp;
+               ss->dev = mgp->dev;
+               netif_napi_add(ss->dev, &ss->napi, myri10ge_poll,
+                              myri10ge_napi_weight);
+       }
+       return 0;
+abort:
+       myri10ge_free_slices(mgp);
+       return -ENOMEM;
+}
+
+/*
+ * This function determines the number of slices supported.
+ * The number slices is the minumum of the number of CPUS,
+ * the number of MSI-X irqs supported, the number of slices
+ * supported by the firmware
+ */
+static void myri10ge_probe_slices(struct myri10ge_priv *mgp)
+{
+       struct myri10ge_cmd cmd;
+       struct pci_dev *pdev = mgp->pdev;
+       char *old_fw;
+       int i, status, ncpus, msix_cap;
+
+       mgp->num_slices = 1;
+       msix_cap = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
+       ncpus = num_online_cpus();
+
+       if (myri10ge_max_slices == 1 || msix_cap == 0 ||
+           (myri10ge_max_slices == -1 && ncpus < 2))
+               return;
+
+       /* try to load the slice aware rss firmware */
+       old_fw = mgp->fw_name;
+       if (old_fw == myri10ge_fw_aligned)
+               mgp->fw_name = myri10ge_fw_rss_aligned;
+       else
+               mgp->fw_name = myri10ge_fw_rss_unaligned;
+       status = myri10ge_load_firmware(mgp, 0);
+       if (status != 0) {
+               dev_info(&pdev->dev, "Rss firmware not found\n");
+               return;
+       }
+
+       /* hit the board with a reset to ensure 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");
+               goto abort_with_fw;
+               return;
+       }
+
+       mgp->max_intr_slots = cmd.data0 / sizeof(struct mcp_slot);
+
+       /* tell it the size of the interrupt queues */
+       cmd.data0 = mgp->max_intr_slots * sizeof(struct mcp_slot);
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_SIZE, &cmd, 0);
+       if (status != 0) {
+               dev_err(&mgp->pdev->dev, "failed MXGEFW_CMD_SET_INTRQ_SIZE\n");
+               goto abort_with_fw;
+       }
+
+       /* ask the maximum number of slices it supports */
+       status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_RSS_QUEUES, &cmd, 0);
+       if (status != 0)
+               goto abort_with_fw;
+       else
+               mgp->num_slices = cmd.data0;
+
+       /* Only allow multiple slices if MSI-X is usable */
+       if (!myri10ge_msi) {
+               goto abort_with_fw;
+       }
+
+       /* if the admin did not specify a limit to how many
+        * slices we should use, cap it automatically to the
+        * number of CPUs currently online */
+       if (myri10ge_max_slices == -1)
+               myri10ge_max_slices = ncpus;
+
+       if (mgp->num_slices > myri10ge_max_slices)
+               mgp->num_slices = myri10ge_max_slices;
+
+       /* Now try to allocate as many MSI-X vectors as we have
+        * slices. We give up on MSI-X if we can only get a single
+        * vector. */
+
+       mgp->msix_vectors = kzalloc(mgp->num_slices *
+                                   sizeof(*mgp->msix_vectors), GFP_KERNEL);
+       if (mgp->msix_vectors == NULL)
+               goto disable_msix;
+       for (i = 0; i < mgp->num_slices; i++) {
+               mgp->msix_vectors[i].entry = i;
+       }
+
+       while (mgp->num_slices > 1) {
+               /* make sure it is a power of two */
+               while (!is_power_of_2(mgp->num_slices))
+                       mgp->num_slices--;
+               if (mgp->num_slices == 1)
+                       goto disable_msix;
+               status = pci_enable_msix(pdev, mgp->msix_vectors,
+                                        mgp->num_slices);
+               if (status == 0) {
+                       pci_disable_msix(pdev);
+                       return;
+               }
+               if (status > 0)
+                       mgp->num_slices = status;
+               else
+                       goto disable_msix;
+       }
+
+disable_msix:
+       if (mgp->msix_vectors != NULL) {
+               kfree(mgp->msix_vectors);
+               mgp->msix_vectors = NULL;
+       }
+
+abort_with_fw:
+       mgp->num_slices = 1;
+       mgp->fw_name = old_fw;
+       myri10ge_load_firmware(mgp, 0);
 }
 
 static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
@@ -3015,7 +3573,6 @@ 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 dac_enabled;
@@ -3030,7 +3587,6 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 
        mgp = netdev_priv(netdev);
        mgp->dev = netdev;
-       netif_napi_add(netdev, &mgp->napi, myri10ge_poll, myri10ge_napi_weight);
        mgp->pdev = pdev;
        mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
        mgp->pause = myri10ge_flow_control;
@@ -3076,11 +3632,6 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
        if (mgp->cmd == NULL)
                goto abort_with_netdev;
 
-       mgp->fw_stats = dma_alloc_coherent(&pdev->dev, sizeof(*mgp->fw_stats),
-                                          &mgp->fw_stats_bus, GFP_KERNEL);
-       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;
@@ -3117,26 +3668,24 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
        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 = dma_alloc_coherent(&pdev->dev, bytes,
-                                               &mgp->rx_done.bus, GFP_KERNEL);
-       if (mgp->rx_done.entry == NULL)
-               goto abort_with_ioremap;
-       memset(mgp->rx_done.entry, 0, bytes);
-
        myri10ge_select_firmware(mgp);
 
-       status = myri10ge_load_firmware(mgp);
+       status = myri10ge_load_firmware(mgp, 1);
        if (status != 0) {
                dev_err(&pdev->dev, "failed to load firmware\n");
-               goto abort_with_rx_done;
+               goto abort_with_ioremap;
+       }
+       myri10ge_probe_slices(mgp);
+       status = myri10ge_alloc_slices(mgp);
+       if (status != 0) {
+               dev_err(&pdev->dev, "failed to alloc slice state\n");
+               goto abort_with_firmware;
        }
 
        status = myri10ge_reset(mgp);
        if (status != 0) {
                dev_err(&pdev->dev, "failed reset\n");
-               goto abort_with_firmware;
+               goto abort_with_slices;
        }
 
        pci_set_drvdata(pdev, mgp);
@@ -3181,24 +3730,27 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
                dev_err(&pdev->dev, "register_netdev failed: %d\n", status);
                goto abort_with_state;
        }
-       dev_info(dev, "%s IRQ %d, tx bndry %d, fw %s, WC %s\n",
-                (mgp->msi_enabled ? "MSI" : "xPIC"),
-                netdev->irq, mgp->tx.boundary, mgp->fw_name,
-                (mgp->wc_enabled ? "Enabled" : "Disabled"));
+       if (mgp->msix_enabled)
+               dev_info(dev, "%d MSI-X IRQs, tx bndry %d, fw %s, WC %s\n",
+                        mgp->num_slices, mgp->tx_boundary, mgp->fw_name,
+                        (mgp->wc_enabled ? "Enabled" : "Disabled"));
+       else
+               dev_info(dev, "%s IRQ %d, tx bndry %d, fw %s, WC %s\n",
+                        mgp->msi_enabled ? "MSI" : "xPIC",
+                        netdev->irq, mgp->tx_boundary, mgp->fw_name,
+                        (mgp->wc_enabled ? "Enabled" : "Disabled"));
 
        return 0;
 
 abort_with_state:
        pci_restore_state(pdev);
 
+abort_with_slices:
+       myri10ge_free_slices(mgp);
+
 abort_with_firmware:
        myri10ge_dummy_rdma(mgp, 0);
 
-abort_with_rx_done:
-       bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
-       dma_free_coherent(&pdev->dev, bytes,
-                         mgp->rx_done.entry, mgp->rx_done.bus);
-
 abort_with_ioremap:
        iounmap(mgp->sram);
 
@@ -3207,10 +3759,6 @@ abort_with_wc:
        if (mgp->mtrr >= 0)
                mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
 #endif
-       dma_free_coherent(&pdev->dev, sizeof(*mgp->fw_stats),
-                         mgp->fw_stats, mgp->fw_stats_bus);
-
-abort_with_cmd:
        dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
                          mgp->cmd, mgp->cmd_bus);
 
@@ -3231,7 +3779,6 @@ 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)
@@ -3246,19 +3793,15 @@ static void myri10ge_remove(struct pci_dev *pdev)
        /* avoid a memory leak */
        pci_restore_state(pdev);
 
-       bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
-       dma_free_coherent(&pdev->dev, 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
-       dma_free_coherent(&pdev->dev, sizeof(*mgp->fw_stats),
-                         mgp->fw_stats, mgp->fw_stats_bus);
-
+       myri10ge_free_slices(mgp);
+       if (mgp->msix_vectors != NULL)
+               kfree(mgp->msix_vectors);
        dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
                          mgp->cmd, mgp->cmd_bus);
 
@@ -3291,6 +3834,15 @@ 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) {
+               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;
+       }
+
        return pci_register_driver(&myri10ge_driver);
 }